General In Vitro Method to Analyze the Interactions of Synthetic

Article pubs.acs.org/Biomac

General In Vitro Method to Analyze the Interactions of Synthetic Polymers with Human Antibody Repertoires Anandakumar Soshee,†,‡ Stefan Zürcher,§ Nicholas D. Spencer,§ Avraham Halperin,‡ and Clément Nizak*,‡,∥ ‡

Laboratory of Interdisciplinary Physics, UMR5588 Grenoble Université 1/CNRS, Grenoble, France Laboratory of Biochemistry, UMR 7084 ESPCI ParisTech/CNRS, Paris, France § Laboratory for Surface Science and Technology, Department of Materials, ETH Zurich, Wolfgang-Pauli-Strasse 10, 8093 Zürich, Switzerland ∥

S Supporting Information *

ABSTRACT: Recent reports on the hitherto underestimated antigenicity of poly(ethylene glycol) (PEG), which is widely used for pharmaceutical applications, highlight the need for efficient testing of polymer antigenicity and for a better understanding of its molecular origins. With this goal in mind, we have used the phage-display technique to screen large, recombinant antibody repertoires of human origin in vitro for antibodies that bind poly(vinylpyrrolidone) (PVP). PVP is a neutral synthetic polymer of industrial and clinical interest that is also a well-known model antigen in animal studies, thus allowing the comparison of in vitro and in vivo responses. We have identified 44 distinct antibodies that bind specifically to PVP. Competitive binding assays show that the PVPantibody binding constant is proportional to the polymerization degree of PVP and that specific binding is detected down to the vinylpyrrolidone (VP) monomer level. Statistical analysis of anti-PVP antibody sequences identifies an amino-acid motif that is shared by many phage-display-selected anti-PVP antibodies that are similar to a previously described natural anti-PVP antibody. This suggests a role for this motif in specific antibody/PVP interactions. Interestingly, sequence analysis also suggests that only a single antibody chain containing this shared motif is responsible for antibody binding to PVP, as confirmed upon systematic deletion of either antibody chain for 90% of selected anti-PVP antibodies. Overall, a large number of antibodies in the human repertoires we have screened bind specifically to PVP through a small number of shared amino acid motifs, and preliminary comparison points to significant correlations between the sequences of phage-display-selected anti-PVP antibodies and their natural counterparts isolated from immunized mice in previous studies. This study pioneers the use of antibody phage-display to explore the antigenicity of biotechnologically relevant polymers. It also paves the way for a fast, cost-effective, and systematic in vitro analysis, thus reducing the need for animal immunization experiments. Moreover, identifying the encoding DNA sequence of polymer-binding antibodies via phage-display enables future applications of a molecular biology approach to protein−polymer conjugation, based on protein−antibody fusion.

■

INTRODUCTION

Here we propose a general, quantitative, fast, and costeffective in vitro method to analyze the interactions of a watersoluble polymer of interest with a large human-antibody repertoire, avoiding recourse to animal immunization. Our strategy is based on antibody phage-display. This is a wellestablished in vitro technique to screen large recombinantantibody libraries of human origin mimicking human immune repertoires in order to generate antibodies to biological molecular targets for fundamental, diagnostic, and therapeutic applications.10 This technique relies on standard molecular biology and biochemistry reagents and equipment, avoiding costly animal experiments and allowing large-scale studies. Phage-display screens typically consist of three rounds of in

1−4

With the exception of poly(vinylpyrrolidone) (PVP), the interactions of synthetic, neutral, water-soluble polymers with antibodies have received little attention. Yet, these interactions are of importance in pharmacological applications,5 such as PEGylation,6 in which neutral polymers are used to prolong the circulation times of protein drugs by slowing renal clearance and screening interactions with blood proteins. Poly(ethylene glycol) (PEG) is the most frequently used polymer for such applications, given its typically repulsive interactions with proteins and reputation for weak antigenicity.6,7 However, several alarming reports have recently described accelerated blood clearance of PEGylated drugs due to hitherto underestimated PEG antigenicity.5,8,9 These reports highlight the need for an efficient assessment method for polymer antigenicity, as well as for a better understanding of its molecular origins. © 2013 American Chemical Society

Received: September 11, 2013 Revised: November 25, 2013 Published: December 3, 2013 113

dx.doi.org/10.1021/bm401360y | Biomacromolecules 2014, 15, 113−121

Biomacromolecules

Article

com).26−28 In brief, the plates were first oxygen-plasma treated for 2 min, incubated for 30 min in a 0.1 mg/mL solution of AziGrip4, and washed three times with ultrapure water. A film of PVP was cast in each well from a 25 mg/mL PVP solution in ethanol. After air-drying overnight, the wells were UV−C illuminated for 2 min. Excess nonbound PVP was then removed by immersion of the plates in ultrapure water overnight and thorough rinsing under a stream of ultrapure water. This procedure led to a dense film of covalently bound PVP of approximately 10−20 nm in thickness. Selection of Anti-PVP Recombinant Antibodies by Antibody Phage-Display. Phage-display screens for both Griffin and Tomlinson libraries were performed essentially according to the standard protocol from Source BioScience (Cambridge, U.K.; http://lifesciences. sourcebioscience.com/media/143421/tomlinsonij.pdf) and our own previous work.29 In brief, 1012 to 1013 phages were prepared by superinfection with helper M13 KO7 phage (GE Healthcare, Pittsburgh, PA) of exponentially growing TG1 E. coli (Source Bioscience, Cambridge, U.K.) in 2xTY medium supplemented with ampicillin (100 μg/mL) and glucose (1% w/v). The infected TG1 culture was grown overnight at 30 °C in 2xTY + ampicillin + kanamycine (50 μg/mL). The culture was centrifuged for 10 min at 10000g to remove bacteria. Phages in the culture medium were precipitated by the addition of 1/4 volume of a solution of PEG of average MW 8000 (30% w/v)/NaCl (1.5 M) at 4 °C, and resuspension in 1 mL of phosphate-buffered saline solution (PBS). To eliminate polystyrene-binding clones from the libraries, negative selection was performed by incubating the amplified phages suspended in PBS + Tween20 (0.1% w/v) + milk (2% w/v Marvel fat free milk, Premier Foods, St Albans, U.K.) in nontreated polystyrene 3 cm diameter Nunc Petri dishes (Thermo Fisher Scientific, Waltham, MA) for 1 h at room temperature. Both preadsorbed libraries were then incubated on a rocker, allowing the phage solution to be shaken evenly for 1 h at room temperature in two separate PVP-coated Petri dishes, which had been blocked prior to selection with PBS + milk (2% w/v) for 1 h at room temperature. Next, the phage solution was removed from both Petri dishes. Petri dishes were subsequently washed 10 times for 5 s with PBS + Tween20 (0.1% w/v; first round) and 20 times (subsequent rounds) to remove nonspecific binders. Bound phages were eluted using 1 mL of fresh solution of 100 mM triethylamine for 20 min and neutralized with 500 μL of Tris/HCl buffer (1 M, pH 7.2). Eluted phages were recovered by infection of an excess of exponentially growing TG1 E. coli cells (14 mL of a 2xTY culture at O.D.600 nm = 0.5) for titration and phage preparation for subsequent rounds of selection. Expression of Selected Recombinant Antibodies. A total of 94 selected antibody clones from both libraries were picked randomly at the third and fourth selection rounds and arrayed into 2 mL 96-well plates. Soluble antibody production was initiated by inoculating 10 μL of an overnight saturated culture (2xTY + ampicillin +1% w/v glucose at 37 °C) into 1 mL of 2xTY + ampicillin +0.1% w/v glucose at 37 °C. Induction of antibody expression (under the control of the Lac promoter in the Griffin and Tomlinson library vectors) was performed by adding 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) to the culture reaching O.D.600 nm = 0.5, followed by further incubation for 18 h at 30 °C. Soluble antibodies (possessing both a VH chain and a VL chain, or either one of them) were secreted into the culture medium at a yield of 1−10 mg/L. The culture supernatant was then directly used as a primary antibody for the ELISA tests. The size, solubility, and yield of expression of antibodies were checked by dot blot (with an anti-His6 tag antibody, since the Griffin and Tomlinson library vectors harbor a C-terminal His6 tag sequence downstream of the antibody sequence) and SDS-PAGE analysis of purified antibodies (Supporting Information, Figure 5). For this purpose, the expression conditions were scaled up to obtain a culture of 500 mL. Antibodies were purified on cobalt-ion resin (CL-Cobalt Chromatrix from Jena Bioscience distributed by Euromedex, Strasbourg, France) through their His6 tag. Briefly, a 500 mL culture was centrifuged at 10000g for 10 min and the pellet containing bacteria and debris was discarded. The culture medium was then passed 3 times at 4 °C (the procedure is done in a cold room from then on until the end) onto a column containing 1

vitro selection performed in a week, while animal immunization protocols can take several months. Moreover, phage display has been successfully applied to the generation of peptides11−13 and antibodies14−16 that bind specifically to nonbiological targets (metals, metal ions, semiconductors, magnetic metal oxides, conductive polymers, and carbon nanotubes). This enables the analysis of peptide or protein interactions with nonbiological surfaces and molecules17,18 as well as applications in materials science.19,20 We have tested whether antibody phage-display is applicable to a whole new class of nonbiological targets: watersoluble synthetic polymers. We have used widely distributed antibody repertoires and protocols (Tomlinson and Griffin repertoires21,22) to ensure that our approach is both easily accessible and reproducible. As a model polymer target, we have chosen poly(vinylpyrrolidone) (PVP), which is a synthetic, neutral, watersoluble polymer of pharmaceutical/industrial interest. For example, PVP is widely used as an excipient under the trademark Povidone. PVP is also a well-characterized model antigen in immunological studies in animals,1,3,4,23 which allows the comparison between our in vitro results and available in vivo data. PVP behaves in vivo as a so-called thymusindependent type 2 antigen. PVP-immunized mice essentially produce IgM and very little IgG, even upon a second injection, and the VH-CDR3 hypervariable regions of 16 distinct PVPreactive antibodies have been sequenced.4 These available sequences of natural anti-PVP antibodies provide information about amino acids that are present in the antibody hypervariable loops involved in antibody/PVP interactions in vivo. PVP is thus by far the best-characterized antigen among synthetic, neutral, water-soluble polymers and therefore a suitable candidate to test our strategy. Apart from information on the antigenicity of polymers, the phage-display technique affords the prospect of a molecularbiology approach to protein−polymer conjugation as an alternative to the chemical-synthesis methods that are traditionally utilized.24 This new approach would rely on the fusion of the DNA sequence of phage-display-selected, polymerbinding antibodies to the DNA sequence of a protein. As we shall detail in the discussion, this introduces the possibility of polymeric material design based on antibodies. A similar strategy has already been utilized for biomolecule immobilization on gold surfaces.16,25 The implementation of this approach to protein−polymer conjugation remains to be explored.

■

EXPERIMENTAL SECTION

Materials and Methods. All chemicals were purchased from Sigma-Aldrich (St Louis, MO) unless otherwise specified. Deionized water of resistivity 18 MΩ.cm was produced with a Milli-Q system (Merck Millipore, Billerica, MA). 2xTY medium was prepared by dissolving 16 g tryptone, 10 g yeast extract, 5 g NaCl (tryptone and yeast extract from USBIO distributed by Euromedex, Strasbourg, France) in 1 L of deionized H2O and autoclaving for 15 min at 120 °C. Competitors used in competitive ELISA experiments were poly(vinylpyrrolidone) of average MW 10000, 40000, 360000, or 1300000, 1-vinyl-2-pyrrolidone, 1-methyl-2-pyrrolidone, N-methyl-2-piperidone, ethylvinylether, poly(2-ethyl-2-oxazoline) of average MW 50000, and poly(acrylamide) of average MW 5000000. PVP-Functionalized Polystyrene 96-Well Plates. High-molecularweight PVP (average MW 1300000) was immobilized onto nontreated polystyrene Greiner 96-well plates (Thermo Fisher Scientific, Waltham, MA) and nontreated polystyrene 3 cm diameter Nunc Petri dishes (Thermo Fisher Scientific, Waltham, MA) using deep UV photobinding by means of AziGrip4, a surface-binding, nitrenegenerating polymer (SuSoS AG, Dübendorf, Switzerland, www.susos. 114


Biomacromolecules

Article

Figure 1. Selection of anti-PVP antibodies by phage display. (a) The fragment V of antibodies comprises the variable regions of heavy and light chains (respectively VH and VL), the association of which is generally thought to shape the antigen-binding pocket. VH and VL each possess three complementary determining regions (CDRs) corresponding to hypervariable sequences (stretches of 5−10 amino-acids, typically 30% of the fragment V sequence overall) that determine binding specificity. Recombinant antibodies of the single-chain fragment V (scFv) format result from the linking of VH and VL via a flexible glycine/serine linker. scFvs recapitulate the binding specificity/chemistry of their natural-antibody counterparts. (b) Fusion of a scFv-encoding DNA sequence to that of a phage surface protein results in the display of this scFv on the phage surface, which provides a physical link between a phage-displayed scFv antibody and its encoding DNA sequence inside the phage capsid. (c) Large, random antibody libraries mimicking natural immune repertoires are screened in vitro by phage-display, that is, alternating cycles of selection against an immobilized target antigen and amplification of selected antibody-encoding DNA sequences via phage infection of E. coli host bacteria (2 days per cycle). In the repertoires used in the present study, most residues in the CDR regions of scFv antibodies are random, while other residues (in the framework regions) are kept relatively constant. Here the target antigen is the synthetic, neutral, water-soluble polymer poly(vinylpyrrolidone) (PVP), which is covalently attached to a Petri dish via a photoactive adhesion promoter27,28 (the precise physical state of the polymer is unknown; polymer chains are nonspecifically linked to the surface and their representation here is arbitrary). During selection, phages that display antibody clones that do not interact strongly with the immobilized target are washed away, while phages displaying anti-PVP antibodies are selected. As we report here, scFv antibody libraries may contain a small proportion of antibody clones lacking either VH or VL that may get selected. After generally 3 or 4 cycles of phage-display selection and amplification in bulk, selected antibody clones are arrayed in 96-well plates and tested individually for binding to the target of interest, here PVP. mL of cobalt-ion resin that had been washed with 10 mL of PBS. The column was then washed with 10 mL of PBS + 15 mM imidazole and eluted with 5 mL of PBS + 250 mM imidazole. The 500 μL fractions were collected, dialyzed against PBS, and analyzed by SDS-PAGE. Importantly, we have observed that Ni-NTA resins are often inappropriate for such purification because they become saturated with the medium components and provide a much lower yield and purity than cobalt-ion resins. Direct and Competitive ELISA Tests for Antibody Binding to PVP. Two wells without bacteria but with induction media served as negative controls. A total of 100 μL of each of the 94 antibody-cloneculture supernatants was used for ELISA. Polystyrene Greiner 96-well plates containing immobilized PVP (or not, for control plates) were blocked with PBS + milk (2% w/v) for 1 h. Next, soluble antibodies from each clone were incubated in wells for 1 h at room temperature. An optional incubation with PBS + Tween20 (0.1% w/v) + milk (2%

w/v) of duration between 1 min and 12 h was applied at this stage to estimate the apparent kinetic dissociation rate constant of anti-PVP antibodies. Detection was performed by subsequent 45 min incubations with anti-His6 mouse monoclonal antibody diluted at 1:1000 in PBS + 0.1% w/v Tween20 + 2% w/v milk (clone HIS1, Sigma-Aldrich, St Louis, MO) and an antimouse horseradishperoxidase-conjugated polyclonal secondary antibody raised in goat, diluted at 1:3000 in PBS + Tween20 + milk (from Invitrogen, distributed by Life Technologies, Grand Island, NY). A 5 s washing step with PBS + Tween20 (0.1% w/v) was carried out between successive incubation with the selected antibody supernatant, the antiHis6 antibody, and the secondary antibody in each well. These incubations were followed by a peroxidase colorimetric assay using tetramethylbenzidine (TMB) as a substrate. 100 μL of a sodium acetate buffer (0.1 M, pH = 6.0) supplemented with TMB (100 mg/L) and H2O2 (0.02% v/v) was added to wells and the enzymatic assay was 115


Biomacromolecules

Article

Figure 2. Anti-PVP antibodies bind proportionally to PVP polymerization degree DP and are sensitive down to one monomer unit. (a) Competitive ELISA was performed in the presence of PVP of various polymerization degrees (DP = 100, 400, 3600, or 13000 monomer units long) or VP monomer (DP = 1) in the bulk solution. Increasing concentrations of both PVP and VP (x axis) in the bulk solution induce a decrease in antibody binding to immobilized PVP (y axis), confirming the specificity of antibody clone T3A2 binding to PVP and identifying the VP monomer as a direct target of this antibody (error bars represent ± SD). (b) C50 represents the monomer concentration of competitor in the bulk solution at which 50% of the antibody is displaced from immobilized PVP (error bars represent ± SD). For a given competitor in the bulk solution, the lower the C50, the higher the antibody binding affinity. For DP > 3600, C50 does not depend on DP, C50(3600 < DP < 106) = 11 mM ± 0.5 mM. For DP = 100, C50 is 3.9 times higher than for DP > 3600, C50(DP = 100) = 43 ± 4 mM. For DP = 1, the 1-vinyl-2-pyrrolidone monomer (VP, C6H9NO, MW = 111 g· mol−1) turned out to displace the antibody from immobilized PVP for all anti-PVP antibodies tested. For VP (DP = 1), C50 is 21 times higher than for DP > 3600, C50(VP) = 234 ± 25 mM, but significantly lower than that of all other chemicals tested, including very similar compounds such as Nmethyl-2-piperidone (NMP), C50(NMP) = 387 ± 10 mM, and 1-methyl-2-pyrrolidone (MP), C50(MP) = 548 ± 25 mM. C50(VP) is 1.7-fold lower than C50(NMP), p = 0.06, and 2.4-fold lower than C50(MP), p = 0.04 (Student’s t test), demonstrating exquisite binding specificity. The T3A2 antibody also binds much more strongly to PVP than to poly(2-ethyl-2-oxazoline) (PEOxa) of the same polymerization degree, DP = 5000, C50(PEOxa, DP = 5000) = 808 ± 142 mM. C50(PVP, DP = 5000) is 62-fold lower than C50(PEOxa, DP = 5000), p = 0.01. stopped after 10 min by adding 50 μL of 1 M HCl to each well followed by immediate absorbance detection by imaging with a CCD camera. For competitive ELISA experiments, the procedure was identical except that the corresponding competitor was coincubated at the indicated concentration in the bulk solution together with the antiPVP antibody. The amount of antibody bound to immobilized PVP was quantified through binding of horseradish-peroxidase-conjugated secondary antibodies to the anti-PVP antibodies, followed by a colorimetric peroxidase enzymatic reaction assay detected by a CCD camera producing 8-bit images (intensity coded by integers from 0 to 255). Illumination was performed with a 395 mm × 250 mm, 2 × 8W Rex light panel lamp (Rex Leuchtplatten, Blaustein, Germany), on top of which the 96-well plates were sitting. Data acquisition was performed using a Canon G10 CCD camera (Canon France, Courbevoie, France) controlled by a PC Running Remote Capture software (Canon,

Tokyo, Japan) at 10 ms exposure time. Image quantification was performed using ImageJ software (http://rsbweb.nih.gov/ij/) with our own macros. We first measured the average intensity of a negative control well that did not contain antibodies (but did contain colorimetric assay reagents). Then we subtracted the average intensity of the negative control well from the average intensity inside each well of the 96-well plate. Technically, each well score is bounded between −255 and 255, although in practice after subtraction, the intensity ranges approximately between 0 and 100, which ensures that detection is far from saturation. We used these pixel intensities as arbitrary units. Graphs were performed with plot (by M. Wesemann, http://plot. micw.eu/). Student t-tests yielding p-values were performed with R (http://www.r-project.org/), error bars on plots representing standard deviation. Experiments were carried out in duplicate, at least twice on separate occasions. From each ELISA curve we determined C50, the bulk solution concentration of competitor at which half of the antibody is displaced 116


Biomacromolecules

Article

from immobilized PVP. Since the precise amount and state of immobilized PVP is not known, we have no access to the absolute value of antibody-soluble competitor binding constants. However, since immobilized PVP and scFv concentrations are uniform in all wells, we can compare C50 values. A rough estimate of the dissociation rate constant can be computed by assuming a first-order exponential decay of the concentration of antibody bound to immobilized PVP during the additional incubation with buffer. The amount of PVP-bound antibody is proportional to exp(−t/τ), where the dissociation rate constant is 1/τ. Since the signal is unchanged after 12 h, exp(−t/τ) ∼ 1 for t = 12 h, hence, τ > 12 h, so 1/ τ < (12 × 3600)−1 s−1. The dissociation rate constant is thus smaller than 10−4 s−1. Statistical Analysis of Recombinant Antibody Sequencing Data. Based on results from ELISA experiments, 96 antibody clones out of 111 showing strong binding to PVP were randomly picked and sequenced. Clones were named TnXY or GnXY, where T or G correspond to Tomlinson or Griffin library, n is the selection round, and XY are the clone coordinates on the arrayed 96-well plate. Sequencing was performed at GATC Biotech, Germany, by using the pHEN-SEQ primer (90 out of 96 successful sequences). Sequences were aligned using the ClustalW 2.1 web-based algorithm (http:// www.ebi.ac.uk/Tools/msa/). The sequence logo was produced using the weblogo (v. 2.8.2) web-based application (http://weblogo. berkeley.edu/logo.cgi). The same procedure was applied to 40 antibody clones that were picked at random from the initial Tomlinson library. The sequences of natural anti-PVP antibodies were translated and aligned using the same ClustalW 2.1 web-based algorithm. Deletion of VH and VL from Antibody Clones Possessing both a VH Chain and VL Chain. Clones that lack either VH or VL retain all the restriction sites for independently subcloning VH or VL and are functional. The vectors of both Griffin and Tomlinson libraries are identical except in the linker sequence. To perform VL deletion of VH− VL clones, the VH fragment of every VH−VL clone was extracted by NcoI/XhoI digestion and ligated into the backbone of an antibody possessing only VH (the VH fragment of which had been extracted by NcoI/XhoI digestion). In a similar way, VH deletion was performed by subcloning the SalI/NotI-extracted VL fragment of every VH-VL antibody into the backbone of a SalI/NotI-digested antibody clone possessing only a VL chain. Subcloning efficiency was verified by sequencing using the pHenSeq primer. These deleted clones were again tested by ELISA for PVP binding. Their correct expression as soluble antibodies secreted in the culture medium was systematically confirmed by dot blot analysis.

clones after 3 and 4 rounds of selection. These were arrayed into 96-well plates containing two negative control antibody clones and tested by ELISA in their secreted, non-phagedisplayed form for binding to PVP. The majority of selected antibody clones (111 out of 188 tested, 59%) were found to bind strongly to immobilized PVP. Binding specificity was assessed by performing two series of control experiments: (i) three negative control antibodies that had been previously isolated from the same repertoires and that are directed against other targets (two antibodies from the Griffin repertoire directed against protein targets, NM-myosin IIA and giantin,30 and one antibody from the Tomlinson repertoire directed against gold) did not bind immobilized PVP, thus ruling out nonspecific interactions between antibody scaffolds present in these repertoires and PVP (Supporting Information, Figure 1); (ii) anti-PVP antibodies did not bind to negative control targets (non-PVP-coated plates, plates treated with adhesion promoter Azigrip4 in the absence of PVP, see Supporting Information, Figure 2, and metal surfaces, data not shown). Binding specificity was further confirmed in competitive ELISA assays with increasing concentrations of polymers in the bulk solution. Polymer competitors were either PVP or negative control polymers that have an empirical formula close to that of PVP but with different functional groups (Supporting Information, Figure 3). Binding of the selected anti-PVP antibodies to immobilized PVP decreased when increasing the concentration of PVP in the bulk solution (Figure 2a). No other polymers tested were found to displace anti-PVP antibodies from immobilized PVP in our competitive binding assay, apart from poly(2-ethyl-2-oxazoline) (PEOxa) at a concentration of 450 mM in the bulk solution, hence, with a much lower efficiency than PVP of the same polymerization degree (Supporting Information, Figure 3, and Figure 2b). In this context, we note that PEOxa is distinct among the polymers investigated in that its monomers do not incorporate cyclic motifs. In a control experiment, we have checked that PVP or VP at the highest bulk solution concentration used in our competitive assay did not displace antibodies directed against other targets from their immobilized targets (Supporting Information, Figure 4). The high binding specificity of antiPVP antibodies for PVP and VP (see below) in comparison to polymers and compounds that have a similar empirical formula but with different functional groups is reminiscent of the high binding specificity of antibodies to biomolecules such as proteins31,32 and DNA,33,34 as well as that of antibodies to solid crystalline surfaces.15 In drawing this analogy, we emphasize the differences between biopolymers and synthetic homopolymers. Proteins are dense heteropolymers, while PVP is a swollen homopolymer. After confirming the binding specificity of selected anti-PVP antibodies, we addressed their binding affinity. In our ELISA assay antibody binding to PVP was not affected by an additional incubation for up to 12 h with buffer alone following antibody incubation and preceding detection with secondary antibodies. Assuming exponential decay, this suggests an upper bound for the kinetic dissociation rate constant of less than (12 h)−1 ∼ 10−4 s−1, which is typical of very strong binders, such as nanomolar affinity antibodies directed against proteins. PVP polymer chains display many contiguous antibodybinding sites that are composed of an undetermined number of monomer repeats. An anti-PVP-backbone antibody may bind to a large number of equivalent sites along the homopolymer chain. Accordingly the antibody/PVP-chain binding-affinity

■

RESULTS We have independently screened two distinct recombinant antibody repertoires of human origin (the Tomlinson and the Griffin repertoires) against immobilized PVP chains covalently attached to Petri dishes (Figure 1). Antibodies of these two repertoires are in the single-chain-fragment V (scFv) format. The scFv is one of the smallest functional recombinant antibody formats, in which the heavy and light chain of the fragment V (VH and VL) are directly linked through a flexible glycine/serine-rich peptide (Figure 1a). scFv antibodies are monovalent and therefore can bind only one antigen at a time. scFv antibodies recapitulate the binding specificity of their IgG counterparts, which are divalent. In both repertoires, residues defining the complementary determining regions (CDRs) of VH and VL, that is, antibody hypervariable loops that are responsible for specific interactions with antigens, are random. In contrast residues in framework regions, that is, those responsible for proper antibody scaffold structure, are relatively constant. Both repertoires yielded overall more than 100 antibody clones that bind PVP. A total of 94 antibody clones were randomly picked for each screen from the pool of selected 117


Biomacromolecules

Article

specifically to PVP (see Supporting Information, Figures 1, 2, and 5). They represent a small fraction ( 3600 (Figure 2b), C50(DP > 3600) = 11 ± 0.5 mM. If we assume that C50 is inversely proportional to the binding-affinity constant of the antibody for soluble PVP, this result is consistent with the theoretical prediction for an antibody binding to a polymer-chain backbone in the dilute regime.35 The agreement with the backbone-binding model suggests the presence along the PVP chains of numerous equivalent binding sites for anti-PVP antibodies. For DP < 3600, C50 increases with decreasing DP, and the onset of the deviation from constant binding seems to occur for DP ∼ 1000−2000, based on interpolation of the data in Figure 2b. For DP = 400, C50 is 1.6 times higher than for DP > 3600, C50(DP = 400) = 18 ± 1 mM. For DP = 100, C50 is 3.9 times higher than for DP > 3600, C50(DP = 100) = 43 ± 4 mM. For DP = 1, the 1-vinyl-2-pyrrolidone monomer (VP, C6H9NO, MW = 111 g·mol−1) turned out to displace all anti-PVP antibodies tested from immobilized PVP, despite its small size comparable to the smallest haptens known to date.36,37 For VP (DP = 1), C50 is 21 times higher than for DP > 3600, C50(DP = 1) = 234 ± 25 mM, but significantly lower than that of all other compounds tested, including very similar compounds such as 1methyl-2-pyrrolidone (MP), C50(MP) = 548 ± 25 mM (Figure 2b, Supporting Information, Figure 3). This suggests that the anti-PVP antibody binding to VP we observe is indeed specific. The 21-fold deviation may be attributed to the weaker binding energy of monomers, the full binding free energy being achieved for an oligomer that is long enough to engage the entire antigen-binding site. The increase of C50 with decreasing DP may reflect chemical “end effects” due to the polymerization of the VP vinyl groups. It may also relate to the translational entropy loss per bound monomer. This issue remains to be clarified and addressed using more quantitative techniques, allowing the direct determination of absolute binding constants. Our approach has identified a large number (111) of recombinant antibodies binding to PVP. This opens up the possibility of statistical analysis of selected antibody sequences in order to identify candidate PVP-binding motifs in antibody sequences. We sequenced 90 out of 111 anti-PVP antibody clones and identified 44 distinct DNA sequences (see Supporting Information for an alignment of selected anti-PVP antibody clone amino-acid sequences). The most represented antibody clones were present up to 5 times among all 90 sequenced clones. Further analysis of the anti-PVP antibody clone sequence alignment in comparison with the sequence alignment of randomly picked antibody clones from the original repertoires suggests that selected anti-PVP antibodies bind to PVP through only one antibody chain. Indeed, a significant proportion of anti-PVP antibody clones lack either V H (Tomlinson library, 2 clones out of 33, 6%) or VL (Griffin library, 5 clones out of 11, 45%). The antibody clones lacking either VH or VL are correctly expressed, soluble and bind 118


Biomacromolecules

Article

have also found enriched in the VH-CDR3 regions of antibodies we have selected in vitro by phage-display from human repertoires (Figure 3a).

■

DISCUSSION Our in vitro antibody phage-display approach provides novel information about the interactions between PVP, a synthetic, neutral, water-soluble polymer, and human antibody repertoires. The polymer PVP is detected specifically by selected anti-PVP antibodies down to the monomer level, and the antibody-PVP binding constant is proportional to PVP polymerization degree DP for DP > 1000. A total of 44 distinct antibodies were isolated out of 90 randomly picked anti-PVP antibodies. Most of them share a conserved aminoacid motif in the VH-CDR3 region, containing tyrosine, glycine, aspartic acid and asparagine residues, which is therefore a candidate motif for binding to PVP. Almost all selected antiPVP antibodies (90%) bind to PVP through only one antibody chain, VH or VL. Our approach is applicable to any other synthetic polymers. It is fast (phage-display screens take 1 week) and cost-effective in comparison to animal immunization experiments, by several orders of magnitude. The anti-PVP antibodies we have selected from two repertoires of human origin exhibit a PVP-binding constant that is proportional to PVP polymerization degree DP for DP > 1000, as predicted by theory for antibodies binding to a polymer backbone.35 Anti-PVP antibodies also bind the VP monomer, in spite of its small size (MW = 110 g·mol−1). The monovalent anti-PVP antibodies we report possess each a single antigen-binding site and may bind at numerous identical antibody-binding sites along the PVP backbone. The antibodyPVP interactions are specific down to the VP monomer level even though binding affinity is weaker for VP than for PVP of DP > 1000 per monomer unit. Each antibody-binding site comprises a number of monomer units. Monomers have a weaker binding energy, and the full binding free energy is achieved for an oligomer that is long enough to engage the entire antigen-binding site. Binding to PVP is highly targetspecific, since anti-PVP antibodies bind much less to compounds that are very similar to PVP and VP, such as MP, and no binding is detected in the case of control targets (such as noncoated plates). Binding is also antibody-sequence specific. Indeed, antibodies isolated from the same library against other targets do not bind to immobilized PVP. The amino-acid sequence of non-PVP-binding antibodies is identical to that of anti-PVP antibodies at most positions and in particular in the framework regions. The binding specificity is thus due to the few variable positions in the antibodies and in particular in the VH-CDR3. We discuss this point in detail below. These results indicate that human repertoires contain antibodies that bind specifically to synthetic polymers, in particular to synthetic polymer backbones. The conserved amino-acid motif we have identified in most selected anti-PVP antibodies, as well as the other distinct amino-acid regions that were present in other antibody clones, provide information about possible interaction mechanisms at the molecular level. Tyrosine, serine, and glycine residues are generally found at high frequency in CDR regions, where they are believed to play a major role in interactions with neutral epitopes of protein antigens.38 Our analysis suggests that they are also involved in antibody interactions with neutral polymers. Aspartic acid and asparagine are not as frequently found in CDRs as tyrosine, serine, and glycine, and their strong

Figure 3. Anti-PVP antibodies contain a conserved amino-acid motif and bind through only one antibody chain. (a) The sequence logo represents the information content of every possible amino acid at every position within CDR2 and CDR3 regions of anti-PVP scFv antibodies we have selected from the Tomlinson library. All positions of both CDR1 regions as well as six positions of CDR2 and CDR3 regions (positions 2, 6, and 9 of VH-CDR2, 2, and 3 of VL-CDR2, 5 of VL-CDR3, respectively, I, G, T, A, S, and P residues) are fixed in the Tomlinson library and display scores of nearly 4 bits in anti-PVP antibody clones as well as antibody clones that were randomly picked from the initial repertoires. All random positions of the initial repertoire display scores below 1.8 bits (threshold defined by a dashed line). All four positions of anti-PVP antibody VH-CDR3 regions display scores that are higher than this threshold as well as the first and last positions of anti-PVP antibody VH-CDR2 regions, highlighting strong selection at these positions of tyrosine (Y), serine (S), glycine (G), aspartic acid (D), and asparagine (N) residues. No significant selection is detected for VL-CDRs of anti-PVP antibodies in comparison to antibodies randomly picked from the initial repertoire. (b) The role of both antibody chains was directly tested by deleting VH and VL of 11 anti-PVP antibody clones possessing both a VH chain and a VL chain. In all cases, VH recapitulates antibody binding to immobilized PVP, while VL does not bind to PVP (error bars represent ± SD).

The PV8 antibody is coclustered in the coalignment analysis with the in vitro selected antibodies containing the YGDN motif, and its VH-CDR3 region displays 79% identity and 86% similarity with that of the T3D7 antibody clone (that is also in the same cluster). The natural murine PV1 and PV4 antibodies (respectively IgM and IgG3 isotypes) are coclustered with two antibodies isolated from the Griffin repertoire, G4F6 and G3D9, albeit with weaker identity and similarity scores (the highest ones being between G4F6 and either PV1 or PV4, 54% identity and 54% similarity). In addition, the VH-CDR3 regions of natural murine anti-PVP antibodies are significantly enriched in tyrosine (Y), glycine (G) and serine (S) residues that we 119


Biomacromolecules

■

selection in our screens suggests that they may be specifically responsible for antibody/PVP interactions. The VH-CDR3 regions of 16 PVP-reactive B cells of mice immunized with PVP were sequenced and can serve as a starting point for the comparison of our in vitro results and in vivo data.4 The in vivo sequence data set is very limited in comparison to our in vitro data set, since only 16 sequences of natural antibodies are available. In addition, these natural antibodies are of a different species, mostly of the IgM isotype, and their published sequences are restricted to the VH-CDR3 regions. Moreover, there is no information about antibody binding affinity or specificity, the relative contributions of VH and VL for binding to PVP, and the dependence of antibody/PVP binding as a function of PVP polymerization degree for these natural antiPVP antibodies. Nevertheless, the VH-CDR3 of one of the natural murine IgM antibodies identified in the in vivo study (the PV8 IgM antibody) displays a strong similarity to the YGDN motif-containing antibodies we have selected in vitro by phage-display from human repertoires. Besides, 3 out of 5 residues (tyrosine, serine, glycine) we have identified in the CDR regions of in vitro selected antibodies are also present at high levels in the VH-CDR3 regions of 16 natural anti-PVP antibody sequences. This suggests a significant correlation between the VH-CDR3 sequences of natural and in vitro selected anti-PVP antibodies. Given the progress of antibody and sequencing technologies since the publication of this report on mouse anti-PVP antibodies, it is now possible to extend the in vivo data set for a larger scale comparison with in vitro results, using for example humanized mouse models or PVPreactive human B cells. Our results point to systematic binding of antibodies to PVP through only one antibody chain, even in antibodies possessing both a VH chain and a VL chain. Binding of antibodies through only one antibody chain to protein targets has already been described,39 can be optimized,40 and has now become an important strategy in the production of therapeutic antibodies. However, there is no report to our knowledge demonstrating systematic binding through only one antibody chain for antibodies isolated from the same or similar repertoires upon selection against a protein target. This unexpected finding might apply either specifically to PVP, or more generally to homopolymer targets, that is, macromolecules that display the same antibody-binding site many times. The validity of our results is corroborated by the fact that two distinct antibody repertoires were screened independently and qualitatively yielded the same results. It is likely that the same features apply to natural anti-PVP antibodies, since the recombinant scFv format recapitulates the binding properties of natural IgG antibodies (albeit in the monovalent form and thus with different association/dissociation kinetics and binding constants). The PV8 murine anti-PVP antibody, which is coclustered in the sequence analysis with antibodies that we have selected in vitro and shown to bind through only one antibody chain, is a good candidate to test this hypothesis. Regarding the molecular mechanisms of synthetic polymer antigenicity, this would imply that the potential number of antibodies in repertoires binding to PVP is a priori larger than for antigens requiring both antibody chains to bind (which is generally assumed for most antigens) since several antibodies might bear similar heavy or light chains.

Article

CONCLUSIONS

Our observations of the systematic binding through only one antibody chain and the presence of an amino-acid motif shared by many distinct antibody clones suggest that many anti-PVP antibodies from human repertoires bind through a small number of shared mechanisms. Our results provide significant evidence that in vitro selection correlates with in vivo immunization experiments in terms of the amino-acid composition of the antigen-binding regions of antibodies. It will be very interesting to conduct a more extensive comparison between natural in vivo response and antibody-phage-display in vitro response to PVP and extend the approach to other synthetic polymers. Extensive statistical analysis of the selected antibody sequences offers great promise in terms of a quantitative, cost-effective, and rapid in vitro estimation of the interactions between polymers and human antibody repertoires. Our study is based on an original combination of immunological techniques and polymer science that allows the characterization of antibody/polymer material interactions through screening of large antibody libraries. Without any prior knowledge of possible mechanisms of polymer material/ antibody interactions, our approach provides a rapid method to characterizing those interactions for a wide antibody repertoire with high statistical significance and in vivo relevance, and further, to unveiling possible interaction mechanisms. For instance, antibody phage display could be used downstream of polymer material design to rapidly test a panel of candidate polymer/copolymer materials and compare their in vitro antigenicity profiles. As noted in the introduction, our results suggest the possibility of a molecular-biology approach to protein−polymer conjugation based on genetic fusion of antipolymer antibodies to other proteins. A similar concept was utilized by Watanabe et al. for the immobilization of biological molecules on gold surfaces25 but its potential for protein−polymer conjugation is yet to be investigated. In this context it is first helpful to note the following observations: (i) Genetic fusion is currently used to produce multimeric formats of antibody fragments varying from bi- to tetravalent.41−43 This technology also permits one to engineer bi and trispecific antibody multimers.44 To engineer such antibodies one needs knowledge of the DNA sequence coding the antibodies, as can be obtained from the phagedisplay experiments discussed in our current study. (ii) Antipolymer antibodies may bind the backbone of a homopolymer or its end groups.45 (iii) The existence of antipolymer antibodies is established for at least two homopolymers, namely, PEG and PVP. With these observations in mind, we can suggest some of the potential options afforded by “molecular-biology conjugation”: (i) Fusion of antiPEG antibody to therapeutic protein provides an alternative route to reversible PEGylation, hitherto only achieved by chemical means;46 (ii) Anti-PEG-anti-PVP bispecfic scFv antibodies in an aqueous solution of PVP and PEG can give rise to polydisperse PVP-PEG block copolymers when both antibodies are end binders, or to PEG-PVP gel when both antibodies are backbone binders; (iii) The use of trivalent trispecific antibody fragments may thus permit the design of a PVP-PEG hydrogel with functionalized junctions. This incomplete list illustrates the interest in exploring antipolymer antibodies using phage display to enable a novel biological approach to the design of polymeric materials. 120


Biomacromolecules

■

Article

(16) Watanabe, H.; Nakanishi, T.; Umetsu, M.; Kumagai, I. J. Biol. Chem. 2008, 283, 36031−36038. (17) Tamerler, C.; Oren, E. E.; Duman, M.; Venkatasubramanian, E.; Sarikaya, M. Langmuir 2006, 22, 7712−7718. (18) So, C. R.; Tamerler, C.; Sarikaya, M. Angew. Chem., Int. Ed. 2009, 48, 5174−5177. (19) Lee, Y. J.; Yi, H.; Kim, W.-J.; Kang, K.; Yun, D. S.; Strano, M. S.; Ceder, G.; Belcher, A. M. Science 2009, 324, 1051−1055. (20) Dang, X.; Yi, H.; Ham, M.-H.; Qi, J.; Yun, D. S.; Ladewski, R.; Strano, M. S.; Hammond, P. T.; Belcher, A. M. Nat. Nanotechnol. 2011, 6, 377−384. (21) de Wildt, R. M.; Mundy, C. R.; Gorick, B. D.; Tomlinson, I. M. Nat. Biotechnol. 2000, 18, 989−994. (22) Griffiths, A.; Williams, S. C.; Hartley, O.; Tomlinson, I. M.; Waterhouse, P.; Crosby, W. L.; Kontermann, R. E.; Jones, P. T.; Low, N. M.; Allison, T. J. EMBO J. 1994, 13, 3245−3260. (23) Andersson, B.; Blomgren, H. Cell. Immunol. 1971, 2, 411−424. (24) Klok, H. A. Macromolecules 2009, 42, 7990−8000. (25) Watanabe, H.; Kanazaki, K.; Nakanishi, T.; Shiotsuka, H.; Hatakeyama, S.; Kaieda, M.; Imamura, T.; Umetsu, M.; Kumagai, I. Langmuir 2011, 27, 9656−9661. (26) Zürcher, S.; Tosatti, S.; Dorcier, A.; Fusco, S.; Lopez, I.. Adhesion promoter based on a functionalized macromolecule comprising photoreactive groups. Patent EP2236524, October 6, 2010. (27) Bielecki, R. M.; Doll, P.; Spencer, N. D. Tribol. Lett. 2012, 49, 273−280. (28) Serrano, Â .; Sterner, O.; Mieszkin, S.; Zürcher, S.; Tosatti, S.; Callow, M. E.; Callow, J. A.; Spencer, N. D. Adv. Funct. Mater. 2013, DOI: 10.1002/adfm.201203470. (29) Nizak, C.; Moutel, S.; Goud, B.; Perez, F. Methods Enzymol. 2005, 403, 135−153. (30) Nizak, C.; Martin-Lluesma, S.; Moutel, S.; Roux, A.; Kreis, T. E.; Goud, B.; Perez, F. Traffic 2003, 4, 739−753. (31) Nizak, C.; Monier, S.; del Nery, E.; Moutel, S.; Goud, B.; Perez, F. Science 2003, 300, 984−987. (32) Vielemeyer, O.; Yuan, H.; Moutel, S.; Saint-Fort, R.; Tang, D.; Nizak, C.; Goud, B.; Wang, Y.; Perez, F. J. Biol. Chem. 2009, 284, 20791−20795. (33) Modi, S.; Nizak, C.; Surana, S.; Halder, S.; Krishnan, Y. Nat. Nanotechnol. 2013, 8, 459−467. (34) Modi, S.; Halder, S.; Nizak, C.; Krishnan, Y. Nanoscale 2013, 10.1039/C3NR03769J. (35) Halperin, A.; Kröger, M. Langmuir 2009, 25, 11621−11634. (36) Chappey, O.; Debray, M.; Niel, E.; Scherrmann, J. M. J. Immunol. Methods 1994, 172, 219−225. (37) Tuomola, M.; Harpio, R.; Mikola, H.; Knuuttila, P.; Lindström, M.; Mukkala, V. M.; Matikainen, M. T.; Lövgren, T. J. Immunol. Methods 2000, 240, 111−124. (38) Birtalan, S.; Zhang, Y.; Fellouse, F. A.; Shao, L.; Schaefer, G.; Sidhu, S. S. J. Mol. Biol. 2008, 377, 1518−1528. (39) Ward, E. S.; Güssow, D.; Griffiths, A. D.; Jones, P. T.; Winter, G. Nature 1989, 341, 544−546. (40) Barthelemy, P. A.; Raab, H.; Appleton, B. A.; Bond, C. J.; Wu, P.; Wiesmann, C.; Sidhu, S. S. J. Biol. Chem. 2008, 283, 3639−3654. (41) Mack, M.; Riethmüller, G.; Kufer, P. Proc. Natl. Acad. Sci. U.S.A. 1995, 92, 7021−7025. (42) Doppalapudi, V. R.; Huang, J.; Liu, D.; Jin, P.; Liu, B.; Li, L.; Desharnais, J.; Hagen, C.; Levin, N. J.; Shields, M. J.; Parish, M.; Murphy, R. E.; Del Rosario, J.; Oates, B. D.; Lai, J.-Y.; Matin, M. J.; Ainekulu, Z.; Bhat, A.; Bradshaw, C. W.; Woodnutt, G.; Lerner, R. A.; Lappe, R. W. Proc. Natl. Acad. Sci. U.S.A. 2010, 107, 22611−22616. (43) Kipriyanov, S. M.; Little, M.; Kropshofer, H.; Breitling, F.; Gotter, S.; Dübel, S. Protein Eng. 1996, 9, 203−211. (44) Holliger, P.; Hudson, P. J. Nat. Biotechnol. 2005, 23, 1126− 1136. (45) Sherman, M. R.; Williams, L. D.; Sobczyk, M. A.; Michaels, S. J.; Saifer, M. G. P. Bioconjugate Chem. 2012, 23, 485−499. (46) Filpula, D.; Zhao, H. Adv. Drug Delivery Rev. 2008, 60, 29−49.

ASSOCIATED CONTENT

S Supporting Information *

Supplemental Figures 1−5 and antibody sequence alignment analysis. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: +33 (0) 1 40 79 45 87. Present Address †

Recombinant Protein Production Group, Microbiology Department, NUI Galway, University Road, Galway, Ireland (A.S.).

Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work was supported by the Centre National de la Recherche Scientifique through its “Prise de Risque Interface Physique-Chimie-Biologie” program. A.S. was the recipient of a postdoc fellowship from the SMINGUE department at Université Grenoble 1. Antoine Dorcier and Elizabeth Duncan (SuSoS AG) assisted us technically in the preparation of PVPfunctionalized polystyrene plates. Purvi Jain performed the control based on gold and an antigold antibody. The Griffin and Tomlinson libraries were provided by the G. Winter laboratory at MRC Cambridge and Geneservice (now at Source Bioscience), respectively. We acknowledge Bahram Houchmandzadeh, Olivier Rivoire, and Erik Geissler for fruitful advice and decisive support to this project, and thank our colleagues of the Biophysics group at LIPhy for many stimulating discussions.

■

REFERENCES

(1) Gill, T. J.; Kunz, H. W. Proc. Natl. Acad. Sci. U.S.A. 1968, 61, 490−496. (2) Andersson, K.; Wrammert, J.; Leanderson, T. Immunol. Rev. 1998, 162, 173−182. (3) Lite, H. S.; Braley-Mullen, H. J. Immunol. 1981, 126, 928−933. (4) Whitmore, A. C.; Haughton, G.; Arnold, L. W. Int. Immunol. 1996, 8, 533−542. (5) Armstrong, J. K.; Hempel, G.; Koling, S.; Chan, L. S.; Fisher, T.; Meiselman, H. J.; Garratty, G. Cancer 2007, 110, 103−111. (6) Veronese, F. M.; Pasut, G. Drug Discovery Today 2008, 5, e57− e64. (7) Abuchowski, A.; van Es, T.; Palczuk, N. C.; Davis, F. F. J. Biol. Chem. 1977, 252, 3578−3581. (8) Liu, Y.; Reidler, H.; Pan, J.; Milunic, D.; Qin, D.; Chen, D.; Vallejo, Y. R.; Yin, R. J. Pharmacol. Toxicol. Methods 2011, 64, 238− 245. (9) Garay, R. P.; El-Gewely, R.; Armstrong, J. K.; Garratty, G.; Richette, P. Exp. Opin. Drug Delivery 2012, 9, 1319−1323. (10) Bradbury, A. R. M.; Sidhu, S.; Dübel, S.; Mccafferty, J. Nat. Biotechnol. 2011, 29, 245−254. (11) Whaley, S. R.; English, D. S.; Hu, E. L.; Barbara, P. F.; Belcher, A. M. Nature 2000, 405, 665−668. (12) Mao, C.; Solis, D. J.; Reiss, B. D.; Kottmann, S. T.; Sweeney, R. Y.; Hayhurst, A.; Georgiou, G.; Iverson, B.; Belcher, A. M. Science 2004, 303, 213−217. (13) Sanghvi, A. B.; Miller, K. P.-H.; Belcher, A. M.; Schmidt, C. E. Nat. Mater. 2005, 4, 496−502. (14) Barbas, C. F.; Rosenblum, J. S.; Lerner, R. A. Proc. Natl. Acad. Sci. U.S.A. 1993, 90, 6385−6389. (15) Artzy Schnirman, A.; Zahavi, E.; Yeger, H.; Rosenfeld, R.; Benhar, I.; Reiter, Y.; Sivan, U. Nano Lett. 2006, 6, 1870−1874. 121


General In Vitro Method to Analyze the Interactions of Synthetic

Recommend Documents