A High Throughput Protein Microarray Approach to Classify HIV Monoclonal Antibodies and Variant Antigens

In recent years, high throughput discovery of human recombinant monoclonal antibodies (mAbs) has been applied to greatly advance our understanding of the specificity, and functional activity of antibodies against HIV. Thousands of antibodies have been generated and screened in functional neutralization assays, and antibodies associated with cross-strain neutralization and passive protection in primates, have been identified. To facilitate this type of discovery, a high throughput-screening tool is needed to accurately classify mAbs, and their antigen targets. In this study, we analyzed and evaluated a prototype microarray chip comprised of the HIV-1 recombinant proteins gp140, gp120, gp41, and several membrane proximal external region peptides. The protein microarray analysis of 11 HIV-1 envelope-specific mAbs revealed diverse binding affinities and specificities across clades. Half maximal effective concentrations, generated by our chip analysis, correlated significantly (P<0.0001) with concentrations from ELISA binding measurements. Polyclonal immune responses in plasma samples from HIV-1 infected subjects exhibited different binding patterns, and reactivity against printed proteins. Examining the totality of the specificity of the humoral response in this way reveals the exquisite diversity, and specificity of the humoral response to HIV.


Microarray fabrication and hybridization
For microarrays, all HIV-1 envelope proteins and MPER analogues were printed at three different concentrations (1.0 mg/mL, 0.1 mg/mL, and 0.01 mg/mL) onto nitrocellulose coated glass FAST slides (Whatman), using an Omni Grid 100 microarray printer (Genomic Solutions). This corresponds to a total protein amount of 1.0 ng, 0.1 ng, and 0.01 ng per spot, respectively. For array Table 1. Sequence alignment of MPER peptides printed on the microarray chip and used for in solution competition assay with HIV-1 JR-FL MPER.

Analog
Sequence References MPER JR-FL 659 ELLELDKWASLWNWFDITNWLWYIK 683 09129 LLELDKWASLWNWFDITNWLWYIKKKK-NH 2 [52] 09122 SLWNWFDITNWLWYIKKKK-NH 2 [24] 09128 ELDKWASLWNWFDITNWLWYIK(IEELKSKIKRIENEIKRIKK) 3 -C22-pT ELDKWASLWNWFNITNWLWYIK(IEELKSKIKRIENEIKRIKK) 3 [25] IZM (IEELKSKIKRIENEIKRIKK) 3  coli lysate is routinely used in our assays to remove background binding especially when working with serum samples. Antibody concentrations for the probing ranged from 10 μg/mL to 0.003 μg/ mL using serial half-log dilution steps. Plasma samples were probed at a dilution of 1:100. Following o/n incubation, slides were washed five times in washing buffer (10 mM Tris (pH 8.0), 150 mM NaCl), containing 0.05% Tween 20, and bound human antibodies were detected by biotin SP-conjugated affini-pure goat anti-human IgG Fc fragment specific secondary antibody (Jackson ImmunoResearch, West Grove, PA), diluted 1/200 in blocking buffer. After 1 h at room temperature (RT), slides were washed three times, and bound antibodies were detected with streptavidin-conjugated SureLight P-3 tertiary reagent (Columbia Biosciences, Columbia, MD), diluted 1/200 in blocking buffer, for 1 h at RT. Slides were then washed 5 times, air dried under brief centrifugation, and stored at 18°C in a desiccator. The arrays were examined with a Perkin Elmer ScanArray Express HT confocal laser scanner at a wavelength of 670 nm and signal intensities were quantified using ProScanArray Express software (Perkin Elmer, Waltham, MA). All signal intensities were corrected for spot-specific background. Only signal intensities above a certain threshold signal (i.e. higher than controls) were considered positive, and used for the calculation of median signal intensities or percentages of bound proteins. Information on the microarray platform is publicly available on NCBI's Gene Expression Omnibus and is accessible through GEO Platform accession number GSE66659.

Specificity ELISA
96-well plates (Costar) were coated with 2 μg/mL (100 ng per well) of respective antigen (gp140, gp120, gp41 or MPER peptide) diluted in PBS and incubated o/n at 4°C. Plates were washed, blocked with 4% non-fat dry milk (NFDM) in PBS containing 0.05% Tween 20, and incubated for 1 h at RT. Serial dilutions (1:4) of mAbs (starting concentration of 10 μg/mL) in 1% NFDM were prepared and added to the washed plates. After 1 h at RT unbound antibodies were removed and bound antibodies were detected by a HRP-labeled goat anti-human Fc gamma specific conjugate (Sigma-Aldrich). Plates were finally washed, developed with TMB (3,3 0 ,5,5 0 -tetramethylbenzidine) substrate, stopped with 2M H 2 SO 4 , and read on a BioTek plate reader at a wavelength of 450 nm. All assays were performed at least twice. Half maximal effective concentration (EC 50 ) was calculated using GraphPad PRISM 5.0 software.

In-solution competition ELISA
The in-solution competition ELISA was performed according to Stanfield and colleagues [30]. In brief, 96-well plates were coated o/n with 200 ng/well neutravidin (Pierce) in PBS at 4°C and blocked with 4% NFDM for 1 h at RT. Different competing MPER peptide analogs (serially diluted 1:3 in 1% NFDM at a starting concentration of 0.04 mM) were mixed with a constant concentration of biotinylated peptide bio-0705 (0.1 μg/ml) or bio-57 (0.1 μg/ml) and the corresponding human mAb 4E10 (0.5 μg/ml) or mAb 2F5 (0.2 μg/ml), respectively. The antibody/ peptide mixture was incubated for 2 h at 37°C, subsequently transferred to the neutravidin plates and incubated at 37°C for 20 min. Bound antibodies were detected with HRP labeled goat anti human F(ab) 2 conjugate (Pierce). After 1 h at RT, wells were developed, stopped and read as described above. The concentration of competitor peptide corresponding to a halfmaximal signal (IC 50 ) was determined by nonlinear fit interpolation of the resulting binding curve (GraphPad PRISM 5.0 software). Each peptide competitor was tested in duplicate in at least two separate experiments.

Virus generation and neutralization assay
Pseudotyped virions were generated by co-transfecting 5 x 10 5 293T cells with a mixture of pSG3ΔEnv backbone plasmid (4 μg), HIV-1 JR-FL, HIV-1 JR-FL D674N , JR-2 or JR-2 D674N envelope complementation plasmid (2 μg), and polyethyleneimine (PEI) at a DNA/PEI ratio of 1/3. Two days post transfection, virus containing supernatant was harvested and cleared by centrifugation at 4000 rpm for 10 min. TZM-bl single round infectivity assay was performed as described elsewhere [31]. Briefly, pseudotyped virus was added at a 1:1 ratio to serially diluted (1:3) mAbs (starting at 100 μg/mL) and incubated at 37°C. After 1 h, TZM-bl reporter cells were added (1:1 by volume) at 1×10 4 cells/well in a final concentration 10 μg/mL DEAE-dextran, and incubated for 48 h at 37°C. Plates were processed, and developed with luciferase assay reagent, according to the manufacturer's instructions (Promega). Luminescence in relative light units (RLUs) was measured using a microplate luminometer (BioTek). All experiments were performed at least in duplicate. The extent of virus neutralization in the presence of antibody was determined at the 50% inhibitory concentration (IC 50 ) in the absence of mAb.

Antibody probing reveals diverse binding patterns against HIV-1 specific antigens
To examine binding attributes of multiple HIV-1 envelope specific antibodies simultaneously, a total of 15 HIV-1 multi-clade envelope proteins, including five gp140, four gp120, one gp41 protein, and six gp41 MPER analogs, were spotted onto microarray chips. All MPER peptides, 09129, 09122, 08023, 09128, and C22-pT (Table 1), were analyzed for mAb 2F5 and mAb 4E10 binding by an in-solution competition assay prior to printing on the protein microarray chip. Analog 09128, a modified version of MPER analog C22-pT, showed remarkably enhanced reactivity to both mAbs 2F5 (IC 50  Based on their reactivity profiles, MPER analogs 09128, C22-pT, 09129, 09122, and 08023 were subsequently used in the chip analysis to screen for 2F5-and 4E10-like antibodies. Each protein and peptide was printed in three concentrations: 1.0 mg/mL, 0.1 mg/mL, and 0.01 mg/mL, corresponding to 1.0 ng, 0.1 ng, and 0.01 ng per spot. After antibody probing and signal quantification, antigens were sorted by envelope and clade into three major clusters (gp140, gp120, and gp41/MPER). Additionally, probed mAbs were grouped into six categories based on their epitope specificity: 1) variable loop 3 (V3) antibodies, 2) CD4bs antibodies, 3) glycan-specific antibodies 4) gp41 antibodies, 5) MPER specific antibodies, and 6) control antibodies. For the generation of the heat map in Fig 1, only antibody signals at non-saturating antigen concentrations were considered in order to decrease nonspecific antibody binding due to high protein concentrations. The optimal HIV-1 protein concentration for mAbs was 0.1 mg/mL, or 0.1 ng per spot. The heat map in Fig 1 shows signal intensities that reveal a diverse binding pattern for each antibody class across the tested antigens.

Variable loop specific antibodies
To compare antibodies within each subclass, percentages and median signal intensities of bound envelope proteins were calculated, and plotted in Fig 2. Percentages of bound envelope proteins were calculated based on a sample size (N) of N = 5 for the gp140 cluster, N = 4 for the gp120 cluster, and N = 5 for the gp41/MPER cluster. The scrambled control peptide 08023 of the gp41/MPER cluster was excluded from the calculation. As indicated in Fig 2A and 2B, the V3 loop-specific mAbs 447-52D and F425 B4e8 exhibited strong cross clade coverage, and intermediate signal intensities throughout the gp140 cluster and gp120 cluster. Highest signal intensities for both antibodies were observed with clade B HIV-1 BaL gp120 and clade B HIV-1 SF162 gp140, followed by intermediate clade A HIV-1 UG21 gp140 and clade F HIV-1 BR29 gp140 binding in case of mAb 447-52D (Fig 1), suggesting a predominant clade B specificity of both V3-loop specific antibodies. In contrast, intermediate to weak binding was observed against  HIV-1 CN54 gp140 and HIV-1 CN54 gp120, whereas very weak binding was detected against HIV-1 93TH975 gp120 (447-52D). None of the V3 loop-specific mAbs reacted against clade D HIV-1 UG21 gp140 and clade C HIV-1 96ZM651 gp120. The strong cross clade reactivity of mAbs 447-52D and F425 B4e8 is consistent with the moderate neutralization breadth observed against HIV-1 primary isolates. Weak poly-reactivity (i.e. signal intensity higher than background level) against HIV-1 JRFL gp41 in the gp41/MPER cluster was observed with mAb F425 B4e8. Some of the poly-reactivity observed might be due to the distinct binding mechanism of mAb F425 B4e8, compared to other V3-specific antibodies like 447-52D. Contact between mAb F425 B4e8 and its linear epitope in the tip of the V3 loop is mainly made through sidechain contacts involving only two residues, thus explaining the considerable cross clade reactivity of this antibody [32].

CD4bs specific antibodies
The CD4bs specific antibodies F105, VRC01, and VRC03 revealed high to weak reactivity across the gp140 cluster and the gp120 cluster at a protein concentration of 0.1 ng/spot ( Fig  2A). Broadly neutralizing mAbs VRC01 and VRC03 exhibited intermediate to weak median cross clade signal intensities against the gp140 and gp120 cluster ( Fig 2B). In contrast, weakly neutralizing mAb F105 showed a strong median cross clade signal intensity against the gp120 cluster, followed by weak binding against the gp140 cluster. Best binding of mAbs VRC01, VRC03, and F105 was achieved against clade B HIV-1 SF162 gp140, and clade B HIV-1 BaL gp120 as well as clade C HIV-1 CN54 gp120 in case of VRC01 (Fig 1). The broadly neutralizing mAb b12 interacted predominantly with clade B in the gp120 cluster. However, the interaction with HIV-1 BaL gp120 was rather weak (Fig 1). Weak poly-reactivity against HIV-1 JRFL gp41, similar to F425 B4e8, was observed with mAb F105.

Glycan specific mAb 2G12
MAb 2G12, the only glycan specific mAb analyzed in this panel, was highly cross-reactive against all tested gp140 and gp120 glycoproteins (Fig 2A). Highest median signal intensities were found against the gp120 cluster ( Fig 2B). As indicated in Fig 1, mAb 2G12 revealed strong binding against clade B HIV-1 BaL gp120 and HIV-1 SF162 gp140, clade C HIV-1 CN54 gp120, and intermediate binding signals against clade A HIV-1 UG37 gp140 and clade E HIV-1 93TH975 gp120. The signal intensities of mAb 2G12 in the gp140 cluster are in good agreement with a recent study where broadly neutralizing mAb 2G12 was able to neutralize isolates from various clades with varying efficiency [33,34]. Minimal reactivity of mAb 2G12 was found against HIV-1 JRFL gp41. However, the signal was below our signal intensity cutoff level of 0.4 x 10 3 , and only visible at the highest antibody concentration (10 μg/mL) tested.

MPER specific antibodies
As shown in Fig 1, MPER antibody 4E10 reacted only minimally with HIV-1 SF162 gp140 in the gp140 cluster, and as expected no signal intensities were observed against the gp120 cluster (Fig 1). In contrast, mAb 2F5 did not interact with envelopes of the gp140 or the gp120 cluster. All tested MPER antibodies strongly interacted with HIV-1 JRFL gp41. Weak binding was observed against full-length MPER peptide 09129, and trimeric full-length MPER peptide 09128. Additionally, mAb 4E10 revealed strong signal intensity against MPER analog 09122. No binding was found against control peptide 08023 and C22-pT, which is in good agreement with the in solution capture assay shown in S2 Fig.
Cluster analysis of antibody specificity against gp140 and gp41 constructs corroborates distinct binding patterns To evaluate the microarray chip for specific antibody binding pattern, a hierarchical cluster analysis was performed. As shown in Fig 3, reactivity of 10 mAbs against 6 HIV antigens consisting of five multi-clade (A, B, C, D, F) gp140 molecules and a clade B gp41 molecule was compared and plotted into a heat map. Antibodies were probed on separate arrays using 8 serial half-log dilutions ranging from 10 μg/mL to 0.003 μg/mL for each mAb. To evaluate antibody-binding patterns, the highest antigen concentration of 1.0 mg/mL (1.0 ng per spot) was used. Although, we observed unspecific binding for some of the antibodies (e.g. F425 B4e8 against gp41 JRFL ) at the highest concentrations we found that binding does not impact the analysis, as the binding patterns of the lower antibody concentrations contribute the most. The results are summarized on the heat maps in Fig 3. Unbiased hierarchical clustering revealed a distinct pattern of reactivity for each of the 6 antigens (Fig 3A and 3B). The anti-gp41 antibody 5F3 was reactive against all tested antigens, followed by 240-D which exhibited a strong reactivity against clade A, B, and C constructs. In contrast, MPER specific mAbs, 4E10 and 2F5 reacted preferentially with gp41 and gp140 from clade B, and gp140 from clade D. Out of the six gp120 specific antibodies tested, the two V3 loop specific mAbs 447-52D and F425-B4e8 showed high cross clade reactivity, followed by glycan specific mAb 2G12, and CD4bs specific mAb VRC01. In sharp contrast, CD4bs antibodies b12 and VRC03 reacted predominantly with clade B. The reactivity of the mAbs was also clustered into distinct groups according to their published classification (Fig 3B). The CD4bs mAbs (VRC01, VRC03 and b12) clustered together and the V3 loop specific mAbs (447-52D and F425-B4e8) formed their own distinct cluster. The glycan specific mAb 2G12 came out as a singleton, whereas the gp41 specific mAbs clustered together with the MPER specific mAbs forming a distinct pair. This result is in good agreement with the published classification of each antibody and illustrates how chipbinding data, followed by hierarchical cluster analysis, may be applied for the identification and classification of unknown antibody specificities.

Microarray binding profile of mAbs correlate significantly with binding profile from ELISA experiments
To confirm the binding data obtained from protein microarray analysis, mAbs were further evaluated by serial dilutions, both on the chip as well as in a conventional ELISA setup. EC 50 values were calculated from both platforms and plotted into a 3-color scale heat map. As indicated in Fig 4, binding patterns were remarkably consistent. Next, EC 50 values from each group were further evaluated for significant correlations. Spearman's rank correlation coefficient (ρ)

Equilibrium dissociation constant (Kd) calculation reveals various binding affinities towards tested antigens
To further define binding affinities, Kd's of antibody-antigen interaction from binding curves (S3 Fig) were calculated, using a one-site binding algorithm (GraphPad PRISM 5.0). For each mAb, the corrected Bmax (Bmax/mg antigen printed) and the affinity constant (Ka) ( Table 2) were determined (1/Kd). All three classes of gp120 specific antibodies displayed varying binding affinities towards gp140 antigens while showing no detectable binding to HIV-1 JRFL gp41 ( Table 2). Besides the lack of binding to gp41, the CD4bs antibodies b12 and VRC03 showed low or no binding affinities towards the gp140 antigens printed, with the exception of HIV-1 SF162 gp140 for which they exhibited weak affinity ( Table 2). The glycan specific mAb 2G12 exhibited intermediate binding affinities toward all gp140 antigens except gp41. On the contrary, the gp41 and MPER specific mAbs analyzed revealed high affinities for both gp41 and gp140 antigens from clades A, B, C, and D, but did not show any binding affinity against HIV-1 BR29 gp140 from clade F ( Table 2).

Probing the chip with HIV-1 infected human plasma yields patient specific binding patterns
We probed and analyzed plasma samples from 28 subjects with chronic HIV-1 infection on suppressive ART to investigate potential clinical applications of the microarray chip. Plasma samples were diluted 1/100 prior to probing. The optimal antigen concentration used for statistical analysis and heat map generation was 0.01 mg/mL or 0.01 ng per spot. As shown in Fig 5, plasma from HIV infected subjects demonstrated different antibody binding intensities across the tested envelope proteins. Twelve subjects revealed intermediate median signal intensities throughout the gp140 cluster, thus indicating a high cross-clade specificity for these antigens. This trend was similar to the positive control HIVIG (pool of purified HIV-1 infected donor IgG). Four additional subjects (7, 9, 15, and 17) showed marginally lower median signal intensities. The remaining twelve subjects revealed substantially lower (weak) median signal intensities. Based on the median signal intensities within each subtype of the gp140 cluster, patients were predominantly reactive to clade B followed by clades D, A, C, and F. A similar trend in reactivity was found for the gp120 cluster where subjects mainly reacted with clade B followed by clade C (HIV-1 CN54 gp120). Minimal reactivity was observed with HIV-1 96ZM651 gp120 and HIV-1 93TH975 gp120. Binding pattern data of the chip analysis provide clear support for the notion that most of the subjects were originally infected with clade B HIV-1. All patients showed intermediate to weak reactivity against HIV-1 JRFL gp41 suggesting utmost specificity at the tested concentration. Additional binding was found against MPER peptide 09129 and 09122, although at rather weak signal intensities. Only subject 2 recognized peptides C22-pT and 09128. None of the subjects reacted with the scrambled control peptide 08023 indicating that antibody response against the other MPER analogs is highly specific. The polyclonal antibody control IVIG (pool of purified IgG from healthy donors) revealed minor antigen interaction within the gp120 cluster. The signal intensities of IVIG served as a cutoff level for positive binding of plasma samples (Fig 5). Patient sera were evaluated for their ability to neutralize a relatively sensitive Tier 1 isolate HIV-1 SF162 (Fig 6A). When we correlated the binding pattern of HIV-1 SF162 gp140 and HIV-1 BaL gp120 we found a significant correlation between antibody binding signals and IC 50 values (HIV-1 SF162 gp140 and HIV-1 SF162; P = 0.0021, ρ value -0.5556; HIV-1 BaL HIV-1 Microarray Analysis gp120 and HIV-1 SF162 : P = 0.0003, ρ value -0.6322) (Fig 6B and 6C). Noteworthy, when we analyzed at the median signal intensity across the gp140 cluster (clades A, B, C, D, and F) the correlation between binding and neutralization was even more pronounced (P < 0.0001, ρ value -0.6760), indicating enhanced neutralization by cross clade specificity (Fig 6D).

Discussion
During the last decades, a major focus in HIV-1 research has been directed towards the isolation of broadly neutralizing antibodies [34,[36][37][38][39]. Recent improvements in antibody screening platforms and antibody isolation techniques from HIV-1 infected patients have led to the discovery of many potent and broadly neutralizing antibodies [40][41][42]. Binding specificity assays and standardized neutralization assays [33,34] are used to understand antibody breadth and activity. Furthermore, growth in this research activity, demands development of assays that maximize the use of sampled material, while reducing the number of samples for multi-parameter measurements. In fact, the National Institute of Allergy and Infectious Diseases provides an AIDS/HIV specimen repository, where particular scrutiny is placed on requests to use limited human specimens with clinical background from infants, paired mother-to-child transmission cases, newly seroconverted individuals, and long-term non-progressors. In order to streamline HIV-1 vaccine discovery and simultaneously evaluate binding against many different envelope proteins, more efficient and high throughput methods are needed. Here, we developed a novel high throughput technology for characterizing antibodies. We printed selected HIV-1 envelope proteins of different clades onto a protein microarray chip. Using this technique we were able to evaluate five recombinant gp140 envelope proteins, four HIV-1 Microarray Analysis gp120 envelope proteins, one gp41 protein, and a total of 5 MPER peptides, two of which were in a putative trimer stalk conformation (09128 and C22-pT) [43]. The recombinant gp140 proteins were primarily produced in CHO cells, except HIV-1 SF162 gp140, which was produced in HEK293 cells. For the expression of the gp120 constructs HEK293 cells and insect cells were used. No information could be found for the expression of HIV-1 96ZM651 gp120. Therefore, the variety of expression hosts could in some cases explain differences in binding, due to possible differences in glycosylation patterns or other differences in the production of proteins in different cells [44]. For example, the glycan dependent mAb 2G12 exhibited a strong signal (i.e. 40 times higher) against the HIV-1 CN54 gp120 construct expressed in insect cells, compared to the HIV-1 CN54 gp140 construct expressed in CHO cells. In contrast, VRC01 or 447-52D revealed only minor differences in signal intensity between the two constructs, probably indicating differences in the glycosylation pattern of the different host systems. However, the gp140 cluster should allow for a robust cross clade comparison, as four out of five constructs were expressed in CHO cells.
Our chip binding analysis revealed extensive cross clade reactivity of V3 loop specific mAbs 447-52D and F425 B4e8 against envelope proteins of the gp140 and the gp120 cluster. Both antibodies are placed among the more broadly neutralizing V3 antibodies and share a similar binding epitope [45]. In the case of mAb 447-52D, it is the well-conserved GPGR motif at the tip of the V3 loop of clade B isolates, thus increasing the neutralization and binding breadth compared to other V3 loop specific antibodies [46]. In contrast, more recently discovered broadly neutralizing variable loop specific mAbs, like the V1/V2 specific mAb PG9, require highly complex glycan dependent quaternary epitopes [47,48]. A moderate to high degree of cross reactivity against gp140 and gp120 envelopes was also visible with the potent and broadly neutralizing mAb VRC01, within the group of CD4bs antibodies. Interestingly, next to VRC01, the rather poorly neutralizing CD4bs mAb F105 exhibited a high degree of binding specificity against the gp140 cluster. This observation is in good agreement with other studies showing high reactivity of mAb F105 against a broad range of viruses despite having poor neutralization capability [49,50].
Interestingly, some of the tested gp120 specific mAbs revealed poly-reactivity against gp41, including mAbs F425 B4e8 and F105 (Fig 1). Signal intensities of the latter mAbs measured on the chip followed a dilution curve. However, the signals turned into noise after the mAbs reached a concentration of 1.0 μg/mL. In sharp contrast, mAb 2G12 only exhibited minimal signal at the highest antibody concentration. A similar trend was found with the gp41 specific antibody 240-D, which revealed reactivity towards HIV-1 BaL gp120. Interestingly, we did not observe poly-reactivity of F105 and 2G12 against HIV-1 JRFL gp41 in our ELISA approach. However, mAb F425 B4e8 showed a slightly elevated OD 450 at the highest antibody concentration (10 μg/mL) tested. It is important to note that the protein density on the ELISA is about 15 times less than protein density on the microarray chip, thus explaining some of the poly-reactivity observed. MPER specific mAbs 2F5 and 4E10 reacted predominantly with gp41 and MPER analogs and with minor extent to HIV-1 SF162 gp140.
MPER analog C22-pT was proposed to be in a prefusion state where critical residues for 2F5 and 4E10 binding are buried within the interface of the helices [25]. This was in good agreement with our in-solution-binding assay where 2F5 and 4E10 minimally reacted with the C22-pT wild type peptide. However, when we introduced a N674D mutation into C22-pT (09128), binding was substantially increased particularly for mAb 4E10 (more than 80 fold enhancement in IC 50 compared to control peptide 94.1). This could be due to enhanced exposure of critical 2F5 and 4E10 contact residues within the trimeric stalk. Although our experimental setup mostly enriches for enhanced 2F5 and 4E10 binding by down selecting those peptides that bind known MPER antibodies, it has to be pointed out that this selection may not be ideal if the method is to be used to capture binding by unknown specificities, where a larger number of antigens may add more relevant information.
To further assess the role of the MPER prefusion state on antibody recognition and subsequent neutralization, a D674N mutation identical to the C22-pT protein was introduced into HIV-1 isolates JRFL and JR-2 to render the viruses more resistant to neutralization. Pseudotyped viruses (wild-type and D674N mutants) were tested against several gp41 and gp120 specific antibodies in a single round infectivity assay. The improved neutralization observed with 2F5 and 4E10 (S2 Table) was a rather unexpected result, in light of the observations of Liu and colleagues [25]. However, our results could be explained by conformational changes in the MPER due to increased flexibility at the hinge region and thus better exposure to MPER specific antibodies in the context of the Env trimer. In contrast, gp41 specific antibody D5 and the gp120 specific antibodies 2G12 and b12 were minimally affected by the point mutation. A negative effect of the D674N mutation was observed with mAb Z13e1, which completely lost neutralization capacity at the highest concentration tested. This is in good agreement with a previous study where D674 has been identified as one of the key residues for Z13e1 binding [21].
Unbiased hierarchical cluster analysis revealed important information about antibody binding characteristics including: 1) antigen specificity, 2) clade specificity, and 3) binding affinity. Moreover, antibodies were grouped into distinct clusters according to their epitope specificity (e.g. gp41 specific antibodies and gp120 specific antibodies), which in our case represents an additional quality control and feature of the microarray chip. This approach applied to unbiased screening of mAbs can be easily adapted in a high throughput, automated manner to efficiently compare panels of well-characterized mAbs against dozens of antigens and rapidly classify uncharacterized mAbs according to antigen binding patterns. In addition, it will rapidly generate actionable data on large numbers of mAbs while consuming small amounts of valuable reagents.
When we first compared our microarray chip data with data obtained through a direct binding ELISA approach, we found a high conformity in binding intensities and OD 450 values. To compare the two platforms more efficiently we calculated a correlation coefficient between the determined EC 50 values generated from binding curves. We found a strong correlation between the data of the microarray and the ELISA approach indicating that our chip analysis is a valuable tool for antibody binding characterization.
However, in some cases low correlation exists between the ELISA and chip generated EC 50 values. For example 447-52D reacted with clade C envelope HIV-1 96ZM651 gp120 on the ELISA platform but an EC 50 could not be calculated on the microarray platform. We believe this was most likely due to a combination of antigen accessibility and protein stability over time. We found a higher affinity for HIV-1 96ZM651 with mAb 447-52D than with VRC01. However, signal intensities of VRC01 against HIV-1 96ZM651 on the chip were several folds higher than with mAb 447-52D (data not shown). We observed up to 40-fold variation in EC 50 between the two platforms. Possible factors that could influence the goodness of curve fit and thus EC 50 calculations were: 1) the antigen concentration spotted onto the chip, 2) the access to antibody specific epitopes, 3) the antibody dilution range, and 4) the overall stability and quality of printed antigens.
Nevertheless, we believe that optimization of this novel technology will yield an excellent tool for comprehensive antibody profiling and characterization in a high throughput manner. For example we envision a high throughput-screening platform for the determination of antibody epitopes by printing a comprehensive set of gp120 mutants or a full set of overlapping peptides covering envelope or distinct regions thereof. To overcome the issues with quaternary structure dependent antibodies such as PG9 or PGT151, and to differentiate between neutralizing and non-neutralizing antibodies, SOSIP trimers [51] could be mutated and printed in a next step. This assay could then be used to determine binding profiles in addition to the characterization of broadly neutralizing antibodies typically done in micro-neutralization assays using very limited quantities of micro-culture wells. Additionally, this technology could be applied to screen clinical samples (plasma and serum) and further our understanding of the evolving humoral immune response among HIV-1 infected persons at different stages of disease and antiretroviral treatment as demonstrated by our results. Last but not least, our chip analysis could be used to evaluate HIV-1 specific antibody profiles among vaccinated individuals from clinical trials, or from animal immunization studies, in a fast and accurate manner. This could be very useful in combination with SOSIP trimers as they allow differentiating between neutralizing and non-neutralizing immune responses. Binding curves of HIV-1 gp120-specific mAbs F425-B4e8, 2G12, VRC01, and gp41-specific mAb 4E10 were generated from two independent measurements. Different concentrations of antibody (i.e. half log dilutions starting at 10 μg/mL) were titrated on microarray chips with printed HIV-1 envelope antigens. The dissociation constant, Kd for each antibody-antigen interaction was determined using GraphPad PRISM 5.0 one-site binding algorithm and the corresponding affinity constants (1/Kd) calculated. (TIF) S1 Table. Clade specificity and expression hosts of all recombinant HIV-1 envelope proteins used for chip generation. All proteins except gp41 were obtained through the NIH AIDS Reagent Program. Clade specificity and expression host systems are indicated. HIV-1 SF162 gp140 was expressed as a trimer. (DOCX) S2 Table. Neutralization capacity (IC 50 in μg/mL) of gp41 and gp120 specific antibodies after site directed mutagenesis at MPER amino acid position 674 (D to N mutation). MAb Den 3 was used as a negative control. Neutralization assays were performed at least twice. (DOCX)