Label-Free, Single Protein Detection on a Near-Infrared Fluorescent

LETTER pubs.acs.org/NanoLett

Label-Free, Single Protein Detection on a Near-Infrared Fluorescent Single-Walled Carbon Nanotube/Protein Microarray Fabricated by Cell-Free Synthesis Jin-Ho Ahn,† Jong-Ho Kim,† Nigel F. Reuel, Paul W. Barone, Ardemis A. Boghossian, Jingqing Zhang, Hyeonseok Yoon, Alice C. Chang, Andrew J. Hilmer, and Michael S. Strano* Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, United States

bS Supporting Information ABSTRACT: Excessive sample volumes continue to be a major limitation in the analysis of protein
protein interactions, motivating the search for label-free detection methods of greater sensitivity. Herein, we report the first chemical approach for selective protein recognition using fluorescent single-walled carbon nanotubes (SWNTs) enabling label-free microarrays capable of single protein detection. Hexahistidine-tagged capture proteins directly expressed by cell-free synthesis on SWNT/chitosan microarray are bound to a Ni2þ chelated by NR,NR-bis(carboxymethyl)-L-lysine grafted to chitosan surrounding the SWNT. The Ni2þ acts as a proximity quencher with the Ni2þ/SWNT distance altered upon docking of analyte proteins. This ability to discern single protein binding events decreases the apparent detection limit from 100 nM, for the ensemble average, to 10 pM for an observation time of 600 s. This first use of cell-free synthesis to functionalize a nanosensor extends this method to a virtually infinite number of capture proteins. To demonstrate this, the SWNT microarrays are used to analyze a network of 1156 protein
protein interactions in the staurosporine-induced apoptosis of SH-SY5Y cells, confirming literature predictions. KEYWORDS: Protein microarray, nanotube, protein
protein interaction, single-walled carbon nanotubes, label-free detection

D

espite significant progress, protein microarrays remain limited by protein availability and minimum sample requirements, which are dictated by both prelabeling protocols and the detection limits of fluorometric analysis. The labor associated with the creation of protein libraries to comprise such arrays also limits their widespread application. Further, protein stability during array storage compromises robustness.1
3 This has motivated the development of technologies capable of directly transducing protein binding, therefore eliminating the need for intensive sample preparation, labeling, and pretreatment, and hence reducing sample volume requirements. While the DNA microarray is largely successful in terms of economic deployment and widespread usage, the protein equivalent has lagged behind, according to several recent reviews.4
6 Near-infrared (nIR) fluorescent single-walled carbon nanotubes (SWNTs)7 form a class of sensors developed by our lab8,9 and others10,11 to detect primarily small molecule analytes; however, there is no previously reported, generic approach to extend these sensors for the detection of protein
protein interactions. In this work, we demonstrate a synthetic approach that enables the production of label-free protein microarrays capable of single protein detection by using cell-free synthesis onto SWNTs. Cell-free protein synthesis is the in vitro protein expression using cell extract and represents a potentially powerful yet unexplored synthetic tool for nanotechnology research. This ability to resolve single protein binding r 2011 American Chemical Society

events decreases the apparent detection limit by nearly 104 orders of magnitude at an observation time of 600 s. Such SWNT-based protein microarrays can be utilized for the high throughput analysis of protein signaling networks, as we demonstrate for the case of staurosporine-induced apoptosis of SHSY5Y cells. There has been significant progress in using label-free detection methods applied to protein microarrays. One motivating factor has been the avoidance of radioactive materials3 or conjugated labels,12 both of which can be perturbative or require special handling. Success has been demonstrated using novel photonic approaches, including fiber-optic waveguides,13 surface plasmon resonance imaging,14 and optical microcavities.15 As an alternative to less economical optical components, nonoptical methods include electrical transduction using nanowires16,17 and resonant mechanical cantilevers;18 however multiplexing array signals become an engineering challenge with these approaches. While promising, many of these approaches cannot be scaled to single protein detection limits, and none has demonstrated such a limit to date. Received: March 28, 2011 Revised: May 12, 2011 Published: May 31, 2011 2743

dx.doi.org/10.1021/nl201033d | Nano Lett. 2011, 11, 2743–2752

Nano Letters SWNTs, rolled cylinders of graphene, have several advantages as potential fluorometric detectors of protein binding. They possess band gap photoluminescence (PL)19 in the nIR that has no photobleaching threshold,20,21 thus permitting long exposure/ integration times. The photoemission can be sensitive to electrondonating or -withdrawing analytes22
24 or those that change the local dielectric constant,25,26 causing a solvatochromic shift.27,28 While our laboratory and others have developed SWNT PL sensors for important biochemical analytes,22,28,27,29,23,11,24,30,31 a generic scheme for detecting protein
protein interactions has not been previously developed. Such a mechanism would enable the application of SWNTs to label-free protein microarrays. Recent work has also shown that the binding of small molecules that quench the nanotube emission can be detected even at the single molecule level,11,32 but this mechanism has not previously been extended to proteins or other macromolecules. In this mechanism, the nanotube detects the stochastic fluctuations of single molecules adsorbing or desorbing to a single nanotube in real time, allowing one to measure both forward and reverse binding rate constants, the ratio of which is the inverse equilibrium or affinity constant. Extending single molecule detection to larger macromolecules, including proteins, is an aim of the current work for the obvious application to label-free microarrays with low detection limits. Recently, the creation of protein arrays using cell-free protein expression offers significant advantages over conventional protein synthesis in which individual proteins are synthesized, purified, and spotted in segregated, but parallel, processes.33,34 In situ immobilization of proteins during synthesis directly on the array eliminates the need for separate protein preparation and purification. Several protein array systems using cell-free protein synthesis have been reported to overcome the limitations of conventional protein synthesis, such as the nucleic acid programmable protein array (NAPPA),35 the DNA array to protein array (DAPA),36 and a protein array from modified cell-free transcription and translation on the chip surface.37 Despite obvious advantages, these cell-free approaches to protein microarray synthesis have not yet been utilized in label-free technologies, which are able to report protein
protein interaction in real time without labeling. In this work, we demonstrate a label-free protein array based upon SWNT embedded within a chitosan (CHI) matrix bearing an NR,NR-bis(carboxymethyl)-L-lysine (NTA) chelator. With this scaffold, Ni2þ can bind and tether a hexahistidine tagged (His-tag) capture protein directly produced by cell-free synthesis on each spot of a SWNT microarray. Without any prior processing of the analyte, the binding event is directly transduced by a decrease in the SWNT nIR fluorescence intensity. This modulation results from changes in the intermolecular distance between SWNT and the Ni2þ ion, which acts as a proximity quencher of the SWNT photoluminescence. We use cell-free synthesis to extend this mechanism to a large array of 1156 protein
protein interactions in a relevant signaling network important in apoptosis. Upon expression, nascent proteins with C-terminal Histags are directly immobilized by Ni-NTA functional groups on the SWNT/CHI array. Real-time protein
protein interactions can then be detected by monitoring changes in the fluorescent emission from each isolated SWNT. These changes are stochastic fluctuations and report single protein binding events in a manner previously demonstrated for several small molecule analytes.9,11,32

LETTER

The extension of single molecule detection limits to protein quantification in the context of microarrays may significantly advance protein science. Results and Discussion. Our approach to label-free sensing of protein
protein interactions (Figure 1a) is to first suspend SWNTs in solution with CHI, selected for its chemical stability, resistance to nonspecific adsorption of proteins, and biocompatibility.38 SWNTs were directly ultrasonicated and centrifuged (Figure S1, Supporting Information). Then the supernatant was mixed with additional CHI solution (2 wt %) containing glutaraldehyde (0.25%, vol/vol). The resulting suspension was spotted on a patterned glass slide and allowed to cross-link for 6 h at 25 °C in a humidified chamber. Figure 1b shows an optical micrograph of the SWNT/CHI array, and its corresponding nIR fluorescence image was obtained with a Si CCD camera. Each spot exhibits intense and discrete nIR fluorescent spectra as shown in Figure S2 (Supporting Information) with highly uniform spectral properties. After the SWNT/CHI array was carboxylated with succinic anhydride, NTA was coupled to carboxylic acid followed by chelation of Ni2þ. The addition of the chelating Ni2þ ion to the grafted NTA on SWNT/CHI array results in a partial fluorescence quenching similar to the excited state quenching from other divalent ions.39 We find that His-tagged proteins tend to dock in such a way as to alter the distance between the Ni2þ and the SWNT, thereby allowing for their detection. The S30 extract for cell-free protein expression was added to each spot of the SWNT/CHI array after Ni
NTA group attachment, followed by the addition of PCRamplified DNA coding for each protein to initiate protein synthesis.40 Since each in vitro synthesized protein has a C-terminal His-tag, we find that nascent proteins can be in situ immobilized directly on the SWNT/CHI array with Ni
NTA. We distinguish this component as the capture protein and its subsequent binding partner as the analyte protein in this work. To reduce nonspecific binding onto the array, proteins known to bind to the Ni
NTA group were removed from the E. coli S30 extract.41 Significant contamination from wondrous histidine-rich protein (WHP) such as SlyD (29 kDa) is commonly observed when using immobilized metal affinity chromatography (IMAC).42,43 To eliminate nonspecific binding, the E. coli S30 extract was pretreated with Ni
NTA resin prior to protein expression. Eleven E. coli proteins, spanning a range of molecular weights, were chosen as model proteins for the proof of concept of this SWNT-based protein array. Instead of cloning each of the fusion constructs in the expression vector, the PCR-amplified DNAs were used directly as the expression templates for protein synthesis, thereby eliminating the time- and labor-intensive cloning steps. PCR-amplified linear genes can be used as expression templates in a cell-free protein synthesis system, which facilitates the rapid preparation of expressible genes for high-throughput protein synthesis.44
48 The genes coding for each of the 11 proteins were prepared by a two-step PCR method (Figure S3a, Supporting Information). In the first-round PCR, target ORFs were amplified from E. coli K12 genomic DNA with gene specific primers flanked with an overlapping region (Table S1, Supporting Information). The PCR products were purified by gel extraction and used for the second-round PCR, in which the full-length expression templates were synthesized by fusing ORF with regulatory elements including the T7 promoter, ribosomal binding site, T7 terminator, and stop codon (Figure S3b,c, Supporting Information). The cell-free protein synthesis was performed on each SWNT/CHI spot in a humidified chamber at 37 °C for 2744

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters

LETTER

Figure 1. Schematic of a label-free protein array based upon fluorescent single-walled carbon nanotubes. (a) Array fabrication using SWNT/CHI and in situ generation of individually addressed capture proteins using cell-free protein synthesis for label-free optical detection of protein interactions. A SWNT/CHI suspension is spotted on glass and functionalized with Ni-NTA to bind His-tag-containing capture proteins. Cell-free extract and PCRamplified DNA coding for each protein were added to each spot for protein expression and in situ immobilization. (b) Optical and nIR fluorescence image of the SWNT/CHI array. (c) Signal transduction mechanism for label-free detection of protein
protein interactions: a nIR fluorescence change from the SWNT occurs when the distance between the Ni2þ quencher and SWNT is altered upon analyte protein binding.

2 h (Figure 1), after which the array was washed three times with PBS buffer (10 mM, pH 7.4). nIR fluorescence spectra from SWNT in each spot were measured before and after cellfree translation, with an average decrease of 40% after His-tag protein expression and binding (Figure 2a). Controls with DNA absent show no fluorescence change (Figure 2a, control, and Figure S4, Supporting Information). Hence, the capture proteins are effectively expressed by cell-free synthesis on each spot of the array, with direct immobilization by the Ni
NTA group through complexation between Ni2þ and

the His-tag residue of proteins. The other eight proteins tested show the same pattern with slight intensity variations (Figure S5a
h, Supporting Information). Additional evidence that the fluorescence quenching is caused by selective immobilization of the expressed His-tag proteins is provided by experiments where each spot was subsequently treated with imidazole (250 mM) to dissociate His-tag from the Ni
NTA functionalized surface, demonstrating a complete restoration after the treatment (Figure S5i, Supporting Information). Hence, we conclude that the Ni
NTA group can 2745

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters

LETTER

Figure 2. Selective recognition of protein
protein interactions on the SWNT/CHI array. (a) nIR fluorescence spectra of Ni2þ/NTA functionalized SWNT/CHI spotted on the array in response to protein immobilization and subsequent interaction with an anti-His-tag antibody, showing diminution upon capture protein binding and an increase after binding of the anti-His-tag antibody. (b) The nIR fluorescence change (ΔI/Io) after immobilization of 12 distinct capture proteins and subsequent binding of the anti-His-tag antibody. (c) Calibration curve based on the integrated spectral intensity of a single spot as a function of antibody concentration (Kd = 0.56 μM). nIR fluorescence spectra were acquired for 5 s using 785 nm excitation (85 mW at the sample).

allow the SWNT to report the binding and unbinding of the capture protein. Next, we investigated whether the SWNT/CHI array is able to detect the binding of an analyte protein that recognizes the immobilized capture protein. As a model for this interaction, the anti-His-tag antibody was added to each spot containing immobilized His-tag proteins. As shown in Figure 2a,b, the fluorescence intensity increases after addition of the anti-His-tag antibody while no significant change is observed for the control without the His-tagged protein, indicating that the SWNT/CHI array can detect the interaction with the protein. This is the first label-free sensing platform utilizing SWNT fluorescence for detecting analyte
protein interactions. If the integrated fluorescent spot containing SWNT (30 ng/ spot) is utilized to report the binding, the response is similar to that of a conventional array utilizing fluorescent tags. In this ensemble measurement, the fluorescence response of the SWNT collectively can be analyzed as a function of protein concentration (Figure 2c). For illustration, the anti-His-tag antibody was added to each spot on which His-tagged EGFP was immobilized, and a typical sigmoidal equilibrium response curve is observed (Kd = 0.56 μM), yielding an apparent detection limit of 100 nM. In this ensemble measurement, the entire noise background is integrated with the signal. We note that the Kd value for this pair is less than that of a typical antibody binding due to steric hindrance around the binding site attached to the array surface, as previously reported.49,50 We also show that the positive fluorescence response to anti-His-tag antibody is attributed to the displacement of the capture proteins as observed by Western blot (Figure S6, Supporting Information), which results in the increase in the distance of the quencher (Ni2þ) from SWNT. To investigate the limitations on the detection limit of our mechanism, we increased the resolution of imaging on the CHI spot to enable single nanotube interrogation (Figure 3a). Single SWNT fluorescence spectroscopy has been previously demonstrated and is quite robust with essentially an infinite photobleaching lifetime.21 SWNTs do not intrinsically blink, as is the case with quantum dots.51 In previous work, it was shown that when monitored at the single SWNT level, the photoemission will stochastically quench and dequench only in the presence of specific quenching molecules based on their recognition at the interface.11,32 These fluctuations therefore correlate with the adsorption and desorption of individual quenching molecules.

Our primary interest is in using these fluctuations to deduce the average concentration of analyte, CA, molecules above the sensor, and so the quantity of interest is the pseudo-first-order rate constant for adsorption, kf. k i CA kf 1 ¼ ¼ Kd kr kr

ð1Þ

This first-order rate constant can then be used to calculate the concentration of the analyte protein, but note that it may reflect several influences unique to the array conditions (steric hindrance, diffusion limitations, etc.). It is therefore not itself a measure of intrinsic affinity between analyte and capture proteins, but is useful for quantification. To examine whether this single molecule detection scheme can extend to our mechanism of protein detection, nIR imaging of the SWNT/CHI spots was conducted (Figure 3a), allowing us to record in parallel the emission changes from single SWNT responses to protein binding. The nIR fluorescence of SWNT is stable before the addition of analyte in PBS (Figure 3b and Figure S7a, Supporting Information). First, we investigated the fluorescence response to chelation of single Ni2þ ions by NTA on the SWNT/CHI microarray. After Ni2þ was added, one hundred time-intensity traces of the SWNT fluorescence were obtained by measuring the intensity moment of four pixel spots recorded at 1 frame per second. The Ni2þ is a quencher and also binds to the NTA group on the SWNT/CHI matrix. Excitons within the excursion radius of the Ni
NTA chelate are quenched in the vicinity of the binding event, resulting in stepwise fluorescence quenching (Figure 3c and Figure S7b, Supporting Information). The appearance of multiple steps in the traces reflects sequential adsorption events per SWNT spot. The intensity changes are quantized according to the exciton diffusion length of approximately 90 nm11 with a 1 μm long SWNT having up to 1000 nm/ 90 nm ∼ 11 quenching states per diffraction limited spot, each distinguishable in a histogram of fluorescence intensities. Histagged EGFP was then added to SWNT/CHI bearing Ni
NTA, and a similar analysis was conducted with 100 traces. As shown in Figure 3d, single-step fluorescence variations are observed, which arise from the additional quenching of excitons by Ni2þ upon binding of His-tagged EGFP (Figure S7c, Supporting Information). Each trace yields a narrow histogram of non-normalized intensity 2746

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters

LETTER

Figure 3. Single molecule detection of protein
protein interaction on SWNT/CHI microarray. (a) Near-infrared (nIR) fluorescence image of a SWNT/CHI spot showing emission from single isolated SWNT sensors distributed within the CHI matrix. (Inset) nIR emission of a single SWNT. (b) Representative fluorescence time trace (red) without protein addition as a control, showing that SWNT emission is stable with zero mean deflection. The all points histogram (right) shows a single Gaussian distribution. (c) Representative fluorescence time trace (red) for addition of Ni2þ to the NTAbearing SWNT/CHI spot, showing stepwise quenching response. The all points histogram (right) indicates several quenching steps. (d) Representative fluorescence time trace (red) for the addition of His-tag EGFP (3.61 μM) to the SWNT/CHI microarray bearing Ni
NTA, demonstrating additional stepwise fluorescence quenching. The all points histogram (right) also shows several quenching steps. (e) Representative fluorescence time trace (red) for the addition of anti-His-tag antibody (100 nM), showing the clear stepwise fluorescence increase. This stepwise increase response indicates single protein
protein interaction on SWNT. Black lines in all traces denote the fitted trace from error-minimizing step-finding algorithm. (f) The number of transitions for the addition of anti-His-tag antibody at five different concentrations (from 100 nM to 10 pM), showing that the number of transitions increases with increasing concentration of the query protein. This indicates that the number of protein binding events increases with increasing concentration. (g) Histograms of the rate constants (kf) for the His-tag EGFP/anti-His-tag antibody interaction at different concentrations, showing that the rate constant distribution shifts to the left as the concentration of anti-His-tag antibody decreases. This demonstrates that the single molecule detection approach enables us to extract the dynamics of protein
protein interaction from the equilibrium state of ensemble measurement. The pinkcolored bars indicate the center of fitting. 2747

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters

LETTER

Figure 4. Detection of protein
protein interactions. (a) Representative nIR fluorescence spectra of SWNT before and after protein addition. Using Jun, Fos, CDK4, and p16 as queries, selective binary protein
protein interaction was detected through fluorescence intensity changes. (b) nIR fluorescence change (ΔI/I0) for detection of capture protein binding and subsequent interaction with the analyte proteins on SWNT/CHI array. (c) Interaction map, showing the binding network of CDK4, p16, Jun, and Fos. All possible interactions were tested (indicated by the arrow). The arrows show the direction (from capture to analyte) of the interaction with positive interaction signals indicated by red arrows. (d) Mathematical simulation of the detection mechanism using eqs 1 and 2. Sample trajectories of Ni2þ diffusing yet constrained by a 1.65 nm CHI tether. A greater well depth energy results in smaller displacements at greater frequencies. (e) Increasing the well depth results in a tighter distribution of equilibrium positions. (f) Simulated calibration curve relating fluorescence efficiency, E, with increasing well depth, KT via the F€orster quenching mechanism.

changes (in the right side of Figure 3d), indicating that binding of single His-tagged EGFP to the Ni-NTA group in the SWNT/ CHI array can be detected. After applying a stochastic step-fitting algorithm32,52 utilizing a Markov model, the histograms of the rate constants for binding of His-tagged EGFP are obtained as shown in Figure S7d (Supporting Information). Next, single molecule detection of the antibody
protein binding was investigated after adding the anti-His-tag antibody at different concentrations from 100 nM to 10 pM to His-tagged EGFP bound on the SWNT/CHI microarray. Note that in this concentration range, the corresponding ensemble measurement cannot distinguish the interaction above noise (Figure 2c). However, as shown in Figure 3e, stepwise fluorescence increases were observed upon addition of the antibody (100 nM), in which a single-step response reports binding of a single anti-His-tag antibody with His-tagged EGFP on the SWNT (Figure S7e, Supporting Information). Note that, without this addition, the fluorescence remains invariant (Figure 3b), and therefore, the response is a measure of this protein
protein interaction.

The number of transitions per area for single protein
protein interactions increases with increasing concentration of anti-His-tag antibody (Figure 3f). Note that we can still observe a statistically significant number of transitions for single protein interactions even at 10 pM, compared to the control without anti-His-tag antibody. The result highlights the advantage of detecting discretized analyte binding events. We also note that the platform is able to extract the dynamics of protein adsorption (kf and kr) from the equilibrium state, which may be useful in the future to identify and analyze weaker protein interactions. From a typical fluorescence trace, kf can be calculated for each SWNT in the image. As shown in the histograms of rate constants (Figure 3g), their distribution shifts to the left as the concentration of anti-His-tag antibody decreases, indicating a decrease with decreasing analyte concentration as expected. To further demonstrate the detection of protein
protein interactions using this mechanism, we produced human derived His-tag proteins by on-chip cell-free expression and then investigated 2748

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters

LETTER

Table 1. Distance between Ni2þ and SWNTa capture protein Δd1 (nm) Δd2(nm) analyte protein for Δd2 calculation Ack

0.17863


0.06980

anti-His-tag Ab

Dnak

0.21011


0.03993

anti-His-tag Ab

FbaA

0.13853


0.02607

anti-His-tag Ab

GlyA

0.17494


0.05045

anti-His-tag Ab

LpdA

0.17908


0.03964

anti-His-tag Ab

RpoA

0.15028


0.01080

anti-His-tag Ab

RplB

0.11432


0.02945

anti-His-tag Ab

RpsB Tsf

0.20878 0.18831


0.06084
0.01899

anti-His-tag Ab anti-His-tag Ab

Ada

0.15081


0.02568

anti-His-tag Ab anti-His-tag Ab

Cdd

0.15809


0.02948

CDK4

0.07655

0.06338

p16

p16

0.08120

0.05174

CDK4

JUN

0.07671

0.04646

FOS

FOS

0.09560

0.02362

JUN

Δd1 and Δd2 were estimated using the F€orster resonance energy transfer equation (4) and the experimentally observed change in fluorescence quenching assuming approximately 1 nm for the F€orster radius. a

the resulting protein
protein interactions upon addition of their known binding partner (Figure 4). We first tested two binary protein interactions using well-known interacting pairs, CDK4p16 and Jun-Fos. The first capture proteins denoted in red were expressed by adding PCR DNA with His-tag sequence to the cellfree lysate (Figure 4a). After in situ expression of the capture protein containing the His-tag, the second query protein, without the His-tag sequence, was added to the capture protein bound spots, and the fluorescence response was monitored. As shown in Figure 4, the fluorescence is diminished after expression of the capture proteins with His-tag on the SWNT/CHI array. In addition, after treatment of the analyte proteins with the query proteins, the fluorescence further decreases. This is opposite to the results of the His-tag protein and anti-His-tag antibody interaction, suggesting that the specific interactions between p16 and CDK4 (or Jun and Fos) cause further fluorescence decrease (Figure 4b and Figure S8, Supporting Information). These results indicate that the SWNT/ CHI array system is able to report the specific interactions between proteins without additional labeling (Figure 4c). The opposite fluorescence response for these protein interactions, compared to the His-tag protein and anti-His-tag antibody pair described above, is consistent with the proposed mechanism of detection described below. We also note that for p16-CDK4 and Jun-Fos interactions, recognition occurs via displayed peptide sequence instead of the Ni
NTA/His-tag region, which could push the Ni2þ ion closer to the SWNT surface. For consistency, a F€orester quenching model was used to calculate the distances between the Ni2þ ion and the SWNT, based upon the observed fluorescence responses; the results are reported in Table 1. The magnitudes of these changes are consistent, and on average, less than a protein diameter. The distances corresponding to 10 monomer CHI units and the Ni
NTA moiety were estimated from a Hyperchem molecular model. Geometry optimization was performed in the presence of water at 300 K for 1 ps resulting in a distance between SWNT and NTA of 1.65 nm (Figure S9, Supporting Information). The distance changes are also calculated after adding the receptor protein and analyte protein based on the F€orster resonance energy transfer to a nonemissive state (Table 1).

The mechanism of detection involves the distance between the Ni2þ and the nanotube fluorophore (Figrue 4d
f). As with many divalent ions, Ni2þ is an excited state quencher of the nanotube photoluminescence.39,53 The Ni2þ bound to the CHI can be viewed as a freely diffusing entity constrained by the potential well created by the CHI tether. The equilibrium position, p, of the quencher can be found by integrating the sum of the random diffusive and restorative displacements over a series of simulated time steps (Δτ) Δpi ¼ ( r1

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Df ðpi
1 Þ ðΔτÞ 2DðΔτÞ þ kT

ð2Þ

where D is the diffusivity of the complex (10
10 m2/s), r1 is a uniform random number, and f is the restorative force as a function of the current displacement, pi = pi
1 þ Δpi based on a Lennard-Jones 6
12 potential. " # pt 6 pt 12 f ðpÞ ¼
48ε 7
13 ð3Þ p p Here, pt is the location of vanishing restorative force (∼1.65 nm). In this scheme, the Ni2þ, tethered to the CHI, makes periodic excursions biased toward p > pt, and returns to the well minimum with a frequency that increases with the well depth, ε, as shown in Figure 4d. A loosely bound Ni2þ (small ε) will spend more time at the periphery of the potential well, causing minimal quenching. When a protein docks to the capture protein, it can increase the energy of the potential well via increases in electrostatic or van der Waals attractive forces to the complex. This has the effect of decreasing the mean distance of the Ni2þ in the nanotube, resulting in a quenching response. In this way, the probability distribution of position tends to narrow with increasing well depth (Figure 4e). These transient changes in position can be related to the optical signal from the nanotube. The fluorescence efficiency, E, is related to the displacement by assuming a F€orster transfer mechanism to quenching states with F€orster integral, Ro (of order unity). E¼

1 Ro 1þ p

!6

ð4Þ

Equations 2
4 simulate the dynamic trajectory of the Ni2þ complex subjected to the potential well of depth ε. An asymmetric potential well, such as one formed from competing electrostatic and van der Waals attractive forces in DLVO theory ensures that the mean p > pt for all cases. A decrease in E is observed when a captured protein increases the well depth, which it can do by strengthening either electrostatic or van der Waals forces. Figure 4f displays a model calibration curve relating efficiency E with increasing well depth, ε. The magnitude of the decrease in E is similar to what we observe experimentally. Conversely, it is consistent that any subsequent docking to the complex can also decrease ε, causing restoration. A more detailed knowledge of the potential well and protein contributions to it may allow responses to be predicted a priori. Lastly, we directly illustrate the utility of this approach by analyzing an extensive protein
protein network of practical interest. The use of cell free synthesis in the fabrication of the capture protein means that we can extend the detection to virtually any analyte protein of interest. We demonstrate this 2749

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters

LETTER

Figure 5. Protein interaction network involved in staurosporine induced apoptosis in SH-SY5Y cells. (a) Heat map representation of protein
protein interaction (34  34). Each protein was numbered from 1 to 34 and protein
protein interactions of each capture protein i to the analyte protein j were examined. (b) Consistency check by comparison of upper triangle and lower triangle shown in (a). The signals for all interactions were calculated by normalization of fluorescence change. Each SWNT fluorescence was individually taken after capture protein binding and subsequent analyte protein addition. A threshold (T) of protein binding interaction is shown as dotted green lines. Three classes of responses were observed, i.e., i = j < T, i = j > T, i 6¼ j > T and a portion where (i,j) and (j,i) both greater than T was considered as protein
protein interaction. (c) Protein interaction map as determined by our SWNT/CHI protein array. Interactions identified in this analysis are shown as light blue lines. This interaction network comprises 34 human proteins connected by 54 interactions including homodimer interaction. (d) Fluorescence response of homodimer protein interaction. A threshold of protein binding interaction is shown (dotted green line). (e) Interaction comparison between our experimental data and the predicted result by Deighton et al.

by considering the system of Short et al.54 who propose candidate proteins related to apoptosis induced by staurosporine treatment in a human neuroblastoma derived cell line (SH-SY5Y). After examination of representative differential protein expression following the treatment, a proposed binding network was generated using statistical methods in silico by Deighton et al.55

using Ingenuity Pathway Analysis software. In this work, we screen these proposed candidate proteins for actually observable pairwise interactions using our label free microarray, hence informing the previously reported predictions. We constructed a 34  34 protein
protein interaction matrix and investigated it using the SWNT/CHI microarray described in this work. Each of 2750

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters these 34 apoptosis related proteins were used as capture proteins tested against the same proteins as analyte probes to generate this 3434 protein interaction matrix. The result is shown in Figure 5a as a heat map. We first note that the approximately symmetrical map indicates that in the majority of cases, the tethering of the capture protein (with His-tag) does not influence the affinity of analyte capture (no His-tag). To examine this symmetry in more detail, we numbered each protein from 1 to 34, and examined the interactions of each capture protein i to the analyte j. Specifically, one can compare the upper triangular (i,j) and lower triangular (j,i) components of the heat map matrix for consistency and for responses above an empirically determined threshold response, T (Figure 5b). We see three classes of responses. In the first, we observe a population where clearly the (i,j) and (j,i) are in agreement and less than T, corresponding to no interaction. We see a substantial portion where (i,j) and (j,i) are both greater than T, indicating protein
protein interaction. Examples, include BAG4-HSPA4,56 STIP1-DNAJ,57 and PLS1-HSP7058 interactions which were all identified previously (Figure S10, Supporting Information). Overall, we confirm approximately 40 of the interactions predicted by Deighton et al.55 and identify 14 additional interactions. A summary of the network deduced from our data appears in (Figure 5c). We have found 54 binary interactions among the protein in the complex we tested, averaging 3.14 binary interactions per protein molecule. We further probed the heat map matrix for homomultimer formation, by analyzing along the diagonal. We find excellent agreement between known, literature homomultimer formation and our responses (Figure 5d,e and Figure S11, Supporting Information). We do find a small population of ambiguous (i,j) responses (where i 6¼ j > T). These differences can arise because of steric hindrance of the His-tag protein. Once the His-tagged capture protein is immobilized, only the opposing face of the protein can freely participate in binding with the analyte proteins.59 In conclusion, we present a label-free detection platform for the analysis of protein
protein interactions, using fluorometric sensors employing SWNT. Such sensors can detect as few as single protein interactions. We report the first chemical approach to selective protein recognition using fluorescent SWNT enabling label-free microarrays capable of single protein detection. A His-tagged capture protein is coupled to a Ni2þ chelated by a NTA group grafted to chitosan surrounding the SWNT. The Ni2þ acts as a proximity quencher with the Ni2þ/SWNT distance altered upon docking of the analyte protein. We show that this ability to discern single protein binding events decreases the apparent detection limit from 100 nM for the ensemble average to 10 pM for an observation time of 600 s. We also use the arrays to analyze a network of 1156 protein
protein interactions in the staurosporine-induced apoptosis of SH-SY5Y cells, both confirming and augmenting literature analysis. Methods. Synthesis of Ni-NTA Functionalized SWNT-CHI Array. SWNT (10 mg) was added into 10 mL of the CHI solution (0.25 wt %), and the solution was sonicated for 20 min in an ice bath. The SWNT dispersion was centrifuged for 3 h at 13000g and the supernatant was decanted to the reservoir. A 150 μL portion of SWNT suspension was well mixed with 1 mL of the CHI solution (2 wt %) and followed by glutaraldehyde (0.25%, vol/vol). The resulting mixture (1 μL/spot) was spotted on the patterned glass functionalized with poly-L-lysine (Tekdon Inc.) and allowed to stand for 6 h at 25 °C. After the SWNT/CHI array was washed with H2O, the array was dipped in the succinic anhydride solution (0.1 M, NMP) containing N,N-diisopropyethylamine

LETTER

(DIEA, 0.1 M) for 2 h at 25 °C. After the array was washed with NMP and H2O, the carboxylic acid on CHI was activated with EDC-HCl and NHS (0.1 M) for 1.5 h, and the SWNT/CHI array was washed with H2O several times. The array was then treated with the NTA solution (0.1 M, in PBS pH 8.0) for 3 h at 25 °C. Nickel sulfate (100 mM) was added to the SWNT/CHI array for 1 h at 25 °C, and the mixture was washed with H2O several times. Cell-Free Protein Synthesis and On-Chip Protein Expression. The standard reaction mixture for cell-free protein synthesis reactions consisted of the following components in a total volume of 10 μL: 57 mM of Hepes-KOH (pH 8.2), 1.2 mM of ATP, 0.85 mM each of CTP, GTP, and UTP, 0.64 mM of cAMP, 90 mM of potassium glutamate, 80 mM of ammonium acetate, 12 mM of magnesium acetate, 34 μg/mL of L-5-formyl-5,6,7,8tetrahydrofolic acid (folinic acid), 1 mM each of 20 amino acids, 0.17 mg/mL of E. coli total tRNA mixture (from strain MRE600), 2% PEG (8000), 67 mM of creatine phosphate (CP), 5.6 μg/mL of creatine kinase, 3 μL of the S30 extract, and 0.5 μL of PCR products. For protein network experiment, cell-free reaction was slightly modified based on the method reported previously.60,61 Detection of Capture Protein Binding and Protein
Protein Interaction. Reaction mixtures (10 μL) for cell-free protein expression were added to each array wells. The fluorescence response of each SWNT/CHI well was measured to monitor the immobilization of His-tag protein and detect the protein
protein interactions. To initiate protein synthesis, 0.5 μL PCRamplified DNAs were added to each well in a humidified chamber at 37 °C and incubated for 2 h. The arrays were washed three times for 10 min each with PBS buffer (pH 7.4, 100 mM) at room temperature and then PL spectra were taken. To analyze protein
protein interaction on SWNT/CHI array, anti-His-tag antibody or cell-free expressed protein mixtures were added to each well and then PL spectra were taken to compare the intensity change.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional experimental details and figures. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]. Author Contributions †

These authors equally contributed to this work.

’ ACKNOWLEDGMENT This work was supported by a Beckman Young Investigator Award to M.S.S. and the National Science Foundation. A seed grant from the Center for Environmental Health and Science at MIT is also appreciated. J.H.A. and J.H.K. are grateful for the postdoctoral fellowship from the Korea Research Foundation Grant funded by the Korean Government (MOEHRD) (NRF2009-352-D00061; KRF-2007-357-D00086). ’ REFERENCES (1) MacBeath, G.; Schreiber, S. L. Science 2000, 289 (5485), 1760–3. 2751

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752

Nano Letters (2) Zhu, H.; Bilgin, M.; Bangham, R.; Hall, D.; Casamayor, A.; Bertone, P.; Lan, N.; Jansen, R.; Bidlingmaier, S.; Houfek, T.; Mitchell, T.; Miller, P.; Dean, R. A.; Gerstein, M.; Snyder, M. Science 2001, 293 (5537), 2101–5. (3) Ramachandran, N.; Srivastava, S.; LaBaer, J. Proteomics: Clin. Appl. 2008, 2 (10
11), 1444–1459. (4) Hall, D. A.; Ptacek, J.; Snyder, M. Mech. Ageing Dev. 2007, 128 (1), 161–167. (5) Joos, T. Expert Rev. Proteomics 2004, 1 (1), 1–3. (6) Wolf-Yadlin, A.; Sevecka, M.; MacBeath, G. Curr. Opin. Chem. Biol. 2009, 13 (4), 398–405. (7) O’Connell, M. J.; Bachilo, S. M.; Huffman, C. B.; Moore, V. C.; Strano, M. S.; Haroz, E. H.; Rialon, K. L.; Boul, P. J.; Noon, W. H.; Kittrell, C.; Ma, J. P.; Hauge, R. H.; Weisman, R. B.; Smalley, R. E. Science 2002, 297 (5581), 593–596. (8) Kim, J. H.; Ahn, J. H.; Barone, P. W.; Jin, H.; Zhang, J. Q.; Heller, D. A.; Strano, M. S. Angew. Chem., Int. Ed. 2010, 49 (8), 1456–1459. (9) Jin, H.; Heller, D. A.; Kalbacova, M.; Kim, J. H.; Zhang, J. Q.; Boghossian, A. A.; Maheshri, N.; Strano, M. S. Nat. Nanotechnol. 2010, 5 (4), 302–U81. (10) Welsher, K.; Liu, Z.; Sherlock, S. P.; Robinson, J. T.; Chen, Z.; Daranciang, D.; Dai, H. J. Nat. Nanotechnol. 2009, 4 (11), 773–780. (11) Cognet, L.; Tsyboulski, D. A.; Rocha, J. D. R.; Doyle, C. D.; Tour, J. M.; Weisman, R. B. Science 2007, 316 (5830), 1465–1468. (12) Chen, Z.; Tabakman, S. M.; Goodwin, A. P.; Kattah, M. G.; Daranciang, D.; Wang, X. R.; Zhang, G. Y.; Li, X. L.; Liu, Z.; Utz, P. J.; Jiang, K. L.; Fan, S. S.; Dai, H. J. Nat. Biotechnol. 2008, 26 (11), 1285–1292. (13) Hughes, R. C.; Ricco, A. J.; Butler, M. A.; Martin, S. J. Science 1991, 254 (5028), 74–80. (14) Lokate, A. M. C.; Beusink, J. B.; Besselink, G. A. J.; Pruijn, G. J. M.; Schasfoort, R. B. M. J. Am. Chem. Soc. 2007, 129 (45), 14013–14018. (15) Armani, A. M.; Kulkarni, R. P.; Fraser, S. E.; Flagan, R. C.; Vahala, K. J. Science 2007, 317 (5839), 783–787. (16) Zheng, G. F.; Patolsky, F.; Cui, Y.; Wang, W. U.; Lieber, C. M. Nat. Biotechnol. 2005, 23 (10), 1294–1301. (17) Zhong, Z. H.; Wang, D. L.; Cui, Y.; Bockrath, M. W.; Lieber, C. M. Science 2003, 302 (5649), 1377–1379. (18) Burg, T. P.; Godin, M.; Knudsen, S. M.; Shen, W.; Carlson, G.; Foster, J. S.; Babcock, K.; Manalis, S. R. Nature 2007, 446 (7139), 1066–1069. (19) O’Connell, M. J.; Bachilo, S. M.; Huffman, C. B.; Moore, V. C.; Strano, M. S.; Haroz, E. H.; Rialon, K. L.; Boul, P. J.; Noon, W. H.; Kittrell, C. Science 2002, 297 (5581), 593. (20) Li, Q.; Zhang, X.; DePaula, R. T.; Zheng, L.; Zhao, Y.; Stan, L.; Holesinger, T. G.; Arendt, P. N.; Peterson, D. E.; Zhu, Y. T. Adv. Mater. 2006, 18, 3160–3163. (21) Heller, D. A.; Baik, S.; Eurell, T. E.; Strano, M. S. Adv. Mater. 2005, 17 (23), 2793–2798. (22) Barone, P. W.; Baik, S.; Heller, D. A.; Strano, M. S. Nat. Mater. 2005, 4 (1), 86–92. (23) Kim, J.; Heller, D. A.; Jin, H.; Barone, P. W.; Song, C.; Zhang, J.; Trudel, L. J.; Wogan, G. N.; Tannenbaum, S. R.; Strano, M. S. Nat. Chem. 2009, 1, 473–481. (24) Satishkumar, B. C.; Brown, L. O.; Gao, Y.; Wang, C. C.; Wang, H. L.; Doorn, S. K. Nat. Nanotechnol. 2007, 2 (9), 560–564. (25) Perebeinos, V.; Tersoff, J.; Avouris, P. Phys. Rev. Lett. 2004, 92 (25), 257402. (26) Walsh, A. G.; Vamivakas, A. N.; Yin, Y.; Cronin, S. B.; Unlu, M. S.; Goldberg, B. B.; Swan, A. K. Nano Lett 2007, 7 (6), 1485–1488. (27) Heller, D. A.; Jeng, E. S.; Yeung, T. K.; Martinez, B. M.; Moll, A. E.; Gastala, J. B.; Strano, M. S. Science 2006, 311 (5760), 508. (28) Jeng, E. S.; Moll, A. E.; Roy, A. C.; Gastala, J. B.; Strano, M. S. Nano Lett. 2006, 6 (3), 371–375. (29) Heller, D. A.; Jin, H.; Martinez, B. M.; Patel, D.; Miller, B. M.; Yeung, T. K.; Jena, P. V.; H€obartner, C.; Ha, T.; Silverman, S. K. Nat Nanotechnol 2008, 4, 114–120.

LETTER

(30) Welsher, K.; Liu, Z.; Daranciang, D.; Dai, H. Nano Lett. 2008, 8 (2), 586–590. (31) Cherukuri, P.; Bachilo, S. M.; Litovsky, S. H.; Weisman, R. B. J. Am. Chem. Soc. 2004, 126 (48), 15638–15639. (32) Jin, H.; Heller, D. A.; Kim, J. H.; Strano, M. S. Nano Lett. 2008, 8 (12), 4299–4304. (33) He, M. Y.; Stoevesandt, O.; Taussig, M. J. Curr. Opin. Biotechnol. 2008, 19 (1), 4–9. (34) Ramachandran, N.; Larson, D. N.; Stark, P. R. H.; Hainsworth, E.; LaBaer, J. FEBS J. 2005, 272 (21), 5412–5425. (35) Ramachandran, N.; Hainsworth, E.; Bhullar, B.; Eisenstein, S.; Rosen, B.; Lau, A. Y.; Walter, J. C.; LaBaer, J. Science 2004, 305 (5680), 86–90. (36) He, M.; Stoevesandt, O.; Palmer, E. A.; Khan, F.; Ericsson, O.; Taussig, M. J. Nat. Methods 2008, 5 (2), 175–177. (37) Tao, S. C.; Zhu, H. Nat. Biotechnol. 2006, 24 (10), 1253–1254. (38) Yan, L. Y.; Poon, Y. F.; Chan-Park, M. B.; Chen, Y.; Zhang, Q. J. Phys. Chem. C 2008, 112 (20), 7579–7587. (39) Brege, J. J.; Gallaway, C.; Barron, A. R. J. Phys. Chem. C 2009, 113 (11), 4270–4276. (40) Angenendt, P.; Kreutzberger, J.; Glokler, J.; Hoheisel, J. D. Mol. Cell. Proteomics 2006, 5 (9), 1658–1666. (41) Kim, T. W.; Oh, I. S.; Ahn, J. H.; Choi, C. Y.; Kim, D. M. Protein Expression Purif. 2006, 45 (2), 249–254. (42) Scholz, C.; Eckert, B.; Hagn, F.; Schaarschmidt, P.; Balbach, J.; Schmid, F. X. Biochemistry 2006, 45 (1), 20–33. (43) Mitterauer, T.; Nanoff, C.; Ahorn, H.; Freissmuth, M.; Hohenegger, M. Biochem. J. 1999, 342, 33–39. (44) Yabuki, T.; Motoda, Y.; Hanada, K.; Nunokawa, E.; Saito, M.; Seki, E.; Inoue, M.; Kigawa, T.; Yokoyama, S. J. Struct. Funct. Genomics 2007, 8 (4), 173–91. (45) Swartz, J. R.; Jewett, M. C.; Woodrow, K. A. Methods Mol. Biol. 2004, 267, 169–82. (46) Ahn, J. H.; Chu, H. S.; Kim, T. W.; Oh, I. S.; Choi, C. Y.; Hahn, G. H.; Park, C. G.; Kim, D. M. Biochem. Biophys. Res. Commun. 2005, 338 (3), 1346–1352. (47) Ahn, J. H.; Keum, J. W.; Kim, D. M. J. Proteome Res. 2008, 7 (5), 2107–2113. (48) Woodrow, K. A.; Swartz, J. R. Proteomics 2007, 7 (21), 3870–3879. (49) Khan, F.; He, M. Y.; Taussig, M. J. Anal. Chem. 2006, 78 (9), 3072–3079. (50) Nieba, L.; NiebaAxmann, S. E.; Persson, A.; Hamalainen, M.; Edebratt, F.; Hansson, A.; Lidholm, J.; Magnusson, K.; Karlsson, A. F.; Pluckthun, A. Anal. Biochem. 1997, 252 (2), 217–228. (51) Shimizu, K. T.; Neuhauser, R. G.; Leatherdale, C. A.; Empedocles, S. A.; Woo, W. K.; Bawendi, M. G. Phys Rev B 2001, 6320, (20), -. (52) McKinney, S. A.; Joo, C.; Ha, T. Biophys. J. 2006, 91 (5), 1941–1951. (53) Brege, J. J.; Gallaway, C.; Barron, A. R. J. Phys. Chem. C 2007, 111 (48), 17812–17820. (54) Short, D. M.; Heron, I. D.; Birse-Archbold, J. L. A.; Kerr, L. E.; Sharkey, J.; McCulloch, J. Proteomics 2007, 7 (17), 3085–3096. (55) Deighton, R. F.; Kerr, L. E.; Short, D. M.; Allerhand, M.; Whittle, I. R.; McCulloch, J. Proteomics 2010, 10 (6), 1307–1315. (56) Lau, P. P.; Chan, L. J. Biol. Chem. 2003, 278 (52), 52988–52996. (57) King, F. W.; Wawrzynow, A.; Hohfeld, J.; Zylicz, M. EMBO J. 2001, 20 (22), 6297–6305. (58) Jones, S. L.; Wang, J.; Turck, C. W.; Brown, E. J. Proc. Natl. Acad. Sci. U.S.A. 1998, 95 (16), 9331–9336. (59) Seurynck-Servoss, S. L.; Baird, C. L.; Rodland, K. D.; Zangar, R. C. Front. Biosci. 2007, 12, 3956–64. (60) Oh, I. S.; Kim, D. M.; Kim, T. W.; Park, C. G.; Choi, C. Y. Biotechnol. Prog. 2006, 22 (4), 1225–1228. (61) Oh, I. S.; Lee, J. C.; Lee, M. S.; Chung, J. H.; Kim, D. M. Bioprocess Biosyst. Eng. 2010, 33 (1), 127–132. 2752

dx.doi.org/10.1021/nl201033d |Nano Lett. 2011, 11, 2743–2752