Recognition of Proteins by Crystallization Patterns in an Array of

The National Center for Biodefense and Infectious Diseases, George Mason University, Manassas, Virginia 20110, and. Institute of Theoretical and Exper...
0 downloads 0 Views 607KB Size
Download PDF
Anal. Chem. 2006, 78, 258-264

Recognition of Proteins by Crystallization Patterns in an Array of Reporter Solution Microdroplets Victor N. Morozov,*,†,‡ Nikolai N. Vsevolodov,† Adam Elliott,† and Charles Bailey†

The National Center for Biodefense and Infectious Diseases, George Mason University, Manassas, Virginia 20110, and Institute of Theoretical and Experimental Biophysics, Russian Academy of Science, Puschino, Moscow region, 142290, Russia

A new technique is described for specific recognition of protein analytes by observing protein-induced changes in the drying/crystallization patterns (DCP) of an array of microdroplets containing solutions of different reporter substances. Recognition is based on a difference in interaction of the protein analyte crystalline elements (planes, edges, defects, etc.) in the growing reporter crystals. Using a set of natural L-amino acids as reporters and denoting the amino acid solutions displaying substantial protein-induced changes in the DCP as “1” and those that show no or small changes as “0”, a digital binary code was determined for several proteins at multiple concentrations. It was demonstrated that globular proteins can be reliably identified using this code as a “signature” when only 2-100 ng of protein was added to amino acid microdroplets. Most bioanalytical procedures are based on the recognition of analytes with natural specific probe molecules: enzymes, antibodies, or complementary DNA. Recognition of the analyte is determined by a limited number of interactions (van der Waals, hydrogen, and ionic bonds), which are spatially organized within a binding site on the probe molecule that complement the receptive groups and atoms on the analyte molecule. One major practical limitation of all such detection techniques resides in the instability of the biological probe molecules, which limits their shelf life and requires special storage conditions. Another known principle of analyte recognition is utilized by physical methods where spectral characteristics (NMR, IR, MSMS, and others) are used to distinguish between molecules, provided a database of spectra for all possible analytes is available. The use of physical methods is, however, limited by their complexity, which increases the cost per assay due to the need for expensive equipment and trained personnel. Is it possible to recognize complex biological substances without using fragile biological probe molecules or complex physical instruments? One such alternative is provided by chromatography, in which tiny differences in affinity of analytes to a specific resin result in their separation. A set of numbers characterizing the mobility of an analyte in a series of columns with different packing material could potentially serve as a * Corresponding author. Tel: 703-993-4294. E-mail: [email protected]. † George Mason University. ‡ Russian Academy of Science.

258 Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

“signature” of this analyte. It is, however, technically difficult to repeatably determine such parallel chromatographic measurements in the multiple columns necessary for recognition. Small differences in the binding of analytes to different surfaces can be easily detected in an array of organic and inorganic crystals. Such arrays provide a rich variety of surfaces for analyte adsorption since each crystal exposes a unique mosaic of chemical groups on its planes, edges, and defects. Moreover, crystals themselves may serve as detectors of analyte binding. It is well documented that crystal growth is highly sensitive to the presence of impurities, which may inhibit their growth by changing the probability of seed formation or may modify crystal morphology.1-9 Since the end result of such modifications is easily detectable by optical imaging, an array of different reporter substances could be tested for sensitivity to a specific analyte and the changes detected, expressed in a certain “code”, could be used as the analyte “signature”. It should be mentioned that a number of proteins have evolved to modify organic and inorganic crystals. Thus, several unrelated proteins have been found in arctic fishes that are capable of binding to certain planes in ice crystals and inhibiting ice formation in biological fluids.10,11 Many crystallization inhibitors have also been described in animals and in humans: nephrocalcin in urine, gallstone protein in bile, and lithostathine in pancreatic juice.12 It was shown that lithostathine inhibits nucleation and modifies morphology of calcite by preferable binding to a certain edge in (1) Berkovitch-Yellin, Z.; Mil, J. V.; Addadi, L.; Idelson, M.; Lahav, M.; Leiserowitz, L. J. Am. Chem. Soc. 1985, 107, 3111-3122. (2) Weiner, S.; Addadi, L. J. Mater. Chem. 1997, 7, 689-702. (3) Clark, R. H.; Campbell, A. A.; Klumb, L. A.; Long, C. J.; Stayton, P. S. Calcif. Tissue Int. 1999 64, 516-521. (4) Cooper, S. J. Cryst. Eng. Commun. 2001, 56, 1-4. (5) Ward, M. D. Chem. Rev. 2001, 101, 1697-1725. (6) Beckera, A.; Beckerb, W.; Marxenb, J. C.; Epplea, M. Z. Anorg. Allg. Chem. 2003, 629, 2305-2311. (7) Jung, T.; Sheng, X.; Choi, C. K.; Kim, W. S.; Wesson, J. A.; Ward, M. D. Langmuir 2004, 20, 8587-8596. (8) Sheng, X.; Jung, T.; Wesson, J. A.; Ward, M. D. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 267-272. (9) Cashell, C.; Corcoran, D.; Hodnett, B. K.; Cryst. Growth Des. 2005, 5, 593597. (10) Houston, M. E.; Chao, H.; Hodges, R. S.; Sykes, B. D.; Kay, C. M.; Frank, D.; Sonnichsen, F. D.; Loewen, M. C.; Davies, P. L. J. Biol. Chem. 1998, 273, 11714-11718. (11) Baardsneses, J.; Jelokhani-Niaraki, M.; Kondejewski, L. H.; Kuiper, M. J.; Kayt C. M.; Hodges, R. S.; Daviesi, P. L. Protein Sci. 2001, 10, 2566-2576. (12) Geider, S.; Baronnet, A.; Cerini, C.; Nitsche, S.; Astier, J. P.; Michel, R.; Boistelle, R.; Berlandi, Y.; Dagorn, J. C.; Verdier, J. M. J. Biol. Chem. 1996, 271, 26302-26306. 10.1021/ac0512857 CCC: $33.50

© 2006 American Chemical Society Published on Web 11/17/2005

the calcite crystal.12 Specialized proteins capable of establishing interactions with certain crystal faces in carbonated apatite and calcite minerals control formation of shells, sea urchin needles, and dental tissue (see excellent review2). Selective adsorption to different surfaces is not limited to specialized proteins: serum albumin and fibrinogen, for example, display quite different preferences for hydrophilic and hydrophobic surfaces.13 To demonstrate how the principle of crystallization pattern modification works, we used a set of common proteins whose biological function has nothing to do with crystal growth modification. We demonstrate in this paper that a simple binary code accounting for the presence or absence of changes in reporter drop crystallization patterns allows one to reliably distinguish between different protein contaminants. In this study natural L-amino acids were tested as reporter substances. It was demonstrated that each protein has a concentration-dependent, unique signature that characterizes the effect of the protein on the drying/ crystallization pattern (DCP) of a set of amino acid solutions. EXPERIMENTAL SECTION Reagents. A set of L-amino acids was obtained from Sigma (kit, LAA-21; minimum 98-99% purity) and used without purification. Chicken egg albumin (Ova), bovine hemoglobin (Hb), hen egg white lysozyme (HEWL), concanavalin A (Con A), bovine serum albumin (BSA), and human apo-transferrin (apo-F) were also obtained from Sigma (St. Louis, MO). Preparation of Proteins. All proteins were thoroughly dialyzed against ultrapure water until the solution conductivity reached a plateau. The conductivity of the dialyzed protein solutions was measured in a 4-µL cell manufactured in the laboratory. Protein concentration in the dialyzed samples was measured with a precision of 15-20% on a homemade quartz crystal microbalance14 (QCM). The microbalance was calibrated by drying a sucrose solution on the quartz resonator. Protein concentrations were measured by drying 0.2-0.4 µL of the dialyzed sample on the QCM. Preparation of Substrates. Microscope slides were cleaned in cold rf plasma at air pressure (0.3-0.7 Torr) at 30 W for 20 s and placed into a vapor of dimethyldichlorosilane (Aldrich) in dry nitrogen for 7-10 min. The hydrophobic layer on the glass surface was then stabilized by baking in an oven at 100-120 °C for 1-2 h. The glass was then covered with a mask manufactured from a perforated aluminum plate (Small Parts) containing an array of round holes, 1.5 mm in diameter, with a distance between the holes of 3 mm. The polished side of the aluminum mask faced the glass surface. The mask was pressed against the slide with binder clips, as shown in the upper part of Figure 1. After treatment in the plasma discharge for 5-10 s, the hydrophobic coating was completely removed from 150 round areas exposed to the holes, while the rest of the slide surface remained hydrophobic. Preparation of Reporter Solutions. Amino acids were dissolved in ultrapure water, and the pH of the solutions was adjusted to 6.8-7.1 with NaOH or HCl solutions. Deposition of Microdroplets. In a typical experiment, a 10 mg/mL amino acid solution was mixed with an equal volume of (13) Roach, P.; Farrar, D.; Perry, C. C. J. Am. Chem. Soc. 2005, 127, 81688173. (14) Morozov, V. N.; Morozova, T. Ya. Anal. Chem. 1999, 71, 1415-1420.

Figure 1. Array of droplets obtained on preactivated patterned glass surface. Upper part: aluminum mask is pressed against a hydrophobic glass surface upon plasma treatment. Middle part: array of amino acid solution droplets on the patterned surface. Lower part: array of amino acid solution DCPs.

Figure 2. Schematic of a semiautomatic applicator for rapid manufacturing of droplet arrays. Key: 1, base; 2, stationary base of the linear motion ball bearing slide; 3, top slide; 4, capillary holder; 5, reed; 6, ratchet plate; 7, tooth; 8, patterned microscope slide; 9, solution capillaries.

dialyzed protein solution. Water was substituted for the protein solution in control. Microdroplets of both solutions were deposited onto hydrophilic spots on the patterned microscopic slide with either a Hamilton microsyringe or a capillary array. In the latter case, a semiautomatic device schematically presented in Figure 2 was used. The patterned slide was fixed onto a linear motion ball bearing slide (Edmund Scientific) and a row of eight capillaries, 1.57 mm o.d. and 40 mm long, was attached to a flexible reed so that slight tapping of the reed caused the capillary tips to touch the slide within the hydrophilic spots and deposit small droplets, ∼0.4 µL in volume (Figure 2). Capillaries were filled manually by placing 20 µL of solution from a pipet tip into each capillary. Each slide included an array of control droplets where water was used in place of the protein solution. The device was placed into a jacket to protect it from dust and air flow. A motor (not shown in Figure 2) steadily moved the top slide and induced oscillatory vibrations in the reed via contact of the tooth (7) with the ratchet (6). Each time the capillary end touched the slide within the hydrophilic area, a microdroplet was released and filled the entire hydrophilic spot. This technique allowed deposition of 18 droplets of 8 different mixtures onto one microscopic slide. The deposition itself took 5 s and produced the array of microdroplets illustrated in Figure 1. No drying of microdroplets was observed during the deposition. Drying Droplets. The microdroplet slide was quickly transferred into a dust-free closed chamber where the droplets were Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

259

dried at 25-30% relative humidity and a temperature 23 °C with no forced ventilation applied. Dynamics of droplet drying were observed by attaching another drying chamber to the microscope platform of the Axiovert 25 Zeiss microscope. The time of drying varied between 3 and 10 min for different amino acid solutions depending on the water activity in the solutions. Imaging Dry Residues. The Axiovert 25 Zeiss microscope was equipped with the Penguin 600CL digital camera connected to a Dell Precision 450 Station. Droplets were imaged in open air at a 50× magnification. It was found that among different modes of optical microscopy tested (transmission, dark-field, polarization, and phase contrast microscopy) phase contrast microscopy yielded images displaying the largest number of distinguishable features. This can be explained by the sensitivity of phase contrast microscopy to the thickness distribution in the transparent DCP. We used phase contrast optical imaging throughout this study. No notable pattern changes were observed upon storage of the amino acid DCP for one year in an air-conditioned room (humidity of ∼30%, temperature 23 °C). RESULTS AND DISCUSSION Choice of Amino Acids and Reproducibility of Crystallization Patterns of Dried Amino Acids. Several obvious criteria have been used in our choice of reporters. A suitable reporter substance should be (i) nontoxic so that it can be used with only minor precautions, (ii) chemically inert with respect to analytes on the time scale of the assay (3-10 min), (iii) inexpensive, (iv) rapidly crystallizable upon drying, (v) crystallizable in multiple forms, and (vi) optically active in order to recognize stereoisomeric analytes such as proteins. Amino acids satisfy all the abovementioned conditions and represent an obvious first choice of reporter substances. All available amino acids were tested in initial experiments in order to find an optimal amino acid solution concentration. DCPs acquired from a series of solutions with amino acid concentrations ranging from 0.1 to 20 mg/mL were evaluated for reproducibility and for the number of distinguishable DCP features. Based on these criteria, a subset of amino acids (Cys, Asp, Glu, Tyr, Arg, His) were discarded from further testing because they failed to produce DCPs. From the remaining amino acid solutions, it was found that concentrations less than 1 mg/mL produced relatively featureless images, probably because the solubility limit is reached later upon drying and less material is available for crystal formation. Solutions with concentrations of 5-10 mg/mL were found to be optimal. On average, 15-18 images were obtained for each amino acid control and each protein/amino acid mixture to determine typical patterns and their variabilities. Of these images, 70-80% displayed identical DCPs, while 10-20% of the images at the beginning and at the end of each series displayed slightly different DCPs, presumably due to variability in the amount of solution deposited onto these spots. This conclusion is supported by an analysis of DCPs obtained from microdroplets deposited by the Hamilton microsyringe: all DCPs were absolutely identical. Although we limited our choice of reporters to amino acids, other substances satisfying the requirements listed above could also be employed. For example, we successfully tested Gly-Gly and Bicin. Buffer components (e.g., TRIS, HEPES, MES, etc.), 260

Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

salts (carbonates, phosphates, etc.) metabolites (glucose, glucose 6-phosphate, etc.), and other inert chemical compounds could also be used. In extending the list of possible reporters, however, one has to account for yet another condition: the crystal lattice of the potential reporter should not be excessively strong (as in the case of ionic crystals) so that the reporter molecules could not displace the analyte from the growing crystal. On the other hand, reporter crystals with low lattice energy (like protein crystals) should also be avoided since their crystallization proceeds slowly. In principle, reporter molecules could be rationally selected for a certain set of analytes by theoretical modeling of interactions of these analytes with the surface elements of the reporter crystals.4,15 Drying/Crystallization Pattern. Formation of a DCP is a complex phenomenon to which both macroscopic and microscopic events in the drying droplet contribute. One DCP feature that often (but not always) changes in response to protein addition is the ringlike deposit along the droplet perimeter seen in Figures 3 and 4. This ringlike deposition results from enhanced evaporation of solvent near the highly convex edge.16 If reporter crystallization happens at an early stage of drying and the crystals are formed in the bulk solution, they are transported by the capillary flow to the droplet edge and concentrated there, forming a ring. Protein added to the droplet can affect this process in several ways: (i) by inhibiting formation of crystals until higher levels of supersaturation are achieved at later stage of drying, at which time the solution layer is too thin to transport crystals; (ii) by inhibiting evaporation from edges due to accumulation of the protein at the periphery; (iii) by increasing solution viscosity; (iv) by blocking the crystallization centers on the glass support; (v) by more effective immobilization of newly formed crystals on the glass substrate. The first three factors are expected to make DCP more uniform, while other two would result in formation of ringlike deposits. One of the significant advantages of using patterned arrays is that all droplets are contained in areas of the same size, thus eliminating the effect of droplet size on the crystallization pattern. In the absence of a patterned array surface, the final size of the DCP would depend on the solution surface activity. Taking into account that all proteins are capable of concentrating at the waterair interface, thus changing the solution surface tension, one would expect such nonspecific changes to occur in DCPs for most protein additives. In addition to the radial distribution of the reporter, proteins affect other DCP features. Thus, Hb promotes formation of large Leu crystals that concentrate on the droplet periphery (Figure 3.). While Ova also promotes formation of Leu crystals, their irregular morphology drastically differs from the elongated crystals radially growing from the edges in the presence of Hb. In contrast to the positive effect of Ova on Leu crystallization, the effect of Ova on Lys crystallization is inhibitory: the rotation of polarized light in the Lys/Ova DCP is dramatically reduced. Increasing the solute concentration upon drying initiates and supports all kinds of processes leading to association of the solutes: aggregation, gelatination, and crystallization. The con(15) Addadi, L.; Geva, M. Cryst. Eng. Commun. 2003, 5, 140-146. (16) Deegan, R. D.; Bakajin, O.; Dupont, T. F.; Huber, G.; Nagel, S. R.; Witten, T. A. Nature 1997, 389, 827-829.

Figure 3. Examples of changes in the DCP of amino acid solutions resulting from addition of proteins at different concentrations. Each 0.4-µL droplet contains a 1/1 (v/v) mixture of the 10 mg/mL amino acid solution and water (control) or protein solution. From left to right: water (control), protein added in concentration of 0.0003, 0.003, 0.03, 0.3, and 3.0 mg/mL. Droplets were deposited with the semiautomatic applicator shown in Figure 2. Droplets dried within 3-5 min. Diameter of each DCP is 1.5 ( 0.02 mm.

Figure 4. Example of binary coding of a protein using DCP in an array of droplets of reporter amino acid solutions. Each 0.4-µL droplet contained amino acids at a concentration of 5 mg/mL, and 100 pg of Ova was added to each amino acid solution. Letter in the lower corner of each image denotes one-letter abbreviation for amino acid: L, leucine; M, methionine; K, lysine; Q, glutamine; P, proline (trans-4-hydroxy) analogue; and F, phenylalanine.

centration and identity of the components, solution viscosity, ionic force, presence of surfactants, and presence of gel- and crystalforming molecules can all potentially affect the DCP. Several authors have pointed to potential analytical applications of these processes by demonstrating that drying of the same multi- or monocomponent liquid under identical conditions resulted in reproducible morphological patterns, while the slightest variation in the composition of a liquid changed the patterns.16,17 Notable changes in the dry patterns of saliva and blood plasma have been documented in sick patients.18,19 It has been suggested that acustomechanical impedance measurements could be used to monitor dynamic processes in a drop drying on a quartz crystal (17) Deegan, R. D. Phys. Rev. E 2000, 61, 475-485. (18) Kharchenko, S. V.; Korneeva G. A.; Vetrov, A. A. Izv. AN USSR, Biol. Issue 1988, 3, 450-454. (19) Rapis, E. G. Zh. Tekh. Fiz. 2002, 72, 139-142.

resonator.20 This method has been tested using plasma, urine, and saliva from sick and healthy people. We must emphasize that our approach is substantially different from those described above: instead of relying on specific changes in the mechanical properties of the drying drop or specific features in the crystal pattern as analyte signatures, we determine signatures by observing effects of the analyte on the DCP in multiple reporter solutions. Protein Binary Code from Amino Acid DCPs. The series of images in Figure 3 illustrate how proteins affect the DCP of amino acid solutions. When compared with control images in which water was added instead of protein solution, one can clearly see that the affect of a protein on the DCP is concentration (20) Yakhno, T. A.; Yakhno, V. G.; Sanin, A. G.; Shemelev, I. I. Biophysics 2002, 47, 1101-1105.

Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

261

Figure 5. Reproducibility of DCPs in a series of droplet controls (upper) and in droplets to which 100 pg of HEWL (middle) and BSA (bottom) was added. Other conditions as in the caption to Figures 3 and 4.

dependent. Although the DCPs change with increasing protein concentration, they remain different from control DCPs. Figure 4 illustrates the differences in DCPs of several amino acids in response to the addition of 100 pg of Ova. It can be seen that Ova changes the DCP of Leu, Met, and Lys, while the DCP of three other amino acids (Gln, Pro, Phe) are only minimally affected and mimic the variations seen in control DCPs. Though one could create a database containing all of the observed DCP modifications caused by a series of proteins for every reporter substance, the resulting database would be unwieldy in size and would require that all control and test samples be prepared under standard drying conditions. A simpler approach involves determining only whether the DCP changes due to the presence of a particular protein while ignoring the specific details of the change. Such changes could be easily recognized despite slight variabilities in DCPs. Figure 5 presents an illustration of the typical variability observed in control DCPs and in the DCPs of reporter/protein solutions. Despite minor variability in DCPs of individual droplets, one can unequivocally conclude that DCPs of Lys/HEWL are almost identical to those of controls, while DCPs of Lys/BSA are notably different from the control DCPs. Assigning the amino acids displaying no or small changes a value of “0” and those that show drastic unambiguous changes in the DCP a value of “1” yields a “binary” code for each of the proteins tested at equal (w/v) concentrations. Table 1 summarizes our results analyzing the effects of 5 proteins on DCPs of 12 different amino acids in terms of the binary code. It is seen that addition of 100 pg of protein results in substantial DCP changes in 3-9 reporters. BSA, known for its ability to bind to many ligands in serum, affected the DCP of 9 out of the 12 amino acids studied. Are Binary Codes Additive? It was natural to expect that the binary codes for two proteins added to an array of reporter solutions might be additive, i.e., the protein mixture should affect the DCP if at least one component alone has an effect (1+ 0 ) 0 + 1 ) 1), and no changes in the DCP should occur if neither component affects the DCP by itself (0 + 0 ) 0). As shown in 262 Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

Table 1. Presentation of the Data Describing the Effect of Proteins on Crystallization Patterns of Amino Acids as a “Binary Code”a

a Proteins that strongly affect the crystallization pattern of an amino acid are assigned a value of 1, while those showing negligible effects are assigned a value of 0. Final amino acid concentration was 5 mg/ mL. b A 100-pg sample of protein was added to a 0.4-µL droplet of 5 mg/mL amino acid. In the case of protein mixtures, 50 pg of each protein was added to the droplet. c P denotes proline (trans-4-hydroxy) analogue.

Table 1, the protein binary codes are additive and this opens the possibility of applying the new recognition approach to mixtures of analytes with known codes, provided the code is long enough. Concentration Dependence of Binary Codes. The data presented in Table 1 describe the effects of proteins present in similar (w/v) concentrations. How does the protein concentration affect the code? We addressed this question by varying the concentration of several proteins and adding them to amino acid reporters kept at a constant concentration. One can see from the results presented in Table 2 that the protein codes change as expected: at very low protein concentration, the code degenerates into a string of zeros, while at high concentration, proteins affect the DCP of each amino acid, resulting in a code of all ones. In a concentration range of 0.010.3 mg/mL, the binary codes for each protein studied are unique

Table 2. Binary Codes for Different Concentrations of Four Proteinsa

Table 3. Matrix Code for Ova, Accounting for Several DCP Features Affected or Unaffected by the Presence of This Analytea

a The 0.1 mg/mL Ova was added in equal volume to a 10 mg/mL microdroplet of each amino acid. b Binary units in the matrix code have the same meaning as in the binary code: 1 indicates changes in the feature, 0 indicates a lack of change or an insignificant change. P denotes proline (trans-4-hydroxy) analogue.

Table 4. Blind Recognition of Unknown Substances by Binary Codesa a The concentration of protein solution added in equal volume to each amino acid solution is indicated. The concentration of all amino acids was 10 mg/mL. P denotes proline (trans-4-hydroxy) analogue.

and could be used to identify these proteins. When dealing with a protein solution of unknown concentration, one needs, thus, to either determine protein concentration independently or test a series of protein dilutions and compare the codes generated at different concentrations with a database. Redundancy of Codes, Different Codes. Theoretically, the binary code resulting from 12 reporters provides 212 ) 4096 different combinations of binary units. We do not know at present how redundant the codes are, i.e., if certain reporters are intrinsically more sensitive to impurities than others (e.g., Lys is affected by most proteins studied, as seen from Table 2). More extended screening is required. However, redundancy seems to not pose a serious theoretical problem since the number of independent reporters could be easily increased to ∼100, yielding ∼1030 unique binary codes. It is worth noting that the total variety of possible antibodies (estimated from all gene combinations) is 17 billion, a number that corresponds to a combinatorial capacity of a 35 reporter binary code. The Cambridge structural database now contains 335 276 structures of organic crystals (http:// www.ccdc.cam.ac.uk/products/) from which one could choose reporter molecules. Though many of these compounds are insoluble in water, reactive to protein molecules, or otherwise unsuitable for use as the reporters, still a considerable fraction of the compounds will satisfy the reporter conditions. “Matrix Code”. The potential redundancy of binary codes could also be overcome by switching to codes with higher dimensions. As seen in Figures 3 and 4, reporters display changes in different DCP features in response to certain analytes (e.g., radial distribution of crystals in the DCP, crystal morphology, the presence or absence of a ring at the droplet edge, etc.,). Ascribing values of 0 and 1 to each such feature within a reporter’s DCP produces a multidimensional code for each analyte that is described by a matrix. An example of such a matrix code for Ova is presented in Table 3. In comparison to its binary code, the

a Protein, with the concentration indicated in the second column, was mixed with an equal volume of a 10 mg/mL amino acid solution. P denotes proline (trans-4-hydroxy) analogue.

matrix descriptor gives notably more information and is predicted to allow discrimination between analytes even when their binary codes are identical. One can see in Table 3 that the binary 1 in the Ova code may originate from quite different DCP features, from their combinations, or from both. Recognition of Proteins in a Blind Experiment. Four protein samples were prepared for blind testing, three of which belonged to the set of proteins with known codes and one of which had not been previously analyzed. The protein samples were subjected to the standard procedure of sample preparation including extensive dialysis against water, determination of the solution conductivity, and determination of the protein concentration using QCM. Protein solutions were then added to an array of amino acid microdroplets and DCPs of the solutions, controls were compared, and their binary codes were determined. Table 4 presents a comparison of the codes determined by this experiment with those found in our database. It can be seen that proteins with known codes had a 92% identity to their database code; i.e., the codes obtained for blindly tested proteins differed from those in our database by 1 digit (different binary units are marked by italics in Table 4) out of 12. The binary code of the unknown protein (human apotransferrin) was different from those in our database at a minimum of four digits. The blind experiment demonstrates that pure proteins can be reliably identified with the present method provided their solution concentration is known. The experiment also demonstrates that recognition is possible even if certain instabilities in DCPs result in variability of codes. It is expected that increasing the number Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

263

of reporters and the code length will increase the identity percentage and thus lead to greater accuracy in identification CONCLUDING REMARKS We do not claim that the method presented here will be capable of replacing existing techniques soon, since there are still many questions to be addressed before one can think of designing a practical assay based on DCPs of arrayed reporter droplets. Will this approach be applicable to more practical situations where proteins are mixed with other biological macromolecules? How sensitive are protein codes to protein folding status, to the presence of metabolites, salts, and other low molecular weight substances? Part of these problems may be solved by applying special preparation procedures. For example, removal of salts and metabolites by dialysis and adjustment of pH will reduce interference from these components.

264

Analytical Chemistry, Vol. 78, No. 1, January 1, 2006

Nevertheless, the approach provides a new avenue in the development of methods for identification of chemical compounds and for their characterization. It is expected, for example, that proteins characterized by similar codes will have similar structures since they preferably bind the same set of surface elements in the reporter crystals. ACKNOWLEDGMENT The authors gratefully acknowledge support from DOE Grant DE-F C52-04NA25455. We sincerely thank Dr. Tim Born (George Mason University) for the critical reading of the manuscript and the valuable comments. Received for review July 20, 2005. Accepted October 25, 2005. AC0512857