WO2010065531A1 - Single molecule protein screening - Google Patents

Abstract

Methods for improving single molecule protein analysis are disclosed. These methods can be used for discovery of new biomarkers, quantitation, and high throughput screening. By way of this invention, surface bound peptides are able to be directly sequenced using a modified Edman degradation followed by detection, e.g., labeled antibody detection. High throughput screening is enabled using pools of molecules (e.g., labeled antibodies) to identify and quantitate individual protein analytes in a biological sample.

Description

SINGLE MOLECULE PROTEIN SCREENING

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority to U.S. Patent Application Serial No. 61/118,676, filed on December 1, 2008, under 35 U.S.C. § 119, the contents of which are hereby incorporated by reference in its entirety.

BACKGROUND

Early detection is important in the treatment of many diseases, e.g., cancer. If cancer could be routinely diagnosed at its earliest stages, it would have an enormous impact on patient care. This result is best illustrated by the example of cervical cancer. Mortality from this disease has dropped 74% since the introduction of the Pap test, a cell- based assay that detects the early signs of cervical cancer. Studies of other cancers have also shown that early detection dramatically improves the 5 -year survival rates (FIGURE

1).

There are two important challenges that must be overcome to make a systematic effort to discover novel protein biomarkers. First, better tools to analyze low-abundance proteins should be developed. While a small number of proteins, such as the albumins, are present at high concentrations in the blood, the vast majority of blood proteins are present at very low levels. In fact, circulating biomarkers are likely to be present at femtomolar or lower concentrations, beneath the detection threshold of most tools used for biomarker discovery. Second, there is a need for a discovery tool that can accurately quantify the absolute abundance of multiple proteins, because many existing biomarkers are only predictive of disease when their expression levels are precisely measured, a rule that will probably also hold true for new biomarkers. Existing tools for biomarker discovery, such as mass spectrometry, provide semi-quantitative, relative measurements of protein abundance. A technology that can analyze a number of proteins in a sensitive and quantitative manner would be a significant advance.

The ability to detect and quantify proteins has lagged behind the ability to analyze nucleic acids. Closing this gap by developing more sensitive and quantitative protein analysis methods would greatly aid efforts to understand cellular processes (Ghaemmaghami S. et al, Nature 2003, 425, 737-741; Cohen A. A. et al., Science 2008, 322, 1511-1516) and to search for protein biomarkers that reveal disease state (Omenn G.S. et ai, Proteomics 2005, 5, 3226-3245; Anderson L. /. Physiol. (London, U.K.) 2005, 563, 23-60). The application of single molecule detection (SMD) methods to proteins holds great promise in this regard for five reasons: (1) Recent advances have made SMD methods inexpensive, robust, and reliable (Harris T. D. et al, Science 2008, 320, 106-109; Roy R. et al, Nat. Methods 2008, 5, 507-516). (2) SMD methods can enable the detection of low-abundance proteins (Sauer M. et ai, J. Appl. Phys. B: Lasers Opt. 1997, 65, 427-431; Li, L. et al, Anal. Chem. 2008, 80, 3999-4006; Loscher F. et al, Anal. Chem. 1998, 70, 3202-3205), which is especially important because the poor sensitivities of current proteomic methods are limiting progress in the area of biomarker discovery (Moul J. W. Clin. Prostate Cancer 2003, 2, 87-97; Munkarah A. et al., Curr. Opin. Obstet. Gynecol. 2007, 19, 22-26). (3) SMD methods can enable protein quantification by employing single molecule counting, which can be significantly more accurate than bulk methods (Wold B. and Myers R. M. Nat. Methods 2008, 5, 19-21; Marioni J. C. et al., Genome Res. 2008, 18, 1509-1517). (4) SMD methods can enable analysis of protein-protein interactions by detecting single-molecule colocalization (Wallrabe H. and Periasamy A. Curr. Opin. Biotechnol. 2005, 16, 19-27). (5) SMD methods for proteins affixed to a surface could enable highly multiplexed immunoassays. For example, by creating -20 overlapping pools of labeled antibodies using a logarithmic pooling strategy like the one used to decode bead-based random microarrays (Gunderson K. L. et al., Genome Res. 2004, 14, 870-877), a single assay could detect the protein targets of all 6,000 nonredundant human proteome antibodies (Berglund L. et al., MoI. Cell. Proteomics 2008, 7, 2019-2027) with only -20 binding rounds.

There are several issues that have limited the feasibility of single molecule immunoassays. One is the lack of a good surface for the SMD of surface-immobilized proteins. A second issue to single molecule immunoassays is the dissociation of antibodies from their individual targets during imaging. Many antibodies rapidly dissociate from their ligands in solution, but surface dissociation is often slower. It is not known whether the surface dissociation rates of antibodies will enable the sensitive detection of single ligand molecules. Finally, a single molecule immunoassay must be able to sample large number of molecules in each experiment to ensure accurate protein quantification and to maximize the dynamic range.

Antibody-based methods are routinely used for protein quantification. Highly sensitive sandwich ELISAs can quantify the abundance of a single protein down to zeptomolar quantities. Since sandwich ELISAs platforms are not amenable to scaling up to multiple targets, antibody microarrays have aimed to fill this gap. Yet they do not obtain the sensitivity of sandwich ELISAs. Promising new alternatives exist in the field of protein quantification. Proximity ligation and bead arrays provide sensitivity comparable sandwich ELISAs and can be multiplexed. However, due to the need for two antibodies per target protein, and for specialized reagents (ligation oligos and bead labeling, respectively) they may not be scalable to proteome-wide throughput.

Therefore, a need exists for methods that allow more effective single molecule protein analysis of biological samples.

SUMMARY

The present invention provides, at least in part, methods for improving single molecule analysis (e.g., detection) of proteins from a sample. In one aspect, the invention provides methods for analyzing proteins (e.g., biomarkers) in a sample, e.g., identifying one or more proteins and/or measuring the expression levels of one or more proteins, e.g., to diagnose diseases (e.g., cancer), e.g., by protein end sequencing. In another aspect, the invention features methods for analyzing (e.g., measuring) the expression levels of proteins (e.g., biomarkers) in a sample, e.g., using pooled molecules (e.g., antibodies). Thus, methods useful for improving single molecule protein analysis are provided herein.

Accordingly, in one aspect, the invention features a method for single molecule protein analysis, e.g., Digital Analysis of Proteins by End Sequencing (DAPES). The method includes the steps of: (a) preparing a substrate (e.g., glass, fused silica, or glass or silica deposited with a metal film, such as, but not restricted to titanium, gold, or aluminum, e.g., etched to make an array) with a first molecule; (b) providing a sample (e.g., a biological sample, e.g., from a patient) comprising one or more analyte proteins; (c) fragmenting analyte proteins into peptides, e.g., utilizing a protease (e.g., proteinase K) or cyanogen bromide; (d) attaching the peptides directly to the substrate or indirectly to the substrate through the first molecule; (e) modifying the N-terminal amino acid of the fragments, e.g., with phenylisothiocyanate (PTC); (f) incubating substrate with a first group of one or more labeled molecules (e.g., antibodies, peptide aptamers, RNA aptamers, DNA aptamers, or engineered proteins (e.g., fused to a fluorescent protein (e.g., GFP))); (g) detecting individually resolvable labeled molecules; (h) repeating steps (f) and (g) 0, 1, 2, 5, 10, 25, 50, 75, 100 or more times with additional groups of labeled molecules; (i) removing the modified N-terminal amino acid; and (j) repeating steps (e) through (i) two or more times using at least a second group of one or more second molecules, wherein at least a partial amino acid sequence of said peptide is determined.

In another aspect, the invention features a method for single molecule protein analysis, e.g., Digital Analysis of Proteins by End Sequencing (DAPES). The method includes the steps of: (a) preparing a substrate (e.g., glass, fused silica, or glass or silica deposited with a metal film, such as, but not restricted to titanium, gold, or aluminum, e.g., etched to make an array) with a first molecule; (b) providing a sample (e.g., a biological sample, e.g., from a patient) comprising one or more analyte proteins; (c) fragmenting analyte proteins into peptides; (d) attaching directly or indirectly peptides to the substrate or the first molecule; (e) incubating substrate with a first group of one or more labeled molecules (e.g., antibodies, peptide aptamers, RNA aptamers, DNA aptamers, or engineered proteins (e.g., fused to a fluorescent protein (e.g., GFP))); (f) detecting individually resolvable labeled molecules; (g) repeating steps (e) and (f) 0, 1, 2, 5, 10, 25, 50, 75, 100 or more times with additional groups of labeled molecules; (h) modifying the N-terminal amino acid of the fragments; (i) removing the modified N- terminal amino acid; and (j) repeating steps (e) through (h) two or more times using at least a second group of one or more second molecules, wherein at least a partial amino acid sequence of said peptide is determined.

Embodiments of the aforesaid methods may include one or more of the following features.

In one embodiment, the preparing step comprises coating the substrate with a protein, e.g., Bovine Serum Albumin (BSA) (e.g., acetylated BSA). In certain embodiments, the coating additionally comprises gelatin. In another embodiment, the preparing step comprises coating the substrate with a cross-linked polyethylene glycol (PEG), e.g., a multiarm PEG. The coating of the substrate can be covalent. For example, the coating can be coupled to a thiol moiety and/or an epoxide moiety on the substrate. In yet another embodiment, the preparing step comprises coating the substrate with a self- assembled monolayer.

In one embodiment, the first molecule is labeled. For example, the label can be a fluorophore. In certain embodiments, the first molecule label interacts with the second molecule label, e.g., a fluorophore, quantum dot, or nanoparticle. In another embodiment, there is at least one molecule in common between first group and second group of labeled second molecules.

In one embodiment, the detecting step produces an image, e.g., a fluorescence image (e.g., acquired using Fluorescence Resonance Energy Transfer (FRET), Total Internal Reflection Fluorescence (TIRF), or Zero Mode Waveguide (ZMW)). In another embodiment, the compilation of the images makes a digital profile, e.g., a digital profile that identifies the analyte proteins.

In one embodiment, the preparing step comprises preparing the substrate with aminoethyl or aminopropyl modification.

In one embodiment, the modifying step comprises modifying N-terminal amino acid with phenylisothiocyanate (PTC). In another embodiment, the antibody specifically recognizes PTC modified amino acids or generally recognizes groups of PTC modified amino acids. In certain embodiments, the groups of labeled molecules can distinguish hydrophobic terminal amino acids, positively charged amino acids, negatively charged amino acids, and small amino acids. In yet another embodiment, the labeled molecule is removed prior to step of (h).

In one embodiment, the fragmenting step (c) is omitted (i.e. the analyte proteins are not fragmented but attached directly to the substrate).

In yet another aspect, the invention features a method for single molecule protein analysis, e.g., Digital Analysis of Proteins Using Pooled Antibodies (DAPPA). The method includes the steps of: (a) preparing a substrate (e.g., glass, fused silica, or glass or silica deposited with a metal film, such as, but not restricted to titanium, gold, or aluminum, e.g., etched to make an array) with a first molecule (e.g., an antibody); (b) providing a sample (e.g., a biological sample, e.g., from a patient) comprising one or more analyte proteins; (c) attaching analyte proteins (e.g., antibodies and/or peptides) directly or indirectly to the substrate or the first molecule; (d) incubating substrate with a first group of one or more labeled second molecules (e.g., antibodies, peptide aptamers, RNA aptamers, DNA aptamers, or engineered proteins (e.g., fused to a fluorescent protein (e.g., GFP))); (e) detecting individually resolvable labeled second molecules; (f) removing the labeled second molecules from the substrate; and (g) repeating steps (d) through (f) two or more times using at least a second group of one or more second labeled molecules.

In one embodiment, the preparing step comprises coating the substrate with a protein, e.g., Bovine Serum Albumin (BSA) (e.g., acetylated BSA). In certain embodiments, the coating additionally comprises gelatin. In another embodiment, the preparing step comprises coating the substrate with a crosslinked PEG, e.g., a multiarm PEG. The coating of the substrate can be covalent. For example, the coating can be coupled to a thiol moiety and/or an epoxide moiety on the substrate. In yet another embodiment, the preparing step comprises coating the substrate with a self-assembled monolayer.

In one embodiment, the first molecule is labeled. For example, the label can be fluorophore. In certain embodiments, the first molecule label interacts with the second molecule label, e.g., a fluorophore, quantum dot, or nanoparticle. In another embodiment, there is at least one molecule in common between first group and second group of labeled second molecules.

In one embodiment, the detecting is an image, e.g., a fluorescence image (e.g., acquired using Fluorescence Resonance Energy Transfer (FRET), Total Internal Reflection Fluorescence (TIRF), or Zero Mode Waveguide (ZMW)). In another embodiment, the compilation of the images makes a digital profile, e.g., a digital profile that identifies the analyte proteins.

The methods described herein can be readily formatted into a high throughput screening. High throughput screening can be enabled using pools of labeled molecules (e.g., antibodies) to identify and quantitate individual protein analytes in a biological sample. In one embodiment, a plurality of samples is analyzed in the high throughput screening. In another embodiment, a plurality of labeled molecules is used for detection in the high throughput screening.

All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety.

Other features, objects, and advantages of the invention will be apparent from the description and drawings, and from the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

FIGURE 1 is a table showing the dramatic improvement of survival rates by early detection of cancer. Source: Omenn and American Cancer Society.

FIGURES 2A-2C depict the Digital Analysis of Protein by End Sequencing (DAPES) Protocol. FIGURE 2A depicts the addition of phenylisothiocyanate to the immobilized peptides on the slide. FIGURE 2B shows that phenylisothiocyanate reacts with the N-termini of the immobilized peptide to form a phenylthiocarbamoyl derivative. FIGURE 2C depicts the removal of the terminal amino acid by lowering the pH and heating the slide. The cycle is repeated to sequence the next amino acid.

FIGURE 3 depicts the off -rate of antibodies bound to single protein molecules. Cy5 labeled antibodies were bound to Cy3 labeled proteins. The slide was kept under constant flow and imaged at various time points to observe dissociation of the complex. After 60 hours, the antibodies were stripped from the slide.

FIGURE 4 depicts the ELISA results for a polyclonal antibody titrated against various dipeptide motifs. The ED motif is the only one that shows significant reactivity (shown as triangles). Motifs that showed no reactivity were KK, RR, EE, KR, KE, KD, RE, RD, RK, EK, DK, ER, DR, DE, and DD.

FIGURE 5 depicts an example of the substrate coating used to reduce nonspecific binding.

FIGURE 6 depicts an example of the Digital Analysis of Proteins Using Pooled Antibodies (DAPPA) strategy.

FIGURE 7 depicts the detection of single protein molecules by antibody binding. Cy3 labeled proteins are immobilized to the slide, imaged (top panel), and then bound with Cy5 labeled antibody (bottom panel). The images are converted to binary form and a 2D correlation is performed. There is significantly more overlap than expected by chance, demonstrating that the antibodies bind the immobilized ligand. Scale Bar = 10mm.

FIGURE 8A is a schematic illustration of the single molecule immunoassay. A chemically adsorbed BSA surface was prepared by reacting BSA with an epoxide-coated glass slide within a flow cell. Unreacted epoxides were quenched, and the BSA was activated for sample immobilization by EDC/NHS. The protein sample (circles) was immobilized to the BSA surface, and unreacted sites were passivated. The flow cell was probed with fluorescently labeled antibody and imaged.

FIGURE 8B depicts the raw TIRF image of Cy5-labeled antibodies (scale bar = 50 μm) that illustrates the nonuniform TIRF illumination.

FIGURE 8C depicts the image processing by standard, single value thresholding allowed only a small portion of the raw image (the brightest spots) to be used for molecule identification.

FIGURE 8D depicts the image processing by iterative thresholding allowed for most of the raw image (regardless of intensity) to be used for molecule identification.

FIGURE 8E depicts nonspecific adsorption of antibodies onto 12 surface protocols. Molecules were counted in 5 x 1,000 μm images, and units were converted to picograms per cm assuming a 155 kDa molecular weight. The chemically adsorbed BSA surfaces suppressed nonspecific adsorption the most.

FIGURES 9A-9C are images demonstrating single antibody detection. To test whether fluorescence objects were in fact single molecules, antibodies of two different colors were mixed together and the number of instances was quantified. Two overlapping objects were observed on the surface. No significant overlap (p = 0.73 Fisher' s Exact Test) was observed between antibodies labeled with Cy3 (FIGURE 9A) or Cy5 (FIGURE 9B) when images are merged (FIGURE 9C) (scale bar = 10 μm). This indicates that each fluorescence object represents a single antibody molecule.

FIGURES 10A-10B depicts the attachment efficiency. The EDC/NHS heterobifunctional crosslinking system can effectively activate BSA molecules on the surface to immobilize target proteins. FIGURE 1OA depicts the number of protein molecules attached to the surface per 2,000 μm with and without EDC/NHS surface activation. FIGURE 1OB depicts images of protein molecules attached to the surface (top) without EDC/NHS surface activation and (bottom) with EDC/NHS (scale bar = 10 μm).

FIGURE 11 depicts the determination of protein accessibility (detection efficiency). The image series illustrates target immobilization, antibody binding, and correlation detection. Each frame is an image of the same position in the flow cell (scale bar ) 2 μm) and shows -21 of the ~10 targets analyzed in each binding experiment, (i)

After target protein immobilization, images of the Cy3 -labeled proteins were acquired (top), and the positions of the proteins were determined by software (bottom), (ii) The surface was probed with antibody, images of bound Cy5-labeled antibodies were acquired (top), and positions of the antibodies were determined by software (bottom), (iii) Positions of the targets (green) and antibodies (red) were overlaid. Yellow pixels represent the colocalized molecules, indicating antibody-bound proteins, (iv) The correlogram analysis of this flow cell indicated that protein and antibody colocalization was nonrandom (i.e., antibodies were specifically binding to targets).

FIGURE 12 depicts the negative control for binding. When protein detection was performed using a nonspecific target protein, the correlogram analysis shows a random distribution of correlations, indicating no specific binding.

FIGURE 13 depicts the protein accessibility (detection efficiency) as a function of antibody concentration. Specific binding (solid) was calculated by subtracting the nonspecific binding (dotted) from total binding (dashed). As much as ~ 70% of the target molecules can be specifically bound, enabling efficient protein detection.

FIGURE 14 depicts the insignificant dissociation of surface-bound antibody: target complexes over 48 hours. Antibody binding onto immobilized Cy3- targets was performed and the number of antibody: target complexes was counted. The flow cell was washed over 48 hours and the number of complexes was analyzed every 8 hours. The number of complexes was plotted over time. A decay of the number of antibody: target complexes over time was not observed, so there is likely an antibody- surface interaction. FIGURE 15 depicts protein rebinding. The image series illustrates two binding rounds separated by a photobleaching step (scale bar = 2 μm). Top: Raw images of Cy3 (green) and Cy5 (red) channels. Yellow represents merged channels. Bottom: Analyzed positions of protein molecules (green), antibody molecules (red), and colocalized antibody-protein molecules (yellow). Circles indicate proteins that were bound in both rounds, (i) First round of antibody binding detects immobilized proteins, (ii) Antibodies are photobleached. (iii) Second round of antibody binding detects many of the same proteins. The results indicate that surface allows for two successful rounds of binding.

FIGURE 16 depicts single molecule protein quantification. Solid line: A linear relationship between the number of antibody molecules and the number of protein molecules on the surface was demonstrated when detecting a purified protein sample. Linear fit R = 0.988; coefficient of variation = 1-7%; lower limit of detection 55 molecules per 1,000 μm image (1.4 pg cm^" ). Dashed line: accurate quantification was achieved in a complex protein sample. Detection of target protein spiked into undiluted rabbit serum produces a quantification curve that deviates only slightly from quantification of the purified sample. No increase in background was observed when detecting in serum.

FIGURE 17 depicts quantification of endogenous IgG in serum. Using single molecule protein quantification, the total IgG levels of a rabbit were measured at various time points after immunization. Top: Single molecule counting at three time points (scale bar = 5 μm). A 70.0% increase ((8.1%) in IgG levels was detected between preimmunization and week 4, and a 11.7 ^'% increase ((4.4%) between weeks 4 and 5. Bottom: Bar graph representations of the above single molecule counting data and of ELISA validation data (black = preimmunization; gray = week 4; white = week 5). Deviation of single protein counting measurements from ELISAs were at most 4.2%. The single-molecule counting data and ELISA data were normalized to the preimmunization time point.

FIGURE 18 depicts an efficient strategy for multiplexed protein detection. DETAILED DESCRIPTION

Protein biomarkers are proteins whose expression levels can be used to detect the presence of disease, predict the future onset of disease, diagnose the severity of disease, or monitor disease progression. There are several protein biomarkers currently in use. For example, prostate-specific antigen (PSA) is a serum protein secreted by the prostate, and elevated levels of PSA are a marker for prostate cancer. PSA-based tests are also useful to monitor for prostate cancer recurrence. Similarly, elevated levels of Alpha Fetoprotein (AFP) and CA- 125 are indicators for hepatocellular carcinoma and ovarian cancer, respectively. These biomarkers have not made a major impact on health care because tests based on PSA or CA- 125 are limited by their low specificities, and hepatocellular carcinoma is so rare that routine screening is not cost effective. However, these examples hint at what is possible if better biomarkers can be found, e.g., a simple blood test that can be used by primary-care physicians to routinely screen the general population and detect cancer at its earliest stages.

Digital Analysis of Proteins by End Sequencing (DAPES), a technology that promises to accurately measure genome- wide protein abundance, is described herein. DAPES will be cost-effective, highly sensitive, and quantitative and is a method comprising, for example:

(a) preparing a substrate with a first molecule;

(b) providing a sample comprising one or more analyte proteins;

(c) fragmenting analyte proteins into peptides;

(d) attaching directly or indirectly peptides to the substrate or the first molecule;

(e) modifying the N-terminal amino acid of the fragments;

(f) incubating substrate with a first group of one or more labeled antibodies;

(g) detecting individually resolvable labeled antibody molecules;

(h) repeating steps (f) and (g) from 0 to 100 times with additional groups of labeled molecules;

(i) removing the modified N-terminal amino acid; and

(j) repeating steps (e) through (i) two or more times using at least a second group of one or more second molecules, wherein at least a partial amino acid sequence of said peptide is determined. For example, to perform DAPES, a large number (-1O⁹) of protein molecules are denatured and cleaved into peptides. These peptides are covalently attached to a glass surface and their amino acid sequences are determined in parallel using a method related to Edman Degradation. In this method, the immobilized peptide molecules are covered with a solution containing phenylisothiocyanate, which reacts with the N-terminus of each peptide to form a stable phenylthiocarbamoyl derivative (PTC-amino acid). The slide is washed and the identity of the terminal amino acid of each peptide molecule is determined through the single molecule detection of antibodies that specifically bind the different PTC-amino acid derivatives. The terminal amino acid is then removed by raising the temperature and lowering pH, and the cycle is repeated to sequence 5-15 amino acids from each peptide on the slide. The absolute concentration of every protein in the original sample can then be calculated based on the number of different peptide sequences observed.

DAPES quantifies protein levels by sequencing the N-termini of millions of immobilized protein molecules in parallel. In DAPES, the following steps are performed (FIGURES 2A-2C): 1) The protein sample is cleaved into peptides by enzymatic or chemical treatment, and these peptides are immobilized on the surface of a microscope slide. 2) Phenylisothiocyanate (PITC), the reagent used in Edman Degradation, is added to the slide and this reacts with the N-terminal amino acid of each peptide to form a phenylthiocarbamoyl derivative (PTC-amino acid, FIGURE 2B). This reaction product is stable at neutral pH. 3) Next, the identity of the N-terminal amino acid of each peptide is determined by performing, for example, 20 rounds of antibody binding, detection, and stripping. In the first round of binding, dye-labeled antibodies that specifically bind both the phenyl group of the phenylthiocarbamoyl derivative and the side chain of one amino acid (e.g., arginine) are used. Because this antibody binds the bulky phenyl group as well as the arginine side chain, it will not bind any internal arginines. Therefore, any protein on the slide that is bound must have an arginine at its N-terminus. After imaging with single molecule resolution, the antibodies are stripped (FIGURE 3) and this procedure is repeated with antibodies that will detect the other PTC-amino acid derivatives. After 20 rounds of binding, detection, and stripping, the identity of the N-terminal amino acid of every peptide molecule on the slide will have been determined. 4) The terminal amino acid is then removed by raising the pH and temperature. 5) Steps 3-5 are repeated to sequence 12-20 amino acids from the N-terminus of each peptide molecule on the slide. The PITC chemistry used in DAPES is the same as that used in Edman Degradation and is efficient and robust (>99% efficiency). Since DAPES operates on single molecules, it will not suffer from the dephasing issues inherent in Edman Degradation.

In some embodiments, the order of the modifying step and the detection step can be reversed.

To quantify the abundance of each protein, all of the peptide sequences obtained by performing DAPES (as many as 500 million per experiment) are mapped back to the human proteome. This map-back procedure is quite robust to errors, so even if a given peptide sequence contains deletions or ambiguous residues there is still a high probability of identifying the correct protein. The number of peptide sequences that map back to a particular protein is then divided by the total number of possible peptides that could be produced by that protein to yield a quantitative measure of the abundance of that protein.

Since the majority of proteins are blocked at their amino-termini, it is important to fragment the sample into peptides before performing DAPES (step (c), above). There are a number of different methods that can be used to fragment the protein samples into peptides. One method of choice is a partial digestion of the sample using a protease with broad substrate specificity (e.g., proteinase K), followed by cleavage with cyanogen bromide. Cyanogen bromide cleaves polypeptides at methionines, leaving a C-terminal homoserine lactone group which can be covalently attached to aminoethyl or amionopropyl-derivatized slides. This method of attachment is attractive because it leaves the N-termini of the peptides free for sequencing. With this strategy, a fraction of the peptides generated by each protein will not react with the slide, but this fraction can be accurately determined a priori from the protein's sequence.

One desirable DAPES strategy utilizes 20 unique antibodies that recognize each of the 20 PTC-amino acids derivatives. However, DAPES can achieve excellent results with only 4 antibodies that can distinguish hydrophobic terminal amino acids, positively charged amino acids, negatively charged amino acids, and small amino acids. In this case, about 7 more cycles of sequencing are performed. The PTC-amino acid moiety bears a strong resemblance to a dipeptide motif - the phenyl group looks approximately like one side chain, and the terminal amino acid provides the other. For example, antibodies that specifically bind dipeptide motifs in different sequence contexts are shown, see FIGURE 4, an ELISA curve of a polyclonal serum that binds to X-X-X-E-D-X-X-X, but does not bind any of the other 15 two amino acid combinations tested, including the closely related motif X-X-X-E-E-X-X-X, which differs from the target ligand by a single carbon group. Polyclonal antibodies against ER, DE, and KD dipeptides were also highly specific.

The DAPES technologies described herein can be used as a discovery tool to find new disease biomarkers and to analyze the abundance levels of all human proteins. Once good biomarkers have been found, it will be important to cost-effectively measure the expression levels of a smaller number (100-1000) of selected proteins to diagnose disease. Towards this goal, a related technology, Digital Analysis of Proteins Using Pooled Antibodies (DAPPA) to measure the expression levels of -1000 pre-selected proteins for about $10-$20 dollars, is described herein.

DAPPA works by taking antibodies that have been raised against individual proteins (e.g., any commercially available monoclonal or polyclonal antibody), labeling them with a fluorescent dye, such as Cy5, and using these to detect single protein molecules attached to a solid surface, for example see FIGURE 5. First, bovine serum

HO _^- albumin is used to block the surface. Next, ethanolamine ^Nh2 is used to cap the remaining reactive sites. Then, EDC cross-linker attaches IgG and anti-goat antibody captures the protein. Many decode methods can be used in DAPPA, so that 1000 biomarkers can be quantified with only 10 rounds of antibody binding, imaging, and removal.

For example, the DAPPA method may comprise:

(a) preparing a substrate with a first molecule;

(b) providing a sample comprising one or more analyte proteins;

(c) attaching analyte proteins directly or indirectly to the substrate or the first molecule;

(d) incubating substrate with a first group of one or more labeled second molecules;

(e) detecting individually resolvable labeled second molecules; (f) removing the labeled second molecules from the substrate; and

(g) repeating steps (d) through (f) two or more times using at least a second group of one or more second molecules.

The illustrative claims appended hereto are intended to form part of the specification as though fully reproduced therein. Additionally, the sequences herein may be chemically synthesized, e.g., artificially produced, or may be isolated or produced using other suitable methods.

Additional features, aspects and examples will be apparent to the person of ordinary skill in the art, given the benefit of this disclosure including, for example, and the specific examples described below

EXAMPLES

Procedures for quantifying single protein molecules affixed to a surface by counting bound antibodies are described herein. For example, key parameters, image acquisition and processing, nonspecific antibody adsorption, sample immobilization, sample accessibility, and surface dissociation, were optimized in a systematic way to enable a single molecule detection of surface-immobilized proteins, e.g., a quantitative immunoassay. Specifically, a chemically adsorbed bovine serum albumin (BSA) surface was found to facilitate the efficient detection of single target molecules with fluorescent antibodies, and these antibodies bound for lengths of time sufficient for imaging billions of individual protein molecules. This surface displayed a low level of nonspecific protein adsorption so that bound antibodies could be directly counted without employing two- color coincidence detection. Endogenous protein abundance was accurately quantified in serum samples by counting bound antibody molecules. The number of antibody molecules that were quantified related linearly to the number of immobilized protein molecules (K = 0.98), and the precision (1-5% CV) facilitated the reliable detection of small changes in abundance (7%). Thus, the procedures described herein allowed for single, surface-immobilized protein molecules to be detected with high sensitivity and accurately quantified by counting bound antibody molecules. Further, flow cells could be probed multiple times with antibodies, suggesting the feasibility to perform multiplexed single molecule immunoassays. EXAMPLE 1: Detection of Cancer Biomarkers by DAPPA

To illustrate the pooling strategy, below is described how DAPPA can be used to detect 10 known cancer biomarkers (CA- 125, PSA, B-HCG, AFP, VEGF, IL-4, IL-10, IL-I alpha, TNF alpha, and IL-7) using four rounds of antibody binding (see FIGURE 6). First, antibodies against each of the 10 biomarkers are labeled with a fluorescent dye (e.g., Cy5). Each antibody is then assigned a number based on the protein that it binds. For example, anti-CA-125 is assigned the number 1, anti-PSA is assigned the number 2, and so on, for each of the 10 antibodies. The numbers are then converted to binary (e.g., anti- CA = 1 = 0001, anti-PSA = 2 = 0010, anti-BHCG = 3 = 0011, etc), and the different antibodies are mixed into four pools based on their binary representation. The first pool consists of antibodies that have a " 1 " in the leftmost column of their binary representation (e.g., IL-I alpha = 8 = 1000, TNF alpha = 9 = 1001, IL-7 = 10 = 1010). The second pool consists of antibodies with a " 1 " in the second column from the left in their binary representation, and so on.

To measure the levels of the 10 biomarkers, a blood sample (or urine, etc) is pipetted onto an activated slide, so that the proteins become covalently attached to the surface. Next, the first pool of antibodies is added to the slide and bound proteins are imaged using a fluorescence microscope (see FIGURE 7). The antibodies are removed and the procedure is repeated with the other 3 pools. Each of the 10 kinds of target proteins will bind an antibody in at least one of the four rounds of binding. To determine the identities of the bound protein molecules, the images are analyzed and a "1" is assigned to the protein if it binds an antibody in that round, and a "0" is assigned if it does not. In this fashion, each protein molecule will produce a binary number that gives its identity. For example, if one protein molecule on the slide is bound in rounds 1 and 3 but not in 2 and 4, then this molecule produces the binary number 1010 (round 1 is the leftmost bit), which is the decimal number 10, the number assigned to IL-7. By counting the number of molecules on the slide with signature 1010, the amount of IL-7 in the sample can be quantified. This procedure can be very robust to false positive and false negative events if a few additional antibody pools are used. DAPPA scales extremely well: in this example, DAPPA was used to quantify the abundance of 10 proteins using 4 pools of antibodies, but with only 6 more pools, > 1000 proteins could be analyzed.

EXAMPLE 2: Iterative Thresholding Improves Detection of Labeled Antibody Molecules

There are a number of formats and methods by which the single molecule detection (SMD) of biomolecules can be achieved (Walter N. G. et ah, Nat. Methods 2008, 5, 475-489). Total internal reflection fluorescence microscopy was chosen as the basis for the single molecule immunoassays. A small amount of protein sample was attached to the surface of a flow cell and probed with fluorescent antibodies. Unbound antibodies were removed and the bound antibodies were directly imaged (FIGURE 8A).

To verify that the detection system could achieve single molecule resolution of fluorescent antibody molecules affixed to glass, Cy3-labeled antibodies were mixed with identical antibodies labeled with Cy 5. The mixture was diluted and reacted to an epoxide-coated glass slide. TIRF imaging with Cy3 and Cy5 channels was performed. The positions of the Cy3 and Cy5 antibodies were determined using software and overlaid (FIGURE 9). The Cy3-labeled molecules did not colocalize with Cy5-labeled molecules more than would be expected by chance (p = 0.78, Fisher's Exact Test), demonstrating that the fluorescence objects detected were not clusters of antibodies (which would have been detected as colocalized molecules) but single antibody molecules.

In these initial experiments, considerably fewer fluorescent antibodies were observed at the edges of the field of view relative to the center of the image. This is due to the nonuniform laser illumination intrinsic to the Nikon optical design (FIGURE 8B). Since this nonuniform illumination greatly reduces the number of antibody molecules that can be analyzed in a single field of view using standard, single value thresholding, an automated and unbiased image processing technique, named "iterative thresholding", was developed (see EXAMPLE 12). The algorithm uses local thresholds to compensate for the lower intensities at the edges and is able to accurately identify the locations of fluorescent antibodies independent of their position within the field of view. The performance of this technique was tested by comparing images processed with the iterative thresholding algorithm with the same images analyzed by single value thresholding. The iterative thresholding algorithm (FIGURE 8D) identified the positions of 14-fold more antibodies per field of view (1408% + 420%), on average, than the standard method (FIGURE 8C) while introducing few false positives with respect to the raw data (sensitivity = 99.96% + 0.07%, specificity = 98.47% + 0.76%). Thus iterative thresholding substantially increased the efficiency of fluorescent antibody analysis using objective TIRF and provided a foundation for our protein quantification method.

EXAMPLE 3: Study of Surface Nonspecific Adsorption

Minimizing the nonspecific adsorption of antibodies to surfaces is critical for the development of single molecule immunoassays because it causes false positive events, decreasing the accuracy and sensitivity of the assays. To find the best surface for single- molecule immunoassays, a systematic search of the literature identified surfaces that were shown to have minimal interactions with antibodies. Surface chemistries previously used for SMD (Heyes C. D. et aL, J. Phys. Chem. B 2004, 108, 13387-13394) and for biosensors (Lange K. and Rapp M. Anal. Biochem. 2008, 377, 170-175; Akkoyun A. et aL, Biosens. Bioelectron. 2002, 17, 655-664; Piehler J. et aL, Biosens. Bioelectron. 1996, 11, 579-590; Masson J. F. et aL, Langmuir 2005, 21, 7413-7420) as well as several that might exhibit low levels of nonspecific protein adsorption, were chosen. Some protocols were followed directly from the literature while others, such as the chemically adsorbed BSA protocols, were modified (see EXAMPLE 12).

The nonspecific adsorption of antibodies for 12 different surface chemistries was quantified. A glass slide was loaded into a flow cell, treated according to a particular surface protocol, and exposed to Cy5-labeled antibody. The unbound antibodies were washed away, and the number of adsorbed antibody molecules was quantified by single molecule counting (FIGURE 8E). Since no ligand was present on the surface of the slide, each surface-bound antibody represented a nonspecific adsorption event. 122-

9 ,600 antibodies were observed per 1,000 μm , (3-68 pg cm -^"2\ ). A chemically adsorbed BSA surface (first developed by Heyes C. D. et aL, J. Phys. Chem. B 2004, 108, 13387- 13394 and modified here to allow adsorption to the glass via epoxide crosslinking and capping) showed the least amount of nonspecific binding. Dextran, aminodextran, and linear polyacrylamide (LPA) surfaces showed moderate adsorption. Among the LPA surfaces, polymers of lower molecular weight outperformed those of higher molecular weight. CM-dextran, glucose, IgG, amino-PEG, and PEG performed the worst.

These finding were consistent with those reported by Heyes C. D. et ai, J. Phys. Chem. B 2004, 108, 13387-13394 who found that chemical immobilization of BSA onto a glass surface provided great reduction in nonspecific adsorption of streptavidin molecules. By atomic force microscopy, it was shown that this surface was highly homogeneous, supporting the hypothesis that the 75 kDa BSA protein creates a neutral, hydrophilic layer that sterically hinders proteins from nonspecifically adsorbing to the sticky silicon dioxide below. On the basis of the performance of the adapted BSA surface, the chemically adsorbed BSA surface was selected for further characterization.

It was found that a chemically coated BSA surface to be a successful in its ability to reduce nonspecific adsorption. However, previously only blocking ability over a 10 minute period was tested (Heyes C. D. et ai, J. Phys. Chem. B 2004, 108, 13387-13394). Much longer experimentation times are often needed for immunoassays, because antibody binding can take several hours to reach equilibrium. Therefore, the amount of nonspecific adsorption that accumulated over a two hour antibody binding period was measured. The nonspecific adsorption increased linearly with antibody concentration and did not saturate at relevant concentrations.

EXAMPLE 4: Robust Immobilization of Protein Ligands on BS A-Coated Glass

It is important that a single molecule immunoassay surface allows for the robust anchoring of ligand molecules. However, it was not clear whether the low background BSA surface discussed above could provide enough functional groups for the attachment of a protein sample. Therefore, to test how efficiently proteins would anchor to the BSA surface using the heterobifunctional cross-linking system l-ethyl-3-(3- dimethylaminopropyl)carbodiimide hydrochloride (EDC) and N-hydroxysuccinimide (NHS), a BSA surface in a flow cell was prepared and the free carboxyl groups on the BSA molecules were activated with EDC and NHS. The flow cell was washed to remove unbound cross-linker and then exposed to Cy3-labeled protein to immobilize the proteins via their primary amines. The flow cell was washed again to remove unbound protein molecules, unreacted cross-linking sites were quenched, and the flow cell was imaged.

Cross-linking proteins to the BSA surface allowed for a 10-fold increase in the number of protein molecules affixed to the surface compared to the surface without EDC/NHS activation (950% + 52%). Also, the proteins were able to be attached at over 1,000 molecules per field of view: a density that allows for highthroughput single- molecule sampling (FIGURE 10). Thus, the EDC/NHS system was able to effectively activate the BSA surface and attach a protein sample. The chemically adsorbed BSA surface with EDC/NHS sample immobilization provided the surface chemistry for all subsequent experiments (FIGURE 8A).

Protein sample attachment is enabled by generating peptide bonds between the solvent-accessible carboxyl groups of the BSA and the primary amine groups of the target proteins. This contrasts the approach of some single molecule studies which have relied on biotin streptavidin linkage (Heyes C. D. et ai, J. Phys. Chem. B 2004, 108, 13387-13394; Braslavsky L. et ai, Proc. Natl. Acad. ScL U.S.A. 2003, 100, 3960-3964). The method described herein does not rely on prelabeling samples by biotinylation, instead taking advantage of endogenous lysine residues present on most proteins. Therefore this approach may provide a more universal way of attaching heterogeneous biological samples.

EXAMPLE 5: Efficient Detection of Single Protein Molecules by Antibody Binding

The accuracy of a single molecule immunoassay depends on the accessibility of target molecules to antibodies; inaccessible ligands will not be detected or counted. There are several mechanisms that can prevent an antibody from binding a ligand immobilized on a solid substrate. Steric, electrodynamic, and thermodynamic variables can hinder binding when repulsive forces of the surface overcome the attractive forces of the antibody-protein complex. Kinetics can also hinder binding if a free energy barrier is sufficiently high to prevent docking on relevant time scales (Heyes C. D. et ah, MoI. Biosyst. 2007, 3, 419-430). To determine what degree these variables affect the accessibility of target molecules to antibodies in the system described herein, the following experiments were performed. To analyze the binding of target molecules by antibodies, a dual-color, single molecule protein accessibility assay was performed (FIGURE 11). Here, the target proteins were labeled with Cy3 and the antibodies were labeled with Cy5. A BSA surface was prepared within a flow cell, and the target proteins were immobilized on the surface. The reactive cross-linking sites were capped, and a preantibody image was acquired. Then, the surface was probed with antibodies, washed away unbound antibodies, and an image was taken. The positions of the antibodies were compared with the positions of the proteins imaged beforehand by overlaying their locations. To verify that the colocalization of proteins and antibodies was a result of specific binding, the correlation between protein and antibody positions was measured and the correlation for randomness was tested (see EXAMPLE 12). The correlogram in FIGURE 11 indicates that antibody binding was specific and not due to chance correlation. (To confirm the specificity of binding, the protein accessibility assay was also performed using a nonspecific target protein with which the antibodies should have had no affinity and a correlogram showing no significant correlations was observed (FIGURE 12).)

To quantify ligand accessibility, the fraction of proteins that were colocalized with antibodies was measured. Then this protein accessibility assay was performed for different antibody concentrations. The total fraction of proteins bound by antibodies is shown by the dashed line in FIGURE 13. To better determine the amount of specific binding, nonspecific binding was estimated based on the observed antibody density and subtracted that from the total binding (see EXAMPLE 12). The dotted line shows the estimated fraction of proteins that overlapped with antibodies as a result of nonspecific binding, while the solid line shows the fraction of specifically bound ligand molecules.

The accessibility curve follows the behavior of fractional occupancy that is expected from binding theory. When 1 μg/mL antibody is used, -70% of the target molecules were specifically bound by antibodies. From these results, single protein molecules can be efficiently detected by counting bound antibody molecules.

In a single molecule immunoassay on solid support, as with some other high throughput molecular biology instruments, the cost effectiveness is determined by the time spent during image acquisition. The more molecules that can be analyzed per hour and moreover, per image, the less costly the instrument becomes. Therefore an automated image-processing algorithm that increases the number of molecules that can be analyzed per image, by 14-fold while maintaining 100% specificity, was developed.

EXAMPLE 6: Surface Dissociation of Antibodies

This example is to determine whether the ligand molecules that failed to be detected in the protein accessibility experiments described above were not detected because they were never bound by antibodies or if they were initially bound by antibodies but the complexes dissociated before imaging. Antibody-ligand interactions are known to have dissociation half-lives in solution ranging from minutes to several hours. However, the surface dissociation rate may be slower due to surface-antibody interactions that stabilize the complex. Therefore, an experiment was designed to measure the surface dissociation rate of antibodies bound to single ligand molecules.

To measure the surface dissociation rate of the antibodies, antibodies were allowed to bind to target proteins that were immobilized on the surface of the flow cell, as previously described. The surface was imaged to determine the starting number of antibody-ligand complexes and then a continual wash was performed to remove unbound antibodies from the flow cell. The surface was imaged every 8 h over a 48 h period. At each time-point the number of antibody-ligand complexes that were lost relative to the starting time point was quantified, and from this the surface dissociation of the antibodies was measured.

Nearly all (>90%) of the colocalized pairs of proteins and antibodies remained intact for 48 h at room temperature (FIGURE 14). Furthermore, antibody dissociation did not follow exponential decay over this time period. Together, these results suggest a strong antibody- surface interaction. The high stability of bound antibodies also explains how single ligand molecules with high efficiency could be detected (i.e., FIGURE 13) even though the flow cell was thoroughly washed.

The half-life of a typical antibody-ligand complex can be as short as several minutes in solution. Such rapid dissociation would pose a serious barrier to the development of a solid phase, single molecule immunoassay because antibodies would be washed off of the surface of the flow cell before they could be detected. Fortunately, surface interactions appear to stabilize antibody-ligand interactions.

99 Using the observed surface dissociation rate, the dynamic range that can theoretically be achieved was calculated. If ligand molecules are immobilized at a density of 1,000 target molecules per image and 10 images are acquired per second (a rate possible with the current generation of charge-coupled device cameras), then one can acquire images of 1,000 x 0.9 x 10 x 60 x 60 x 48 = 1.5 billion target molecules while retaining 90% of the antibodies on the surface. Thus, the observed surface dissociation rate will support a dynamic range of 9 orders of magnitude. This suggests that it should be possible to develop single-molecule immunoassays with a high dynamic range.

EXAMPLE 7: Dual-Round Protein Binding

Ligand rebinding in successive binding rounds could be used to increase detection specificity or to enable efficient sample multiplexing (Gunderson K.L. et al. , Genome Research 2004, 14, 870-877). However, as the surface dissociation experiments illustrated, it was difficult to remove bound antibodies from the surface. This was true even after washing using with a low pH buffer as well as various antibody eluting reagents (data not shown). Therefore, the possibility of rebinding ligands was explored by "erasing" antibodies from the surface via photobleaching. Rebinding after photobleaching might be possible because the antibodies used was polyclonal and could theoretically bind multiple epitopes on a single ligand.

One hurdle to performing multiple binding rounds with an intermediate photobleaching step is that the antibodies that bind in the first round could competitively inhibit the binding of antibodies in subsequent rounds. To test whether competitive binding would be a major phenomenon, Cy3-labeled ligand molecules were probed with Cy5-labeled antibodies as described above and the positions of the bound antibodies were acquired. Then the antibodies were photobleached with 640 nm light before performing a second round of binding with the same antibody. (Target molecules were not bleached.) If antibodies competitively inhibited the second round of binding, then any ligand molecules that were bound in both rounds should not have been observed.

2,829 Cy3-labeled ligand molecules were observed. Of these, 1,497 proteins were bound in round 1, 1,146 were bound in round 2, and 526 (18.6%) were bound twice. Assuming independent binding in round 1 and round 2, 21% of the ligands would be expected to be bound twice. Thus, approximately 87% of the proteins bound in round 1 were available for binding in round 2 (FIGURE 15). This result supports the feasibility of performing multiple rounds of single-molecule protein detection.

EXAMPLE 9: Quantification of Single Molecules by Antibody Binding

To perform a quantitative immunoassay, single, immobilized protein molecules were counted by detecting bound antibodies. Varying amounts of Cy3-labeled protein were affixed onto the surface and the number of immobilized target molecules was quantified by imaging. Then the surface was probed with Cy5-labeled antibodies and the total number of bound antibodies was counted after washing.

The solid line in FIGURE 16 illustrates the relationship between number of antibody molecules and number of protein molecules affixed to the surface. In the range of 55 to 1,676 target molecules per 1,000 μm image, a linear relationship between the number target molecules and antibodies was observed. The lower limit of detection (LOD) of 55 molecules per 1,000 μm image (1.4 pg cm^" ) was achieved by acquiring only five images. It should be possible to detect lower quantities of surface-bound proteins by acquiring greater numbers of images (Li L. et ai, Anal. Chem. 2008, 80, 3999-4006). Given the sample immobilization efficiency and this LOD, proteins in solution down to 100 pM could be detected. In this proof-of-principle study, the attachment efficiency was not maximized but doing so should increase the detection sensitivity (Li L. et al, Anal. Chem. 2008, 80, 3999-4006).

The standard curve displays high correlation (K = 0.98), and precision between 1% and 5% CV was obtained. By acquiring only five images, abundance changes down to 7% can be robustly detected; 99% confidence intervals around each data point were generated, and the widest interval was a 7% deviation. This result demonstrates the utility of digital quantification.

The Cy3 -labeled target protein was also quantified in the presence of serum. Here, Cy3-labeled target protein was spiked at varying concentrations into neat rabbit serum. The complex mixture, including target and nontarget proteins, was immobilized to the BSA surface. The surface was probed with fluorescently labeled antibody and the number of target proteins versus the number of antibodies on the surface was quantified. Similar results to the purified protein detection curve were obtained, demonstrating the robustness of the method in the presence of a complex biological fluid (FIGURE 16, dashed line). The LOD in serum was 390 molecules per 1,000 μm² (10 pg cm^"2) corresponding to a target starting concentration of 1 μg/mL. By comparison, the total concentration of the serum was 74 mg/mL (by dry weight). Therefore, despite the overabundance of serum proteins, the serum introduced almost no background. This indicates that single antibody, direct binding can be used to make specific detection measurements in a highly complex biological fluid.

EXAMPLE 10: Accurate Quantification of an Endogenous Serum Protein

To apply the method described herein to quantify endogenous protein in a biological sample, the amount of total IgG in blood of a rabbit was quantified at various time points after immunization. Serum samples were diluted in PBS, immobilized to flow cell surfaces, and probed with anti-rabbit IgG Cy5-antibody. Then the antibodies remaining on the flow cell surface were quantified after washing.

A 70.0% increase (+8.1%) in total IgG between preimmunization and week 4, as well as a subtle 11.7% increase (+4.4%) between weeks 4 and 5, were detected (FIGURE 17). The single molecule quantitation measurements matched bulk measurements obtained by ELISA, deviating from the gold-standard by at most 4.2% (see EXAMPLE 12). This demonstrates the accuracy of single molecule quantitation in complex, real-world samples.

EXAMPLE 11: Efficient Multiplexing

An important goal for SMD is to perform multiple rounds of antibody binding on a solid surface to allow for efficient multiplexing (Gunderson K.L. et al. , Genome Research 2004, 14, 870-877). This would be achieved by encoding each binding pool with a predetermined mixture of antibodies, so that n protein targets could be quantified in ~log₂« binding rounds (FIGURE 18). The ability to perform multiple rounds of binding would also enable error-checking, since antibodies would get a second pass at detecting a particular target. Additionally, analysis of protein-protein interactions would follow easily from such an approach, since interacting proteins will be present at the same positions on the flow cell. Toward this goal, the serial detection of proteins has been demonstrated by two rounds of antibody binding. A photobleaching step after the first round of binding was used to erase surface-associated fluorescence prior to the second hybridization. Photobleaching was used because the rate at which specifically bound antibodies dissociated from the surface was low enough that it was difficult to completely remove them from the flow cell in a reasonable amount of time. Using this approach, the majority (87%) of proteins that were expected to be bound in two binding rounds were in fact bound twice, indicating that competitive binding by the bleached, surface-bound antibodies was minimal. This lends support to the feasibility of multiple rounds of antibody binding and detection, with each round separated by a photobleaching step. (Alternatively, one could use a cleavable linker between the antibody and fluorophore, which would enable dye removal by exposure to a reducing agent or to light (Mitra R. D. et aL, Anal. Biochem. 2003, 320, 55-65).)

Washing with surfactants and denaturants may allow for better removal of bound antibodies from their targets. For example, it was demonstrated that the efficient stripping of antibodies from Western blots without disrupting protein attachment (Yeung Y. G. and Stanley E. R. Anal. Biochem. 2009, 389, 89-91). To develop such a protocol in a single- molecule setting will require a low-background surface that is also surfactant-compatible (the surfaces described here are not). One type of low background surfactant-compatible surfaces can be surfaces that utilize multiarm PEG nanogels (Tessler L. A. et aL, 2009, in preparation)

Some obstacles for developing a multiplexed single molecule immunoassay might include, e.g., the need to characterize each antibody- Ii gand pair beforehand in order to ensure that the concentration of antibody used in the immunoassay is high enough to ensure maximal binding to its immobilized ligand since each antibody-ligand pair might have variable affinities. However, as antibody production and characterization becomes more standardized, it will become possible to obtain large numbers of well-characterized antibodies. For example, the Human Antibody Initiative has already generated and curated antibodies against over 6,000 human proteins, and they aim to expand the collection to the entire human proteome within the decade (Berglund L. et aL, MoL Cell. Proteomics 2008, 7, 2019-2027). Solid phase single-molecule immunoassays could provide a way to leverage such antibody collections toward high-throughput proteomic applications.

EXAMPLE 12: Materials and Methods

Imaging. All experiments were performed on a Nikon TE-2000 inverted microscope fitted with a total internal reflection fluorescence (TIRF) illuminator (Nikon, Melville, NY). Two lasers, 532 nm/75 mW and 640 nm/40 mW were used for fluorescence excitation (Compass 215M, Cube-40C, Coherent, Santa Clara, CA). Illumination of the sample was controlled through a computer animated shutter (Prior Scientific, Rockland, MA). The 532 nm laser beam was attenuated by a ND 2 neutral density filter (Nikon, Melville, NY). The two beams were coupled into one end of an optical fiber cable using a dichroic mirror (Z532BCM, Chroma, Brattleboro, VT), with the other end of the cable attached to the TIRF illuminator. Before reaching the objective, each laser beam passed through a band-pass filter: HQ545/30 for the green laser and D635/30 for the red laser (Chroma, Brattleboro, VT). Objective type total internal reflection was achieved through a 6Ox TIRF oil objective with index of refraction 1.49 (Nikon, Melville, NY). The chemistry of the assay was performed in a flow cell (see Fluidics) mounted onto the microscope stage. When the lasers are experiencing TIR, an evanescent wave decays exponentially at the glass-water interface into the flow cell to a distance of about 300 nm. TIRF allows for the excitation of only surface-bound fluorophore-labeled antibodies and therefore reduces the overall fluorescence background. The emitted photons from the labeled antibodies were collected by the objective and passed through a dichroic mirror (custom Cy3/Cy5, Semrock, Rochester, NY) and an emission filter for either the green channel (HQ610/75, Chroma, Brattleboro, VT) or the red channel (LP02-647RU-25, Semrock, Rochester, NY). Light was then detected by a charge coupled device (CoolSnap ED, Roper Scientific, Tucson, AZ) which imaged a 140 μm by 100 μm (1,400 pixels x 1,000 pixels) region of the surface.

Immediately prior to image acquisition, the flow cell was washed with 600 μL of PBS and loaded with 600 μL of oxygenscavenger and blink-reduction system32 to prevent dyes from photobleaching and blinking. Then images were acquired in the red and green fluorescence channels at five different positions across the length of the flow cell, with 0.5 s exposure. Custom software written in Metamorph (Molecular Devices, Sunnyvale, CA) and Matlab (Mathworks, Natick, MA) was used to analyze the locations and intensities of the fluorescent molecules.

Fluidics. The analysis substrate was a 40 mm diameter, no. 1.5 glass slide (Erie Scientific, Waltham, MA). The substrate was epoxide-derivatized by the vendor unless otherwise specified in Preparation of Surfaces. The slide was loaded into a flow cell (FSC2, Bioptechs, Butler, PA) fitted with perfusion ports to allow for reagents to be passed over the surface. Reagents were flowed through by a custom-made negative pressure vacuum pump.

Target Proteins. The target proteins were polyclonal goat IgG molecules labeled with an average of eight Cy3 dyes per molecule. The nonspecific target proteins used as a negative control in the target binding accessibility assay were polyclonal rabbit IgG molecules labeled with Cy3. Proteins were obtained from Abeam (Cambridge, MA).

Serum Samples. The serum sample used for the spike-in quantification experiment was obtained from rabbit. The serum samples used for the endogenous protein quantification experiment were from preimmunized, week 4, and week 5 rabbits in an antibody production protocol (for an unrelated study) during which rabbits were immunized with antigen and adjuvant. All serum samples were obtained from 21st Century Biochemicals (Marlboro, MA).

Antibodies. The antibodies used in all experiments with the exception of the endogenous protein quantification experiment were polyclonal anti-goat, labeled with Cy5. The antibodies used to detect endogenous rabbit IgG were polyclonal anti -rabbit, labeled with Cy5. All antibodies were obtained from Abeam (Cambridge, MA).

Preparation of Surfaces. A total of 12 surfaces were generated by protocols taken directly from or adapted from surface blocking protocols in the literature (Jung, S. et al, ChemBioChem 2006, 7, 900-903; Heyes C. D. et al, J. Phys. Chem. B 2004, 108, 13387-13394; Sofia S. J. et al, Macromolecules 1998, 31, 5059-5070 ; Akkoyun A. et al, Biosens. Bioelectron. 2002, 17, 655-664; Piehler J. et al., Biosens. Bioelectron. 1996, 11, 579-590; Piehler J. et al, Bioelectron. 2000, 15, 473-481; Taylor S. et al, Nucleic Acids Res. 2003, 31(16):e87; Masson J. F. et al, Langmuir 2005, 21, 7413-7420; Yanker D. M. and Maurer J. A. MoI Biosyst. 2008, 4, 502-504; Scott E. A. et al, Biomaterials 2008, 29, 4481-4493). Nine of the surface chemistries were generated by the chemical attachment of primary amine groups of a polymer or small molecule to epoxidederivatized glass. The epoxide-coated glass was loaded into the flow cell and washed in 600 μL of phosphate buffered saline pH 7.3 (PBS). The glass was reacted with one of the following solutions in PBS for 1 h at room temperature: 1% bovine serum albumin (BSA) (Fisher Scientific, Pittsburgh, PA), 1% BSA/0.1% cold water fish skin gelatin (Aurion, The Netherlands), 1 M glucose, 10% linear polyacrylamide (LPA) MW 1500 Da, 10% LPA MW 10 kDa, 10% LPA MW 1 MDa, 100 mg/mL amino-PEG (Sigma- Aldrich, St. Louis, MO), 200 μg/mL rabbit IgG (Abeam, Cambridge, MA), and 100 mg/mL aminodextran MW 500 kDa (CarboMer, San Diego, CA). These surfaces were then capped with 1 M Tris pH 8.0 for 20 min. Two of the surfaces were generated by the noncovalent adsorption of a polymer to the glass. Here the epoxide-coated glass was first capped with ethanolamine-HCl pH 8.0 for 20 min and then treated with one of the following solutions in PBS for 1 h at room temperature: 100 mg/mL dextran MW 5 kDa and 1% PEG MW 8 kDa (Sigma- Aldrich, St. Louis, MO). The carboxymethyl (CM) dextran surface was generated as previously described (Akkoyun A. et al. , Biosens. Bioelectron. 2002, 17, 655-664).

Measuring Nonspecific Adsorption. A flow cell containing the surface to be tested was loaded with 600 μL, 100 ng/mL Cy5 antibody. The surface was exposed to the antibody in the dark for 25 min at room temperature. Then, unbound antibodies were removed with a 600 μL PBS wash, and the flow cell was imaged as described above.

Immobilizing Protein Samples. A chemically adsorbed BSA surface was formed as described above, and the surface was activated with 0.2 M l-ethyl-3-(3- dimethylaminopropyl)carbodiimide hydrochloride (EDC) and 0.05 M N- hydroxysuccinimide (NHS) (Pierce, Rockford, IL) in sodium phosphate buffer pH 5.8 (SPB) for 10 min. Free EDC and NHS was washed away with 600 μL of SPB.

The attachment of the protein sample of interest to the activated surface was as follows. For immobilization of purified target protein, 100 ng/mL (unless otherwise specified) of target protein in PBS was loaded into the flow cell. To generate a standard curve of detection, dilutions of target protein in PBS were loaded into the flow cell. To generate a standard curve of detection for target protein in the presence of serum, dilutions of target protein were spiked-in to whole rabbit serum, and the spiked-in serum was loaded into the flow cell. To detect endogenous IgG in serum, whole rabbit serum was diluted 1:105 in PBS and loaded into the flow cell.

Proteins samples that were loaded into the flow cell were allowed to react with the surface for 10 min at room temperature, in the dark. Then, unbound proteins were removed with a 600 μL PBS wash, and unreacted EDC-NHS sites on the BSA surface were quenched with 1 M Tris pH 8.0 for 20 min.

Antibody Binding and Oxygen Scavenging. After a protein sample was immobilized onto the flow cell surface (as described above), the Cy5-labeled antibody was loaded into the flow cell at 100 ng/mL (unless otherwise noted) in PBS and incubated for 2 h in the dark at room temperature. Unbound antibodies were washed away, and the flow cell was imaged.

Analyzing Images by Iterative Thresholding. Software was developed to determine the positions of single fluorescent molecules that overcome the limitations of the nonuniform illumination region inherent to the Nikon TIRF Illumination. A Metamorph Journal that reads in a 12-bit TIFF was created and acts as follows. Intensity thresholds were iterated over from 0 to 4,050 by increments of 50. For each intensity threshold, objects were defined as sets of pixels that 1) have intensity values are greater than the threshold, 2) are contiguous with other pixels within that object and 3) comprise an area between 2 and 20 pixels. The limits of 2 and 20 gave the best precision. Then for each object found at each intensity threshold, the X-Y locations of the object centroids were outputted. This set of 2-D points was imported into MATLAB. A binary output image, of ones and zeros, which contained a 3 x 3 pixel square of ones centered on each of the 2-D points, was created. These binary squares were used to represent the locations of fluorescent molecules for analyzing their positions and abundance.

To measure sensitivity and specificity of iterative thresholding (IT), several raw images of Cy5 antibody molecules were obtained and analyzed by IT and by single value thresholding (SVT), and analyzed versions using each method were outputted. To obtain sensitivity, the fraction of objects present in the SVT image that were also present in the IT image was measured. To obtain specificity, the fraction of objects present in the IT image that were present in the raw image was measured. False positives with respect to the raw image were identifiable by the lack of likeness to a Gaussian point-spread function.

The strong single peak in the correlogram indicates that the correlation is significant. Had the correlation obtained been by chance occurance of overlapping molecules, many peaks of similar height to the highest peak should be expected. In an example, the starting point for the intensity threshold was zero and the intensity threshold increased by 50 intensity units throughout the entire spectrum of intensities (0 - 4096). For each intensity iteration, objects were identified by contiguity of pixels and objects that were too large (> 20 pixels) or too small < 2 pixels) to be single antibodies were removed. The coordinates of the centroid of each of the objects that passed the filter were saved. During the iterations in which the intensity threshold was low, this method identified the dimmest objects, which were located at the periphery of illumination. As higher intensity thresholds were iterated through, objects closer to the middle were identified. As the last iterations were being conducted, only the brightest objects in the middle were identified. Finally, the algorithm used the locations obtained during the iterative thresholding to generate an output image that has each of the fluorescent objects represented by equally sized objects of 4 x 4 pixels.

Measuring Protein Detection Efficiency. For analyzing the dual-color assays, the acquired Cy5 image (antibody) and the corresponding Cy3 image (protein) were processed by the IT algorithm into binary images. Then, a 316 pixel x 316 pixel (31.6 μm x 31.6 μm) region from the Cy5 and Cy3 images was chosen to calculate the antibody-protein correlation. The correlation between red and green channels was calculated as follows (Equation 1):

Here, R and G are matrices of ones and zeros, representing the binary image of size 316 x 316 pixel . To correct for stage shifting, the Cy5-Cy3 image pair was allowed to be offset with respect to each other in order to find the alignment that produced the maximum correlation (the true alignment). Once the true alignment was found, the software counted the number of proteins that overlapped antibodies and divided that by the total number of proteins. This ratio was defined as the fractional accessibility or binding efficiency.

Testing for Random Correlations and Specificity of Binding. To determine whether the Cy5-Cy3 image correlation was a random event, the correlation derived from Equation 1 was compared to the background distribution of correlations for Cy5-Cy3 image pairs that were offset in the X and Y directions. To do this, we chose a 316 x 316 pixel² region from the Cy3 image as the "base". The correlation of the base with regions of the Cy5 image that were misaligned by a translational offset was computed. Offsets between -100 to +100 pixels (with respect to the true alignment) in both the X and Y directions were scanned. After these 40,000 correlations were computed, they were plotted as a function of the X and Y offsets. To interpret the Z-axis of the correlograms, level surfaces correspond to the background distribution of correlation values, and peaks correspond to correlations that are significantly nonrandom. In binding experiments in which the antibody was specific for its target (FIGURE 11), a high peak was seen around the offset (0, 0). Therefore correlation for the true alignment was nonrandom, and binding was specific. By contrast, when a nonmatching protein target was used, no peak appeared (FIGURE 12), indicating randomness between Cy3 and Cy5 channels (and no specific binding).

Estimating Nonspecific Binding Based on the Observed Antibody Density. The total number of antibodies remaining on the surface after washing was used to estimate the frequency of antibody-ligand correlations that occurred merely by chance overlap of molecules - the false positive (FP) rate. The FP rate was defined as the probability that a randomly chosen pixel will be within a radius 2.5 pixels from an antibody pixel. This probability follows a Poisson process, where the parameter lambda is the frequency of antibody pixels out of the total number of pixels. Therefore, for the number of antibodies on the surface A, and total pixel area of the image T,

~ >

T . (21

Quantifying Total IgG in Rabbit Serum by ELISA. The three rabbit serum samples were used as coating antigens. The detection antibody was polyclonal anti- rabbit antibody conjugated to alkaline-phosphatase (Abeam, Cambridge, MA). Polystyrene microtiter plates (Immulon 2HB) were obtained from Thermo Fisher Scientific (Waltham, MA). Washes were performed using Labsystems Multidrop 384 (Beckman Coulter, Fullerton, CA). Detection of the fluorogenic substrate, (4- methylumbelliferyl phosphate, Sigma Aldrich, St. Louis) was performed on the microtiter plate flourimeter Synergy HT (Biotek, Winooski, VT).

An indirect enzyme-linked immunosorbent assay (ELISA) was performed as described (Hornbeck P. Curr Protoc Immunol 2001, Chapter 2, Unit 2 1). For each serum sample, two dimensional titrations were performed to determine the optimal dilutions of coating antigen and detection antibody. A dynamic range of detection that spanned the signals of all three serum samples was achieved using the following dilutions. Coating antigens - 1:64,000. Detection antibody - 625 ng/ml.

Efficient multiplexing. An efficient strategy for multiplexed protein detection log ₂ ή was illustrated here, in which n proteins may be quantified in binding rounds, where c is the number of independent fluorescence channels used for antibody detection. This logarithmic encoding is based on the method by Gunderson et ah, used to decode bead-based random microarrays (Gunderson K. L. et ah, Genome Res. 2004, 14, 870- 877). A 7-plex assay, using 1 fluorescence channel, was described herein, as a small- scale example.

First, each protein is assigned a unique digital signature. Next, fluorescent antibodies for each protein are pooled into combinations that are determined by the columns of the digital signatures. (In the example above, the three columns of the signatures dictate the compositions of the three "antibody pools".) Then, immobilized proteins are probed by one antibody pool per binding round. In each binding round, proteins of different species are bound and detected. In between binding rounds, antibodies are stripped. After probing with all of the antibody pools (three in this example), the history of binding at each position on the slide is analyzed. In this manner, each position on the flow cell becomes represented by a binding signature. Finally, the pre-assigned digital signatures are used to decode the flow cell positions into protein identities. Moreover, the number of occurrences of each signature is counted to determine protein abundance. To illustrate the decoding procedure, in this example there are two positions on the flow cell that have the binding history 1-0-0 (i.e. bound in round 1, unbound in round 2, and unbound in round 3). This signature corresponds to Protein 4, so the number of instances of that signature on the flow cell (two), indicates the abundance of Protein 4.

Equivalents

The present invention is not to be limited in scope by the specific embodiments described herein. Indeed, various modifications of the invention in addition to those described herein will become apparent to those skilled in the art from the foregoing description and accompanying figures. Such modifications are intended to fall within the scope of the appended claims.

Claims

We claim:

1. A method comprising:

(a) preparing a substrate with a first molecule;

(b) providing a sample comprising one or more analyte proteins;

(c) attaching the analyte proteins directly to the substrate or indirectly to the substrate through the first molecule;

(d) incubating substrate with a first group of one or more labeled second molecules;

(e) detecting individually resolvable labeled second molecules;

(f) removing the labeled second molecules from the substrate; and

(g) repeating steps (d) through (f) two or more times using at least a second group of one or more second molecules.

2. The method of claim 1, wherein the substrate is glass or fused silica.

3. The method of claim 2, wherein the glass or silica is deposited with a metal film.

4. The method of claim 3, wherein the metal film is titanium, gold, or aluminum.

5. The method of claim 3, wherein the metal film is etched to make an array.

6. The method of claim 1, wherein the preparing step comprises coating the substrate with a protein.

7. The method of claim 6, wherein the coating comprises Bovine Serum Albumin (BSA).

8. The method of claim 7, wherein the BSA is acetylated.

9. The method of claim 6, wherein the coating additionally comprises gelatin.

10. The method of claim 1, wherein the preparing step comprises a cross-linked polyethylene glycol (PEG).

11. The method of claim 10, wherein the PEG comprises a multiarm PEG.

12. The method of claim 6, wherein the coating of the substrate is covalent.

13. The method of claim 12, wherein the coating is coupled to a thiol moiety or epoxide moiety on the substrate.

14. The method of claim 1, wherein the preparing step comprises coating the substrate with a self-assembled monolayer.

15. The method of claim 1, wherein the analyte proteins are covalently attached directly to the coating moiety.

16. The method of claim 1, wherein the analyte proteins are antibodies or peptides.

17. The method of claim 1, wherein the first molecule is labeled.

18. The method of claim 17, wherein the label is a fluorophore.

19. The method of claim 17, wherein the first molecule label interacts with the second molecule label.

20. The method of claim 1 , wherein the second molecule label is a fluorophore, quantum dot, or nanoparticle.

21. The method of claim 1, wherein there is at least one molecule in common between first group and second group of labeled second molecules.

22. The method of claim 1, wherein the first molecule is an antibody.

23. The method of claim 1, where the second molecule is an antibody.

24. The method of claim 1, where the second molecule is a peptide aptamer.

25. The method of claim 1, where the second molecule is a DNA aptamer.

26. The method of claim 1, where the second molecule is an RNA aptamer.

27. The method of claim 1, where the second molecule is an engineered protein.

28. The method of claim 27, wherein the engineered protein is fused to a fluorescent protein.

29. The method of claim 1, wherein the detecting step produces an image.

30. The method of claim 29, wherein the image is a fluorescence image.

31. The method of claim 30, wherein the image is acquired using Fluorescence Resonance Energy Transfer (FRET), Total Internal Reflection Fluorescence (TIRF), or Zero Mode Waveguide (ZMW).

32. The method of claim 29, wherein the compilation of images make a digital profile.

33. The method of claim 32, wherein the digital profile identifies the analyte proteins.

34. A method comprising:

(a) preparing a substrate with a first molecule;

(b) providing a sample comprising one or more analyte proteins; (c) fragmenting the analyte proteins into peptides;

(d) attaching the peptides directly to the substrate or indirectly to the substrate through the first molecule;

(e) modifying the N-terminal amino acid of the fragments;

(f) incubating the substrate with a first group of one or more labeled molecules;

(g) detecting individually resolvable labeled molecules;

(h) repeating steps (f) and (g) from 0 to 100 times with additional groups of labeled molecules.

(i) removing the modified N-terminal amino acid; and

(j) repeating steps (e) through (i) two or more times using at least a second group of one or more second molecules. wherein at least a partial amino acid sequence of said peptide is determined.

35. The method of claim 34, wherein the substrate is glass or fused silica.

36. The method of claim 35, wherein the glass or silica is deposited with a metal film.

37. The method of claim 36, wherein the metal film is titanium, gold, or aluminum.

38. The method of claim 36, wherein the metal film is etched to make an array.

39. The method of claim 34, wherein the preparing step comprises coating the substrate with a protein.

40. The method of claim 39, wherein the coating comprises Bovine Serum Albumin (BSA).

41. The method of claim 40, wherein the BSA is acetylated.

42. The method of claim 39, wherein the coating additionally comprises gelatin.

43. The method of claim 34, wherein the preparing step comprises a cross-linked polyethylene glycol (PEG).

44. The method of claim 43, wherein the PEG comprises a multiarm PEG.

45. The method of claim 39, wherein the coating of the substrate is covalent.

46. The method of claim 45, wherein the coating is coupled to a thiol moiety or epoxide moiety on the substrate.

47. The method of claim 34, wherein the preparing step comprises coating the substrate with a self-assembled monolayer.

48. The method of claim 34, wherein the labeled molecules comprise one or more antibodies.

49. The method of claim 34, wherein the labeled molecules comprise one or more peptide aptamers.

50. The method of claim 34, wherein the labeled molecules comprise one or more DNA aptamers.

51. The method of claim 34, wherein the labeled molecules comprise one or more RNA aptamers.

52. The method of claim 34, wherein the labeled molecules comprise one or more engineered proteins.

53. The method of claim 52, wherein the engineered protein is fused to a fluorescent protein.

54. The method of claim 34, wherein the labeled molecules are labeled with a fluorophore, quantum dot, or nanoparticle.

55. The method of claim 34, wherein the detecting step produces an image.

56. The method of claim 55, wherein the image is a fluorescence image.

57. The method of claim 56, wherein the image was acquired using Fluorescence Resonance Energy Transfer (FRET), Total Internal Reflection Fluorescence (TIRF), or Zero Mode Waveguide (ZMW).

58. The method of claim 34, wherein preparing step comprises preparing the substrate with aminoethyl or aminopropyl modification.

59. The method of claim 34, wherein the fragmenting step utilizes a protease.

60. The method of claim 59, wherein the protease is proteinase K.

61. The method of claim 34, wherein the fragmenting step utilizes cyanogen bromide.

62. The method of claim 34, wherein the modifying step comprises modifying N- terminal amino acid with phenylisothiocyanate (PTC).

63. The method of claim 34, wherein fragmenting step is omitted.

64. The method of claim 34, wherein the labeled molecules specifically recognizes PTC modified amino acids.

65. The method of claim 34, wherein the labeled molecules generally recognizes groups of PTC modified amino acids.

66. The method of claim 34, wherein the groups of labeled molecules can distinguish a group of amino acids consisting of hydrophobic terminal amino acids, positively charged amino acids, negatively charged amino acids, and small amino acids.

67. The method of claim 34, wherein the removing step comprises removing modified N-terminal amino acid utilizing heat, changes in pH, or both.

68. The method of claim 34, wherein the labeled molecule is removed prior to step (h).

69. A method comprising:

(a) preparing a substrate with a first molecule;

(b) providing a sample comprising one or more analyte proteins;

(c) fragmenting analyte proteins into peptides;

(d) attaching the peptides directly to the substrate or indirectly to the substrate through the first molecule;

(e) incubating substrate with a first group of one or more labeled molecules;

(f) detecting individually resolvable labeled molecules;

(g) repeat steps (e) and (f) from 0 to 100 times with additional groups of labeled molecules.

(h) modifying the N-terminal amino acid of the fragments;

(i) removing the modified N-terminal amino acid; and

(j) repeating steps (e) through (h) two or more times using at least a second group of one or more second molecules. wherein at least a partial amino acid sequence of said peptide is determined.

70. The method of claim 69, wherein the substrate is glass or fused silica.

71. The method of claim 70, wherein the glass or silica is deposited with a metal film.

72. The method of claim 71, wherein the metal film is titanium, gold, or aluminum.

73. The method of claim 71, wherein the metal film is etched to make an array.

74. The method of claim 69, wherein the preparing step comprises coating the substrate with a protein.

75. The method of claim 74, wherein the coating comprises Bovine Serum Albumin (BSA).

76. The method of claim 75, wherein the BSA is acetylated.

77. The method of claim 74, wherein the coating additionally comprises gelatin.

78. The method of claim 69, wherein the preparing step comprises a cross-linked polyethylene glycol (PEG).

79. The method of claim 78, wherein the PEG comprises a multiarm PEG.

80. The method of claim 74, wherein the coating of the substrate is covalent.

81. The method of claim 80, wherein the coating is coupled to a thiol moiety or epoxide moiety on the substrate.

82. The method of claim 69, wherein the preparing step comprises coating the substrate with a self-assembled monolayer.

83. The method of claim 69, wherein the labeled molecules comprise one or more antibodies.

84. The method of claim 69, wherein the labeled molecules comprise one or more peptide aptamers.

85. The method of claim 69, wherein the labeled molecules comprise one or more DNA aptamers.

86. The method of claim 69, wherein the labeled molecules comprise one or more RNA aptamers.

87. The method of claim 69, wherein the labeled molecules comprise one or more engineered proteins.

88. The method of claim 87, wherein the engineered protein is fused to a fluorescent protein.

89. The method of claim 69, wherein the labeled molecules are labeled with a fluorophore, quantum dot, or nanoparticle.

90. The method of claim 69, wherein the detecting step produces an image.

91. The method of claim 90, wherein the image is a fluorescence image.

92. The method of claim 91, wherein the image was acquired using Fluorescence Resonance Energy Transfer (FRET), Total Internal Reflection Fluorescence (TIRF), or Zero Mode Waveguide (ZMW).

93. The method of claim 69, wherein preparing step comprises preparing the substrate with aminoethyl or aminopropyl modification.

94. The method of claim 69, wherein the fragmenting step utilizes a protease.

95. The method of claim 94, wherein the protease is proteinase K.

96. The method of claim 69, wherein the fragmenting step utilizes cyanogen bromide.

97. The method of claim 69, wherein the modifying step comprises modifying N- terminal amino acid with phenylisothiocyanate (PTC).

98. The method of claim 69, wherein fragmenting step is omitted.

99. The method of claim 69, wherein the labeled molecules specifically recognizes PTC modified amino acids.

100. The method of claim 69, wherein the labeled molecules generally recognizes groups of PTC modified amino acids.

101. The method of claim 69, wherein the groups of labeled molecules can distinguish a group of amino acids consisting of hydrophobic terminal amino acids, positively charged amino acids, negatively charged amino acids, and small amino acids.

102. The method of claim 69, wherein the removing step comprises removing modified N-terminal amino acid utilizing heat, changes in pH, or both.

103. The method of claim 69, wherein the labeled molecule is removed prior to step (h).