PhD Candidacy Report

Anna Folinsky

California Institute of Technology

April 7, 2005

Contents

I Bacterial Disease Detection with Electronic Nose Systems

Progress Report 2

II Lung Cancer Detection Using An Electronic Nose

In Field Proposal 26

III Exploration Of A Binding Site In the ORL1 Receptor Using

Site-Directed Mutagenesis

Out of Field Proposal 44

1

Part I

Bacterial Disease Detection with

Electronic Nose Systems

Progress Report

2

3

Abstract

Vapor detectors based on “electronic nose” technology have been explored as po-tential diagnostic tools for a plethora of diseases. Here, chemical and physical systemsare designed to allow an electronic nose to serve as a detector for the bacteriologicaldiseases of bacterial vaginosis and urinary tract infection. A variety of sensor sub-strates are brought together to deal with the complexity of the problem, and a newsystem is created that will allow the smaller, less reliable sample size, to be dealt withsuccessfully.

Introduction

“Electronic noses” are chemical vapor detection systems, sharing certain characteristics with

the mammalian olfactory system.1 Similarly to natural olfaction, electronic nose systems

comprise a number of individual sensors (2-80, or even more). Unlike lock-and-key detection

systems, sensors in noses (natural and artificial) form what is called a cross-reactive array.

Each sensor responds to a variety of analytes, and in turn, each analyte elicits a response

from several sensors. The pattern of these responses from the sensor array (Fig.1) is then

sent to the processing device (the brain in biological systems, or a computer in laboratory

systems), processed by one of a variety of methods, and the results then interpreted.

Unlike natural systems, in which olfactory neurons are the sole detector basis, a wide

variety of substrates have been used as the sensing component in e-nose systems. The exact

signaling mechanism which is measured varies across all of them, but virtually all sensor

substrates respond by measuring uptake of the analyte of interest. This signal is transduced

to a numerical value, and is then passed to a processing system. Actual mechanisms used to

date include measuring the mass-dependent resonance frequency shift of coated quartz crystal

microbalances (QCMs)2 and surface acoustic wave devices (SAWs),3 measuring resistance

changes due to swelling of various polymer systems4, or observing the optical shifts from

4

Figure 1: Differentiation between odorants: (a) an array of broadly cross-reactive sensorsin which each individual sensor responds to a variety of odors; (b) pattern of differentialresponses across the array produces a unique pattern for each odorant or odor.

fluorescent compounds that can be observed with a CCD.5

Vapor detection systems based on electronic nose technology have been found useful

in a variety of fields. They have been used extensively in the food and drink industries,6

explored as nerve agent detectors,7 and studied in a variety of situations for environmental

monitoring, including the space shuttle,8 and other places where the ability to robustly deal

with complicated vapor mixtures is called for.

Medical diagnosis is another field where the power of this technology is being called into

play. A wide variety of diseases and infections create volatile markers released by the body, in

skin, urine, breath, and elsewhere.9 Some of these compounds are byproducts of physiological

changes wrought by the diseases themselves (alkanes on the breath breast cancer,10 ketones

on the breath in diabetes11), but the compounds detected in cases of bacterial infection are

often bacterial metabolites (as discussed below).12 As gas chromatography (GC) and later

mass spectrometry (MS) techniques became available the ability to quantitatively determine

which compounds were produced in the disease states became possible.12

In the cases of many of these diseases, diagnosis is time-consuming, invasive to the patient,

5

and expensive. Electronic nose systems have great potential to serve as early diagnostic

testing devices for several of these diseases. The need for more in-depth biological tests for

confirmation of these diseases will not be obviated, but a simple, inexpensive, lightweight

first level screen could be enormously beneficial, by narrowing in on only those who need the

further levels of testing.

Towards this end, electronic nose systems have been studied for diagnosis of a variety of

disease states. They have been examined as detectors for diabetes,13 lung cancer,14 urinary

tract infections(UTIs),15 rhinosinusitis,16 tuberculosis17 and bacterial vaginosis,18 mostly

with promising results.

Our group has agreements in place to explore detection of bacterial vaginosis and UTI.

The general method for diagnosing these diseases involves taking a biological sample (of

vaginal fluid or urine, respectively), culturing the samples for 24+ hours, and examining the

sample under a microscope to count the number of bacterial colonies formed.18,19 This can

take several days (adding in transport time, and heavy workload in the analytic laboratories),

requires a trained technician, and often ends up being used in cases where no disease is found.

An electronic nose-based quick detection system would cut back greatly on the number of

samples sent to the labs, more easily clear those not infected, and allow those who then

exhibit a higher chance of infection to quickly be started on an antibacterial regimen that

they otherwise might have had to wait several days for.

There have been initial reports of using electronic nose systems to detect both bacterial

vaginosis and cases of UTI,15,18 but in both of those cases the samples still required heating

and incubation. We believe that one of the greatest assets of using an electronic nose system

will be its ability to detect these disease states in near real time (on the order of minutes), and

6

thus are working towards being able to detect the analytes of interest from the un-incubated

samples.

We have begun designing a system to detect these characteristic vapors from biological

samples, by headspace testing of the samples with our electronic nose systems. This has

required developing novel configurations of vapor delivery systems and sensor substrates

previously tested, but not used extensively, in our labs. Progress has been measured to date,

but many details of the system have been worked out.

Background

Diseases

Bacterial Vaginosis

Bacterial vaginosis (BV) is very commonly found in women of childbearing age. No specific

bacteria have been causally implicated, but a general shift from the normal lactobacillus

mix to a mixed vaginal anaerobic flora including Gardnerella vaginalis, Bacteroides spp.,

and Mobiluncus spp.20 is seen. Several studies have reported finding amines in the vaginal

fluid of BV patients,21,22 and a gas chromatographic study showed this more quantitatively,

by positively correlating increased levels of phenethylamine, putrescine, cadaverine, and

tyramine in derivatised samples from BV patients.23

There has also been a brief report of a commercial e-nose system having initial limited

success in differentiating treated samples from patients with and without BV.18 However,

their system was non-optimized, only yielded a positive predictive value of 61.5%, and re-

7

quired sample pre-treatment. Based on the above, we feel that the noted amine signature

and lack of further e-nose results make BV a promising candidate for our technology.

Urinary Tract Infections

UTIs are caused by a variety of bacteria, but by far the most common cause is Escherichia

coli (with reports ranging from 58-90% of causation).19 There have been several GC studies

of headspace over E. coli infected urine samples,24,25 which were generally successful in

identifying samples. However, not only were these methods time and labor intensive (a

large cause of GC methods not being further pursued as a diagnostic technique), all of these

studies required artificial stimulation of the E. coli bacteria to induce ethanol production,

which was then readily detected via GC.

One study used a commercial electronic nose coupled with a neural network system to

distinguish UTI infected urine samples from controls.15 Their results were encouraging, with

the trained system able to correctly identify 18 out of 19 unknown UTI cases. However,

again, this system is not optimized in any way. The use of a commercial sensor system does

not afford them control over their sensor substrates and they therefore cannot change it to

improve performance, although they failed to report the identity of any of the analytes they

believed they were detecting. Furthermore, the testing scheme used still required a 4-5 hour

incubation period of the samples.

E. coli have been extensively studied, however, and one of the many things known about

them is that they produce many different fatty acids.26 While most of the fatty acids are

not particularly volatile (especially at room temperature), it is quite possible that some of

the shorter carboxylic acids could be found in the headspace above the samples. This gives

8

us a handle to begin actual design of a sensor system to detect these samples.

Sensor Substrates

There have been many different sensor substrates explored in the last decade in our lab.27–30

Most of these have been further explored in these experiments for use as disease diagnostics.

The most commonly used in our lab, and the best explored, consist of polymer/carbon

black (CB) composites. The carbon black particles serve as the conducting medium in the

sensor layer. As the composite is exposed to analyte, the vapor partitions into the polymer.

This leads to swelling of the composite and causes the average distance between carbon black

particles to increase, which in turn causes an increase in the resistance of the sensor. The

degree to which the analyte sorbs depends on the interactions between it and the sensor,

which varies due to specific makeup of the analyte and polymer. This swelling response

has been shown to be linear with respect to both concentration and mass loading from the

limit of detection to the point where the percolative effects cause a dramatic increase of

resistance.31 As this happens only at very high mass loadings, it is easy to keep within these

boundaries.

Substrates similar to these, but using a dendrimeric instead of of a polymeric basis

have recently been explored.28 These substrates have a high density of basic or acidic func-

tional groups (amine, protonated amine, and protonated carboxylates) (Fig 2). It was pro-

posed that the ability of these substrates to undergo proton-transfer reactions with acidic

or basic analytes would yield a much increased differential resistance change compared to

the traditional non-dendrimeric polymer sensors. It was found that the amino-terminated

polypropylenimine (DAB-AM) dendrimers showed an enhancement of sensitivity on the or-

9

Figure 2: Structures of the PAMAM and DAB-AM dendrimers, and a protonated terminalamino group. Each generation number represents one further exponential level of growth

der of 103 towards carboxylic acids, when compared to the non-dendrimeric polymer/CB

composites. Similarly, protonated DAB-AM and polyamidoamine (PAMAM) dendrimers

showed an equivalent increase in sensitivity towards volatile amine vapors.

Another substrate demonstrating increased sensitivities to amines are acid doped polyani-

line compounds.27 Acid doping of the chain to its half-oxidized, “emeraldine base” form

induces a conformational change , aligning the aniline rings such that their pi-systems over-

lap, and the polyaniline becomes electrically conductive. These sensors show responsive-

ness to analytes not only via swelling mechanisms, but also through their ability to be

doped/dedoped. They do show increased sensitivity to amines, and were included in the

initial experiments of this project. However, they show a distressing tendency to become

permanently dedoped, and rendered completely insulating. For this reason, they are not

10

Figure 3: Structure of normal and doped polyaniline. HFIP = hexafluoroisopropanol.

being explored any further at this time as a disease diagnostic sensor.

Even more recently, our lab has begun examining the properties of non-polymeric com-

pounds for sensing.32 Using much smaller, non-volatile compounds greatly increases the

potential density of functional groups and provides for a much greater library of materials

for use than is possible when restricted to solely polymeric compounds. When coupled with

carbon black, these perform comparably to, and in some cases, better, than polymer/CB

systems. Little work has been done to date to further optimize the properties of these sys-

tems, but their inclusion in these experiments should be able to aid with the specificity and

sensitivity that will be required.

Principal Components Analysis

The data initially taken from the experimental setup are a long stream of resistances read

from each sensor over time. From this data, for each exposure to an analyte from each

11

sensor, we extract a value:

∆Rmax

Rb

(1)

Where Rb is the baseline resistance of the sensor prior to exposure, and ∆Rmax is the

maximum change in steady state resistance (Fig 4)

Figure 4: Generalized response of one sensor to a single analyte

This yields R = {rij}, an m×n matrix of sensor values, where n is the number of sensors,

m is the number of exposures, and rij represents the response of the jth sensor to the ith

exposure of analyte, as shown in Eqn 1. This still leaves us with the problem of having data

in n-space, which is difficult to interpret and visualize. Principal components analysis (PCA)

33 is a multivariate statistical technique employed to reduce the dimensionality of the data,

and make it more amenable to interpretation. This is a common method used in pattern

analysis, and has been extensively used and reviewed in the electronic nose literature.34 This

is the analytic method used to date on this project, although others may be used later as

necessary.

The matrix R is first preprocessed such that each column in the matrix is normalized

and autoscaled (ie, centered about the mean and defined to have unit standard deviation,

resulting in a final matrix D = {dij}. First the rij values are normalized, creating the matrix

12

Q = {qij} which helps correct for differences in solvent vapor pressure.

qij =rij∑

j

rij

(2)

These normalized values are then autoscaled, such that they are both mean-centered and

set to have a standard deviation of unity.

dij =qij − qj

σj

(3)

Here, qj and σj represent the mean and standard deviation of each sensor j to all analytes

presented to it. This matrix D = {dij} is then diagonalized (ie, multiplied by its transpose)

to obtain a correlation matrix M.

M = DT · D (4)

The eigenvalues and eigenvector matrix of M are then obtained. The n eigenvectors of

the eigenvector matrix V are mutually orthogonal. We multiply this n× n matrix V by the

data matrix D to obtain our matrix of principal components, P, an m× n matrix, in which

each row is still associated with a particular analyte exposure, and each column is now a

principal component of the data, in which the maximal variance between the members of the

original data set is found in the first principal component, the maximal remaining variance

found in the second component, and so on. The corresponding eigenvalues of M tell us how

much of the total variance is to be found in each principal component.

13

P = DV (5)

The maximal amount of variance is now front loaded into the first few principal compo-

nents, allowing us to much more easily visualize the information in the data in just 2 or 3

dimensions, rather than the full n-dimensionality of the original sensor set.

Experimental

All sensors used in these experiments consist of the detective layer sprayed across a gap

between two conductive gold leads. The gold layer is deposited on top of a chromium layer,

for adhesion purposes, with a glass slide acting as support for the sensor. Some experiments

were done with all pairs of sensor leads being deposited on a single microscope slide, in

which case the slide has 15 pairs of leads on it, with a gap of ∼0.5 mm between the two

leads (Fig 5a). The larger sensors are ∼5 mm in width pieces cut from a larger slide, and

have a gap of ∼1 mm between the conductive leads.

The sensor slides are used in a chamber designed specifcally for them that allows much

smaller volumes and flow rates to be dealt with(Fig 5b,c). Through experimentation, a flow

system with computer switching, but manual flow and sample control has been designed

(Fig 6). The larger sensors are housed in an airtight chamber, with sensors placed sequen-

tially along the path of the vapor flow. The vapor delivery systems consists of mass flow

controllers connected to solvent bubblers, which allow controlled flow rates of saturated an-

alyte vapor to be mixed with background laboratory air. In this manner, the system can

generate analytes at fractional partial pressures from 0.5% P/P0 to near saturation levels.

14

Figure 5: a)View of the sensor slide itself. b)End on view of the slide in the slide chamber.c) Angled view, showing the slide in the chamber, and direction of airflow. After Briglin et.al.35

The electronic apparatus necessary to acquire the data signals consists simply of a com-

puter controlled multiplexer and multimeter to read the resistances across the sensors, and

the computer itself to record the data. The data are then removed to other computers for

analysis, which is done primarily with MATLAB.

15

Background flow

WaterBubbler

a)ClearPath

b)Sample Path

3−way 3−way

sample vial

blank vial

sensor chamber

Figure 6: Schematic of the headspace sampling technique for use with urine samples. Theflow is run through a water bubbler to create a saturated water background, and computerswitched to run through either a background pathway, or through the headspace of thesample

Results

This project has moved in somewhat of a circular nature. We began at the end, seeing if the

problem was amenable to immediate attack of biological samples. Upon discovering that the

complexity of the situation made that not possible, we moved back step by step such that

we are now finally back to the beginning of the problem, starting to move forward again.

We received urine samples from Dr. Harvey Kasdan at ProIris , Inc. These were broken

up into four categories: a) “smelly” samples, b) filtered, then spiked with Level 1 E. coli, c)

filtered, then spiked with Level 2 E. coli, and d) filtered, clean samples.

GC-MS was done of the headspace of the samples as delivered, but no response was seen.

In comparison, saturated headspace over a pure butylamine sample had a clear peak, a 1-in-3

dilution had a weak peak, and a 1-in-20 dilution didn’t show up at all.(Figs. 7 and 8)

We attempted to run these samples by injecting one mL of the headspace into a sealed

16

Figure 7: GC/MS of headspace over pure butylamine

0 2 4 6 8 10 12 14 16 18 201

1.5

2

2.5

3

3.5

4

4.5

5x 10

5

time(min)

coun

ts

GC/MS of 1−in−3 dilution of butylamine headspace

Butylamine peak

Figure 8: GC/MS of 1-in-3 dilution of butylamine headspace (Rise at the end is from thecolumn itself)

17

0 0.5 1 1.5 2 2.5

x 104

5500

6000

6500

7000

7500

8000

time(s)

resis

tance

(ohm

s)

PAMAM 3.5 sensor

Open

close

Injection

Figure 9: Response of a PAMAM 3.5 sensor to opening and closing of the chamber. Thespikes in the middle of the peaks are temporary responses to increases in pressure frominjection of the sample

sensor slide chamber (after realizing that the larger chamber would be entirely inappropriate

for a sample size of that volume) , containing a slide with nine sensors sprayed on it (Table 1).

The chamber was then purged by opening and closing of the chamber itself. However, the

sensors proved to show scant response to the actual injection, with much greater response

for the chamber openings and closing. (Fig 9)

Polyaniline Dendrimer PolymerPAni/HCl PAMAM 3.0 Poly(vinylstearate)PAni/MSA PAMAM 3.0 w/1:1 MSA Poly(ethylene-co-vinylacetate)PAni/NSA PAMAM 3.5 Hydroxypropylcellulose

Table 1: Sensor substrates used on all sensor slides. All polyanilines are doped to a 2:1PAni/acid ratio. All dendrimers are mixed with 50% carbon black, and all polymers aremixed with 40% carbon black.

We realized that static headspace was not going to work and that we needed to move on

to a flow system more like the ones we had worked with previously. However, testing urine

samples is somewhat different than being able to generate, at will, an essentially infinite

18

0 1000 2000 3000 4000 5000 6000 70000.7

0.8

0.9

1

1.1

1.2

1.3x 10

4

time(s)

resis

tance

(ohm

s)

PAMAM 3.5 sensor

Sample flowBlank

Figure 10: Response of a PAMAM 3.5 sensor to samples, with a 5 mL/min flow rate

stream of controlled analyte vapor. We set up a system to allow constant flow through

the sensor slide chamber (similar to Fig 6). One side of the pathway is the “background”

pathway flow, and remains untouched during a run. The sample vial has a sealed cap with

a PTFE coated septum in it, and these vials are alternated by hand. Between each sample

run a blank vial of the same size is swapped into the same position, to allow the sample

pathway time to purge.

A sample run through this system was initially promising, with sharp sensor responses

that cleared out nicely when a blank was applied to the system (Fig 10), but it short order

it became apparent that this was mostly water response. When exposed to a saturated

background flow of water, we saw no response at all to eight urine sample exposures (Fig 11).

At this stage, we realized this would be more complex than initially hoped, and moved one

step back. Using the same slide system, with continued use of a saturated water background,

we moved on to more controllable analytes. In these cases, we could see clear responses to

the headspace of 1- and 2.5% solutions of butylamine in water (Fig 12).

19

0 1000 2000 3000 4000 5000 6000 7000 80001

1.2

1.4

1.6

1.8

2

2.2

2.4x 10

4

time(s)

resis

tance

(ohm

s)PAMAM 3.5 sensor

First exposure to backgroundl

Figure 11: Response of a PAMAM 3.5 sensor to samples, with a 5 mL/min flow rate of sat-urated water background. The dip seen is the response to the first switch to the backgroundpathway, which had to reach the saturated water levels

0 0.5 1 1.5 2 2.5

x 104

1

1.2

1.4

1.6

1.8

2

2.2

2.4

2.6

2.8x 10

5

time(s)

resis

tance

(ohm

s)

HPC sensor

1% sample

2.5% sample

Figure 12: Response of an HPC/40%CB sensor to the headspace of two concentrations ofbutylamine in water. The first arrow marks the 1% sample, and the second the 2.5% sample

20

Continued experimentation with the small scale system, however, revealed inconsistency

of responses, and more importantly, extremely slow response times (on the order of thousands

of seconds), which yielded responses that were extremely difficult to deal with analytically.

We decided to start over again, and build a system towards this final goal, rather than seeking

to achieve it all at once. However, all these initial stages did garner invaluable information

that will be quite useful when it comes time to go back to the biological samples.

We then moved back to the larger, computer controlled system, where analyte deliv-

ery could be more carefully calibrated. It was at this time that we moved away from the

polyaniline sensors as substrates, and began to incorporate the non-polymeric sensors into

the array.

The larger system, as we had hoped, let us obtain much cleaner responses, upon which

we could actually perform data analysis. A system of ten sensors let us discriminate between

repeated exposures to butylamine, diethylamine, and isopropanol, all at 1% partial pressure

(Fig 13)

After that initial success, we most recently turned to look at these analytes when a

saturated water background was present. (Fig 14). Analyte discrimination is reasonable,

although not quite as clear as we would wish , we can definitely see that the sensors can tell

apart “water” from “water + analyte”. This is quite promising, in that we now know that

the sensors are not just responding to the large amount of water vapor presented to them,

and that there is real potential to differentiate between the analytes presented, even if they

are largely overshadowed by the other vapor.

21

−5 −4 −3 −2 −1 0 1 2 3−3

−2

−1

0

1

2

3

4

5

6

7

PC1

PC2

PCA of 1% P/P0 exposures

butylaminediethylamineiPrOH

Figure 13: First two principal components of sensors exposed to 1% P/P0 of analytes vapor.Total flow rate was 2.5 L/min.

−8 −6 −4 −2 0 2 4−3

−2

−1

0

1

2

3

4

5

PC1

PC2

PCA of 1% P/P0 exposures

butylaminediethylamineiPrOHacetic acidwater

Figure 14: First two principal components of sensor responses to analytes with a saturatedwater background, and also a pure water background.

22

Future Work

Now that the system has been setup to begin the sampling, future work is laid out fairly

clearly. The first step in the ongoing work will be to finalize sensor systems that can robustly

discriminate between various carboxylic acids, or various amines. Once that is complete, the

systems will need to be further tested on their discriminatory ability between different mix-

tures within those groups (ie, varying either the mixture itself, or different concentrations

of mixtures). Further GC-MS studies using purge-and-trap or other preconcentration tech-

niques will be undertaken with actual biological samples to determine more accurately which

mixtures need to be detected. Then this differentiation ability will need to be shown with

respect to a water background, which is the situation we would expect to find in the case of

almost any biological sample.

Once the samples and mixtures with a water background can be clearly determined,

the system will be switched to one of headspace sampling, on the much smaller scale that

we would use with biological samples. This will lead to the testing of actual biological

samples, first returning to those which are spiked with known bacterial concentrations, and

then moving on to samples from actual suspected ill patients, leading eventually to our final

goal of being able to robustly and easily discriminate between samples which show signs of

bacterial infection and those which belong to healthy patients.

Bibliography

[1] Lewis, N. S. Accounts of Chemical Research 2004, 37(9), 663–672.

[2] Fu, Y.; Finklea, H. O. Analytical Chemistry 2003, 75(20), 5387–5393.

[3] Ballantine, D. S.; Rose, S. L.; Grate, J. W.; Wohltjen, H. Analytical Chemistry 1986,58(14), 3058–3066.

[4] Partridge, A. C.; Jansen, M. L.; Arnold, W. M. Materials Science & Engineering C-

Biomimetic and Supramolecular Systems 2000, 12(1-2), 37–42.

[5] James, D.; Scott, S. M.; Ali, Z.; O’Hare, W. T. Microchimica Acta 2005, 149(1-2),1–17.

[6] Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Trends in Biotechnology 1998,16(6), 250–258.

[7] Hopkins, A. R.; Lewis, N. S. Analytical Chemistry 2001, 73(5), 884–892.

[8] Ryan, M. A.; Lewis, N. S. Enantiomer 2001, 6(2-3), 159–170.

[9] Hayden, G. F. Postgraduate Medicine 1980, 67(4), 110–.

[10] Phillips, M.; Cataneo, R. N.; Ditkoff, B. A.; Fisher, P.; Greenberg, J.; Gunawardena,R.; Kwon, C. S.; Rahbari-Oskoui, F.; Wong, C. Breast J 2003, 9(3), 184–91.

[11] Smith, M.; Levinson, B.; Smith, L. G. Lancet 1982, 2(8313), 1452–1453.

[12] Pavlou, A. K.; Turner, A. P. Clinical Chemistry & Laboratory Medicine 2000, 38(2),99–112.

[13] Mohamed, E. I.; Linder, R.; Perriello, G.; Daniele, N. D.; Poppl, S. J.; Lorenzo, A. D.Diabetes, Nutrition & Metabolism - Clinical & Experimental 2002, 15(4), 215–21.

[14] Natale, C. D.; Macagnano, A.; Martinelli, E.; Paolesse, R.; D’Arcangelo, G.; Roscioni,C.; Finazzi-Agro, A.; D’Amico, A. Biosensors and Bioelectronics 2003, 18(10), 1209–1218.

[15] Pavlou, A. K.; Magan, N.; McNulty, C.; Jones, J. M.; Sharp, D.; Brown, J.; Turner, A.P. F. Biosensors and Bioelectronics 2002, 17, 893–899.

23

BIBLIOGRAPHY 24

[16] Mohamed, E. I.; Bruno, E.; Linder, R.; Alessandrini, M.; Girolamo, A. D.; Poppl, S. J.;Puija, A.; Lorenzo, A. D. An Otorrinolaringol Ibero Am 2003, 30(5), 447–57.

[17] Pavlou, A. K.; Magan, N.; Jones, J. M.; Brown, J.; Klatser, P.; Turner, A. R. Biosensors

& Bioelectronics 2004, 20(3), 538–544.

[18] Chandiok, S.; Crawley, B. A.; Oppenheim, B. A.; Chadwick, P. R.; Higgins, S.; Persaud,K. C. Journal of Clinical Pathology 1997, 50(9), 790–1.

[19] Wilson, M. L.; Gaido, L. Clin Infect Dis 2004, 38(8), 1150–8.

[20] Spiegel, C. A. Clinical Microbiology Reviews 1991, 4(4), 485–502.

[21] Chen, K. C. S.; Forsyth, P. S.; Buchanan, T. M.; Holmes, K. K. Journal of Clinical

Investigation 1979, 63(5), 828–835.

[22] Chen, K. C. S.; Amsel, R.; a. Eschenbach, D.; Holmes, K. K. Journal of Infectious

Diseases 1982, 145(3), 337–345.

[23] Wolrath, H.; Forsum, U.; Larsson, P. G.; Boren, H. Journal of Clinical Microbiology

2001, 39(11), 4026–31.

[24] Manja, K. S.; Rao, K. M. Journal of Clinical Microbiology 1983, 17(2), 264–6.

[25] Coloe, P. J. Journal of Clinical Pathology 1978, 31(4), 361–4.

[26] Magnuson, K.; Jackowski, S.; Rock, C. O.; Cronan, J. E. Microbiological Reviews 1993,57(3), 522–542.

[27] Sotzing, G. A.; Phend, J. N.; Grubbs, R. H.; Lewis, N. S. Chemistry of Materials 2000,12(3), 593–+.

[28] Detection and classification of volatile organic amines and carboxylic acids using arraysof carbon-black dendrimer composite vapor detectors (submitted). Gao, T.; Tillman,E.; Lewis, N. S.

[29] Lonergan, M. C.; Severin, E. J.; Doleman, B. J.; Beaber, S. A.; Grubb, R. H.; Lewis,N. S. Chem. Mater. 1996, 8(9), 2298–2312.

[30] Tillman, E. S.; Koscho, M. E.; Grubbs, R. H.; Lewis, N. S. Analytical Chemistry 2003,75(7), 1748–1753.

[31] Severin, E. J.; Doleman, B. J.; Lewis, N. S. Analytical Chemistry 2000, 72(4), 658–668.

[32] Array-based vapor sensing using chemically sensitive, carbon black - monomeric organicmolecule resistors (submitted). Gao, T.; Woodka, M.; Brunschwig, B. S.; Lewis, N. S.

[33] Duda, R. O.; Hart, P. E. Pattern classification and scene analysis; Wiley: New York,,1973.

BIBLIOGRAPHY 25

[34] Jurs, P. C.; Bakken, G. A.; McClelland, H. E. Chemical Reviews 2000, 100(7), 2649–2678.

[35] Briglin, S. M.; Freund, M. S.; Tokumaru, P.; Lewis, N. S. Sensors and Actuators B-

Chemical 2002, 82(1), 54–74.

Part II

Lung Cancer Detection Using An

Electronic Nose

In Field Proposal

26

27

Abstract

Electronic nose systems have been used both in and out of laboratory settings fora wide variety of applications. One of the larger general initiatives is towards usingthem in biomedical applications, notably the detections of analytes which may becorrelated with disease states. This proposal outlines a system to detect biomarkersfound in the breath which are associated with lung cancer, one of the leading causesof death in the USA. Use of standard polymer/carbon black sensors from our lab willbe augmented with novel sensing technologies from our lab, in an effort to obtain thenecessary sensitivity and discriminatory power for the task. These should be able todifferentiate diseased and healthy states, and potentially separate different stage cancerpatients.

Introduction

Lung Cancer

It has long been known that certain diseases produce volatile compounds that can be smelled

on the breath, or elsewhere on the body. Two of the oldest, most common examples are the

scents of ketones on the breath of diabetic patients, and the smell of “freshly baked bread” on

the skin of patients with typhoid.1 More recently, GC studies have shown quantitatively that

many diseases produce distinct patterns of volatile compounds that could potentially be used

as biomarkers for these diseases.2–5 Among these, lung cancer has been shown by several GC-

MS studies to display elevated levels of several volatile organic compounds (VOCs) on the

breath, mostly C4 to C20 monomethylated alkanes, in addition to certain benzene derivatives,

6,7 although there is some variaton between studies as to precisely which compounds are

most indicative. On one occasion, Phillips et al identified styrene, 2-methylheptane, and

decane and three of the most important compounds for discrimination;8 in another, butane,

3-methyltridecane, and 4-methyloctane2 were most informative.

Lung cancer is one of today’s leading health problems. It was the third leading cause

28

of death in the USA in 2002, behind only heart and cerebrovascular diseases. Estimates

indicate that 172,570 new cases will be reported in 2005, and 163,510 deaths from lung

cancer are expected the same year. The current 5-year survival rate for lung cancer is 15%,

but this rate rises to 49% if the cancer is discovered when it is still localized. Unfortunately,

only 16% of lung cancer cases are discovered while they are still localized.9 There is clearly

a great need to improve early detection ability for this disease.

Bronchoscopy, biopsy, and sputum cytology are the current most common ways to di-

agnose lung cancer, but these methods can occasionally miss tumors, and are dependent

on tumor size.10 There are also some reports of new methods for earlier detection, such

as fluorescence bronchoscopy,11 spiral CT scanning,12,13 PCR sputum assays,14 and using

computers to aid in the analysis of chest radiographs.12 All of these, however, are expensive

and time consuming. A non invasive breath test would have great potential as a widespread

screen.

Electronic Nose Systems

The vapor sensing method we pursue in our lab involves the use of arrays of sensors. No sensor

is designed to respond specifically towards an individual compound. Instead, each sensor

is broadly responsive to a variety of odorants. Each analyte produces a distinct fingerprint

from the array of broadly cross-reactive sensors (Fig.1). Pattern recognition algorithms

can then be used to obtain information on the identity, properties, and concentration of

the exposed vapor.15–18 In this respect, our system resembles that used in the mammalian

olfactory system, in which each olfactory receptor responds to a wide variety of odorants,

19 and our array of sensors may be seen as analogous to the array of receptors in the nasal

29

epithelium. Due to this similarity, our system is sometimes designated as an “electronic

nose”.

Figure 1: Differentiation between odorants: (a) an array of broadly-cross reactive sensorsin which each individual sensor responds to a variety of odors; (b) pattern of differentialresponses across the array produces a unique pattern for each odorant or odor.

A variety of signal transduction mechanisms have now been implemented to construct

electronic nose systems. Surface acoustic wave devices (SAWs),20 metal oxide sensors,21 con-

ducting organic polymers,22 polymer coated quartz crystal microbalances (QCMs),23 poly-

mer coated micro-machined cantilevers,24 thin film capacitors,25 and polymer composite

chemically sensitive resistors26 have all been used.

Except for metal oxide sensors, they all share the trait that the analyte in the vapor

phase is sorbed into a film coated onto the transducer, and the differential sorption of the

various analytes into an array of such films generates the distinct pattern produced by the

devices. Differences then result from varying the sorptive coating, general convenience of use

and ability to reduce noise. Additionally, some variance is found in the complexity of the

required data analysis and data reduction algorithms.

Our focus to date has been on sensors prepared from composites of conductive carbon

black particles interspersed with insulating polymers. These are low power, inexpensive,

30

lightweight, and allow for simple transduction circuitry as well as simple data analysis meth-

ods. The sorption of a vapor induces swelling in the polymer, which reduces the number of

conductive pathways in the film, and increases the resistance of the film, which is readily

measured (Fig. 2).

Figure 2: Conceptual image of swelling induced resistance changes of conducting compositevapor sensors

At small swellings, the sensors are highly reversible and respond rapidly to the presence of

analyte vapors, demonstrated reproducibly over thousands of cycles over a period of months,

to a variety of organic vapors. The thin films have been used to obtain response times in

essentially real time, with rise and fall times of less than 100ms in most cases and less than

20 ms in some systems27 (limited by the diffusivity of small molecules permeating through

low glass transition, rubbery, polymeric films at room temperature).

This proposal will be centered upon this technology, to be augmented by newer sensor

substrates from our lab.

31

Proposed Studies

Rationale

This proposal will use the various sensor systems we have been developing to detect lung

cancer biomarkers in patients’ breath. GC studies showing that several VOCs were elevated

in the breath of lung cancer patients were able to use discriminant analysis on the alveolar

gradient of the GC-MS information to distinguish between healthy and cancerous patients.

2,3 Certain compounds of interest (styrene, decane, and isoprene among them) are found at

1-20 parts per billion (ppb) in healthy breath, but are seen at levels from 10-100 ppb in

cancerous patients.28

Yu and coworkers used a GC column coupled with a SAW sensor and a neural network to

discriminate between healthy and diseased patients,29 with preconcentration. Another study

used an array of coated QCM sensors, with no preconcentration,30 to detect lung cancer in

patients prior to surgical tumor removal. They made no effort to quantify the analytes

detected, nor correlate them to stage of disease. However, they did show that the signature

returned to a “healthy” pattern approximately one month after removal of the tumor in a

pair of cases, providing further evidence that the analytes are related to the disease.

Coated QCMs are comparable in sensitivity to SAW devices, however, carbon black /

polymer sensors have been shown to be more sensitive towards alkanes and related VOCs;31

therefore the initial step of detection should be easily obtainable.

Our lab has already demonstrated the ability to detect and discriminate between all

the low molecular weight straight-chain alkanes, as seen in Fig. 3 and to quantify their

concentrations based on the amplitude of recorded patterns.32 Separately, it has been shown

32

that patterns for alkanes and those for a variety of aromatics can be differentiated.33 We

have also shown that even very similar alkane mixtures can be discriminated.26 A binary

mixture of n-heptane at P/Po = 0.0011 (64 ppm) and n-hexane at P/Po = 0.00090 (192

ppm) has been differentiated from a binary mixture of n-heptane at P/Po = 0.00090 (58

ppm) and n-hexane at P/Po = 0.0011 (235 ppm) with 95% correct discrimination, using an

optimal subset of a chemiresistive detector array. Similarly, a mixture of 1-propanol at P/Po

= 0.0025(72 ppm) and 2-propanol at P/Po = 0.0025 (150 ppm) could be differentiated with

98% correct classification rate from a mixture of 2-propanol at P/Po = 0.0027 (164 ppm)

and 1-propanol at P/Po = 0.0023 (64 ppm).

Current Methods

Human breath is composed mostly of water vapor, with all other analytes existing only

as minor substituents. This fact implies the need to deal not only with the low levels

of the analytes of interest but also with the high background level of water. While the

level of water is considered a primary obstacle for commercial sensor arrays, whose polar,

inherently conducting polymer sensors (made from inherently conductive materials, such

as polyaniline, polypyrrole, polythiophene, etc) are highly sensitive to water vapor, our

approach of using composites of inorganic conductors and sorptive insulating organic phases

allows development of chemiresistive sensors that are relatively insensitive to water vapor.19

There are also a pair of newer sensing substrates being tested, expected to yield improved

sensor classification. The first of these is based on composites of homogenous or blended

organic nonvolatile molecules with conductors such as carbon black.34 These sensors have

been constructed from either pure compounds or mixtures of moderate length monomeric

33

Figure 3: Principal components data from a 20-detector array exposed 5 times to of thelabeled analytes, each at 0.005 - 0.03 P/Po, containing 99% of the total variance. Theellipsoids contain 99% of the data for each analyte. All presentations were in each setrandomized over all repetitions

organic molecules used as binders, mixed with carbon black. The sensors show fast response

time, good reversability, and high stability. Furthermore, they show the ability to discrimi-

nate and classify both similar and different types of analytes, even at low concentrations in

air, compared to similar sensors based on polymer binders. They also allow an even broader

choice of substrates, and also allow us to achieve a higher density of functional groups in the

thin films than we are allowed by using the polymer composites. Principal component anal-

ysis of the sensor array responses showed clear distinctions between different types of vapors

34

allowing easy classification. This approach is particularly applicable to the development of

biosensors where the ability to detect low concentrations of specific types of compounds is

required.

The second approach uses ligand-capped Au nanoparticles. Thiol capped gold nanopar-

ticles in the range of 2-10 nm have been synthesized and tested as sorption based detectors

for different analytes.35 Thiols investigated varied by chain length, polarity, and functional

group. Effects of chemical functionalization of the thiol on sensor sensitivity , specificity, and

stability were studied. Sensor specificity varied with the thiol functional group. Most thio-

late gold nanoparticles chemiresistors showed good stability over three months, with longer

chain length thiols having better stability than short chain lengths, and a high ability to

classify and discriminate analytes according to their polarity and vapor pressure. In specific

cases, the selectivity and discrimination of such arrays were superior to arrays of carbon

black/polymer composite sensors.

Recent work in our lab has shown that such sensors can differentiate between mixtures

of long chain alkanes, at very low concentrations, in the presence of saturated water va-

por. Two mixtures of nonane and hexadecane were tested in background flows of saturated

water vapor at room temperature, to roughly simulate human breath (human breath is at

34.5o C saturated water vapor; however, the difference is minor for these demonstration

purposes). Standard exposures consisted of a background stream of water-vapor saturated

air which contained one of two predetermined mixtures of nonane and hexadecane. The

low concentration mixture had nonane at P/Po = 0.0010 (4 ppm) and hexadecane at P/Po

= 0.010 (40 ppb) in saturated water vapor, and was representative of the key alkane VOC

components of a healthy patient’s breath, while a higher concentration mixture of nonane

35

at P/Po = 0.0050 (20 ppm) and hexadecane at P/Po = 0.050 (200 ppb) in saturated water

vapor was used to be representative of the key elevated concentration VOC components that

have been reported as diagnostic of lung cancer. Sixty exposures to each breath type were

performed at flows of 5 L/min and data were obtained for carbon black/polymer compos-

ites, carbon black/monomer composites, and ligand capped Au nanoparticles. Other than

selection for sensors we knew empirically to provide generally good responses, the sensors

were not optimized for the task.

Figure 4: Principal componenent plots indicating the ability to distinguish between “healthy”and “cancerous” patients. (a) 6 sensors of poly(vinylstearate),poly(ethylene-co-vinyl ac-etate), and poly(ethylene vinyl alcohol), all 40 wt% carbon black (CB). (b) 4 sensors of 2-5nm Au colloids, capped with hexanethiol or 6-mercapto-1-hexanol. (c) 8 sensors of lauricacid, tetracosane, and tetracosanoic acid, with 30 wt% dioctyl phthalate, and tetracosanoicacid, all with 75 wt% CB

36

Figure 4 presents the principal component plots for the normalized data of the three sen-

sor classes used in the preliminary experiment. The principal component plots are simply

transformations of the multivariate responses of the sensor array into orthogonal directions

which statistically capture the most variance between the data, and therefore allow conve-

nient visualization of the differentiation ability of the sensor array for the analytes of interest,

through a reduction in dimensionality of the data. As shown in Fig 4, discrimination between

“cancerous” and “healthy” breath was possible with each class of sensor arrays. The best

discrimination ability was exhibited by the carbon black/monomer composite sensors, which

can be attributed either to the larger number of those sensors, or to the unique properties

of these sensors. For the other two classes, increasing the diversity of chemiresistors (in an

array of sensors) and modifying the physical properties of their building blocks is expected

to give better discrimination. Since our sensors operate solely by detecting changes, we will

be able to detect these analytes even in the presence of a constant background of other

components of an analyte mixture (as has been previously demonstrated in a variety of sit-

uations for nerve agent simulants at low concentrations in the presence of many different

background ambients).36 This, and the prior data, demonstrate the ability of our sensors to

discriminate between levels, and mixtures, of straight chain alkanes that have been reported

to be signatures of lung cancer in breath, even in the presence of saturated water vapor, and

without preconcentration.

37

Experimental

We will receive 4 L samples of lung air in inert Tedlar bags from our clinical partner, Dr.

David E. McCune, Chief, Clinical Studies Service, and Director of Clinical Trials at the

Madigan Army Medical Center, Tacoma, Washington. The Army is one of the largest health-

care providers in the USA, and this could facilitate ready expansion of the number of testing

sites if such is warranted further in the study. Vapor sampling will be done by extended

breath sampling, in which the patient will breathe into the collection apparatus for 10-30

min. The first two minutes of breath sample will be discarded, due to likely contamination

of upper respiratory air. The later deep lung air will be retained for testing purposes. These

samples will be collected with a straw, or other suitable tube in the patient’s mouth, that is

connected to the collection bag. These subjects will be diverse, and background information

will be obtained from patients regarding such factors as age, sex, race, smoking history, etc,

which will allow us both to provide for diversity in our data, and also to match each patient

with a control subject. We will take these samples back to Caltech and flow them into a

home-made chamber containing our sensors, at a relatively slow flow rate (approx. 100-300

mL/min), to obtain sensor response data, and will also use these samples in a parallel GC-MS

study.

The targeted VOCs are diverse enough that comparison studies will be needed to validate

the results we will obtain with our chemiresistor sensor arrays. We will do this via a GC

study in parallel with acquiring the sensor response data. A portion of each sample will be

retained for use with our GC-MS system, coupled with an automatic headspace sampler. As

standard headspace/gas sampling techniques have sensitivities only in the ppt-ppm range,37

38

we will use purge and trap thermal desorption or solid phase micro-extraction methods to

allow GC-MS to attain the necessary ppb detection levels to provide the information on the

VOC level and composition of our samples.

We will run several parallel sensor studies to both validate the responses and to optimize

the testing method. The initial method will involve flowing the as-received samples through

our sensor chambers and a comparison of the breath sample results with the response to

a clean background of laboratory air. We will only use sensors having minor sensitivity to

water (as those used in Fig. 4). Additionally, in parallel analyses, we will remove the water

from the samples by running the analyte flow through a desiccant chamber before presenting

the flow to the sensors. Both methods will be compared critically for their performance

with the breath samples. Preconcentration will also be used on the samples to produce

higher concentrations of the VOCs to be analyzed. Another key step will be to create

artificially the VOC sample concentrations indicated by the GC-MS, most likely through

successive dilutions of the saturated vapor phase of the pure VOCs, in order to create the

extremely low concentration levels needed. That mixture will then be exposed to the sensors

to determine whether the breath sample response pattern is indeed due to the VOCs detected

by the GC-MS method, or whether the sensor pattern arised from different volatile breath

biomarkers. The first set of samples will be taken from patients with known conditions, to

allow us to calibrate and perfect the sampling and signal processing methods. Subsequent

samples will be analyzed by personnel who are blind as to whether the patient has lung

cancer (as diagnosed by conventional methods) or not.

Ideally, we would be able to use this data to also classify patients on the basis of stage

of their cancer. This would clearly be a great aid in diagnosing lung cancer at the clinically

39

more useful earlier stages. Some information suggests this may not be possible, however.

One study suggests that the elevated VOC levels are comparable in all stages of the cancer,

because they are not associated with the size of the cancerous tumor, but instead are a

signature of oxidative stress related to a change in body chemistry as a result of development

of the cancerous condition.2 With findings consistent with this hypothesis, a study in India

was performed on 108 patients with abnormal chest radiographs who were scheduled for

bronchoscopy.38 The breath of these patients was assayed by GC-MS. Lung cancer was

confirmed histologically in 60 patients. A combination of 22 VOCs, predominantly alkanes,

alkane derivatives, and benzene derivatives, discriminated between patients with and without

lung cancer, regardless of stage (all p < 0.00003). For stage 1 (initial)39 lung cancer, a

combination of 22 VOCs had 100% sensitivity and 81.3% specificity. Cross validation of the

combination correctly predicted the diagnosis in 71.7% of the patients with lung cancer and

66.7% of those without lung disease.

Sensors with good shelf life and homogeneous response can be fabricated in batch form in

our lab. When using spray coating, batches of sensors typically have ≤ 10% difference in their

baseline resistances and ≤ 15% variation in their response during a usage period over several

months.19 Data analysis will be performed initially using standard chemometric methods

such as principal components analysis and linear discriminant analysis, which are easily

implemented using commercial software packages, such as MATLAB, with algorithms that

already exist and have been extensively previously used in our lab. Although these methods

have proven to be quite useful for analyte discrimination, more sophisticated algorithms

(e.g. supervised or unsupervised neural networks, or a variety of non-linear methods) will

be employed if necessary. However, if, as suggested, the biomarkers are entirely due to static

40

levels of production due to oxidative stress, that is also of great interest. While we won’t

be able to differentiate, we will still be able to detect at early stages, which would still be

enormously beneficial as an early detection screen.

Bibliography

[1] Cone, T. E. Pediatrics 1968, 41(5), 993.

[2] Phillips, M.; Cataneo, R. N.; Cummin, A. R. C.; Gagliardi, A. J.; Gleeson, K.; Green-berg, J.; Maxfield, R. A.; Rom, W. N. Chest 2003, 123(6), 2115–2123.

[3] Phillips, M.; Cataneo, R. N.; Ditkoff, B. A.; Fisher, P.; Greenberg, J.; Gunawardena,R.; Kwon, C. S.; Rahbari-Oskoui, F.; Wong, C. Breast J 2003, 9(3), 184–91.

[4] Kaji, H.; Hisamura, M.; Saito, N.; Murao, M. Journal of Chromatography 1978, 145(3),464–468.

[5] Olopade, C. O.; Zakkar, M.; Swedler, W. I.; Rubinstein, I. Chest 1997, 111(4), 862–865.

[6] Gordon, S. M.; Szidon, J. P.; Krotoszynski, B. K.; Gibbons, R. D.; Oneill, H. J. Clinical

Chemistry 1985, 31(8), 1278–1282.

[7] Preti, G.; Labows, J. N.; Kostelc, J. G.; Aldinger, S.; Daniele, R. Journal of

Chromatography-Biomedical Applications 1988, 432, 1–11.

[8] Phillips, M.; Gleeson, K.; Hughes, J. M. B.; Greenberg, J.; Cataneo, R. N.; Baker, L.;McVay, W. P. Lancet 1999, 353(9168), 1930–1933.

[9] Jemal, A.; Murray, T.; Ward, E.; Samuels, A.; Tiwari, R. C.; Ghafoor, A.; Feuer, E. J.;Thun, M. J. CA Cancer J Clin 2005, 55(1), 10–30.

[10] Schreiber, G.; Mccrory, D. C. Chest 2003, 123(1), 115S–128S.

[11] Banerjee, A. K.; Rabbitts, P. H.; George, J. Thorax 2003, 58(3), 266–271.

[12] Henschke, C. I.; Yankelevitz, D. F. Radiologic Clinics of North America 2000, 38(3),487.

[13] Kaneko, M.; Eguchi, K.; Ohmatsu, H.; Kakinuma, R.; Naruke, T.; Suemasu, K.;Moriyama, N. Radiology 1996, 201(3), 798–802.

[14] Boiselle, P. M.; Ernst, A.; Karp, D. D. American Journal of Roentgenology 2000, 175(5),1215–1221.

[15] Geladi, P.; Kowalski, B. R. Analytica Chimica Acta 1986, 185, 1–17.

[16] Kowalski, B. R.; Bender, C. F. Analytical Chemistry 1972, 44(8), 1405–.

41

BIBLIOGRAPHY 42

[17] Duda, R. O.; Hart, P. E. Pattern classification and scene analysis; Wiley: New York,,1973.

[18] Burns, J. A.; Whitesides, G. M. Chemical Reviews 1993, 93(8), 2583–2601.

[19] Lewis, N. S. Accounts of Chemical Research 2004, 37(9), 663–672.

[20] Ballantine, D. S.; Rose, S. L.; Grate, J. W.; Wohltjen, H. Analytical Chemistry 1986,58(14), 3058–3066.

[21] Srivastava, R.; Dwivedi, R.; Srivastava, S. K. Sensors and Actuators B-Chemical 1998,50(3), 175–180.

[22] Partridge, A. C.; Jansen, M. L.; Arnold, W. M. Materials Science & Engineering C-

Biomimetic and Supramolecular Systems 2000, 12(1-2), 37–42.

[23] Fu, Y.; Finklea, H. O. Analytical Chemistry 2003, 75(20), 5387–5393.

[24] Lang, H. P.; Baller, M. K.; Berger, R.; Gerber, C.; Gimzewski, J. K.; Battiston, F. M.;Fornaro, P.; Ramseyer, J. P.; Meyer, E.; Guntherodt, H. J. Analytica Chimica Acta

1999, 393(1-3), 59–65.

[25] Willing, B.; Kohli, M.; Muralt, P.; Oehler, O. Infrared Physics & Technology 1998,39(7), 443–449.

[26] Burl, M. C.; Sisk, B. C.; Vaid, T. P.; Lewis, N. S. Sensors and Actuators B-Chemical

2002, 87(1), 130–149.

[27] Briglin, S. M.; Lewis, N. S. Journal of Physical Chemistry B 2003, 107(40), 11031–11042.

[28] Personal communication. Wang, P. 2005.

[29] Detection volatile organic compounds in breath as markers of lung cancer using a novelelectronic nose. Yu, H.; Xu, L.; Cao, M.; Chen, X.; Wang, P.; Jiao, J.; Wang, Y. 2003.

[30] Natale, C. D.; Macagnano, A.; Martinelli, E.; Paolesse, R.; D’Arcangelo, G.; Roscioni,C.; Finazzi-Agro, A.; D’Amico, A. Biosensors and Bioelectronics 2003, 18(10), 1209–1218.

[31] James, D.; Scott, S. M.; Ali, Z.; O’Hare, W. T. Microchimica Acta 2005, 149(1-2),1–17.

[32] Severin, E. J.; Doleman, B. J.; Lewis, N. S. Analytical Chemistry 2000, 72(4), 658–668.

[33] Sisk, B. C.; Lewis, N. S. Sensors and Actuators B-Chemical 2003, 96(1-2), 268–282.

[34] Array-based vapor sensing using chemically sensitive, carbon black - monomeric organicmolecule resistors (submitted). Gao, T.; Lewis, N. S.

[35] Briglin, S. M.; Gao, T.; Lewis, N. S. Langmuir 2004, 20(2), 299–305.

BIBLIOGRAPHY 43

[36] Matzger, A. J.; Vaid, T. P.; Lewis, N. S. In Proc. SPIE Vol. 3710, p. 315-320, Detection

and Remediation Technologies for Mines and Minelike Targets IV, Abinash C. Dubey;

James F. Harvey; J. Thomas Broach; Regina E. Dugan; Eds., pages 315–320, 1999.

[37] http://sisweb.com/reference/applnote/app-39.htm.

[38] http://www.indiandoctors.com/iaso/oncoinfo.htm.

[39] http://www.cancer.org/docroot/CRI/content/CRI 2 4 3X How is lung cancer staged26.asprnav=cri.

Part III

Exploration Of A Binding Site In the

ORL1 Receptor Using Site-Directed

Mutagenesis

Out of Field Proposal

44

45

Abstract

ORL1, an “orphan” member of the opioid neuroreceptor family, and its endoge-nous peptide agonist, nociceptin, are implicated in a great number of physiological andneurological activities. Study of the ORL1-nociceptin complex revealed its behaviorto be quite different from that of the traditional opioid receptor systems. Molecularmodeling of the system has proposed multiple distinct binding sites, and experimenta-tion has supported many of these. In the current proposal, site-directed mutagenesistechniques will be employed to further characterize the specificity of the binding of thesecond extracellular loop of ORL1 with the nociceptin-[8-13] core.

Introduction

Background

In 1994 and 1995, a number of groups reported initial cloning of a receptor that was closely

homologous to traditional opioid receptors (Table 1). The clones were from a number of

species, and were typical G protein coupled receptors (GPCRs) , with seven transmembrane

domains (TMs). They showed approximately 50% homology with the traditional µ-, κ-, and

δ-opioid receptors, with the TM regions showing homology levels of up to 80%.

Species Nomenclature

Mouse KOR-31,2

MOR-C3

Rat LC1324

XOR15,6

Ratxor17

C38

ROR-C9,10

Human ORL111

Table 1: The names and species given to the initial reports of ORL1 receptor clones

This novel receptor was responsive to very few opioids, and the affinities shown were much

lower than with traditional receptors.11 However, in 1995, two labs independently reported

46

the discovery of its endogenous agonist, nociceptin.12,13 The receptor itself is alternately

referred to as ORL1 (Opioid Receptor-Like 1), or NOP (Nociceptin Opioid Receptor).

Nociceptin is a heptadecapeptide (FGGFTGARKSARKLANQ). It resembles dynorphin

A, the natural agonist of the κ-opioid receptor (YGGFLRRIRPPKLKWDNQ), both contain-

ing seventeen residues, being strongly cationic and having six common amino acid residues

(as underlined). These similarities were seen as positive evidence of correct identification.

Activation of the ORL1 receptor by nociceptin inhibits cAMP synthesis12,13 and closes

voltage-gated calcium channels,14 a trait it has in common with the normal coupling of

opioid receptors. However, the effects of nociceptin-ORL1 binding can differ, or, sometimes,

even oppose standard opioid pharmacological effects. The binding has been shown to produce

hyperalgesia in rats,12,13 and can also inhibit stress or opioid induced analgesia.15 The system

has also been implicated in tolerance pathways for other opioids. Nociceptin, applied by

itself, shows effects in many systems. It can induce a decrease in locomotor activity, impair

spatial learning, , block fear and anxiety, attenuate withdrawal symptoms, and facilitate

lordosis in rats.16 This wide variety of effects has caused great interest in this system, with

the evident possibility of serving as a pharmacological base for dealing with a broad array

of human physiological and neurological functions.

Binding

Similarities between nociceptin (noc) and dynorphin A (dyn) suggested a similar functional

architecture between the normal complexes of the two,12 despite the specificity of their

binding (nociceptin has 500-1000 greater fold affinity for the ORL1 receptor than for the

κ-opioid receptor and the reverse is true for dynorphin A.17) However, it quickly became

47

clear that the two ligands interacted with their receptors in different fashions. Tyr1 in dyn is

necessary for its activity towards opioid receptors ([Phe1]-dyn is much less active),18 whereas

[Tyr]1-noc and nociceptin are equally potent at the ORL1 receptor.19

Furthermore, several studies have shown that the noc-[1-13]-amide is the smallest no-

ciceptin fragment displaying the same potency, affinity, and selectivity as nociceptin.20,21

Noc-[1-13] is notably longer than dyn-[1-7], the smallest fully active fragment of dynorphin

A.22 Additionally, the N-terminal noc-[6-17] fragment displays significant affinity and activ-

ity with ORL1.23 Furthermore, several highly basic hexapeptides have been identified24 that

show nanomolar potency at the ORL1 receptor. All of this evidence points to the highly

basic noc-[8-13] sequence (R8KSARK13) being required for biological activity, as compared

to dynorphin A, in which the N-terminal Y1GGF4 sequence, the “opioid message” sequence,

is required to produce a biological response.22

There have been some efforts to model the noc-ORL1 complex, starting from the avail-

able crystallographic structure of bovine rhodopsin,25,26 to which the ORL1 receptor has

an identity of 18%. These models propose that the N-terminal FGGF-tetrapeptide binds

in a largely conserved TM pocket, bound by the equivalent of the opioid binding pocket

found in other opioid receptors. Noc-[5-7] is proposed to bind at the second extracellular

loop (EL2) - transmembrane helix interface. This interaction provides some conformational

constraints that would make the binding of dynorphin A more difficult with this receptor,

such that it would have unfavorable sidechain contacts. This begins to explain some of the

observed specificity of the ORL1 receptor towards nociceptin.. Finally, the positively charged

noc-[8-13] core is, in their model, shown to interact mostly with the highly acidic EL2 loop

(Figure 1).

48

Figure 1: Nociceptin-ORL1 complex.26 The image was produced using SETOR and SETOR-PLOT programs. EL2 is the loop at the top right, and nociceptin is shown in heavy atomvector form.

49

A great deal of work has been done to characterize the pocket binding site. Several

small molecule ligands have been synthesized, and they appear to bind solely within the

TM pocket.27,28 Additionally, it has been determined that a specific few point mutations in

the proposed binding pocket area are sufficient to give the ORL1 receptor much improved

opioid binding29 without removing the potency of the noc-ORL1 binding. These observations

showcase the relation of the TM binding pocket to the traditional opioid binding pocket, but

do not explain much about the nociceptin specificity.

The binding of noc-[5-7] at the EL2-TM interface is given weight by a structure activity

study using dyn/noc hybrids showing positions 5 and 6 to be major determinants of ORL1

and κ-opioid receptor specificity.30 A chimera bearing the dyn 5 and 6 positions bound

preferentially to the κ-opioid receptor, and the similar hybrid, but when the chimera had

the 5 and 6 positions from noc, it preferred the ORL1 receptor.

The final, potentially most important, binding region includes the putative interactions of

noc-[8-13] with EL2 of the ORL1 receptor. The molecular model posits that the interactions

here are an absolute requirement for activation by nociceptin. The results propose a list of

hydrogen bonding interactions between nociceptin and the EL2 region based on this model

(Table 2). The necessity of the EL2 interactions have been given a solid backing by several

studies. In one set,31,32 κ-opioid receptor/ORL1 chimeras were constructed. Replacement of

the κ-opioid EL2 loop with that from ORL1 system restored response to nociceptin, which

had been sluggish at best. Furthermore, a site directed mutagenesis study33 was performed

with the general goal of disrupting the putative electrostatic interactions in the bound com-

plex. The Glu194-Asp-Glu-Glu197 was replaced by the (nominal) structural equivalent from

the µ-opioid receptor (Arg213-Glu-Gly-Ser216). The binding affinity of this mutant receptor

50

Molecule Donor Acceptor MoleculeResidue Group Group Residue

Nociceptin Arg 8 -Nη2

H2(−)Oδ1 Asp 195 ORL1 Receptor

Nociceptin Arg 8 -NεH (−)Oδ1 Asp 195 ORL1 ReceptorNociceptin Arg 8 -NεH (−)Oε2 Glu 197 ORL1 Receptor

Nociceptin Lys 9 -NζH(+)3 O=C< Gln 208 ORL1 Receptor

ORL1 Receptor Glu 203 -NH O=C< Ser 10 Nociceptin

Nociceptin Arg 12 -Nη1

H2(−)Oε1 Glu 196 ORL1 Receptor

Nociceptin Lys 13 -NζH(+)3

(−)Oε1 Glu 194 ORL1 Receptor

Nociceptin Lys 13 -NζH(+)3

(−)Oε Glu 194 ORL1 Receptor

Table 2: Nociceptin and ORL1 EL2 intermolecular hydrogen bonds, after Topham et. al.

for nociceptin decreased 10-fold, and the biological affinity decreased 40-fold.

Alanine scans of nociceptin (in which each residue is replaced by alanine) showed that

positions 8 and 12 (two of the proposed residues for H-bonding interactions with EL2) greatly

decreased inhibition of forskolin-mediated cAMP accumulation.34

It is thus generally proposed that the interactions in the TM binding pocket are analogous

to the opioid “message”, which is necessary for receptor activation in the κ-opioid receptor.

In ORL1, however, the noc-[8-13]/EL2 interactions serve as the “address domain” function

of the binding,32 and are more necessary for activation. These interactions are thought to

bind the receptor into its conformationally active form.

Proposed Studies

Despite all of this work, an expanding amount of information about the noc-receptor complex,

and an growing library of ORL1 agonists and antagonists, little has been done to gain a

more detailed understanding of the interactions between nociceptin and the EL2 region of

the receptor. Beyond the gross scale studies above, in which large sections of the loop were

51

replaced, or the entire proposed binding section of nociceptin was swapped, there has been

only one experiment showing more precise detail.19 During determination of the shortest

viable subsection of nociceptin, the noc-[1-11] chain was shown to be entirely inactive in

binding affinity tests, showing that of the noc-[8-13] core, the noc[12] and noc[13] residues

are specifically necessary. This data aligns with the molecular model, which shows those two

residues to interact with Glu194 and Glu196 of ORL1.

We will use site-directed mutagenesis techniques to modify the EL2 section of the ORL1

receptor such that the proposed H-bonding interactions with nociceptin can be examined

more systematically. We will explore the size- and polarity-dependent effects by mutating

residues one at a time, in three different ways, as outlined below. Status of the mutants will

be examined by both specific binding assays and biological activity measurements. This will

allow us to see whether disruption of any specific proposed electrostatic interaction has an

effect, and if so, what that effect is. As comparison to these results, the implicated residues

of nociceptin will also be mutated, to determine how disrupting interactions from the other

side will affect binding and activation. Ideally, changing an H-bonding interaction in one

direction should be confirmed in the other direction.

Specifically, we will focus on the Glu194-Asp-Glu-Glu197 segment found at the N-terminal

end of EL2. Each glutamate residue will, in turn, be replaced by an aspartate, and the

reverse will be performed for the aspartate residue. These residues differ by only one carbon

atom in the side chain, and their pKa values of their side chains are very similar, such that

they are both fully ionized at physiological pH levels (glu pKa = 4.07, asp pKa = 3.90,

35 and Figure 2). The added carbon atom changes the length of the chain in its extended

conformation by ∼1.25 A. A normal hydrogen bond distance is on the order of 2.7 - 3.1 A,36

52

H2N CH C

CH2

OH

O

C

OH

O

H2N CH C

CH2

OH

O

C

NH2

O

H2N CH C

CH2

OH

O

CH2

C

OH

O

H2N CH C

CH2

OH

O

CH2

C

NH2

O

H2N CH C

CH2

OH

O

CH CH3

CH3

aspartic acid

glutamic acid

asparagine glutamineleucine

Figure 2: The structures of the 5 amino acids involved in the site-directed mutagenesis

so a change in length of over 1 A should represent a significant divergence. The loop section

and the peptide are expected to be somewhat flexible; our interests lie in determining to

what extent this is so.

Both Asp and Glu residues will be replaced by their respective amides, Asn and Gln.

The carboxylic acid side chains both have a full negative charge in vivo, which aids greatly

in their hydrogen bonding ability (Table 3). While the dipole moment of aspartate is lower

than that of asparagine, it is still calculated to have a higher hydrogen affinity. This should

allow for a somewhat subtle changing of only the electrostatic properties of the bonding, and

should permit determination of how much hydrogen acceptor ability is needed in EL2.

The third modification will be a stepwise swapping out of each of those four residues

for leucine. Leucine’s sidechain is entirely nonpolar, and is highly positively hydropathic

(hydropathy of 3.8; Glu, Gln, Asp, and Asn all have hydropathies of -3.5).38 It is most

similar in size to aspartate, but will be used to replace the glutamate residues as well,

53

despite the size disparity. Adding another carbon to the chain length in order to get a more

sterically similar acid would require a whole different set of biochemical techniques. The

information about size gained from replacing the Glu residues with Asp residues should aid

in understanding the effects of creating the Glu→Leu point mutation, in terms of which

effects come from size differences and which result from hydrophobicity.

This last step will be repeated at the implicated sites in nociceptin (residues 8,9,12 and

13, two arginines and two lysines). In both cases, this is analogous to doing a traditional

alanine-scan, in which each residue is changed to alanine, to see if any effects are disrupted.

In the case of the receptor, leucine is used instead for the reasons above, and within the

ligand, leucine will again be used for reasons of symmetry with the ORL1 mutations.

Experimental

Site-Directed Mutagenesis

All mutations will be done using a commercially available kit, such as the GeneTailor from

Invitrogen. Briefly, the procedure is as follows: the DNA to be mutated (in this case, the

gene encoding the ORL1 receptor) is obtained. Two oligonucleotide primers are designed,

overlapping at the point of interest. One of these primers contains the mutation of choice.

Amino Acid HA DMAsn 0.95 13.43Asp 1.62 8.72Gln 1.02 10.32Glu 20.09 364.92

Table 3: HA = hydrogen bond acceptor score, DM = dipole moment, in debye. All num-bers are for ionized amino acids, and all data were obtained in the HyperChem modelingenvironment.37

54

These are placed into solution, along with polymerase enzyme and the necessary dideoxynu-

cleotides. The DNA is then denatured, and multiple rounds of polymerase chain reaction

(PCR) are undertaken. Each cycle produces only one unmutated strand of DNA, so there are

very quickly many more mutated strands present than unmutated. After the desired number

of cycles (40 is often considered adequate), PCR is stopped, the DNA is annealed, and the

mutated genes are transfected into cells. In prior experiments, a common choice was to use

the pcDNA3 plasmid vector to transfect human embryonic kidney 293 cells (HEK293),27 or

Chinese Hamster Ovary cells (CHO).30,32

Characterization

Radioligand Binding Assays

Membrane fractions will be prepared from transfected cells, homogenized, and centrifuged.

Competitive binding analysis will be performed with [3H]-nociceptin. The manner of mea-

suring binding is as follows: half of the receptor samples are incubated with varying concen-

trations of the radio-labeled peptide. The other half of the samples are incubated with the

labeled peptides and a large excess of unlabeled. Since the amount of unlabeled peptide is so

much greater, it should occupy virtually all of the receptor binding sites. Any noted binding

from the radio-labeled agonist is thus due to non-specific binding, which we do not want to

count. The samples are filtered after incubation, scintillation fluid is added, and radiation is

counted using a scintillation counter. Non-specific binding is subtracted from total binding,

and the remaining counts can be attributed to specific binding. This is then plotted against

concentration of the labeled agonist. This specific binding is then compared to that of the

55

wild-type ORL1 receptor.

cAMP Assay

While it is posited that that the TM binding site is more necessary for activation of the

receptor, characterization of how well the mutant can work is also of some interest. To this

end, the inhibition of forskolin-mediated cyclic adenosine monophosphate (cAMP) accumu-

lation will be measured and compared to the normal receptor. Transfected cells will be

incubated in a buffered solution with forskolin, and with varied concentrations of agonist.

The reactions are then stopped, and the cells are stored at freezing temperatures. cAMP

content if then determined using one of a variety of methods, all available in commercially

available kits.

Bibliography

[1] Pan, Y. X.; Cheng, J.; Xu, J.; Pasternak, G. W. Regulatory Peptides 1994, 54(1),217–218.

[2] Pan, Y. X.; Cheng, J.; Xu, J.; Rossi, G.; Jacobson, E.; Ryanmoro, J.; Brooks, A. I.;Dean, G. E.; Standifer, K. M.; Pasternak, G. W. Molecular Pharmacology 1995, 47(6),1180–1188.

[3] Nishi, M.; Takeshima, H.; Mori, M.; Nakagawara, K.; Takeuchi, T. Biochemical and

Biophysical Research Communications 1994, 205(2), 1353–1357.

[4] Bunzow, J. R.; Saez, C.; Mortrud, M.; Bouvier, C.; Williams, J. T.; Low, M.; Grandy,D. K. Febs Letters 1994, 347(2-3), 284–288.

[5] Wang, J. B.; Johnson, P. S.; Imai, Y.; a. M. Persico.; a. Ozenberger, B.; Eppler, C. M.;Uhl, G. R. Febs Letters 1994, 348(1), 75–79.

[6] Wick, M. J.; Minnerath, S. R.; Lin, X. Q.; Elde, R.; Law, P. Y.; Loh, H. H. Molecular

Brain Research 1994, 27(1), 37–44.

[7] Chen, Y.; Fan, Y.; Liu, J.; Mestek, A.; Tian, M. T.; Kozak, C. A.; Yu, L. Febs Letters

1994, 347(2-3), 279–283.

[8] Lachowicz, J. E.; Shen, Y.; Monsma, F. J.; Sibley, D. R. Journal of Neurochemistry

1995, 64(1), 34–40.

[9] Fukuda, K.; Kato, S.; Mori, K.; Nishi, M.; Takeshima, H.; Iwabe, N.; Miyata, T.;Houtani, T.; Sugimoto, T. Febs Letters 1994, 343(1), 42–46.

[10] Keith, D.; Maung, T.; Anton, B.; Evans, C. Regulatory Peptides 1994, 54(1), 143–144.

[11] Mollereau, C.; Parmentier, M.; Mailleux, P.; Butour, J. L.; Moisand, C.; Chalon, P.;Caput, D.; Vassart, G.; Meunier, J. C. Febs Letters 1994, 341(1), 33–38.

[12] Meunier, J. C.; Mollereau, C.; Toll, L.; Suaudeau, C.; Moisand, C.; Alvinerie, P.;Butour, J. L.; Guillemot, J. C.; Ferrara, P.; Monsarrat, B.; Mazarguil, H.; Vassart, G.;Parmentier, M.; Costentin, J. Nature 1995, 377(6549), 532–535.

[13] Reinscheid, R. K.; Nothacker, H. P.; Bourson, A.; Ardati, A.; Henningsen, R. A.;Bunzow, J. R.; Grandy, D. K.; Langen, H.; Monsma, F. J.; Civelli, O. Science 1995,270(5237), 792–794.

56

BIBLIOGRAPHY 57

[14] Connor, M.; Yeo, A.; Henderson, G. British Journal of Pharmacology 1996, 118(2),205–207.

[15] Mogil, J. S.; Grisel, J. E.; Reinscheid, K. K.; Civelli, O.; Belknap, J. K.; Grandy, D. K.Neuroscience 1996, 75(2), 333–337.

[16] Mogil, J. S.; Pasternak, G. W. Pharmacological Reviews 2001, 53(3), 381–415.

[17] Meunier, J. C.; Mouledous, L.; Topham, C. M. Peptides 2000, 21(7), 893–900.

[18] Walker, J. M.; Moises, H. C.; Coy, D. H.; Baldrighi, G.; Akil, H. Science 1982,218(4577), 1136–1138.

[19] Shimohigashi, Y.; Hatano, R.; Fujita, T.; Nakashima, R.; Nose, T.; Sujaku, T.; Saigo,A.; Shinjo, K.; Nagahisa, A. Journal of Biological Chemistry 1996, 271(39), 23642–23645.

[20] Guerrini, R.; Calo, G.; Rizzi, A.; Bianchi, C.; Lazarus, L. H.; Salvadori, S.; Temussi,P. A.; Regoli, D. Journal of Medicinal Chemistry 1997, 40(12), 1789–1793.

[21] Dooley, C. T.; Houghten, R. A. Life Sciences 1996, 59(1), PL23–PL29.

[22] Chavkin, C.; Goldstein, A. Proceedings of the National Academy of Sciences of the

United States of America-Biological Sciences 1981, 78(10), 6543–6547.

[23] Butour, J. L.; Moisand, C.; Mazarguil, H.; Mollereau, C.; Meunier, J. C. European

Journal of Pharmacology 1997, 321(1), 97–103.

[24] Dooley, C. T.; Spaeth, C. G.; BerzeteiGurske, I. P.; Craymer, K.; Adapa, I. D.; Brandt,S. R.; Houghten, R. A.; Toll, L. Journal of Pharmacology and Experimental Therapeutics

1997, 283(2), 735–741.

[25] Broer, B. M.; Gurrath, M.; Holtje, H. D. Journal of Computer-Aided Molecular Design

2003, 17(11), 739–754.

[26] Topham, C. M.; Mouledous, L.; Poda, G.; Maigret, B.; Meunier, J. C. Protein Engi-

neering 1998, 11(12), 1163–1179.

[27] Rover, S.; Adam, G.; Cesura, A. M.; Galley, G.; Jenck, F.; Monsma, F. J.; Wichmann,J.; Dautzenberg, F. M. Journal of Medicinal Chemistry 2000, 43(7), 1329–1338.

[28] Rover, S.; Wichmann, J.; Jenck, F.; Adam, G.; Cesura, A. M. Bioorganic & Medicinal

Chemistry Letters 2000, 10(8), 831–834.

[29] Meng, F.; Ueda, Y.; Hoversten, M. T.; Taylor, L. P.; Reinscheid, R. K.; Monsma, F. J.;Watson, S. J.; Civelli, O.; Akil, H. Molecular Pharmacology 1998, 53(4), 772–777.

[30] Lapalu, S.; Moisand, C.; Mazarguil, H.; Cambois, G.; Mollereau, C.; Meunier, J. C.Febs Letters 1997, 417(3), 333–336.

BIBLIOGRAPHY 58

[31] Lapalu, S.; Moisand, C.; Butour, J. L.; Mollereau, C.; Meunier, J. C. Febs Letters 1998,427(2), 296–300.

[32] Mollereau, C.; Mouledous, L.; Lapalu, S.; Cambois, G.; Moisand, C.; Butour, J. L.;Meunier, J. C. Molecular Pharmacology 1999, 55(2), 324–331.

[33] Mouledous, L.; Topham, C. M.; Mazarguil, H.; Meunier, J. C. Journal of Biological

Chemistry 2000, 275(38), 29268–29274.

[34] Reinscheid, R. K.; Ardati, A.; Monsma, F. J.; Civelli, O. Journal of Biological Chemistry

1996, 271(24), 14163–14168.

[35] Dawson, R. M. C. Data for biochemical research; Clarendon Press: Oxford, 3rd ed.,1986.

[36] Voet, D.; Voet, J. G. Biochemistry; Wiley: New York, 2nd ed., 1995.

[37] Young, V. R.; Ajami, A. M. Journal of Nutrition 2001, 131(9), 2449S–2459S.

[38] Kyte, J.; Doolittle, R. F. Journal of Molecular Biology 1982, 157(1), 105–132.