+ All Categories
Home > Documents > Molecular Interferometric Imaging for Biosensor Applications

Molecular Interferometric Imaging for Biosensor Applications

Date post: 23-Sep-2016
Category:
Upload: dd
View: 215 times
Download: 0 times
Share this document with a friend
11
1680 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007 Molecular Interferometric Imaging for Biosensor Applications Ming Zhao, Xuefeng Wang, Gregory M. Lawrence, Patricio Espinoza, and David D. Nolte (Invited Paper) Abstract—Molecular interferometric imaging (MI2) is a common-path interferometric imaging technique for detecting protein binding to surfaces. The experimental metrology limit is 10 pm/pixel longitudinal resolution at 0.4-µm diffraction-limited lateral resolution, corresponding to 1.7 attogram of protein, which is only 8 antibody molecules per pixel, near to single-molecule detection. The scaling mass sensitivity at the metrology limit is 5 fg/mm. We demonstrate a protein microarray application in a 128-multiplex immunoassay. Assay applications include prostate specific antigen (PSA) at a detection limit of 60 pg/mL and the cy- tokine interleukin-5 (IL-5) at a detection limit of 50 pg/mL. Real- time binding assays using MI2 enable the study of reaction kinetics of antibodies exposed to antigen, and the binding of antibody Fc regions to protein G. Index Terms—Biomedical optics, biosensor, immunoassay, inter- ferometry, label-free, protein microarray. I. INTRODUCTION I N THE field of label-free biosensors [1], there is currently a gap between the number of analytes that can be mea- sured in a biological sample using existing systems, compared to the number of measurements required to understand the pro- teomic signature of health and disease. The measurement prob- lem of proteomics is immense. A single cell can have as many as 10 000 expressed proteins, and each protein interacts with three or four others, on average, in cascaded networks of protein interactions [2], [3]. Optical biosensors [4] have a potential ad- vantage for this problem because of the intrinsic parallelism of light that allows multiple channels to be illuminated and de- tected simultaneously. Some imaging biosensors rely on this parallelism to some degree, but most do not tap the full resource that this parallelism represents. In principle, it is possible to have over a million independent optical modes per square millimeter. There are two challenges to utilizing this millionfold resource for optical biosensors. The first challenge is the preparation of the assay. In the case of mi- croarrays, the individual spatial modes need to be patterned with Manuscript received September 19, 2007; revised October 15, 2007. This work was supported by QuadraSpec, Inc. through the Purdue Research Foundation. M. Zhao, X. Wang, and D. D. Nolte are with the Department of Physics, Purdue University, West Lafayette, IN 47907 USA (e-mail: zhaom@physics. purdue.edu; [email protected]; [email protected]). G. Lawrence and P. Espinoza are with QuadraSpec, Inc., West Lafayette, IN 47906 USA (e-mail: [email protected]; PEspinoza@quadraspec. com). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTQE.2007.911002 recognition molecules, either antibodies, peptides, or proteins. This represents a technological challenge that still has far to go. The second challenge, which is the readout of the million opti- cal modes, has already been partially met. High-end pixel arrays today have 10 million pixels, which is sufficient to measure the full-mode density of an area 3 mm on a side. An important part of this challenge is the generation of a robust signal proportional to the amount of protein bound to the surface of the biosensor within a single pixel. The most robust means of performing direct optical detection is common- path phase-quadrature interferometry [5], in which a signal carrying optical phase information from thin protein layers is combined with a reference wave that has a fixed relative phase of 90 to produce an intensity shift proportional to the protein-induced phase. Interferometry performs the function of a phase-to-intensity transducer. The common-path approach establishes and maintains quadrature independent of mechan- ical vibrations, making the system highly stable and of low noise. Common-path interferometric approaches to protein detec- tion have used spinning-disk interferometry in the form of laser scanning on the biological compact disk (BioCD) [6]. The BioCD uses single-mode illumination by a focused laser that scans bound protein on a spinning disk. There are sev- eral approaches to establishing and maintaining phase quadra- ture on a BioCD, including microdiffraction [7], [8], adaptive optics [9], phase contrast [10], [11], and in-line [12]. The ad- vantages to high-speed sampling on a spinning platform include suppression of temporal 1/f noise, fast scan times, large-area mi- croarrays, and high multiplexing. Conversely, the single-mode illumination used on the BioCD does not access the intrin- sic parallel advantage of optical detection afforded by pixel arrays. In this paper, we describe a full-field protein imaging ap- proach called molecular interferometric imaging (MI2) that uti- lizes the full parallel advantage of a pixel detector while rely- ing on common-path in-line phase quadrature identical to the in-line BioCD. The signal and reference waves are produced locally from a single optical mode and share the same optical path to maintain a stable relative phase. The shot-noise-limited sensitivity of the technique approaches the single-molecule range for moderate molecule sizes. The theoretical operation of MI2, based on in-line quadrature interferometry, is described in Section II, followed in Section III with the technical details of the microscope system and silicon substrates. Immunoassays are described in Section IV with applications to prostate-specific 1077-260X/$25.00 © 2007 IEEE
Transcript
Page 1: Molecular Interferometric Imaging for Biosensor Applications

1680 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007

Molecular Interferometric Imagingfor Biosensor Applications

Ming Zhao, Xuefeng Wang, Gregory M. Lawrence, Patricio Espinoza, and David D. Nolte

(Invited Paper)

Abstract—Molecular interferometric imaging (MI2) is acommon-path interferometric imaging technique for detectingprotein binding to surfaces. The experimental metrology limit is10 pm/pixel longitudinal resolution at 0.4-µm diffraction-limitedlateral resolution, corresponding to 1.7 attogram of protein, whichis only 8 antibody molecules per pixel, near to single-moleculedetection. The scaling mass sensitivity at the metrology limit is5 fg/mm. We demonstrate a protein microarray application in a128-multiplex immunoassay. Assay applications include prostatespecific antigen (PSA) at a detection limit of 60 pg/mL and the cy-tokine interleukin-5 (IL-5) at a detection limit of 50 pg/mL. Real-time binding assays using MI2 enable the study of reaction kineticsof antibodies exposed to antigen, and the binding of antibody Fcregions to protein G.

Index Terms—Biomedical optics, biosensor, immunoassay, inter-ferometry, label-free, protein microarray.

I. INTRODUCTION

IN THE field of label-free biosensors [1], there is currentlya gap between the number of analytes that can be mea-

sured in a biological sample using existing systems, comparedto the number of measurements required to understand the pro-teomic signature of health and disease. The measurement prob-lem of proteomics is immense. A single cell can have as many as10 000 expressed proteins, and each protein interacts with threeor four others, on average, in cascaded networks of proteininteractions [2], [3]. Optical biosensors [4] have a potential ad-vantage for this problem because of the intrinsic parallelism oflight that allows multiple channels to be illuminated and de-tected simultaneously. Some imaging biosensors rely on thisparallelism to some degree, but most do not tap the full resourcethat this parallelism represents.

In principle, it is possible to have over a million independentoptical modes per square millimeter. There are two challengesto utilizing this millionfold resource for optical biosensors. Thefirst challenge is the preparation of the assay. In the case of mi-croarrays, the individual spatial modes need to be patterned with

Manuscript received September 19, 2007; revised October 15, 2007. Thiswork was supported by QuadraSpec, Inc. through the Purdue ResearchFoundation.

M. Zhao, X. Wang, and D. D. Nolte are with the Department of Physics,Purdue University, West Lafayette, IN 47907 USA (e-mail: [email protected]; [email protected]; [email protected]).

G. Lawrence and P. Espinoza are with QuadraSpec, Inc., West Lafayette,IN 47906 USA (e-mail: [email protected]; [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JSTQE.2007.911002

recognition molecules, either antibodies, peptides, or proteins.This represents a technological challenge that still has far to go.The second challenge, which is the readout of the million opti-cal modes, has already been partially met. High-end pixel arraystoday have 10 million pixels, which is sufficient to measure thefull-mode density of an area 3 mm on a side.

An important part of this challenge is the generation of arobust signal proportional to the amount of protein bound tothe surface of the biosensor within a single pixel. The mostrobust means of performing direct optical detection is common-path phase-quadrature interferometry [5], in which a signalcarrying optical phase information from thin protein layersis combined with a reference wave that has a fixed relativephase of 90◦ to produce an intensity shift proportional to theprotein-induced phase. Interferometry performs the functionof a phase-to-intensity transducer. The common-path approachestablishes and maintains quadrature independent of mechan-ical vibrations, making the system highly stable and of lownoise.

Common-path interferometric approaches to protein detec-tion have used spinning-disk interferometry in the form oflaser scanning on the biological compact disk (BioCD) [6].The BioCD uses single-mode illumination by a focused laserthat scans bound protein on a spinning disk. There are sev-eral approaches to establishing and maintaining phase quadra-ture on a BioCD, including microdiffraction [7], [8], adaptiveoptics [9], phase contrast [10], [11], and in-line [12]. The ad-vantages to high-speed sampling on a spinning platform includesuppression of temporal 1/f noise, fast scan times, large-area mi-croarrays, and high multiplexing. Conversely, the single-modeillumination used on the BioCD does not access the intrin-sic parallel advantage of optical detection afforded by pixelarrays.

In this paper, we describe a full-field protein imaging ap-proach called molecular interferometric imaging (MI2) that uti-lizes the full parallel advantage of a pixel detector while rely-ing on common-path in-line phase quadrature identical to thein-line BioCD. The signal and reference waves are producedlocally from a single optical mode and share the same opticalpath to maintain a stable relative phase. The shot-noise-limitedsensitivity of the technique approaches the single-moleculerange for moderate molecule sizes. The theoretical operation ofMI2, based on in-line quadrature interferometry, is described inSection II, followed in Section III with the technical details ofthe microscope system and silicon substrates. Immunoassays aredescribed in Section IV with applications to prostate-specific

1077-260X/$25.00 © 2007 IEEE

Page 2: Molecular Interferometric Imaging for Biosensor Applications

ZHAO et al.: MOLECULAR INTERFEROMETRIC IMAGING FOR BIOSENSOR APPLICATIONS 1681

antigen (PSA) and the cytokine interleukin-5 (IL-5). Theability of MI2 to perform real-time binding in a manner similarto surface plasmon sensors is described in Section V.

II. PRINCIPLE OF MI2

A. Inline Quadrature

Protein immobilized on a dielectric surface may be treated asan additional dielectric film that changes the reflection coeffi-cient r of the surface. By fabricating a disk with an appropriatereflected phase and amplitude, the reflectance change due to theprotein can be maximized. For a protein layer of thickness h,the modified reflectivity is

r′ = r +

[(rp − r)(1 − rrp)(

1 − r2p

) + r

]4πnp

λih (1)

where r′ is the protein-modified reflection coefficient, rp is thereflection coefficient of the medium–protein interface, np is therefractive index of the protein layer, and λ is the wavelength ofthe probe light. This equation simplifies to

r′ = r + iπ

(n2

m − n20)(1 + r)2

nm λh = r

((1 − iK

1 + r2

r

h

λ

)(2)

where h is the thickness of the protein layer, and nm is therefractive index of the medium in which the dielectric surfaceis immersed. The medium can be air or water. The constantK = −π(n2

m − n20)/nm is real-valued.

MI2 measures the protein density or the thickness of proteinlayer by monitoring the reflectance change ∆R = |r′|2 − |r|2 .The optical setup adopts in-line (IL) common-path interferom-etry in which the relative phase is nearly 90◦ between the partialreflections from the protein and from the underlying substrate.The reflectance change caused by the presence of a protein layeris given by

∆RIL =

(∣∣∣∣1 − iK(1 + r2)

r

h

λ

∣∣∣∣2

− 1

)|r|2

= 2K Im((1 + r2)/r)|r|2 h

λ+

∣∣∣∣K (1 + r2)r

∣∣∣∣2 (

h

λ

)|r|2

≈ 2K Im((1 + r2)/r)|r|2 h

λ(3)

in which the quadratic term is negligible when the height ofthe protein layer is a few nanometers, much smaller than thewavelength of light.

This equation reaches a maximum when r = ±i/√

3. Thesensitivity of protein detection would be maximized at thisworking condition, called the quadrature condition for in-linedetection. In practice, it is not necessary to fabricate a disk tohave exactly this reflectivity.

In MI2 measurements, the quantity that is being measuredis the ratio of the intensity change caused by the presence of

Fig. 1. Responsivity to protein at different oxide thicknesses on silicon asa function of wavelength. The operating wavelength for this paper is 630 nmusing an oxide thickness of 120 nm.

protein relative to the reflected intensity of the substrate, ∆I/I

∆IIL

I=

2Kh Im(r + 1/r)|r|2λ|r|2 =

2K Im(r + 1/r)λ

h. (4)

From this equation, the measured ∆I/I is proportional to theheight of the protein layer, and can be converted to a surfaceprotein height with a conversion factor that is determined by thesubstrate reflection coefficient.

B. Substrate Structure

A simple substrate with which to work is a thermal siliconoxide on silicon. The oxide layer acts as a spacer, and the re-flected light from the bottom of the oxide layer interferes withlight reflected from the top of the oxide layer where protein isimmobilized. The relative phase of the two waves is determinedby the thickness of the oxide layer, and changed by the presenceof protein. For a two-layer system, the protein-free reflectioncoefficient r is given by

r =r1 + r2e

−2iδ

1 + r1r2e−2iδ(5)

where r1 is the reflection coefficient of the air–oxide interface,and r2 is the reflection coefficient of the oxide–silicon interface.The optical phase shift δ of the incident light that travels throughthe oxide layer is given by

δ =2πn1d cos φ1

λ(6)

where n1 is the refractive index of the silicon dioxide, d isthe thickness of the oxide layer, and φ1 is the angle of therefracted light in the oxide layer. Fig. 1 shows the dependenceof the system response as a function of wavelength and oxidethickness. For our experiments, we use a 630-nm wavelength,and an oxide thickness of 120 nm. The protein response forthis condition is calculated by (4) to be 1.6%/nm for a proteinlayer with a refractive index of 1.44 at 630 nm. Because ofthe nonresonant nature of the detection, this conversion ratio is

Page 3: Molecular Interferometric Imaging for Biosensor Applications

1682 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007

Fig. 2. Protein responsivity as a function of increasing numerical aperture for7×, 20×, and 40× magnifications. The data values are experimental comparedto the calculated response.

relatively stable. If the refractive index of protein changes from1.44 to 1.5, or the thickness of the substrate SiO2 layer changesfrom 120 to 125 nm, the protein response changes from 1.6%to about 1.8%. This conversion ratio is only used to convertthe measured relative intensity change ∆I/I to a surface proteinheight and does not affect relative measurements.

C. Angular Bandwidth

The theoretical responsivity is based on normal-incident light.However, when a microscope objective is used, the numericalaperture (NA) of the objective accepts oblique-incident lightand needs to be considered. For oblique-incident light with anrefraction angle of φ1 , the response of the in-line quadratureinterference is approximately

∆RIL = 2Kh

λIm

(r(φ1) +

1r(φ1)

)|r(φ1)|2 cos φ1 (7)

in which r(φ1) is also a function of the refraction angle. Whena microscope objective with a large NA is used, the resultingresponse is integrated over the collection angle because of thelarge angular bandwidth, which reduces the system responsivity.Fig. 2 shows the system response as a function of NA for auniform distribution of incident light intensity. The data in thefigure are the measured protein response using objectives withNA of 0.12 (7×), 0.45 (20×), and 0.65 (40×).

III. EXPERIMENTAL SYSTEM

A. MI2 Experimental Apparatus

The detection system of MI2 is shown in Fig. 3. Monochromiclight from a light source is directed into a reflective microscope(Leica DMR) and focused onto the sample by the microscopeobjective. The reflected light is collected by the same objec-

Fig. 3. System layout comprised of an illumination source, an objective lens,the target substrate, and a CCD camera. The incident and reflected light aresurface-normal and require no polarization elements. The phase relationship forin-line interferometric quadrature is established and maintained by a thermaloxide layer on the silicon substrate.

tive and imaged onto a camera that captures the images on acomputer.

Two light sources have been used in our experiments. Oneis a 10-mW LED with a center wavelength at 630 nm and a30-nm bandwidth (eLED). The top of the LED is flattened toobtain more uniform illumination. The other light source is anOriel model 66187 halogen lamp with 10-nm bandpass filtersat wavelengths 635, 532, and 440 nm. The LED is more stablethan the halogen lamp, with a smaller intensity drift.

The light from the light source is directed into the microscopeand focused onto the sample by microscope objectives with7×, 20×, or 40× magnification. The 7× objective is a 4×Nikon 0.17-m corrected objective, which when coupled withthe infinity-corrected Leica microscope produces an effectivemagnification of 7×. The corresponding pixel resolutions are2.2, 0.72, and 0.36 µm, respectively, which are calibrated by atest chart with 100-µm features.

The charge-coupled device (CCD) camera is a 12-bitQImaging 4000R interline camera, with 2048 × 2048 7.4-µmpixels. Because of the interline format of the pixel array, the ef-fective size of each pixel is 14.8 µm. At 40× magnification, thiscorresponds to 0.37 µm on the sample. The Arie diffraction limitat a wavelength of 630 nm is 0.39 µm, which is approximatelyequal to the pixel resolution at 40×. The full-well depth of theCCD is 40 000 electrons, with a read-out noise of 10 electrons.The measured photon transfer curve of the CCD camera showedthat the noise of the CCD pixels has a baseline at 10 electrons,and increases as the square-root of the number of electrons perwell up to nearly the full well. This demonstrates that the noiseof the CCD is dominated by photon shot-noise when operatedclose to full well. Therefore, we operate the CCD typically at30 000 electrons per well per acquisition, to acquire as manyphotons as possible, while avoiding nonlinearity near full well.

Page 4: Molecular Interferometric Imaging for Biosensor Applications

ZHAO et al.: MOLECULAR INTERFEROMETRIC IMAGING FOR BIOSENSOR APPLICATIONS 1683

B. Disk Structure and Preparation

The substrates that we used in our experiments were 100-mmdiameter silicon wafers coated with 120 nm of thermal oxide.The top SiO2 surface was functionalized to bind protein cova-lently. Two surface chemistries were used for the immobiliza-tion of protein onto the surface. One was based on an isocyanatecrosslinker, and the other on an aldehyde-functionalized surface.Both surface chemistries bind protein covalently.

Protein spots were printed using a piezoelectric printer. Theresulting diameter of the spots was about 100–150 µm. After theprint, the spots were incubated for 1 h to immobilize the proteins.The disk was then washed and the remaining functionalizedsurface blocked and passivated.

C. Shearing Interferometry

The intensity modulation produced by a protein layer is small,usually only a few percent of the total reflected light inten-sity. When measured directly, it is dominated by the spatialinhomogeneity of the illumination, which is the low-frequencyspatial 1/f noise of the system and can be much larger thanthe protein-induced intensity modulation. To remove the spa-tial 1/f noise, we perform shearing interferometry, which isshown schematically in Fig. 4(a). The sample is first imagedby the CCD, then the sample is mechanically shifted and im-aged again. The spatial profile of the illumination remains un-changed during the shift, and, therefore, can be normalized.Two images are taken for each dataset, one before (I1) andone after (I2) the shift, and are referenced against each otherby

∆I

I= 2

(I1 − I2

I1 + I2

). (8)

This procedure measures the relative intensity change causedby the shift for each pixel in the field-of-view. The relative in-tensity change is converted to surface height with the conversionfactor calculated previously.

Fig. 4(b) shows the raw data acquired by the CCD camera.The protein spots are buried in the large background variationand are almost invisible. Fig. 4(c) shows the result of shearinginterferometry for the data in Fig. 4(b). The spatial 1/f noise isremoved, and the protein spot can be measured precisely. Theresult contains a positive image of the protein spot and a negativemirror image of the same spot. In this process, the protein spot isreferenced against its adjacent land. Therefore, the result of theshearing interferometry is a surface height profile differenced tothe adjacent land.

In an immunoassay, the binding of protein is analyzed bydifferencing the height profile of the immobilized capture agentbefore and after incubation with a sample containing multipleanalytes. An example is given in Fig. 5. A protein spot wasscanned with a 20× objective before (Fig. 5(a)) and after incu-bation (Fig. 5(b)). The two scan images of the same spot are thenregistered by cross correlation and differenced. The differenceimage is shown in Fig. 5(c).

Fig. 4. Image shifting procedure. (a) Removal of spatial 1/f noise by acquiringtwo images of the same protein spot with a lateral shift of the target betweenimages. The images are differenced and normalized to remove spatial variationsin the microscope illumination. (b) Raw image of 100-µm diameter proteinspots under 20× magnification. The CCD camera operates near full well, withthe protein producing an intensity modulation of only a percent. (c) The resultof shearing interferometry under 20× magnification. The difference image hastwo mirror images of the same protein spot, but referenced against two differentadjacent lands.

Fig. 6(a) shows the probability histograms of the prescan spot,the postscan spot, and the difference. Both the spot print heightin the prescan image and the binding of protein in the differenceimage are nearly Gaussian. Fig. 6(b) shows histograms of anarea of the same size as the spot on the land for the prescan,postscan, and difference, which show that the land is nonreactiveand very flat, with surface height standard deviations less than0.3 nm/pixel.

Page 5: Molecular Interferometric Imaging for Biosensor Applications

1684 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007

Fig. 5. Protein spot imaged as (a) a prescan, (b) postscan, and (c) difference.The spot was anti-IL5 antibody with bound IL-5 that is binding secondaryanti-IL-5 from solution at a concentration of 100 ng/mL. The captured analyteexhibits a surface height increase of 0.21 nm.

D. Metrology Limit

The shearing interferometry procedure removes the spatial 1/fnoise, and the remaining noise in the differential surface profileare the temporal 1/f noise (which is the long-term system drift),and the noise from the CCD camera. When the measurementtime is short compared to the system drift, and the CCD operatesclose to full-well depth, then the measurement of the differentialsurface profile is shot-noise limited.

Because of the shot-noise-limited nature of the CCD detec-tion, we average many images together to obtain a larger photonexposure to reduce the relative shot-noise. Fig. 7 shows theeffect of data averaging on the detection noise of the surfaceheight profile. The detection noise is obtained by differencingconsecutively collected images (with averaging) and calculatingthe standard deviation of the difference. The curves for differentmagnifications are separated because of different intensity-to-height conversion ratios. The absolute intensity noise at differ-ent magnifications is comparable because it only arises fromthe CCD camera, and is not related to the actual optical sys-tem. The noise decreases as the square-root of the number ofaverages taken, suggesting that the noise is random and un-correlated, which is another property of shot-noise. When theintegration time becomes too long, long-term system drift be-gins to dominate, which increases the measurement noise. Thenoise is a minimum at 12 pm for 4096 averages under 40×magnification. We typically use 128 averages at a frame rateof 4 frames/s, which takes 30 s for the acquisition of oneimage. If a faster CCD, or a CCD with a larger well depth,were used, it would be possible to reduce the noise further,as more photons can be collected in the same data acquisitiontime.

The metrology limit of MI2 is calculated using the minimumof the detection noise, which is 12 pm per pixel with 40×magni-fication at 4096 averages at the lateral resolution of 0.4 µm. Theheight sensitivity of 12 pm with 0.4-µm resolution corresponds

Fig. 6. Histograms of the protein spot shown in Fig. 5. Histogram of (a) thespot and (b) the adjacent land. The root-mean-squared mass change per pixelon the land is 120 attograms.

to 2 attogram of protein per pixel assuming a protein density of1.3 g/cm3 . This is approximately 8 antibody molecules with atypical size of 150 kD, which is close to single-molecule detec-tion. The scaling mass sensitivity is calculated by dividing theminimum detectable mass by the square-root of the measure-ment area [9], and is 5 fg/mm in the metrology limit.

This metrology limit is obtained under ideal conditions, with-out dismounting or moving the disk. Therefore, this limit is notachieved in practical assays in which the disk needs to be dis-mounted for sample incubation. When the disk is dismountedand remounted, the random position shift of the sample causesthe derivative of the surface roughness to influence the mea-surement and becomes the dominant source of measurementuncertainty. As a result, in a practical assay under our cur-rent experimental protocol of 128 averages, the measurement

Page 6: Molecular Interferometric Imaging for Biosensor Applications

ZHAO et al.: MOLECULAR INTERFEROMETRIC IMAGING FOR BIOSENSOR APPLICATIONS 1685

Fig. 7. Measurement noise as a function of the number of images averaged.The noise decreases as the square-root of the number of images, correspondingto shot noise. For a large number of averages, long-term system drift begins todominate.

uncertainty increases to about 100 pm/pixel. The scaling masssensitivity is inversely proportional to pixel resolution, and, forpractical assays under 7×, 20×, and 40× magnification, thescaling mass sensitivity is 200, 70, and 40 fg/mm, respectively.Higher magnification results in higher sensitivity, but increasesthe acquisition time.

IV. IMMUNOASSAYS USING MI2

A. Multiplexed 128-Fold Assay

To demonstrate the potential of MI2 to perform highly multi-plexed assays, we performed a multiplexed reverse-phase assayin which immobilized rabbit and mouse IgG were incubatedagainst antirabbit and antimouse IgG in solution. A 120-nmoxide-on-silicon wafer was plasma cleaned and separated intowells. The silicon oxide surface was then functionalized withthe isocyanate coating. Each well was printed with 64 proteinspots that form 16 2 × 2 unit cells that have mouse IgG onone diagonal and rabbit IgG on the other. For the experiment,64 wells were selected and were incubated with samples thathad 64 different concentration combinations of antirabbit IgGfrom goat (Sigma), antimouse IgG from goat (Sigma), and al-bumin from bovine serum (BSA) (Sigma). All samples wereprepared in 10-mM pH 7.4 phosphate-buffered saline (PBS).Each protein had four different concentrations and all permu-tations of these concentrations produced 64 different three-component sample mixtures. With the two different analytesprinted on the disk, this experiment is a 128-plex immunoas-say. The concentrations of antirabbit IgG were 0, 1, 10, and100 ng/mL. The concentrations of antimouse IgG were 0, 10,100, and 1000 ng/mL. The concentrations of BSA were 0, 10,100, and 1000 µg/mL. The disk was incubated statically for1 h; then it was rinsed with distilled water and blown dry bynitrogen.

Fig. 8. Results of the 128-multiplex assay (2 analytes per well over 64 wells).The mouse concentrations were 0, 10, 100, and 1000 ng/mL (increasing down-ward on the red matrix with BSA and rabbit concentrations varying along thehorizontal), and the rabbit concentrations were 0, 1, 10, and 100 ng/mL (in-creasing to the right on the green matrix with BSA and mouse concentrationsvarying along the vertical). The final matrix is combined color (red = mouse,green = rabbit with equal mixture adding to yellow) and intensity (stronger formore bound mass).

The 64 wells selected for this experiment were scanned with7× magnification before incubation and scanned again after the1-h static incubation. The data were analyzed in the same wayas described earlier. For each spot, the postscan and prescandata were registered by cross correlation, and shifted with linearinterpolation. Then, the two images were differenced to obtaina difference image of protein height change for each pixel of theprotein spot, and the difference image was histogramed and fit toa Gaussian distribution to extract the mean height change. The32 mouse and rabbit spots in each well were averaged togetherto obtain the assay response for each well.

The result of the analysis is shown in Fig. 8 as a color mapof the assay responses across the 64 wells. The responses of themouse spots are plotted in the red matrix, with concentrationsof antimouse increasing downward and varying concentrationsof BSA and antirabbit across to the right. The responses of therabbit spots are plotted in the green matrix, with concentrationsof antirabbit increasing to the right in the horizontal directionwith varying concentrations of BSA and antimouse downward.The intensity of the color represents the magnitude of the proteinbinding. For both the mouse and rabbit assay, the binding ofprotein increases with increasing concentration, as can be seenfrom an increasing brightness of the color downward for mouse,and to the right for rabbit. The presence of a high-concentrationprotein background (BSA) does not have a significant effect onthe binding, and there is no systematic trend within each analyteconcentration.

The two responses were averaged over the BSA concentra-tions and combined into a single matrix in which equal inten-sity of red and green produces yellow, and the intensity of thecolor shows the strength of the protein binding. In this ma-trix, the highest concentration for both analytes is yellow (lowerright), showing that the binding of antimouse and antirabbit weresimilar. The color turns more red going left, and more green go-ing up, and with a decreasing intensity for the lower concentra-

Page 7: Molecular Interferometric Imaging for Biosensor Applications

1686 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007

tions (upper left). The limit of detection (LOD) of this 128-plexassay was approximately 50 ng/mL per well. The sensitivity waslimited by spot-to-spot and by well-to-well variability across thedisk. Much more sensitive assays are possible in a monolithicarray format, which we demonstrate next in an assay for PSA.

B. PSA

Prostate cancer (PCa) accounts for 10% of all deaths fromcancer [13]. A major focus of prostate cancer research has beenthe early detection of PCa using serum biomarkers [14]. Wetested the sensitivity of our detection system with a sandwichassay of PSA. In a sandwich assay, a capture antibody is im-mobilized onto the substrate to capture the antigen in a sample.Then, it is incubated against a secondary antibody that bindsto the captured antigen to boost the binding signal. For thisexperiment, a 120-nm silicon oxide disk was plasma cleanedand functionalized with the isocynate surface coating. Then, itwas printed with 25 600 spots in a 256 × 100 radial grid pat-tern. The tangential distance between two spots was 2π/256,and the radial separation of spots was 0.3 mm. The spots wereprinted in 2 × 2 unit cells of target and reference, with goatanti-PSA antibody (Biospacific) as target on one diagonal of theunit cell, and goat antirabbit IgG (Sigma) as reference in theother diagonal. The concentrations of the print solutions were100 µg/mL.

The entire disk was incubated against 10 ng/mL of PSA anti-gen (Fitzgerald Inc.) spiked into a high-concentration proteinbackground of 2-mg/mL BSA using an orbital shaker (VWR)for 3 h. The solution was prepared in 10 mM pH 7.4 PBSbuffer, with 0.05% Tween 20. The 3-h incubation on the orbitalshaker removes potential mass transport problems of surfacecapture and drives the reaction to equilibrium. After incuba-tion with antigen, the disk was rinsed with distilled water andblown dry by nitrogen, and protein spots were scanned with MI2under 20× magnification. The disk was then incubated with100 ng/mL of goat anti-PSA (Biospacific) in PBS buffer with0.05% Tween 20 on the orbital shaker for 30 min. After rinsingwith distilled water and drying by nitrogen, the same proteinspots were scanned again under 20× MI2.

The prescan and postscan of the protein spots were registeredand differenced, and the histograms of the differences for asingle pair of spots are shown in Fig. 9, which shows the bindingof the secondary antibody to the captured antigen. The broaddistributions in the difference are caused by inhomogeneous spotmorphology. Although the distributions are not well separated,more than 15 000 pixels were used in the histograms for eachspot, and, therefore, the standard error for the distributions isreduced by a factor of 120. The resulting height change forthe target spot is 0.613 ± 0.005 nm, and the height change forthe reference spot is −0.195 ± 0.005 nm. The negative valuefor the reference spot is caused by wash-off. The SNR for asingle pair of spots is 160:1. Assuming a linear interpolationfrom 10 ng/mL of PSA, the concentration detection limit is60 pg/mL, where the SNR would be unity, which is well belowthe normal concentration of PSA in human sera. The detectionwas performed in a high-concentration protein background of

Fig. 9. Histograms of pixel values for a single target-reference pair of spotsfor the PSA assay at 20× magnification. The incubation was at 10 ng/mL. TheLOD for a single pair is 60 pg/mL.

2-mg/mL BSA, which did not affect the assay significantly.Because a single pair of 100-µm diameter spots was used forthis analysis, a disk that is printed with 25 000 spots would beable to perform more than 10 000 assays, each with a detectionlimit of 60 pg/mL.

C. IL-5

Detection of IL5 cytokine through sandwich assays was per-formed using MI2. The cytokine IL5 is an important signalingprotein in the immune system. For this experiment, both iso-cyanate and aldehyde surface chemistries were used, and cap-ture antibodies were immobilized by either direct printing or bybinding to protein A/G that was printed onto the surface. Diskswith 120-nm oxide were used for this experiment, and eachdisk was separated into 96 independent wells. The surfaces ofthe disks were functionalized by either isocyanate or aldehydechemistry, and then printed with protein spots. Each well wasthen printed with 64 protein spots that formed 16 2 × 2 unit cellsof target and reference spots on opposite diagonals. On half ofeach disk, the target spots were directly printed with captureantibody (BD Biosciences, TRFK-5), and, on the other half ofthe disc, the target spots were printed with protein A/G. Thewells that were printed with protein A/G were incubated withthe capture antibody. After the immobilization of the captureantibody, different wells of the disks were incubated staticallyfor 2 h with three different concentrations of IL5: 0, 50 pg/mL,and 1 ng/mL. The disks were then scanned with MI2 under 20×magnification. After the scan, the wells were incubated staticallyfor 1 h with secondary antibody (BD Biosciences, JES1-5A10)at a concentration of 100 ng/mL. The disks were scanned againafter the incubation of secondary antibody.

The prescan and postscan were registered and differencedto obtain the binding signal of the secondary antibody to the

Page 8: Molecular Interferometric Imaging for Biosensor Applications

ZHAO et al.: MOLECULAR INTERFEROMETRIC IMAGING FOR BIOSENSOR APPLICATIONS 1687

Fig. 10. IL-5 sandwich assay showing response as a function of antigen con-centration. The limit of detection of the assay is 50 pg/mL.

captured antigen. The height change differences for each pair oftarget and reference spots were calculated, and, for each concen-tration, the differences of all the pairs were averaged together.The results of the analysis for the direct-printed antibodies onthe isocyanate-coated disk are shown in Fig. 10. The error barson the response curve are set by the standard error of the pair re-sponses. At 50 pg/mL, the binding signal is distinguished fromzero by the error bar. The average height change at 50 pg/mL is15 pm, with an error bar of 7 pm.

The aforementioned results were for the directly printed cap-ture antibody. The same analysis was performed for the proteinA/G immobilized capture assays, and the results are particu-larly interesting. After incubating with the secondary antibody,the zero-concentration wells showed a large height increase,whereas, in the higher concentration wells, the height increasewas much smaller. The results are plotted in Fig. 11 as themaximum binding to A/G at zero concentration minus the spotresponse versus increasing antigen concentration. This suggeststhat under this experimental condition, the IL5 competes withthe immobilized protein A/G to bind with the secondary an-tibody. Therefore, it behaves as a competitive assay in whichhigher concentrations of antigen bind more secondary antibodyso that less secondary antibody is bound to the protein A/Gspot. Under this condition, the balance of the reaction is highlysensitive to the change in antigen concentration. At 50 pg/mL,the response difference for the aldehyde-coated disk is 250 pmwith a standard error of 20 pm. Similar results were obtained forthe isocyanate-coated disk, although with a smaller magnitude.The projected concentration detection limit for this competitiveassay of IL-5 against protein A/G is 3 pg/mL.

V. REAL-TIME BINDING KINETICS

The direct imaging character of MI2 enables real-time bind-ing measurements in which the surface is imaged through theincubant. This gives it the capability of observing antibody bind-ing in solution in real-time in a manner similar to surface plas-mon resonance sensors [15]. The main difference is that surface

Fig. 11. Competitive assay in which the antigen competes against the proteinA/G to bind the Fab secondary. Plotted is the maximum binding to A/G at zeroconcentration minus the spot response versus increasing antigen concentration.

plasmon resonance probes the interface using an evanescentfield, while, in MI2, the optical field penetrates the full fluidthickness above the surface. However, if the fluid sample is ho-mogeneous and nonturbid, it causes no degradation in the sen-sitivity of the measurements other than a decrease in the proteincontrast by a factor of 4 because water has a higher refractive in-dex than air. [We measured a decrease in contrast equal to 4, andcalculated the refractive index of protein by (np − nw )/(np −1) = 0.25, which yields np = 1.44.] Although the probe beamtravels through the fluid, if the fluid moves faster than analyteis depleted, only the incremental mass binding at the surfaceis detected because it is a differential measurement over time.One drawback of this approach relative to surface plasmon res-onance is that if the fluid is static, then the differential signalmay be reduced by the depletion of analyte above the antibodyspot. For this reason, we use actively flowing sample in all ourexperiments. This has not only the advantage of sweeping fluidaway from the spot before it is depleted, but also enables thesurface binding kinetics to be free of transport limitations.

In the initial kinetic experiments using MI2, a well on a sili-con wafer was covered with a 0.17-mm-thick microscope coverglass. Fluid was injected under the cover slip on one side, anddrained from the other. The cover glass was attached by double-sided tape to make the height of the flow cell approximately100 µm. The flow rate of the fluid was controlled by a valve, andwas typically about 100 µm/s for our experiments. The proteinspots were imaged directly through the cover slip. In this type ofreal-time detection, the disk is not shifted, and the data that aretaken at different times are normalized by the first image to re-move the spatial 1/f noise. The slow temporal drift in the illumi-nation background is removed by software. This measurementobtains differential surface height change as a function of time.

As a proof of principle, we measured the real-time bind-ing of rabbit IgG against goat antirabbit IgG with secondary

Page 9: Molecular Interferometric Imaging for Biosensor Applications

1688 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007

Fig. 12. Antigen exposure study with secondary amplification. The kineticcurves in (a) show the effect of increasing antigen incubation prior to theapplication of the secondary antibody. Longer incubation produces higher on-rates and larger saturation. The saturated relative height change is shown in (b)as a function of antigen exposure.

amplification. For this study, six wells were printed with goat an-timouse IgG (Sigma) and goat antirabbit IgG (Sigma), and wereincubated against two different concentrations of rabbit IgG,at 2 and 20 µg/mL, with different exposure times of 1, 3, and10 min. They were then incubated against 20 µg/mL of an-tirabbit IgG as the secondary antibody, and the resulting signalchange for the specific spots is plotted in Fig. 12(a). Longerincubation time caused higher captured antigen concentrationon the specific spots, and caused a faster height increase whenincubated against the secondary antibody. The oscillations in thereal-time binding are caused by the drift of interference fringesfrom the light reflected from the glass cover slip. Because therelative intensity change caused by the protein binding is ofthe order of 0.1% of the total intensity, the interference fringescause a nontrivial effect. We were able to reduce the effect of theinterference fringes by improving the attachment of the coverslip and the mechanical stability of the flow system. They canbe further reduced using antireflection-coated cover slips.

Fig. 13. Real-time binding in response to a series of protein exposures. Thedisk is printed with protein G that binds the Fc region of IgG. The first stepshows the binding of antibody to protein G. The second step is the binding ofantigen. The third step is the binding of the secondary sandwich antibody.

The saturation height of the binding curves in Fig. 12(a) is pro-portional to the captured antigen, which is plotted in Fig. 12(b)as a function of the antigen exposure. Longer exposure resultsin higher saturation height and higher reaction rate.

The process of real-time binding to printed protein G spotsis shown in Fig. 13. The vertical axis is the height change, andthe error bar is set by the standard error of 20 different proteinG spots in the field-of-view. An 84-well disk was printed withprotein G, and one well was selected for the real-time bindingexperiment. PBS buffer was first flowed through the flow cell,which sets the baseline of the measurements and drives theprotein spots to equilibrium. Then, the fluid was switched to20 µg/mL of antirabbit IgG. The antirabbit IgG was capturedonto the protein G spot, which saturated at a height of 3 nm. Then20 µg/mL of rabbit IgG was flowed through the system. Theimmobilized antirabbit IgG captured rabbit IgG in the solution,and gained 1 nm of height. Finally 20 µg/mL of antirabbit IgGwas applied again into the system and bound with the capturedrabbit IgG. The binding increased the spot heights by 5 nm. Allsolutions were prepared in 10-mM PBS buffer, and PBS bufferwas briefly applied during each change in protein solution tomake sure that the protein solutions did not mix together.

The successive step heights after each protein application inFig. 13 relate to the biological or chemical activity of the immo-bilized chemistry and capture molecules. The first step is causedby the binding of the Fc regions of the antibody to the printedprotein G. This step height is approximately 3 nm, correspond-ing approximately to a half-coverage of captured antibody. Thesecond step is caused by presenting antigen to the bound anti-body. This step height is only about 1 nm, suggesting that onlyabout one in three antibody molecules is capturing the antigen.The final step height is caused by the sandwich secondary an-tibody with a much higher step height of approximately 6 nm.This large step suggests that there are approximately six an-tibodies attached to each captured antigen. This secondary

Page 10: Molecular Interferometric Imaging for Biosensor Applications

ZHAO et al.: MOLECULAR INTERFEROMETRIC IMAGING FOR BIOSENSOR APPLICATIONS 1689

amplification may be possible because the secondary antibody isa polyclonal antibody that can bind to more than one epitope onthe antigen. It is further possible to produce polymer chains ofpolyclonal antibodies with the antigen acting as the crosslinker.Considerable future work is needed to study these effects, andto verify that the overlying fluid is not participating significantlyin the observed intensity changes.

VI. CONCLUDING REMARK

The revolution in digital cameras over the past decade hasprovided increasingly more powerful CCD cameras at approxi-mately level cost, making massively parallel channel acquisitionaccessible and commonplace in the laboratory and in the mar-ketplace. The CCD cameras are participating increasingly in thefield of biosensors, for instance, being used in imaging ellipsom-etry and in surface plasmon resonance systems, as well as in theconventional arena of fluorescent array readers. In this paper,we have introduced a new digital camera-based biosensor thatis particularly easy to implement to detect protein on modifiedsilicon surfaces with molecular sensitivity nearly down to thesingle-molecule limit.

The metrology limit of MI2 is 10-pm/pixel longitudinal res-olution at 0.4-µm diffraction-limited lateral resolution, corre-sponding to 1.7 attogram of protein. This is only about eightantibody molecules per pixel, which is near to single-moleculedetection. The scaling mass sensitivity of this technique is de-fined as the mass per square-root of detection area. This scalingsensitivity is an intrinsic property of the detection system, and isnot related to the area over which data are averaged. For instance,surface plasmon sensors typically quote values of 1–10 pg/mm2

as their smallest detectable surface mass density. However, thearea of their sensor is implicitly included in this value. In the caseof MI2, the intrinsic scaling at the metrology limit is 5 fg/mm.If an area of 1 mm is used to average the data, this correspondsto 5 fg/mm2 , which is 20–200 times more sensitive than typicalsurface plasmon sensors. This molecular-scale mass sensivity isdue to the shot-noise-limited performance of the CCD camera,combined with the nonresonant character of the in-line quadra-ture that makes it particularly low-noise and stable.

The metrology limit of MI2 is primarily of academic interest,while applications involve extensive handling of the disks, in-cluding wet chemistry. In the case of practical immunoassays onour specific surface chemistries, the scaling sensitivity increasesto approximately 40 fg/mm for assays under 40×magnification,and 200 fg/mm for assays under 7×magnification. This last sen-sitivity is still better than for most surface plasmon resonancesensors. Using this sensitivity, we have performed an assay forPSA with a detection limit of 60 pg/mL per spot pair in thepresence of high background. With 10 000 spot pairs on a disk,this would provide a significant resource for high-throughputassays. Conversely, if all the spots on the disk were used for asingle assay, the limit of detection would decrease to approxi-mately 1 pg/mL, which is at the most sensitive range for proteinsin signaling and regulatory pathways. Clearly, the simplicity ofMI2 combined with its molecular sensitivity show promise forfuture applications in high-throughput proteomics.

REFERENCES

[1] P. B. Luppa, L. J. Sokoll, and D. W. Chan, “Immunosensors—principlesand applications to clinical chemistry,” Clin. Chim. Acta, vol. 314, pp. 1–26, 2001.

[2] C. L. Tucker, J. F. Gera, and P. Uetz, “Towards an understanding ofcomplex protein networks,” Trends Cell Biol., vol. 11, pp. 102–106,2001.

[3] P. Uetz and R. L. Finley, “From protein networks to biological systems,”FEBS Lett., vol. 579, pp. 1821–1827, 2005.

[4] G. Gauglitz, “Direct optical sensors: Principles and selected applications,”Anal. Bioanal. Chem., vol. 381, pp. 141–155, 2005.

[5] X. Wang, M. Zhao, and D. D. Nolte, “Common-path interferometric de-tection of protein on the BioCD,” Appl. Opt., to be published.

[6] D. D. Nolte and F. E. Regnier, “Spinning-disk interferometry: TheBioCD,” Opt. Photon. News, vol. 15, no. 10, pp. 48–53, 2004.

[7] M. M. Varma, H. D. Inerowicz, F. E. Regnier, and D. D. Nolte, “High-speedlabel-free detection by spinning-disk micro-interferometry,” Biosens.Bioelectron., vol. 19, pp. 1371–1376, 2004.

[8] M. M. Varma, D. D. Nolte, H. D. Inerowicz, and F. E. Regnier, “Spinning-disk self-referencing interferometry of antigen–antibody recognition,”Opt. Lett., vol. 29, pp. 950–952, 2004.

[9] L. Peng, M. Varma, H. D. Inerowicz, F. E. Regnier, and D. D. Nolte,“Adaptive optical BioCD for biosensing,” Appl. Phys. Lett., vol. 86,pp. 183902-1–183902-3, 2005.

[10] M. Zhao, W. Cho, F. Regnier, and D. Nolte, “Differential phase-contrastBioCD biosensor,” Appl. Opt., vol. 46, pp. 6196–6209, 2007.

[11] M. Zhao, D. D. Nolte, W. R. Cho, F. Regnier, M. Varma, G. Lawrence, andJ. Pasqua, “High-speed interferometric detection of label-free immunoas-says on the biological compact disc,” Clin. Chem., vol. 52, pp. 2135–2140,2006.

[12] D. D. Nolte and M. Zhao, “Scaling mass sensitivity of the BioCD at0.25 pg/mm,” in Proc. SPIE 6380: Smart Med. Biomed. Sens. Technol. IV,Boston, MA, 2006, p. 63800 J.

[13] H. Ozen and S. Sozen, “PSA isoforms in prostate cancer detection,” Eur.Urol. Suppl., vol. 5, p. 495, 2006.

[14] S. P. Balk, Y.-J. Ko, and G. J. Bubley, “Biology of prostate-specific anti-gen,” J. Clin. Oncol., vol. 21, pp. 383–391, 2003.

[15] J. Homola, “Present and future of surface plasmon resonance biosensors,”Anal. Bioanal. Chem., vol. 377, pp. 528–539, 2003.

Ming Zhao was born in Taiyuan, Shanxi, China. He received the Bachelor’sdegree in physics from Tsinghua University, Beijing, China, in 2003. He is cur-rently working toward the Ph.D. degree in physics in the Department of Physics,Purdue University, West Lafayette, IN.

His current research interests include optical detection methods forbiosensing.

Xuefeng Wang was born in Jiangxi Province, China, on June 22, 1980. Hereceived the Bachelor’s and Master’s degrees in physics from Tsinghua Univer-sity, Beijing, China, in 2001 and 2004, respectively.

Currently, he is a Research Assistant in the Department of Physics, PurdueUniversity, West Lafayette, IN. His current research interests include fluores-cence and interferometry combination detection methods for protein microarray.

Gregory M. Lawrence was born in Grand Forks, ND, on September 25, 1961.He received the Graduate degree in Arts and Sciences from the University ofNorth Dakota, Grand Forks, in 1989. In 1992, he pursued graduate studies atPurdue University, West Lafayette, IN, with the primary focus on medicinal andorganic chemistry under Dr. D. Bergstrom.

In 1997, he pursued an opportunity to start a company based on novelinterfacially active resins and polymers. In 2002, he was a Senior Scientistat Polymer Technology Systems, Indianapolis, IN, where he was engaged inresearch on creating a point-of-care diagnostic system for the determinationof cholesterol from high-density lipoproteins, low-density lipoproteins, totalcholesterol, and triglycerides. He has also had a consulting career in formulatingagricultural adjuvant blends. Currently, he is a Senior Product DevelopmentChemist at Quadraspec, Inc., West Lafayette. His current research interestsinclude general immunological assay development including immobilizationmethodologies.

Page 11: Molecular Interferometric Imaging for Biosensor Applications

1690 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 13, NO. 6, NOVEMBER/DECEMBER 2007

Patricio Espinoza received the Ph.D. degree in electrical engineering from theUniversity of Pennsylvania, Philadelphia, in 2001, with the focus on processdevelopment.

He has been engaged in research on optical metrology, ceramics, stereolithog-raphy, multiphysics simulations, and silicon microfabrication. Currently, he isthe Director of Process Development at Quadraspec, Inc., West Lafayette, IN.His current research interests include the development of enabling technologiesfor label-free, multiplexed immunoassays.

David D. Nolte received the Ph.D. degree in physics from the University ofCalifornia at Berkeley, Berkeley, in 1988.

Thereafter, he was engaged in postdoctoral research at Bell Laboratories,Murray Hill, NJ. Currently, he is a Professor of Physics at Purdue Univer-sity, West Lafayette, IN. He is a Technical Founder of QuadraSpec, Inc., WestLafayette, a small business that is commercializing biological compact disk(BioCD) interferometric immunoassays. He is the author of Mind at LightSpeed (Free Press, 2001), which focuses on the future of optical computing. Hiscurrent research interests include the field of biophotonics, optical coherenceimaging, and BioCD technology.

Prof. Nolte is a Fellow of the Optical Society of America and the AmericanPhysical Society. He was the recipient of the Herbert Newby McCoy Award.


Recommended