WHISPERING GALLERY MODE BIOPARTICLE SENSING …research.poly.edu/~sarnold/paper/Pub_704_6242.pdf ·...

Ph.D. Thesis 2009 Biomedical Engineering

D. Keng – Whispering Gallery Mode bioparticle sensing and transport

- i -

WHISPERING GALLERY MODE BIOPARTICLE

SENSING AND TRANSPORT

DISSERTATION

Submitted in Partial Fulfillment

Of the Requirements for the

Degree of

DOCTOR OF PHILOSOPHY (BIOMEDICAL ENGINEERING)

at the

POLYTECHNIC INSTITUTE OF NEW YORK UNIVERSITY

by

Ta Kang (David) Keng

May 2009

Approved by:

____________________________________

Department Head

____________________________________

Date

Copy No. _____



- ii -

COPYRIGHT

Copyright by

Ta Kang (David) Keng

And

MicroParticle PhotoPhysics Lab (MP3L)

© 2009



- iii -

GUIDANCE COMMITTEE APPROVALS



- iv -

MICROFILM OR OTHER COPIES OF THIS DISSERTATION

MAY BE OBTAINED FROM:

UMI Dissertations Publishing

Bell & Howell Information and Learning

300 North Zeeb Road

P.O. Box 1346

Ann Arbor, Michigan 48106-1346



- v -

VITA

Ta Kang (David) Keng was born on June 17, 1983 in Moorhead, Minnesota. He obtained

a B.S. in Electrical Engineer in 2005, and a M.S. in Biomedical Engineering in 2007 from

Polytechnic University. As an undergrad, he took Prof. Stephen Arnold’s honor’s physics

class and then he joined Arnold’s MicroParticle PhotoPhysics Lab (MP3L) in summer

2002 as a research assistant. He was particularly interested in the instrumentation and

automation of the Whispering Gallery Mode Biosensor (WGMB) system. He developed

many gadgets turning the WGMB into a robust research tool for discovery. During his

MS studies, he developed a microfluidic system enabling the WGMB to perform in situ

surface chemistry for unlabeled specific detection of virus. He was accepted into the

Polytechnic-SUNY Downstate Biomedical Engineering PhD program in 2007, after

obtaining his MS degree. His PhD thesis topic was WGM single bioparticle sensing and

transport. He co-authored a publication on single Influenza A virus detection in the

summer of 2008. After that, he focused his research on the WGM Carousel, which

utilizes the cavity enhanced nearfield optical force to sense and actively trap individual

particles. The WGMC effect is revolutionary and it solves numerous biosensing

challenges in one shot, including: Active transport within the boundary layer (enhanced

detection rate), uniform sensor response (particle mass spectrometer in liquid), and

particle-surface interaction (surface probe with nanometer resolution). During his MP3L

stay, he generated seven refereed publications, one book chapter, one provisional patent,

and possibly more.

June 11, 2009



- vi -

DEDICATION

TO

MY FELLOW HOMO SAPIENS SAPIENS



- vii -

ACKNOWLEDGMENT

I would like to give special thanks to my thesis advisor Prof. Stephen Arnold for giving

me a chance to join the cutting edge research. I thank him for his academic and non-

academic guidance, his dedication to science, and his kindness to his students and lab

members. He is the best guidance professor.

I would like to also give special thanks to thesis co-advisor Prof. Iwao Teraoka for his

care to his student.

I would like to thank my co-workers and affiliates in the MP3L and in the Polytechnic:

(in approximate chronological order in which I met them)

Prof. Neil Wotherspoon, Dr. Mazia Khoshsima, Dr. Mayumi Noto, Dr. Frank Vollmer,

Dr. Steve Holler, Dr. Jason Guan, Dr. Ivan Selesnick (committee), Dr. Ravi Ramjit, Dr.

Volkan Otugen, Dr. Richard Seasholtz, Dr. Grigory Adamovsky, Ophir Gaathon, Dr.

Charles Martucci, Momchil Mihnev, Jelena Culic- Viskota, Dr. Noel L.Goddard, Dr.

Kolchenko, Dr. Bruce Garetz (committee), Minnie Chan, Dr. Subrata Saha (committee),

Suzy McAnanama, Monica Agarwal, Dr. Siyka Shopova, Dr. Walter Zurawsky

(committee), Raaj Rajmangal, Mariya Gelman, and many others.

I would like to thank my parents, my sister, and the rest of the family members for their

full support, special thanks to my aunt Jenny Chen, who died of cancer and who made me

set my goal to be a biomedical engineer. I would like to thank all my friends, and special

thanks to my girlfriend Ying Chen for her support.



- viii -

ABSTRACT

We are in a constant battle with viral-born diseases.1 (eg. Swine flu, avian flu, HIV,

HepC, West Nile, smallpox, and etc.) To prevent worldwide pandemics, rapid

identification and sensitive detection are keys to intervention. Here we present a label-

free, real-time single virus detection platform derived from a fusion between physics,

telecom fiber-optics, microfluidics, and biochemistry. The result, the Whispering Gallery

Mode (WGM) biosensor, has demonstrated real-time single Influenza-A virion

detection,2 and this on-going development may be the most sensitive gadget ever built for

identifying individual virions.3 The WGM biosensor detects the virus by transducing the

virus’ optical polarizability into an optical resonance shift. The virus binds to the

antibody decorated sensor surface, and the natural interaction allows the virus particle to

be sensed without the need for labeling. The individual virus’ mass is also obtained from

the measured polarizability and adds an additional dimension in identifying the virus. In

addition, the resonant nature of the WGM allows the light to build up inside the sensor,

and the accumulated light forms a near-filed optical tweezers. This intense optical field

breaks the limitation of the diffusing-only transport within the boundary layer, and

actively grabs the particle in the surrounding fluids into the sensing region and

contributes to enhancing the particle detection rate by >100 fold.4 This active force gives

WGM biosensor unmatched advantages in guaranteeing the uniform sensor response, and

provides a new method for studying surface binding in real time using WGM

fluctuations.5

1 P.W. Ewald, “Mastering Disease in the Next Fifty Years,” Ed. John Brockman, (Vintage Books, 2002). 2 F. Vollmer, S. Arnold, D. Keng, Proc. National Academy of Science, 105, 20701-20704 (2008) 3 S. Arnold, R. Ramjit, D. Keng, V. Kolchenko, I. Teraoka, Faraday Discussion, 137, 65-83 (2008) 4 S. Arnold, D. Keng, S.I. Shopova, S. Holler, W. Zurawsky, and F. Vollmer, Optics Express, 17, 6230-

6238 (2009) 5 D. Keng, S.R. McAnanama, I. Teraoka, and S. Arnold, Appl. Phys. Lett., 91, 103902 (2007)



- ix -

TABLE OF CONTENT

1 INTRODUCTION .................................................................................................... 1

1.1 THE GENERAL THEME FOR THIS THESIS .........................................................................................1 1.2 THE FOCUS OF THE THESIS.............................................................................................................1

2 WGM BIOSENSOR HEURISTICS........................................................................ 2

2.1 RAY TRAPPING ..............................................................................................................................2 2.2 WAVE TRAPPING ...........................................................................................................................2 2.3 PERTURBATION AND RESONANT WAVELENGTH SHIFT ...................................................................3 2.4 EXPERIMENTAL APPROACH ...........................................................................................................4 2.5 QUALITY FACTOR..........................................................................................................................5 2.6 SINGLE PARTICLE PERTURBATION .................................................................................................5 2.7 WGM EVANESCENT PROFILE ........................................................................................................6 2.8 SINGLE HIV1 VIRUS DETECTION LIMIT..........................................................................................8 2.9 WGM CAROUSEL .........................................................................................................................9 2.10 WGM CAROUSEL AS AN INTERFACIAL NANO-PROBE ..................................................................12

3 DESCRIPTION OF LIST OF PUBLICATIONS................................................ 15

4 WGM BIOSENSING SYSTEM OVERVIEW..................................................... 18

4.1 INTRODUCTION ...........................................................................................................................18 4.2 ELECTRICAL / OPTICAL ...............................................................................................................19 4.3 MICROFLUIDIC / MACROFLUIDIC ................................................................................................20 4.4 OVERALL PERFORMANCE ............................................................................................................21

5 OPTICS – LASER CALIBRATION..................................................................... 22

5.1 DFB LASER WAVELENGTH DEPENDENCE ....................................................................................22 5.2 CALIBRATION PROCEDURE AND PROGRAM..................................................................................24 5.3 CALIBRATION COEFFICIENTS FITTING AND ERROR ESTIMATE ......................................................26

6 OPTICS – TAPERED FIBER ............................................................................... 28

6.1 SINGLE MODE OPTICAL FIBER......................................................................................................28 6.2 FIBER TAPERING METHOD ...........................................................................................................28 6.3 FIBER TAPERING APPARATUS ......................................................................................................29 6.4 MOUNTING OF A TAPERED FIBER.................................................................................................35



- x -

7 OPTICS – MICROSPHERE FABRICATION.................................................... 39

7.1 FLAME MELTING LATHE APPARATUS...........................................................................................39 7.2 CO2 LASER MELTING APPARATUS................................................................................................43

8 MICROFLUIDIC SYSTEM.................................................................................. 46

8.1 BASIC FUNCTION REQUIREMENT FOR MICROFLUIDIC...................................................................46 8.2 REPLICA MOLDING PROCESS........................................................................................................46 8.3 MOLD MASTER DESIGN ...............................................................................................................48 8.4 CNC ZEROING AND CUTTING PROCEDURE...................................................................................50 8.5 MOLDING PROCESS .....................................................................................................................56 8.6 OVERVIEW AND COST ANALYSIS .................................................................................................57

9 MACROFLUIDIC SYSTEM................................................................................. 58

9.1 BASIC FUNCTION OF THE MACROFLUIDIC SYSTEM.......................................................................58 9.2 PUMP AND TUBING SELECTION ....................................................................................................59 9.3 PUMP CONTROL SOFTWARE AND HARDWARE ..............................................................................62

10 LABVIEW PROGRAM – DIP TRACKING ....................................................... 64

10.1 BASIC FUNCTION OF THE PROGRAM.............................................................................................64 10.2 PROGRAM STRUCTURE ................................................................................................................64 10.3 SOFTWARE/HARDWARE INTERFACE............................................................................................65 10.4 TRIGGER TIMING AND NOISE REDUCTION ....................................................................................70 10.5 DIP DETECTION AND WAVELENGTH CONVERSION .......................................................................71 10.6 DIP TRACKING CORE ALGORITHM................................................................................................72 10.7 LASER TEMPERATURE CONTROL AND INTERFACE .......................................................................74 10.8 RUN TIME UTILITY – LOG KEEPER ...............................................................................................74 10.9 CHANGING LASER WITH MATCHING CALIBRATION FILE...............................................................75 10.10 OVERVIEW AND PRECISION .........................................................................................................75

11 PUMP PROBE COUPLING AND COMPACT MICROFLUIDIC .................. 76

11.1 CONTINUOUS FIBER COUPLING METHOD .....................................................................................76 11.2 PUMP PROBE COUPLING METHOD ................................................................................................77 11.3 DROP CHANNEL ON THE PLANAR SURFACE..................................................................................78 11.4 HIGH Q – TUNABLE COUPLING ....................................................................................................78 11.5 ZERO BACKGROUND ...................................................................................................................79 11.6 MINIATURIZATION OF THE MICROFLUIDIC...................................................................................80 11.7 CONSTRUCTION...........................................................................................................................80



- xi -

11.8 TAPER FIBER MANUFACTURING – HF METHOD............................................................................80 11.9 MICROFLUIDIC WAVEGUIDE CHIP DESIGN: INTRODUCTION .........................................................83 11.10 PREFABRICATED FIBER TAPERS - MULTI-FIBER PARALLEL ETCHING ...........................................84 11.11 WAVEGUIDE CHIP FABRICATION ................................................................................................86 11.12 APPLICATION – ZRO2 COATED SILICA MICROSPHERE COUPLING .................................................87 11.13 OVERALL ....................................................................................................................................89

12 LIST OF ABBREVIATIONS ................................................................................ 90

13 REFERENCES........................................................................................................ 91



- xii -

TABLE OF FIGURES

Figure 2-1 Light caught in a bottle by total internal reflection, where n1 and n2 are the

refractive index of the sphere and the surrounding medium....................................... 2

Figure 2-2 (a) Wave representation of the WGM, and perturbation – note that the wave

penetrate beyond the dielectric boundary, this is the origin for the evanescent field.

(b) Layer perturbation of the WGM, note that the wave moves outward, and the

wavelength is extended to re-close the orbit in phase................................................. 3

Figure 2-3 Waveguide excitation of a WGM, and typical spectrum of many WGM dips.

WGM orbit represented in red, note that multiple orbits can be formed which

corresponds to multiple dips. The microfluidic channel (not shown) provides liquid

coverage to the sphere and the tapered fiber............................................................... 4

Figure 2-4 (a) Dip spectrum before and after adsorption. (b). Dip trace for continuous

adsorption to saturation............................................................................................... 5

Figure 2-5 Intensity distribution with latitude near the sphere equator. The latitudinal

width is approximately 5 μm while the radial intensity falls off in approximately 150

nm. .............................................................................................................................. 7

Figure 2-6 Limit of detection (LOD) in terms of number of HIV1 for a given microsphere

R.................................................................................................................................. 8

Figure 2-7 Water absorption spectrum. Note that the minimum water absorption occurs at

400nm. With the laser wavelength reduction from 1300nm to 780nm, the water

absorption is 100x less; from 1300nm to 400nm, 10,000x less.................................. 9

Figure 2-8 Near field optical trap. (a) Evanescent decay of the light at the dielectric

boundary. (b) Light intensity represented as ray diagram, thicker and longer red

vectors represent stronger light. As the rays are being refracted, the change in

momentum of each produces a light-force (in black). The imbalance in force drives

the particle toward the high intensity region (e.g. the sphere surface) and also

propels the particle in the direction of the light propagation. ................................... 10

Figure 2-9 WGM Carousel. (a) A nanoparticle scattering light within the Carousel. In the

image, the microsphere edge is out of focus. Note that the particle is traveling in the



- xiii -

light direction. (b) The electronics sample the resonant wavelength, and the dip trace

shows a delimited wavelength shift. The fluctuations in the dip trace correspond to

the radial direction Brownian motion, and can be observed as blinking in scattered

light. With higher trapping power, the particle is drawn closer to the surface, and the

dip trace fluctuate closer to the green delimited line. ............................................... 11

Figure 2-10 Potential plot derived from the histogram of the dip trace fluctuation. (a) the

histogram, (b) the potential plot and the fit of the total potential. ............................ 13

Figure 2-11 Potential comparison with ionic screening by introducing additional NaCl

into the solution. The maximum count in the histogram is closer to the surface as the

ionic strength is increased and the surface repulsive potential is screened. ............. 14

Figure 4-1 WGM Biosensor system with an illustration of the microfluidic setup. Tapered

fiber is UV glued on a standard microscope glass slide (1˝x3˝) and capped by a

PDMS microfluidic channel. The zoom-in shows the fiber taper region and the

microsphere coupling. Note that the light is launched from the left, and a scattering

spot is observed on the right edge of the sphere. ...................................................... 18

Figure 4-2 WGM Biosensor system with detailed electronic connections....................... 19

Figure 4-3 Temperature control block and the tapered fiber UV glued on to a glass slide.

................................................................................................................................... 20

Figure 5-1 The interior of a 780nm Eagleyard DFB laser, note that this is not a telecomm

laser and does NOT have an isolator built in. The laser diode output is focused to a

fiber secured by a ceramic fiber holder, and the output intensity is monitored using a

photodiode and a ball lens on the other end of the laser chip. The entire assembly is

secured on a Peltier cooler. A thermistor is positioned to the top left of the fiber

holder. The butterfly package’s pin spacing is 0.1˝. ................................................. 22

Figure 5-2 Diagram for DFB laser calibration. The LabVIEW program controls and

monitors both the laser driver and wavelength meter through GPIB interface. The

laser driver drives the laser with a specific combination of diode drive current and

temperature (i and T), and the laser outputs a wavelength to be measured by the

wavelength meter and the readout is recorded by the LabVIEW program. This

calibration procedure is performed many times to measure wavelengths produced by



- xiv -

multiple combinations of (i, T) to form a full set of sample for calibration coefficient

fitting......................................................................................................................... 23

Figure 5-3 Wavelength meter. The input accepts FC/PC optical fibers, the input range is

set to match the intended wavelength, and the GPIB address is set to 3. The intensity

indicator should be maintained in the green region of the analog bar indicator for the

entire current/temperature span to prevent faulty reading. Light can be attenuated

with external attenuator if necessary......................................................................... 24

Figure 5-4 The user interface of a LabVIEW program used for calibration of a DFB laser

(Laser_cal.llb). The wavelength outputs appears to change in many ramps. Each

ramp consists of multiple current tuning steps (13 steps, from 30 mA to 90 mA at 5

mA increment, one sample dot per step) at a fixed temperature. With the

configuration shown, each ramp is 0.5oK apart. (e.g. 15oC, 15.5oC, 20oC, …) In this

setting, the relaxation time is significantly shorter than recommended, which leads

to a bumpy ramp and calibration error. For this particular laser, the current tuning

yields ~0.1 nm / 60 mA, and the temperature tuning yields ~0.1 nm / 2 oK............. 25

Figure 5-5 Laser driver. For calibration purpose, the modulation (MOD) BNC input is

left open. The current and temperature are strictly controlled by LabVIEW using

GPIB. GPIB address is set to 1. ................................................................................ 25

Figure 5-6 MATLAB program (LaserCal.exe) for linear fit. The calibration coefficients

will be stored in the same folder as the data file. The RMS error on the right shows

the accuracy of the fitting. ........................................................................................ 26

Figure 6-1 Image of a tapered fiber reconstructed from 33 individual microscopy images.

(The image index number is in red.) The 0.5 mm spacing marker on the top was

printed on a transparency using a laser printer and it is placed adjacent to the fiber

when the images were taken. The printed marker also serves as reconstruction

landmarks – two adjacent images can be aligned by overlapping the ink particle

granule patterns. The measured waist diameter is 3±0.5 μm with 15 mm of pulled

length (described later). The non-tapered portion is 125 μm. The image was zoomed

vertically 3x for clarity.............................................................................................. 29

Figure 6-2 Flame design. (a) Gas configuration, green denotes oxygen, blue for propane,

and red for the mixture. The flame is stabilized by the pinched section of the nozzle.



- xv -

(b) Two views on the flame, and the actual flame. Top view – The fiber is positioned

off centered. Side view – The flame’s position relative to the fiber can be lowered to

increase the temperature at the fiber; flame gets hotter at the top. The brass flame

guard is 14 mm in diameter. ..................................................................................... 30

Figure 6-3 Fiber puller microcontroller diagram for PIC16F84....................................... 31

Figure 6-4 Fiber pulling apparatus. The arrows indicate the stage movements. The linear

actuator is to the right of the “flame stage” tag. Fiber is removed from the v-groove

for clarity; see the next figure for a comparison. The T-joint feeds the mixed gas to

the nozzle. Stage 1 is used only when a thinner fiber is required. Optical table has

tapped holes 1˝ apart. ................................................................................................ 31

Figure 6-5 Pulling mechanism for stage 2. The fiber is positioned in the v-groove and is

secured by a magnet.................................................................................................. 34

Figure 6-6 Vial containing a Cytop solution, a pipette, and a holding device with a

template slide. The substrate glass slide is mounted on top of the template slide. A 2

μL pipette deposits the Cytop solution onto the substrate indicated by the black

marker dots on the template slide. ............................................................................ 35

Figure 6-7 Mounting a fiber on a substrate glass slide with UV glue. Alignment screws

s1, s2, and s3 ensure that the slide can be reproducibly positioned on the gluing

platform. A drop of UV glue is applied on each edge of the glass slide. The flame is

left on to heat the platform. The warm platform lowers the UV glue viscosity and

helps the glue to flow into the polymer-slide gap. Note the double image of the

tapered fiber produced by the reflection at the glass surface.................................... 36

Figure 6-8 CCD video system for fiber alignment. (a) A microscope/CCD is mounted on

a long horizontal translation stage inspects the entire length of the fiber. (b) Images

of the fiber taper image and its reflection, and half-distance between the images is

the fiber-slide distance, 125μm. (c) End of polymer-stripped section of the fiber.

This image clearly shows that the polymer is secured on the slide surface, and the

remaining silica sits parallel to the surface. .............................................................. 37

Figure 6-9 UV curing to secure the fiber on the substrate................................................ 38

Figure 7-1 Automatic flame lathe and real time monitoring through a microscope. The

yellow arrow indicates the fiber feed direction......................................................... 40



- xvi -

Figure 7-2 Automatic flame lathe. A belt driving mechanism rotates the fiber chuck. The

stage release solenoid mechanism is in the engaged position as shown here. When

disengaged, the chuck is recessed by one inch towards the microscope eyepiece.

This allows the sphere to be reproducibly positioned inside the flame, and released

quickly at the push of a button on the controller box. The microscope is mounted on

the same stage as the chuck so that the sphere size can also be determined after the

solenoid release the stage. Inset: stepper motor controller. ...................................... 41

Figure 7-3 Microsphere size measurement system and some sample microspheres ........ 42

Figure 7-4 CO2 laser melting setup. The CO2 laser is focused through a ZnSe lens and

monitored by a microscope/CCD system. The CO2 output can be changed by the

knob on the laser controller box................................................................................ 43

Figure 7-5 Optical setup for the CO2 laser. A reflector is placed in front of the focal point,

allowing the sphere to be heated from both sides. By translating the microsphere

closer to or further away from the reflector, the microsphere can be made to bend

toward or away from the reflector. The reflector is set at a slight angle to avoid

reflection back to the laser. ....................................................................................... 44

Figure 7-6 Close-up view of the CO2 setup. Inset: the laser forms a microsphere. The

filter glass is placed below the microscope objective to protect the CCD. Note a

glow of a silica microsphere and the white smoke accumulated on the beam

reflector. This can be easily cleaned with an ethanol wipe....................................... 44

Figure 8-1 The master for MF replica molding. (a) The exploded view showing all the

required components (M0-M3) of the master. (b) The master is assembled by

sequentially stacking the four plates, M3~M0, from top to bottom. M0 contains

tapped holes for the 1˝ long 8-32 screws to secure the entire stack. The 1˝ long 21

gauge hypodermic needle is inserted into the master for molding the through-hole.47

Figure 8-2 CAD program and the generated G-code. The tool path are indicated by

arrows, and the G-code can be scrolled through line by line to check the accuracy.

The U shaped MF channel is designed in the middle, surrounded by a rectangular

superstructure............................................................................................................ 48

Figure 8-3 Key design features of M1. All the cutouts are highlighted in blue, and the

through-hole is in red. The fiber cutouts are 0.02˝ high, and the microsphere



- xvii -

entrance and sample inlet cutouts are 0.04˝ high. The superstructure’s outer edge are

2˝×0.5˝. Fluidic channel corners are rounded to minimize the dead volume. The two

fluidic walls are designed to be 0.13˝ apart, accounting for the end mill width of

0.03˝, to have the final fluidic channel width of 0.1˝................................................ 49

Figure 8-4 Physical CNC tool parameter defined in the CAD, material top is set to zero.

................................................................................................................................... 50

Figure 8-5 CNC end mill after Z zeroing. A 0.50˝ thick aluminum stock is used to level

and raise the acrylic closer to the top of the acrylic holding chuck.......................... 51

Figure 8-6 CNC control software run under Linux operating system (Shirline CNC

(inch)). Here a slightly different microfluidic allows two spheres to be coupled to

one tapered fiber; note the additional cutout. The end mill currently is located at the

resting position right now, X=0, Y=0, and Z=0.3 (in inches). The A Axis is the

rotation option that is not being used. The G-code file is loaded through

AUTO/Open, and is then executed line by line. The line being executed is

highlighted in red. This figure is showing the last line being executed. The backplot

view allows the user to monitor the end mill position in real time. In addition to the

microfluidic G-code, hole.txt is also executed, producing 4 pilot holes in the corners.

................................................................................................................................... 52

Figure 8-7 Cutting in progress. Water is added as lubricant and coolant to prevent the end

mill from melting the acrylic. ................................................................................... 53

Figure 8-8 Cutting completed. The end mill moves back to the resting position, z=0.3˝. 53

Figure 8-9 Close-up view of M1 after machining. ........................................................... 54

Figure 8-10 Close-up view of M0 after machining. ......................................................... 54

Figure 8-11 Close-up view of M2 after machining. M2 has a rectangular cavity cutout. 55

Figure 8-12 Close-up view of M3 after machining. The injection holes are two-step

drilled to fit a Luer tip syringes. The drain hole is positioned at one corner of the

PDMS cavity to collect and push out trapped bubbles. ............................................ 55

Figure 8-13 Bubble removing procedure once PDMS is filled: First the bolted master sits

flat (left) so the bubble would move to the top. Then, tilting the master to push the

bubble out of the drain hole. ..................................................................................... 56



- xviii -

Figure 9-1 Macrofluidic and Microfluidic interface. A buffer tubing is connected to the

PDMS microfluidic with a stainless steel hypodermic needle that has a gauge

identical to the one that was used to mold the through-hole. The sample is injected

using a removable 25 gauge needle held down by a hairclip. The sample inlet can

slide in and out through the PDMS cutout. The drain is connected to a weak

vacuum. All connections are press fit, and no glues or adhesives are used.............. 58

Figure 9-2 Macrofluidic system. The valve/plunger diagram for sample loop loading and

injection is also shown. A 0.22 μm filters is used on the buffer line........................ 59

Figure 9-3 Macrofluidic setup with a laptop computer control. The blue T-valves are

fitted at the pump output. The T-valve handle direction indicates the stop branch.

Here, the sample loading syringe and buffer refilling are stopped. The data logging

screen showing a dip in the fiber transmission spectrum is to the right edge of the

figure. ........................................................................................................................ 60

Figure 9-4 pump control program. The plunger direction control (the three position

control adjacent to the syringe indicator) can be used to drive the plunger up,

neutral, or down. The number of steps delivered, the buffer and sample volume, and

various other control parameters are also displayed. ................................................ 61

Figure 9-5 Pump drive mechanism and header pin definitions. Although not being used in

the program design, HM is the optical chopper signal output, which could be used to

indicate the home position of the plunger and the valve. ......................................... 63

Figure 10-1 The three layer flow chart for the dip trace program. ................................... 64

Figure 10-2 The main LabVIEW program (main.vi) used for dip trace. Each section has

its functional description labeled and commented. These labels will be referred to

throughout the chapter. Various parts of this program will be zoomed in and

described in details in the rest of the chapter. ........................................................... 66

Figure 10-3 L1, configured to be triggered by the falling edge of the TTL ..................... 67

Figure 10-4 Input channels (blue) and the operation of main.vi. The spectrum (having 2

dips) is sampled for every other ramp separated by an idling time [symbol …]. The

dip tracking algorithm is performed during the sample idling. Both edges of the

display spectrum is truncated (shaded) in the analysis to avoid complications from

the transient. .............................................................................................................. 68



- xix -

Figure 10-5 L2a and L2g section for converting the voltage data into power and laser

current through their respective transfer functions. .................................................. 69

Figure 10-6 The delayed start on the triggering frame. The TTL triggering delay and the

resulting error has a 1:1 relationship. This figure is an exaggeration, and the actual

delay is on the order of 1ms or so (1% of the frame). .............................................. 70

Figure 10-7 Dip fitting routine and conversion of the dip location from index (L2b) to

current (L2c) , then to wavelength (L2d).................................................................. 71

Figure 10-8 Single dip lock and track (L2f). User can select one of the dips from a

spectrum.................................................................................................................... 72

Figure 10-9 Dip tracing procedure illustrated with Dselect in red, and the other dips in blue.

Step 0 shows the preceding spectrum, and step 1, 2, and 3 display the current

spectrum. A new Dselect’ is assigned in steps 2,3. All the dips in the spectrum are

located in step 1, but only the dip closest to Dselect is designated as Dselect′. .......... 73

Figure 10-10 An example of the dip trace and log keeper, taken from a virus detection

experiment................................................................................................................. 75

Figure 11-1 : Continuous fiber coupling, side and top views. Arrows indicate light

propagation direction. Solid line: strong coupling; dashed line: weak coupling.

(Figures not to scale)................................................................................................. 76

Figure 11-2 : Pump-probe configuration with two fast tapered fibers, side view ............ 77

Figure 11-3 : Other possible pump and probe coupling configuration............................. 77

Figure 11-4 : Field of the excited microsphere, bottom view, not to scale. Notice the

divergent light exiting the pump fiber. ..................................................................... 79

Figure 11-5 : HF etching process to manufacture a taper. Top green layer is silicon oil,

and the bottom blue layer is HF solution. ................................................................. 81

Figure 11-6 : The meniscus at the HF (bottom) and silicone oil (top) interface. A taper is

being formed above the interface.............................................................................. 82

Figure 11-7 : Top: Close-up view of the taper region, 8 μm per division. Below: Fiber

profile and light leakage test, 15 μm per division. The effective region of the fiber is

evident by a scattering of a red laser. Both photos were taken in air. ...................... 82

Figure 11-8 : Dual pump-probe coupling configuration with two microspheres ............. 84



- xx -

Figure 11-9 : Etched fibers. Two fibers are etched simultaneously. The taper profiles are

similar. ...................................................................................................................... 84

Figure 11-10 : Two fibers glued in a parallel configuration. Notice two spacer fibers with

jacket in the middle................................................................................................... 85

Figure 11-11 : Assembled waveguide chip. Two pairs of tapers glued onto a glass slide

................................................................................................................................... 86

Figure 11-12 : Microspheres coupled to tapers in a waveguide chip ............................... 87

Figure 11-13 : Fiber coupling to a ZrO2 coated silica microsphere in air. Note the shallow

dip. ............................................................................................................................ 87

Figure 11-14 : Pump-probe coupling to the same ZrO2 coated microsphere. Note the

easily detectable peaks, and large values of Q.......................................................... 88



- 1 -

1 Introduction

1.1 The general theme for this thesis

Early disease prevention has enhanced efficacy if pathogen detection can be accomplish-

ed both sensitively and rapidly.1 Label-free, specific, ultra-sensitive biosensor design

enjoys the distinct advantage of minimum sample preparation and is therefore a prime

focus to many. In particular, leveraging the existing commercial optical fiber technology

and natural ligand analyte specific interaction, Whispering Gallery Mode (WGM)

resonant optical biosensors demonstrate various biosensing applications at ~picomolar

(pM) levels. 2 , 3 , 4 , 5 However, the ultimate biosensing involving discrete counting of

extremely dilute analytes, at femtomolar (fM) or attomolar (aM) levels, otherwise known

as “single particle detection,” was an open challenge when I started my graduate career. I

took on the challenge, and single particle detection became my main theme for this thesis.

At the end, the goal was met 6 with elegant discoveries that have allowed us to

characterize the analyte-surface physical chemistry and guided us to design an ultimate

biosensor with capabilities far beyond the obvious.7

1.2 The focus of the thesis

Since many of our research results are already published, I will try to document the

experimental techniques that were utilized but were not published due to page limits. The

documentation of the techniques as well as the publications that are attached at the end of

this thesis should paint a detailed picture of the flow of experiments and thoughts. The

intention here is to provide a technical complement to the published research and not to

re-write what has already been published. In this way the experimental art will not be

lost.



- 2 -

2 WGM Biosensor Heuristics

2.1 Ray trapping

A Whispering gallery Mode (WGM) is formed by “capturing light in a bottle” – Light

can be trapped by total internal reflection (TIR) in a dielectric circular structure such as a

sphere, cylinder or disk as shown in Figure 2-1. This TIR forces the light to take on a

polygonal path and preserves its energy by dielectric confinement. In our case, we use a

highly symmetrical glass microsphere as our light bottle because of its simplicity for

fabrication and analysis. Regardless of the material or shape of choice, capturing light in

a confined structure can lead to some very interesting and useful effects such as the

optical resonance.

Figure 2-1 Light caught in a bottle by total internal reflection, where n1 and n2 are the refractive

index of the sphere and the surrounding medium

2.2 Wave trapping

With the wave-particle duality principle, light can also be represented as a particle or

wave. In the wave representation, the orbit is closed in-phase forming a Whispering

Gallery Mode (WGM) as shown in Figure 2-2a. If the light is coherently pumped into the

sphere, the constructive interference can continuously add energy. Pumping in this way

leads to resonance, and produces a necklace of waves wrapped around the circumference.

This condition has an orbital wavelength λs. The wave nature of the light allows the

photon to penetrate beyond the dielectric boundary and into the surroundings as an



- 3 -

evanescent field. This field interacts with bio-particles in the surrounding medium, and

this interaction, in turn, changes the resonance condition and corresponding orbital

wavelength. To understand this in a geometrical way, imagine a layer of adsorbed

particles that have the same dielectric properties as the underlying sphere (Figure 2-2). A

necklace having the same number of waves must grow in circumference in order to

satisfy the resonance condition. This requires driving the resonator with a longer

wavelength.

Figure 2-2 (a) Wave representation of the WGM, and perturbation – note that the wave penetrate

beyond the dielectric boundary, this is the origin for the evanescent field. (b) Layer perturbation of

the WGM, note that the wave moves outward, and the wavelength is extended to re-close the orbit in

phase.

2.3 Perturbation and resonant wavelength shift

In other words, the orbital wavelength (λs) has to be increased by Δλs to re-close the orbit

in phase. In Figure 2-2b, a uniform layer perturbation (ex. adsorbed protein) with

thickness t is shown. The resulting Δλs is geometrically proportional to the layer

thickness,

Δλs / λs = t / R Equation 2-1

where R is the orbit radius. Note that this equation assumes the added layer carries the

same refractive index as the sphere itself. The layer perturbation can also be induced by



- 4 -

thermal expansion8 or mechanical stress9, which would increase the effective sphere size.

In this case, our particular interest is biosensing – the perturbation due to water-born

bioparticle (analyte ex.: proteins, virus, and bacteria).10 We should see how this is carried

out experimentally.

2.4 Experimental approach

Figure 2-3 Waveguide excitation of a WGM, and typical spectrum of many WGM dips. WGM orbit

represented in red, note that multiple orbits can be formed which corresponds to multiple dips. The

microfluidic channel (not shown) provides liquid coverage to the sphere and the tapered fiber.

A simplified system diagram is shown in Figure 2-3.5 In practice, the sphere is fabricated

by melting a silica fiber tip using a flame or a CO2 laser. The softened silica naturally

balls up and forms a microspheroid under surface tension, while the remaining fiber

serves as a convenient handle. In order to observe the WGM, a flame-tapered fiber

positioned near the sphere serves as a waveguide to stimulate and interrogate the WGM.

For biosensing, both the sphere and fiber are immersed in water in a microfluidic

channel. The evanescent field of the 3-micron-thin fiber overlaps the evanescent field of

the sphere, which allows light to be tunneled into the WGM. This process is reversible,

allowing the light to return to the fiber. Each tunneling event leads to a 90 degree phase

shift, so that overall the light returning from the sphere is 180 degrees out of phase with

the light that was not originally coupled. This destructive interference can be observed by

the photodiode (PD) as an intensity dip as the laser light is tuned through each WGM.

The wavelength tuning of the distributed feedback laser (DFB) allows the multiple dips



- 5 -

to be clearly resolved in a scan spanning only 0.2nm. The laser wavelength at the dip

minimum is proportional to the orbital wavelength (through dielectric constants

associated with the sphere and solute) and will be defined as λr. Figure 2-4a shows a dip

spectrum, and the shifted spectrum. Through continuous monitoring of the dips, λr can be

plotted against time, forming a dip trace associated with analyte adsorption to saturation,

as shown in Figure 2-4b.

Figure 2-4 (a) Dip spectrum before and after adsorption. (b). Dip trace for continuous adsorption to

saturation

2.5 Quality factor

While analyzing a dip trace, the ability to detect minimum Δλr governs how small of a

perturbation one can interpret. The “noise” in measuring Δλr is related to the dip

linewidth γ, defined by the full width at half maximum – For larger linewidth in relation

to the resonance wavelength shift, the uncertainty in determining Δλr increases. In

practice, by using a parabolic fitting routine, Δλr as small as 1/100 of γ can be resolved.

Similar to other electrical and mechanical resonators, the optical resonance linewidth is

determined by the system energy leakage rate. The quality factor (Q), defined as Q= λr/γ,

can be used as a sensitivity comparison between different systems: WGM (~107),

piezoelectric crystal resonant circuit (~105), and mechanical micro-cantilever (~105)11.

With such a high Q for a WGM sensor, detection of minute perturbations caused by a

single virus binding event is possible.

2.6 Single particle perturbation

For single particle adsorption, the perturbation of the orbit occurs locally near the particle

and it is different from the layer perturbation depicted in Figure 2-2b. However, the



- 6 -

overall effective orbit is lengthened and requires a positive shift in λs. When the particle

diffuses into the evanescent field, the laser tracking the resonance shifts by Δλr, which is

related to the location and size of the particle. Because the evanescent field can extend

hundreds of nm into the solution, the analyte can be sensed remotely without the need to

physically bind or touch the sphere surface. The Δλr produced by the WGM interaction

with a single nano-particle may be written as

2

2

( )2 ( )

analyteex

cavity cavity

E r

E r dVαλ

λ εΔ

=∫

Equation 2-2

where αex is the excess polarizability of an analyte particle, and the later term is a ratio

describing the relative light intensity in the cavity versus the portion that polarizes the

analyte. The polarizability αex is directly proportional to the mass of the nano-particle.

Since most proteins are made out of the same group of amino acids with similar overall

density, aex is therefore a good estimate of the protein size and mass. In addition the

compacted DNA and RNA in a virus have similar dielectric properties to protein. Now

we will analyze the evanescent intensity distribution (the ratio term), which is somewhat

more complicated.

2.7 WGM evanescent profile

It is best to think of the microsphere geometrically as a miniature Earth, with a system of

parallels and meridians. The ribbon of the WGM is launched at the equator (Figure 2-5),

however the mode is that of a wave with a wave function. Angular quantization rules

distribute this wave-function in patterns known as spherical harmonics. The simplest of

these has a square modulus (i.e. intensity) with a single lobe in latitude. This lowest order

angular mode has a nearly Gaussian shape extending by a degree or so above and below

in latitude. For our spheres which are about 80 μm in diameter, the Gaussian width is

about 5 μm. The intensity falls off exponentially in the radial direction with a

characteristic length of about 150nm, forming a ribbon of light, Figure 2-5.



- 7 -

Figure 2-5 Intensity distribution with latitude near the sphere equator. The latitudinal width is

approximately 5 μm while the radial intensity falls off in approximately 150 nm.

Considering that the shift in Eq.2-2 is proportional to the intensity at the nanoparticle, a

considerable variation in Δλr will result from the particle binding at a random position in

latitude. Because of this, WGM has a drawback that even though single particle is

expected to produce discrete a Δλr, the sensor response can vary considerably. This non-

uniformity problem is later resolved with the invention and implementation of a WGM

Carousel mechanism, which utilizes light forces to focus adsorption to the highest

intensity at the equator. For now, we will investigate the ideal case – the maximized Δλr

created by one equator-bound HIV virus, and theoretically analyze the possibility for

detection.



- 8 -

2.8 Single HIV1 virus detection limit

Figure 2-6 Limit of detection (LOD) in terms of number of HIV1 for a given microsphere R

Considering the empirical minimum detectable Δλr is about γ/100, we estimate the limit

of detection for the WGM sensor as displayed in terms of least number of equator-bound

HIV1 virions (NLOD) in Figure 2-6. Each HIV1 virion has a mass of 8×10-16 grams.5 The

reader should note that under ideal conditions: using a microspherical cavity of 40 μm

radius and a driving wavelength near 780nm, we anticipate sensitivity down to 1/10 the

mass of a single HIV1 particle (NLOD=0.1). At larger microsphere radii, we anticipate a

loss in sensitivity that varies as R5/2. This is a result of the denominator in Eq.2-2 which is

approximately in proportion to the microsphere volume. Below 40 μm, sensitivity is once

again lost due to light leakage through the high curvature surface the microsphere

equator.

A key reason for the enhanced sensitivity at 780nm (comparing to 1300nm) is

associated with the absorption spectrum of water. Overtone vibrations increase water

absorption in the near infrared (Figure 2-7), causing an increase in WGM linewidth γ and

a decrease in wavelength shift detection sensitivity. As a consequence the sensitivity for

wavelength shift detection increases by 100× as the laser wavelength decreases from

1300nm to 780nm. Although a further increase in sensitivity is expected down to 400 nm,

there are no DFB lasers currently available below 780nm. Although the 400 nm case is



- 9 -

not shown in Figure 2-6, we would expect the use of such a laser to enhance sensitivity

by an additional factor of 100x, leading to the detection of a single bio-particle with

1/1000 the mass of HIV. Our approach certainly seems promising, but the sensitivity

analysis in Figure 2-6 demands the bio-particle be bound at the equator for the signal

strength to be maximum. With a freely diffusing virus such as HIV1, exclusive equatorial

binding seems impossible, unless an “angel” force were available; a deterministic force

that can transport a particle to the equator. As fate would have it, just such an “angel”

force was discovered.

Figure 2-7 Water absorption spectrum. Note that the minimum water absorption occurs at 400nm.

With the laser wavelength reduction from 1300nm to 780nm, the water absorption is 100x less; from

1300nm to 400nm, 10,000x less.

2.9 WGM Carousel

We normally think of transport in terms of 19th century concepts of fluid dynamics. With

the inception of lasers and the interest of nano-science, there is more awareness of the

ability of light to generate forces that become particularly useful in manipulating bio-

particles. Our recent discovery that light can draw particles to the active sensing region

and propel them along the equator rescues miniature sensors from the “starvation” of

diffusion. This is where the WGM Carousel mechanism makes a remarkable contribution



- 10 -

to biosensing. The high Q nature of the WGM causes stored energy to build up

resonantly, which leads to the guiding of bio-particles to the equator with less than 0.2

mW of drive power. The intense light within the orbit along with other interfacial forces

gives rise to the Carousel. It works because the photon momentum flux tangent to the

equator falls off exponentially in the radial directions as shown in Figure 2-8. The rapid

decay of this photon momentum flux in the radial evanescent region generates an

interaction that propels particles forward while drawing them to the equatorial surface

(Figure 2-8).

Figure 2-8 Near field optical trap. (a) Evanescent decay of the light at the dielectric boundary. (b)

Light intensity represented as ray diagram, thicker and longer red vectors represent stronger light.

As the rays are being refracted, the change in momentum of each produces a light-force (in black).

The imbalance in force drives the particle toward the high intensity region (e.g. the sphere surface)

and also propels the particle in the direction of the light propagation.

As the intensity decays in the radial direction, r, the momentum flux density decreases.

This can be depicted by the ray diagram, Figure 2-8b, – As a particle enters this non-

uniform field, the evanescent light is refracted. The change of momentum in the refracted

photon requires a force (Newton’s 2nd Law), and the reaction to this force (Newton’s 3rd

Law) pushes the particle toward the highest intensity at the surface, and propels the

particle forward along the light direction. This light-force accompanied by an interfacial

electrostatic repulsion (to be described later) and causes the particle to be trapped radially

and propelled to circumnavigate in the WGM orbit as in Figure 2-9, hence the term

“WGM Carousel.” The Carousel trap dimensionally reduces transport by comparison

with free-diffusion leading to more rapid detection.



- 11 -

While the particle is propelled in the Carousel, if a binding site is available, the

analyte may attempt to bind with each revolution, or it can escape the trap due to a

stochastic thermal kick. The Carousel’s radial binding energy Uc , not to be confused with

the chemical binding energy of an antibody, is controlled by the optical power P driving

the resonator. The Carousel is considerably different in its approach to transport to

chemical binding sites.

Figure 2-9 WGM Carousel. (a) A nanoparticle scattering light within the Carousel. In the image, the

microsphere edge is out of focus. Note that the particle is traveling in the light direction. (b) The

electronics sample the resonant wavelength, and the dip trace shows a delimited wavelength shift.

The fluctuations in the dip trace correspond to the radial direction Brownian motion, and can be

observed as blinking in scattered light. With higher trapping power, the particle is drawn closer to

the surface, and the dip trace fluctuate closer to the green delimited line.

The Carousel’s drive takes the place of a fluid drive in a micro-fluidic system; nano-

particles are not driven by fluidics, but by a light force. It is also both large and small;

large in the sense that it is open to analyte, and small in terms of the evanescent Carousel

volume confining the trap; the trap volume is ~25 fL with an invisible outer wall. In

addition, the response of the sensor is uniform.



- 12 -

The response of the sensor at binding is proportional to the local intensity, and since the

particle is trapped within the Carousel in the pre-binding state, the particle has a high

probability to bind within the intensity maximum of the Carousel. By increasing the

optical binding potential to overcome an electrostatic repulsive surface potential, the

particle stays trapped only within the Carousel but is repelled at other latitudes. This

manipulation of potential allows the particle to be preferentially attracted to the WGM

orbit, and therefore produces a uniform sensor response. This uniform signal is directly

proportional to the analyte’s polarizability, and therefore to its volume/mass, which

allows for mass spectrometry of nanoparticles in solution.

The size/mass spectrometry can also be realized without the need for binding. In a low

ionic strength solution, the sphere’s repulsive surface is not screened and can reach tens

of nm away from the surface. However, Brownian fluctuations allow the particle to

undergo stochastic changes in its height from the surface, so that there is a finite

probability for “colliding” with the surface. This momentary state produces a maximum

in wavelength shift from which the polarizability of the particle can be determined. Most

of these collisions do not lead to binding but rather a delimited stream of fluctuations in

the wavelength shift, as shown in Figure 2-9. Not only can the maximum in these

fluctuations lead to particle’s size, but the distribution of the fluctuation can be used to

extract the radial potential energy profile.

2.10 WGM Carousel as an interfacial nano-probe

Wavelength shift statistics provide a means for following the particle’s position with

time. Since the wavelength shift is in proportion to the evanescent intensity at a particle

center h+a (see Figure 2-10), and this intensity decays exponentially with h+a, it is

possible to utilize this shift as a nanoscopic proximity sensor:

r

( ) exp[ ] = exp[ ]( ) exp[ ]

h + a h+ a L h La a L

λλ

Δ=

Δr -

--

( ) Equation 2-3

where L is the characteristic evanescent length, h is the particle-microsphere interfacial

separation, and a is the particle’s radius. With this relationship, the wavelength shift

statistics can be transformed into separation statistics, as shown in Figure 2-10a. In

addition, by assuming thermal equilibrium, the Boltzman relation can utilize these



- 13 -

statistics for determining the interfacial potential energy, U(h)= -kBT ln[p(h) / p(href)],

where p(h) is the probability density to be at separation h, and href is a separation that

U(href) is taken to be zero. Figure 2-10b shows the resulting interaction potential energy.

The plot reveals that the combined surface repulsive potential and the optical binding

potential form a stable minimum at h = 35 nm. This figure reveals two contributions to

the interaction energy.

Figure 2-10 Potential plot derived from the histogram of the dip trace fluctuation. (a) the histogram,

(b) the potential plot and the fit of the total potential.

The optical binding energy is generated through the gradient of the evanescent intensity.

This force is similar to the force at work in optical tweezers. It is manifestly

conservative12 and falls off with the intensity of the field. Consequently, its characteristic

is on the order of the evanescent length L. In addition, there exists a repulsive potential

between the negatively charge groups on the microsphere surfaces and the negatively

charged nanoparticle. This combination leads to a balanced radial force at 35 nm.

Although the optical gradient force is independent of ionicity, the repulsive force has an

associated Debye length λb, which is sensitive to the ionic strength of the environmental

solution. By increasing the ionicity of the solution with added salt, the λb is decreased,

and the potential minimum shifts toward the surface, as shown in Figure 2-11.



- 14 -

Figure 2-11 Potential comparison with ionic screening by introducing additional NaCl into the

solution. The maximum count in the histogram is closer to the surface as the ionic strength is

increased and the surface repulsive potential is screened.

Further increase of salt allows the particle to be trapped even closer and be bound by a

short range Van der Waals attraction. The binding is found to be strong as evident by the

inability of the particle to be unbound after a neutral water rinse. By adjusting the

combination of ionic strength and optical power, it becomes possible to fully isolate the

equator for binding. This discovery paves the way for light induced functionalization of

surfaces.

Lastly, we would also like to point out that optical induced thermal convection

may also be involved in the long range transport. The resonator’s high Q can build up

>10W of power within a microsphere with a 50 μm radius. Although the water’s optical

absorption coefficient is small, the built-up high power may generate sufficient heat

causing a convective flow in water. This convection flow can reach mm or cm out into

the solution, and gives the WGM Carousel an additional invisible arm to grab particles

into the sensing region.



- 15 -

3 Description of list of publications Here is a list of publication that I was involved in. The publications are attached at the

end of the thesis. The important scientific values and my contribution to each specific

publication are listed as well.

7. Whispering Gallery Mode Carousel – a photonic mechanism for enhanced

nanoparticle detection in biosensing

S. Arnold, D. Keng, S. Shopova, S. Holler, W. Zurawsky, F. Vollmer

Optics Express, 17, 6230-6238 (2009)

This work (Aug 2008 ~ Feb 2009) investigates the WGM near field optical tweezers

capabilities. This research was constructed to provide an opto-mechanical transport

mechanism in the classical stagnant boundary layer nm above the biosensor surface.

Many important scientific discoveries are contained in this publication: WGM Carousel

effect, single nanoparticle –WGM sensor surface interaction, Debye layer manipulation

through ionic strength control, single nanoparticle sizing and mass spectroscopy, and

possibility for light induced functionalization. I designed, built, and conducted the

experiments as well as data and theoretical analysis.

6. Single virus detection from the reactive shift of a whispering-gallery mode

F. Vollmer, S. Arnold, and D. Keng

PNAS, 105, 20701-20704 (2008)

This work (June 2003 ~ June 2008) was an on-going research since my undergrad years.

The initial single particle results were first observed by F.V. at Harvard, and I later

reproduced the results with virus simulants and helped to analyze the data. This is the

first paper demonstrating single virus (Influenza A) resolution capability using WGM

reactive sensing principle.



- 16 -

5. MicroParticle photophysics illuminates viral bio-sensing

S. Arnold, R. Ramjit, D. Keng, V. Kolchenko and I. Teraoka

Faraday Discussions, 137, 65-83 (2008)

This work (Dec 2006 ~ Dec 2007) was the first publication to demonstrate WGM specific

virus detection in a microfluidic channel. The entire wet surface chemistry was done in

situ within the micro-channel and being monitored at all times. I design, built, and

conducted experiments as well as data analysis. In particular, I invented a procedure to

fabricate the microfluidic channel.

4. Resonance fluctuations of a whispering gallery mode biosensor by particles

undergoing Brownian motion

D. Keng, S. R. McAnanama, I. Teraoka, and S. Arnold

Appl. Phys. Lett., 91, 103902 (2007)

This work (March 2007 ~ Oct 2007) investigated the density fluctuation of colloids

within the WGM evanescent region. The mean size of the colloids was estimated through

the autocorrelation of the fluctuation. This was the first demonstration of in-solution

WGM stochastic analysis. I design, built, and conducted experiments as well as data

analysis.

3. Detection of Protein Orientation on the Silica Microsphere Surface Using

Transverse Electric/Transverse Magnetic Whispering Gallery Modes

M. Noto, D. Keng, I. Teraoka, and S. Arnold

Bio. Phys. J., 92, 4466-4472 (2007)

This work investigates the molecular orientation on the surface by measuring the shift

difference between the TE and TM polarization of light. I design and built the

experimental apparatus, and helped with the experiments and data analysis.



- 17 -

2. Molecular weight dependence of a whispering gallery mode biosensor

M. Noto, M. Khoshsima, D. Keng, I. Teraoka, V. Kolchenko, and S. Arnold

Appl. Phys. Lett., 87, 223901 (2005)

This work investigates the resonance shift caused by protein mono-layer non-specific

adsorption. The shift was found to be in proportion to the molecular weight. I wrote the

data acquisition and analysis software for the experiments, and helped with the

experimental apparatus and data analysis.

1. Nanolayer characterization through wavelength multiplexing of a microsphere

resonator

M. Noto, F. Vollmer, D. Keng, I. Teraoka, S. Arnold

Opt. Lett., 30, No. 5 (2005)

This work investigates the difference in resonance shift of two source wavelengths to the

same perturbation. The evanescent penetration depth is a function of wavelength, and by

monitoring the shifts from the two lasers (1300nm and 780nm), the signal can be used as

a nanoscopic ruler to measure the protein mono-layer thickness. I wrote the data

acquisition and analysis software for the experiments.



- 18 -

4 WGM Biosensing system overview

4.1 Introduction

The WGM biosensing experimental setup is composed of three main subsystems: optical,

electrical (hardware and software), and microfluidic. Figure 4-1 shows a typical

experimental setup5, and each experiment uses a slightly different setup. In Figure 4-2, a

complimentary system diagram is shown that includes the actual electrical and optical

connections used in the experiment. This was excluded from the original diagram.

In this chapter, a brief system overview is provided to serve as the outline of this thesis.

The details of system components that I have invented, designed, modified, and built are

then described in the rest of the chapters. Here we will start the system overview with the

description of electrical and optical components, followed by the microfluidic.

Figure 4-1 WGM Biosensor system with an illustration of the microfluidic setup. Tapered fiber is UV

glued on a standard microscope glass slide (1˝x3˝) and capped by a PDMS microfluidic channel. The

zoom-in shows the fiber taper region and the microsphere coupling. Note that the light is launched

from the left, and a scattering spot is observed on the right edge of the sphere.



- 19 -

Figure 4-2 WGM Biosensor system with detailed electronic connections.

4.2 Electrical / Optical

In Figure 4-2, a function generator produces a 10Hz ramp signal that modulates the laser

driver current. The laser driver serves as a transimpedance amplifier, converting the

modulating input voltage from the function generator into a ramping laser current. The

laser driver also controls the laser diode temperature. The pigtailed DFB laser’s

wavelength output (λ) depends on both the driving current (i) and the diode temperature

(T). The wavelength λ can be approximately tuned at 0.1 nm/oK and/or 2~3pm/mA, by

changing T and i respectively. These number are rough estimates, since λ(T, i) is non-

linear and different for each laser. The wavelength calibration is performed using a

commercial wavelength meter based on Michelson prior to the experiment, and the

calibration coefficients are stored in the computer (PC). During the experiment, the

computer monitors the modulation voltage and the diode temperature, then computes the

laser wavelength using on the stored calibration coefficients.

The pigtailed laser is connected to an external isolator (30dB), a polarization controller,

and a variable attenuator (these are not shown in Figure 4-1) to condition the light

launching into the tapered fiber. The tapered fiber is produced by heat pulling with a

propane/oxygen flame and then the tapered fiber is UV glued (depicted as green

rectangles) onto a microscope slide as shown in Figure 4-3. The fiber’s polymer coating

is left intact at the glue region and serves as an offset to prevent the tapered section from

physically touching the glass surface. The intensity through the taper is monitored by a



- 20 -

detector and data acquisition (DAQ) system. The DAQ is triggered by the sync signal

from the ramp function generator, so that each ramp corresponds to one wavelength scan

(e.g. resulting in one spectrum). The fiber and glass slide assembly is capped by a

microfluidic channel and then placed on a custom made Peltier thermal control block

with temperature stability of 0.01 oK.

Figure 4-3 Temperature control block and the tapered fiber UV glued on to a glass slide.

4.3 Microfluidic / Macrofluidic

The transparent microfluidic channel is made of polydimethylsiloxane (PDMS) and

replica molded. The acrylic mold master is micromachined by a computer numerical

controlled (CNC) milling machine. To make the microfluidic channels, the PDMS

prepolymer in liquid phase is injected into the master cavity and then thermally cured at

70 oC for an hour. The cured microfluidic channel is then removed from the master. The

master can be reused to mold more replicas. The cured hydrophobic PDMS adheres to the

glass slide forming a water tight seal. The microfluidic channel has pre-designed cutouts



- 21 -

in the walls to accommodate the tapered fiber and a microsphere. To prevent water

leakage at cutouts, a hydrophobic fluorinated polymer, Cytop, is diluted in

hexafluorobenzene (HFB) and pre-deposited onto the glass slide at the cutout locations

before UV gluing the fiber. After the deposition, the HFB quickly evaporates, leaving a

chemically inert 13 thin Cytop film behind. After capping the fiber with the PDMS

microfluidic channel, liquids are injected.

The liquid delivering system (i.e. macrofluidic) consists of two computer controlled

syringe pumps, one dedicated for buffer and the other one for sample. To prevent sample

cross contamination, a 2 m long Teflon tubing sample loop is used – the sample resides

only in the sample loop (and not in the syringe itself) prior to injection. A valve system

was designed to allow manual sample loading and cleaning without a need to reverse the

direction of the pump’s syringe plunger, thereby preventing inaccurate delivery possibly

caused by pump drive backlash.

The microfluidic drain is connected to a weak vacuum to carry away excess liquid. A

wide channel opening at the drain port provides a smaller flow resistance compared with

other smaller cutouts, and therefore guides the flow toward the drain direction. (The

hydrophobic Cytop at the cutouts also serves a similar purpose of guiding the flow and

preventing leakage.) When the liquid is injected, the computer program is turned on to

monitor the dip trace.

4.4 Overall performance

The DAQ and a LabVIEW program monitor the entire system. The dip trace is plotted

after measuring both the ramp voltage and the detector intensity in each scan. Although

each ramp should be identical, it is not in practice. Voltage and time drifts of the ramp do

exist, and monitoring allows the program to compensate the drift.

Overall, the sensing system is capable of tracking λr with 0.000 001nm (10-6 nm)

precision out of the nominal 1300 nm, this amounts to a noise figure of 1 part per billion.



- 22 -

5 Optics – Laser calibration

5.1 DFB laser wavelength dependence

The monolithic nature of the distributed feedback (DFB) laser allows its output

wavelength to be tuned robustly at MHz bandwidth.14 DFB lasers are single frequency

lasers that have a grating built in within the semiconductor gain medium to generate a

single frequency output. This non-mechanical tuning (as opposed to external cavity

lasers) is crucial to the stability required in precision spectroscopy. In addition, the

general public’s interest in telecomm has turned the DFB lasers (1300 nm, 1500 nm) into

a commodity item, and driving the unit price down to $30 each. Furthermore, all the fiber

optic accessories (e.g. variable attenuator, optical isolator, polarization controller, phase

modulators) are also available at low prices. Although extremely inexpensive at

telecomm band, the research band non-telecomm lasers (<1000nm) are much more

expensive ($2500). These research band DFB are used in narrow range high precision

spectroscopy for atomic physics and gas sensing purposes.

Figure 5-1 The interior of a 780nm Eagleyard DFB laser, note that this is not a telecomm laser and

does NOT have an isolator built in. The laser diode output is focused to a fiber secured by a ceramic

fiber holder, and the output intensity is monitored using a photodiode and a ball lens on the other

end of the laser chip. The entire assembly is secured on a Peltier cooler. A thermistor is positioned to

the top left of the fiber holder. The butterfly package’s pin spacing is 0.1˝.



- 23 -

Our interest is to move away from the telecomm band to a shorter wavelength (~700 nm

or shorter), where water absorption minimizes and therefore the resonator’s Q is

sufficiently high for single particle detection. Regardless of the laser’s nominal

wavelength, the output wavelength of the laser must be calibrated.

DFB laser’s wavelength λ(T, i) can be tuned by changing the drive current (i) or the

diode temperature (T). The tuning relationship is non-linear, and calibration is required.

The goal is to obtain a calibration equation λ(T, i). Once λ(T, i) is obtained, λ can then be

accurately computed if T and i are measured. Here, the computation of λ assumes that the

calibration does not change with time. The calibration is generally performed in static

configuration at constant current and temperature, although the drive current in actual

spectroscopy experiments constantly ramps. An advanced method, involving an atomic

standard and a temperature invariant ring resonator, has been invented to calibrate the

laser dynamically, but the discrepancy between the two calibration methods is small and

can be ignored as long as the laser is ramped at a constant frequency. Here we will

explain the procedure of the static calibration.

The laser is calibrated with an automatic calibration system. Figure 5-2 shows its diagram

and the data flow.

Figure 5-2 Diagram for DFB laser calibration. The LabVIEW program controls and monitors both

the laser driver and wavelength meter through GPIB interface. The laser driver drives the laser with

a specific combination of diode drive current and temperature (i and T), and the laser outputs a

wavelength to be measured by the wavelength meter and the readout is recorded by the LabVIEW

program. This calibration procedure is performed many times to measure wavelengths produced by

multiple combinations of (i, T) to form a full set of sample for calibration coefficient fitting.



- 24 -

5.2 Calibration procedure and program

A LabVIEW program (Laser_cal.llb), Figure 5-4, controls the laser driver (ILX

Lightwave, LDC3714B) through General Purpose Interface Bus (GPIB) communication.

The program is set to drive the laser at various values of the current and temperature,

while the laser wavelength λ is measured by a picometer wavelength meter (Advantest,

TQ8325). The LabVIEW program sets the laser current and temperature to user-specified

values, one combination at a time, while the wavelength meter measures λ with a 1 pm

front panel resolution. The error in wavelength is ~3 pm, and this error is partially caused

by the transients in the laser current temperature settings. To minimize the error, the

program allows a delay in reading the wavelength after each jump in the

temperature/current. Typical delay period is 10 s for the temperature jump, and 2 s for the

current jump. The program is set to current tune and then temperature tune.

Figure 5-3 Wavelength meter. The input accepts FC/PC optical fibers, the input range is set to match

the intended wavelength, and the GPIB address is set to 3. The intensity indicator should be

maintained in the green region of the analog bar indicator for the entire current/temperature span to

prevent faulty reading. Light can be attenuated with external attenuator if necessary.



- 25 -

Figure 5-4 The user interface of a LabVIEW program used for calibration of a DFB laser

(Laser_cal.llb). The wavelength outputs appears to change in many ramps. Each ramp consists of

multiple current tuning steps (13 steps, from 30 mA to 90 mA at 5 mA increment, one sample dot per

step) at a fixed temperature. With the configuration shown, each ramp is 0.5oK apart. (e.g. 15oC,

15.5oC, 20oC, …) In this setting, the relaxation time is significantly shorter than recommended, which

leads to a bumpy ramp and calibration error. For this particular laser, the current tuning yields ~0.1

nm / 60 mA, and the temperature tuning yields ~0.1 nm / 2 oK.

Figure 5-5 Laser driver. For calibration purpose, the modulation (MOD) BNC input is left open. The

current and temperature are strictly controlled by LabVIEW using GPIB. GPIB address is set to 1.



- 26 -

The output data is then stored in a text file in a user specified folder. A line of lists is

generated: i (mA), T (oC) and λ (nm), tab delimited for each combination. A typical data

file is shown as follows:

30.000 15.000 1058.685

35.000 15.000 1058.692

40.000 15.000 1058.703

45.000 15.000 1058.712

… … …

90.000 15.000 0.000

5.3 Calibration coefficients fitting and error estimate

The data file is then used by a polynomial linear fitting program (LaserCal.exe, written

by Steve Holler) to extract the coefficients for the calibration equation. Sometimes, a

faulty wavelength meter reading may occur that results in a wavelength measurement of

zero (as shown in red above, 0.000). The latter is usually caused by insufficient or excess

intensity at the wavelength meter input, which should be noticeable in the measured

wavelengths in Figure 5-4 (the figure shown does not contain such a point). The intensity

check should be performed for the current tuning extremes (30~90mA for Figure 5-4)

before using the LabVIEW program for automatic calibration.

Figure 5-6 MATLAB program (LaserCal.exe) for linear fit. The calibration coefficients will be stored

in the same folder as the data file. The RMS error on the right shows the accuracy of the fitting.



- 27 -

The fault-free data file can then be fitted by the MATLAB program, LaserCal.exe, as

shown in Figure 5-6. The program lists the fitting coefficients and the RMS error of the

fit. RMS error may be useful to check the robustness for each calibration run for the same

type of laser. The λ(T, i) used is as follows:

2 2 2 2 2 2λ(T,I) = a+b(T)+c(T )+d(I)+e(IT)+f(IT )+g(I )+h(I T)+i(I T ) Equation 5-1

A typical set of a~i coefficients for a 1059nm DFB laser is fitted as follows:

a = 1.0576815570658734E+003

b = 6.3604475851531769E-002

c = 1.9449452609848349E-005

d = 1.2937459442356864E-003

e = 3.9358779100440927E-006

f = 3.2008675901520519E-007

g = 2.9109110189266548E-006

h = -3.6653440692733332E-009

i = -2.0342809866634700E-009

These coefficients are then saved automatically as coeffs.txt in the original data file

folder by LaserCal.exe. Although the coefficient’s precision is higher than necessary, the

λ(T, i) computation is straightforward and no noticeable computer delay is generated

when this precision is used. The laser calibration is completed at this point, and the

coeffs.txt will be used specifically for the calibrated laser in the WGM experiments.



- 28 -

6 Optics – Tapered Fiber

6.1 Single mode optical fiber

For WGM excitation and detection, a tapered optical fiber is used as a waveguide to

evanescently couple light in and out of the microsphere.17 A typical telecommunication

grade single mode fiber (SMF) has a doped 9 μm core embedded in a pure silica cladding

with overall diameter of 125 μm. An additional acrylic polymer coating covering the

silica serves as the protection layer, making the total diameter 250 μm. In the normal

operating condition, the light is well shielded and guided in the core which is virtually

lossless. (0.3 dB/km).15 In our case, we want to expose the light by forcing the light into

the water medium as an evanescent wave.

6.2 Fiber tapering method

After stripping the polymer coating, there are two methods to reduce the diameter of the

cladding to expose the light – One is a chemical etch16, and the other is flame tapering.17

In practice, the hydrofluoric acid (HF) chemical etching is not favored for several

reasons: A high scattering loss from the etched rough surface (over 90%, the roughness

also leads to fragility), a long etching time (4 hrs+), and the toxicity. In contrast, the

pulled fiber, shown in Figure 6-1 , has a flame polished surface with low loss (<10%),

short preparation time (<10 minutes), high mechanical strength (tensile strength tested to

be ~10 GPa), and does not require a chemical fume hood. (Side notes: The 10 GPa is

measured by hanging a 1 gram load on a tapered fiber having a 3 μm waist diameter,

which is about 7 times the tensile strength compare to steel.) This high tensile strength

allows the tapered fiber to be easily handled.

The fiber tapering procedure was used by many other research groups, first applied to

WGM in 1997.17 I’ve learned how simple it was to pull a fiber from Mazur’s group at

Harvard.18 After the visit, I spent some time practicing pulling a fiber by hand using a

Bunsen burner, and then I designed an apparatus to do it automatically. This apparatus

was first used in the layer enhancement work19, and then a second version was designed

to produce a faster taper in a shorter distance so it can be fitted within a microfluidic



- 29 -

channel. The later uses a microflame to reduce the total pulling length, and is described in

what follows.

Figure 6-1 Image of a tapered fiber reconstructed from 33 individual microscopy images. (The image

index number is in red.) The 0.5 mm spacing marker on the top was printed on a transparency using

a laser printer and it is placed adjacent to the fiber when the images were taken. The printed marker

also serves as reconstruction landmarks – two adjacent images can be aligned by overlapping the ink

particle granule patterns. The measured waist diameter is 3±0.5 μm with 15 mm of pulled length

(described later). The non-tapered portion is 125 μm. The image was zoomed vertically 3x for clarity.

6.3 Fiber tapering apparatus

The fiber tapering apparatus pulls a fiber asymmetrically while heating the fiber above its

softening temperature of ~1200 oC. The final tapered fiber profile is determined by the

nature of the flame, the pulling distance and velocity.

A propane/oxygen flame can be controlled by gas flow and nozzle design. The laminar

premixed flame design is shown in Figure 6-2a. Propane is regulated at 25psi with a flow

rate of 10.5mL/min, while the oxygen is regulated at 10psi with flow rate adjusted for

optimum pulling. The theoretical oxygen/fuel ratio is 5:1 (5O2+C3H8→3CO2+4H2O),

which produces a flame temperature of 2800oC. However, this temperature would melt

the silica instead of softening it and oxidize the stainless steel nozzle. Therefore, oxygen

is reduced in the mixture to produce a cooler flame that looks like Figure 6-2b. The

nozzle is made with 1.5 inch long, 16 gauge stainless steel hypodermic tubing. The

nozzle is produced by pinching the tubing to have ½ of the cross sectional area at 3 mm

below the nozzle opening. The pinched section increases the local gas velocity and

stabilizes the flame. A brass flame guard serves multi-purpose from reducing ambient air

fluctuation, and distributing the excess heat, to holding the nozzle in place. The flame is

hottest within light-blue inner cone, and gets progressively colder away from the cone.



- 30 -

The fiber is positioned off-centered (in top view) in the flame in between the inner and

the colder outer cone, Figure 6-2b. This arrangement prevents the flame’s hot upstream

from lifting the thin tapered fiber at the final stage of the pulling process. The manual

micrometer on the flame stage adjusts the flame’s vertical position relative to the fiber,

and thereby controlling the temperature and the viscosity of the softened silica during the

pulling process. The position is adjusted until the fiber softens and pulling starts. The

approximate vertical position is shown in Figure 6-2b, side view.

Figure 6-2 Flame design. (a) Gas configuration, green denotes oxygen, blue for propane, and red for

the mixture. The flame is stabilized by the pinched section of the nozzle. (b) Two views on the flame,

and the actual flame. Top view – The fiber is positioned off centered. Side view – The flame’s position

relative to the fiber can be lowered to increase the temperature at the fiber; flame gets hotter at the

top. The brass flame guard is 14 mm in diameter.

Prior to pulling, polymer coating is stripped from a 2 cm section of the fiber. The fiber is

then positioned on a pair of opposing magnetic v-grooves, with the stripped section

centered. The pulling process is choreographed by a microcontroller. (PIC16F84,

Microchip, previous version used BS2, Parallex) The microcontroller program executes

the following steps:

1. Engage the flame to the fiber, and pre-heat the fiber for 2 seconds

2. Pulls the fiber at 15.0 mm/60 s

3. Disengage the flame



- 31 -

Figure 6-3 Fiber puller microcontroller diagram for PIC16F84.

Figure 6-4 Fiber pulling apparatus. The arrows indicate the stage movements. The linear actuator is

to the right of the “flame stage” tag. Fiber is removed from the v-groove for clarity; see the next

figure for a comparison. The T-joint feeds the mixed gas to the nozzle. Stage 1 is used only when a

thinner fiber is required. Optical table has tapped holes 1˝ apart.



- 32 -

*** PicBASIC Pro program developed on PIC16F84. ***

'* Name : FiberPuller.BAS

'* Author : David Keng

'* Notice : Copyright (c) 2008 MP3L

'* : All Rights Reserved

'* Date : 4/23/2008

'* Version : 1.0

J var word

T2 var byte ‘time for motor 1

T3 var byte ‘time for motor 2

TD var byte

TRISA = 12

if porta.2 = 1 or porta.3 = 1 then

T2=0

T3=0

'M1, M2 button check

while porta.2 = 1 or porta.3 = 1

if porta.2 = 1 then

T2=T2+1

endif

if porta.3 = 1 then

T3=T3+1

endif

high portb.6

pause 300

low portb.6

pause 700

wend

'write to EEPROM

write 1, T2

write 2, T3

endif

'Starting the main program

st:

'Read EEPROM

read 1, T2

read 2, T3

'Linear actuator FORWARD

For J = 0 to 200

HIGH portb.0

LOW portb.1

LOW portb.2

LOW portb.3

pause 5

LOW portb.0

LOW portb.1

HIGH portb.2

LOW portb.3

pause 5

LOW portb.0

HIGH portb.1

LOW portb.2

LOW portb.3

pause 5

LOW portb.0

LOW portb.1

LOW portb.2

HIGH portb.3

pause 5

next J

'Linear actuator STOP

low portb.0

LOW portb.1

LOW portb.2

LOW portb.3

'pre-heating delay

pause 2000

'motor puller

high portb.4

high portb.5

if T2>T3 then

TD=T2-T3

for j=1 to T3

high portb.6

pause 300

low portb.6

pause 700

next J

low portb.4



- 33 -

for j=1 to TD

high portb.6

pause 300

low portb.6

pause 700

next J

low portb.5

endif

if T3>T2 then

TD=T3-T2

for j=1 to T2

high portb.6

pause 300

low portb.6

pause 700

Next J

low portb.5

for j=1 to TD

high portb.6

pause 300

low portb.6

pause 700

Next J

low portb.4

endif

if t3<>t2 then

ELSE

for j=1 to T2

high portb.6

pause 300

low portb.6

pause 700

next J

low portb.4

low portb.5

endif

'Linear actuator BACKWARD

For J = 0 to 200

LOW portb.0

LOW portb.1

LOW portb.2

HIGH portb.3

pause 5

LOW portb.0

HIGH portb.1

LOW portb.2

LOW portb.3

pause 5

LOW portb.0

LOW portb.1

HIGH portb.2

LOW portb.3

pause 5

HIGH portb.0

LOW portb.1

LOW portb.2

LOW portb.3

pause 5

next J

'Linear actuator STOP

low portb.0

LOW portb.1

LOW portb.2

LOW portb.3

'indicator ON program finished

high portb.6

stop

This program is used to control the 2 automatic pulling

motors with user programmable pulling time and the flame

engage/disengage linear actuator. The program read

EEPROM addr 1 and 2 to drive motor 1 and 2 for T2 or T3

seconds. To program T2 and T3, M1 and M2 switches are

press and hold for the amount of time for motor 1 and motor

2 respectively. By pressing and hold M1 or/and M2 switches

while press and release the PRG switch, the microcontroller

switches to programming mode. (PRG starts the program.)

The LED will blink every second while M1 or M2 is pressed.

By counting the desired number of blinks and release M1 and

M2, the amount of pulling time for each motor will be stored

in the EEPROM. After programming, the T2 and T3 are

stored and the microcontroller performs the programmed task

once. The program will be in “playback mode”, reading and

executing T2 and T3 values in the EEPROM if neither M1

nor M2 is pressed after PRG release which starts the

program.



- 34 -

The flame stage is controlled by a linear actuator with a built-in clutch (Haydon, bi-polar

stepper, ball bearing). A standard H-bridge transistor circuit was used in conjunction with

the microcontroller to move the actuator forward and backward for engaging and

disengaging the flame.

The pulling stage (e.g. stage 2) is linked to a DC motor with built-in gear reduction

through a rubber shaft couplers and a telescoping shaft, Figure 6-5. The DC motor is a

standard servo motor with position sensor and restriction removed for continuous turning.

The rotation speed of the DC motor is set constant by the 5V supply voltage. In addition,

low viscosity machine oil is used as lubricant in the micrometer to reduce the motor’s

load. To protect the motor and provide a soft start for the pulling process, a clutch made

from rubber tubing was designed as a part of the shaft coupler. The system pulls a fiber to

a ~3 μm diameter waist at the flame region. Additional pulling is required to produce a

thinner fiber.

Figure 6-5 Pulling mechanism for stage 2. The fiber is positioned in the v-groove and is secured by a

magnet.

In single particle experiments, the microsphere has a radius as small as 35 μm. Efficient

coupling of light into these high-curvature spheres requires that the fiber be thinned to ~2

μm diameter20. The procedure used is a combination of manual pulling (stage 1, 6 mm/60

s) followed by automatic pulling (stage 2, as described earlier).



- 35 -

The manual pulling reduced the fiber down to 40 μm. During the manual pulling steps,

stage 2 (e.g. DC motor not used) is detached from the motor. The stage 1’s manual

pulling length is adjusted to obtain a desired final fiber thickness. For example, the

procedure listed above manually pulls 6mm and then automatically pulls 15 mm,

resulting in a 2 μm diameter fiber (denoted as 6mm/2μm, 15mm automatic pulling is

common for all procedures, therefore omitted in the notation). Other fiber diameters can

also be made with good reproducibility: 7 mm/1 μm, 6 mm/2 μm, and 5 mm/3 μm. Note

that the fiber diameter tapers in an exponential manner, so the manual pulling length must

be well controlled to produce a fiber with a desired final diameter. After the pulling

procedure, the tapered fiber is mounted on a microscope glass slide.

6.4 Mounting of a tapered fiber

The mounting process uses a Cytop coated glass slide to serve as a fiber mounting

substrate. Cytop is fluorinated polymer and it is hydrophobic and chemically inert

(similar to Teflon). It is applied to the intended cutout locations for the microfluidic

channels to prevent water leakage by droplet coating. The coating agent was prepared by

diluting by a factor of 10 Cytop stock (Asahi Glass Company, CTL-107 M) with

hexafluorobenzene (HFB). A template slide and a holding device are used to indicate the

spots where the Cytop solution should be applied, Figure 6-6.

Figure 6-6 Vial containing a Cytop solution, a pipette, and a holding device with a template slide. The

substrate glass slide is mounted on top of the template slide. A 2 μL pipette deposits the Cytop

solution onto the substrate indicated by the black marker dots on the template slide.



- 36 -

A glass slide is mounted on top of the template slide in the holding device. Each

microfluidic channel design has its own unique cutout locations and therefore requires a

corresponding template slide. Only a small amount of the coating agent is required: 2 μL

can coat 4 spots, which is shown in Figure 6-6. After the pipette deposition, HFB

evaporates in ~20 s, and the remaining Cytop appears as white spots due to rough surface

induced by evaporation. The chemical coating treatment is complete at this point, and the

glass slide is ready for fiber mounting.

Figure 6-7 Mounting a fiber on a substrate glass slide with UV glue. Alignment screws s1, s2, and s3

ensure that the slide can be reproducibly positioned on the gluing platform. A drop of UV glue is

applied on each edge of the glass slide. The flame is left on to heat the platform. The warm platform

lowers the UV glue viscosity and helps the glue to flow into the polymer-slide gap. Note the double

image of the tapered fiber produced by the reflection at the glass surface.

The tapered fiber is mounted on to the substrate by UV glue (Norland, NOA81). The

alignment screws (s1~s3) ensures the substrate slide to be reproducibly positioned, Figure

6-7, while the glue is applied on the polymer coating region of the fiber at each edge of

the glass slide. A CCD video monitors the fiber alignment, Figure 6-8.



- 37 -

Figure 6-8 CCD video system for fiber alignment. (a) A microscope/CCD is mounted on a long

horizontal translation stage inspects the entire length of the fiber. (b) Images of the fiber taper image

and its reflection, and half-distance between the images is the fiber-slide distance, 125μm. (c) End of

polymer-stripped section of the fiber. This image clearly shows that the polymer is secured on the

slide surface, and the remaining silica sits parallel to the surface.

The tapered portion is secured to the substrate slide, parallel to the surface: Any contact

between the tapered fiber and the substrate would induce a significant transmission light

loss, and must be avoided. Here is an elegant method that has been invented and

implemented to deal with the problem.

In order to avoid direct contact between the tapered fiber and the substrate slide, a shim

stock is required to offset the fiber above the substrate surface. The existing polymer

coating of the fiber can serve this purpose. During the commercial manufacturing

process, an acrylic polymer coats the cladding to produce a 250 μm thick coating with

tight tolerance (<4 μm). This coating provides us a built-in shim stock. When partially

stripped fiber is mounted on the substrate, the tapered region should sit ~125 μm above

the surface offset by the acrylic coating. (The spooling stress generated during the fiber

manufacturing process causes the fiber to bend slightly.) When UV glue drops are

applied, the glue’s capillary force pulls the polymer coating toward the glass slide,

leaving a reproducible microscopic thin film of UV glue in between. The glass substrate

platform remains heated to reduce the glue’s viscosity, allowing the glue to flow along



- 38 -

the entire length of the polymer coated region. The glue is left free flowing for two

minutes, ensuring good coverage before inspection.

Under CCD inspection, fiber appears as dual images – one is the real image, and the other

is its reflection by the glass slide, Figure 6-8b. The dual images give us an accurate fiber

height caliper – the distance between the two images is twice the fiber-substrate distance.

Inspection is performed along the entire length of the tapered region, and a slight tension

can be applied by adjusting stage 2 in case the fiber does not sit parallel to the substrate.

After inspection and adjustment, a 365 nm UV lamp (Spectronics corp., 5″ tube) is then

turned on for three minutes to cure the UV glue, Figure 6-9.

Figure 6-9 UV curing to secure the fiber on the substrate.

The curing completes the mounted fiber assembly. The final assembly typically looks

identical to the un-cured version shown in Figure 6-7. The glue may turn slightly yellow

if over-cured. (This would not affect the experiment.) The overall process (fiber polymer

stripping, cleaning, fiber pulling, and gluing) takes about ten minutes, and the assembly

can be made with a micro-precision reproducibility since all the raw materials (e.g. the

glass slide and the fiber) are commercially mass-produced, and the UV gluing process

self-assembles the entire structure. This waveguide assembly can be stored for later use,

or to be mounted on temperature controlled platform for experiments. After the assembly

is mounted, a PDMS microfluidic channel caps it to form the complete WGM sensing

chamber.



- 39 -

7 Optics – Microsphere Fabrication Two methods were developed to fabricate microspheres: Flame melting and CO2 laser

melting21. In either method, the starting material is a segment of optical fiber, polymer

stripped and cleaned with ethanol to remove polymer debris. The end of the fiber

segment is heated with either a flame or CO2 laser above its softening point. The surface

tension of the melted silica forms a spheroid naturally.

For CO2 laser melting, a ZnSe lens focuses the laser light to serve as a spot heating

source. The laser time average output power can be conveniently adjusted through pulse

width modulation. This CO2 laser method produces no flame residues, and therefore is

preferred for making microspheres of size 300 μm or less. The CO2 laser’s 10 μm light

attenuates quickly in the silica. Therefore, for a large microsphere, only its front surface

is heated, whereas the backside receives only a conducted heat. The CO2 laser’s non-

uniform heating causes unwanted eccentricity in the microsphere; the flame melting is

preferred for fabricating large microspheres.

The fiber segment is positioned 2 mm inside the flame while a custom made lathe rotates

the fiber slowly so that heat is applied from all sides. This lathe technique cannot be

easily implemented for CO2 melting because of tolerance problems – A slight bend in the

fiber can displace the laser’s small ~30 μm wide focal spot. Therefore, a different

technique was used to allow uniform heating on both front and back side of the

microsphere. In what follows, both the CO2 laser melting and the flame melting

apparatuses are described.

7.1 Flame melting lathe apparatus

Most of the functions of flame melting apparatus are automated. The size of the produced

sphere is programmable with a diameter tolerance of 5 μm. The working principle is

based on preservation of material volume – The volume of the portion of the fiber pushed

into the flame is equal to the volume of the microsphere. A programmable stepper motor

drives the translation stage that pushes the fiber into the flame, and the delivered length is

controlled by the number of motor steps or revolutions. Since the volume is preserved,

the sphere size control is more precise for large spheres’ size. The sphere’s volume grows

as the cube of the radius and yet the melted fiber volume grows as the delivered length.



- 40 -

The lathe has a belt-driven fiber chuck. A modified continuously turning servo motor

drives the fiber chuck. The chuck is modified to have stainless steel tubing as a fiber

holder on the flame side to help stabilize the fiber during the lathe rotation. The pulleys in

the belt reduction are machined from Teflon so they are heat-resistant and serve as an

emergency clutch to protect the motor. The fiber chuck is spring loaded in the fiber feed

direction. The pulley and the spring-loaded chuck eliminate any backlash from the chuck

turning mechanism.

Figure 7-1 Automatic flame lathe and real time monitoring through a microscope. The yellow arrow

indicates the fiber feed direction.

A propane-oxygen mixture produces a 3 cm high flame from a twist-on stainless nozzle

that is designed to be replaceable. Unwanted oxidization of the nozzle can be observed as

an orange glow near the base of the flame, and the nozzle eventually would require

replacement. The melted silica glows in the flame, and can be viewed and sized through a

microscope against a LED backlight as the microsphere is being formed.



- 41 -

Figure 7-2 Automatic flame lathe. A belt driving mechanism rotates the fiber chuck. The stage

release solenoid mechanism is in the engaged position as shown here. When disengaged, the chuck is

recessed by one inch towards the microscope eyepiece. This allows the sphere to be reproducibly

positioned inside the flame, and released quickly at the push of a button on the controller box. The

microscope is mounted on the same stage as the chuck so that the sphere size can also be determined

after the solenoid release the stage. Inset: stepper motor controller.

The stepper motor controller is fully programmable. There are two parameters for

forming spheres: The stepper motor feeding speed and the feeding length. Both

parameters can be increased or decreased through a bi-directional toggle switch on the

controller. The stored parameter values can be checked by pressing the status button; the

number of beeps from a piezo speaker represents the feed length and the feed speed. The

speaker also serves as an alarm producing a continuous beeping when the microsphere

forming process is completed at the end of the programmed fiber feed. The forming

process can be initiated by press and holding down the start button (same as status button,

a short press reports the status while a long press starts the forming process). The lathe

motor switch and the solenoid release control are also located on the same controller.



- 42 -

Figure 7-3 Microsphere size measurement system and some sample microspheres

After the sphere is formed, the sphere size can be characterized through a microscope

metrology tool. A webcam fitted on a microscope tube images the microsphere against a

white LED backlight. When the contrast is increased in a grayscale image with high

contrast, the edge of the microsphere becomes closer to black, and the background white.

The image acquisition is controlled by a LabVIEW program. An edge detection

algorithm determines the sphere-background boundary. The boundary points are then

fitted by a circle to estimate the diameter of the microsphere with micron-precision. The

process takes less than 30 s to complete.



- 43 -

7.2 CO2 laser melting apparatus

The fabrication of small microspheres of diameter less than 200 μm requires a CO2 laser

as a point heat source to melt a tapered silica fiber. The CO2 laser is preferred to the

flame for these sphere sizes because a thinned fiber bends in the flame and creates

undesired non-symmetricity. In contrast, the CO2 laser heats the silica taper through light

absorption with little upstream of air.

Figure 7-4 CO2 laser melting setup. The CO2 laser is focused through a ZnSe lens and monitored by a

microscope/CCD system. The CO2 output can be changed by the knob on the laser controller box.

The 10 W CO2 pulse laser is powered by a 28V DC supply. The mean intensity is set to

be around 3 W by adjusting the pulse width (1/3 of a full turn of the laser controller

knob). The laser light is focused by a ZnSe lens with a 4 cm focal length. Since the 10

μm wavelength light is strongly absorbed in silica penetrating only 30 μm deep, the light

heats the formed microsphere from one side. The uneven heating causes the formed

sphere to bend toward the light source. To correct the bending, a stainless steel beam

reflector is placed within the focal point of the ZnSe lens. The reflector heats the sphere

on the other side, minimizing the uneven heating.



- 44 -

Figure 7-5 Optical setup for the CO2 laser. A reflector is placed in front of the focal point, allowing

the sphere to be heated from both sides. By translating the microsphere closer to or further away

from the reflector, the microsphere can be made to bend toward or away from the reflector. The

reflector is set at a slight angle to avoid reflection back to the laser.

Figure 7-6 Close-up view of the CO2 setup. Inset: the laser forms a microsphere. The filter glass is

placed below the microscope objective to protect the CCD. Note a glow of a silica microsphere and

the white smoke accumulated on the beam reflector. This can be easily cleaned with an ethanol wipe.

A CCD camera monitors in real time forming process. A glass filter in front of the

camera’s objective lens protects the CCD and attenuates the heat-induced glowing light.

During the forming process, a focused CO2 laser may vaporize silica, forming a white



- 45 -

smoke that is blocked by the glass filter. The glass filter can be swiveled to allow easy

mounting and dismounting of the fiber taper. As a safety precaution, an acrylic sliding

cover was designed to absorb possible scattered laser light. A concrete block is placed

behind the beam reflector as a safety redundancy.

The forming process starts by manually pulling a fiber taper down to 10 μm diameter as

the starting silica stock. The polymer coating debris should be cleaned with an ethanol

wipe before the tapering process. An additional ethanol wipe after the tapering also helps.

The taper stock is then placed onto the V-groove, and an XYZ stage is used to deliver the

taper stock into the heating region. The microsphere size can be controlled by the feeding

length. A well formed sphere produces an incandescent even glow when placed close to

the heating region. By comparison, a defect such as a trapped air bubble causes a more

intense local scattering and can be easily identified when the microsphere is glowing. The

defect occurs more frequently when the microsphere is formed too quickly with laser

power exceeding 3 W or when the taper stock is not well cleaned. The glow inspection is

a reliable real-time test for the sphere quality – A microsphere with scattering usually

cannot resonate with Q > 105.



- 46 -

8 Microfluidic system

8.1 Basic function requirement for microfluidic

The microfluidic channel (MF) is designed to handle liquids surrounding the microsphere

and the tapered fiber for chemical modification of the microsphere surface or analyte

injection. A variety of liquid handling requirements in different experiments can be

satisfied through flexible design and fabrication process of MF and a short turn-around

time. Furthermore, different units fabricated using the same design should be

reproducible. The MF must be built with chemically inert materials. In order to meet such

requirements, replica molding is the most efficient way to fabricate the MF.22 The replica

molding we adopted here has been routinely used to mold a lithographed master (often

referred to as “soft lithography.”) However, our microspheres are as large as 0.5 mm in

diameter, and the surrounding fluidic channel depth is as deep as 1 mm, and they are too

large for the etching process used in conventional photolithography. Therefore a new

technique was developed to produce a large structure: The master is made by micro-

machining.

8.2 Replica molding process

The replica molding process can be divided into two parts: fabrication of the mold

master, and molding the MF replica using the mold master. Here we will start with

description of the fabrication of the master. The present thesis involved producing more

than one hundred designs of masters in the initial MF development stage. The one

described below contains most of the design features that are necessary for virus

detection experiments. The master (Figure 8-1) is composed of 4 machined acrylic plates

(0.204˝×1.5˝ ×3.0˝) M0~M3, sequentially stacked (with M0 at the bottom) and bolted

together by four screws in the corners. In addition, a hypodermic needle is inserted to

mold one through-hole, which would be used as a buffer injection port. The through-hole

is molded to guarantee a tight seal between the microfluidic and the macrofluidic

(described in the next chapter). The stacked master creates a cavity allowing the liquid

polydimethylsiloxane (PDMS) to fill in. The filled master is then heated to cure and

solidify the PDMS forming a transparent mold. Therefore, by design, the master is a



- 47 -

negative of the final molded MF. The process of the master fabrication is designed to be

simple. The entire master fabrication takes about four hours, is performed in-house, and

is machined with micron-precision using minimum effort.

Figure 8-1 The master for MF replica molding. (a) The exploded view showing all the required

components (M0-M3) of the master. (b) The master is assembled by sequentially stacking the four

plates, M3~M0, from top to bottom. M0 contains tapped holes for the 1˝ long 8-32 screws to secure

the entire stack. The 1˝ long 21 gauge hypodermic needle is inserted into the master for molding the

through-hole.



- 48 -

Figure 8-2 CAD program and the generated G-code. The tool path are indicated by arrows, and the

G-code can be scrolled through line by line to check the accuracy. The U shaped MF channel is

designed in the middle, surrounded by a rectangular superstructure.

8.3 Mold master design

The design process starts with the M1, the plate which defines the walls and all the key

features for the fluidic channel. A computer aided design (CAD) software (BobCADv21),

as shown in Figure 8-2, is used to help design M1 and generate the tool path instructions

(G-code) for a CNC milling machine (Sherline 5400). The CNC then machines an acrylic

blank plate following the instruction thus forming M1. Figure 8-3 illustrates all the key

features designed in the CAD. The M1 consists of a U-shaped fluidic channel holding

~200μL of liquid, various cutouts for a microsphere, a fiber, and a sample inlet, a buffer

injection through-hole, and the surrounding 2˝×0.5˝ rectangular superstructure. Here are a

few guidelines on designing the M1.



- 49 -

Figure 8-3 Key design features of M1. All the cutouts are highlighted in blue, and the through-hole is

in red. The fiber cutouts are 0.02˝ high, and the microsphere entrance and sample inlet cutouts are

0.04˝ high. The superstructure’s outer edge are 2˝×0.5˝. Fluidic channel corners are rounded to

minimize the dead volume. The two fluidic walls are designed to be 0.13˝ apart, accounting for the

end mill width of 0.03˝, to have the final fluidic channel width of 0.1˝.

The master is a negative of the MF. Therefore, all the lines in the CAD design represent

the walls for the fluidic path, and not the fluid path itself. The distance betweens the two

walls defines the actual width of the microfluidic channel. The cutting tool also has a

physical width, and should be added to the intended channel width in the CAD design.

The minimum microfluidic channel width is only limited by the CNC tolerance,

vibration, backlash, and the cutting tool tolerance, but not the width of the cutting tool.

The microfluidic channel width can be built much thinner than the tool width itself if

desired, with 200 μm being the practical limit. In our fabrication, a 0.030˝ diameter end

mill is used as the cutting tool, and a channel width of 0.1˝ is used. Therefore, the design

wall distance should be 0.1˝+0.030˝=0.13˝, as indicated in Figure 8-3. The designed walls

will be in physical contact with the glass slide substrate once the PDMS mold is formed

and used. The U-shaped microfluidic channel has two 90o bends designed to damp fiber

vibration caused by fluid motion. The rounded corners at the bends are designed to

reduce flow dead volume. A superstructure surrounding the U-shaped channel is designed



- 50 -

to increase the PDMS-glass contact area, allowing better PDMS-glass adhesion. The

cutout should be sufficiently high to allow microspheres to pass through with tolerance.

The depth of the wall is determined by the end mill cutting depth, and a simple tool

calibration procedure has been developed to set the cutting depth with micron precision.

Figure 8-4 Physical CNC tool parameter defined in the CAD, material top is set to zero.

8.4 CNC zeroing and cutting procedure

The CNC itself does not have an absolute Z reference (coordinates defined in Figure 8-2).

Therefore, we need to create a reference to relate the virtual dimensions in the CAD

design with the physical dimensions used by the CNC. This latter is accomplished by

setting both the material top as Z=0 (Figure 8-4), and zeroing the end mill (described

later). With this setting, to design a MF wall with a height of 0.06˝, the tool path is

programmed to cut at Z= –0.06˝, as shown by the G-code in Figure 8-2. The rapid plane

is set to Z= +0.3˝ in the CAD (Figure 8-4), which is the end mill – acrylic plate distance

when the tool is not cutting. (This is also known as the end mill resting height.)

Therefore, by setting the physical end mill resting height to be at Z= +0.3˝, the wall depth

can be cut correctly.

The CNC end mill zeroing procedure must be performed when a new end mill is

installed, or when a new acrylic plate with a different thickness is used as a stock plate.

(Typical acrylic plate dimension is 0.204˝×1.5˝ ×3.0˝.) Because individual plates are cut

from a large stock plate, the individual plates should have an identical thickness with

tolerance less than 0.5 mil. The zeroing procedure is performed as follows:



- 51 -

Figure 8-5 CNC end mill after Z zeroing. A 0.50˝ thick aluminum stock is used to level and raise the

acrylic closer to the top of the acrylic holding chuck.

CNC Z zeroing procedure:

0. Disable the CNC stepper motor control switch

1. Lower the end mill manually in Z direction

2. Stop at an arbitrary integer in Z that does not allow an end mill–acrylic contact

3. This should close the end mill–acrylic gap to be less than 50 mils

4. Set loose the end mill set screw, allowing the end mill to contact the acrylic top

5. Tighten the set screw after contact

6. The material top is now zeroed

7. Enable the CNC stepper motor control switch and set Z=0 in the CNC control

8. CNC controller drives the Z to 0.3˝

9. End of the zeroing procedure

The zeroing steps 0–6 allow the Z to only be lowered, which minimizes the error due to

backlash in the Z direction. The G-code produced in the CAD program can now be



- 52 -

transferred to the CNC controller and be executed. While cutting, water should be added

on the acrylic plate to cool and lubricate the end mill and bring away the chips. The end

mill feed speed and plunge speed is programmed in the CAD to be 0.1˝/min (Figure 8-4).

This speed determines the wall roughness and the amount of total cutting time required:

The higher the speed, the larger the roughness, and the shorter the time to cut. The

0.1˝/min used is a good compromise.

Figure 8-6 CNC control software run under Linux operating system (Shirline CNC (inch)). Here a

slightly different microfluidic allows two spheres to be coupled to one tapered fiber; note the

additional cutout. The end mill currently is located at the resting position right now, X=0, Y=0, and

Z=0.3 (in inches). The A Axis is the rotation option that is not being used. The G-code file is loaded

through AUTO/Open, and is then executed line by line. The line being executed is highlighted in red.

This figure is showing the last line being executed. The backplot view allows the user to monitor the

end mill position in real time. In addition to the microfluidic G-code, hole.txt is also executed,

producing 4 pilot holes in the corners.



- 53 -

Figure 8-7 Cutting in progress. Water is added as lubricant and coolant to prevent the end mill from

melting the acrylic.

Figure 8-8 Cutting completed. The end mill moves back to the resting position, z=0.3˝.

In what follows, the fabrication procedure for each plate (e.g. M1, M0, M2, M3) will be

briefly described.



- 54 -

Figure 8-9 Close-up view of M1 after machining.

After a blank plate is loaded into the CNC with the Z zeroing procedure, the G-code for

M1 generated by the CAD is loaded in to the CNC controller. The cutting progress can be

monitored through the Linux-based CNC controller software. After the G-code is

executed and the cutting is completed, another stand-alone G-code (hole.txt) is executed

to produce four pilot holes for screws. The M1 is then removed from the CNC, and the 32

mil fluidic through-holes and the 8-32 screw clearance holes are drilled with a drill press

using the pilot holes generated. The M1 is now complete.

Figure 8-10 Close-up view of M0 after machining.



- 55 -

M0 is also produced with hole.txt on the CNC followed by drilling and tapping 8-32

holes, in the same way as M1 was fabricated. The M0 is only used as a supporting plate

for the screws and does not have fluidic features. The M2 and M3 also follow the same

treatment to have their 8-32 clearance holes made.

Figure 8-11 Close-up view of M2 after machining. M2 has a rectangular cavity cutout.

Figure 8-12 Close-up view of M3 after machining. The injection holes are two-step drilled to fit a

Luer tip syringes. The drain hole is positioned at one corner of the PDMS cavity to collect and push

out trapped bubbles.

After the screw holes are produced, the M2 is cut with a manual milling machine to

produce a large cavity cutout in the middle of the plate. The cavity dimension is made to



- 56 -

be larger than the superstructure designed in the M1. The M2 cavity cutout must cover all

the features on the M1, allowing a proper fill for PDMS molding.

Lastly, M3 is fabricated. To produce liquid through-holes in the M3 – M1, M3, and M0

are sequentially stacked and bolted. The already drilled liquid through-hole on the M1

serves as a drilling template for the M3. To produce PDMS pouring and draining holes,

#22 and #32 drill bits are used. The #22 hole provides a matching fit for the standard

Luer syringe, while the #32 gives a smaller diameter, so that the PDMS can be broken off

at this place after curing. The finished master is then cleaned with detergent and allowed

to dry completely before PDMS pouring.

Figure 8-13 Bubble removing procedure once PDMS is filled: First the bolted master sits flat (left) so

the bubble would move to the top. Then, tilting the master to push the bubble out of the drain hole.

8.5 Molding process

PDMS (Sylgard 184) and its catalyst are mixed at 10:1 mixing ratio (20 mL/2 mL) in a

60 mL syringe. A 24mL mixture is usually sufficient to fill the master three times with

excess. After de-gasing under a weak vacuum in the syringe, the PDMS is injected into

the master. The master is left to sit flat for five minutes and then tilted for another 5

minutes to allow the trapped bubble to collect at one corner and then be pushed out of the

drain hole, as shown in Figure 8-13. After the bubble removal, the filled master is cured

in an oven for 1 hour at 70oC. The PDMS can also be cure at room temperature as well,

allowing 3 hours of working time after mixing.

After thermal curing, the cured mold is removed from the master, and the master is left to

cool before cleaning. The molded MF is shown in Figure 9-1. The cured mold is stored in



- 57 -

a closed container with fluidic channels facing down to avoid dust. The leftover PDMS

debris on the master can be easily cleaned with an ethanol wipe. However, performing

ethanol cleaning before the master completely cools will cause a high thermal stress on

the acrylic and crack the master, and therefore must be avoided.

8.6 Overview and cost analysis

Overall, the master fabrication takes 3–4 hours, and the material cost is less than $1.

(Price is for the four plates of acrylic, four screws, and one syringe needle). The mold

takes one hour to cure, and material cost is about eight cents. (Price is for the 8 mL of

PDMS). For equipments, the CNC cost $2500, and the CAD is $500. The master can be

reused for about 100+ times if it is cleaned properly (e.g. cooling before cleaning). In

addition, the PDMS mold can also be reused by ultrasound cleaning in H2O and ethanol

mixture (1:1) for 10 minutes if necessary. However, for precision measurement (e.g. virus

detection), a new PDMS mold is used per experiment to avoid cross contamination.



- 58 -

9 Macrofluidic system

9.1 Basic function of the macrofluidic system

The design of the fluidic control and the interface to the microfluidic, often referred to as

a macrofluidic system, is critical to the data quality of the WGM biosensor. More

specifically, a good design should minimize the tapered fiber’s mechanical vibration

caused by the fluid motion to reduce the overall WGM system noise. In addition, the

system should be designed to be versatile, precise, easy to operate and user friendly.

Figure 9-1 illustrates the macrofluidic to microfluidic connections and interfaces. No

adhesives or sealants are used in any connection, which is made possible by the precision

replica molding of the PDMS microfluidic. This chapter will describe the design

principle for the macrofluidic system used in our WGM virus biosensing experiment.

Figure 9-1 Macrofluidic and Microfluidic interface. A buffer tubing is connected to the PDMS

microfluidic with a stainless steel hypodermic needle that has a gauge identical to the one that was

used to mold the through-hole. The sample is injected using a removable 25 gauge needle held down

by a hairclip. The sample inlet can slide in and out through the PDMS cutout. The drain is connected

to a weak vacuum. All connections are press fit, and no glues or adhesives are used.



- 59 -

Since fluid pulsation leads to vibrations in the tapered fiber, diaphragm pumps and

peristaltic pumps should be avoided. Instead, the macrofluidic system employs two

micro-stepping syringe pumps, one dedicated for buffer delivering, and the other one

dedicated for sample delivering. The pump outputs are always air damped to reduce the

pulsation caused by the micro-stepping. A typical system is shown in Figure 9-2, and also

in Figure 9-3.

Figure 9-2 Macrofluidic system. The valve/plunger diagram for sample loop loading and injection is

also shown. A 0.22 μm filters is used on the buffer line.

9.2 Pump and tubing selection

The pumps used (Kloehn, #50300) have a suitable delivering resolution of ~1 μL/step.

The step resolution depends on volume of the syringe used. In our system, the plunger

mechanics has 24000 total available steps for its full travel, and the syringe used has 25

mL, yield 1.04μL per step. The polycarbonate valve v1 allows closed-loop buffer

replenishing of the pump1 syringe without exposure to the environment. (Valve v2 on the

sample pump syringe also serves a similar purpose.) All the fluid paths are made with

semi-transparent thin wall Teflon tubing (Smallparts, gauge 22), which is connected to

pumps and valves through 21-gauge blunt-tipped hypodermic needles. A buffer Teflon

tubing is connected to the PDMS microfluidics channel via a 21 gauge needle, and the

sample line is inserted into the channel via a 21gauge/25 gauge concentric needles. The



- 60 -

25 gauge needle is small enough to fit the designed cutout, and also increases the back

pressure (described later).

Table 9-1 Guide for tubing and needle diameter. A 21G needle fits tightly with a 25G needles. The

tolerance is ~0.5 mil.

gauge ID (mils) OD (mils) Function

22 tubing 30 30.5 PTFE, expandable to fit a 21G needle

21 needle 20 32 Used to mold the PDMS through-hole

25 needle 10 20 Can also be used as a bare fiber holder

For buffer delivering, the pump output is connected to a bubble trap which also serves as

a fluid pulsation damper. The fluid delivered into the bubble trap builds a slightly

positive air pressure inside the trap which pushes the liquid out of the hermetically sealed

trap. The trapped air is compressible and therefore regulates the output fluid pressure. A

valve, v1, is used to connect the replenishing buffer reservoir to the pump so that the

fluidic path is fully enclosed. In normal buffer delivering, the plunger travels only

upwards and therefore there is no backlash on the plunger mechanics.

Figure 9-3 Macrofluidic setup with a laptop computer control. The blue T-valves are fitted at the

pump output. The T-valve handle direction indicates the stop branch. Here, the sample loading

syringe and buffer refilling are stopped. The data logging screen showing a dip in the fiber

transmission spectrum is to the right edge of the figure.



- 61 -

Sample delivery requires that the sample to be drawn into the sample loop first. To avoid

a change in the direction of the pump plunger and to guarantee accurate fluid delivering,

the sample pump has a more complicated configuration with an additional manual sample

loading syringe. The sample is loaded by a manual syringe according to the v2 and v3

setting as shown in Figure 9-2. The sample resides only within the sample loop, and the

buffer inside the syringe pump pushes out the sample. In addition, an air gap is purposely

loaded between the sample and the buffer during the sample loading. The air gap

prevents sample/buffer mixing and also dampens the pulsation. The concentric needles

used at the sample loop increase the sample back pressure, compressing the air gap to

dampen the liquid pulsation. To clean the sample loop, a buffer can be introduced from

the buffer reservoir into the manual syringe through an appropriate valve configuration,

and then manually dispensed to clean the loop.

Figure 9-4 pump control program. The plunger direction control (the three position control adjacent

to the syringe indicator) can be used to drive the plunger up, neutral, or down. The number of steps

delivered, the buffer and sample volume, and various other control parameters are also displayed.



- 62 -

9.3 Pump control software and hardware

The pump is controlled through a custom built program with a graphical user interface,

see Figure 9-4. The syringe pump was originally designed to be controlled through now

obsolete RS485 communication with command line-style user interface. It was difficult

to use effectively. Therefore, the new custom program is made with LabVIEW that

commands the pump directly by controlling the TTL inputs on the stepper motor driving

chip. The direct TTL control is done as follows:

The original microcontroller board is removed from the pump, and all the header pins on

the motor drive board are identified. Each stepper motor can be controlled by a set of

three pins: Enable (EN), direction (DIR) , and step command (CLK). Our system does

not use the pump’s valve motor, and only the plunger motor is used. The LabVIEW

program generates the TTL signals through the computer’s parallel port.

The LabVIEW’s front panel allows the user to control the number of steps to be delivered

as well as the period of each step (Figure 9-4, dT). The EN and DIR signals can be

controlled on the front panel as the plunger direction control. The instrumentation panel

displays the amount of liquid remaining in each syringe. The program allows pre-loading,

the next command can be programmed before the current command is completed,

allowing continuous flow delivery.

To summarize, the macrofluidic system was made with minimum parts and designed to

keep the liquid path fully enclosed to avoid contamination. All the sample-containing

parts used (e.g. Teflon tubing and stainless steel hypodermic needle) are chemically inert

and disposable. Although the system could be made to be fully automatic and

programmable, the semi-automatic macrofluidic design described here was sufficient for

the virus detection experiments.



- 63 -

Figure 9-5 Pump drive mechanism and header pin definitions. Although not being used in the

program design, HM is the optical chopper signal output, which could be used to indicate the home

position of the plunger and the valve.



- 64 -

10 LabVIEW program – dip tracking

10.1 Basic function of the program

The dip tracking program is used to track the minimum position of a resonant dip in the

fiber transmission spectrum. The accuracy and robustness of the tracking algorithm

determine the quality of the experimental data. Since the invention of the tracking

technique, the algorithm has been modified extensively to reduce noise and improve the

sampling rate. The algorithm should be able to locate the dip minimum with resolution

better than the sampling resolution, and should also track the time-dependent shift of the

resonance dip. The core of the algorithm is a parabolic fitting routine which can locate

the minimum with a resolution better than 1/50 of the dip’s linewidth. A state-machine

logic is then used to follow the dip position in each spectrum over time. Below is a

description of the hardware and software configurations for this program. This chapter

should serve as a detailed LabVIEW reference.

Figure 10-1 The three layer flow chart for the dip trace program.

10.2 Program structure

The program can be divided into three layers: Data sampling, core tracking algorithm,

and the added utility, as shown in Figure 10-1. Figure 10-2 shows the back panel of the

LabVIEW program, main.vi. This program has been simplified to contain all the



- 65 -

necessary ingredients to carry out a virus detection experiment. In practice, this main.vi is

often modified for a specific experiment, but the concept does not change. Here, the laser

is driven by a function generator’s 10 Hz positive sloped sawtooth ramp. For each ramp

produced, the transmission intensity is monitored by a power meter. Both the laser

driving ramp and the transmission intensity are simultaneously sampled through

individual analog input channels of the DAQ.

Each section of the main.vi has been commented and given a label of Lxx. The first

number represents the approximate layer each section belongs to in the flowchart; for

example: L2x indicates a section is in the 2nd layer, the core algorithm. A detailed

explanation will be given below by first discussing the functionally related sections as an

entity, followed by commenting on individual sections. Here we will start with the first

layer, L1x, the data sampling.

10.3 Software/Hardware interface

The data sampling layer is responsible for sampling the voltage at the DAQ input

channels (L1). The voltage signals are translated into corresponding physical quantities

(L2a, L2g). For example, to sample the optical power measured by the detector (L2a), a

transfer function (either provided by the manufacture of the detector, or obtained through

separate calibration measurement) in the LabVIEW converts the power-meter generated

voltage into the power measured. The transferring of signal to physical measurement can

be checked. Usually, a measurement instrument has a dedicated front panel display. The

translated signal obtained in LabVIEW must agree with the instrument front panel

display within reasonable tolerance. A similar sampling and conversion routine is also

performed to obtain the DFB drive current (L2g). However, for DFB laser temperature

monitoring, the sampling route is through GPIB digital communication generated by the

laser driver box itself (L3b). Therefore, the sampling is exact, and no transfer function is

required. After all the sampling voltage is converted into physically meaningful numbers,

the converted data is then ready for the next layer of the program, the core algorithm for

dip tracking.



- 66 -

Figure 10-2 The main LabVIEW program (main.vi) used for dip trace. Each section has its

functional description labeled and commented. These labels will be referred to throughout the

chapter. Various parts of this program will be zoomed in and described in details in the rest of the

chapter.



- 67 -

Figure 10-3 L1, configured to be triggered by the falling edge of the TTL

The L1 is triggered by the function generator’s sync TTL signal, and more specifically,

by the falling edge of the signal. To sample the spectrum driven by the 10 Hz ramp, the

ADC sampling must have matching parameters. The channels 0, 1, and 2 (ch0, ch1, ch2

for short) are programmed to have an input range of -5V to +5V, with a 16 bit resolution.

The ch0 is connected to the power detector; ch1 monitors the laser ramp; and ch2 is left

as a spare for an additional power detector. The connection and their expected signals are

illustrated in Figure 10-4. The sampling rate is set on the main.vi front panel to be 10

kS/s (10 kSamples/sec), and with 1 kS/scan to match the 10 Hz ramp. That means that

there are 1000 sample points for every ramp scan (10 kS/s / 1 kS/scan = 1 kS/ramp). The

L1 takes 100 ms to execute and then relays the gathered data to L2a and L2g. Note here

that L1 alone takes 100 ms to execute and the rest of the program also takes time.

Therefore, each sampling takes 200ms: The data sampling is performed on the every

other ramp, and the rest of the algorithm operation is performed on every intervening

ramp. In effect, spectrum is collected at 5 Hz. The 100 ms idling time allows the dip trace

algorithm to be carried out.



- 68 -

Figure 10-4 Input channels (blue) and the operation of main.vi. The spectrum (having 2 dips) is

sampled for every other ramp separated by an idling time [symbol …]. The dip tracking algorithm is

performed during the sample idling. Both edges of the display spectrum is truncated (shaded) in the

analysis to avoid complications from the transient.

After the ch1 and ch2 are sampled, they are sent to L2a and L2g. The first step in these

sections is to exclude the ramp transient; only the middle 800 samples are used.

The conversion function for the power detection L2a was obtained previously for the

particular wavelength used, since the detector photo efficiency depends on the

wavelength. The average of the detected voltage is computed and displayed to allow the

user to check if there are sufficient quantization bits covering the sampled voltage. The

op-amp used within the power meter has a null offset in high gain settings. Therefore, an

intensity NULL at laser powered off should be measured and compensated to yield the

correct power.



- 69 -

Figure 10-5 L2a and L2g section for converting the voltage data into power and laser current

through their respective transfer functions.

To monitor the laser drive current in L2g, the laser driver’s transimpedance gain (in

mA/V) is measured (there is about 0.6% discrepancy between the nominal 10 mA/v and

the measured 10.0585 mA/v) for the particular laser driver used. The data of the laser

current in mA vs. voltage are fitted by a linear regression to obtain the slope (m), and the

intercept (b).



- 70 -

10.4 Trigger timing and noise reduction

The linear fit is crucial in reducing the system noise: Ideally, the DAQ sampling is

triggered when a falling edge in the TTL signal is detected. However, because Microsoft

Windows is not a real-time operating system, there is always an arbitrary delay between

the time when chTRIG triggers and the time when ch0 and ch1 start the sampling

process, as shown in Figure 10-6. As a result, the spectrum is time-shifted and gives an

error in the dip location. The delay is on the order of 1 ms, or 1/100 of the sampling

frame. For a 0.2 nm wavelength scan, 1% is significant (2pm), and must be compensated.

The latter is accomplished by simultaneously measuring the laser driving ramp, since the

ramp would be sampled with an identical delay. The fitted ramp slope (m) and intercept

(b) obtained form a linear fit of the ramp are used in the algorithm to compensate for the

error.

Figure 10-6 The delayed start on the triggering frame. The TTL triggering delay and the resulting

error has a 1:1 relationship. This figure is an exaggeration, and the actual delay is on the order of

1ms or so (1% of the frame).



- 71 -

10.5 Dip detection and wavelength conversion

We shift our attention back to the fiber transmission signal and explain the dip fitting

routine. The routine is illustrated in Figure 10-7. First, the transmitted power signal

passes through a peak/valleys detector (L2b), which detects peaks and dips by fitting

routine user selected number of points of a parabola. This fitting routine locates and fits

all of local maxima/minima that are above/below a threshold level. (The transmission

spectrum is typically recorded as dips, but the peak option is made available to the user.)

These peaks/dips are represented by their raw sampling index. Note that the index is not

limited to an integer, allowing 1/50 linewidth resolution with only 20 points spanning the

dip.

Figure 10-7 Dip fitting routine and conversion of the dip location from index (L2b) to current (L2c) ,

then to wavelength (L2d).

Next, the fitted points are converted to laser current using m and b, and then converted to

the wavelength. The wavelength conversion requires computing λ(T, i), which is based

on the laser calibration coefficients. The laser temperature is measured by the laser driver

and it is obtained through GPIB communication (L3c). In what follows, one of these dips

will be tracked to form a dip trace.



- 72 -

10.6 Dip tracking core algorithm

Figure 10-8 Single dip lock and track (L2f). User can select one of the dips from a spectrum.

The dip selection (LOCK) routine (L2f), as shown in Figure 10-8, is used to select one

dip (Dselect) out of the all the dips fitted (Dall, bolded to represent an array). The process

begins with the user selecting a dip from the spectrum. The selected dip is then used as

the starting point for the following state-machine logic procedure for tracking, see Figure

10-9.

Dip tracing procedure

0. A user selected dip (or from shift register, explained later) as Dselect

1. Find the dip closest to Dselect in the current spectrum

2. Designate the closest dip as the new Dselect′

3. Display the new Dselect′ and send the new Dselect′ to the spectrum obtained in the next

scan

4. Process complete, repeat 0-3



- 73 -

Figure 10-9 Dip tracing procedure illustrated with Dselect in red, and the other dips in blue. Step 0

shows the preceding spectrum, and step 1, 2, and 3 display the current spectrum. A new Dselect’ is

assigned in steps 2,3. All the dips in the spectrum are located in step 1, but only the dip closest to

Dselect is designated as Dselect′.

The algorithm for the dip tracing procedure is written as follows:

select selectD ' =min( -D )allD Equation 10-1

select selectD =D ' Equation 10-2

The LOCK subroutine in Figure 10-8 uses the above algorithm. The dip location unit is

specified by wavelength. The dip being tracked, Dselect, is also represented in terms of

wavelength and it is derived through all measurable parameters (e.g. laser current m, b,

and laser temperature), and therefore is absolute and can be compared from one spectrum

against the next spectrum. Equation 10-2 is implemented in LabVIEW as a shift register,

L2l and L2m, in which the preceding Dselect is transferred to the new spectrum. The fitted



- 74 -

points (Dall) as well as the locked point (Dselect) are displayed in the spectrum on the

program’s front panel.

The fiber transmission spectrum is displayed as an XY plot, with X = λ(T, i), and Y =

detected power. All of the Dall and Dselect are superimposed on to the plot. The real time

dip trace is shown as a function of program running index in place of the time to reduce

the program complexity. However, when the dip trace is saved as a two column format

text file by the user, the first column is the real time and the second column is the dip

wavelength (L3g, real time obtained through L2m). There are also other utility features in

the program.

10.7 Laser temperature control and interface

An important utility is the laser temperature control and monitoring section (L3b). The

L3b sends a GPIB command to the laser driver to control the laser temperature and reads

the temperature immediately afterwards. The laser temperature can be set with 0.1oK

resolution and be read on the laser driver’s front panel with a precision of 0.01oK;

however, the proportional feedback loop used in the laser driver is capable of 0.0005oC

short term (2 hours) stability. Therefore, the L3b is set to run once first for temperature

change and the user should manually run L3b again to monitor the temperature after the

laser temperature is stabilized (usually 10 s later). The transient in the wavelength created

by the laser temperature change is usually insignificant (< 50 fm), as long as the laser

temperature is slowly changed by 0.1oC at a time.

10.8 Run time utility – Log keeper

Another useful utility function is the log keeper. The log keeper allows the users to

document their experiments with an automatic time stamp. Figure 10-10 illustrates an

example of a dip trace obtained from a virus detection experiment. All the chemical

preparation steps are documented with corresponding time stamps.



- 75 -

Figure 10-10 An example of the dip trace and log keeper, taken from a virus detection experiment.

10.9 Changing laser with matching calibration file

Another utility is L3a, which loads the laser calibration coefficients. Although a single

DFB is usually used throughout a series of experiments, this utility allows the laser to be

swapped and the corresponding calibration coefficients be loaded with ease. The

calibration loading procedure immediately allows a new laser to be used once executed.

The data saving routines (L3d, L3e, L3f, and L3g) save the data. The user can

save all the data from one experiment in a folder with a user specified name (L3d). A

spectrum is automatically saved every time a new line is added to a log keeper or be

saved by user demand (L3f). Each saved spectrum is given a filename carrying a time

stamp to the nearest second. For example, spectrum_1543.txt means the spectrum is

saved at time 1543 s.

10.10 Overview and precision

Overall, the program is capable of tracking a dip for over 12 hours, with precision of 1 fm

(depending on the Q of the resonance dip), and a tracking range limited only by the laser.



- 76 -

11 Pump probe coupling and Compact Microfluidic

11.1 Continuous fiber coupling method

A typical configuration for microsphere WGM excitation is tapered fiber coupling, see

Figure 11-1.17 A tapered fiber (green) created by flame tapering or HF acid etching

evanescently couples the guided light into and out of the microsphere, and the

transmission intensity is measured. This continuous fiber coupling configuration has been

used extensively as a scientific discovery tool for biosensing and achieved a great

success. However, there are many engineering limitations for using it in some biosensing

applications because of a tapered fiber and the strict requirement for the refractive index

matching to the microsphere. The Fiber Effective Length (FEL) is defined as the length

of the thin fiber that has its evanescent field exposed available for coupling. Although the

tapered fiber produced previously has an effective length of only 10 mm, this length is

still too long for many applications that involve miniaturization of microfluidic channels,

multiple-fiber multiplexing, and drop channel implementation.

Figure 11-1 : Continuous fiber coupling, side and top views. Arrows indicate light propagation

direction. Solid line: strong coupling; dashed line: weak coupling. (Figures not to scale)



- 77 -

11.2 Pump probe coupling method

A different coupling technique called “pump-probe” has been developed to address these

limitations, see Figure 11-2. The name derives from the design that uses two fast tapers

which function as near field probes. The fast tapers are co-aligned facing each other on a

planar substrate to excite (pump) and pick up (probe) the resonating light in the

microsphere. Each taper has an EFL of 500 μm or less. In this configuration, the pump

fiber (blue) excites the resonance, while the probe fiber (red) picks up a small fraction of

the resonating light for resonance detection. Therefore, the output of the probe fiber is

similar to a “drop channel” in an add-drop filter. A 50 μm gap between the two fibers

prevents the light from transmitting directly.

Figure 11-2 : Pump-probe configuration with two fast tapered fibers, side view

This pump and probe technique can be used on other WGM cavity geometries as well.

Other alignment configurations are also possible, see Figure 11-3.

Figure 11-3 : Other possible pump and probe coupling configuration

The main advantage of the pump-probe configuration is its compactness, robustness, and

ease of construction. The FEL of the taper can be as short as a few hundred μm, allowing



- 78 -

the entire structure shown in Figure 11-2 to be packaged in a flowcell of just a few μL.

The short taper serves as a natural spring and generates frictional force when coupled to a

microsphere. This spring-loaded system exerts a sufficient frictional force to resist

mechanical vibration, and the flow drag in the microfluidic. As was the case with

continuous fiber coupling configuration, this pump-probe configuration can be easily

assembled on a planar glass substrate. There is no need to worry about breakage of a long

continuous piece of a fiber since the pump and probe fibers are already taper-terminated.

Furthermore, the probe fiber signal provides a zero background drop channel spectrum

that continuous fiber coupling configuration cannot offer.

11.3 Drop channel on the planar surface

The traditional add-drop configuration is achieved using two fibers. The latter design is

not practical for device implementation, because two fibers have too many degrees of

freedom, and require multiple translation stages for alignment. Such a requirement

compromises reliable coupling and makes robust miniaturization impracticable. The new

pump-probe configuration has a fewer degrees of freedom because both tapers sit on a

planar substrate: the two tapers are automatically placed at the same level. For system

assembling, the only observation tool required is a microscope. This reduction in the

degree of freedom facilitates the manufacturing process and make it more reproducible

compared with the conventional configuration. This pump-probe planar waveguide

design may also be implemented by standard photolithography technique for mass

production.

11.4 High Q – tunable coupling

In general, the tapered fiber itself adds coupling loss, thus lowering the Q. To minimize

the loss, the coupling between the microsphere and the fiber can be weakened by leaving

a small gap in between. However, for practical biosensors, the tapered fiber needs to be in

physical contact with the microsphere at all times to eliminate the gap variation caused by

mechanical vibration. With this physical contact constraint, the only possible way to raise

the Q is by off-equator coupling, as shown by the dashed line in Figure 11-1.



- 79 -

The new pump-probe configuration, in contrast, has more degrees of freedom for tuning

the coupling parameter. The positions of the pump and probe fiber, and the coupling

length can be adjusted to maximize the Q. We can use numerical calculations to find the

optimal coupling parameters if necessary.

11.5 Zero Background

For a practical biosensor, or any other sensor system, reducing background noise and

interference is crucial for sensitivity. In the continuous fiber coupling configuration, the

transmission spectrum is the only sensor signal, from which we can extract the shift in the

resonance wavelength. However, the ramping of the laser current causes a ramp in the

laser output power, and a ramp in the fiber transmission spectrum is a sloped background

that cannot be easily removed by signal processing. In principle, it is possible to

normalize the transmitted intensity by pre-recording the background ramp, but it would

be better if the background can be eliminated altogether.

Ideally, the probe fiber should not receive any light directly from the pump fiber, and

only receive the light from the WGM orbit in the sphere. One of the possible layouts is

illustrated in Figure 11-4.

Figure 11-4 : Field of the excited microsphere, bottom view, not to scale. Notice the divergent light

exiting the pump fiber.

In Figure 11-4, the pump fiber (blue) excites the resonance in the sphere (thick blurred

red band). The tapered terminal launches a far field pattern as indicated. The probe fiber

(red) is positioned off the WGM orbit to minimize the perturbation on the WGM. If the

probe fiber were in-line with the pump fiber, then the microsphere would bridge the gap,

causing the transmission ramp to be visible in the probe fiber signal.



- 80 -

A zero background, peak-only spectrum can facilitate many interesting measurements:

locating each resonance with a threshold filter, direct Q measurement, TE/TM

discrimination, and etc.

Resonance peaks can be very easily isolated by a simple threshold filter. The Q

(or linewidth) measurement can also be easily performed through parabolic fit, since

there is no background ramp involved. Transmission peaks associated with TE/TM

polarization can be easily distinguished with zero background.

11.6 Miniaturization of the microfluidic

The short taper of the pump probe configuration allows a sample volume as little as 1 μL.

11.7 Construction

Tapered fibers can be produced with two methods: flame pulling and HF etching. The

flame pulled fiber has a tapering length of about 3 mm. This long length makes the

tapered fiber extremely flexible and therefore cannot be used. The HF etching process

was developed to fabricate reproducible fiber tapers with micron scale precision.

11.8 Taper fiber manufacturing – HF method

The HF process uses multi-step etching to produce tapered fibers in 40 minutes. Figure

11-5 illustrates major steps involved. The original HF tapering was developed for near-

field probe spectroscopy.23, 24



- 81 -

Figure 11-5 : HF etching process to manufacture a taper. Top green layer is silicon oil, and the

bottom blue layer is HF solution.

In Figure 11-5, step (a) mechanically strips off the fiber acrylic jacket from the fiber end.

The strip length is 10 mm. In step (b), the fiber is lowered in a polystyrene cuvette (1 cm2

cross section) containing 1.5 mm high 48% (w/w) aqueous HF (blue) covered with 2 mm

thick of silicone oil (green).24 The silicone oil prevents vaporization of HF, and limits the

etching to occur only below the liquid interface. A meniscus at the liquid interface,

Figure 11-6 panel b1, automatically etches the fiber into a taper as shown in Figure 11-6

(b2). The meniscus diminishes with a decreasing fiber diameter, and thus creating a linear

taper. The fiber diameter is monitored with a microscope. The etching process is stopped

when diameter decreases to 4 μm. This step usually takes 28 minutes. The etching

process is stopped by lifting the fiber taper from the cuvette, and rinsing with methanol.

The fiber diameter is then measured under a microscope. Additional etching is performed

if the fiber diameter is greater than 4 μm. Once the desired fiber diameter is obtained, the

fiber is again lifted off the cuvette and be rinsed with methanol. Then the fiber is trimmed

to length by lifting the etched cone portion 500 μm above the interface, step (c1). This

500 μm section will be protected by the silicone oil, and the remaining fiber in the HF

solution will be completely etched away, as shown in (c2). After step (c2), the tapered

fiber is removed from the cuvette and rinsed with methanol to remove the residue silicone



- 82 -

oil and HF. The entire etch process takes about 40 minutes, and the HF solution can be

reused up to ten times.

Figure 11-6 : The meniscus at the HF (bottom) and silicone oil (top) interface. A taper is being

formed above the interface.

Figure 11-7 : Top: Close-up view of the taper region, 8 μm per division. Below: Fiber profile and

light leakage test, 15 μm per division. The effective region of the fiber is evident by a scattering of a

red laser. Both photos were taken in air.



- 83 -

The typical tapered fiber has a profile as shown in Figure 11-7 and Figure 11-9. Notice

the linear taper has a tapering angle of 11.5o, resulting in a 380 μm tapering length. The

EFL is ~550 μm.

After the fiber is etched, a simple light leakage test is conducted by coupling a

656 nm laser into the fiber. As shown in Figure 11-7, the light is only visible where the

fiber diameter is smaller then the fiber core diameter, ~8 μm. Close-up view reveals a

continuous and smooth transition between the taper region and the core-exposed effective

region of the fiber, suggesting a low tapering loss. This agrees with the fact that there is

no light diffraction observed at the transition region. The diameter of the effective region

decreases from ~7 μm at the transition region (corresponds to of 8.7~8.4 in the scale in

Figure 11-7) slowly down to ~3 μm at the fiber tip. The small diameter explains the

additional light leakage due to dust and surface roughness. The slow tapering along the

effective length may be a result of the variable etching rate caused by HF concentration

gradient near the interface.

Note that for a flame tapered fiber, the core is stretched thin and the light is primarily

guided in the cladding modes. With the etched taper, the reduction of the cladding does

not change the core guiding of the light.

11.9 Microfluidic waveguide chip design: Introduction

For a microfluidic WGM sensor system, one of the challenges is to minimize the overall

fluidic volume. Since the previous design uses a continuous fiber to couple light, the

entire taper must be immersed in the same medium (usually water) as the one that

surrounds the microsphere to avoid refractive index mismatch. This length has been

about 10 mm for both the HF etched fiber and the flamed pulled fiber, which translates to

20 μL as the minimum volume for a microfluidic channel.

Having the short taper enables a new approach to the design of the microfluidic,

since the etched probe can be made to be 500 μm or shorter. Figure 11-8 is one of many

coupling configurations that can be made and integrated into a fully functioning

microfluidic system with a volume of just a few μL.



- 84 -

Figure 11-8 : Dual pump-probe coupling configuration with two microspheres

In this dual pump-probe (DPP) coupling configuration, a pair of prefabricated fiber tapers

pumps and probes two microspheres. The pre-fabricated fiber tapers are etched with the

same characteristics and they are pre-aligned, thus making the DPP assembly easy.

11.10 Prefabricated fiber tapers - Multi-fiber parallel etching

Figure 11-9 : Etched fibers. Two fibers are etched simultaneously. The taper profiles are similar.



- 85 -

The fibers shown in Figure 11-9 is etched in a unique parallel configuration, so that

multiple fibers can be etched together, as shown in Figure 11-10.

Figure 11-10 : Two fibers glued in a parallel configuration. Notice two spacer fibers with jacket in

the middle.

In this configuration, two fibers can be etched in close proximity, thus producing a

matched taper profile and effective length with less than 10 μm tolerance in all

dimensions. The parallel configuration can be achieved by just self assembling. No micro

positioning tool is required.

Four fibers, two jacket-stripped fibers and two jacketed spacer fibers, are bonded

together using UV adhesive on a substrate. A layer of scotch tape is applied on a glass

slide and serves as a non-sticky substrate for the curing. The UV adhesive can be peeled

off easily from the scotch tape after it is cured. Two stripped fibers (~30 cm long, 10 mm

stripped), and two fully jacketed spacer fiber (~3 cm long) are placed on the substrate,

and a droplet (~5 μL) of UV glue (Norland, NOA81) is applied. The UV glue produces a

capillary force and aligns the fibers against the substrate and against each other. The UV

adhesive is then cured under a UV lamp (365 nm) for 30 s. After curing, while still on the

substrate, additional droplets of UV glue are applied at the two ends of the spacer fibers

to strengthen the assembly, and are then cured for 60 seconds. The assembly is then

peeled off from the substrate, and ready to be etched.

The two spacer fibers used to maintain the distance between the two stripped

fibers produce a ~750 μm gap (core to core distance), which is important for parallel

etching. If only one or no spacer fiber were used, then the capillary action would raise the



- 86 -

meniscus between the two fibers and distort the meniscus profile, producing an

asymmetric taper and shortening the taper region. A short and asymmetrical taper would

increase the optical loss.

After assembling, the core to core distance is then checked with a microscope. A

typical distance measured is 760 ± 20 μm. The manufacture specifies the jacketed fiber

with a diameter of 245 ± 5 μm, and the jacket-cladding concentricity of < 12 μm, which

results in the expected value of 735 ± 15 μm. This estimated value does not include the

fiber concentricity error and the UV film thickness in between the fibers.

With the HF etching, short fiber tapers can be made with micron precision. The

next step is to construct an ultra-compact microfluidic cell.

11.11 Waveguide Chip Fabrication

Two sets of etched tapers are needed to produce a pump-probe coupling in a waveguide

chip. Since the core-to-core distance for different fiber sets may vary, two sets of tapered

fiber are chosen to have the best match. With this matched-pair method, the performance

of DPP will not be compromised by the core-to-core distance variability in the

prefabrication process.

A glass slide is used as an assembly substrate. First, one set of the pre-fabricated

fiber tapers is UV cured onto the substrate. Before curing, the UV glue produces capillary

action to pull the tapers toward the substrate surface. After the alignment, the UV glue is

cured. Then the other set of the prefabricated fiber tapers is placed onto the substrate. The

uncured UV glue holds the second set of the fibers on to the substrate so it can be aligned

with the first under a microscope. After the alignment, the UV glue is cured to finish the

assembling procedure, as shown in Figure 11-11.

Figure 11-11 : Assembled waveguide chip. Two pairs of tapers glued onto a glass slide



- 87 -

Figure 11-12 : Microspheres coupled to tapers in a waveguide chip

11.12 Application – ZrO2 coated silica microsphere coupling

The pump-probe approach is very useful for detecting WGM which normally cannot be

coupled to a through fiber because of the resonator-waveguide refractive index mismatch.

Below is a spectrum comparison of Zirconia (Zr) coated silica microsphere using

the pump-probe approach and the conventional through fiber coupling approach.

Figure 11-13 : Fiber coupling to a ZrO2 coated silica microsphere in air. Note the shallow dip.



- 88 -

As one can see, the fiber offers poor coupling because of the index refraction mismatch.

Since little light is coupled into the microsphere, there is not enough light returning to the

fiber to produce a strong interference at the coupling region, and therefore the dip is

shallow.

One way to decrease the mismatch is to lower the refractive index of silica by

fluorine doping, which requires that the microsphere be built using expensive fluorinated

fiber. Even though this method was demonstrated for a polystyrene (PS) coated silica

microsphere, it is difficult to use for microspheres with a higher refractive index coating.

For zirconia (ZrO2) coating, the high refractive index (~1.9) cannot be compensated for

easily, and therefore continuous fiber coupling gives a large propagation constant

mismatch.

The pump-probe approach, however, does not require interference condition at the

coupling region to reveal a dip as detectable signal. Thus, even if a small amount of light

is coupled into the microsphere, the probe fiber can still pick up WGM easily, and the

zero background signals can be amplified if necessary.

Figure 11-14 : Pump-probe coupling to the same ZrO2 coated microsphere. Note the easily detectable

peaks, and large values of Q.



- 89 -

This peak spectrum is obtained using the same ZrO2 coated microsphere as the one used

for Figure 11-13. Comparing the two figures, one can see that the peaks are easy to

identify for the pump-probe coupling. In addition, the Q values are not affected using the

pump probe coupling method.

11.13 Overall

The pump probe coupling method takes more effort to produce, but it allows microfluidic

miniaturization without a need for lithography. Its most useful application may be for

resonator coupling that has mismatch in refractive index.



- 90 -

12 List of Abbreviations DAQ – Data acquisition

DFB – distributed feedback (laser)

DIO – Digital Input/Output

DPP – Dual Pump-Probe

EFL – Effective Fiber Length

(length of the tapered fiber which has its evanescent field exposed)

MC – Microcontroller

TTL – Transistor Transistor logic

PC – Personal Computer

PS – Polystyrene

PWM – Pulse width modulation

RSP – reactive sensing principle

SNR – Signal to Noise Ratio

TIR – Total Internal Reflection

WGM – Whispering Galley Mode

ZrO2 – Zirconia



- 91 -

13 References All the scientific references are included in the individual publications attached. The

references here are relevant to the engineering of the experimental setup.

1 P.W. Ewald (2002) In The Next Fifty Years, ed. Brockman J (Vintage Books, New

York), pp 289–301. 2 S. Arnold, M. Khoshsima, I. Teraoka, S. Holler, F. Vollmer, Shift of Whispering

Gallery Modes in Microspheres by Protein Adsorption, Opt. Lett., 28, 272-274 (2003) 3 F. Vollmer, D. Braun, A. Libchaber, M. Khoshsima, I. Teraoka, S. Arnold, Protein

Detection by Optical Shift of a Resonant Microcavity, Appl.Phys.Lett., 80, 4057(2002) 4 F. Vollmer, S. Arnold, D. Braun, I. Teraoka, A. Libchaber, Multiplexed DNA

Quantification by Spectroscopic Shift of Two Microsphere Cavities, Bio. Phys. J., 85,

1974-1979, 2003 5 S. Arnold, R. Ramjit, D. Keng, V. Kolchenko and I. Teraoka, MicroParticle

photophysics illuminates viral bio-sensing, Faraday Discussions, 137, 65-83 (2008) 6 F. Vollmer, S. Arnold, and D. Keng, Single virus detection from the reactive shift of a

whispering-gallery mode, PNAS, 105, 20701-20704 (2008). 7 S. Arnold, D. Keng, S. I. Shopova, S. Holler, W. Zurawsky, and F. Vollmer, Whispering

Gallery Mode Carousel – a photonic mechanism for enhanced nanoparticle detection in

biosensing, Optics Express, 17, 6230-6238 (2009) 8 G. Guan, S. Arnold and M. V. Otugen, Temperature measurements using a microoptical

sensor based on whispering gallery modes, AIAA J., 44, 2385 (2006) 9 T. Ioppolo, and M. V. Otugen, Pressure Tuning of Whispering Gallery Mode

Resonators, J. Opt. Soc. Am. B, 24, 2721-2726 (2007) 10 F. Vollmer, and S. Arnold, Whispering-gallery-mode biosensing: label-free detection

down to single molecules, Nature Methods, 5, 591 -596 (2008) 11 T. P. Burg, M. Godin, S. M. Knudsen, W. Shen, G. Carlson, J. S. Foster, K. Babcock,

and S. R. Manalis, Weighing of biomolecules, single cells and single nanoparticles in

fluid, Nature, 446, 1066-1069 (2007)



- 92 -

12 Y. Roichman, B. Sun, Y. Roichman, J. Amato-Grill, and D. G. Grier, “Optical forces

arising from phase gradients,” Phys. Rev. Lett., 100, 013602 (2008) 13 CYTOP specification sheet 14 NEL electrics DFB specification sheet 15 Corning SMF28e fiber specification sheet 16 J. P. Laine, B. E. Little, H. A. Haus, Etch-eroded fiber coupler for whispering-gallery-

mode excitation inhigh-Q silica microspheres, Photonics Technology Letters, IEEE, 11,

1429 – 1430 (1999) 17 J. C. Knight, G. Cheung, F. Jacques, and T. A. Birks, "Phase-matched excitation of

whispering-gallery mode resonances by a fiber taper," Opt. Lett., 22, 1129-1131 (1997). 18 L. Tong, R. R. Gattass, J. B. Ashcom, S. He, J. Lou, M. Shen, I. Z. Maxwell, and E.

Mazur, Subwavelength-diameter silica wires for low-loss optical wave guiding, Nature,

426, 816-819 (2003) 19 O. Gaathon, J. Culic-Viskota, M. Mihnev, I. Teraoka, and S. Arnold, Enhancing

sensitivity of a whispering gallery mode biosensor by subwavelength confinement,

Applied Physics Letters, 89, 223901 (2006) 20 J. C. Knight, G. Cheung, F. Jacques, and T. A. Birks, Phase-matched excitation of

whispering-gallery-mode resonances by a fiber taper, Opt. Lett., 22, 1129 (1997) 21 L. Collot, V. Lefèvre-Seguin, M. Brune, J. M. Raimond and S. Haroche, Very High-Q

Whispering-Gallery Mode Resonances Observed on Fused Silica Microspheres,

Europhys. Lett., 23, 327-334 (1993) 22 Y. Xia and G. M. Whitesides, Soft lithography, Annu. Rev. Mater. Sci., 28, 153–84

(1998) 23 D. W. Pohl, W. Denk, M. Lanz, Optical stethoscopy: image recording with resolution

λ/20, Applied Physics Letters, 44, 651 (1984). 24 P. K. Wong, T. H. Wang, and C. M. Ho, Optical fiber tip fabricated by surface tension

controlled etching, Solid-State Sensor, Actuator and Microsystems Workshop, 94-97

(2002)



- 93 -

The following attachments are my refereed publications associated with this thesis.

Whispering gallery mode carousel – a photonic mechanism for enhanced nanoparticle detection

in biosensing

S. Arnold1*

, D. Keng1, S. I. Shopova

1, S. Holler

2, W. Zurawsky

1, and F. Vollmer

3

1MicroParticle PhotoPhysics Lab, Polytechnic Institute of NYU, Brooklyn, New York 11201, USA 2Novawave Technologies, Redwood Shores, California 94065

3The Rowland Institute, Harvard University, Cambridge, Massachusetts 02142, USA *Corresponding author: [email protected]

Abstract: Individual nanoparticles in aqueous solution are observed to be attracted to and orbit within the evanescent sensing ring of a Whispering Gallery Mode micro-sensor with only microwatts of driving power. This Carousel trap, caused by attractive optical gradient forces, interfacial interactions, and the circulating momentum flux, considerably enhances the rate of transport to the sensing region, thereby overcoming limitations posed by diffusion on such small area detectors. Resonance frequency fluctuations, caused by the radial Brownian motion of the nanoparticle, reveal the radial trapping potential and the nanoparticle size. Since the attractive forces draw particles to the highest evanescent intensity at the surface, binding steps are found to be uniform.

2009 Optical Society of America

OCIS codes: (230.3990) Micro-optical devices; (040.1880) Detection; (020.7010) Laser trapping

References and links

1. A. Ashkin and J. M. Dziedzic, “Optical Trapping and Manipulation of Viruses and Bacteria,” Science 235, 1517-1520 (1987).

2. F. Vollmer and S. Arnold, “Whispering-gallery-mode biosensing: label-free detection down to single molecules,” Nature Methods 5, 591-596 (2008).

3. A. M. Armani, R. P. Kulkarni, S. E. Fraser, R. C. Flagan, and K. J. Vahala, “Label-free, single-molecule detection with optical microcavities,” Science 317, 783-787 (2007).

4. T. M. Squires, R. J. Messinger, and S. R. Manalis, “Making it stick: convection, reaction and diffusion in surface-based biosensors,” Nature Biotechnol. 26, 417-426 (2008).

5. J. C. Knight, G. Chung, F. Jacques, and T. A. Birks, “Phase-matched excitation of whispering-gallery-mode resonances by a fiber taper,” Opt. Lett. 22, 1129-1131(1997).

6. L. Yang, T. Carmon, B. Min, S. M. Spillane, and K. J. Vahala, “Erbium-doped and Raman microlasers on a silicon chip fabricated by the sol-gel process,” Appl. Phys. Lett. 86, 091114 (2005).

7. S. Arnold, M. Khoshsima, I. Teraoka, S. Holler, and F. Vollmer, “Shift of whispering-gallery modes in microspheres by protein adsorption,” Opt. Lett. 28, 272-274 (2003).

8. L ≈ (λ/4π)(ns2-nm

2)-1/2, D = 2nm2 (2ns)

1/2(nnp2 – nm

2)/(ns2 – nm

2)(nnp2 + 2nm

2), where ns, nm, and nnp are the refractive indices of the microsphere (1.45), aqueous medium (1.33), and nanoparticle (1.5 for virus and 1.59 for polystyrene; D = 1.50 and 2.26 respectively).

9. A. Ashkin, J. M. Dziedzic, J. E. Bjorkholm, and S. Chu, “Observation of a Single-Beam Gradient Force Optical Trap for Dielectric Particles,” Opt. Lett. 11, 288-290 (1986).

10. Y. Roichman, B. Sun, Y. Roichman, J. Amato-Grill, and D. G. Grier, “Optical forces arising from phase gradients,” Phys. Rev. Lett. 100, 013602 (2008).

11. J. N. Izraelachvili, Intermolecular And Surfaces Forces. 173-191 (Academic Press, Inc. , San Diego, CA, 1987).

12. I. Teraoka and S. Arnold, “Theory of resonance shifts in TE and TM whispering gallery modes by nonradial perturbations for sensing applications,” J. Opt. Soc. Am. B 23, 1381-1389 (2006).

13. F. Vollmer, S. Arnold, and D. Keng, “Single Virus Detection from the Reactive Shift of a Whispering-Gallery Mode,” Proc. Natl. Acad. Sci. USA 105, 20701-20704 (2008).

14. The translation from a size to a mass spectrum requires knowledge of mass density.

(C) 2009 OSA 13 April 2009 / Vol. 17, No. 8 / OPTICS EXPRESS 6230#107949 - $15.00 USD Received 25 Feb 2009; revised 31 Mar 2009; accepted 31 Mar 2009; published 1 Apr 2009

http://mp3l.org/abs/073abs.htm

http://mp3l.org/abs/073abs.htm

15. H. J. Yang, S. D. Moore, B. S. Schmidt, M. Klug, M. Lipson, and D. Erickson, “Optical manipulation of

nanoparticles and biomolecules in sub-wavelength slot waveguides,” Nature 457, 71-75 (2009)

1. Introduction

Light forces interacting with mechanical systems provide a unique tool for studying small biological objects [1]. In the case of a whispering-gallery-mode (WGM) bio-sensor [2] where the intensity within the evanescent volume is built up resonantly a light-force may answer a puzzling question. Measured binding rates of bioparticles in aqueous solution using a toroidal WGM bio-sensor [3] at ultra-low concentrations appear to be about one hundred times higher than calculated based on diffusive and convective transport theory [4]. Traditionally Brownian motion of ultra-low concentration analytes crossing the boundary layer has been considered as a major hurdle for the practicality of miniature bio-sensors [4]. There is no comprehensive model to explain the physical mechanism of enhanced binding rates in the case of WGM bio-sensors. Here we report an observation and analysis of an optical mechanism that enhances the transport rate to the sensing volume of a microspherical silica resonator by more than 50x. Polystyrene nanoparticles are drawn toward this volume by evanescent optical gradient forces generated with just a few microwatts driving the resonator. At low ionic strength an addition electrostatic force repels the nanoparticle from the surface, contributing to a radial trap. Here the particle finds itself in the tangential momentum flux of the WGM and is driven to orbit by scattering forces. The radial trapping potential is elucidated from fluctuations in the micro-cavity’s resonance frequency, allowing the use of the Carousel as a surface-potential nanoprobe. The maximum fluctuation enables the size and mass of the trapped nanoparticle to be determined without binding, suggesting that the WGM Carousel mechanism can be used for size/mass spectrometry in solution. At a considerably high ionic strength the electrostatic field is screened to a much shorter depth, and the particle is drawn closer to the surface where it is caught by a van der Waal interaction and binds. Resonance shifts due to these binding events are found to be steps having uniform heights.

2. The Whispering Gallery Mode Carousel Phenomenon

Nanoparticles suspended in an aqueous environment normally appear to be in Brownian motion. However, we observe in the vicinity of a bare silica microsphere (oblate microspheroid, eccentricity < 5%, equatorial radius R ≈ 50 µm) excited into a circulating WGM (quality factor Q ~ 10

6), nanoparticles as small as 140 nm radius (a) are trapped for

hundreds of seconds in orbit within the sensing volume with driving light power P ≈ 50 µW. As shown in Fig. 1(a), these nanoparticles appear to circumnavigate in the direction that light takes within the WGM. The nanoparticle concentration was ≈ 1 fM in D2O. D2O (Aldrich, 99.9%) was used to minimize absorption loss in the infrared. The particle recorded in the video was seen to orbit for over two revolutions before escaping, Fig. 2.

A tapered fiber which coupled power into the microsphere was positioned a few microns to one side of the equator. At resonance, a dip was observed in the power transmitted through

the fiber at wavelength λr as the laser was tuned. The power P driving the WGM was estimated from this dip depth [5, 6]. In addition to the deterministic propulsion, the trapped particle is also under the influence of Brownian motion, revealed as a blinking of the elastic

scattering signal from the particle, as well as by the delimited fluctuations in λr (Fig. 1(b)). In what follows we will show that the physical interpretation of these fluctuations reveals the trapping potential and the size of the nanoparticle. This potential is responsible for increased transport of target nanoparticles to the sensing volume.

Fractional fluctuations in the resonance wavelength from the background level ∆λr/λr are clearly due to perturbations in the WGM as the result of nanoparticle’s interaction with the microcavity, and are equal at each instant to the ratio of the energy polarizing the particle Wp to the energy in the cavity Wc (reactive sensing principle, RSP) [7],

∆λr/λr = Wp /Wc (1)

#107949 - $15.00 USD Received 25 Feb 2009; revised 31 Mar 2009; accepted 31 Mar 2009; published 1 Apr 2009

Fig. 1. WGM-Carousel-Trap. (a) WGM excited in a microsphere (radius R = 53 µm) with Q =

1.2×106 by a 1060nm tunable laser using fiber-evanescent-coupling. The resonance wavelength is tracked from a dip in the transmitted light (PD). An elastic scattering image shows a

polystyrene particle (radius a = 375 nm) trapped and circumnavigating at 2.6 µm/s using a

drive power of 32 µW. (b) A particle is sensed through resonance wavelength fluctuations ∆λr that identify its size/mass. These fluctuations are recorded from before the particle enters the Carousel-trap until after it escapes ≈ 6 min later.

Fig. 2 (Media 1) This is a sped-up video (16× real time) of a single nanoparticle (a = 375 nm) being trapped and propelled by the WGM momentum flux. The fiber is coupled to the microsphere (R = 48 µm) by contact slightly off the equator on the backside. The WGM has Q = 1.5×106, and is driven with a power P = 25 µW. Light travels in the fiber from right to left (WGM scatter can be seen on the left edge of the microsphere). The trapped particle is observed through elastic scattering as a bright spot in front and in back of the microsphere. The ring pattern around the bright spot is caused by diffraction by the microscope objective. The nanoparticle is trapped, and propelled for just over two revolutions with a period of 140s before escaping. The particle appears to move faster on the backside due the transverse magnification in the microsphere image.

The shift ∆λr is therefore independent of the power driving the resonator but is proportional in a dipole approximation to the ratio of the intensity at the nanoparticle’s center r

c to the energy

in the mode; ∆λ

r(r

c) ∝ E

0

2 (rc) ε(r)E

0

2 (r ) , where 0

E is the electric field amplitude and

ε(r) is the modal permitivity. For the lowest order angular wave excited in our experiments

the intensity function is symmetrical about the equator with a Gaussian-like shape, and falls

20µµµµm


http://www.opticsexpress.org/viewmedia.cfm?URI=oe-17-8-6230-1

off at “latitudes” on either side with a characteristic width w ≈ 6 µm. In contrast, in the radial direction the intensity falls off as the square of a spherical Hankel function which is well approximated by a decaying exponentially with a much shorter “evanescent length” L ≈ 150 nm [8]. Images of the particle’s orbit show it travelling along the equator with a root mean

square transverse displacement to either side of < 1.5 µm. The time to diffuse over this distance for our typical nanoparticle is several seconds, whereas the observed fluctuations in

∆λr occur much faster: over a time scale associated with diffusion through a length ~ 100 nm. Consequently, these fluctuations are due to changes in the interfacial separation h between the nanoparticle’s surface and the surface of the microsphere, with the maximum fluctuation

occurring at h ≈ 0 (i.e. green line in Fig. 1(b)). Translating other wavelength shift levels into h

values is critical to our analysis. Fortunately, this translation is easily implemented. Based on the RSP

[7] the wavelength shift for the nanoparticle’s center at h+a to that on the surface is

∆λr(h + a)

∆λr(a)

= exp[- (h + a) L ]

exp[- a L ]= exp[- h L ] . (2)

Equation (2) enables wavelength shift statistics to be transformed into separation statistics. The results are particularly revealing.

3. Trapping potential well

Figure 3(a) shows the separation histogram taken on a nanoparticle (a = 140 nm) that circumnavigated a microsphere for just over two orbits. A pronounced maximum is seen at

h ≅ 35 nm from the sensing surface. The peak is indicative of a surface repulsion that

becomes more evident by translating these separation statistics into a potential curve using equilibrium statistical mechanics.

Under the assumption of thermal equilibrium the Boltzmann distribution relates the

potential U(h) to the probability density p(h); U (h) = −k

BT ln[ p(h) p(h

ref)] , where href is a

reference separation for which U (href) = 0. Figure 3(b) shows the result. The particle is clearly trapped in a radial potential well with its minimum 35 nm from the surface as it is driven to orbit by the WGM’s tangential momentum flux. These potential points were fit by a sum of

two exponentials: a short-range repulsive interaction ( )/ 6.2exp - / 17.6 nmBsU k T h= , and a

long-range attractive interaction ( )/ -8.0exp - / 142.7 nmBpU k T h= . The latter supports our

hypothesis that the particle’s motion is principally radial since its characteristic length of 143 nm is close to the evanescent length in the radial direction (146 nm). The attractive force arising from this potential is similar to the gradient force in optical tweezer experiments, for which the potential in the equatorial plane is expected to be the negative of the polarization

energy, U p (h) ≈ − (αex 4) E0

2(a) exp(−h / L) where αex is the nanoparticle’s polarizability [9].

Indeed, a series of experiments show that the value of this “polarization potential” at the surface Up(0) is proportional to the power P entering the mode. The gradient force is aided in keeping the particle on an equatorial track by a transverse phase-gradient contribution [10]. The positive potential Us is independent of power, and appears to be due to repulsion between ionized silanol groups on the bare silica surface (pH = 7), and the negatively charged polystyrene nanoparticle (the particles used were slightly sulfonated). The characteristic

length of Us is close to the Debye length λD arrived at from the measured conductivity of our

medium [11], λD ≈ 20 nm, assuming monovalent ions. By varying the ionic conductivity of the solution one can effectively change the range of Us. In contrast, Up is independent of ionic conductivity and reaches much deeper into the solution. In effect the combined potential forms a “sink-hole” that draws particles toward the optimal region in the sensing volume.


Fig. 3. Separation histogram and trapping potential. (a) separation histogram collected from a single tapping event of a polystyrene (PS) particle (from mean radius <a> =140 nm hydrosol).

The WGM with Q = 7.3×105 was excited with P = 233 µW at λ ≈1060 nm in a microsphere

with R = 44 µm. The statistics were comprised of 1000 points. (b) Potential plot arrived at from the histogram in (a). These points are fit to a sum of two potentials (in red).

It is important to point out that not all forces in the optical problem may be considered conservative. [10] Our description of a potential associated with the separation statistics is strictly meant to apply to conservative forces in the equatorial plane.

The value of the polarization potential at zero separation, Up(0), may be calculated directly

in terms of the maximum wavelength shift (∆λr)max = ∆λr(a), the power P driving the mode, and its resonant Q by using the RSP, Eq. (1). [7] The polarization energy at zero separation Wp(0) = - Up(0) and the energy in the cavity Wc is the driving power P times the photon

lifetime Q/ωr. Consequently, the potential at zero separation is

U

p(0) = − ∆λ

r( )max

P Q / 2πc( ), (3)

where c is the speed of light. Whereas (∆λr)max is independent of P or Q, the attractive potential grows as their product. If we suppose Us is very short range, then the minimum

power to perceive trapping, Pmin , can be estimated by setting (0)p BU k T≈ ;

P

min≈ k

BT (2πc) / [Q (∆λ

r)

max] . (4)

Since all of the parameters on the right in Eq. (4) are measurable, the thermal escape hypothesis is testable by measuring Pmin.

A series of five experiments were performed in order to detect the minimum power to keep a particle in orbit. In each the power driving a WGM was lowered as an orbiting particle’s velocity was measured from a video recording. Figure 4 shows the results of one of

these experiments for which the power was lowered from 42 µW over a period of 1200 s. The

particle was lost as the power reached 7.3 µW. At this power the normalized potential from

Eq. (3) (0) 1p BU k T ≈ , consistent with thermal escape (as indicated by the top horizontal


scale in the Fig. 4). The recorded velocities in the Fig. 4 do not appear to be heading toward an intercept at the origin as might be expected. The reason lies in the fact that although the momentum flux at a given height decreases in proportion to drive power, the flux seen by the particle falls more rapidly, since the particle moves outward as the drive power decreases. The other four experiments showed similar results. The picture that evolves is of a particle attracted to the orbit and rapidly fluctuating radially above the equator by Brownian forces. This trapping mechanism also leads to enhanced detection rates in the WGM biosensor.

Fig. 4. Particle velocity as a function of drive power P. A nanoparticle of radius a = 375 nm

was trapped in a Carousel of a microsphere with R = 45 µm and Q = 1.5×106. The power was

gradually reduced over a period of 1200 s. Upon reaching 7.3 µW the particle escapes within 10 s, as seen by imaging and through the cessation of wavelength fluctuations. The upper horizontal scale is calculated from Eq. (3).

Carousel trapping is expected to enhance binding rate detection in the following manner: The enhanced transport increases accumulation of particles in the sensing volume, and forced re-visitations by a particle to the surface increases the probability for finding a binding site. To measure the enhancement of the transport rates we compared the case of pure diffusion driven particles by operating at an arbitrarily low polarization potential |Up(0)|d ≈ 0.01 kBT to the case where the polarization potential was near the threshold for escape, |Up(0)|e ≈ 1kBT. These experiments were carried out for particles with a = 250 nm and for a concentration of 6 fM. In each case we counted the number of visitations to the sensing volume by registering

the number of wavelength shift pulses exceeding 0.25(∆λr)max. For |Up(0)|d we detected only one visitation to the sensing volume in 1 hour. However for |Up(0)|e, 49 visitations ~ 1 s in duration were detected in 1 hour. As the potential was increased to |Up(0)| > 5kBT, particles were strongly trapped in the Carousel, and accumulation of multiple particles over time became unavoidable. With |Up(0)| above 2kBT trapping of a single particle for several minutes became highly probable, which allowed us to observe the delimited fluctuation. This limit,

where the particle temporarily “touches” the surface, (∆λr)max provides a means for the determination of the particle size/mass.

The maximum wavelength shift signal (∆λr)max registered for a given laser-resonator combination appears to depend only on the size and dielectric properties of the orbiting nanoparticle. This signal occurs when the particle encounters the greatest evanescent field. As the drive power is raised more of these events occur, but the largest have the same limiting value. Theory was constructed early on that related this wavelength shift to the particle’s polarizability and size by using the RSP [7, 12], but its confirmation has only come recently


with the observation and measurement of single wavelength shift steps in non-specific binding experiments [13]. However, these steps were random in size, associated with particles adsorbing at different latitudes on the sphere’s surface where the sensing response can vary over orders of magnitude. In the Carousel mechanism, the particles are attracted to the equator, and therefore produce uniform response in the delimited fluctuation. This allows a nanoparticle to be sized without the need for binding. The RSP

[7] provides an illuminating

equation for describing the wavelength shift [13]

( )1/ 2

3

5 / 2 max

r

r

a LD a

Re

λλ −∆ ≃ (5)

where D is a dimensionless dielectric factor equal to 2.26 for polystyrene in water. [8] Table 1

shows a comparison of particles sizes determined from (∆λr)max, for separate experiments on single polystyrene nanoparticles caught in Carousel traps by inverting Eq. (5), with the mean size reported for a statistical number of particles by the manufacturer.

Table 1. Nanoparticle Sizing by WGM Carousel. Size determined for each of four Carousel trapped nanoparticles

from their delimited wavelength shift (∆λr)max using Eq. (5) (far right) as compared with the mean size given by the manufacturer for the associated hydrosol <a>.

<a> (nm)

σ = 5% λ (nm)

Nominal R ± 1.5

(µm)

(∆λr)max ± 0.02 (pm)

a (nm) from RSP

140 ± 7 1059 43 0.25 158 ± 12

245 ± 12 1059 51 0.32 228 ± 19

245 ± 12 1312 56 0.42 247 ± 17

375 ± 19 1312 56 0.67 350 ± 21

Although the experiments were for resonators of different sizes and driven by different lasers, the nanoparticle size obtained by inverting Eq. (5) agreed with the manufacturer’s mean size <a> within the uncertainties in the experiment and the standard deviation in the manufactured hydrosols. This clearly opens the door for a nanoparticle size/mass spectrometer

in solution [14]. With a microsphere for which R = 40 µm and Q ≈ 107

individual bioparticles

having a mass of HIV (600 attograms, a ≈ 50 nm) should be easily sizable with P ≈ 50 µW at

λ ≈ 780 nm. For a power of 2 mW using the same resonator a smaller virus with a = 15 nm (mass ≈ 15 attograms, e.g. Poliovirus) is within reach.

Finally we return to the subject of binding. A particle caught in a Carousel orbit is in a pre-binding state which is easily converted to a binding state by reducing the range of the electrostatic repulsion and thereby allowing the optical force to pull a nanoparticle to the surface where the intrinsic van der Waal attraction can take hold. Decreasing the range of the electrostatic repulsion is accomplished by increasing the conductivity of the solution. Figure 5 shows two separate experiments on individual particles caught in Carousels for which the solution conductivities differed by an order of magnitude. The experiment at higher conductivity clearly shows the separation between the nanoparticle and the surface to be substantially reduced. An analysis of these separation statistics gave a repulsive potential for the low conductivity case corresponding to 0.5mM NaCl of

( )/ 6.2exp - / 17.6 nmBsU k T h= , whereas in the higher conductivity case corresponding to

5 mM NaCl ( )/ 4.9exp - / 6.1 nmBsU k T h= . The reduction in the range of the repulsive

potential by a factor of 2.9 is close to the expected reduction in the Debye length. The later is known to be inversely proportional to the square root of the salt ionic strength, (reduction in range by 3.2). [11]


Fig. 5. Particle separation histograms for two different NaCl concentrations (0.5 mM and 5 mM). Note that the particle is closer to the surface for higher salt concentration, indicated by the peak position of the statistics.

By adding 20 mM of NaCl to our D2O solution its conductivity was increased 40×. Nanoparticles (a = 375 nm) were trapped in the Carousel and bound to the surface. Figure 6(a)

shows the first three binding events registered as uniform steps in ∆λr. Although spatially random particle adsorption leads to a distribution of step heights which vary by more than an order of magnitude [13], the constancy of the step heights in Fig. 6(a) shows that the equator can be spatially isolated for binding. Such characteristics were also demonstrated with smaller nanoparticles (a = 140 nm). Figure 6(b) shows an experiment in which several of these particles bind to the Carousel surface from a 10 fM solution over a period of 20 minutes. In this case, there were two parallel paths, corresponding to a mode for which quantum number

1m l= − , indicating that the Carousel effect exists for higher order angular modes as well.

Fig. 6. (a) First three binding steps of nanoparticles (a = 375 nm) on a microsphere with R = 45

µm and P = 150 µW, Q = 2×105, Note the uniformity in step height. Red dash separation is set to 0.45 pm. (b) Image of a = 140 nm particles trapped and bound in the Carousel orbit, R = 39

µm.

4. Conclusions

In conclusion, the WGM Carousel mechanism provides underlying physics which answers the question posed in Ref. 4. We clearly see that detection rates are not limited by diffusion, and

20µµµµm 20µµµµm


can be increased by ~ 100×. In addition we discover that the new light-force mechanism provides a sensitive means for sizing individual particles and detecting their interactions with the sensor’s surface. The effects produced by Carousel trapping should be difficult to avoid for viral sized nanopartices such as HIV or Influenza A, since the power needed to form the

Carousel is < 200 µW. This power is orders of magnitude smaller than the trapping power reported with linear optical waveguides (~ 250 mW) [15]. Analytical and experimental studies show that our low trapping power is due to resonant build up within the spherical microcavity structure. In addition, by controlling the ionic strength and the trapping optical power, the particles are shown to bind preferentially within the Carousel. All of this provides optical WGM sensors with a distinct advantage not afforded to non-optical devices since the attractive potential reaches out into a solution and draws nanoparticles to the optimal sensing region unabated by ionic screening. In addition, proteins are also expected to interact with the Carousel, and light-force assisted functionalization of the resonator’s equator is possible.

Acknowledgments

S. A. thanks D. G. Grier of NYU for useful discussions. This work was principally supported by the National Science Foundation - Division of Bioengineering and Environmental Systems Grant No: 0522668. D.K. thanks an NYU-POLY seed grant for partial support. F.V. was supported by a Rowland Junior Fellowship.


Single virus detection from the reactive shiftof a whispering-gallery modeF. Vollmera,1, S. Arnoldb,1, and D. Kengb

aThe Rowland Institute, Harvard University, Cambridge, MA 02142; and bMicroParticle PhotoPhysics Lab, Polytechnic Institute of New York University,Brooklyn, NY 11201

Edited by Robert H. Austin, Princeton University, Princeton, NJ, and approved November 10, 2008 (received for review September 9, 2008)

We report the label-free, real-time optical detection of Influenza Avirus particles. Binding of single virions is observed from discretechanges in the resonance frequency/wavelength of a whispering-gallery mode excited in a microspherical cavity. We find that themagnitude of the discrete wavelength-shift signal can be suffi-ciently enhanced by reducing the microsphere size. A reactivesensing mechanism with inverse dependence on mode volume isconfirmed in experiments with virus-sized polystyrene nanopar-ticles. By comparing the electromagnetic theory for this reactiveeffect with experiments, the size and mass (5.2 1016 g) of abound virion are determined directly from the optimal resonanceshift.

biosensor influenza optical resonance

V irus particles are a major cause for human disease, and theirearly detection is of added urgency since modern day travel

has enabled these disease agents to be spread through popula-tions across the globe (1). Fast and early detection on site of anoutbreak requires biosensors where ideally individual viral par-ticles produce a quantitative signal. Here, we report the obser-vation of discrete changes in frequency of whispering gallerymodes (WGMs) as Influenza A virions bind to a microspherecavity. A ‘‘reactive’’ perturbation of the resonant photon state isconfirmed in measurements with similar-sized polystyrene (PS)particles near a wavelength of 1,310 nm: The frequency/wavelength shift signal follows a strong dependence on cavitycurvature near the predicted R5/2 scaling (2), providing amechanism for increasing signal by limiting modal volume. Byreducing the microsphere radius to just 40 m and operatingat a more favorable wavelength near 760 nm where reducedwater absorption is expected to enhance sensitivity (3), bindingsteps of individual Influenza A (InfA) virions are seen that easilyexceed the experimental noise level. Analytic equations arederived that relate discrete changes in resonance wavelength tothe size and mass of adsorbed virions. Although field effecttechniques using nanofibers (4) and interferometric approachesbased on light scattering (5) have demonstrated single virionsensing in the past, reactive WGM sensing adds new dimensionsto what can be learned: The measured wavelength shift enablesone to quantitatively identify the virion size and mass.

Experimental ApproachWGM resonances are perturbed toward longer wavelength asparticles with polarizability in excess to that of water adsorb toa microsphere cavity. Individual binding events have beentheorized to produce discrete steps in a time-trace of thewavelength shift signal (2). To probe for single binding events weimmerse a silica microsphere in a suspension of polystyreneparticles (PS) with radius a 250 nm (Fig. 1). The PS particlesare diluted in PBS to final concentrations 10–50 fM. A tunabledistributed feedback laser (DFB) (1,311 nm nominal wave-length) excites WGMs by evanescent coupling from a taperedoptical fiber. A WGM mode is detected as a Lorentzian-shapedtrough in a spectrum acquired by a photodetector that measurestransmission through the fiber as the wavelength of the laser is

tuned (6). Fig. 1 Inset shows the typical transmission spectrumfor a WGM excited in silica microsphere (here radius R 50m) immersed in aqueous solution. The linewidth 5 pm, asdetermined from the full width at half-maximum, corresponds toa Q factor Q / 2.6 105, primarily limited by overtonevibrational absorption of water in the near infrared. Micro-spheres are fabricated from tapered optical fiber tips that aremelted in a focused 10-W CO2 laser (7). Immediately afterfabrication, the microsphere-on-a-stem is mounted on the sam-ple cell and immersed in PBS solution. A small box encloses thesample cell to limit air f low and stabilize the ambient humiditylevel. A transmission spectrum is acquired every 20 ms, and theresonance wavelength is determined from a parabolic minimumfit to the Lorentzian-line, typically with precision 1% of thelinewidth. The dip-trace is displayed as fractional shift in wave-length /.

Fig. 2A shows a trace of / for a 250 nm PS particlesinteracting with a microsphere with R 45 m. Steps of variousheights are clearly visible against the background cavity noiseindicating adsorption of individual PS particles. We believe thatthe spikes such as those shown in Fig. 2B are associated withunsuccessful adsorption attempts. Fig. 2C shows the step statis-tics. A maximum step height can be distinguished. With theability to detect discrete steps and identify optimal step heightwithin a dip trace for a given nanoparticle radius, we can nowdirect our attention toward measuring the dependence of theoptimal shift signal on the microsphere curvature. We experi-ment with different-sized microspheres (R 44–105 m) andplot maximum step heights versus curvature (i.e., 1/R, Fig. 3).Interestingly, we find a strong dependence of the fractionalwavelength shift on the cavity radius, scaling as R2.67. This is ingood agreement with electromagnetic theory associated withreactive WGM sensing, for which steps have been predicted forsingle protein binding, with maximum step heights in proportionto R2.5 (2). The largest step heights are predicted for proteinparticles binding to the equator, whereas a 1/R dependence (2)is expected for a shift due to a random surface density. (6) Thesmall discrepancy in the exponent (2.67 vs. 2.5) in our PSparticle experiments may be explained by an increase in theevanescent interaction caused by a slight lengthening in theevanescent field depth L as the microsphere size is decreased,and is associated with large particles adsorbed for which a/L 1. For particles that are small compared with the evanescent fielddepth, such as protein or InfA virus, our optimal wavelengthshifts are consistent with the reactive theory, but the weak signal

Author contributions: F.V. and S.A. designed research; F.V. performed research; F.V.contributed new reagents/analytic tools; F.V., S.A., and D.K. analyzed data; and F.V. andS.A. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

1To whom correspondence may be addressed. E-mail: [email protected] [email protected].

© 2008 by The National Academy of Sciences of the USA

www.pnas.orgcgidoi10.1073pnas.0808988106 PNAS December 30, 2008 vol. 105 no. 52 20701–20704

PHYS

ICS

BIO

PHYS

ICS

associated with these smaller particles prohibits us from acquir-ing data over a wide range of microsphere sizes.

Having identified a means for increasing the shift magnitudeby reducing microcavity size, we set out to optimize the micro-sphere system for the detection of single InfA virions. InfAvirions have an average radius a 50 nm (8) and a refractiveindex below that of PS. Although our microcavity fabricationmethod would allow us to take further advantage of the largecurvature enhancement of the wavelength shift signal by formingsmaller cavities, the success of this approach is limited: furtherreduction of cavity size also increases cavity leakage so that anygain in sensitivity due to shift enhancement is offset by abroadening of the resonance linewidth. Instead, we find thatfurther increase in sensitivity is possible if we scale down thewavelength and the cavity. A wavelength 1,311 nm is alsofavorable due to less absorption in water. Following this ap-proach, we use DFB laser with 763 nm and excite a WGMwith Q 6.4 105 in R 39-m microspheres. We inject InfAvirions at concentration of 10 fM directly into a PBS filledsample cell, since the virions are known to adsorb to silica (9).The dip-trace of the resonance wavelength InfA/ in Fig. 4

reveals clear steps associated with binding of single viral particlesand one unbinding event. The signal-to-noise ratio (InfA/noise 3) can be further improved upon by signal processingschemes such as Median filtering; the solid line in Fig. 4 showsa median filter of rank 3. A similar median filter has been appliedin the green curve in Fig. 2 A.

High-sensitivity measurements of individual virus particles aremade possible by reducing modal volume, a principle that shouldalso apply to other WGM cavity geometries (10) and othernon-WGM microcavities (e.g., photonic crystals with ‘‘defects’’)(11). Below, we show that the steps heights already recorded forthe microsphere geometry can be described analytically by usingthe reactive theory, and that the viral size and mass may beidentified.

Reactive Sensing MechanismReactive sensing relies on the fact that work is done by theevanescent field of a microcavity as a polarizable nanoparticlemoves from a distant position to the microcavity surface. As aresult, the energy of light in the resonator is reduced. With thenumber of cavity photons conserved, the frequency of each

Fig. 1. Excitation of an equatorial WGM of a microsphere by evanescent coupling to a guided wave in a tapered optical fiber. Resonance positions are detectedas dips in the transmitted light T at particular laser wavelengths.

Fig. 2. Nanoparticle wavelength shift. (A) Dip-trace for a 250 nm PS particles interacting with a microsphere with R 45 mm. (B) A 15 second zoom-in. (C)Wavelength step statistics.

20702 www.pnas.orgcgidoi10.1073pnas.0808988106 Vollmer et al.

resonant photon is shifted by r in accordance with refs. 2and 3.

–hr ex

2Erv, t2 , [1]

where E(rv, t)2 is the time average of the square of the field atthe nanoparticle’s position rv due to a single photon resonantstate. We have assumed that the particle is small in relation tothe wavelength, and has an isotropic excess polarizability ex

including local field effects. Under these conditions, Eq. 1 shouldwork for any microcavity geometry. By dividing the shift infrequency by the single photon energy –hr on the left and by thevolume integral of the associated electromagnetic energy densityon the right, we arrive at a simple expression for the fractionalfrequency shift (2),

r

r

ex/0 E0rv 2

2 rr E0r 2dV

, [2]

where E0 is the electric field amplitude, and r(r) is the dielectricconstant throughout the cavity. Although the field in Eq. 1 isassociated with a single photon, this restriction does not apply toEq. 2, since both the numerator and denominator are separatelyproportional to the number of photons; the reactive effect isindependent of intensity. In addition, the volume integration inthe denominator suggests that the reactive effect should varyinversely with volume. This insight although approximate, isnone-the-less almost correct. In what follows we describe ourtheoretical results for the resonance shift of a microsphericalcavity. From this point onward we will express the resonanceshift as a shift in free space wavelength in accord with theexperimental data (i.e., r/r r/r).

Evaluation of Eq. 2 for a microspherical cavity is mostelegantly carried out for the lowest order WGM launchedaround the equator of a glass sphere of radius R.2 Such a modespreads symmetrically to either side of the equator with aGaussian-like profile, causing nanoparticles adsorbing above orbelow the equator to have a diminished shift relative to theequatorial shift. The maximum shift (i.e., equatorial) for ananoparticle of radius av adsorbing on the equator is found to be

r

r

max

Dav

3

R5/2 r1/2 eav/L, [3]

where L is the characteristic length of the evanescent field, andD is dimensionless dielectric factor* associated with both themicrosphere and nanoparticle. The wavelength shift enhance-ment that results from the reduction in microsphere size is clearlyseen in Eq. 3 to be proportional to R5/2, in good agreement withexperiment (Fig. 3). The exponential factor on the right resultsfrom the variation of the evanescent field over the radius of thenanoparticle. One can invert this transcendental equation toobtain the radius av of the adsorbed particle;

av a0

1 a0

3L

, [4]

where a0 is

a0 R5/6r

1/6

D1/3 r

r

max

1/3

. [5]

Numerical studies indicate that Eq. 4 deviates from the exactsolution to Eq. 3 by 1% for av 90 nm, and R 30 m.

DiscussionEq. 3 provides a clear statement with respect to the dependenceof the wavelength shift on microsphere curvature, and ourpolystyrene experiments agree well with respect to the exponent.Further evidence for the reactive mechanism results by compar-ing the nominal radii of the particles used in our experimentswith the radii arrived from the maximum measured wavelengthshifts (using Eq. 4). These results are summarized in the table.We note that our measurements agree well with Eq. 4 for a 250 nm. We also would like to point out that the estimate ofradius is a lower limit since only few particles bind directly on theequator. The mass mv of the InfA virion can now be evaluatedfrom its volume times its density to be 5.2 1016 g, inagreement with results for InfA’s molecular weight found fromthe sedimentation in density gradients based on a statisticalnumber of viruses (3 108 g/mol) (12).

*L (/4) (ns2 nm

2 )1/2 and D 2nm2 (2ns)1/2(nnp

2 nm2 )/(ns

2 nm2 )(nnp

2 2nm2 ), where ns, nm,

and nnp are the refractive indices of the microsphere (1.45), aqueous medium (1.33), andnano-particle (1.5 for virus and 1.59 for polystyrene).

Fig. 3. Maximum step height vs. microsphere curvature for polystyreneparticles with radius a 250 nm.

Fig. 4. Shift signal for InfA. The data were acquired for a microsphere withR 39 mm, and a DFB laser having a nominal wavelength of 763 nm.

Vollmer et al. PNAS December 30, 2008 vol. 105 no. 52 20703

PHYS

ICS

BIO

PHYS

ICS

ConclusionsWe have shown that adsorption of individual nanoparticlesproduce discrete changes in resonance frequency/wavelength ofa WGM. We confirm the reactive sensing mechanism (Eq. 1 and2) and use its signal enhancement by reducing the size ofmicrospherical cavities. Detection of individual InfA virions inaqueous buffer is then demonstrated directly from steps in thewavelength shift signal. Both the virus size and mass are iden-tified from maximum step height associated with binding nearthe equator. This work may be considered a forerunner forreactive biosensing with microcavities of ultrasmall modal vol-ume such as ‘‘defects’’ within photonic crystals (11), since the Eq.1 and 2 should still apply.

The number of attempts needed to bind is a direct reflectionof the affinity of the bioparticle to a surface. No effort has beenmade in this first demonstration to extract ligand-receptoraffinities, however, such experiments will be soon to follow. Inaddition the tumbling of a nearly spherical virus is not expectedto lead to significant time dependent variations in the wavelengthshift signal, however, rod like virus will couple through a tensor(3) interaction to the electromagnetic field and should lead tosignificant time variations, allowing one to gain more detailsconcerning affinities associated with orientation. Fortunately,our approach has considerable bandwidth since a WGM cansense temporal variations down to the photon life Q/, whichis 1010 s for the Q values reported in the current work.Furthermore, the genus of a virus may be determined frommeasurements that are sensitive to a virion’s shape in additionto its size and mass. Considering the quantitative nature of thephysical interaction, and its large available bandwidth, the futureof single particle reactive WGM experiments is expected to beexpansive. The approach based on microspheres has advantagesover other cavity geometries and label-free sensing mechanismswhere a closed-form analysis of the wavelength shift signal maynot be possible (13).

MethodsPurified and formalin-inactivated human InfA virus A/PR/8/34 is purchased in4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (Hepes) buffer fromCharles River Laboratories. The virus sample is passed through a 0.2 m nylonfilter to remove aggregates and then concentrated in a speed vac. WGMsample cells (Fig. 1) are assembled from o-rings (McMaster–Carr, 12.5 mmdiameter) glued to a 48 65 mm no. 1 coverslip (Gold Seal). A microsphere-on-a-stem is then glued to the o-ring so that the sphere resides in the centerof the cell without touching the coverslip. The sample cell is mounted on aninverted microscope and visually inspected. The midsection of a taperedoptical fiber, which is held between the posts of a u-shaped metal holder, ispositioned in contact with the equator of the sphere (Fig. 1), using micrometerscrews. Alignment is adjusted by optimizing the coupling to WGM modes byslightly moving the point-of-contact between taper and sphere within theequatorial region. To check for viral aggregates and estimate concentration,an aliquot of the purified virion solution is stained with DiIC membrane dye(Invitrogen) and fluorescently imaged using a cooled charge-coupled device(CCD) camera (Cooke). Similarly, fluorescent carboxylated polystyrene parti-cles (Invitrogen) are checked for aggregates before use. Microspheres ofcontrollable size are fabricated from tapered optical fiber (Corning; SMF-28).The taper is held on a stage above a reflecting Aluminum surface and meltedin the focal spot of a focused 10-W CO2 laser (Synrad). The process is inspectedon a CCD camera while progressively more silica can be melted by pushingmore fiber into the beam where surface tension immediately forms a sphere.Tapered SMF-28 fiber for evanescent coupling to microspheres is fabricated bypulling of the fiber on a motorized stage while softening its midsection in abutane/nitrous-oxide flame (Microtorch; Azure Moon Trading Corp.). Thetaper is typically 5 cm in length with a thinnest diameter 2 m for couplingat 760 nm wavelength. DFB lasers are purchased from Thorlabs (1,311 nmwavelength) and Eagleyard (763 nm wavelength) and optically isolated (iso-lators purchased from Thorlabs and OFR). A free-space coupler (Thorlabs)equipped with a 20x objective is used to couple to the SMF-28 fiber. ILXLightwave controllers are used to tune the laser diode by sweeping the currentwith a sawtooth-shaped function.

ACKNOWLEDGMENTS. This work was supported by a Rowland Junior Fellow-ship (to F.V.) and National Science Foundation Division of Bioengineering andEnvironmental Systems Grant 0522668 (S.A.).

1. P.W. Ewald (2002) In The Next Fifty Years, ed Brockman J (Vintage Books, New York),pp 289–301.

2. Arnold S, Khoshsima M, Teraoka I, Holler S, Vollmer F (2003) Shift of whispering gallerymodes in microspheres by protein adsorption. Opt Lett 28:272–274.

3. Arnold S, Ramjit R, Keng D, Kolchenko V, Teraoka I (2008) Microparticle photophysicsilluminates viral biosensing. Faraday Discuss 137:65–85.

4. Patolsky F, et al. (2004) Electrical detection of single viruses. PROC NATL ACAD SCI USA101:14017–14022.

5. Ignatovich FV, Novotny L (2006) Real-time and background-free detection ofnanoscale particles. Phys Rev Lett 96:0139011–0139014.

6. Vollmer F, et al. (2002) Protein detection by optical shift of a resonant microcavity. ApplPhys Lett 80:4057–4059.

7. Collot L, Lefevre-Seguin V, Brune M, Raimond JM, Haroche S (1993) Very high-Q whispering-gallery mode resonances observed on fused silica microspheres. Europhys Lett 25:327–334.

8. Lamb RA, Krug RM (2001) In Fundamental Virology, eds Knipe DM, Howley PM(Lippincott Williams & Wilkins, Philadelphia), pp 725–728.

9. Bresler SE, et al. (1975) Purification of influenza-viruses on wide-pore glass columns.Acta Virologica 19:190–196.

10. Vollmer F, Arnold S (2008) Whispering-gallery mode Biosensing: Label-free detectiondown to single molecules. Nat Methods 5:591–596.

11. Song BS, Noda S, Asano T, Akahane Y (2005) Ultra-high-Q photonic double-heterostructure nanocavity. Nat Mater 4:207–210.

12. Reimer CB, Baker RS, Newlin TE, Havens ML (1966) Influenza virus purification with thezonal ultracentrifuge. Science 3:1379–1381.

13. Armani AM, Kulkarni RP, Fraser SE, Flagan RC, Vahala KJ (2007) Label-free, single-molecule detection with optical microcavities. Science 317:783–787.

Table 1. Measurement of size and mass for different particles

Particle, radiusWavelength (),

nmCavity radius (R),

mMax. expt. step,

(/)max

Radius fromEq. 4 (a), nm

PS, 250 nm, 5% 1,311 44 5 107 211PS, 100 nm, 5% 763 30 2.2107 100InfA, SEM* 45–55 nm 763 39 1.5108 47

SEM, scanning electron microscope.

20704 www.pnas.orgcgidoi10.1073pnas.0808988106 Vollmer et al.

MicroParticle photophysics illuminates viral

bio-sensing

S. Arnold,* R. Ramjit, D. Keng, V. Kolchenko and I. Teraoka

Received 26th February 2007, Accepted 11th April 2007First published as an Advance Article on the web 17th July 2007DOI: 10.1039/b702920a

The authors present an approach for specific and rapid unlabeled detection

of a virus by using a microsphere-based whispering gallery mode sensor that

transduces the interaction of a whole virus with an anchored antibody.

They show theoretically that this sensor can detect a single virion below the

mass of HIV. A micro-fluidic device is presented that enables the

discrimination between viruses of similar size and shape.

Introduction

None of civilization’s socio-political catastrophes (e.g. world wars) have caused anequivalent destructive effect on the world’s population as biological pandemics.1

Exponentially growing pathogens are difficult to contain and eliminate unless theycan be detected early on. Some years ago, one of us (S.A.) reflected on this problemas a friend was dying from a viral infection. His friend’s diagnosis came too late;real-time methods for testing for the virus were not available. A decision was madeto direct the MicroParticle PhotoPhysics Lab toward finding a solution. This paperrepresents its first expose.Our approach is to sense bio-particles using the high sensitivity afforded

by the perturbation that an adsorbed molecule has on high Q (4107) opticalresonances of a microparticle.2 In particular, bio-particle adsorption will be sensedfrom the associated shift in resonance frequency.3,4 We are interested in the opticalspectroscopy of microparticles, but not in the spectroscopy of the nanoscopic bio-particles (e.g. protein, DNA, virus). Although optical spectroscopy can be particu-larly useful for small molecules, bio-particles such as protein are difficult todistinguish by optical spectroscopy, since they are considerably larger and sharecommon vibrational and electronic states, as a consequence of being made up of thesame 20 amino acids. The same can be said for distinguishing between strands ofDNA, since they have four nucleotides in common. Biology is a great teacher in thisrespect. Through all the eons of evolution, biology has not taken a spectroscopicapproach; there is no light in our bodies. Instead nature evolves bio-nano-probesthat specifically grab onto protein, DNA and foreign invaders throughphysio-chemical interactions. Our approach is to use these bio-nano-probes assurface-bound recognition elements and the microparticle to transduce (report)the interaction.We are not interested in labeling the bio-particle with markers (e.g. fluorescent

tags). Labels can structurally and functionally interfere with the assay, may not bespecific and block our goal of real-time detection by involving an additional step.5

MicroParticle PhotoPhysics Lab (MP3L), Polytechnic University, 6 MetroTechCenterBrooklyn, NY 11201, USA. Web: http://www.poly.edu/microparticle. E-mail:[email protected]; Fax: +1 718 260 3139; Tel: +1 917 568 6549

PAPER www.rsc.org/faraday_d | Faraday Discussions

Faraday Discuss., 2008, 137, 65–83 | 65This journal is c The Royal Society of Chemistry 2007

There are other unlabeled biosensors such as the Surface Plasmon Resonance (SPR)device, however it has limited sensitivity, which has not substantially improved fromits inception.6,7 Our ultimate goal is to demonstrate the unlabeled sensing of a singlebio-particle.We are also not interested in identifying a virus from a multi-step analysis of

geonomic information within its protein coat. Instead, we will seek to identify thewhole virus by tranducing the immobilization that takes place when a coat proteinon its surface interacts with a complementary antibody anchored to the micro-particle surface.

Resonant sensors-general considerations

Each and every oscillator, whether a mass on a spring, a violin string, the thorax of acricket or a Fabry–Perot cavity has the common property of resonance. If they aredriven by a harmonic source, the square of the oscillator’s amplitude |A|2

(i.e., energy) will demonstrate a Lorentzian-shaped frequency response with max-imum at or and linewidth g (Fig. 1). At the same time they can be sensitive toperturbations. Allowing dust to fall on a violin string causes its tone to be reduced by|Do|. One can imagine putting bio-nano-probes on the violin string in order to detectspecific dust (e.g. anthrax spore). However, one is apt to find in a world awash withnoise that the frequency shift may not be sufficient for real time detection. Theprincipal difficulty is in measuring a frequency shift much smaller than the linewidth(Fig. 1).We will characterize this competition by a ‘‘measurement acuity factor’’,

F = |Do|min/g where |Do|min is the smallest measurable frequency shift. Clearly,smaller F is better, but more difficult to achieve. One thing is certain: for a given Fthe minimum frequency shift that can be measured is proportional to the line width,so that the fractional minimum shift |Do|min/or = Fg/or. Whereas F and or arecontrolled principally by the non-dissipative physics of the oscillator and thebandwidth of the detection system, g is principally controlled by dissipation. Toreduce the minimum measurable shift, one can reduce dissipation. By convention wewill represent the linewidth-to-frequency ratio by 1/Q, where Q is the so-calledQuality factor; g/or = 1/Q. With this definition

|Do|min/or = F/Q. (1)

The larger Q, the smaller the dissipation and the smaller the dust particle that can bedetected. Since a mechanical system like a violin string agitates the fluid around it, itis a far from an ideal sensor; dissipation is assured. By the same token, aFabry–Perot cavity with metalized mirrors has dissipation due to Ohmic losses onreflection. The least dissipative reflection in this regard is Total Internal Reflection(TIR), in which light propagating in a medium with refractive index n1 is reflected ata sufficient angle y from a medium with a lower refractive index n2. TIR is, inprinciple, without loss. Light that bounces off the interior surface of a sphere

Fig. 1 The frequency response of an oscillator before and after a perturbation.

66 | Faraday Discuss., 2008, 137, 65–83 This journal is c The Royal Society of Chemistry 2007

while executing a polygonal orbit has this appeal. Such an orbit [Fig. 2, withy 4 sin1(n2/n1)] is known as a Whispering Gallery Mode (WGM).The WGM in Fig. 2 has no apparent loss. This implies an infinite Q. However, it is

well known that Q has limits. The largest Q for a WGMmode currently measured ina silica microsphere is B1010.8 Clearly, Fig. 2 is misleading, light does not havestationary states in a dielectric. Photons orbiting in the polygon are only partiallytrapped. They can ‘‘tunnel’’ into free space modes. Before providing a morecomplete description of a WGM, we will attempt to obtain a heuristic estimatefor the shift of its resonant frequency due to a perturbing layer.

WGM layer perturbation: heuristic approach

There is always a wave-particle duality associated with the photon as there is for theelectron. Instead of the description in Fig. 2 of a particle bouncing against theinterior of the microsphere, we now turn to a wave description for estimating thesensitivity of the microsphere resonance frequency to adsorption. Fig. 3(a) showsthis point of view. It represents the mode by a wave that circumnavigates near thesurface of a sphere of radius R and returns in phase.This picture, which is analogous to a Bohr–de Broglie atom, may be appropriately

called a Photonic Atom.9 By this analogy, the angular momentum of the photon ischaracterized by a quantum number l that is equal to the number of wavelengths inthe orbit. In what follows, this wave description will be used to obtain an estimatefor the minimum thickness of an adsorbed layer that is required to produce ameasurable shift of the mode’s frequency. Later, this quantum analog will beexpanded to determine the fields and probability densities associated with thesephotonic modes.Let us suppose that a material having the same dielectric constant as the

microsphere adsorbs on the sphere’s surface to a thickness t (Fig. 3(b)). This layercauses all lengths in Fig. 3(b) to be scaled up, while l for the given mode remainsinvariant. As a result the wavelength increases so that its fractional increase is the

Fig. 2 WGM within a dielectric microsphere.

Fig. 3 (a) Photonic AtomMode; (b) Anticipated wavelength change caused by the addition ofa spherically-symmetric layer.


same as the fractional increase in radius; Dl/l = t/R. Since frequency andwavelength are inversely related,

|Do|/or= t/R. (2)

The smallest measurable thickness is found by combining eqn (1) and (2),

tmin = F R/Q. (3)

For a conservative estimate for tmin, we take the measurement acuity F = 1,R = 100 mm and Q = 106, for which our minimum thickness is 0.1 nm (the sizeof a hydrogen atom). With a greater effort we have found experimentally that F canbe reduced to 1/50, corresponding, for our example, to a minimum detectablethickness t = 0.002 nm. This thickness is approximately one hundredth the size of ahydrogen atom! Although too small to be physical, it certainly shows the promise ofthe microsphere as an adsorption sensor.Our analysis thus far is strictly heuristic. Although it only applies to a

spherically-symmetric layer adsorbed on a homogeneous sphere and does not allowthe layer to have a different refractive index from the microsphere, it provides a greatdeal of guidance. Perhaps the most important rule, for our homogeneous sphericalsubstrate, is that the smallest amount of adsorbate is detected by minimizing theratio R/Q.Before we can address the problem of single bio-particle sensitivity, we must first

discuss a theory for the effect of local dielectric perturbations on microsphereresonances.This requires an understanding for the fields associated with the resonances. We

will approach this subject through a quantum analog using a pseudo-potentialapproach. Following the discussion of the theory for perturbation of microsphereresonances by bio-particles, we will present our virus experiments and show specificreal time detection of a virus from the frequency shift of a microcavity resonance, forthe first time.

WGM field: A quantum analog approach

Ultra-high Q WGM resonances were first seen in the microparticle area in opticallevitation measurements in air.10 Their interpretation came throughMie theory, which includes both incident and scattered fields.11 An easierapproach, pioneered by Nussenzvieg,12 utilizes a quantum analog and may be bestdescribed as a Photonic Atom (PA) model.9 Here, we briefly describe the model forcompleteness. For anyone who has studied the quantum mechanics of hydrogen thisapproach should be familiar. More importantly, it is concise and particularlyphysical.In the PA particle description, the photon is seen as a particle that orbits by

bouncing off the interior wall of the dielectric microsphere (Fig. 4a). This ‘‘quantumbilliard’’13 possesses quantized orbital angular momentum l just as the electron in aBohr atom.The PA wave description of a Transverse Electric (TE) mode is constructed by

inserting an electric field of the form

E = AL [cr (r) Yl,m/r] (4)

into the electromagnetic vector Helmholtz equation, where L is a dimensionlessangular momentum operator, L= ir r, Yl,m is a spherical harmonic and A is aconstant. As a result the radial wavefunction cr is found to follow a one dimensionalSchrodinger equation

d2cr

dr2þ ðEeff Veff Þcr ¼ 0; ð5Þ


where the effective energy Eeff = k02, the square of the free space wave vector and the

effective potential

Veff = k02[1 n(r)2] + l(l + 1)/r2. (6)

The first term in this potential describes dielectric confinement for a radiallyvariable refractive index n(r), while the second represents centrifugal repulsion.Fig. 4 connects up the particle (4a) and wave (4b) points of view for a homogeneoussphere.The analog particle with effective energy k0

2 is caught in a potential ‘‘pocket’’ inVeff as it oscillates radially between the classical turning points at rmin and R.However, since the particle is a quantum analog it can tunnel through the barrierextending from R to r2. Within the barrier the probability density falls. Thisnearly-exponential fall between r = R and r = r2 is known in electromagnetictheory as the region of the evanescent field (the analog probability density isproportional to the square modulus of the electromagnetic field, E* E). Anyprobability density left at the end of the barrier leads to energy loss in the form ofan outward radiating spherical wave. Consequently, unlike the electron in thehydrogen atom, confinement in a dielectric Photonic Atom cannot be complete;the intrinsic Quality factor is finite.The probability density illustrated for the mode in Fig. 4b has one peak in the

potential pocket and is known as a first order mode (n = 1). At higher energies,modes may form with more interior peaks, corresponding to higher order numbers.In all, four attributes are required to describe a mode: radial order number n, angularmomentum quantum number l, z-component of angular momentum m and polar-ization P (TE or TM). As in hydrogen,m can have integer values betweenl and l. Ageneralized mode will be labeled as Pv

l,m. For a sphere having a uniform refractiveindex, the radial wavefunctions cr in the interior and beyond are so called Riccati–Bessel functions;

cr = r jl(nsk0r) for r r R, (7)

cr = [jl(nsk0R)/hl(nmk0R)] rhl(nmk0r) for r Z R, (8)

where jl and hl are spherical Bessel and spherical Hankel functions and ns and nmare the refractive indices in the microsphere and the surroundingmedium. Mode energies are found, as in quantum mechanics, by matchingthe logarithmic derivatives of the Cr functions on either side of the interfacialboundary.

Fig. 4 The inter-relationship between the WGM model and the Photonic Atom Model for amicrosphere having a uniform refractive index.


Local dielectric perturbations of WGMs

Molecules that approach the surface of a microsphere interact with a WGM as theyenter the evanescent field. This oscillating field polarizes molecules, and as aconsequence causes a frequency shift of the mode.Fields due to photons are polarized due to the photon’s spin. If a microsphere is in

a single photon resonant state of energy ho, it will have an associated semi-classicalfield E(r, t) = Re[E0(r)e

iot]. With a bio-particle outside the sphere at position rj, aninteraction will occur; the bio-particle will be polarized and develop an oscillatingdipole moment in excess of the displaced solvent, Dp(t) = Re[Dp0e

iot]. The time-averaged energy required to polarize the bio-particle serves as the perturbation thatshifts the photon energy of the resonant state by

hDo ¼ 1

2hDpðtÞ E0ðrj; tÞit ¼

1

4Re½E0ðrjÞ Da# j E0ðrjÞ ð9Þ

where Da#

is the bio-particle’s excess polarizability tensor. By dividing this energy

shift by the energy of the mode

hor ¼ ð1=2ÞZ

eðrÞE0ðrÞ E0ðrÞdV ð10Þ

we arrive at a useful expression for the shift associated with a single bio-particleinteraction,

Door

j

¼ Re½E0ðrjÞ Da#j

E0ðrjÞ2ReðrÞE0ðrÞ E0ðrÞdV

ð11Þ

where e(r) is the dielectric function within the microparticle and in its surroundings(absent the bio-particle). Eqn (11) provides a much more general result than hadbeen presented previously.14 By simply switching from a real to imaginary operatorin eqn (11), one may also obtain an expression for the change in the linewidth Dg ofthe resonant mode due to molecular absorption,

Dgor

j

¼Im½E0ðrjÞ Da#j

E0ðrjÞ2ReðrÞE0ðrÞ E0ðrÞdV

ð12Þ

For our experiments, D can be assumed to have no imaginary part, since experimentswill be carried out at low enough energies to avoid absorption by protein or DNA.In this way, our measurements can be used to obtain added information about thesebio-particles.So long as one uses photon energies well below excited electronic states, water-

soluble proteins have dielectric properties that deviate less than 1% from one toanother. In addition, they also share very similar mass densities. On this basis thetrace of the polarizability tensor is proportional to the volume of the protein,15 andto the mass. This allows the Whispering Gallery Mode Biosensor (WGMB) to enjoya distinct advantage. The resonance shift contributed by a protein containsmolecular weight and size information. This is distinct from sensing schemes thatuse labels or detection involving protein charge.16,17 Another distinct advantage iscontained within the form of the interaction.Since the shift is proportional to E0ðrjÞ Da#j

E0ðrjÞ, a non-spherical proteinmolecule when adsorbed on a surface will show a different shift for a field polarizedperpendicular to the interface (TM polarization) than for the parallel case(TE polarization). This approach has recently been used for determining theorientation of the protein Bovine Serum Albumin (BSA) on a silica microspherein biological buffer.18

Many viruses that infect our cells are nearly spherical. For example HIV, HPVand HSV are icosahedral. This is fortunate since the excess polarizability tensor for


such a shape is essentially diagonal with identical elements (i.e. isotropic) and thebasic interaction between the field and the bio-particle at rj in eqn (11) may bewritten as

E0ðrjÞ Da#j E0ðrjÞ ¼ DajE0ðrjÞ E0ðrjÞ ð13Þ

On this basis, the shift due to a single bio-particle perturbation is

Door

j

¼ DajE0ðrjÞ E0ðrjÞ

2ReðrÞE0ðrÞ E0ðrÞdV

: ð14Þ

The shift caused by a large number of bio-particles is most easily computed byassuming that the field at a particular bio-particle is not influenced by thecontributions from its neighbors,19 so that eqn (14) can be summed over a numberdensity r(r),

Door

¼

RrðrÞDaðrÞE0ðrÞ E0ðrÞdV2ReðrÞE0ðrÞ E0ðrÞdV

: ð15Þ

By taking a more condensed view (i.e., r(r)c l3), the factor r(r) D a(r) E0(r)withinthe integrand in eqn (15) may be replaced by De (r) Ein (r), where De (r) and Ein (r) arethe excess permittivity and the ‘‘true’’ local field at r respectively. With thisMaxwellian approach, a truly continuum equation for the frequency shift perturba-tion evolves

Door

¼

RDeðrÞEinðrÞ E0ðrÞdV

2ReðrÞE0ðrÞ E0ðrÞdV

ð16Þ

This is a familiar result obtained from traditional perturbation theory for dielectricparticles within metallic cavities20 and near dielectric cavities.21 Our approach ofstarting from a molecular property is somewhat non-traditional, however, it leavesus with several useful results for perturbations that are discrete (eqn (11)) orcontinuous (eqn (15)).Eqn (15) may be applied to analyte molecules nearby in solution as well as those

adsorbed. The former, which is normally referred to as refractive index sensing, is oflittle interest to our current investigation.There are alternative ways to utilize the above theory in dealing with adsorbed

virus. If the adsorbed virus particle is much smaller than the evanescent field lengththen the field in its vicinity may be considered to be uniform, and one may expecteqn (14) to apply with |rj| = R. However, if the virus particle is comparable to orlarger than the evanescent field length then one must consider it to be in a non-uniform field. A model for its excess dielectric form may then be input for De (r) ineqn (16) and the numerator evaluated.Although the numerators in eqn (11), (14), (15) and (16) depend on the specific virus

being adsorbed and the evanescent field characteristics, the mode energy integral withinthe denominators in all of these equations are the same and may be simply estimated.In what follows, we will evaluate the shift due to a single bio-particle with radius a ladsorbed on a sphere resonating in an equatorial TE mode; TEv

l,l. This mode is ofparticular interest since it is possible to excite it selectively using an optical fiber.For a high Q resonance, the mode energy is dominated by the interior energy.22

Consequently, we will approximate the denominator of eqn (14) by limitingthe integration only to the interior. From eqn (4) and (7), the field in the interiorwithin an equatorial mode E0 = Ain jl (ns k0r) L Yll, and consequently eqn (14)becomes

Door

j

Daj½jlðnsk0RÞ2jLYll j2

2esR

jlðnsk0rÞ½ 2r2drR

LYll

2 sinðyÞdydf ð17Þ


for l c 1, |LYll|2p |Yll|

2.23 In addition, the spherical harmonic is normalized withrespect to solid angle,

HYllj j2dO ¼ 1, which reduces eqn (17) to

Door

j

Daj½jlðnsk0RÞ2jYll j2

2esR

jlðnsk0rÞ½ 2r2dr: ð18Þ

On resonance, the integral in the denominator of eqn (18) may be asymptotically(2pR/l c 1) related to the surface value of jl

2 through

Z R

0

½jlðnsk0rÞ2r2dr ffiR3

2½jlðnsk0RÞ2

n2s n2mn2s

: ð19Þ

where nm is the relative permittivity of the surrounding medium.24 By combiningeqn (19) with eqn (18), the shift due to a single bio-particle appears,

Door

j

DajjYllðrjÞj2

e0R3ðn2s n2mÞ: ð20Þ

Where e0 is the permittivity of free space (e0 = es/ns2), and the unit position vector rj

is used to denote the angular orientation at which the spherical harmonic isevaluated. This result does not depend on the resonance order, but possesses astrong dependence on polar angle and microsphere radius. In particular, thespherical harmonic in eqn (20) has a square modulus that peaks at (y = 901),producing a ‘‘band’’ of sensitivity around the equator (Fig. 5). With eqn (20) inhand, it is now possible to estimate the Limit Of Detection (LOD).We will first concentrate on this equatorial region, since particles adhering to this

region produce the largest possible shift. For a small number of virus particles N, wewill assume that each particle produces an equivalent shift. To estimate the smallestdetectable number NLOD we allow the accumulated shift |Do| using eqn (20) to beequal to |Do|min in eqn (1);

NLOD R3ðn2s n2mÞF

ðDa=e0ÞjYllðp=2Þj2Q: ð21Þ

NLOD is controlled by a number of factors. Since purified silica is readily available(i.e., optical fiber) and easily shaped, we choose silica for the microsphere. Weassume the virus to be blood borne, so the medium will be essentially aqueous. F isfixed by the measurement system and fluctuation theory, however, we find that a10 Hz bandwidth can be achieved with F = 1/50, and we will use this number in

Fig. 5 Depiction of the intensity associated with an equatorial mode (l = m).


subsequent calculations. Microsphere size plays a critical role. It explicitly affects theR3 factor in the numerator and implicitly affects the |Yll (p/2)|

2Q factor in thedenominator. At a fixed wavelength, the spherical harmonic factor increases withl and therefore increases with R. For large l, |Yll (p/2)|

2 B R1/2, so R3/|Yll (p/2)|2 B

R5/2. As for Q, the story is more complicated. Q is a measure of the rate at which amode’s energy decays (i.e. g = o/Q). The paths for decay are numerous, includingintrinsic loss rates due to tunneling (o/Qint), absorption (o/Qabs), Raman scattering(o/QRaman) and Rayleigh scattering (o/QRaleigh, including surface roughness); (o/Q)= o/Qint + o/Qabs + o/QRaman + o/QRaleigh. Surface scattering is expected tolimit the ultimate Q for a large silica sphere in vacuum (B1012).25 For ourexperiments in aqueous buffer solution, Q for spheres with radii from 20 mm to200 mm is limited to values orders of magnitude below the ultimate vacuum value,due to loss of dielectric contrast (i.e., increased tunneling) and aqueous absorption;

1

Q 1

Qintþ 1

Qabs: ð22Þ

As R increases above 20 mm, both tunneling and absorption decrease, causing 1/Q todecrease faster than R5/2. As a result, NLOD decreases with increasing R. Buttunneling decreases much more rapidly than absorption, causing it to have aminority influence as the size further increases. This causes NLOD to begin toincrease. Fig. 6 shows a theoretical plot for NLOD for HIV virus particles as afunction of the silica sphere radius for two wavelength regions, 1300 nm and 780 nm.The substantially smaller NLOD values in the 780 nm region are due to considerablysmaller absorption by water at this wavelength. At either wavelength, we predictvery high sensitivity, with NLOD dipping below one tenth at 780 nm for a micro-sphere approximately 40 mm in radius.Fig. 6 was constructed from an estimate of the HIV polarizability. Although the

excess polarizability of HIV has not been measured, it is not difficult to estimate Dafrom the virus’ mass. Viruses are principally composed of protein and protein havethe convenient property of raising the refractive index of an aqueous solution bynearly the same amount for the same mass concentration; the differential refractiveindex of protein in an aqueous buffer dn/dc E 0.18 cm3 g1 (visible).26 Tobacco

Fig. 6 Smallest theoretical detectable number of HIV virus particles adsorbed on theequatorial rim of a homogeneous silica WGMB as a function of microsphere radius R forF = 1/50 and wavelength near 1300 nm or 780 nm.


Mosaic Virus (TMV) is one of the few examples for which dn/dc has been measuredand the correspondence with protein verified (l= 488 nm).27 Although dispersion isprojected to lower dn/dc by 0.006 between 780 nm and 1300 nm,28 we have neglectedthis difference and used the visible value for our calculations. The excess polariz-ability is related to dn/dc and the molecular mass m through

Dae0¼ 2nm

dn

dcm: ð23Þ

HIV is a fusion virus with a lipid bilayer surrounding its capsid. From the densityand average size of the viral particle (110 nm), we estimate its mass to be 8 1016 gm and Da/e0, from eqn (23), to be 4 104 mm3.Although we will capture virus particles from a fluid flow, we are currently not at

the point of having them deposit only at the micro-sphere equator. Instead, theentire spherical surface will be functionalized. Here, the more relevant question is thetheoretical limit of detection associated with surface mass density, sm,Lod, which isderived by adding individual frequency shifts (eqn 20) from bio-particles placed atrandom positions on the surface and setting this overall shift to the minimummeasurable shift (eqn (1)). Before calculating sm,Lod we will first obtain the shift dueto a uniform layer of bio-particles.The frequency shift due to a random layer of bioparticles is found by summing

eqn (20) over random position coordinates on the microsphere surface. This discretesum may be made continuous by defining a number of adsorbates per unit solidangle dN/dO = s4pR2/4p and integrating over solid angle,

Doj jor¼ Da s

e0ðn2s n2mÞR: ð24Þ

The result has the same 1/R dependence that we heuristically obtained in eqn (2).The two equations may be further related by describing the bioparticles as having arefractive index nbp and forming a layer of thickness t in which the protein volumefraction is f. With an effective medium approach Da s/e0 = f(nbp

2 nm2)t, and

consequently

Doj jor¼

f ðn2bp n2mÞðn2s n2mÞ

t

R: ð25Þ

A uniform silica layer on silica corresponds to f= 1 and nbp = ns for which eqn (25)agrees with eqn (2). We can now calculate the limit of detection for surface massdensity by setting s= sm,Lod/m and in eqn (24). The results for 760 nm and 1300 nmare shown in Fig. 7.It should be noted that we have indicated the typical baseline sensitivity levels for

other label-free sensing techniques in Fig. 7. Interestingly, the sensitivity for theWGMB at 1300 nm for a relatively large microsphere (200 mm) is potentially betterthan the typical baseline sensitivity for the Quartz Crystal Microbalance (QCM),SPR,7 or Micro-Cantilevers (MC),29 and the minimum on the 780 nm curve beats thebest of these by more than two orders of magnitude.In what follows, we will describe our experimental approach and present our

results on specific virus sensing.

Experimental approach

There are two essential parts to our sensor design: the WGM resonator and thedriver that stimulates it.The WGM resonator is formed from silica by rotating the end of a bare

telecommunication fiber in an oxygen–propane micro-flame. The end of the glasssoftens and flows into a spheroidal shape. Our spheroids are oblate with aneccentricity B6% and equatorial radii between 75 and 200 mm.


The use of spheroids may come as a surprise, since the theory in the last sectionwas developed for spheres. However, our theory applies equally well to modes withorbits near the equator of a spheroid so long as the perturbation does not change theeccentricity.The resonator is driven by evanescently coupling it at its equator to a tapered

optical fiber wave guide. Excitation of a micro-spheroid mode is signalled by a dip inthe transmission through the fiber.30 The fiber is flame tapered adiabatically from adiameter of 125 mm to a skirt width of B2 mm using an asymmetric pulling device,31

and directed perpendicular to the spheroid’s axis of symmetry (z, Fig. 8a). In thisconfiguration, energy is most efficiently coupled into the equatorial mode (m = 1).Other modes with orbits tilted from the equatorial plane have |m| values less than l,and as a consequence of the spheroidal shape differ in circumference and resonancefrequency (Fig. 8b).

Fig. 7 Smallest theoretical detectable uniform mass density on a silica WGMB as a function ofmicrosphere radius R for F = 1/50 and wavelength near 1300 nm or 780 nm.

Fig. 8 (a) The optical configuration. (b) A recorded transmission spectrum in water for a silicaspheroid having a 200 mm equatorial radius. Adjacent resonances differ by |Dm| = 1.


The fiber transmission spectrum taken in water in Fig. 8b has a complete rangecomparable to the resolution of a good fluorescence spectrometer (0.2 nm), however,for our experiments the resolution is much higher. It is ultimately controlled by thelinewidth of a Distributed Feedback Laser (DFB, o0.00001 nm) operating near1312 nm. This DFB laser is tuned directly from its power supply, by driving the laserwith a saw tooth current source. As the forward current increases, both the outputpower and wavelength of the laser increase. The lower fiber transmission spectrum inFig. 8b reflects this tuning approach with the resonant dips dropping from a sawtooth backbone. The upper curve is constructed by normalizing to constantintensity. Adjacent resonances differ by |Dm| = 1.The perturbation in the frequency/wavelength of a resonance in a spectrum such

as Fig. 8b is tracked by the movement of a resonance dip by the use of a five pointparabolic fit and frequency/wavelength positions are recorded every 200 ms as a ‘‘diptrace’’. For a resonance having a Q B106, the fluctuations in the base line of the diptrace have an rms value as small as 0.00002 nm for a time period of 5 s.Two fluidic configurations were chosen. For non-specific binding experiments the

sphere and fiber are immersed in a 1 cm3 volume containing a stirred aqueous buffersolution. For specific binding experiments it is necessary to look at both adsorptionand desorption rates. For this purpose, a much smaller micro-fluidic cell with amotorized fluid exchange system was developed. This device will be described inmore detail along with our specific binding experiments.We will next focus on surface preparation for non-specific adsorption experiments

and attempt to detect virus particles in this way. Following this, we will turn to thepreparation of a surface for specific adsorption and demonstrate specific detectionthough the discrimination between similar viruses.

Surface modificatoin: non-specific sensing

Our interest is in functionalizing a silica surface for either non-specific or specificsensing. In all cases reported in this paper, we start by cleaning the surface in anoxygen plasma. In the presence of background humidity, silanol groups (Si–OH)form at the surface. Our surfaces will be functionalized by reacting the protrudinghydroxyl groups covalently with the methoxy/ethoxy components of linking com-pounds known as silanes.The silane compound chosen for non-specific sensing is one that leaves the silica

surface with an opposite charge to the net charge on the adsorbate. Most proteinsacquire a negative charge at a biological pH (B7.4). To produce a positively chargedsurface we react the protruding hydroxyl groups with 3-aminopropyltriethoxysilane(APTES), since its amino terminus is protanated to NH3

+ as a consequence of thepKa of NH2 (B8). APTES is deposited from the vapor phase in a partial vacuum.32

After bringing the vacuum chamber to atmospheric pressure, the microspheres aretransferred to an oven at 120 1C, where the silane rug is cured (i.e. cross-linked). Thechemical progression is shown in Fig. 9.

Non-specific sensing of virus

For a preliminary test of our sensor, we non-specifically adsorbed BSA on itssurface. At sufficient concentration (4100 nM) it produced a Langmuir-likesaturation with a frequency shift consistent with eqn (24), based on the molecularweight of BSA (66.4 103 Da) and a maximum surface density built up by randomsequential adsorption.33 Although BSA is relatively small in comparison with a virusparticle, similar experiments using carboxylated polystyrene beads with diameters of100 nm (mass B3.5 108 Da) gave good agreement with eqn (24).For virus adsorption, we chose a virus that kills E. coli but is friendly to humans as

a model system for both non-specific and specific sensing. The RNA virus known asMS2 is icosahedral in shape, as is HIV, although it has a far smaller size with a


diameter of only 23.6 nm and a mass of 3.6 106 Da. MS2 phage has a pI of 3.9,34

making it suitable at pH 7.4 to take on a negative charge.Virus stock was obtained from American Type Culture Collection (ATCC,

Manassas, VA). MS2 phage was propagated in E. coli K91 host and incubated onLB plates (courtesy of N.L. Goddard, Harvard University). After filtration, the viruswas resuspended in PBS and stored at 4 1C.Our experiment is begun by injecting 30 mL of MS2 solution [3 109 pfu (plaque

forming units)] into 1 mL of PBS buffer (pH = 7.4) in the sample cell of our non-specific sensing system. As a micro-stirring bar homogenizes the solution to aconcentration of 5 pM, all resonance dips shift toward longer wavelengths. Thedip trace represented in Fig. 10 by plotting the normalized shift RDl/l, shows theadsorption of virus in real time and is indicative of monolayer formation. We cantrack the density of adsorbed virus by inverting eqn (24),

s ¼ RDllðn2s n2mÞDa=e0

: ð26Þ

With appropriate values for the refractive indices of silica (ns = 1.452) and the buffer(ns = 1.32), and by calculating Da/e0 from eqn (23) (3.7 1018 cm3), we find thatthe surface density associated with the equilibrium shift in Fig. 10 is se = 2.6 1010

cm2. This layer is far from compact. In our previous experiments with protein,surface densities approached the maximum density for random sequential adsorp-tion, srsa,max.

33 This density is 55% of the inverse ‘‘foot print’’ area (srsa,max = 0.55sfp = 0.55 /(pa2) = 1.2 1011 cm2). The reason for the small surface density atequilibrium in Fig. 11 lies in the balance between adsorption and desorption rates.Although there was not enough incubated virus to make runs at higher concentra-tions, measurements made at lower concentration reveal that the equilibriumwavelength shift is in proportion to concentration [v] up to 5 pM (insert, Fig. 10).This may be understood from Langmuir’s isotherm. Here, the fraction of maximumsurface coverage at equilibrium ye = ka [v]/(ka [v] + kd), where ka is the adsorptionrate constant and kd is the desorption rate. For our equilibrium shift to beproportional to [v], ka[v] must be considerably less than kd, so that the Langmuirisotherm has the approximate form ye = ka[v]/kd. By making the reasonable

Fig. 9 Amino-silanization using APTES on cleaned microsphere support.


assumption that ye C se/srsa,max, we estimate an upper limit for the equilibriumconstant Ke = ka/kd B1011 M1.So far, we have shown that we can detect the binding of virus particles to a

microsphere from the shift in wavelength of a whispering gallery mode and extractthe associated equilibrium constant. Our ultimate goal, specific sensing, will requirethe ability to follow the desorption process. This requires a microfluidic system andthe design of a surface that can be specific to MS2.Before discussing our microfluidic system, it should be pointed out that the mass

density sensitivity associated with the rms baseline signal in Fig. 10 is consistent withour calculation of sm,LOD in Fig. 7 for R = 200 mm; 0.2 ng cm2.

Micro-fluidic system for specific sensing

Specific sensing will be tested through the difference in desorption rates associatedwith different virus on a surface functionalized with antibody. This can reasonablybe done by utilizing a microfluidic flow system that incorporates the microsphereand a tapered optical fiber.Our specific sensing system is shown in Fig. 11. It consists of a tapered fiber bound

by UV adhesive to a glass slide as a foundation. Below this foundation is athermoelectric stage and above it is a polymer cap with multiple entry points. Thistransparent cap, formed by moulding poly-dimethyl siloxane (PDMS) in amicromachined master, is accesible to fluids, the microsphere, laser excitation andan optical detection port. The cross section of the microfluidic channel is 1.5 mm 2.5 mm and has a total volume of 100 mL.A fluidic system was designed to handle the various solutions used in our

experiments. This system was supplied by two syringe pumps that are directed bya custom LabVIEW driver program. Each pump is fitted with a 25 mL syringehaving 24000 steps of total travel (1 mL per step). The step rate is controlled by thedriver that delivers liquid over a range from 1 step per 20 d to 100 steps per 1 s. Atypical experiment is run at 38 mL min1. One pump is dedicated to the buffer usedfor washing in between each sample injection. The other is the sample pump thatpushes the contents of a 700 mL teflon sample loop into the fluidic channel. The loop

Fig. 10 Dip trace for non-specific MS2 adsorption for a 5 pM concentration. The insert showsthe equilibrium shift isotherm below 5 pM.


is washed with buffer between each sample injection, ensuring that the syringe is freeof cross-sample contamination.

Surface modification: specific virus detection

Anti-MS2 (in PBS, pH 7.4, Tetracore) antibodies were covalently linked to themicrosphere using amine–carboxyl coupling chemistry (Fig. 12). Briefly, sphereswere cleaned by oxygen plasma for 4 min. Immediately after, spheres were immersedin 2% 3-(Triethoxysilyl)propylsuccinic anhydride in ethanol/acetic acid for 2 minthen rinsed with ethanol to yield anhydride groups on the silica. Spheres were driedat 115 1C for 20 min. After this, a sphere was coupled to fiber inside the flowcell andslightly basic buffer (pH 8.3) was introduced into the cell, which hydrolyzes theanhydride to create a carboxyl rug across the sphere’s surface (Fig. 13).The sphere was allowed to incubate in the flow cell for 45 min. The rest of the

assembly is followed by the dip trace in Fig. 13. Next, activation of carboxyl groupsby EDC/NHS provided for a semi-stable amine, highly-reactive ester surface. 200 mLof 0.4 M 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC) was mixed with

Fig. 11 Microfluidic system for sensing.

Fig. 12 Specific surface chemistry.


200 mL 0.1 MN-hydroxysuccinimide (NHS) and injected into the flow channel. NHSesters react with amine groups on the antibody to form covalent links. Underconstant flow, the surface was allowed to react for 5 minutes. After rinsing with PBS,MS-2 antibodies were injected at a concentration of 620 nM and allowed to incubatefor 60 min with flow off. From the overall shift of 5 nm and the molecular weight ofthe antibody (155 000 Da), the surface density was estimated from eqn (26) to be1.1 1012 cm2, about 5 antibodies within an MS2 footprint. Unbound antibodieswere then washed away and the surface was blocked with 1 M ethanolamine (pH8.5). Remaining antibodies are covalently bound to the surface. After a thoroughrinse, surfaces were ready for specific virus detection. All steps involving flow wereperformed at a flow rate of 38 mL min1. The setup and microsphere surface are nowready for virus specificity detection.

Specific virus detection

In order to discriminate between different viruses, we designed a cycling scheme thatincluded another E. coli virus, Phix174, as a negative control. Phix174 is a DNAvirus of icosahedral shape and has a mass and diameter of 6.2 106 Da35 and28.4 nm, respectively. Its coat proteins contain epitopes different from the MS2virus, and therefore should not exhibit specific binding to our anchored anti-MS2.The experiment proceeds as follows. At a flow rate of 38 ml min1, 500 ml of MS2

phage at a final concentration of 5 pM was flowed through the microfluidic channeland then the pump was turned off. MS2 was allowed to bind to the surface for 5 min.We observed that all resonant dips shifted to longer wavelengths, indicatinginteraction between the modified surface and virus (top trace, Fig. 14). Just after,PBS was flowed into the channel for 10 minutes. We observed a resonance frequencyshift towards smaller wavelength, indicating that weakly-bound virus was present onour surface. However the signal did not return to the value of pre-injected MS2. Theresidual surface bound MS2 after this wash was calculated (eqn (26)) to be sv = 9 109 cm2. Once again, this surface density is far less than the maximum density forrandom sequential adsorption, even though there are an average of five antibodieswithin an MS2 footprint. Under the supposition that few antibodies are in a positionand orientation to be viable, a decision was made to lower the concentration and

Fig. 13 Dip trace following the preparation of a silica microsphere with antibody.


repeat the experiment. This required an antibody-preserving regeneration step,which will be discussed later. The lower trace in Fig. 14 shows this second experimenton the same microsphere, at half the previous MS2 concentration. It should be notedthat although the buffer wash caused virus to be removed at the higher concentrationrun, at 2.5 pM there was no evidence of desorption. We conclude that oursupposition is correct; only a small fraction of the antibodies are viable for specificadsorption.The very useful regeneration step previously noted deserves some attention at this

point. It is accomplished by introducing 10 mM glycine into the flow channel(pH 2.0) and allowing the microsphere to bathe in this solution for 5 min. Theglycine solution competitively disrupts the electrostatic portion of theantibody–antigen interaction.36 During this time, the residual signal associated withspecifically bound virus is eliminated without eliminating the covalently boundantibody. To complete the regeneration step, PSB is flowed into the channel forB10 min. At this point, the microsphere is ready for reuse with another virus(e.g. the control Phix174).With the same microsphere in place, we next tested our sensor against Phix174

(Fig. 15). Phix174 was injected into the cell at 5 pM and allowed to bind as above.After injection, the frequency shifted toward longer wavelength, reachingequilibrium before the pump was turned off. About 100 seconds later PBS waspumped into the channel and the wavelength dropped toward the baseline, indicat-ing that Phix174 non-specifically binds to our surface. The MS2 trace from Fig. 14 isreproduced in Fig. 15 for comparison. This figure clearly shows that the WGMBtechnique can discriminate between the non-specific binding by Phix174 and specificbinding by MS2. The experiments were repeated several times with similar results.

Conclusions

By using perturbation ideas from microparticle photophysics, combined withbiochemical surface functionalization and microfluidics, we have found a meansfor specifically sensing a virus. The axial symmetry associated with a slightlyperturbed sphere allows for simple analytical equations that connect up wavelengthshifts with the polarizability and surface density of the adsorbed virus. By using

Fig. 14 MS2 virus experiment at 5 pM (upper) and after regeneration at 2.5 pM (lower).


chemical regeneration, our microparticle-inspired real time sensing approach allowsus to follow virtually all aspects of the surface modification and viral sensing on thesame sphere in situ.Although our calculations looked in part at the question of single virion sensing,

our experiments concentrated on the demonstration of specific detection usinghigher surface densities. This is not a limitation of baseline noise. Rather, theproblem involves not being able currently to spatially localize adsorption solely tothe equator of the sphere. We are currently designing a photolythographic techniquethat will allow us to place antibodies only along the equator. Once implemented, ashorter wavelength DFB laser will be employed in order to test our single virioncalculations in Fig. 7. We expect the dip trace to contain a randomly-spacedstaircase. Based on the agreement between our surface saturation measurementsand theory (see the section titled ’’Non-specific sensing of virus’’), the height of eachstep should reveal the molecular weight of the adsorbate.We have merely scratched the surface in our use of the WGMB technique. Within

a flow cell there can be a multitude of resonant microcavities each functionalizedwith different antibodies for different viruses. One can do this while multiplexingboth the fiber and fluid channel. Other cross configurations will surely be imagined.

Acknowledgements

This research was supported by NSF through the Division of Bioengineering andEnvironmental Systems, Grant No. 0522668. We thank N. L. Goddard, HarvardUniversity, for incubating the MS2 virus used in our experiments.

References

1 P. W. Ewald, ‘‘Mastering Disease’’, in The Next Fifty Years, ed. John Brockman, VintageBooks, 2002, pp. 289–301.

2 A. Serpenguzel, S. Arnold and G. Griffel, Opt. Lett., 1995, 20, 654–656.3 S. Arnold, Am. Sci., 2001, 89, 414–421.4 F. Vollmer, D. Braun, A. Libchaber, M. Khoshsima, I. Teraoka and S. Arnold, Appl.

Phys. Lett., 2002, 80, 4057–4059.5 R. Rich and D. Myszka, J. Mol. Recognit., 2005, 18, 1–39.

Fig. 15 Comparison between Phix174 and MS2 experiments on the same sphere.


6 U. Jonsson, L. Fagerstam, B. Ivarsson, B. Johnsson, R. Karlsson, K. Lundh, S. Lofas, B.Persson, H. Roos and I. Ronnberg, BioTechniques, 1991, 11, 620.

7 R. Karlsson and R. Stahlberg, Anal. Biochem., 1995, 228, 274–280.8 M. L. Gorodetsky, A. A. Savchenkov and V. S. Ilchenko, Opt. Lett., 1996, 21, 453–455.9 S. Arnold, J. Camunale, W. B. Whitten, J. M. Ramsey and K. A. Fuller, J. Opt. Soc. Am.

B, 1992, 9, 819–824.10 A. Ashkin and J. M. Dziedzic, Phys. Rev. Lett., 1997, 38, 1351.11 P. Chylek, J. T. Kiehl and M. K. W. Ko, Phys. Rev., 1978, A18, 2229.12 H. M. Nussenzveig, Comments At. Mol. Phys., 1989, 23, 175.13 H. G. L. Schwefel, H. E. Tureci, A. Douglas Stone and R. K. Chang, in Optical Processes

in Microcavities, ‘‘Progress in Asymmetric Resonant Cavities: Using shape as a designparameter in dielectric microcavity laser’’, ed. R. K. Chang and A. J. Campillo, WorldScientific, Singapore, 1996.

14 S. Arnold, M. Khoshsima, I. Teraoka, S. Holler and F. Vollmer, Opt. Lett., 2003, 28,272–274.

15 F. Frolich, Theory of Dielectrics, Oxford University Press, London, 1958, p. 28.16 W. U. Wang, C. Chen, K. Lin, Y. Fang and C. M. Lieber, Proc. Natl. Acad. Sci. U. S. A.,

2005, 102, 3208.17 F. Patolsky, G. Zheng, O. Hayden, M. Lakadamyali, X. Zhuang and C. M. Lieber, Proc.

Natl. Acad. Sci. U. S. A., 2004, 101, 14017.18 M. Noto, D. Keng, I. Teraoka and S. Arnold, Biophys. J., 2007, 92, 4466–4472.19 I. Teraoka and S. Arnold, J. Appl. Phys., 2007, 101, 0235051.20 R. P. Harrington, Time-Harmonic Electromagnetic Fields, Wiley-IEEE Press, 2001.21 I. Teraoka and S. Arnold, J. Opt. Soc. Am. B, 2006, 23, 1381–1389.22 D. Q. Chowdhury, S. C. Hill and M. M. Mazumder, IEEE J. Quantum Electron., 1993, 29,

2553.23 J. D. Jackson, Classical Electrodynamics, Wiley, New York, 1962, 753.24 M. Khoshsima, ‘‘Perturbation of Whispering Gallery Modes in Microspheres by Protein

Adsorption: Theory and Experiment’’, PhD thesis, Polytechnic University, New York,2004.

25 M. L. Gorodetsky, A. D. Pryamikov and V. S. Ilchenko, J. Opt. Soc. Am. B, 2000, 17,1051–1057.

26 S. H. Armstrong, Jr, M. J. E. Budka, K. C. Morrison and M. Hasson, J. Am. Chem. Soc.,1947, 69, 1747.

27 H. J. Coles, B. R. Jennings and V. J. Morris, Phys. Med. Biol., 1975, 20, 310.28 C. E. Alupoaei, J. A. Olivares and L. H. Garcia-Rubio, Biosens. Bioelectron., 2004, 19,

893.29 T. P. Burg and S. R. Manalis, Appl. Phys. Lett., 2003, 83, 2698.30 G. Griffel, S. Arnold and A. Serpenguzel, Opt. Lett., 1996, 21, 695–697.31 D. Keng, ‘‘Asymmetric fiber puller for biosensor experiments’’, in preparation, 2007.32 K. H. Choi, J. P. Bouroin, S. Auvray, D. Esteve, G. S. Duesberg, S. Roth and

M. Burghard, Surf. Sci., 2000, 462, 195–200.33 M. Noto, M. Khoshsima, D. Keng, I. Teraoka, V. Kolchenko and S. Arnold, Appl. Phys.

Lett., 2005, 87, 223901.34 L. R. Overby, G. H. Barlow, R. H. Doi, M. Jacob and S. Spiegelman, J. Mol. Biol., 1966,

91, 442–448.35 A. B. Burgess, Proc. Natl. Acad. Sci. U. S. A., 1969, 64, 613–617.36 D. Ilchmann, D. Helbig, H. Gohler, M. Stopsack, H. J. Thiele andW. Hubl, J. Clin. Chem.

Clin. Biochem., 1990, 28, 677–681.


Resonance fluctuations of a whispering gallery mode biosensorby particles undergoing Brownian motion

D. Keng, S. R. McAnanama, I. Teraoka, and S. Arnolda

MicroParticle PhotoPhysics Laboratory, Polytechnic University, Brooklyn, New York 11201

Received 31 July 2007; accepted 8 August 2007; published online 4 September 2007

Nanoparticles suspended in the vicinity of a whispering gallery mode WGM biosensor aredetected from fluctuations in the driving light-guide transmission. These fluctuations are describedby Brownian particles perturbing the resonance wavelength in reaction to being polarized by theWGM’s evanescent field. Comparison between the autocorrelation of the measured fluctuations andtheory provides a first order approximation for the nanoparticle size and lays the basis for futurestudies of interfacial dynamics. With this advance, the WGM biosensor goes beyond low-frequencymeasurements of adsorption and desorption and into a world which has been dominated byfluorescence correlation spectroscopy, but without labels. © 2007 American Institute of Physics.DOI: 10.1063/1.2778351

There is an extant revolution in label-free biosensingbrought on by the use of whispering gallery mode WGMresonances. Nanoparticles in the vicinity of a resonator sen-sitively perturb its characteristics e.g., resonance wave-length. Already, resonance wavelength shift data have en-abled the label-free specific sensing of protein,1,2 DNA,3 andvirus,4 as well as the characterization of nanolayer growth5

and solvent refractive index changes.6,7 These sensing appli-cations have been implemented by simply measuring the dccomponent of the resonance wavelength shift signal. Herein,we show that the broadband “noise” on this signal providesphysical information not available from its dc component.Stochastic effects associated with molecules undergoingBrownian motion near the resonator surface are responsiblefor this noise. This observation allows this all-photonic sen-sor to enter a world that has been dominated by total internalreflection fluorescence correlation spectroscopy TIR-FCS,8

without the need for fluorescent labels. In what follows, wecharacterize the noise in resonance wavelength fluctuationsof a spherical microcavity bathed in a solution containingnanoparticles of viral size, and demonstrate that nanoparticlesizes may be estimated from the analysis of this noise.

Let us consider the situation depicted in Fig. 1a. Ananoparticle diameter d1–200 nm in an aqueous solu-tion just outside a spherical microresonator radius R50–500 m diffuses through the evanescent field of aWGM. The WGM is driven by coupling the microsphereevanescently to light guided through an optical fiber.9 Eachnanoparticle diffusing within the evanescent field near theequator of the microsphere influences the resonance wave-length as a result of polarization by the field.10 As the freespace wavelength of the laser driving the fiber is sweptacross the resonance, the WGM is revealed as a dip in thetransmitted intensity T through the fiber11 Fig. 1b. Theresonance wavelength at the center of this Lorentzian dip, r,fluctuates as a result of the interaction with this particle andothers in the surrounding solution. Temporal changes of theresonant wavelength are transduced and amplified by posi-tioning the laser wavelength at bias on one side at the maxi-

mum slope of the dip and recording the intensity, as shown inFig. 1b.

In previous studies, we followed the slow variation ofthe resonance dip position using a scanning approach torecord the fiber transmission spectrum.1 In this way, the dipposition was updated every 0.1 s. With the bias scheme inFig. 1, we can follow variations in the resonance wavelengthat a much faster pace, limited only by our data acquisitionsystem, which samples the transmitted intensity with 16 bitprecision at 200 kHz. The noise with Brownian particlespresent in the surrounding solution will be different in com-parison to the neat buffer, as shown in Figs. 1c and 1d.Since the noise is stochastic, it will be characterized by itsautocorrelation function ACF. Below, we describe someexperimental details.

The WGM biosensor consisted of a silica microsphere200 m radius coupled to a phase-matched silica optical

aElectronic mail: [email protected]

FIG. 1. Color online Measurement principle for transducing WGM reso-nance fluctuations: a Brownian particle passing through the evanescentfield of a WGM in a microsphere coupled to a fiber. b Resulting transmis-sion showing a fluctuating WGM resonance dip. c and d Recordedintensity traces at bias as a function of time t with d=37 nm and withoutBrownian particles, respectively.

APPLIED PHYSICS LETTERS 91, 103902 2007

0003-6951/2007/9110/103902/3/$23.00 © 2007 American Institute of Physics91, 103902-1Downloaded 04 Jan 2008 to 128.122.149.42. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp

http://dx.doi.org/10.1063/1.2778351

http://dx.doi.org/10.1063/1.2778351

http://dx.doi.org/10.1063/1.2778351

fiber12 within a microfluidic cell molded from silicone, asshown in Fig. 2. A 1300 nm pigtailed distributed feedbacklaser was connected to the input end of the fiber, and itswavelength was controlled through the drive current.11

Noise associated with nanoparticles was first discernedwhile sensing bacterial virus. To understand the phenom-enon, we chose polystyrene particles having a carboxylatedsurface and mean diameters of 37, 103, and 219 nm Poly-sciences, Inc., as measured by dynamic light scatteringN4Plus, Coulter, as viral simulants.

After recording resonance fluctuations using a filteredphosphate buffered saline PBS solution pH=7.4 for 1 s,the ACF of the signal was computed. This procedure wasrepeated 50 times with each new ACF added to the previousones in order to form an average. Following this, particles ofa given size suspended in PBS were injected into the mi-crofluidic channel Fig. 2 using a digitally controlled sy-ringe pump, and the resonance wavelength was seen to shifttoward a longer wavelength, indicative of adsorption. Whenthe wavelength shift stabilized, the ACF of the signal wascalculated as in the case of the buffer. Then, the microfluidicchannel was washed with the buffer solution using a seconddigitally controlled pump Fig. 2. As the particles were re-moved from the solution surrounding the microsphere, theresonance wavelength changed only slightly, indicative ofstrong nonspecific binding. The rms value of the resonancewavelength fluctuations essentially returned to its value be-fore the particles were injected into the channel, indicatingthat the observed fluctuations were principally associatedwith particles diffusing in solution. Experiments for eachnew particle size utilized a new microsphere, but otherwisefollowed the same procedure.

The initial decay in the normalized ACF, c /c0, forthe three particle sizes is shown in Fig. 3 as a function ofdelay time for the first 1 ms. Each ACF is seen to falllinearly with time following the empirical equationc /c0=1− for 1 The inset shows that the decayrate is approximately proportional to the inverse diameterof the particles: /d, with =13.1103 nm/s.

The results in Fig. 3 may be understood from the Brown-ian motion of particles in the evanescent field near the sur-face of the microsphere. A given particle at position r causes

a resonance wavelength shift r as a reaction to beingpolarized by the evanescent field Er , t.10 The shift is pro-portional to the scalar product of the induced dipole and theelectric field, and is consequently proportional to the squaremodulus of the field Er2. The accumulated resonancewavelength shift from many particles is most easily ac-counted for by utilizing a number density r , t and per-forming a simple sum,

rt r,tEr2dV . 1

Before evaluating Eq. 1 it is useful to consider the charac-teristic lengths of our system. The “ribbon” of intensity nearthe equator Fig. 1 has three characteristic lengths: the cir-cumference 600 m, the width along a meridian 3m, and the evanescent intensity length L0.2 m. Con-sidering that the time to diffuse through a given length isproportional to its square, the short time behavior demon-strated in Fig. 3 should be overwhelmingly controlled by L.This allows us to pare down the integral in Eq. 1 to essen-tially one dimension,

FIG. 2. Color online Microfluidic system incorporating a microsphere resonator and a coupling fiber.

FIG. 3. Color online Normalized autocorrelation of resonance wavelengthfluctuations for several particle diameters d. Each follows c /c0=1− ,for 1 Inset: plot of vs 1/d.

103902-2 Keng et al. Appl. Phys. Lett. 91, 103902 2007

Downloaded 04 Jan 2008 to 128.122.149.42. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp

rt ,texp− /Ld , 2

where is the distance from the surface. The normalizedACF is then

cc0

=rr0

r02

= 0

d0

d,,0exp− + /L .

3

Fortunately, this integral has been worked out for a self-similar problem in TIR-FCS.8 It is the problem of fluores-cence fluctuations from particles diffusing in the evanescentfield at the back of a prism. The solution involves derivingthe density autocorrelation , ,0 from the diffu-sion equation under a prescribed boundary condition, andcarrying out the integration. For a reflecting boundary, con-sistent with having saturated the surface with particles hav-ing the same charge as those diffusing, the solution is13

cc0

= 1 − 2ReexpReerfcRe1/2 + 2Re

1/2

,

4

where Re=D /L2, with D being the diffusion coefficient. Forshort times 1/Re, c /c0=1−Re allowing us toidentify our experimental with Re. The approximate pro-portionality found between and 1/d would also be ex-pected based on bulk diffusion for which D follows theStokes-Einstein S-E relation D=kBT / 3 d, where kBTis the thermal energy and the solvent viscosity. Theassumption of S-E diffusion also allows us to calculate ,=kBT / 3 L2, from which we can estimate the evanes-cent field length without using optics; L= kBT / 3 1/2.Using our experimental value for =13.1103 nm/s, weextract L=193 nm. Optical calculations based on a micro-sphere of 200 m in radius with a refractive index for silicaof 1.452 and water of 1.32 for the first order radial mode14

gives L=188 nm, in good agreement with the estimate pro-duced by comparing fluctuation theory with experiment.

Although S-E diffusion is a good approximation for de-scribing our experimental results, it is wanting. This is best

seen by inverting our previous discussion and asking howaccurately we can determine nanoparticle diameters from thesimple S-E based equation,

d = kBT

3 L2 1

. 5

With L taken as before to be 188 nm, the diameters arrived atfrom Eq. 5 for our three standard diameters of 37, 103, and219 nm are 39, 111, and 309 nm, respectively. Here, we seegood agreement for the smallest size and a progressive de-viation as the nanoparticle size is increased. This is likelydue to hydrodynamic effects near the phase boundary.13 Dif-fusion is expected to slow as one approaches an interface,and this effect increases with particle size.15

This letter should leave little doubt concerning the roleof Brownian motion in the resonance wavelength fluctua-tions of a WGM. It may be considered a harbinger for usingthe WGM biosensor for studying the dynamics of nanopar-ticles near interfaces.

This research was supported by NSF through Division ofBioengineering and Environmental Systems Grant No.0522668. The authors thank Steve Holler of Novawave Tech-nologies Inc. and Ralph v. Baltz of Universität Karlsruhe forvaluable discussions.

1F. Vollmer, D. Braun, A. Libchaber, M. Khoshsima, I. Teraoka, and S.Arnold, Appl. Phys. Lett. 80, 4057 2002.

2A. M. Armani, R. P. Kulkarni, S. E. Fraser, R. C. Flagan, and K. J. Vahala,Science 317, 783 2007.

3F. Vollmer, S. Arnold, D. Braun, I. Teraoka, and A. Libchaber, Biophys. J.85, 1974 2003.

4S. Arnold, R. Ramjit, D. Keng, V. Kolchenko, and I. Teraoka, FaradayDiscuss., 137, DOI: 10.1039/b702920a.

5M. Noto, F. Vollmer, D. Keng, I. Teraoka, and S. Arnold, Opt. Lett. 30,510 2005.

6N. M. Hanumegowda, C. J. Stica, B. C. Patel, I. White, and X. Fan, Appl.Phys. Lett. 87, 201107 2005.

7C. Y. Chao and L. J. Guo, Appl. Phys. Lett. 83, 1527 2003.8T. E. Starr and N. L. Thompson, Biophys. J. 80, 1575 2001.9A. Serpengüzel, S. Arnold, and G. Griffel, Opt. Lett. 20, 654 1995.

10S. Arnold, M. Khoshsima, I. Teraoka, S. Holler, and F. Vollmer, Opt. Lett.28, 272 2003.

11G. Griffel, S. Arnold, D. Taskent, A. Serpengüzel, J. Connolly, and N.Morris, Opt. Lett. 21, 695 1996.

12J. C. Knight, G. Cheung, F. Jacques, and T. A. Birks, Opt. Lett. 22, 11291997.

13J. K. Pero, E. M. Haas, and N. L. Thompson, J. Phys. Chem. B 110,10910 2006.

14R. E. Benner, P. W. Barber, J. F. Owen, and R. K. Chang, Phys. Rev. Lett.44, 475 1980.

15H. Brenner, Chem. Eng. Sci. 16, 242 1961.

103902-3 Keng et al. Appl. Phys. Lett. 91, 103902 2007

Downloaded 04 Jan 2008 to 128.122.149.42. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp

Detection of Protein Orientation on the Silica Microsphere Surface UsingTransverse Electric/Transverse Magnetic Whispering Gallery Modes

Mayumi Noto, David Keng, Iwao Teraoka, and Stephen ArnoldMicroparticle Photophysics Laboratory, Polytechnic University, Brooklyn, New York 11201

ABSTRACT The state of adsorbed protein molecules can be examined by comparing the shifts in a narrow line resonancewavelength of transverse electric (TE) and transverse magnetic (TM) whispering gallery modes (WGM) when the moleculesadsorb onto a transparent microsphere that houses WGM. In adsorption of bovine serum albumin (BSA) onto an aminopropyl-modified silica microsphere, the TM/TE shift ratio indicated highly anisotropic polarizability of BSA in the direction normal to thesurface, most likely ascribed to anchoring the heart-shaped protein molecule by one of its tips. The polarization-dependent res-onance shift was confirmed when the surrounding refractive index was uniformly changed by adding salt, which would simulateadsorption of large objects.

INTRODUCTION

The quantity and quality of protein adsorbed on a surface is

of a great concern (1–4). The amount of the surface-bound

protein can be evaluated by various methods (2–11). How-

ever, methods to find the state of the adsorbed molecules are

not well established, except that submolecular information

can be obtained using spectroscopic methods (8–10). There

are controversies about the state and orientation of adsorbed

molecules, even for often studied proteins such as serum

albumin (2,4,11). Their conformation may depend on the

surface—whether it is hydrophobic or ionic (positively or

negatively charged)—and on the pH of the immersing aque-

ous phase. This article proposes using resonance shifts of

photonic whispering gallery modes (WGM) as a method to

determine the state of adsorbed protein.

A transparent microsphere can accommodate WGM in the

vicinity of the sphere surface. The light propagates near the

curved surface by total internal reflection. Resonance is

achieved when the light path closes upon itself in phase after

one cycle. If the diameter of the sphere is sufficiently large

compared with the wavelength, the resonance can have a nar-

row width. The Q value of a silica microsphere in water at a

1.3-mm wavelength can be as large as 2 3 106 (12). A much

greater Q value, exceeding 108, is reported for a toroidal reso-

nator at 680 nm (13).

In each reflection along the circular path of WGM, the

light seeps into the surroundings as an evanescent wave. The

wavelength of the sharp resonance is sensitive to small changes

in dielectric property in the immediate neighborhood of the

transparent microsphere (14). The changes include adsorp-

tion of molecules onto the microsphere and a change of re-

fractive index (RI) in the surrounding medium. The shift of

the wavelength upon adsorption of biomolecules onto the

microsphere has been heralded as the most sensitive detector

ever made possible without the necessity for fluorescent label-

ing (12,15–18). Detection of a single protein molecule is

considered within reach (15). Recent prediction (19) and dem-

onstration (20) of enhanced sensitivity by a high RI coating

has paved the way for the difficult detection. The sensor’s

capability is not limited to estimating the surface density of

adsorbed molecules. Independent detection of the resonance

shifts for two polarization modes—transverse electric (TE) and

transverse magnetic (TM)—is expected to allow us to esti-

mate the orientation of adsorbed anisotropic molecules (21).

Each WGM is specified by l, m, n, and polarization (22). lrepresents the number of waves in a circular orbit, m (¼ l,l 1 1, . . . , l) is the azimuthal index, and n is equal to the

number of peaks in the radial function of the electric field

intensity, thus specifying the radial mode. The polarization is

either TE or TM. The wavelength at resonance is determined

by l, n, and polarization. In a perfect spherical resonator,

modes of different m are degenerate. The shift of resonance

wavelength in response to the environmental changes depends

also on l, n, and polarization (23). It was recently demonstrated

that the observed shifts of TE modes due to RI changes in the

surroundings were in agreement with the shifts calculated

using the indices evaluated for the microsphere used (24).

More than a decade ago, Folan distinguished TE and TM

shifts of WGM in a small polystyrene sphere levitated elec-

trodynamically in air (25). Folan examined the change in the

scattering spectrum as water condensed onto the polymer

sphere for the two polarizations, but the difference between

the two shifts was insignificant within experimental error.

In this report, we use side coupling of a core-exposed single-

mode fiber to induce both TE and TM polarizations in a silica

microsphere and measure the wavelength shifts when proteins

are added to the surrounding fluid to adsorb onto the sphere

surface. We find that the shifts are different for TE and TM

and the ratio of the two shifts provides information on the state

of adsorbed protein. We confirmed the polarization-sensitive

Submitted December 15, 2006, and accepted for publication February 15,

2007.

Address reprint requests to Iwao Teraoka, E-mail: [email protected].

Mayumi Noto’s present address is Nantero Inc., Woburn, MA 01801.

2007 by the Biophysical Society

0006-3495/07/06/4466/07 $2.00 doi: 10.1529/biophysj.106.103200

4466 Biophysical Journal Volume 92 June 2007 4466–4472

shifts by adding NaCl to the surroundings to cause a uniform

increase of RI. The latter situation simulates adsorption of

large objects such as mitochondria (17).

Recently developed dual polarization interferometry (DPI)

(26–29) can provide information on the state of surface-

bound proteins. DPI uses two polarizations of light trans-

mitted through a pair of planar waveguides to find the RI and

thickness of the adsorption layer, which in turn provide

information on the protein conformation. To achieve a high

sensitivity comparable to that of the surface plasmon reso-

nance (SPR) instrument, DPI uses a large sensor area, ;150

mm2. Our WGM sensor has a much smaller sensor area,

typically ,0.005 mm2, yet easily surpasses the sensitivity of

DPI and SPR in terms of adsorbed mass per unit area while

retaining the capability to find the state of the adsorbate.

More importantly, the WGM sensor allows easy interpreta-

tion of the resonance shift in terms of molecular parameters

(21), without the need to assume an adsorption layer of uni-

form RI and thickness (26). Neutron reflectivity (7) is an-

other method that macroscopically characterizes the adsorbed

molecules as a whole, but its sensitivity and usefulness are

limited.

Theoretical background

A plain microsphere of radius a and uniform relative per-

mittivity er1 ¼ n21 is placed in a uniform medium of er2 ¼ n2

2

(n1 . n2). When the wavelength, l, of WGM is much shorter

than a, the electric field, E(r), of the WGM is mostly

confined to the interior of the microsphere. However, the

evanescent field extends into the surroundings to the depth

of ;(l/2p)(n21 n2

2)1/2, which polarizes the molecules in

the immediate neighborhood of the microsphere surface. The

resonance wavelength of WGM shifts from l0 to l0 1 Dl,

when small molecules (much smaller than l/n2) adsorb onto

the sphere surface to displace a part of the surrounding me-

dium. The adsorbed molecules are polarized by E(jrj ¼ a1),

where a1 indicates the exterior side of the sphere surface.

The fractional shift Dl/l0 is equal to the ratio of the po-

larization energy in the adsorbed molecules to the total mode

energy (12,15,21). In general, the TE and TM modes exhibit

different shifts, as the TE mode has only a tangential com-

ponent, Et(a) as E(a1), whereas the TM mode has also a

normal component, En(a1). For uniform adsorption of Np

molecules of excess polarizability, a, at a low surface den-

sity, the fractional shift of either mode is given as (21)

Dl

l0

¼ NpÆattE2

t ðaÞ1 annE2

nða1 Þæ2e0

Rer EðrÞ EðrÞdr

; (1)

where att and ann are the polarizability tensor components in

directions of Et and En, respectively, Æ. . .æ indicates the

average on the sphere surface (for Et and En) or the average

with respect to the configuration of adsorbed molecules (for

att and ann), e0 is the vacuum permittivity, and the volume

integral in the denominator covers the entire space (relative

permittivity er is er1 in the sphere; er2 elsewhere).

We showed earlier that the denominator in Eq. 1 is equal

to 4pa3e0(er1 er2) Æ[Et(a)]2æ for the TE mode (21,23). Then,

the fractional shift of the TE mode is given as

DlTE=l0 ¼ NpðÆattæ=e0Þ½4pa3ðer1 er2Þ1

: (2)

The shift is identical for all radial modes (n ¼ 1, 2, . . .).Since Np a2 for a given surface density of adsorbates,

Dl/l0 a1. There is a weak dependence of DlTE/l0 on

l0 through wavelength dispersions of att, er1, and er2. For

the TM mode, the denominator in Eq. 1 is calculated as

4pa3e0(er1 er2)(Æ[Et(a)]2æ 1 (er2/er1)Æ[En(a1)]2æ) (21,23).

The expression for the TM shift, DlTM, is then obtained. The

ratio DlTM/DlTE is

DlTM=DlTE ¼ ðA1 1 Æannæ=ÆattæÞ=ðA1 1 er2=er1Þ (3)

for each pair of WGM having the same l and n. Here, A1 [

Æ[Et(a)]2æ/Æ[En(a1)]2æ is the intensity anisotropy ratio of the

evanescent field of the TM mode (right on the sphere sur-

face). When l 1, the following approximation is useful

(21):

A1 ffi 1 ðn2k0a=lÞ2; (4)

which may be further approximated as A1 ffi 1 er2/er1. The

second approximation is not good except for the first-order

radial modes (n ¼ 1). The shift ratio given by Eq. 3 is in-

sensitive to the dimension of the adsorbed molecule, as long

as it is sufficiently smaller than l0/n2.

We now evaluate Eq. 3 for a molecule of volume Vp and

uniform, isotropic relative permittivity erp ¼ n2p. We consider

five geometries of the molecule—a sphere, a rod (cylinder)

standing vertically on the surface, a rod lying on the surface,

a disk standing on the surface (edge-on), and a disk lying

on the surface (face-on), as illustrated in Fig. 1. For now,

we assume a low surface density of adsorbed molecules. The

effect of interference from the dipoles induced at nearby

adsorbed molecules will be discussed toward the end of this

section.

For each of the five geometries, ann is definite: Æannæ ¼ann. For a sphere, a standing rod and a lying disk, att, is also

definite. For a lying rod and a standing disk, the orientation

of the molecule on the surface relative to Et varies from

FIGURE 1 Five geometries of particles on a surface: (a) sphere, (b) stand-

ing rod, (c) lying rod, (d) standing disk (edge-on adsorption), and (e) lying

disk (face-on adsorption).

Protein Orientation on Silica Surface 4467

Biophysical Journal 92(12) 4466–4472

molecule to molecule. If we assume random orientation of

molecules in the directions tangential to the surface, Æattæ is

the isotropic mean of the orthogonal tensor components in

the directions parallel to the surface.

Table 1 lists Æattæ/(e0Vp) and Æannæ/(e0Vp) for the five

geometries. These expressions were obtained from the

boundary conditions for the electric field across the surface

of the adsorbate. The expressions for the sphere and the two

orientations of the rod were obtained earlier (21). Our early

works also showed that the image dipole induced within the

sphere does not affect the polarization of the adsorbed par-

ticle (21,30). For rods lying on the surface, Æattæ/(e0Vp) is the

isotropic mean of 2er2(erp 1 er2)1(erp ef2) and erp er2,

which gives the listed expression. For edge-on adsorbed

disks, Æattæ/(e0Vp) is the isotropic mean of (er2/erp)(erp er2)

and erp er2. A thin uniform layer has the same att and ann

as those for isolated disks lying on the surface. The two geo-

metries are listed in the last row of the table.

Table 1 also summarizes results for the shift ratio due to

the shape anisotropy and lists its values for er1 ¼ 1.4522

(silica), er2¼ 1.322 (water), a¼ 171 mm, l0¼ 1.312 mm, l¼1170, and erp ¼ 1.552, where Eq. 4 was used for A1. Here,

the RI of protein at 1.32 mm was estimated as 1.55 from the

value of 1.57 at 589 nm (31). More accurate calculation of

the shift ratio employing the numerical method described

earlier (21) gives the same values. The shift ratio for the

sphere represents the ratio of the field intensities of the two

modes on the surface and is nearly equal to 2 er2/er1. For

the other geometries, the ratio is 2 er2/er1 ¼ 1.18 at np¼ n2

and deviates from that value with an increasing np. The shift

ratio .1.18 indicates ann . att, which is likely due to mole-

cules of an anisotropic shape standing on the surface. The

ratio smaller than 1.18 indicates a geometry of the adsorbate

extending parallel to the surface. The capability of the WGM

sensors to provide information on the molecular orientation,

independent of the size of the molecule, will be useful in

studies on protein adsorption in different solutions and sur-

face environments as well as conformational changes.

The above discussion applies to low surface coverages.

With an increasing surface density, dipolar fields by nearby

particles decrease Æ[En(a1)]2æ but increase Æ[Et(a)]2æ (21).

Earlier (30), we used dipolar approximation to consider the

effect for spherical molecules sequentially and randomly ad-

sorbed onto the surface. Since we do not have a formula for

geometries other than spheres, we adopt a formula for the

spheres. Calculations for spheres of np ¼ 1.55 show that the

effect increases Et(a) by 0.9% and decreases En(a1) by 1.8%

at 15% coverage of the projection of the spheres onto the

surface. As a result, the TE shift is 0.9% greater than the es-

timate given by Eq. 2, and the TM shift is 1.3% less. The

increase in A1 with an increasing surface density causes the

ratio to drop to 1.07 at the highest surface coverage in ran-

dom sequential adsorption (32–34). Concomitantly, the crite-

rion of the shift ratio for the anisotopic polarizability moves

to a smaller value.

MATERIALS AND METHODS

Materials

Bovine serum albumin (BSA; A2153) was purchased from Sigma (St. Louis,

MO). A 50 mmol/L solution of BSA in phosphate buffered saline (PBS) at

pH 7.4 was prepared immediately before resonance shift measurements.

Microspheres of radius around 200 mm were prepared by melting a tip

of a single mode silica fiber (SMF-28; Corning, Corning, NY). Details of

the microsphere preparation can be found elsewhere (12). The surface of

microspheres used in BSA adsorption was modified with aminopropyl-

triethoxysilane. Microspheres used in NaCl experiments were washed in

pyrranhia solution. The diameter of each microsphere was evaluated under

an optical microscope to a reference of the optical fiber of cladding diameter

125 mm.

WGM resonance shift measurement system

The optical part in our measurement system is shown in Fig. 2. We used a

pigtailed butterfly laser (distributed feedback laser; DFB) from NTT Elec-

tronics (NLK1B5E1AA; Saddle Brook, NJ) operating at ;1.31 mm as a

light source. The laser is linearly polarized with an extinction ratio ;1000 as

observed by a photodiode (PDA400; Thorlabs, Newton, NJ) at the end of the

single-mode fiber. Exposing the core of the fiber by etching in hydrofluoric

acid solution decreased the extinction ratio to ;200. However, neither

etching nor contact with the microsphere skewed the polarization. Rotation

of the laser mount around the axis of the output fiber changed the polar-

ization direction. The angle of rotation of the polarizer at the photodiode to

maximize the intensity of transmitted light was measured as the laser mount

was rotated in a step of 10 up to 690 from the unstrained direction. The

required polarizer rotation was nearly identical to the laser rotation. The

standard deviation of the difference between the two angles of rotation was

4.8. The extinction ratio barely changed during the rotation. Thus, we know

what angles of the laser mount cause vertical and horizontal polarizations in

the etched section of the fiber, which in turn will excite TE and TM modes,

respectively, within the microsphere that touches the fiber at its horizontal

equator.

TABLE 1 TM/TE shift ratio in adsorption of small molecules at low densities

DlTM/DlTE

Adsorbate Æattæ/(e0Vp) Æannæ/(e0Vp) Formula At np ¼ 1.55

Spheres 3er2(erp 1 2er2)1(erp er2) 3er2(erp 1 2er2)1(erp er2) (A1 1 1)/(A1 1 er2/er1) 1.18

Rods, standing 2er2(erp 1 er2)1(erp er2) erp er2 [A1 1 (erp 1 er2)/(2er2)]/(A1 1 er2/er1) 1.37

Rods, lying [er2(erp 1 er2)1 1 1/2](erp er2) 2er2(erp 1 er2)1(erp er2) [A1 1 4er2/(erp 1 3er2)]/(A1 1 er2/er1) 1.09

Disks, standing (edge-on) (1/2)(er2/erp 1 1)(erp er2) erp er2 [A1 1 2erp/(erp 1 er2)]/(A1 1 er2/er1) 1.34

Disks, lying (face-on);

thin uniform layer

erp er2 (er2/erp)(erp er2) (A1 1 er2/erp)/(A1 1 er2/er1) 0.90

4468 Noto et al.


The other parts of the measurement system are similar to those described

earlier (18). For this study, the laser wavelength, l, was scanned at 10 Hz by

changing the laser drive current, i, linearly with time. A sawtooth function

generator was used for that purpose. The scan range was ;0.2 nm. The

relationship between l and i was evaluated using an interferometer (Agilent,

Santa Clara, CA; HP3325A). In each scan, the laser intensity and wave-

length increase almost linearly with time, and the two increases are nearly

parallel to each other. When a microsphere is placed in contact with the core-

exposed section of the fiber, destructive interference by the WGM causes

dips in the light intensity at the photodetector. Each dip represents a WGM

of a unique set of indices (l, m, n, polarization).

RESULTS AND DISCUSSION

TE and TM spectra

In each experiment, the position of a microsphere relative to

the fiber was adjusted to produce dips of reasonable depths in

the wavelength scan of the light intensity through the fiber.

Examples of the TE and TM spectra of the photodiode signal

when the DFB laser was scanned by a current sweep, dis-

played in Fig. 3, are similar. Each spectrum has a period of

;0.042 nm, ascribed to splitting of the degenerate azimuthal

modes (m) by a spheroidal shape of the microsphere; Lai

et al. predicted polarization-independent splitting by distor-

tion of the meridional cross section of the microspheroid (35).

There is a major dip and several minor dips in each period.

The major dips are for n ¼ 1. Our scan range of ;0.2 nm

sees just a part of a cluster of the dips having the same l but

different values of m. The minor dips are ascribed to higher

order radial modes (n ¼ 2, 3; l may be different) and a tail

portion of adjacent clusters with n ¼ 1 (l is different by 1).

Adsorption of BSA

One of the microspheres (radius a ¼ 167–203 mm) whose

surface was modified with aminopropylsilane was immersed

in a 980 mL solution of PBS at pH 7.4 that was constantly

stirred and held at 25C. For each adsorption study, a fresh

microsphere was used. Resonance dips in the fiber trans-

mission spectrum were traced as a 20 mL solution containing

BSA (pI ¼ 4.8) was added. The final concentration, 1 mmol/

L, is sufficiently low to make the resonance shift due to the

surroundings’ RI change negligible compared with the shift

due to adsorption but sufficiently high to cause the highest

possible surface coverage for this pH and surface (18). We

expect that the surface coverage is similar for all the mi-

crospheres. Fig. 4 shows typical changes of the TE and TM

spectra. A part of the scan range is zoomed for clarity. The

pair of experiments in Fig. 4 was selected so that the micro-

sphere used for TM is slightly larger than the one used for

TE. The intensity spectrum undergoes a red shift without

changing the overall pattern. In either TE or TM mode, deep

and shallow dips shift nearly equally. The TM shifts are

greater than the TE shifts, although the sphere radius, a, is

greater for TM; the shift is reciprocally proportional to a.

We did two measurements for TE and three for TM using

spheres of different radii. To eliminate the radius dependence

of the fractional shift, we compare the reduced fractional shift,

FIGURE 2 Optical part of microsphere WGM resonance shift measure-

ment system that allows change of the polarization. A zoomed view of the

fiber-microsphere coupling is also shown.

FIGURE 3 Light intensity through the fiber in the wavelength scan for TE

and TM modes. The microsphere of 195-mm radius was immersed in water.

FIGURE 4 Light intensity spectrum before and after injection of BSA.

The radii of the microspheres used for TE and TM modes are 167 mm and

186 mm, respectively. The arrows indicate the shifts for some dips.



k0aDl/l0, where k0 ¼ 2p/l0 (k0a is called a size parameter).

From Eq. 2, we find that k0aDl/l0 is proportional to the pro-

duct of the number density of BSA on the surface and the

polarizability. A similar relationship exists for the TM modes.

In our TE mode experiments, 103k0aDl/l0 is 7.7 6 0.1

(mean 6 SD), regardless of whether the dip is deep or

shallow. For the TM modes, the reduced shift is 10.3 6 0.2.

The shift’s independence of the radial mode agrees with the

theoretical prediction for adsorption of small particles (21).

The ratio of the TM shift to the TE shift is 1.34, which is

close to the ratios for standing rods and standing disks in

Table 1. It is known that the BSA molecule is heart-shaped at

pH¼ 7.4 (36). The result of our TE-TM shift study indicates

adsorption of the heart-shaped molecule by one of its tips. At

pH ¼ 7.4, the protein surface has a high density of N1H3

and COO (4), and the microsphere surface is dense with

N1H3. It is not surprising that the protein anchors to the

aminopropyl surface by facing one of the sections covered

predominantly with COO to the sphere surface. A study of

protease digestion of BSA adsorbed on an unmodified silica

surface at the same pH indicates a similar geometry (4), al-

though sections predominantly covered with N1H3 would

face to SiO on the silica surface in the latter experiment.

The volume of a BSA molecule, Vp, is estimated from the

molecular mass (1.10 3 1019 g) and the specific volume of

BSA, 0.734 g/cm3 (37), as 81.0 nm3. To estimate the surface

coverage of BSA from our experimental data, we use below

a picture of a standing disk with radius R and height H. First,

we note that the above Vp can be equated to a disk of R¼ 4.1

nm, H ¼ 1.5 nm, where the aspect ratio is close to the one

proposed (38) in a phosphorescence study as consistent with

x-ray crystallographic data (36). Then, from Table 1, we

obtain att/e0 as 46.1 nm3, where er2 ¼ 1.322 and erp ¼ 1.552

were used. The surface density of the BSA molecules can be

estimated using the formula

Np

4pa2 ¼ k0a

DlTE

l0

3er1 er2

k0a=e0

: (5)

Since k0 ¼ 2p/l0 ¼ 4.796 mm1, er1 ¼ 1.4522, and

k0a(Dl/l0)TE ¼ 7.7 3 103 in our measurement, Np=ð4pa2Þis estimated as 1.3 3 104 mm3. Therefore, the area fraction

F of the projection of the rectangular cross section 2RH onto

the surface is estimated as F ¼ ½Np=ð4pa2Þ2RH ¼ 0:15.

The latter value is ,1/3 of the highest possible value of F by

spheres, 0.55 (30,32–34).

We could assume another geometry for the BSA mole-

cule, for instance, a standing rod. The molecular dimension

of the rod that gives Vp ¼ 81.0 nm3 is R ¼ 1.9 nm and H ¼7.1 nm, for example. Then, att/e0¼ 44.9 nm3, virtually iden-

tical to the one we obtained for the disk model.

As discussed earlier, F ¼ 0.15 is too low for the dipoles

induced in nearby particles to affect the estimate of F or the

TM/TE shift ratio. Therefore, we do not need to change our

discussion for the surface density and orientation of the ad-

sorbed BSA molecules.

Refractive index change of the surroundings

We tested our polarization-sensitive WGM sensor for a

uniform change of relative permittivity, Dn22, in the sur-

roundings. The change mimics adsorption of particles with a

linear dimension greater than the penetration depth of the

evanescent field. The shift will be greater for the mode with a

greater n, since its evanescent field penetrates deeper into the

surroundings. Numerical calculation of the resonance con-

ditions (23) gives the reduced response, k0aDl/(l0Dn22), of

the TE mode in a microsphere with a ¼ 174 mm at l0 ¼1.312 mm as 2.507 and 2.755 for n ¼ 1 and 2, respectively.

The reduced response of the TM mode will be 2.950 and

3.252 for n ¼ 1 and 2, respectively. The reduced response is

insensitive to a: At a ¼ 196 mm, the response for n ¼ 1 is

;0.9% less than it is at a ¼ 174 mm.

In adding NaCl to the surroundings three times, the shifts

exhibited a complicated pattern, as each radial mode had a

different shift. Two parts of Fig. 5 show 103k0aDl/l0 for TE

and TM modes as a function of NaCl concentration c in PBS

buffer surrounding a plain silica microsphere (a ¼ 174–196

mm). The data were compiled from the shifts of different dips

in a few measurements. Attention was paid not to include

broad dips that apparently consisted of two or more dips at

any stage of NaCl addition; the shape of these dips changed

as more NaCl was introduced.

In Fig. 5, the shift for each dip in successive NaCl addition

follows a straight line through the origin. The slope of the

line is slightly different for each dip. The variations are

ascribed to uncertainties in c and a and fluctuations in the

resonance position. For the TE mode (Fig. 5 a), most of the

data run along the lower solid line, ascribed to the n ¼1 modes. The line gives 103k0aDl/(l0Dc)¼ 0.990 L/g. With

k0aDl/l0¼ 2.507 3 2n2Dn2, we obtain dn/dc¼ 0.150 mL/g,

FIGURE 5 Fractional wavelength shift times the size parameter for (a) TE

and (b) TM modes when a 20 mL, 0.0163 g/mL solution of NaCl is incre-

mentally added to a buffer surrounding a silica microsphere. The lines are

the best fit for two groups of data.

4470 Noto et al.


slightly less than 0.171 mL/g, the value reported for l ¼ 589

nm at 25C (39). We believe that the difference is mostly due

to RI dispersion. Likewise, most of the data run along the

lower solid line in the plot for TM modes (Fig. 5 b). The ratio

of the TM slope to the TE slope is 1.19, close to the theo-

retical value of 1.18.

In both of TE and TM plots, two sets of data are away

from those for n ¼ 1. They are ascribed to the n ¼ 2 modes.

The ratio of k0aDl/(l0Dc) for these sets to that for n ¼ 1 is

1.11 in TE and TM. The ratio compares favorably with the

theoretical values.

CONCLUSIONS

We have demonstrated a simple method to excite TE and TM

modes separately in a microsphere and observe the shifts of

resonance lines when protein molecules adsorb at high

densities. Studies of protein adsorption on different surface

chemistries in different pH at different surface coverages will

help us understand the state of protein on the surface. In

particular, studies at low coverages will be important, as the

TM/TE shift ratio allows us to estimate the polarizability

anisotropy ratio of isolated molecules. The studies will be

facilitated by simultaneous observation of the TE and TM

shifts for a common microsphere which can be accomplished

by feeding the light linearly polarized at ;45 from the TE

direction, splitting the light by the polarizations right before

the photodetector, and measuring the fiber transmission

spectra using two photodiodes. The high sensitivity of the

WGM sensor will allow such measurements at extremely

low coverages of small molecules.

We thank L. Folan for helpful discussion.

This work was supported by the National Science Foundation through

BES0522668.

REFERENCES

1. Nakanishi, K., T. Sakiyama, and K. Imamura. 2001. On the adsorptionof proteins on solid surfaces, a common but very complicated phenom-enon. J. Biosci. Bioeng. 91:233–244.

2. Gray, J. J. 2004. The interaction of proteins with solid surfaces. Cur.Op. Str. Bio. 14:110–115.

3. Robers, M., I. J. A. M. Rensink, C. E. Hack, L. A. Aarden, C. P. M.Reutelingsperger, J. F. C. Glatz, and W. Th. Hermens. 1999. A newprinciple for rapid immunoassay of proteins based on in situ precipitate-enhanced ellipsometry. Biophys. J. 76:2769–2776.

4. Larsericsdotter, H., S. Oscarsson, and J. Buijs. 2005. Structure,stability, and orientation of BSA adsorbed to silica. J. Coll. Interf. Sci.289:26–35.

5. Malmsten, M. 1995. Ellipsometry studies of fibronectin adsorption.Coll. Surf. B. 3:371–381.

6. Sukhishvili, S. A., and S. Granick. 1999. Adsorption of human serumalbumin: dependence on molecular architecture of the oppositelycharged surface. J. Chem. Phys. 110:10153–10161.

7. Lu, J. R. 1999. Neutron reflection study of globular protein adsorptionat planar interfaces. Annu. Rep. Progr. Chem. Sec. C. 95:3–45.

8. Giacomelli, C. E., M. G. E. G. Bremer, and W. Norde. 1999. ATR-FTIR study of IgG adsorbed on different silica surfaces. J. Coll. Interf.Sci. 220:13–23.

9. Giacomelli, C. E., and W. Norde. 2001. The adsorption-desorptioncycle. Reversibility of the BSA-silica system. J. Coll. Interf. Sci. 233:234–240.

10. Noinville, S., M. Revault, M.-H. Baron, A. Tiss, S. Yapoudjian, M.Ivanova, and R. Verger. 2002. Conformational changes and orienta-tion of Humicola lanuginosa lipase on a solid hydrophobic surface: anin situ interface Fourier transform infrared-attenuated total reflectionstudy. Biophys. J. 82:2709–2719.

11. Su, T. J., J. R. Lu, R. K. Thomas, Z. F. Cui, and J. Penfold. 1998. Theconformational structure of bovine serum albumin layers adsorbed atthe silica-water interface. J. Phys. Chem. B. 102:8100–8108.

12. Vollmer, F., D. Braun, A. Libchaber, M. Khoshsima, I. Teraoka, and S.Arnold. 2002. Protein detection by optical shift of a resonantmicrocavity. Appl. Phys. Lett. 80:4057–4059.

13. Armani, A. M., D. K. Armani, B. Min, K. J. Vahala, and S. M.Spillane. 2005. Ultra-high-Q microcavity operation in H2O and D2O.Appl. Phys. Lett. 87:151118.

14. Serpenguzel, A., S. Arnold, and G. Griffel. 1995. Excitation ofresonances of microspheres on an optical fiber. Opt. Lett. 20:654–656.

15. Arnold, S., M. Khoshsima, I. Teraoka, S. Holler, and F. Vollmer. 2003.Shift of whispering-gallery modes in microspheres by protein adsorp-tion. Opt. Lett. 28:272–274.

16. Vollmer, F., S. Arnold, D. Braun, I. Teraoka, and A. Libchaber. 2003.Multiplexed DNA detection by optical resonances in microspheres.Biophys. J. 85:1974–1979.

17. Noto, M., F. Vollmer, D. Keng, I. Teraoka, and S. Arnold. 2005.Nanolayer characterization through wavelength multiplexing of amicrosphere resonator. Opt. Lett. 30:510–512.

18. Noto, M., M. Khoshsima, D. Keng, I. Teraoka, V. Kolchenko, and S.Arnold. 2005. Molecular weight dependence of a whispering gallerymode biosensor. Appl. Phys. Lett. 87:223901.

19. Teraoka, I., and S. Arnold. 2006. Enhancing sensitivity of a whisper-ing gallery mode microsphere sensor by a high-refractive index surfacelayer. J. Opt. Soc. Am. B. 23:1434–1441.

20. Gaathon, O., J. Culic-Viskota, M. Mihnev, I. Teraoka, and S. Arnold.2006. Enhancing sensitivity of a whispering gallery mode bio-sensorby sub-wavelength confinement. Appl. Phys. Lett. 89:223901.

21. Teraoka, I., and S. Arnold. 2006. Theory on resonance shifts in TE andTM whispering gallery modes by non-radial perturbations for sensingapplications. J. Opt. Soc. Am. B. 23:1381–1389.

22. Johnson, B. R. 1993. Theory of morphology-dependent resonances:shape resonances and width formulas. J. Opt. Soc. Am. A. 10:343–352.

23. Teraoka, I., S. Arnold, and F. Vollmer. 2003. Perturbation approachto resonance shifts of whispering galley modes in a dielectric micro-sphere as a probe of a surrounding medium. J. Opt. Soc. Am. B. 20:1937–1946.

24. Hanumegowda, N. M., C. J. Stica, B. C. Patel, I. White, and X. Fan.2005. Refractometric sensors based on microsphere resonators. Appl.Phys. Lett. 87:201107.

25. Folan, L. M. 1992. Characterization of the accretion of material bymicroparticles using resonant ellipsometry. Appl. Opt. 31:2066–2071.

26. Swann, M. J., L. L. Peel, S. Carrington, and N. J. Freeman. 2004. Dual-polarization interferometry: an analytical technique to measure changesin protein structure in real time, to determine the stoichiometry ofbinding events, and to differentiate between specific and nonspecificinteractions. Anal. Biochem. 329:190–198.

27. Freeman, N. J., L. L. Peel, M. J. Swann, G. H. Cross, A. Reeves, S.Brand, and J. R. Lu. 2004. Real time, high resolution studies of proteinadsorption and structure at the solid-liquid interface using dual polari-zation interferometry. J. Phys. Condens. Matter. 16:S2493–S2496.

28. Berney, H., and K. Oliver. 2005. Dual polarization interferometry sizeand density characterisation of DNA immobilisation and hybridisation.Biosens. Bioelectron. 21:618–626.



29. Ricard-Blum, S., L. L. Peel, F. Ruggiero, and N. J. Freeman. 2006.Dual polarization interferometry characterization of carbohydrate-protein interactions. Anal. Biochem. 352:252–259.

30. Teraoka, I., and S. Arnold. 2007. Dielectric property of particles atinterface in random sequential adsorption and its application to whis-pering gallery mode resonance-shift sensors. J. Appl. Phys. 101:023505.

31. Voros, J. 2004. The density and refractive index of adsorbing proteinlayers. Biophys. J. 87:553–561.

32. Onoda, G. Y., and E. G. Liniger. 1986. Experimental determination ofthe random-parking limit in two dimensions. Phys. Rev. A. 33:715–716.

33. Wang, J. S. 1994. A fast algorithm for random sequential adsorption ofdiscs. Int. J. Mod. Phys. C. 5:707–715.

34. Torquato, S. 1995. Mean nearest-neighbor distance in random packingsof hard d-dimensional spheres. Phys. Rev. Lett. 74:2156–2159.

35. Lai, H. M., P. T. Leung, K. Young, P. W. Barber, and S. C. Hill. 1990.Time-independent perturbation for leaking electromagnetic modes inopen systems with application to resonances in microdroplets. Phys.Rev. A. 41:5187–5198.

36. Carter, D. C., and J. X. Ho. 1994. Structure of serum albumin. Adv.Protein Chem. 45:153–196.

37. Brill, A. S., J. S. Olin, and B. M. Siegel. 1962. The specific volume ofdry protein. J. Cell Biol. 13:249–251.

38. Ferrer, M. L., R. Duchowicz, B. Carrasco, J. Garcıa de la Torre, andA. U. Acuna. 2001. The conformation of serum albumin in solution: acombined phosphorescence depolarization-hydrodynamic modelingstudy. Biophys. J. 80:2422–2430.

39. Russo, P. Differential Index of Refraction, dn/dc. http://macro.lsu.edu/HowTo/.

4472 Noto et al.


APPLIED PHYSICS LETTERS 87, 223901 2005

Molecular weight dependence of a whispering gallery mode biosensorM. Noto, M. Khoshsima, D. Keng, I. Teraoka, V. Kolchenko, and S. Arnolda

Microparticle Photophysics Lab(MP3L), Polytechnic University, Brooklyn, New York 11201

Received 6 July 2005; accepted 19 October 2005; published online 23 November 2005

We report on molecular weight dependence measurements for an optical resonance biosensor. Adielectric microparticle is evanescently coupled with an optical fiber for the resonance stimulation,and a shift of the resonance wavelength is measured to monitor protein monolayer formation on themicroparticle surface. Wavelength shifts for proteins over two orders of magnitude in molecularweight are measured. We show that the shift is proportional to molecular weight to the one-thirdpower. Our result demonstrates that the optical resonance biosensor provides protein sizeinformation upon detection. This molecular weight dependency differentiates optical resonancesensing from electrical detection using field-effect transistors. © 2005 American Institute ofPhysics. DOI: 10.1063/1.2137902

Label-free detection of biomolecules has become an ac-tive area of research. New sensing methods have been soughtto enhance the sensitivity and facilitate miniaturization ofdevices. Recent notable approaches are the direct electricaldetection of biomolecules using field-effect transistorsFET.1–4 Single viral particle detection has been demon-strated with a silicon nanowire FET.2 Although FET sensorsallow scalable detection of biomolecules in real time, thesignal transduction relies on charges carried on analyte mol-ecules. The sensor signal alone does not provide the infor-mation on properties such as molecular weight M.

The use of a high-Q optical resonator for biomoleculardetection has been demonstrated for protein adsorption anddeoxyribonucleic acid hybridization with unprecedentedsensitivity.5,6 Based on first-order perturbation theory, wefind that a resonance frequency shift is proportional to theexcess polarizability of an analyte molecule and its surfacedensity.7 Since polarizability scales with the volume of amolecule,8 a resonance frequency shift is expected to containmolecular size information i.e., molecular weight.

In what follows, we investigate the M dependence of aresonance frequency shift. We test five protein samples from5000 to 700 000 g/mol, and measure the shift for proteinmonolayer formation on the microparticle surface. We thenprovide a quantitative analysis of a protein layer to addressthe M dependence.

Our biosenser consists of a distributed feedback laserwith a nominal wavelength of 760 nm Princeton Light-wave as the light source, a sample cell, and a Si detector818SL, Newport to monitor the transmitted light intensityFig. 1. The setup is similar to the one reported earlier,6 butwe modified the sample cell and improved on the tempera-ture control and the solution mixing. The new sample cell isfashioned from a disposable polystyrene cuvette cut to 1.5cm height. A single-mode optical fiber 780 HP, Nufern ispassed through the cuvette and secured with epoxy resin. Thesample cell is placed on a modified laser mount, whose tem-perature is maintained at 25 °C by a laser controller LDC-3742B, ILX Lightwave.

To stimulate a resonance in a microparticle also knownas whispering gallery mode WGM, a section of the optical

a
Electronic mail: [email protected]
0003-6951/2005/8722/223901/3/$22.50 87, 22390Downloaded 11 Dec 2005 to 128.122.88.167. Redistribution subject to

fiber is acid eroded to a final diameter 4 m in the samplecell, thus exposing the evanescent filed. We fabricate a silicamicroparticle equatorial radius R200 m of an oblateshape with the eccentricity of 0.05 by melting a single-mode optical fiber in a butane/nitrous oxide flame. The mi-croparticle is positioned to make contact with the etched sec-tion of the fiber. Optical resonances in the microparticle aredetected as dips in the transmitted light intensity when thewavelength is scanned repeatedly at 2 Hz between 24 and 32mA using a saw-tooth-shaped function 3325 A, HewlettPackard. The output wavelength has a drive current imA dependence of nm=761.0185+0.0043i+1.010−5i2. The LABVIEW program records the transmissionspectrum and tracks a resonant dip by detecting its positionwith a parabolic minimum fit. A typical quality factor of aresonance in water is 4106.

We selected five water-soluble proteins with M rangingfrom 5000 to 700 000 g/mol: Insulin I5500, Sigma, M=5800, -lactalbumin LA, L6010, Sigma, M =14 300, bo-

FIG. 1. a Experimental setup. b Close-up view of the optical fiber and
the microparticle coupling.
© 2005 American Institute of Physics1-1 AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

http://dx.doi.org/10.1063/1.2137902

http://dx.doi.org/10.1063/1.2137902

223901-2 Noto et al. Appl. Phys. Lett. 87, 223901 2005

vine serum albumin BSA A2153, Sigma, M =66 000,-globulin 191 478, ICN Biomedical, M =152 000, andthyroglobulin T1001, Sigma, M =670 000. All protein stocksolutions were prepared by dissolving protein in 10 mMphosphate buffered saline PBS pH 7.4 except -globulin,which was dissolved in 50 mM PBS at pH 6.

Microparticle surfaces are chemically modified to en-hance protein adsorption onto the surfaces. Amine surface9 isused for all the proteins except -globulin. Carboxylsurface10 is used for the latter.

To measure a resonance shift at monolayer saturation foreach protein sample, we first carry out adsorption isothermexperiments to find the protein concentration necessary formonolayer formation on the microparticle surface.

The sample cell is filled with 980 l PBS, and a freshlymodified microparticle is mounted on the xyz stage and al-lowed to thermally equilibrate with the sample cell solution.20 l of a protein stock solution is injected and the adsorp-tion is monitored by following the resonance dip position.The measurement is carried out under constant mixing and isterminated when the dip position no longer changes. LA,BSA, and thyroglobulin achieve monolayer saturation at0.2 M Fig. 2, whereas insulin and -globulin require 0.5nM and 4.3 M, respectively. The adsorption isotherm fol-lows a Langumir-type pattern, indicating that protein layer isno more than one layer.

Using the monolayer concentrations determined from theisotherm experiments, a wavelength shift is measured threetimes for each protein sample. A fractional wavelength shift /R at monolayer saturation is plotted at logarithmicscale in Fig. 3. The slope is 0.31±0.02. The inset confirmsthat the fractional shift is proportional to M1/3 and the errorbars indicate the spread of data from the three measurements.The linear fit of the M1/3 graph has the slope 0.060±0.003.

Having confirmed the M dependency of /, we pro-ceed with quantitative analysis of protein layers. We start ouranalysis by obtaining the analytic expression for a resonanceshift for a protein layer formation. A layer perturbation cor-responds to adding the dielectric excess nl

2−nm2 of thickness t

to the microparticle surface, where nl and nm are the refrac-tive indices of the layer and the medium, respectively. Fromthe first-order perturbation theory, the fractional wavelength

FIG. 2. Adsorption isotherm of LA, BSA, and thyroglobulin.The dotted lines are a guide for the eyes.

shift isDownloaded 11 Dec 2005 to 128.122.88.167. Redistribution subject to

=

nl2 − nm

2 ns

2 − nm2

t

RL

t1 − exp− t/L , 1

where L is the evanescent field length given by L= /4nef f

2 −nm2 −1/2 , ns is the refractive index of the micropar-

ticle, and neff is the effective index for propagation within theWGM.11 In the limit of a thin layer t /L1, Eq. 1 can beapproximated and we take an effective medium approach toaccount for inhomogeneity of the layer. We obtain

=

fnp2 − nm

2 ns

2 − nm2

t

R, 2

where f is the volume fraction of protein and np is proteinrefractive index.

LA, -lactalbumin, is known to assume a sphericalshape and its adsorption kinetics is successfully modeled bytreating LA as a sphere.12 We treat LA as a model protein.The volume of a protein molecule can be written asMNAp−1, where NA is Avogadro’s number and p is massdensity. Relating this expression to the volume of a sphere,4 /3r3, we find the radius r in terms of M :r= 3M /4NAp1/3. The thickness of a layer packed withspheres is 2r, and Eq. 2 becomes

=

fnp2 − nm

2 ns

2 − nm2

2

R 3

41/3 M

NAp1/3

. 3

Molecular weight dependence is evident in Eq. 3. We com-pare Eq. 3 to our empirical result /R=0.060M1/3.Using p1.37 g/cm3, which is practically constant formost proteins,13 np=1.50,14 nm=1.33, and ns=1.46, we findthe f value for the LA layer is 0.34.

Random packing of spheres onto a planar surfaceachieves lower surface coverage than hexagonal close pack-ing. The simulation study by Torquato15 estimates that themaximum fractional area coverage for the random packing is0.55. Thus, the volume fraction is 0.367. Our LA layer isclose to the theoretical limit.

Although all of our candidate proteins are termed12

FIG. 3. Resonance wavelength shifts against M at logarithmic scale.Insulin, LA, BSA, -globulin, and thyroglobulin. Theline was drawn manually to fit the data points. Inset: Linear plot of /*R against M1/3.

“globular”, LA has been characterized as being spherical. AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

223901-3 Noto et al. Appl. Phys. Lett. 87, 223901 2005

The others vary in shape. It appears from Fig. 3 that theWGM sensor is relatively insensitive to the shape morphol-ogy. The WGM sensor started out as a means for biomolecu-lar detection, yet it now appears that one can make estimatesof protein size.

Research at Polytechnic was supported by a NSF grantNo. BES-0522668. The authors thank Frank Vollmer andSteve Holler for valuable discussions.

1W. U. Wang, C. Chen, K. Lin, Y. Fang, and C. M. Lieber, Proc. Natl.Acad. Sci. U.S.A. 102, 3208 2005.

2F. Patolsky, G. Zheng, O. Hayden, M. Lakadamyali, X. Zhuang, and C. M.Lieber, Proc. Natl. Acad. Sci. U.S.A. 101, 14017 2004.

3J. Fritz, E. B. Cooper, S. Gaudet, P. K. Sorger, and S. R. Manalis, Proc.Natl. Acad. Sci. U.S.A. 99, 14142 2002.

4R. J. Chen, S. Bangsaruntip, K. A. Drouvalakis, N. W. S. Kam, M. Shim,Y. Li, W. Kim, P. J. Utz, and H. Dai, Proc. Natl. Acad. Sci. U.S.A. 100,
4984 2003.
Downloaded 11 Dec 2005 to 128.122.88.167. Redistribution subject to

5F. Vollmer, D. Braun, A. Libchaber, M. Khoshsima, I. Teraoka, and S.Arnold, Appl. Phys. Lett. 80, 4057 2002.

6F. Vollmer, S. Arnold, D. Braun, I. Teraoka, and A. Libchaber, Biophys. J.85, 1974 2003.

7S. Arnold, M. Khoshsima, I. Teraoka, S. Holler, and F. Vollmer, Opt. Lett.28, 272 2003.

8F. Fröhlich, Theory of Dielectrics Oxford University Press, London,1958, p. 28.

9N. Zammatteo, L. Jeanmart, S. Hamels, S. Courtois, P. Louette, L. Hevesi,and J. Remacle, Anal. Biochem. 280, 143 2003.

10Amine-modified microparticles are treated with 0.1 M succinic anhydridein N,N-dimethyl formamide for 15 min, rinsed with the solvent and etha-nol, and dried in the oven for at least 10 min at 70 °C.

11M. Noto, F. Vollmer, D. Keng, I. Teraoka, and S. Arnold, Opt. Lett. 30, 12005.

12A. P. Minton, Biophys. J. 76, 176 1999.13C. R. Cantor and P. R. Schimmel, Biophysical Chemistry Part II W.H.

Freeman and Company, New York, 1980, p. 554.14P. A. Cuypers, W. T. Hermans, and H. C. Hemker, Biochemistry 84, 56

1978.15
S. Torquato, Phys. Rev. Lett. 74, 2156 1995.
AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

510 OPTICS LETTERS / Vol. 30, No. 5 / March 1, 2005

Nanolayer characterization through wavelength multiplexingof a microsphere resonator

Mayumi Noto, Frank Vollmer,* David Keng, Iwao Teraoka, and Stephen Arnold

Microparticle Photophysics Laboratory, Polytechnic University, 6 Metrotech Center, Brooklyn, New York 11201

Received September 16, 2004

We optically characterize nanolayer (,150 nm) formation in situ on a silica microsphere in an aqueous environ-ment by simultaneously following the shifts of whispering-gallery modes at two wavelengths. This approachwas inspired by layer perturbation theory, which indicates that these two measurements can be used to deter-mine independently both the thickness and the optical dielectric constant. The theory is verified for extremecases and used to characterize a biophysically relevant hydrogel nanolayer with an extremely small excessrefractive index of 0.0012. © 2005 Optical Society of America

OCIS codes: 170.4520, 300.6490.

It is common in optics to create nanoscopic dielectriclayers from inorganic insulators, semiconductors, andmetals; however, as biophotonics1 looms more strongly,soft condensed biofunctional layers formed at anaqueous–solid interface and made of DNA, protein,lipids, and hydrogels are gaining strong appeal. Thisappeal is being driven by the need for biosensorsfor clinical and military use and for investigationof biomolecular interactions as they relate to drugdiscovery. Although techniques outside of optics,such as neutron ref lection, have been used to gaugethe thickness of these layers, new methods to monitorthe formation of such layers and to characterize themin a noninvasive manner are greatly needed. Weshow in what follows that wavelength-multiplexingexperiments on microspherical optical cavities canfollow the growth, gauge the thickness, and determinethe optical dielectric constant for adsorbed layers.

Our experimental approach is driven by the pre-dicted theoretical effect that a nanoscopic layer has onthe shifts of whispering-gallery modes (WGMs) in amicrosphere. In particular, the ratio of the shifts be-tween resonances stimulated at separated wavelengthscan be used to determine the thickness of a layer. Inaddition, the dielectric constant of the layer can beevaluated by use of this thickness and the shift at ei-ther wavelength. The theoretical approach we employhere is distinct from recent work2 on individual dipoleperturbations, in which the sensitivity of such a sys-tem for the detection of single biomolecular adsorptionevents was estimated. Although the latter theory hasbeen successfully applied to sensitivity issues associ-ated with the identification of mismatches in DNA,3

for the current work our interest is in the perturba-tions by a dielectric layer.

In what follows we first outline our perturbationtheory. Our approach will be to turn the microsphereperturbation problem into a quantum analog4 and per-turb the analog potential by adding a layer. Thenwe present experiments that test this theory. Finallywe attempt to determine the thickness and refractive-index perturbation for a thin hydrogel layer.

Although the full theory will appear elsewhere,5

here we outline its basic components. Our approach

0146-9592/05/050510-03$15.00/0

resembles first-order perturbation theory in quantummechanics. We concentrate on TE modes. Represent-ing the electric f ield in terms of a scalar functionE LC, where L is a dimensionless angular momen-tum operator, allows us to easily reduce the problemof solving the vector wave equation to the solution ofa Schrödinger-like equation for the radial part of C,Cr.6 The effective energy Eeff for this quantum ana-log is the square of the free-space wave vector Eeff k2

0 ,and the effective potential Veff k2

01 2 n2 1

ll 1 1r2, where n is the radial refractive-indexprofile and l is the angular momentum quantumnumber of a particular mode. A layer perturbationcorresponds to changing n2 from the surface out to athickness t by dn2. The f irst-order perturbation is

dEeff crjdVeffjcr , (1)

where cr is constructed from appropriate quasi-normalized functions.7 After substituting for themajor components in expression (1) we find that thefractional perturbation in effective energy is

dk20

k20

2

Ω2tR

∑dn2

n2s 2 n2

m

∏æ ΩLt

1 2 exp2tLæ

, (2)

where ns and nm are the refractive indices of the sphere(silica, 1.47) and its environment (water, 1.33), respec-tively; R is the sphere radius; and L is the evanes-cent f ield length, with L l4p n2

eff 2 n2m212. The

fractional wavelength shift dll is related to dk20k2

0through dll 212 dk2

0k20, where neff is the ef-

fective index for propagation within the WGM. Sinceour sphere has an 200-mm radius and is much largerthan either of the laser wavelengths, neff varies by only1% between radial modes8 and will be approximatedby its grazing incidence value ns. Equation (2) mayseem awkward; however, it has a particularly simplestructure when one considers that the principal wave-length dependence is contained within the evanescentfield length in the rightmost factor on the right-handside. By a judicious choice of the wavelength regionsto be used, the leftmost factor on the right-hand side

© 2005 Optical Society of America

March 1, 2005 / Vol. 30, No. 5 / OPTICS LETTERS 511

can be considered relatively constant. Consequently,by taking a ratio of the fractional shift at one wave-length l1 to that at a longer wavelength l2, we arrive ata particularly simple expression that provides the de-sign principle for our surface analysis approach. Thisratio S is

S

µdl

l

∂1µ

dl

l

∂2

L11 2 exp2tL1L21 2 exp2tL2

. (3)

For an ultrathin layer (i.e., tL1, tL2 ,, 1), Sapproaches 1, whereas for a thick layer (i.e.,tL1, tL2 .. 1), S approaches L1L2, which with nefftaken constant is just l1l2. For our experimentsthis ratio is 760 nm1310 nm 0.58. For ourchosen wavelengths, S falls off in an approximateexponential fashion in between the two extreme caseswith a characteristic length of tc 192 mm [i.e.,S L1L2 1 1 2 L1L2exp2ttc]. MeasuringS therefore allows us to estimate t. With t in hand,Eq. (2) gives dn2.

We performed wavelength-multiplexing experimentswhile forming nanolayers on a silica microsphere sur-face. Light from two current-tunable distributed-feedback lasers with nominal wavelengths of 760 and1310 nm was coupled to a single-mode fiber (Nufern780-HP) (Fig. 1). A portion of the f iber was acideroded down to a 3-mm diameter to facilitate couplingto the WGMs of a silica microsphere.9 The micro-sphere and fiber were contained within a temperature-controlled 1-ml cuvette containing buffer solution anda magnetic stirrer. Beyond this cuvette the fiberwas led to an InGaAs detector. By scanning bothlasers with a synchronous ramp, we observed thatthe light from each independently stimulates WGMsin the microsphere and yields a distinct transmissionspectrum with a superposition of resonant dips fromeach. By observing which resonances disappear aseither laser is shut off, the resonances are easilyassociated with the 760- and 1310-nm region. In thisway, resonances can be identified and tracked.

As a test of our perturbation theory we constructedtwo experiments at the extreme limits. First we builta monolayer of bovine serum albumin (BSA) 3 nmthick.10,11 The microsphere surface was treated with3-aminopropyltrimethoxysilane, and BSA with afinal concentration of 1 mM was injected into 10-mMphosphate-buffered saline (pH of 7.4). The shiftsof resonances at two wavelengths, l1 760 nm andl2 1310 nm, are shown in Fig. 2(a). The BSAreached Langmuir-like saturation (i.e., monolayerformation10) at dll 1 3 1025. The two resonancesfrom each wavelength region shifted almost the sameamount. As a second test case we injected NaCl intothe water surrounding a sphere. We sequentiallyincreased the salt concentration by 0.1-M incrementsstarting with de-ionized water. Figure 2(b) shows thetime trace of the resonance shifts in dll from eachwavelength region. In this case the fractional shiftat 760 nm is considerably less than that at 1310 nm.Figure 3 summarizes the experimental results and

the layer theory prediction. For a BSA monolayer,tL ,, 1, and S in Eq. (3) should approach 1. The ex-periment yields a slope of 1.04. For the NaCl experi-ment, tL .. 1, and S should approach the ratio ofthe wavelengths, 0.58. The experimental result was0.54. Our limiting tests are in reasonable agreementwith theory. We are now in a position to demonstratethe usefulness of our approach by attempting toevaluate the optical properties of biophysically rele-vant hydrogel.

Poly-L-lysine (PLL) is a hydrogel that takes on ex-tremely positive charge in water and is consequentlyfavored as a means for adsorbing biomolecules witha negative charge. However, the physical properties

Fig. 1. Experimental setup for wavelength multiplexing ofa microcavity. LD, laser diode.

Fig. 2. (a) Resonance shifts at two wavelengths [l1 760 nm (thin curve) and l2 1310 nm (thick curve)]owing to BSA adsorption. (b) Resonance shifts at thesame wavelengths owing to two sequential injections ofNaCl by 0.1-M increments.

512 OPTICS LETTERS / Vol. 30, No. 5 / March 1, 2005

Fig. 3. Plot of dll760 nm versus dll1310 nm for BSAlayer formation (dots) and for six incremental 0.1-M in-jections of NaCl (squares). Points represent experimentalvalues and lines represent the layer perturbation theory.

of PLL are difficult to measure since it deposits ina thin layer with extremely low contrast in a waterenvironment. We used a PLL solution from Sigma(P8920, 0.1% wt.vol. in water, average molecularweight of 225,000 gmol) that is commonly used inbiology to treat glass slides. To generate a layer,40 ml of the PLL solution was injected into 900 mlof phosphate-buffered saline surrounding the micro-sphere. We observed a shift toward longer wave-lengths that saturated in the usual Langmuir fashionfor monolayer formation. However, the fractionalshift at a saturation of 2 3 1026 was well belowanything we had seen previously. Slope S based onthe average of a number of experiments was 0.82.This slope fed back into Eq. (3) gives a thickness of110 nm, which is reasonable considering the molecu-lar structure of the polymer. After substitutingthis thickness into Eq. (2), we determined the waterexcess increment in the optical dielectric constant tobe dn2 0.0033. Consequently, dn 0.0012, whichis indeed small.

We have provided support for our simple layer per-turbation theory, and as a result the WGM resonatorgoes beyond its original promise as a biosensor. Byanalyzing wavelength-multiplexed experiments withthis theory, we can not only monitor the growth ofnanolayers, but we can also determine the opticaldielectric constant for the resulting f ilm. All thishas been done at one polarization (TE) that can beconveniently arranged by matching the polarizationcharacteristics of the lasers and by reducing the lengthof optical f ibers. Surprisingly, so long as both laserslaunch the same polarization into their respectivemodes (i.e., TE or TM), Eq. (3) should be identical foreither polarization. This interesting result can beshown by use of the more electromagnetically detailedwork of Teraoka et al.12

Wavelength-multiplexing experiments have createda new window of opportunity for the WGM resonator.For the first time to our knowledge, a WGM resonatorwas applied to study a commonly used biofunctionallayer. It will be interesting to measure the changein S as a self-assembled monolayer forms on a sur-face. As the layer density increases, the morphologyof the molecules in the layer may change. This phasetransition will lead to a change in the layer thickness,and a real-time measurement of S should reveal thistransition.

Our method can be extended to individual particles.However, the spherically symmetrical theory thatgenerated Eq. (2) cannot be used. Instead a Green’sfunction approach must be applied. This alternatedirection is in the works. The result shows promisefor looking at heterogeneous structures such as ad-sorbed bacteria.

M. Noto and F. Vollmer are grateful for their gradu-ate and postdoctoral support while at Polytech-nic University from the National Science Foundation(BES-0119273). S. Arnold’s e-mail address is [email protected].

*Present address, Rowland Institute, Harvard Uni-versity, Cambridge, Massachusetts 02142.

References

1. P. N. Prasad, Introduction to Biophotonics (Wiley, NewYork, 2003).

2. S. Arnold, M. Khoshsima, I. Teraoka, S. Holler, andF. Vollmer, Opt. Lett. 28, 272 (2003).

3. F. Vollmer, S. Arnold, D. Braun, I. Teraoka, andA. Libchaber, Biophys. J. 85, 1974 (2003).

4. H. M. Nussenzveig, Diffraction Effects in Semiclas-sical Scattering (Cambridge University, Cambridge,England, 1992).

5. S. Arnold, M. Noto, and F. Vollmer, in Frontiers ofOptical Spectroscopy: Investigating Extreme Physi-cal Conditions with Advanced Optical Techniques,B. DiBartolo, ed. (Kluwer Academic, Dordrecht, TheNetherlands, to be published).

6. S. Arnold and S. Holler, in Cavity-Enhanced Spectro-scopies, R. D. van Zee and J. P. Looney, eds. (Academic,San Diego, Calif., 2002), pp. 227–253.

7. E. S. C. Ching, P. T. Leung, and K. Young, in OpticalProcess in Microcavity, R. K. Chang and A. J. Campillo,eds. (World Scientif ic, Singapore, 1996), pp. 1–76.

8. J. C. Knight, G. Cheung, F. Jacques, and T. A. Birks,Opt. Lett. 22, 1129 (1997).

9. J. P. Laine, B. E. Little, and H. A. Haus, IEEE Photon.Technol. Lett. 11, 1429 (1999).

10. F. Vollmer, D. Braun, A. Libchaber, M. Khoshsima,I. Teraoka, and S. Arnold, Appl. Phys. Lett. 80, 4057(2002).

11. T. J. Su, J. R. Lu, R. K. Thomas, Z. F. Cui, andJ. Penfold, J. Phys. Chem. B 102, 8100 (1998).

12. I. Teraoka, S. Arnold, and F. Vollmer, J. Opt. Soc.Am. B 20, 1937 (2003).

Date post:	30-Apr-2018
Category:	Documents
Upload:	nguyendat
View:	225 times
Download:	2 times

WHISPERING GALLERY MODE BIOPARTICLE SENSING …research.poly.edu/~sarnold/paper/Pub_704_6242.pdf ·...

Documents