Detection and differentiation of normal, cancerous,and metastatic cells using nanoparticle-polymersensor arraysAvinash Bajaja, Oscar R. Mirandaa, Ik-Bum Kimb, Ronnie L. Phillipsb, D. Joseph Jerryc, Uwe H. F. Bunzb,and Vincent M. Rotelloa,1

Departments of aChemistry, and cVeterinary and Animal Science, University of Massachusetts, Amherst, MA 01003; and bSchool of Chemistry andBiochemistry, Georgia Institute of Technology, 901 Atlantic Drive, Atlanta, GA 30332

Edited by Laura L. Kiessling, University of Wisconsin, Madison, WI, and approved May 20, 2009 (received for review January 28, 2009)

Rapid and effective differentiation between normal and cancercells is an important challenge for the diagnosis and treatment oftumors. Here, we describe an array-based system for identificationof normal and cancer cells based on a ‘‘chemical nose/tongue’’approach that exploits subtle changes in the physicochemicalnature of different cell surfaces. Their differential interactions withfunctionalized nanoparticles are transduced through displacementof a multivalent polymer fluorophore that is quenched whenbound to the particle and fluorescent after release. Using thissensing strategy we can rapidly (minutes/seconds) and effectivelydistinguish (i) different cell types; (ii) normal, cancerous and met-astatic human breast cells; and (iii) isogenic normal, cancerous andmetastatic murine epithelial cell lines.

fluorescence � gold nanoparticle � sensor � conjugated polymer

Each cell type has unique molecular signatures that distinguishbetween healthy and diseased tissues (1). In the case of

cancers, the distinctions between normal vs. tumor and benignvs. metastatic cells are often subtle. The identification of cellularsignatures for early cancer cell detection is a major hurdle forcancer therapy; the earlier these signatures can be established,the more effectively they can be treated (2). Cancerous cells aredifferentiated from noncancerous ones on the basis of intracel-lular or extracellular (cell surface) biomarkers. Detection meth-ods based on specific recognition of intracellular biomarkers(e.g., DNA/RNA/Proteins) require previous knowledge of spe-cific mutations in DNA/RNA (3) or changes in the regulation ofprotein expression inside the cells. Similarly, detection methodsbased on specific recognition of extracellular (cell surface)biomarkers such as histopathology (4), bioimaging (5), antibodyarrays require prior knowledge of biomarkers on cell surfaces.Observation of overexpressed antigens (6) on tumor cells usingantibody-based platforms have been explored using ELISA (7),surface plasmon resonance (8, 9), nanoparticles (10–13), micro-cantilevers (14), carbon nanotubes (15, 16), and expressionmicroarrays (17). Antibody arrays provide an effective butcomplex approach for cancer detection, diagnosis and prognosis(18), however, there is no single marker or a combination ofbiomarkers that has sufficient sensitivity and specificity todifferentiate between normal, cancerous, and metastatic celltypes (19). Here, we describe a detection system that is based onselective noncovalent interactions between cell surface compo-nents and nanoparticle-based sensor elements that does notrequire any previous knowledge of intracellular or extracellularbiomarkers.

The cell membrane surface consists primarily of a thin layerof amphipathic phospholipids, carbohydrates and many integralmembrane proteins. The amount and types of which differbetween species and according to function of cells (20, 21). Thisresults into distinct cell membrane composition in different celltypes. Therefore, one can predict, however, that there will bephysicochemical (i.e., charge, hydrophobicity etc.) differences

between cell types and between healthy and cancerous cells.Such physicochemical differences could potentially be detectedby an array-based ‘‘chemical nose’’ approach that relies onselective interactions between multiple reporter elements andthe target cell.

In the chemical nose approach, an array of different sensorsis used where every element in the sensor array responds to anumber of different chemicals or analytes (22). A distinct patternof responses produced from a set of sensors in the array providea fingerprint that allows classification and identification of theanalyte (23). The collection of sensors should contain chemicaldiversity to respond to largest possible cross-section of analytes.The specific interactions involved between the reporter elementsand the analyte are noncovalent and reversible. This approachprovides an alternative to ‘‘lock–key’’ specific recognition (24)and has been used to detect metal ions (25), volatile agents (26),aromatic amines (27), amino acids (28, 29), and carbohydrates(25). In recent research we have demonstrated that the displace-ment of fluorescent polymers from differentially functionalizedgold nanoparticles with concomitant restoration of fluorescenceprovides an effective array-based method for the identificationof proteins (30). More recently, we have shown that this meth-odology can be used to differentiate between bacterial speciesand even between different strains of the same species (31). Wereport here a particle-polymer array that distinguishes betweenhealthy, cancerous and metastatic human breast cells, anddifferentiates isogenic healthy and transformed cells.

Results and DiscussionOur detection system is based on conjugates between 3 struc-turally related cationic gold nanoparticles (NP1–NP3, Fig. 1Aand Fig. S1) and the poly(para-phenyleneethynylene) (PPE)polymer PPE-CO2 featuring charge multivalency (32) and mo-lecular wire properties (33) (Fig. 1 A). In these noncovalentconjugates, the nanoparticle quenches the fluorescence of thepolymer. The interactions between nanoparticles and anionicpolymers are noncovalent, and predominantly electrostatic.When mammalian cells were incubated with these nanoparticle-polymer complexes, there is competitive binding between nano-particle-polymer complexes and cell types (Fig. 1B). Because oftheir cationic surface, nanoparticles are expected to interact withphospholipids, membrane proteins and carbohydrates of the cellsurface through both electrostatic and hydrophobic interactions.

Author contributions: A.B., D.J.J., U.H.F.B., and V.M.R. designed research; A.B. and O.R.M.performed research; I.-B.K. and R.L.P. contributed new reagents/analytic tools; A.B.,O.R.M., and V.R. analyzed data; and A.B., O.R.M., D.J.J., U.H.F.B., and V.M.R. wrote thepaper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

1To whom correspondence should be addressed. E-mail: rotello@chem.umass.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0900975106/DCSupplemental.

10912–10916 � PNAS � July 7, 2009 � vol. 106 � no. 27 www.pnas.org�cgi�doi�10.1073�pnas.0900975106

These interactions are responsible for displacement of thefluorophore polymer from the nanoparticle-polymer complexesgenerating a fluorescence response. The nanoparticles are ex-pected to possess different affinities for dissimilar cell surfacesdepending on cell membrane composition and surface of nano-particles. Selective displacement of the polymer from the particleby the cell surface regenerates fluorescence, transducing thebinding event in a ‘‘turn on’’ fashion.

The complex stability constants (KS) and association stoichio-metries (n) for the nparticle-polymer dyads were obtainedthrough nonlinear least-squares curve-fitting analysis (34). Com-plex stabilities vary within 1 order of magnitude (��G � 4.5kJ�mol�1), and the binding stoichiometry ranges from 2.5 forNP2 to 0.9 for NP3 (Fig. S2). After determining the saturationpoint for fluorescence quenching (Fig. S3), the appropriatestoichiometries of particle and polymer were mixed in 5 mMphosphate buffer (pH � 7.4) to yield nanoparticle-PPECO2complexes with a final concentration of polymer of 100 nM andof nanoparticles 10–40 nM. The complexes of PPECO2 andNP1-3 were then incubated with different cell types to determinechanges in fluorescence intensities. We observed increases anddecreases in fluorescence intensities depending on the cell typeand the nature of nanoparticle-polymer complexes. Increasedfluorescence intensities are due to the displacement of thePPECO2 polymer from the NP-PPECO2 complexes by cellsurfaces (Fig. 1B), whereas decreases in the fluorescence inten-sities are due to the quenching of the residual PPECO2 f luores-cence by the cell surfaces. These differences in the fluorescencepatterns depend on the cell type and are reproducible. We haveperformed array-based s sensing using 9 gold nanoparticles thatpossess different head groups and interact differently withpolymers (Fig. S3a). We studied their interactions with thedifferent cell types listed in Table 1, focusing on which particleset can best differentiate between different particles. (see be-low). From studies, we have observed the maximum differenti-ation grouping using 3 nanoparticles NP1-NP3, as establishedthrough jackknifed analysis (Fig. S3b).

Detection of Differences in Cell Types. As an initial test of ourmethod we used 4 different types of human cancer cells: HeLa

(Cervical), HepG2 (Liver), NT2 (Testis) and MCF-7 (Breast).Fig. 2A presents the change in the fluorescence response for thenanoparticle-polymer supramolecular complexes upon additionof the different cancer cell types. Linear Discriminant Analysis(LDA) was used to statistically characterize the fluorescencechanges. This analysis reduced the size of the training matrix (3nanoparticles � 4 cell types � 6 replicates) and transformedthem into canonical factors that are linear combinations of theresponse patterns (3 factors � 4 cell types � 6 replicates). The2 canonical factors contain 96.6% and 3.3% of the variation,

Fig. 1. Molecular structures of nanoparticles and polymers, and schematic offluorophore displacement cell detection array. (A) Molecular structures of thecationic gold nanoparticles (NP1-NP3) and the fluorescent polymer (PPECO2).(B) Displacement of quenched fluorescent polymer (dark green strips, fluo-rescence off; light green strips, fluorescence on) by cell (in blue) with concom-itant restoration of fluorescence.

Table 1. Origin and nature of the normal, cancerous andmetastatic cell lines used in this study.

Cell line Liver HepG2 Cancerous

Human Cervix HeLa CancerousTestis NT2 CancerousBreast MCF10A Normal immortalized

MCF-7 CancerousMDA-MB-231 Metastatic

Mouse BALB/c mice (breast) CDBgeo Normal immortalizedTD CancerousV14 Metastatic

Fig. 2. Detection of human cancerous cell lines. (A) Change in fluorescenceintensities (F � F0) for 4 different cancer cell lines HeLa (Cervical), MCF7(Breast), HepG2 (Liver) and NT2 (Testes) using nanoparticle-polymer supramo-lecular complexes. Each value is average of 6 parallel measurements. (B)Canonical score plot for the two factors of simplified fluorescence responsepatterns obtained with NP–PPECO2 assembly arrays against different mam-malian cell types. The canonical scores were calculated by LDA for the iden-tification of 4 cell lines.

Bajaj et al. PNAS � July 7, 2009 � vol. 106 � no. 27 � 10913

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

CHEM

ISTR

Y

respectively as shown in Fig. 2B. In this plot, each pointrepresents the response pattern for a single cell type to theNP-PPECO2 sensor array. In the canonical f luorescence re-sponse patterns, the different cell types are clustered into 4nonoverlapping groups (95% level confidence ellipses) (Fig. 2B)with standard deviation of �5%. These initial results validateour ability to differentiate cancer cell types phenotypically basedon their surface properties.

Detection of Normal/Cancerous and Metastatic Cells. An importantissue in cancer therapy is assessing whether tissue/cells arehealthy, or either benign or metastatic tumors. We chose 3different human breast cell lines to test our sensor array in thisapplication: MCF10A a normal breast cell line, MCF7 a can-cerous but nonmetastatic cell line, whereas MDA-MB-231 is ametastatic cancer cell line. The 3 cell lines show differentialf luorescence patterns (Fig. 3A); LDA of their response indicatesa 100% accuracy of detection (Fig. 3B).

Detection of Isogenic Cell Types. The above studies suggest that wecan differentiate normal, cancerous and metastatic cell typeswith our sensor array. Each of the 3 cell lines, however, camefrom different individuals. To provide a test bed where individ-ual-to-individual variation is not present, we used 3 isogenic celllines, CDBgeo, TD, and V14 cells. Due to their high geneticsimilarity, isogenic cells are expected to present a particularly

stringent test for detection assays. Each of these isogenic cellswas developed from BALB/c mice, and therefore possesses thesame genotypic background. CDBgeo cells were prepared byretroviral infection with a marker gene encoding the fusion of�-galactosidase and neomycin resistance. These cells exhibitnormal outgrowths when transplanted into mammary fat pads(35). The TD cells were prepared by treating CDBgeo-cellswith 10 ng/mL TGF-� for 14 days. Withdrawal for 5 passagesresulted in a persistent epithelial to mesenchymal transforma-tion: Tumorogenic growth resulted when transplanted. TheV14 cell line was established from a primary mammary tumorarising in BALB/c-Trp-53�/� mice. The cells lack p53 proteinand form aggressive tumors that are locally invasive in mice(36). Fig. 4A presents the change in f luorescence intensities of3 isogenic cell types toward nanoparticle-polymer complexes.The differential response indicates that these supramolecularcomplexes can effectively differentiate isogenic cell types.LDA classifies the cell types into 3 distinct clusters with 2canonical factors containing 83.0% and 17.0% of the variation, with100% identification accuracy among these isogenic cell types (Fig.4B). Taken together, these studies indicate that our method rapidlyand effectively differentiates cell lines based on cell type and diseasestate.

The efficacy of our approach indicates that there are distinctphenotypic differences in the physicochemical properties ofcells. One question that arises is whether there is a response thatis generally indicative of whether a cell is normal or cancerous.

Fig. 3. Detection of normal, cancerous and metastatic human breast cells.(A) Change in fluorescence intensities (F � F0) for 3 breast cell lines of differentnature MCF10A (normal), MCF-7 (cancer) and MDA-MB231 (metastatic) usingnanoparticle-polymer supramolecular complexes. Each value is average of 6parallel measurements. (B) Canonical score plot for the first two factors ofsimplified fluorescence response patterns obtained with NP–PPECO2 assemblyarrays against different mammalian cell types.

Fig. 4. Detection of Isogenic cell types. (A) Change in fluorescence intensities(F � F0) for 3 cell lines of same genotype CDBgeo, TD cell and V14 usingnanoparticle-polymer supramolecular complexes. Each value is average of 6parallel measurements. (B) Canonical score plot for the first two factors ofsimplified fluorescence response patterns obtained with NP–PPECO2 assemblyarrays against different mammalian cell types.

10914 � www.pnas.org�cgi�doi�10.1073�pnas.0900975106 Bajaj et al.

Metaanalysis of our studies using LDA indicates that normalepithelial cell lines CDBgeo and MCF10A were overlapping(Fig. 5a) even though both of these cell lines were isolated frommouse and human respectively. Likewise the metastatic murine(V14) and human MDMBA-231 metastatic cell lines wereclustered, indicating a potential correlation between cell surfaceproperties and disease states of cells.

In summary, we have developed a rapid and effective array-based approach to differentiate between normal and cancerouscell lines. Significantly, full differentiation was achieved usingonly 3 nanoparticle-polymer dyads, indicating that a simplesensor array has ample diagnostic capacity when exposed tomammalian cells. These systems have the potential to help usunderstand the physical changes that occur on the surfaces ofcells in various disease states. Taken together, ‘‘nose’’ based

sensor systems are a fundamentally new way of looking intobiodiagnostic, biophysical and surface science processes involv-ing cell surfaces.

Materials and MethodsNanoparticles (25, 37, 38) (SI Text, Scheme S1, and Fig. S1) and polymers (39)were synthesized as reported previously. All of the cells except MCF10A,CDBgeo, TD and V14 were grown in DMEM media supplemented with 10%FBS and 1% antibiotics in T75 flasks. NT2 cell line was obtained from R. T.Zoeller (Department or Biology, University of Massachusetts, Amherst, MA).CDBgeo, TD and V14 cells were grown in DMEM-F12 media supplementedwith 2% ABS, 25 mM Hepes, 10 �g/mL insulin, 5 ng/mL EGF, 15 �g/mLgentamycin. Cells were washed with DPBS buffer, trypsinized with 1� trypsinand collected in the DMEM media. Fluorescence titration experiments deter-mined the complexation between nanoparticles and PPECO2. Fluorescenceintensity changes at 465 nm were recorded with an excitation wavelength of430 nm. Polymer and stoichiometric amounts of NP1–NP3, as determined bythe fluorescence titration study were diluted with phosphate buffer (5 mm,pH 7.4) to solutions with a final polymer concentration of 100 nM. Eachsolution (200 �L) was placed into a well on the micro plate. After incubationfor 30 min, the fluorescence intensity at 465 nm was recorded with anexcitation wavelength of 430 nm. Next, 100 �L of cell suspension (20,000 cells)was added to each well. After incubation for another 30 min, the fluorescenceintensity at 465 nm was measured again. The fluorescence intensity beforeaddition of the cells was subtracted from that obtained after addition of thecells to record the overall fluorescence response (DI) (Tables S1–S3). Thisprocess was completed for all cell lines to generate 6 replicates of each thatwas subjected to a classical linear discriminant analysis (LDA) using SYSTAT(version 11.0). Each cell line possesses a unique fluorescence response datawith the NP-PPECO2 complex array, because cell interaction with the NP-PPECO2 complex array depends on the cell surface characteristic. Therefore,for each cell, we tested its fluorescence responses against 3 NP-PPECO2 adduct6 times, generating (i) 3 � 6 � 4 matrix for 4 different cancer cell lines HeLa(Cervical), MCF7 (Breast), HepG2 (Liver) and NT2 (Testes) for Fig. 2, (ii) 3 � 6 �3 matrix for 3 breast cell lines of different nature MCF10A (normal), MCF-7(cancer) and MDA-MB231 (metastatic) for Fig. 3, and (iii) 3 � 6 � 3 matrix for3 cell lines of same genotype CDBgeo, TD cell and V14 for Fig. 4. The raw dataobtained were subjected to Linear Discriminant Analysis (LDA) (40, 41) tomaximize the ratio between-class variance to the within-class variance, thusdifferentiate the fluorescence response patterns of the NP-PPECO2 system thecell targets. This analysis reduced the size of the training matrix and trans-formed them into canonical factors that are linear combination of the fluo-rescence response patterns (i) 2 factors � 6 replicates � 4 cell, (ii) 2 factors �6 replicates � 3 cells, and (iii) 2 factors � 6 replicates � 3 cells, respectively. Thecanonical factors contain different percentage of variation and two of themwere plotted in 2D as shown in Figs. 1B and 4B. In a blind experiment, the ratesof fluorescence patterns of new case were first converted to canonical scoresusing discriminate functions established on training samples. Then, Mahal-anobis distances [the distance of a case to the centroid of a group in amultidimensional space, in the current case it is 2-dimensional (42, 43)] of thenew case to the centroid of respective groups (normal or cancerous or meta-static cells) of training samples were calculated. The new case was assigned tothe group with shortest Mahalanobis distance. This processing protocol wasperformed on the SYSTAT 11 program, allowing the assignment of cells tospecific groups.

ACKNOWLEDGMENTS. We thank Professor R. Thomas Zoeller for providingthe NT2 cell lines. This work was supported by the National Science FoundationCenter for Hierarchical Manufacturing at the University of MassachusettsNanoscale Science and Engineering Center and National Science FoundationGrant DMI-0531171; and National Institutes of Health Grant GM077173; Al-lergy and Infectious Grant AI073425; Grants R01-CA095164 and R01-CA105452 (to D.J.J.); and Department of Energy Grant DE-FG02-04ER46141 (toU.H.F.B., R.L.P., I.-B.K. and V.M.R.).

1. Srinivas PR, Kramer BS, Srivastava S (2001) Trends in biomarker research for cancerdetection. Lancet Oncol 2:698–704.

2. Pantel K, Brakenhoff RH, Brandt B (2008) Detection, clinical relevance andspecific biological properties of disseminating tumour cells. Nat Rev Cancer 8:329–340.

3. Jen J, Wu L, Sidransky D (2000) An overview on the isolation and analysis of circulatingtumor DNA in plasma and serum. Ann N Y Acad Sci 906:8–12.

4. Skvara H, Teban L, Fiebiger M, Binder M, Kittler H (2005) Limitations of dermoscopy inthe recognition of melanoma. Arch Dermatol 141:155–160.

5. Gao X, Cui Y, Levenson RM, Chung LWK, Nie S (2004) In vivo cancer targeting andimaging with semiconductor quantum dots. Nat Biotechnol 22:969–976.

6. Borrebaeck C (2006) Antibody microarray-based oncoproteomics. Expert Opin BiolTher 6:838.

7. Ward AM, Catto JWF, Hamdy FC (2001) Prostate specific antigen: Biology, biochemistryand available commercial assays. Ann Clin Biochem 38:633–651.

8. Chou SF, Hsu WL, Hwang JM, Chen CY (2004) Development of an immunosensor forhuman ferritin, a nonspecific tumor marker, based on surface plasmon resonance.Biosens Bioelectron 19:999–1005.

9. Alivisatos P (2004) The use of nanocrystals in biological detection. Nat Biotechnol 22:47–52.10. Soukka T, et al. (2001) Supersensitive time-resolved immunofluorometric assay of

free prostate-specific antigen with nanoparticle label technology. Clin Chem47:1269 –1278.

Fig. 5. Detectionofnormalandcancerouscell lines. (A)Changes influorescenceintensities (F � F0) of noncancerous and cancerous cell types using nanoparticle-polymer supramolecular complexes. Each value is average of 6 parallel measure-ments. (B) Canonical score plot for the first two factors of simplified fluorescenceresponse patterns obtained with NP–PPECO2 assembly arrays against differentnormalandcancerouscell types. Furtherdifferentiationwasobserved inthethirddimension, allowing discrimination each of the species.

Bajaj et al. PNAS � July 7, 2009 � vol. 106 � no. 27 � 10915

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

CHEM

ISTR

Y

11. Nam JM, Thaxton CS, Mirkin CA (2003) Nanoparticle-based bio-bar codes for theultrasensitive detection of proteins. Science 301:1884–1886.

12. Wu G, et al. (2001) Bioassay of prostate-specific antigen (PSA) using microcantilevers.Nat Biotechnol 19:856–860.

13. Chen RJ, et al. (2003) Noncovalent functionalization of carbon nanotubes for highlyspecific electronic biosensors. Proc Natl Acad Sci USA 100:4984–4989.

14. Chen RJ, et al. (2004) An investigation of the mechanisms of electronic sensing ofprotein adsorption on carbon nanotube devices. J Am Chem Soc 126:1563–1568.

15. Wang WU, Chen C, Lin KH, Fang Y, Lieber CM (2005) Label-free detection of small-molecule-protein interactions by using nanowire nanosensors. Proc Natl Acad Sci USA102:3208–3212.

16. Cui Y, Wei Q, Park H, Lieber CM (2001) Nanowire nanosensors for highly sensitive andselective detection of biological and chemical species. Science 293:1289–1292.

17. Campagnolo C, et al. (2004) Real-Time, label-free monitoring of tumor antigen andserum antibody interactions. J Biochem Biophys Methods 61:283–298.

18. Wingren C, Borrebaeck CA (2008) Antibody microarray analysis of directly labelledcomplex proteomes. Curr Opin Biotech 18:55–61.

19. Sanchez-Carbayo M (2006) Antibody arrays: Technical considerations and clinicalapplications in cancer. Clin Chem 52:1651–1659.

20. Singer SJ, Nicolson GL (1972) The fluid mosaic model of the structure of cell mem-branes. Science 175:720–731.

21. Alberts B, Johnson A, Lewis J (2002) in Molecular Biology of the Cell (GarlandPublishing, New York) 4th Ed.

22. Lavigne JL, Anslyn EV (2001) Sensing a paradigm shift in the field of molecularrecognition: From selective to differential receptors. Angew Chem Int Ed 40:3118–3130.

23. Albert KJ, et al. (2000) Cross-reactive chemical sensor arrays. Chem Rev 100:2595–2626.24. Wright AT, Anslyn EV (2006) Differential receptor arrays and assays for solution-based

molecular recognition. Chem Soc Rev 35:14–28.25. Lee JW, Lee JS, Chang YT (2006) Colorimetric identification of carbohydrates by a pH

indicator/pH change inducer ensemble. Angew Chem Int Ed 45:6485–6487.26. Rakow NA, Suslick KS (200) A colorimetric sensor array for odour visualization. Nature

406:710–713.27. Greene NT, Shimizu KD (2005) Colorimetric molecularly imprinted polymer sensor

array using dye displacement. J Am Chem Soc 127:5695–5700.28. Folmer-Andersen JF, Kitamura M, Anslyn EV (2006) Pattern-based discrimination of

enantiomeric and structurally similar amino acids: An optical mimic of the mammaliantaste response. J Am Chem Soc 128:5652–5653.

29. Buryak A, Severin K (2005) A chemosensor array for the colorimetric identification of20 natural amino acids. J Am Chem Soc 127:3700–3701.

30. You CC, et al. (2007) Detection and identification of proteins using nanoparticle-fluorescent polymer ‘‘chemical nose’’ sensors. Nat Nanotechnol 2:318–323.

31. Phillips RL, Miranda OR, You CC, Rotello VM, Bunz UH (2008) Rapid and efficientidentification of bacteria using gold-nanoparticle-poly(para-phenyleneethynylene)constructs. Angew Chem Int Ed 47:2590–2594.

32. Kim IB, Erdogan B, Wilson JN, Bunz UHF (2004) Sugar-poly(para-phenylene ethynylene)conjugates as sensory materials: Efficient quenching by Hg2� and Pb2�. Chem Eur J10:6247–6254.

33. Zhou Q, Swager TM (1995) Fluorescent chemosensors based on energy migration inconjugated polymers: The molecular wire approach to increased sensitivity. J Am ChemSoc 117:12593–12602.

34. Phillips RL, et al. (2009) Gold nanoparticle-PPE constructs as biomolecular materialmimics: Understanding the electrostatic and hydrophobic interactions. Soft Matter5:607–612.

35. Deugnier MA, et al. (2006) Isolation of mouse mammary epithelial progenitor cells withbasal characteristics from the Comma-Dbeta cell line. Dev Biol 293:414–425.

36. Blackburn AC, et al. (2004) Loss of heterozygosity occurs via mitotic recombination inTrp53�/� mice and associates with mammary tumor susceptibility of the BALB/c strain.Cancer Res 64:5140–5147.

37. Brust M, Walker M, Bethell D, Schiffrin DJ, Whyman R (1994) Synthesis of thiol-derivatised gold nanoparticles in a two-phase liquid–liquid system. J Chem Soc ChemCommun 801–802.

38. Hostetler MJ, Templeton AC, Murray RW (1999) Dynamics of place-exchange reactionson monolayer-protected gold cluster molecules. Langmuir 15:3782–3789.

39. Kim IB, Dunkhorst A, Gilbert J, Bunz (2005) Sensing of lead ions by a carboxylate-substituted PPE: Multivalency effects. Macromolecules 38:4560–4562.

40. Engelman L (2004) In Systat 11.0, Statistics (SYSTAT Software, Chicago), pp i301–i358.

41. Jurs PC, Bakken GA, McClelland HE (2000) Computational methods for the analysis ofchemical sensor array data from volatile analytes. Chem Rev 100:2649–2678.

42. Mahalanobis PC (1936) On the generalised distance in statistics. Proc Natl Inst Sci India2:49–55.

43. Gnanadesikan R, Kettenring JR (1972) Robust estimates, residuals, and outlier detec-tion with multiresponse data. Biometrics 28:81–124.

10916 � www.pnas.org�cgi�doi�10.1073�pnas.0900975106 Bajaj et al.

Black plate (318,1)

Detection and identification of proteinsusing nanoparticle–fluorescent polymer‘chemical nose’ sensors

CHANG-CHENG YOU1, OSCAR R. MIRANDA1, BASAR GIDER1, PARTHA S. GHOSH1, IK-BUM KIM2,BELMA ERDOGAN1, SAI ARCHANA KROVI1, UWE H. F. BUNZ2 AND VINCENT M. ROTELLO1*1Department of Chemistry, University of Massachusetts, 710 North Pleasant Street, Amherst, Massachusetts 01003, USA2School of Chemistry and Biochemistry, Georgia Institute of Technology, 770 State Street, Atlanta, Georgia 30332, USA

*e-mail: rotello@chem.umass.edu

Published online: 22 April 2007; doi:10.1038/nnano.2007.99

A sensor array containing six non-covalent gold nanoparticle –fluorescent polymer conjugates has been created to detect, identifyand quantify protein targets. The polymer fluorescence is quenched by gold nanoparticles; the presence of proteins disrupts thenanoparticle –polymer interaction, producing distinct fluorescence response patterns. These patterns are highly repeatable andare characteristic for individual proteins at nanomolar concentrations, and can be quantitatively differentiated by lineardiscriminant analysis (LDA). Based on a training matrix generated at protein concentrations of an identical ultravioletabsorbance at 280 nm (A280 5 0.005), LDA, combined with ultraviolet measurements, has been successfully used to identify 52unknown protein samples (seven different proteins) with an accuracy of 94.2%. This work demonstrates the construction ofnovel nanomaterial-based protein detector arrays with potential applications in medical diagnostics.

The presence of certain biomarker proteins and/or irregular proteinconcentrations is a sign of cancer and other disease states1,2.Sensitive, convenient and precise protein-sensing methodsprovide crucial tools for the early diagnosis of diseases andsuccessful treatment of patients. However, protein detection is achallenging problem owing to the structural diversity andcomplexity of the target analytes. At present, the most extensivelyused detection method for proteins is the enzyme-linkedimmunosorbent assay (ELISA)3. In this system, the captureantibodies immobilized onto surfaces bind the antigen through a‘lock–key’ approach, and another enzyme-coupled antibody iscombined to react with chromogenic or fluorogenic substrates togenerate detectable signals. Despite its high sensitivity, theapplication of this method is restricted because of its highproduction cost, instability and challenges regarding quantification.Although synthetic systems would alleviate some of these concerns,obtaining high affinity and specificity remains quite challenging.

The ‘chemical nose/tongue’ approach provides an alternativefor the sensing protocols that use exclusive analyte–receptorbinding pairs as its basis4. In this strategy, a sensor arrayfeaturing selective receptors, as opposed to ‘lock–key’ specificrecognition, is used for analyte detection. Strategically, the arrayis able to present chemical diversity to respond differentially to avariety of analytes. Over the past few years, this approach hasbeen used to detect a wide range of analytes, including metalions5, volatile agents6, aromatic amines7, amino acids8,9 andcarbohydrates10,11. There have been preliminary studies into theapplication of this strategy to protein sensing, includingHamilton’s porphyrin-based sensors, which are used to identifyfour metal- and non-metal-containing proteins12,13, and Anslyn’s

use of 29 boronic acid-containing oligopeptide functionalizedresin beads to differentiate five proteins and glycoproteinsthrough an indicator-uptake colorimetric analysis14.

The first key challenge for the development of effective proteinsensors is the creation of materials featuring appropriate surfaceareas for binding protein exteriors, coupled with the control ofstructure and functionality required for selectivity. Nanoparticlesprovide versatile scaffolds for targeting biomacromolecules thathave sizes commensurate with proteins15–17, a challengingprospect with small molecule-based systems. Moreover, the self-assembled monolayer on these systems allows facile tuning of arange of surface properties in a highly divergent fashion, enablingdiverse receptors to be rapidly and efficiently produced. Forexample, charged ligand-protected clusters can effectivelyrecognize the protein surface through complementaryelectrostatic and hydrophobic interactions18–20.

The second challenge in protein sensing is the transductionof the binding event. Our strategy for the creation of proteinsensors is to use the particle surface for protein recognition,with displacement of a fluorophore generating the output.As depicted in Fig. 1a, the nanoparticles associate with charge-complementary fluorescent dyes to produce quenched complexes.The subsequent binding of protein analytes displaces the dyes,regenerating the fluorescence. By modulating the nanoparticle–protein and/or nanoparticle–dye association, distinct signalresponse patterns can then be used to differentiate the proteins(Fig. 1b). The fluorescent indicator displacement assay does notrequire special instruments, and its sensitivity (due in large partto the high surface area provided by the nanoparticles) and speedfacilitate protein detection.

ARTICLES

nature nanotechnology | VOL 2 | MAY 2007 | www.nature.com/naturenanotechnology318

Black plate (319,1)

In the current study, we used six readily fabricated structurallyrelated cationic gold nanoparticles (NP1–NP6) to create proteinsensors (Fig. 2a). These particles serve as both selectiverecognition elements as well as quenchers for the polymer. Forour studies, we chose gold rather than other potential corematerials (such as silver) because of its extraordinary stability,in particular its resistance to exchange by amines (forexample, lysine residues)21 and strong quenching ability22.The nanoparticle end groups carry additional hydrophobic,aromatic or hydrogen-bonding functionality engineered to tunenanoparticle–polymer and nanoparticle–protein interactions.For the fluorescent transduction element we used a highlyfluorescent poly(p-phenyleneethynylene) (PPE)23,24 derivative,PPE-CO2 (ref. 25), as a fluorescence indicator. Seven proteinswith diverse structural features including molecular weight andisoelectric point (pI) were used as the target analytes (Fig. 2b).Using these components, we created a competent sensor array,rendering distinct fluorescence response fingerprints forindividual proteins. LDA was performed to identify the proteinpatterns with a high degree of accuracy.

RESULTS AND DISCUSSION

Fluorescence titration was first conducted to assess thecomplexation between anionic PPE-CO2 and cationic goldnanoparticles NP1–NP6. The intrinsic fluorescence of PPE-CO2

was significantly quenched and slightly blue-shifted on additionof all nanoparticles (Fig. 3 for NP3; for the other nanoparticles,see Supplementary Information, Fig. S1). The absorption effectof gold cores was obtained through control experiments usingneutral particles26, and the normalized fluorescence intensities ofPPE-CO2 at 465 nm were subsequently plotted versus the ratio ofnanoparticle to polymer. The complex stability constants (KS)and association stoichiometries (n) were obtained throughnonlinear least-squares curve-fitting analysis (Table 1)19. Complexstabilities vary within approximately one order of magnitude(DDG � 6 kJ mol2 1), and the binding stoichiometry rangesfrom 0.8 for NP6 to 2.9 for NP2. These observations indicate

a

b A

1

2

3

4

B C E F GD

Figure 1 Fluorophore displacement protein sensor array. a, Displacement of

quenched fluorescent polymer (dark green strips, fluorescence off; light green

strips, fluorescence on) by protein analyte (in blue) with concomitant restoration

of fluorescence. The particle monolayers feature a hydrophobic core for stability,

an oligo(ethylene glycol) layer for biocompatibility, and surface charged residues

for interaction with proteins. b, Fluorescence pattern generation through

differential release of fluorescent polymers from gold nanoparticles. The wells on

the microplate contain different nanoparticle–polymer conjugates, and the

addition of protein analytes produces a fingerprint for a given protein.

a

b

S O ON

R9 3Au

S

OO N R

9

3

SO

O

NR

9

3

CH3NP1: R =

CH2CH3NP2: R =

(CH2)5CH3NP3: R =

CH(CH2)5NP4: R =

CH2C6H5NP5: R =

(CH2)3OHNP6: R =

BSA(pI = 4.8, 66.3 kDa)

β-galactosidase(pI = 4.6, 540 kDa)

Cytochrome c(pI = 10.7, 12.3 kDa)

Lipase(pI = 5.6, 58 kDa)

Acid phosphatase(pI = 5.2, 110 kDa)

Alkaline phosphatase(pI = 5.7, 140 kDa)

Subtilisin A(pI = 9.4, 30.3 kDa)

PPE-CO2

O

O–Na+

O

O–Na+O

O

m

Nanoparticle

Figure 2 Structural features of nanoparticles, polymer transducer and

target analytes. a, Chemical structure of cationic gold nanoparticles (NP1–NP6)

and anionic fluorescent polymer PPE-CO2 (m � 12, where m refers to the

number of repeated units in the polymer). b, Surface structural feature and

relative size of seven proteins and the nanoparticles used in the sensing study.

Colour scheme for the proteins: nonpolar residues (grey), basic residues (blue),

acidic residues (red) and polar residues (green).

ARTICLES

nature nanotechnology | VOL 2 | MAY 2007 | www.nature.com/naturenanotechnology 319

Black plate (320,1)

that the subtle structural changes of nanoparticle end groupssignificantly affect their affinity for the polymer. Significantly, allparticle–polymer conjugates were optically transparent over theconcentration range studied.

Once the different binding characteristics of PPE-CO2 withNP1–NP6 were established, the particle–polymer conjugates wereused to sense proteins. The proteins were chosen to have a varietyof sizes and charges, with pI of the seven proteins varying from 4.6to 10.7 and molecular weights ranging from 12.3 to 540 kDa.Within this set there were several pairs of proteins havingcomparable molecular weights and/or pI values, providing achallenging testbed for protein discrimination. In the initialsensing study, 200 ml of PPE-CO2 (100 nM) and stoichiometricnanoparticles NP1–NP6 (the stoichiometric values were takenfrom Table 1) were loaded onto 96-well plates for recording theinitial fluorescence intensities at 465 nm. Under these conditions, itis estimated that .80% of polymer is bound to the nanoparticles,based on the binding constants listed in Table 1, allowingfluorescent enhancement through subsequent displacement.

As illustrated in Fig. 4a, addition of aliquots of protein (5 mM)resulted in a variety of fluorescence responses. By contrast, theaddition of proteins (5 mM) into PPE-CO2 (100 nM) induced

only marginal fluorescence changes (see SupplementaryInformation, Fig. S2), confirming the disruption ofnanoparticle–PPE-CO2 interactions by proteins. BSA, b-galactosidase, acid phosphatase and alkaline phosphatase induceddifferent levels of fluorescence increase, and cytochrome c, theonly metal-containing protein, further attenuated thefluorescence of the systems, presumably through an energy orelectron transfer process27. Lipase and subtilisin A had smaller,but still significant, fluorescence changes for most nanoparticle–PPE systems. Notably, each protein possesses a unique responsepattern. Such an outcome is reasonable, because their interactionwith the protein-detecting array is dependent on surfacecharacteristics such as the distribution of hydrophobic, neutraland charged amino-acid residues. For each protein, we testedits fluorescence responses against the six nanoparticle–PPEassemblies six times, generating a 6 � 6 � 7 matrix.

The raw data obtained were subjected to LDA to differentiatethe fluorescence response patterns of the nanoparticle–PPE

a

b

BSA CC β-Gal Lipase PhosA PhosB

–200

0

200

400

600

800

1,000

ΔI (

a.u.

)

Proteins

NP1NP2NP3NP4NP5NP6

SubA

–25 0 25 50–10

–5

0

5

10

β-Galactosidase

Subtilisin A

Alkaline phosphatase

Acid phosphatase

Lipase

Cytochrome c

BSA

Factor (1), 96.4%

Fact

or (2

), 1.

9%

Figure 4 Array-based sensing of protein analytes at 5 mM. a, Fluorescence

response (DI ) patterns of the NP–PPE sensor array (NP1–NP6) against various

proteins (CC, cytochrome c; b-Gal, b-galactosidase; PhosA, acid phosphatase;

PhosB, alkaline phosphatase; SubA, subtilisin A). Each value is an average of six

parallel measurements. b, Canonical score plot for the first two factors of

simplified fluorescence response patterns obtained with NP–PPE assembly

arrays against 5 mM proteins. The canonical scores were calculated by LDA for

the identification of seven proteins. The 95% confidence ellipses for the

individual proteins are also shown.

0.0 0.5 1.0 1.5 2.00.0

0.2

0.4

0.6

0.8

1.0

445 495 545 5950.0

0.2

0.4

0.6

0.8

1.0

λ (nm)

I (a.

u.)

[NP3] / [PPE-CO2]

I (a.

u.)

Figure 3 Fluorescence intensity changes for PPE-CO2 (100 nM) at 465 nm

on addition of cationic NP3. To eliminate the absorption effect of the gold

core, the fluorescence intensity was calibrated in the presence of relevant

concentrations of tetra(ethylene glycol)-functionalized gold nanoparticles, which

do not associate with PPE-CO2. The inset shows the fluorescence spectra and

the images of PPE-CO2 solution before and after addition of NP3. The arrow in

the inset indicates the direction of spectral changes.

Table 1 Binding constants (logKS) and binding stoichiometries (n ) between

polymer PPE-CO2 and various cationic nanoparticles (NP1–NP6) as determined

from fluorescence titration.

Nanoparticle KS (108 M21) 2DG (kJ mol21) n

NP1 3.0 48.4 2.0NP2 2.1 47.5 2.9NP3 1.7 47.0 1.5NP4 21.0 53.2 1.8NP5 3.6 48.8 2.4NP6 25.0 53.6 0.8

ARTICLES

nature nanotechnology | VOL 2 | MAY 2007 | www.nature.com/naturenanotechnology320

Black plate (321,1)

systems against the different protein targets28. LDA is used instatistics to recognize the linear combination of features thatdifferentiate two or more classes of object or event. It canmaximize the ratio of between-class variance to the within-classvariance in any particular data set, thereby enabling maximalseparability. This analysis reduced the size of the training matrix (6nanoparticles� 7 proteins � 6 replicates) and transformed theminto canonical factors that are linear combinations of theresponse patterns (5 factors � 7 proteins � 6 replicates). The fivecanonical factors contain 96.4%, 1.9%, 0.8%, 0.6% and 0.3% ofthe variation, respectively. The first two factors were visualized ina two-dimensional plot as presented in Fig. 4b. In this plot, eachpoint represents the response pattern for a single protein to thenanoparticle–PPE sensor array.

The canonical fluorescence response patterns of 5 mM proteinsagainst the nanoparticle–PPE sensor array are clustered to sevendistinct groups according to the protein analyte, with no overlapbetween the 95% confidence ellipses. This result demonstratesthat LDA allows the discrimination of very subtle differences inprotein structure. Moreover, LDA provides in-depth quantitativeanalysis of the fluorescence responses of protein analytes. Theassignment of the individual case was based on its Mahalanobisdistances to the centroid of each group in a multidimensionalspace, as the closer a case is to the centroid of one group,the more likely it is to be classified as belonging to that group.The 42 training cases (7 proteins � 6 replicates) can be totallycorrectly assigned to their respective groups using LDA, giving100% accuracy. Furthermore, another 56 protein samples wereprepared randomly and used as unknowns in a blind experiment,where the individual performing the analysis did not know theidentity of the solutions. During LDA analysis, the new caseswere classified to the groups generated through the trainingmatrix according to their Mahalanobis distances. Of 56 cases, 54were correctly classified, affording an identification accuracy of96.4%. This result confirms not only the reproducibility of ourfluorescence patterns, but also the feasibility of practicalapplication of such a nanoparticle-conjugated polymer sensorarray in detection and identification of proteins.

Real-world applications, however, require identification ofproteins at varying concentrations. Varying proteinconcentrations would be expected to lead to the drastic alterationof fluorescence response patterns for the proteins, makingidentification of proteins with both unknown identity andconcentration challenging. To enable the detection of unknownproteins, we have designed a protocol combining LDA andultraviolet (UV) measurements. In this approach, a set offluorescence response patterns were generated at analyte proteinconcentrations that generated a standard UV absorption value at280 nm (A280 ¼ 0.005), the lowest concentration for which theproteins could be substantially differentiated using the givensensor array followed by LDA. Therefore, this concentrationcould also be treated as the detection limit of this assay, withmolar concentrations ranging from 4 nM for b-galactosidase to215 nM for cytochrome c (see Fig. 5 for other proteins). In ourunknown identification protocol, the A280 value of the proteinwas determined, and an aliquot subsequently diluted to A280 ¼0.005 for recording the fluorescence response pattern against theNP–PPE sensing array. Once the identity of the protein wasestablished by LDA, its initial concentration could be determinedfrom the initial A280 value and corresponding molar extinctioncoefficient (1280) according to the Beer–Lambert law.

The fluorescence response patterns where the proteinconcentration is A280 ¼ 0.005 are distinctly different from thosegenerated from 5 mM of proteins, but retain a high degree ofreproducibility (Fig. 5a; see also Supplementary Information,

Table S2). As before, LDA accurately differentiates the proteinpatterns. As shown in Fig. 5b, the canonical fluorescenceresponse patterns display excellent separation, except for a minoroverlap between lipase and subtilisin A. According to the Jack-knifed classification matrix (the classification matrix with cross-validation) in the LDA results, only one subtilisin A sample ismisclassified, affording a classification accuracy of 97.6%(41/42). As a control, analogous analyses were performed usingpolymer ([PPE-CO2] ¼ 100 nM) in the absence of nanoparticles.These studies show that the polymer itself can only substantiallydifferentiate cytochrome c (ref. 27), the metalloprotein, from theother proteins (see Supplementary Information, Fig. S3). For theother six proteins, only 50% classification accuracy is obtainedon the basis of six replicates of measurement, only modestlyhigher than the statistical possibility (that is, 17%). A further

a

b

–20 –10 0 10 20

Factor (1), 81.5%

–15

–5

5

15

Fact

or (2

), 17

.0%

β-Galactosidase

BSA

Acid phosphatase

Alkaline phosphatase

Cytochrome c

Subtilisin A

Lipase

BSA CC β-Gal Lipase PhosA PhosB SubA–200

–150

–100

–50

0

50

100

150

Proteins

ΔI (

a.u.

)

NP1NP2NP3NP4NP5NP6

Figure 5 Array-based sensing of protein analytes with identical absorbance

at 280 nm. a, Fluorescence response (DI ) patterns of the NP–PPE sensor array.

b, Canonical score plot for the first two factors of simplified fluorescence

response patterns obtained with NP–PPE assembly arrays against proteins with

identical absorption values of A ¼ 0.005 at 280 nm. The canonical scores were

calculated by LDA for the identification of seven proteins, with 95% confidence

ellipses for the individual proteins shown. [BSA] ¼ 110 nM; [cytochrome c] ¼

215 nM; [b-galactosidase] ¼ 4 nM; [lipase] ¼ 90 nM; [acid phosphatase] ¼

20 nM; [alkaline phosphatase] ¼ 80 nM; [Subtilisin A] ¼ 190 nM.

ARTICLES

nature nanotechnology | VOL 2 | MAY 2007 | www.nature.com/naturenanotechnology 321

Black plate (322,1)

in-depth examination on the classification accuracy of the polymerin the absence and presence of individual nanoparticles revealedthat the particle–polymer conjugates generally afforded betterdifferentiation abilities than the polymer alone (seeSupplementary Information, Tables S3 and S4), demonstratingthe role of the nanoparticle in providing the differentiationbetween proteins required for effective sensing.

A series of unknown protein solutions were subsequently usedfor quantitative detection. To facilitate solution preparation andUV measurement, the unknown proteins were prepared atvarying concentrations (between 120 nM and 50 mM). Inprinciple, lower concentrations can also be used because thedetection limit of this method is nanomolar. The unknownprotein solutions were submitted to the testing procedures,including determination of A280, dilution of solution to A280 ¼0.005, fluorescence response recording against the sensor array,and LDA. Of the 52 unknown protein samples, only threesamples were incorrectly identified, affording an identificationaccuracy of 94.2% (see Supplementary Information, Table S5, fororiginal data). In addition, the protein concentration wasassessed generally within +5% once it was identified (seeSupplementary Information, Table S5). This resultunambiguously manifests that our sensor array holds substantialpromise for both the identification and quantification ofprotein analytes.

In conclusion, we have demonstrated that the assemblies ofgold nanoparticles with fluorescent PPE polymer provideefficient sensors of proteins, achieving both the detection andidentification of analytes. This strategy exploits the size andtunability of the nanoparticle surface to provide selectiveinteractions with proteins, and the efficient quenching offluorophores by the metallic core to impart efficient transductionof the binding event. Through application of LDA, we are able touse these fluorescence changes to identify and quantify proteinsin a rapid, efficient and general fashion. The robustcharacteristics of the nanoparticle and polymer components,coupled with diversity of surface functionality that can be readilyobtained using nanoparticles, make this array approach apromising technique for biomedical diagnostics.

METHODS

Carboxylate-substituted PPE (PPE-CO2) was synthesized according to a knownprocedure25. The weight- and number-averaged molecular weights of thepolymer are 6,600 and 3,500, respectively. The polydispersity index and degree ofpolymerization of the conjugated polymer are 1.88 and 12, respectively. Thiolligands bearing ammonium end groups were synthesized through the reaction of1,1,1-triphenyl-14,17,20,23-tetraoxa-2- thiapentacosan-25-ylmethanesulphonate with corresponding substituted N,N-dimethylaminesfollowed by deprotection in the presence of trifluoroacetic acid andtriisopropylsilane. Subsequent place-exchange reaction with pentanethiol-coatedgold nanoparticles (d � 2 nm)29 resulted in cationic gold nanoparticles NP1–NP6 in high yields. 1H NMR spectroscopic investigation revealed that the place-exchange reaction proceeds almost quantitatively and the coverage of cationicligands on the nanoparticles is near unity. Bovine serum albumin, cytochrome c(from horse heart), b-galactosidase (from E. coli), lipase (from Candida rugosa),acid phosphatase (from potato), alkaline phosphatase (from bovine intestinalmucosa) and subtilisin A (from Bacillus licheniformis) were purchased fromSigma-Aldrich and used as received. Phosphate buffered saline (PBS, pH 7.4,�1) was purchased from Invitrogen and used as the solvent throughout thefluorescence assays.

In the fluorescence titration study, fluorescence spectra were measured in aconventional quartz cuvette (10 � 10 � 40 mm) on a Shimadzu RF-5301 PCspectrofluorophotometer at room temperature (�25 8C). During the titration,2 mL of PPE-CO2 (100 nM) was placed in the cuvette and the initial emissionspectrum was recorded with excitation at 430 nm. Aliquots of a solution of PPE-CO2 (100 nM) and nanoparticles were subsequently added to the solution in the

cuvette. After each addition, a fluorescence spectrum was recorded. Thenormalized fluorescence intensities calibrated by respective controls(tetra(ethylene glycol)-functionalized neutral nanoparticle with the same core)at 465 nm were plotted against the molar ratio of nanoparticle to PPE-CO2.Nonlinear least-squares curve-fitting analysis was conducted to estimate thecomplex stability as well as the association stoichiometry using a calculationmodel in which the nanoparticle is assumed to possess n equivalent andindependent binding sites19.

In the protein sensing study, fluorescent polymer PPE-CO2 andstoichiometric nanoparticles NP1–NP6 (determined by fluorescence titration;Table 1) were placed into six separate glass vials and diluted with PBS buffer toafford mixture solutions where the final concentration of PPE-CO2 was 100 nM.Then, each solution (200 ml) was respectively loaded into a well on a 96-wellplate (300 ml Whatman Glass Bottom microplate) and the fluorescence intensityvalue at 465 nm was recorded on a Molecular Devices SpectraMax M5 microplate reader with excitation at 430 nm. Subsequently, 10 ml of protein stocksolution (105 mM) was added to each well (final concentration 5 mM), and thefluorescence intensity values at 465 nm were recorded again. The differencebetween two reads before and after addition of proteins was treated as thefluorescence response. This process was repeated for seven protein targets togenerate six replicates of each. Thus, the seven proteins were tested against thesix-nanoparticle array (NP1–NP6) six times, to give a 6 � 6 � 7 training datamatrix. The raw data matrix was processed using classical LDA in SYSTAT(version 11.0). Similar procedures were also performed to identify 56 randomlyselected protein samples.

In the studies featuring unknown analyte protein concentrations, the sensorarray was tested against seven proteins (A280 ¼ 0.005) six times to generate thetraining data matrix. Fifty-two unknown protein solutions were subjectedsuccessively to UV absorption measurement at 280 nm, dilution to A280 ¼ 0.005,fluorescence response pattern recording against the sensor array, and LDA. Afterthe protein identity was recognized by LDA, the initial protein concentration, c,was deduced from the A280 value and corresponding molar extinction coefficient(1280) on the basis of the Beer–Lambert law (c ¼ A280/(1280l)). In theexperimental setup, the protein samples were randomly selected from the sevenprotein species and the solution preparation, data collection and LDA analysiswere performed by different persons.

Received 16 January 2007; accepted 14 March 2007; published 22 April 2007.

References1. Daniels, M. J., Wang, Y., Lee, M.-Y. & Venkitaraman, A. R. Abnormal cytokinesis in cells deficient in

the breast cancer susceptibility protein BRCA2. Science 306, 876–879 (2004).2. Ross, J. S. & Fletcher, J. A. The HER-2/neu oncogene in breast cancer: prognostic factor, predictive

factor, and target for therapy. Stem Cells 16, 413–428 (1998).3. Haab, B. B. Applications of antibody array platforms. Curr. Opin. Biotechnol. 17, 415–421 (2006).4. Albert, K. J. et al. Cross-reactive chemical sensor arrays. Chem. Rev. 100, 2595–2626 (2000).5. Lee, J. W., Lee, J. S., Kang, M., Su, A. I. & Chang, Y. T. Visual artifical tongue for quantitative metal-

cation analysis by an off-the-shelf dye array. Chem. Eur. J. 12, 5691–5696 (2006).6. Rakow, N. A. & Suslick, K. S. A colorimetric sensor array for odour visualization. Nature 406,

710–713 (2000).7. Greene, N. T. & Shimizu, K. D. Colorimetric molecularly imprinted polymer sensor array using dye

displacement. J. Am. Chem. Soc. 127, 5695–5700 (2005).8. Folmer-Andersen, J. F., Kitamura M. & Anslyn E. V. Pattern-based discrimination of enantiomeric

and structurally similar amino acids: an optical mimic of the mammalian taste response. J. Am.Chem. Soc. 128, 5652–5653 (2006).

9. Buryak, A. & Severin, K. A chemosensor array for the colorimetric identification of 20 natural aminoacids. J. Am. Chem. Soc. 127, 3700–3701 (2005).

10. Wright, A. T. & Anslyn, E. V. Differential receptors arrays and assays for solution-based molecularrecognition. Chem. Soc. Rev. 35, 14–28 (2006)

11. Lee, J. W., Lee, J.-S. & Chang, Y.-T. Colorimetric identification of carbohydrates by a pHindicator/pH change inducer ensemble. Angew. Chem. Int. Edn 45, 6485–6487 (2006).

12. Baldini, L., Wilson, A. J., Hong, J. & Hamilton, A. D. Pattern-based detection of different proteinsusing an array of fluorescent protein surface receptors. J. Am. Chem. Soc. 126, 5656–5657 (2004).

13. Zhou, H., Baldini, L., Hong, J., Wilson, A. J. & Hamilton, A. D. Pattern recognition of proteins basedon an array of functionalized porphyrins. J. Am. Chem. Soc. 128, 2421–2425 (2006).

14. Wright, A. T. et al. Differential receptors create patterns that distinguish various proteins. Angew.Chem. Int. Edn 44, 6375–6378 (2005).

15. You, C.-C., Verma, A. & Rotello, V. M. Engineering the nanoparticle–biomacromolecule interface.Soft Matter 2, 190–204 (2006).

16. Rosi, N. L. & Mirkin C. A. Nanostructures in biodiagnostics. Chem. Rev. 105, 1547–1562 (2005).17. Katz, E. & Willner I. Integrated nanoparticle–biomolecule hybrid systems: synthesis, properties, and

applications. Angew. Chem. Int. Edn 43, 6042–6108 (2004)18. Fischer, N. O., McIntosh, C. M., Simard, J. M. & Rotello V. M. Inhibition of chymotrypsin through

surface binding using nanoparticle-based receptors. Proc. Natl Acad. Sci. USA 99, 5018–5023 (2002)19. You, C.-C., De, M., Han, G. & Rotello V. M. Tunable inhibition and denaturation of a-chymotrypsin

with amino acid-functionalized gold nanoparticles. J. Am. Chem. Soc. 127, 12873–12881 (2005).20. Bayraktar, H., Ghosh, P. S., Rotello, V. M. & Knapp, M. J. Disruption of protein–protein

interactions using nanoparticles: inhibition of cytochrome c peroxidase. Chem. Commun.1390–1392 (2006).

ARTICLES

nature nanotechnology | VOL 2 | MAY 2007 | www.nature.com/naturenanotechnology322

Black plate (323,1)

21. Nath, S. et al. Is gold really softer than silver? HSAB principle revisited. J. Nanoparticle Res. 8,

111–116 (2006).

22. Fan, C. et al. Beyond superquenching: hyper-efficient energy transfer from conjugated polymers to

gold nanoparticles. Proc. Natl Acad. Sci. USA 100, 6297–6301 (2003).

23. Bunz, U. H. F. Synthesis and structure of PAEs. Adv. Polym. Sci. 177, 1–52 (2005).

24. Zheng, J. & Swager, T. M. Poly(arylene ethynylene)s in chemosensing and biosensing. Adv. Polym.

Sci. 177, 151–179 (2005).

25. Kim, I.-B., Dunkhorst, A., Gilbert, J. & Bunz, U. H. F. Sensing of lead ions by a carboxylate-

substituted PPE: multivalency effects. Macromolecules 38, 4560–4562 (2005).

26. Aguila, A. & Murray, R. W. Monolayer-protected clusters with fluorescent dansyl ligands. Langmuir

16, 5949–5954 (2000).

27. Sandanaraj, B. S., Demont, R., Aathimanikandan, S. V., Savariar, E. N. & Thayumanavan, S. Selective

sensing of metalloproteins from nonselective binding using a fluorogenic amphiphilic polymer.

J. Am. Chem. Soc. 128, 10686–10687 (2006).

28. Jurs, P. C., Bakken, G. A. & McClelland, H. E. Computational methods for the analysis of chemical

sensor array data from volatile analytes. Chem. Rev. 100, 2649–2678 (2000).

29. Brust, M., Walker, M., Bethell, D., Schiffrin, D. J. & Whyman, R. Synthesis of thiol-derivatisedgold nanoparticles in a two-phase liquid-liquid system. J. Chem. Soc., Chem. Commun.801–802 (1994).

AcknowledgementsThis work was supported by the National Science Foundation (NSF) Center for HierarchicalManufacturing at the University of Massachusetts (NSEC, DMI-0531171), the NSF (VR, CHE-0518487),and the NIH (GM077173). U.B. and I.B.K. thank the Department of Energy for generous financialsupport (DE-FG02-04ER46141).Correspondence and requests for materials should be addressed to V.M.R.Supplementary information accompanies this paper on www.nature.com/naturenanotechnology.

Competing financial interestsThe authors declare no competing financial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

ARTICLES

nature nanotechnology | VOL 2 | MAY 2007 | www.nature.com/naturenanotechnology 323