Quantification of the Epitope Diversity of HIV-1-Specific Binding Antibodies by Peptide Microarrays for Global HIV-1 Vaccine Development

Abstract

An effective vaccine against human immunodeficiency virus type 1 (HIV-1) will have to provide protection against a vast array of different HIV-1 strains. Current methods to measure HIV-1-specific binding antibodies following immunization typically focus on determining the magnitude of antibody responses, but the epitope diversity of antibody responses has remained largely unexplored. Here we describe the development of a global HIV-1 peptide microarray that contains 6,564 peptides from across the HIV-1 proteome and covers the majority of HIV-1 sequences in the Los Alamos National Laboratory global HIV-1 sequence database. Using this microarray, we quantified the magnitude, breadth, and depth of IgG binding to linear HIV-1 sequences in HIV-1-infected humans and HIV-1-vaccinated humans, rhesus monkeys and guinea pigs. The microarray measured potentially important differences in antibody epitope diversity, particularly regarding the depth of epitope variants recognized at each binding site. Our data suggest that the global HIV-1 peptide microarray may be a useful tool for both preclinical and clinical HIV-1 research.

1. Introduction

One of the fundamental challenges in HIV-1 vaccine development is the tremendous diversity of HIV-1 strains worldwide (Korber et al., 2001; Gaschen et al., 2002; Taylor et al., 2008; Barouch and Korber, 2009; Korber et al., 2009; Walker et al., 2011; Ndung'u and Weiss, 2012; Picker et al., 2012; Stephenson and Barouch, 2013). Globally, there are more than a dozen HIV-1 subtypes and hundreds of circulating HIV-1 recombinant forms (CRFs), and between-subtype variation can be as large as 35% (Hemelaar et al., 2006; Taylor et al., 2008; Ndung’u and Weiss, 2012). Several HIV-1 vaccine candidates under development aim to overcome the challenge of HIV-1 genetic diversity either through the choice of HIV-1 antigen sequence or the method of antigen delivery (Stephenson and Barouch, 2013). However, most tools used to assess the immunogenicity of these vaccines focus on measuring the magnitude of HIV-1-specific antibody responses, rather than the epitope diversity and specificity of these responses.

Peptide microarrays are a potential tool for the measurement of antibody diversity against linear epitopes in HIV-1 vaccine studies. This platform has been utilized to characterize antibody binding to linear sequences in multiple fields, including HIV-1 vaccine research (Nahtman et al., 2007; Cerecedo et al., 2008; Gaseitsiwe et al., 2008; Lorenz et al., 2009; Tomaras et al., 2011; Haynes et al., 2012). HIV-1-specific microarrays to date, however, have not included extensive coverage of variable sequences (Karasavvas et al., 2012; Gottardo et al., 2013; Imholte et al., 2013). Here we describe the development of a global HIV-1 peptide microarray that includes 6,564 overlapping linear HIV-1 peptides covering most common HIV-1 variants in the HIV-1 sequence database at Los Alamos National Laboratory (LANL). This microarray includes 6,564 peptides, including an average of 7 peptide variants for each 15 amino acid position in HIV-1 Env, Gag, Nef, Pol, Rev, Tat, and Vif, with up to 95 peptide variants per location within the most variable regions of HIV-1 Env. This epitope diversity on the microarray allows for more precise measurements of the magnitude, breadth and depth of HIV-1-specific binding IgG responses.

2. Methods

2.1.Generation of peptide library

In collaboration with JPT Peptide Technologies (Berlin, Germany), we designed a library of HIV-1 linear peptides that provided optimal coverage of HIV-1 global sequence diversity. We began by downloading the sequence alignment for HIV-1 genes ENV, GAG, NEF, POL, REV, TAT, and VIF from the website of the LANL HIV-1 sequence database (Theoretical Biology and Biophysics, 2009) using the following settings: Alignment type: Web Alignment (all complete sequences); Year: 2009; Region: Pre-defined region of the genome; Subtype: All M Group (A-K + Recombinants); DNA/Protein: Protein; Format: FASTA. Full length proteins of gp120, gp41, p17, p24, Tat, and Nef were used, as were the immunogenic fragments of p2p7p1p6, protease, reverse transcriptase, integrase, Vif, and Ref as published by LANL (Theoretical Biology and Biophysics, 2010) (Table 1).

Table 1.

Composition of global HIV-1 peptide microarray.

GENE	Position HXB2 Start	Position HXB2 End	Protein	# of Input Sequences	# Peptides on Array	Global Coverage [%]
ENV	1	511	gp160_1	2248	2672	50.2
	512	856	gp160_2	2248	1210	65.5
GAG	1	132	p17	3578	517	58.4
	133	363	p24	3578	264	86.2
	364	379	p2p7p1p6_1	3578	20	10.5
	380	394	p2p7p1p6_2	3578	18	17.9
	406	420	p2p7p1p6_3	3578	13	71.7
	428	450	p2p7p1p6_4	3577	50	68.7
NEF	1	205	Nef	2606	667	55.3
POL	57	71	Prot_1	1765	10	88.4
	97	104	Prot_2	1765	13	75.4
	217	231	RT_1	1765	4	90.5
	402	420	RT_2	1765	8	69.3
	447	461	RT_3	1765	13	51.4
	500	514	RT_4	1765	14	86.1
	529	543	RT_5	1765	22	84.9
	594	608	RT_6	1765	17	52
	673	707	RT_7	1765	51	41.5
	716	774	Int_1	1765	98	40
	865	879	Int_2	1765	16	70.3
	924	990	Int_3	1765	68	81.3
REV	3	17	Rev_1	1370	17	26.1
	32	50	Rev_2	1370	14	78.3
	70	116	Rev_3	1370	187	42.6
TAT	1	100	Tat	1286	296	55.5
VIF	33	47	Vif_1	1643	19	9.8
	174	192	Vif_2	1642	38	38.7

From the global sequence database, we selected the individual sequences to be used as peptides that would provide optimal coverage of sequence diversity using the program package MosaicVaccines.1.2.11 from LANL (ftp://ftp-t10.lanl.gov/pub/btk/mosaic/) (Fischer et al., 2007a; Thurmond et al., 2008a). Parameters for the generation of MOSAIC sequences were –s 20 –d=true –T 20 –p 100. Sequence manipulation and processing were performed in R 2.11.1 (http://www.r-project.org/) using the package Biostrings (http:www.bioconductor.org/packages/2.2/bioc/html/Biostrings.html) or using bespoke scripts in python (http:www.python.org/).

Since our goal was to cover the seven most frequent clades (A, B, C, D, G, CRF01_AE, and CRF02_AG), we used a stepwise approach to generating an optimal sequence cocktail. As a first step, the MOSAIC program was used to identify a sequence for each gene product or fragment from each of the 7 most frequent clades, and the resulting 7 sequences were merged into one cocktail. Secondly, we identified 13 additional sequences which showed best coverage without consideration of the clade. These two sequence cocktails were merged into one cocktail and evaluated for gain of coverage for each sequence. All sequences which did not gain more than 0.75% of coverage were removed from the cocktail. Thirdly, MOSAIC sequences were generated for each gene product or fragment, respectively (Fischer et al., 2007b; Thurmond et al., 2008b). For the MOSAIC runs the sequence cocktails generated in the previous step were used as fixed sequences. The resulting cocktails were evaluated in terms of coverage gain. All MOSAIC cocktails which gained less than 1% coverage were removed, and a maximum of 2 MOSAIC sequences was kept in the final cocktail. Figure 1A displays the relationship between the increasing size of the cocktail and the plateauing increase in coverage for gp120.

Figure 1. Generation of a global HIV-1 peptide library.

(A) Cocktail size of Env gp120 and Gag p24 sequences is plotted against percent coverage of global sequence variability. The cocktails consist of the single best covering sequences from each of the 7 frequent clades (left of vertical line), and the 13 best covering sequences as calculated from the complete HIV-1 sequence library and MOSAIC sequences (right of vertical line). (B) Clade- and circulating recombinant form (CRF)-specific peptide sets are shown. Only sets that have >300 peptides are shown out of a total of 135 sets. If a peptide sequence was found in multiple clades/CRFs, then it was counted in multiple sets.

Once we had generated a cocktail of sequences with optimal global coverage, we then generated a library of peptides where all sequences within the cocktail were covered at a minimal number of peptides. One of the sequences was used as a template sequence and processed into 15 amino acid peptides overlapping by 11 amino acids. All other sequences within the cocktail were fragmented into peptide scans of 15 amino acid peptides overlapping by 14 amino acids. Of note, this length of peptide (15 amino acids) covers 83% of known linear antibody epitopes in the LANL immunology database, including the median length of epitopes (11 amino acids) (Theoretical Biology and Biophysics, 2014). Scan-peptides were then aligned onto the scan-peptides of the template. The resulting 5,141 peptides covered all template sequences completely. For ENV, we performed one additional step to assure that every region of the protein was represented on the microarray by adding additional MOSAIC sequences that our group generated in the course of HIV-1 vaccine design (Barouch et al., 2010; Barouch et al., 2013). To overcome the bias of peptides towards conserved regions of the protein, we also included an additional 1,004 peptides from the variable loops V2 and V3 of gp120 in the library.

The final library consisted of 6,654 peptides from 135 different clades or CRFs. CRFs are circulating related variants that have different regions associated with the different major HIV-1 clades (Robertson et al., 2000); for example CRF_02 has sections of clade A and clade G, and is common in West Africa. (For a comprehensive listing of CRF’s see http://www.hiv.lanl.gov/content/sequence/HIV/CRFs/CRFs.html). By capturing A clade diversity, we capture some of the diversity found in the regions of CRF_02 that are A-like, but CRF_02 started with a recombinant founder virus decades ago, and has been spreading and diversifying as a separate lineage (Zhang et al., 2010), and so it will have its own distinctive evolutionary trajectory. Each CRF represents its own lineage, thus by including the CRFs in diversity considerations, not just major clades, we take a more comprehensive and realistic view of global diversity than by a more narrow examination of major clades. Figure 1B shows how many peptides were included in each clade- or CRF-specific peptide set (only sets that contain >300 peptides are shown). If a peptide sequence was found in multiple clades/CRFs, then it was counted in multiple sets. Peptides sets from the seven most frequent clades (A, B, C, D, G, CRF_01, and CRF_02) include >500 peptides each.

2.2 Microarray printing

PepStar peptide microarrays were produced by JPT Peptide Technologies GmbH (Berlin, Germany). All peptides were synthesized on cellulose membranes using SPOT synthesis technology. Subsequent to a final synthesis step attaching a reactivity tag to each peptide’s N-terminus, the side chains were deprotected and the solid-phase bound peptides were transferred into 96-well microtiter filtration plates (Millipore, Bedford, MA, USA). For cleaving the peptides from the cellulose membrane the individual spots were treated with aqueous triethylamine [2.5% (v/v)]. The peptide-containing solution was centrifuge-filtered into daughter plates and the solvent was removed by evaporation under reduced pressure. Quality control measurements using LCMS were performed on random samples of the final library. For transferring the peptides to 384 well plates, the dry peptide derivatives were dissolved in 35 µl of printing buffer and reformatted with automated liquid handling systems. Peptide microarrays were produced using a non-contact high performance microarray printer on epoxy-modified slides (PolyAn; Germany). All peptides and controls were deposited in three identical sub-arrays, enabling analysis of assay homogeneity and reliability of the results. Peptide microarrays were scanned after printing process and statistical values were generated for identification and quality control of each individual spot. Subsequently, peptide microarray surfaces were deactivated using appropriate quenching solutions, washed with water and dried using microarray centrifuges. Resulting peptide microarrays were stored at 4°C until use.

2.2. Sample selection

Thirty-six (36) serum or plasma samples were obtained from previously performed studies in the Barouch laboratory and were selected to represent a spectrum of potential preclinical and clinical uses for the microarray. Samples included (a) plasma from 5 HIV-1-infected individuals from North America with high HIV-1 viral load and 5 HIV-uninfected controls, (b) serum from 5 human subjects vaccinated with one dose of adenovirus 26 (Ad26) vaccine expressing clade A HIV-1 Env and 5 naive human controls, (c) serum from 5 rhesus monkeys vaccinated with 6 doses of clade C HIV-1 Env protein and 2 naïve monkey controls, and (d) serum from 5 guinea pigs vaccinated with 6 doses of clade C HIV-1 Env protein and 3 naïve guinea pig controls.

2.3. Microarray incubation and immunolabeling

Microarray slides were incubated with serum or plasma using the manual method, essentially as described (Masch et al., 2010). Serum or plasma was diluted 1:200 in SuperBlock T20 (TBS) Blocking Buffer (Thermo Scientific). Slides were placed in the individual chambers of a Sarstedt Quadriperm Dish and incubated in 4 mL of diluted serum/plasma for 1 hr at 30° C. Slides were then washed with 5 mL of TBS-Buffer + 0.1%Tween20 for 3 minutes on a shaker at room temperature for 5 washes. Next, slides were incubated with Alexa Fluor 647-conjugated AffiniPure Mouse Anti-Human IgG (H+L) (Jackson ImmunoResearch Laboratories) for human or monkey samples for 1 hr in the dark on a shaker at room temperature. Alexa Fluor 647-conjugated AffiniPure Goat Anti-Guinea Pig IgG (H+L) (Jackson ImmunoResearch Laboratories) was used for guinea pig samples. Slides were then washed 5 times with TBS-Buffer with 0.1%Tween20, and 5 times with deionized water. To dry, slides were placed in a 50 mL conical and spun at 1500 rpm for 5 minutes. Of note, all batches of slides were run in parallel with a control slide that is incubated with secondary antibody alone.

2.3 Microarray image analysis

Slides were scanned with a GenePix 4300A scanner (Molecular Devices), using 635 nm and 532 nm lasers at 500 PMT and 100 Power settings. Images were saved as TIF files. The fluorescent intensity for each feature (peptide spot) was calculated using GenePix Pro 7 software and GenePix Array List (GAL) file, a text file with specific information about the location, size, and name of each feature on the slide. This analysis created a GenePix Results (GPR) file. We then calculated the mean fluorescent intensity across the triplicate sub-arrays (SAs) for each feature; if the coefficient of variation was greater than 0.5, then the mean of the two closest values was used. These calculations were performed with a custom-designed R script “MakeDat_V04” (available as Appendix 1) and R software package 2.15.2. Data was saved as a comma-delimited DAT file usable by Excel (Microsoft). MakeDat_V04 also created scatterplots of the correlation between the feature fluorescent intensities of sub-array 1 and 2, sub-array 2 and 3, and sub-array 1 and 3 as a measure of assay quality (Figure 3).

Figure 3. Quality analysis of raw microarray data.

The signal distribution of a representative microarray is displayed, as well as the correlation between slide sub-arrays. This data is following incubation with serum from an HIV-1-infected subject. SA, subarray; Rq, R squared; Sl, slope; In, intersection with the y axis; MC2, two closest values between the 3 sub-arrays; MC2 1 and MC2 2, each component of MC2.

The threshold value used to define a minimum positive fluorescent intensity was calculated for each slide using the computational tool rapmad (Robust Alignment of Peptide MicroArray Data, available for free at http://tron-mainz.de/tron-facilities/computational-medicine/compmed) and the custom-designed R script “calc_THRES_KS_130711_V02” (available as Appendix 2). This analysis is based on the method as previously described (Renard et al., 2011) and distinguishes those peptide features that carry a signal from those features that only display noise. Data from each individual slide was combined with data from the control slide to create two distributions of data (noise and signal). We then calculated four potential threshold values for positivity with increasing levels of stringency: the false discovery rate cutoff (FDR cutoff), the point at which the chance that signal is noise is P<0.01, 5 standard deviations above the mean of the noise distribution (SD.noise*5), and the point at which the chance that signal is noise is very low at P<10⁻¹⁶.

2.4 Analysis of magnitude, breadth, and depth of antibody binding

The raw magnitude, or fluorescent intensity, of antibody binding to individual peptides (averaged over the 3 sub-arrays as described above) was sorted and categorized by (1) HIV-1 protein, (2) amino acid start position as aligned to HXB2 HIV-1 reference strain, and (3) HIV-1 clade or CRF within which the peptide sequence can be found. This sorting was performed using the custom-designed R script “Table_select_V01” (available as Appendix 3). To correct for any direct binding of the secondary antibody to linear peptides, the fluorescent intensity of antibody binding measured on the control slide was subtracted from the fluorescent intensity of antibody binding measured on the sample slide. Finally, all corrected fluorescent intensities were compared to the calculated threshold for positivity, and all values above the threshold were considered positive (with the rest of the values changed to “0” and considered negative). For these studies, we chose the threshold SD.noise*5.

To calculate the breadth of antibody binding, we evaluated the number of positive peptides for each sample and aligned the peptide sequences to eliminate overlap. If any positive peptide sequences shared 5 or more contiguous amino acids, we assumed that the peptides were recognized by the same antigen-binding site on a single antibody; these overlapping sequences were conservatively defined as a single positive “binding site.” If the first and last overlapping peptide in a string of overlapping peptides shared 4 or less amino acids, we assumed that the peptides were recognized by a minimum of two antibody sites (on either two antibodies or the same antibody). This approach to calculating antibody breadth is based on established methods to calculate T cell breadth, essentially as described in (Stephenson et al., 2012). The primary difference is that the overlapping region for T cells is usually 9 or more amino acids, reflecting the structure of CD4/CD8 T cell binding pockets. For antibodies, the antigen binding-site can range in length, and for conformational epitopes may not be contiguous. For this analysis, 5 amino acids was chosen as the minimum overlap region for antibody binding based on the fact that, apart from a handful of 3 and 4 AA sequences, 5 AA is the minimum length reported for human antibody binding to linear HIV-1 sequences as per the LANL HIV Immunology Database (Theoretical Biology and Biophysics, 2014). Conformational epitopes cannot be directly assessed with linear peptide microarray.

To calculate the depth of antibody responses, we evaluated the overlapping sequences of each binding site and determined the number of unique sequence variations of the binding site that were present. We then calculated the average number of variations/binding site for each sample. We also determined the relative frequency of clade or CRF-specific antibody responses. To do this we first defined distinct clade or CRF peptide ‘sets’ that included any peptide whose sequence had been identified in that clade or CRF (see Figure 1B). If a sequence could be found across multiple clades, it was included in multiple sets. We then calculated the percent of positive peptides within each set to provide a relative measure of clade- or CRF-specific antibody responses that could be comparable across sets of different sizes. To maximize our ability to detect differences in clade- or CRF-specific antibody responses, we restricted this analysis to the variable regions V1V2 and V3 of gp120.

3. Results and Discussion

3.1 Evaluation of coverage of global HIV-1 sequence diversity

In designing this microarray, our goal was to develop a tool to measure the diversity of HIV-1-specific antibody binding to linear HIV-1 epitopes from global sequences. To determine how well the peptide library represented global HIV-1 sequence diversity, we analyzed coverage using the program package MosaicVaccines.1.2.11 as described above. We found that the peptide library covered the majority of sequences in the Los Alamos National Database (Table 1), including gp120 (50.2 %), gp41 (65.5 %), Gag p17 (58.4 %), and Gag p24 (86.2 %). Of note, for some protein regions a small group of 15-mer peptides sufficed to span a reported antibody binding site, but because the site was of high sequence diversity with no conserved sequences, the observed coverage was low (e.g. VIF_1 with 9% coverage reported).

We also evaluated the coverage of gp120 sequences from clades A, B, C, D, G, CRF01_AE, CRF02_AG, and a summary population of all other clades (Figure 2). This analysis demonstrated that for each clade- or CRF-specific sequence, 50% of the sequence (on average) was covered by peptides on the microarray. As expected, in the variable regions of the HIV-1 proteome lower coverage was achieved, as for the variable loops in ENV V1/V2 (HXB2 131–196) or V4 (HXB2 385–418). However, the microarray reached a maximum of 95 peptide variants for each location within the most variable regions of HIV-1 Env, and an average of 7 peptide variants for each location on HIV-1 Env, Gag, Nef, Pol, Rev, Tat, and Vif.

Figure 2. Coverage of HIV-1 peptide library for gp120 by clade.

Top row: A, C, G, CRF02_AG; bottom row: B, D, CRF01_AE, and all other clades. The X-direction of each plot represents the sequence of gp120. Sequences segments included on the microarray are depicted in red. In the Y-direction all sequences for the respective clade from the alignment HIV1_ALL_2009_ENV_PRO.fasta are shown (total 2248). The average coverage (horizontally) for each used HIV-1 sequence is 50%. The evaluation of coverage was performed presuming the same length of all sequences for one protein or fragment within a given clade.

The diversity of linear peptides on the global HIV-1 microarray described here is in contrast to the composition of the predominant HIV-1 peptide microarray previously reported in the literature (Tomaras et al., 2011; Karasavvas et al., 2012; Gottardo et al., 2013; Imholte et al., 2013). This previous microarray includes gp160 subtype consensus sequences from six HIV-1 group M subtypes (A, B, C, D, CRF_01 and CRF_02) and a consensus group M gp160, Con-S. In contrast to the global microarray reported here, this previous microarray contains less than a quarter of the number of peptides (1,423 vs. 6,564), excludes variable sequences by design, and does not include any non-Env proteins, making it potentially less optimal for quantifying HIV-1 antibody epitope diversity.

3.2 Evaluation of microarray assay quality and determination of threshold values for positivity

Given the density of peptides on the microarray (19,692 peptides over 3 triplicate sub-arrays), we designed a program to evaluate the quality of raw microarray data following sample incubation and immunolabeling, as described above. Figure 3 demonstrates representative results of this analysis following microarray incubation with plasma from an HIV-1-infected subject. As shown in this example, the program provides a snapshot of how well the results from each sub-array correlate with each other; in this case the correlation ranged from R²=0.93 to 0.96. We also designed a program to determine a threshold value above which a signal can be considered “positive” (Renard et al., 2011). Figure 4 demonstrates representative results of this analysis when the microarray was incubated with plasma from an HIV-1-infected subject. By providing four potential threshold values with varying stringency, the program allows the user to decide whether his or her analysis will have greater sensitivity or specificity in detecting antibody binding.

Figure 4. Determination of threshold values for positivity.

The signal and noise distribution of data from a sample microarray is displayed, as well as the correlation between slide sub-arrays. Microarray data is following incubation with plasma from an HIV-1-infected subject. FDR, false discovery rate; SD, standard deviation.

3.3 Quantitating and visualizing antibody binding diversity

The goal of this project was to develop a method to both quantitate and visualize antibody binding patterns to diverse HIV-1 sequences for the purpose of HIV-1 vaccine and therapeutic research. To visualize binding patterns, one can plot the magnitude of peptide binding (MFI) by peptide location (starting amino acid position). For instance, Figure 5A demonstrates the gp140-specific binding pattern among HIV-1-infected subjects, where the average MFI per peptide is shown for the 5 subjects. In this example, peak MFI values were observed at the V3 region of gp120 and the CC loop region of gp41, with maximum values about 60,000 MFI, consistent with well-described immunodominant regions in HIV-1 infection (Goudsmit, 1988; Tomaras et al., 2008; Tomaras and Haynes, 2009; McMichael et al., 2010). Among HIV-uninfected controls, there were a handful of nonspecific positive peptides, but peak values did not rise above 4,500 MFI (Figure 5B). For comparison, Figure 5C shows the binding pattern among human subjects vaccinated with a single priming dose of Ad26-EnvA HIV-1 vaccine. Here peak binding values were observed to V1, V2 and V3 linear peptides, with maximum MFIs up to about 12,000. The lower MFI of vaccinees compared to HIV-1-infected subjects is expected given receipt of only one dose of vaccine without subsequent boosting, but were still above those observed in naïve controls (Figure 5D). Of note, sera from naïve controls demonstrated more nonspecific positive peptides than plasma from naïve controls.

Figure 5. IgG binding to linear HIV-1 gp140 peptides in HIV-1-infected subjects and vaccine recipients.

The mean fluorescent intensity (MFI) of positive HIV-1 gp140 peptides are plotted against the amino acid start position as aligned to HXB2 HIV-1 reference strain for (A) 5 HIV-1-infected human subjects with high viral load, (B) 5 HIV-uninfected plasma controls, (C) 5 human subjects following one dose of Ad26-EnvA HIV-1 vaccine, (D) 5 naïve human serum controls, (E) 5 rhesus monkeys following 6 doses of HIV-1 Clade C Env protein, (F) 2 naïve rhesus monkey serum controls, (G) 5 guinea pigs following 6 doses of HIV-1 Clade C Env protein, and (H) 3 naïve guinea pig serum controls. Symbols represent the average MFI from infected or vaccinated subjects (color squares) and controls (black circles); bars represent SEM.

As shown in Figure 5E–H, the peptide microarray can also be used to map antibody binding patterns in two animal models commonly used in HIV-1 vaccine research: rhesus macaques and guinea pigs (Nkolola et al., 2010; Barouch et al., 2012; Barouch et al., 2013; Nkolola et al., 2014). In both examples, animals were vaccinated with 6 serial doses of clade C HIV-1 protein and developed a similar binding pattern, with peak responses at V3. The higher MFIs among vaccinated animals compared to humans is likely due to the increased number of boosts received by the animals. Of note, naïve guinea pig samples demonstrated higher backgrounds than naïve human or monkey samples.

While maps of antibody binding can provide a useful tool to visualize binding patterns, they are less useful for the quantitative comparison of groups or HIV-1 regions. To provide such quantitative data, we calculated the average MFI of peptide binding sorted by region and HIV-1 protein (Figure 6A); magnitude can be compared across subjects or vaccine platforms as long as the dilution factor for the assay is kept constant, as was done in these experiments. As demonstrated in Figure 6A, the microarray can help characterize which regions of the HIV-1 envelope are preferentially targeted. For example, in HIV-1-infected subjects, V3-specific binding was significantly greater than to any other gp120 region (P<0.02 for all comparisons by t-test) and CC loop-specific binding was greater than to any other gp41 region (P<0.002 for all comparisons by t-test). In contrast, human vaccinees did not show a preference for V3 or CC loop responses, although the vaccine included these antigens.

Figure 6. Magnitude, breadth, and depth of HIV-1-specific antibody binding, as measured by global HIV-1 peptide microarray.

Antibody responses for four groups are are plotted by gp120 and gp41 region, including (A) the mean fluorescent intensity of positive peptides (magnitude), (B) the number of binding sites (breadth), and (C) the mean number of epitope variants at each binding site (depth). The four groups include HIV-1-infected subjects (blue), vaccinated human subjects (red), vaccinated rhesus monkeys (green), and vaccinated guinea pigs (purple). Bars represent SEM.

It is also useful to know whether HIV-1-specific antibodies are binding to a limited region of the HIV-1 envelope or if multiple areas are targeted. Figure 6B demonstrates the number of binding sites (“breadth”) by gp120 and gp41 region for our four groups of samples. Here, we can see that while the vaccinated human subjects had relatively low magnitude gp140 binding compared to HIV-1-infected subjects, there was no discernable difference in antibody breadth between the two groups. This ability to distinguish between magnitude and breadth is important in HIV-1 vaccine research. For example, if a particular vaccine candidate elicits low magnitude but broad antibody responses, then one might decide to change the vaccine vector or schedule to boost responses. On the other hand, if the vaccine candidate elicits high magnitude but narrow antibody responses, then one might decide to retain the same vector and schedule, but change the immunogen to broaden the specificity.

We also developed the microarray to measure the cross-clade binding of HIV-1-specific antibodies. Figure 6C demonstrates the mean number of epitope variants per binding site by gp120 and gp41 region for the four groups of samples. This analysis shows that the Ad26-EnvA vaccine elicited comparable depth of V4 binding as seen in the HIV-1-infected subjects (5 vs. 6 variants/binding site, P=NS by t-test), even though the vaccine elicited significantly lower magnitude of V4 binding (1,955 vs. 10,468 MFI, P=0.0031 by t-test). In addition, the depth of V2 binding among vaccinated guinea pigs could not be predicted by magnitude alone. For example, while HIV-1-infected humans and HIV-1-vaccinated guinea pigs had the same magnitude of V2-specific responses (5,998 vs. 7,770 MFI, P=NS by t-test), the vaccinated guinea pigs had significantly greater depth of V2-specific binding (7 vs. 20 variants/binding site, P=0.0161 by t-test). Despite substantial differences in the human and guinea pig studies, this example demonstrates how the microarray can discriminate between magnitude and depth of antibody responses. This information may be highly relevant to HIV-1 vaccine researchers who aim to design a global HIV-1 vaccine capable of blocking acquisition of diverse HIV-1 strains.

We also calculated the relative clade- or CRF-specific binding present for the three most frequent clades (A, B, and C). Figure 7 demonstrates the percent of each clade- or CRF-specific peptide set that was positive for the four groups within the variable regions V1V2 and V3. In Figure 7A, we can see that among vaccinated monkeys and guinea pigs, V1V2-specific responses were increased compared to the other cohorts, and that binding to clades A and C V1V2 peptides predominated, whereas clade B-specific binding was relatively low. This finding likely reflects the fact that both monkeys and guinea pigs received clade C Env immunogens. In contrast, in Figure 7B, we can see that among HIV-1-infected subjects, who had increased V3-specific responses, binding to clade B peptides predominated. This finding presumably reflects the fact that these subjects were from North America and were infected with clade B HIV-1. These data suggest that the microarray may not only be useful for measuring cross-clade immune responses following vaccination, but also may have an application in serotyping HIV-1-infected subjects. Further studies with larger numbers of HIV-1-infected subjects from different regions could test this hypothesis.

Figure 7. Cross-clade HIV-1 antibody binding as measured by global HIV-1 peptide microarray.

The percent of clade- or CRF-specific peptide sets that are positive for V1V2 and V3 regions of gp120. The four cohorts include HIV-1-infected subjects (blue), vaccinated human subjects (red), vaccinated rhesus monkeys (green), and vaccinated guinea pigs (purple). Bars represent SEM.

Finally, we also designed the microarray to assess HIV-1-specific binding across the HIV-1 proteome. In Figure 8A, we demonstrate the magnitude, breadth, and depth of HIV-1-specific binding to gp120, gp41, Gag, Nef, Pol, Rev, Tat, and Vif proteins among 5 HIV-1-infected human subjects. We observed that gp41 (which includes regions from the cytoplasmic tail) has the highest binding magnitude, followed by Gag. Figure 8B shows the antibody binding pattern for Gag among 5 HIV-1-infected subjects; peak values are noted within the p17 region, with very little Gag-specific binding among naïve controls (Figure 8C). Antibody binding to non-Env proteins may be relevant to evaluate vaccine potency and for certain non-neutralizing antibodies (Lewis, 2014). Some studies have also shown an association between anti-p24 responses and virologic control (Weber et al., 1987; Trkola et al., 2004). In addition, the microarray might be useful to assess vaccine-induced seroreactivity in the context of HIV-1 vaccine clinical trials.

Figure 8. IgG binding to linear HIV-1 peptides from across the HIV-1 proteome in HIV-1-infected subjects.

(A) The mean fluorescent intensity (MFI) of positive HIV-1 peptides (magnitude), number of binding sites (breadth), and mean variants per binding site (depth) are shown for multiple HIV-1 proteins for 5 HIV-1-infected human subjects. Bars represent SEM. In addition, the MFI of positive HIV-1 Gag peptides are plotted against the amino acid start position as aligned to HXB2 HIV-1 reference strain for 5 HIV-1-infected human subjects (B) and 5 HIV-uninfected controls (C). Bars represent SEM.

4. Conclusions

As more HIV-1 vaccine candidates progress into clinical trials, it is important to develop new tools to assess the epitope diversity of HIV-1-specific antibodies. Here we report the development of a global HIV-1 peptide microarray based on a library of 6,564 peptides covering the majority of sequences in the Los Alamos National Laboratory HIV-1 sequence database. This microarray provides a method to measure the magnitude, breadth, and depth of IgG binding to linear HIV-1 peptides, allowing for a more in depth analysis of antibody epitope diversity than is currently available. Such knowledge may contribute to improvements in HIV-1 vaccine design and development, or to a better understanding of immune responses to HIV-1 infection. The major limitations are that this assay does not measure conformational antibodies or antibody function. Nevertheless, when used in conjunction with other antibody assays, the microarray assays should prove useful for both preclinical and clinical HIV-1 research.

Appendix 1. MakeDat_V04 R Script

Note, all provided R scripts are available under the terms of the Free Software Foundation’s GNU General Public License, as outlined at http://www.r-project.org/.

PAR_dir <- ".\\"
PAR_val <- "F635.Mean"
PAR_filter <- FALSE
PAR_flags <- c("−100") # "−50", "−75", "−100"
PAR_array <- 1 # 1 or 21

PAR_multiwell <- matrix( c(1,2,7,8,13,14,
                                 3,4,9,10,15,16,
                                 5,6,11,12,17,18,
                                 19,20,25,26,31,32,
                                 21,22,27,28,33,34,
                                 23,24,29,30,35,36,
                                 37,38,43,44,49,50,
                                 39,40,45,46,51,52,
                                 41,42,47,48,53,54,
                                 55,56,61,62,67,68,
                                 57,58,63,64,69,70,
                                 59,60,65,66,71,72,
                                 73,74,79,80,85,86,
                                 75,76,81,82,87,88,
                                 77,78,83,84,89,90,
                                 91,92,97,98,103,104,
                                 93,94,99,100,105,106,
                                 95,96,101,102,107,108,
                                 109,110,115,116,121,122,
                                 111,112,117,118,123,124,
                                 113,114,119,120,125,126) ,ncol=21)

#################################################################################################
######################

readGPR = function(name) {
       headersize <- read.table(name, header=FALSE, skip=1, nrows=1)
  GPR <- read.table(name, header=TRUE, skip=as.numeric(headersize[1,1])+2, sep ="\t",
strip.white=TRUE, colClasses="character")
       GPR <- GPR[order(as.numeric(GPR$Column)),]
       GPR <- GPR[order(as.numeric(GPR$Row)),]
  GPR <- GPR[order(as.numeric(GPR$Block)),]
       return(GPR)
}

getMC2 = function(val1, val2, val3) {
       tmp <- cbind(rbind(abs(val1-val2),abs(val1-val3),abs(val2-
val3)),rbind(mean(c(val1,val2)), mean(c(val1,val3)),
mean(c(val2,val3))),rbind(val1,val1,val2),rbind(val2,val3,val3),rbind(1,1,2),rbind(2,3,3))
       colnames(tmp) <- c("Diff", "MC2", "MC2_1", "MC2_2", "MC2_SA1", "MC2_SA2")
       tmp <- tmp[order(tmp[,1]),]
       return(tmp[1,])
}

getMMC2 = function(val1, val2, val3) {
  val <- mean(c(val1, val2, val3))
  if(sd(c(val1, val2, val3))/val > 0.5){
    val <- getMC2(val1, val2, val3)["MC2"]
  }
  return(val)
}

filterDAT = function(dat) {
  tmp <- NULL
  exclude = c("AA","GGSGGGSDYKDDDDK","printingbuffer","empty")
  for(i in 1:nrow(dat)) {
    if( length(unlist(strsplit(dat[i,"Annotation"],"|",fixed=TRUE))) == 4 & (! (dat[i,"ID"] %in%
exclude)) ) {
       tmp <- c(tmp,TRUE)
    } else {
       tmp <- c(tmp,FALSE)
    }
  }
  return(dat[tmp,])
}

plotplus = function(x ,y ,name ,xl ,yl, MINVAL, MAXVAL) {
  xtmp <- subset(x, !(x==-Inf|y==-Inf))
  ytmp <- subset(y, !(x==-Inf|y==-Inf))
  x <- xtmp
  y <- ytmp
  plot(x, y, main=name, xlab=xl, ylab=yl, xlim=c(MINVAL,MAXVAL), ylim=c(MINVAL,MAXVAL))
  tmp1 <- lm(y~x)
  abline(tmp1, col="red")
  tmp2 <- summary(tmp1)
  pos1 <- MAXVAL - (MAXVAL-MINVAL)/10*1
  pos2 <- MAXVAL - (MAXVAL-MINVAL)/10*2
  pos3 <- MAXVAL - (MAXVAL-MINVAL)/10*3
  text(MINVAL,pos1,pos=4,labels=paste("Rq =",round(as.numeric(tmp2[8]),digits=2)),col="red")
  text(MINVAL,pos2,pos=4,labels=paste("Sl
=",round(as.numeric(tmp1$coefficients[2]),digits=2)),col="red")
  text(MINVAL,pos3,pos=4,labels=paste("In =",round(as.numeric(tmp1$coefficients[1]))),col="red")
}

#################################################################################################
######################

getDAT = function(fname,GPR,BLOCKS=NULL,INDEX=NULL){
  fname <- unlist(strsplit(fname,"\\",fixed=TRUE))
  fname <- fname[length(fname)]
  fname <- substr(fname,1, nchar(fname)-4)
  if(is.null(BLOCKS)){
    ARRAY <- GPR
  }else{
    ARRAY <- subset(GPR, as.numeric(GPR$Block) %in% BLOCKS)
    fname <- sprintf("%s_ARRAY%.2d",fname,INDEX)
  }
  ### dat ###
  SA <- NULL
  SA_LINES <- nrow(ARRAY)/3
  for(i in 1:3){
    SA[[i]] <- ARRAY[(SA_LINES*(i-1)+1):(SA_LINES*(i)),,drop=FALSE]
  }
  if(FALSE %in% unique(c(SA[[1]]$Name == SA[[2]]$Name, SA[[1]]$Name == SA[[3]]$Name))){
    print("ERROR")
    return()
  }
  DAT <- cbind(SA[[1]]$ID, SA[[1]]$Name, SA[[1]]$Annotation, SA[[1]]$Flags, SA[[2]]$Flags,
SA[[3]]$Flags, SA[[1]][[PAR_val]], SA[[2]][[PAR_val]], SA[[3]][[PAR_val]])
  colnames(DAT) <- c("ID", "Name", "Annotation", "SA1_Flags", "SA2_Flags", "SA3_Flags", "SA1",
"SA2", "SA3")
  tmp <- apply(DAT[,c("SA1","SA2","SA3")],c(1,2),as.numeric)
  DAT <- cbind(DAT, apply(tmp,1,mean), apply(tmp,1,median), apply(tmp,1,sd),
apply(tmp,1,sd)/apply(tmp,1,mean), apply(tmp,1,min), apply(tmp,1,max))
  colnames(DAT)[colnames(DAT)==""] <- c("Mean", "Median", "SD", "CV", "Min", "Max")
  MC2 <- NULL
  MMC2 <- NULL
  for(i in 1:nrow(DAT)){
    MC2 <- rbind(MC2,
getMC2(as.numeric(DAT[i,"SA1"]),as.numeric(DAT[i,"SA2"]),as.numeric(DAT[i,"SA3"]))[2:4])
    MMC2 <- c(MMC2,
getMMC2(as.numeric(DAT[i,"SA1"]),as.numeric(DAT[i,"SA2"]),as.numeric(DAT[i,"SA3"])))
  }
  DAT <- cbind(DAT, MC2, MMC2)
  ### corr ###
  tmp[cbind(SA[[1]]$Flags %in% PAR_flags, SA[[2]]$Flags %in% PAR_flags, SA[[2]]$Flags %in%
PAR_flags)] <- NA
  colnames(tmp) <- paste("CORR_",colnames(tmp),sep="")
  DAT <- cbind(DAT,tmp)
  DAT <- cbind(DAT, apply(tmp,1,mean,na.rm=T), apply(tmp,1,median,na.rm=T),
apply(tmp,1,sd,na.rm=T), apply(tmp,1,sd,na.rm=T)/apply(tmp,1,mean,na.rm=T),
apply(tmp,1,min,na.rm=T), apply(tmp,1,max,na.rm=T))
  DAT[DAT==Inf] <- NA
  DAT[DAT==-Inf] <- NA
  DAT[DAT=="NaN"] <- NA
  colnames(DAT)[colnames(DAT)==""] <- c("CORR_Mean", "CORR_Median", "CORR_SD", "CORR_CV",
"CORR_Min", "CORR_Max")
  CORR_MC2 <- NULL
  CORR_MMC2 <- NULL
  for(i in 1:nrow(DAT)){
    VAL <- tmp[i,!is.na(tmp[i,])]
    if(length(VAL)==1){
       CORR_MC2 <- rbind(CORR_MC2, c(VAL,VAL,VAL))
       CORR_MMC2 <- c(CORR_MMC2, VAL)
    }else if(length(VAL)==2){
       CORR_MC2 <- rbind(CORR_MC2, c(mean(VAL),VAL))
       CORR_MMC2 <- c(CORR_MMC2, mean(VAL))
    }else if(length(VAL)==3){
       CORR_MC2 <- rbind(CORR_MC2, getMC2(VAL[1],VAL[2],VAL[3])[2:4])
       CORR_MMC2 <- c(CORR_MMC2, getMMC2(VAL[1],VAL[2],VAL[3]))
    }else{
       CORR_MC2 <- rbind(CORR_MC2, c(NA,NA,NA))
       CORR_MMC2 <- c(CORR_MMC2, NA)
    }
  }
  colnames(CORR_MC2) <- c("CORR_MC2","CORR_MC2_1","CORR_MC2_2")
  DAT <- cbind(DAT, CORR_MC2, CORR_MMC2)
  if(PAR_filter){
    DAT <- filterDAT(DAT)
  }
  DAT <- DAT[order(DAT[,"Name"]),]
  write.table(DAT, file=sprintf("%s.dat",fname), quote=TRUE, sep="\t", row.names=FALSE,
col.names=TRUE)
  ### image ###
  NCOLOR <- 256
  PAL <- rev(heat.colors(NCOLOR))
  MINVAL <- 0
  MAXVAL <- 65535
  MINVAL_LOG <- 2.5
  MAXVAL_LOG <- log10(MAXVAL)
  PAR_valB <- "B635.Mean"
  png(filename=sprintf("%s.png",fname), width=5*200, height=4*200, res=100)
       layout(rbind(
         c(1,3,5,6,7),
         c(1,3,8,9,10),
         c(2,4,11,13,15),
         c(2,4,12,14,16)
    ))
  par(mar=c(0,4,2,1))
  plot(as.numeric(ARRAY$X), as.numeric(ARRAY$Y)*(−1), col=PAL[(NCOLOR-1)/(MAXVAL-
MINVAL)*(as.numeric(ARRAY[[PAR_val]])-MINVAL)+1], pch=15, cex=0.8, xlab="", ylab="", axes=FALSE,
main="Array signal", yaxs="i", xaxs="i", xlim=c(min(as.numeric(GPR$X)), max(as.numeric(GPR$X))),
ylim=c(max(as.numeric(GPR$Y))*(−1), min(as.numeric(GPR$Y))*(−1)))
  tmp <- log10(as.numeric(ARRAY[[PAR_val]]))
  tmp[tmp<MINVAL_LOG] <- MINVAL_LOG
  plot(as.numeric(ARRAY$X), as.numeric(ARRAY$Y)*(−1), col=PAL[(NCOLOR-1)/(MAXVAL_LOG-
MINVAL_LOG)*(tmp-MINVAL_LOG)+1], pch=15, cex=0.8, xlab="", ylab="", axes=FALSE, yaxs="i",
xaxs="i", main="Array signal log", xlim=c(min(as.numeric(GPR$X)), max(as.numeric(GPR$X))),
ylim=c(max(as.numeric(GPR$Y))*(−1), min(as.numeric(GPR$Y))*(−1)))
  plot(as.numeric(ARRAY$X), as.numeric(ARRAY$Y)*(−1), col=PAL[(NCOLOR-1)/(MAXVAL-
MINVAL)*(as.numeric(ARRAY[[PAR_valB]])-MINVAL)+1], pch=15, cex=0.8, xlab="", ylab="", axes=FALSE,
main="Array background", yaxs="i", xaxs="i", xlim=c(min(as.numeric(GPR$X)),
max(as.numeric(GPR$X))), ylim=c(max(as.numeric(GPR$Y))*(−1), min(as.numeric(GPR$Y))*(−1)))
  tmp <- log10(as.numeric(ARRAY[[PAR_valB]]))
  tmp[tmp<MINVAL_LOG] <- MINVAL_LOG
  plot(as.numeric(ARRAY$X), as.numeric(ARRAY$Y)*(−1), col=PAL[(NCOLOR-1)/(MAXVAL_LOG-
MINVAL_LOG)*(tmp-MINVAL_LOG)+1], pch=15, cex=0.8, xlab="", ylab="", axes=FALSE, yaxs="i",
xaxs="i", main="Array background log", xlim=c(min(as.numeric(GPR$X)), max(as.numeric(GPR$X))),
ylim=c(max(as.numeric(GPR$Y))*(−1), min(as.numeric(GPR$Y))*(−1)))
  par(mar=c(4,4,2,1))
  plotplus(as.numeric(DAT[,"SA1"]),as.numeric(DAT[,"SA2"]),"",sprintf("SA %d",1),sprintf("SA
%d",2), MINVAL, MAXVAL)
  plotplus(as.numeric(DAT[,"SA1"]),as.numeric(DAT[,"SA3"]),"",sprintf("SA %d",1),sprintf("SA
%d",3), MINVAL, MAXVAL)
  plotplus(as.numeric(DAT[,"SA2"]),as.numeric(DAT[,"SA3"]),"",sprintf("SA %d",2),sprintf("SA
%d",3), MINVAL, MAXVAL)
  plotplus(log10(as.numeric(DAT[,"SA1"])),log10(as.numeric(DAT[,"SA2"])),"",sprintf("SA %d
log",1),sprintf("SA %d log",2), MINVAL_LOG, MAXVAL_LOG)
  plotplus(log10(as.numeric(DAT[,"SA1"])),log10(as.numeric(DAT[,"SA3"])),"",sprintf("SA %d
log",1),sprintf("SA %d log",3), MINVAL_LOG, MAXVAL_LOG)
  plotplus(log10(as.numeric(DAT[,"SA2"])),log10(as.numeric(DAT[,"SA3"])),"",sprintf("SA %d
log",2),sprintf("SA %d log",3), MINVAL_LOG, MAXVAL_LOG)
  plotplus(MC2[,"MC2_1"],MC2[,"MC2_2"],"",sprintf("MC2 %d",1),sprintf("MC2 %d",2), MINVAL,
MAXVAL)
  plotplus(log10(MC2[,"MC2_1"]),log10(MC2[,"MC2_2"]),"",sprintf("MC2 %d log",1),sprintf("MC2 %d
log",2), MINVAL_LOG, MAXVAL_LOG)
  plot(density(as.numeric(ARRAY[[PAR_val]])), xlim=c(MINVAL,MAXVAL), main="", xlab="Signal")
  plot(density(log10(as.numeric(ARRAY[[PAR_val]]))), xlim=c(MINVAL_LOG,MAXVAL_LOG), main="",
xlab="Signal log")
  tmp <- NULL
  tmp[["SA1"]] <- as.numeric(DAT[,"SA1"])
  tmp[["SA2"]] <- as.numeric(DAT[,"SA2"])
  tmp[["SA3"]] <- as.numeric(DAT[,"SA3"])
  boxplot(tmp, ylim=c(MINVAL,MAXVAL), ylab="Signal")
  tmp <- NULL
  tmp[["SA1"]] <- log10(as.numeric(DAT[,"SA1"]))
  tmp[["SA2"]] <- log10(as.numeric(DAT[,"SA2"]))
  tmp[["SA3"]] <- log10(as.numeric(DAT[,"SA3"]))
  boxplot(tmp, ylim=c(MINVAL_LOG,MAXVAL_LOG), ylab="Signal log")
  dev.off()
}

#################################################################################################
######################

files <- dir(path=PAR_dir, pattern="*.gpr", full.names=TRUE)
files <- files[!file.info(files)$isdir]
files <- files[substr(files,nchar(files)-3,nchar(files)) == ".gpr"]
for(i in 1:length(files)){
  print(sprintf("Processing file %d out of %d",i,length(files)))
  GPR <- readGPR(files[i])
  if(PAR_array==1){
    getDAT(files[i],GPR)
  }else{
    for(j in 1:PAR_array){
       print(sprintf("Processing array %d out of %d",j,PAR_array))
       getDAT(files[i],GPR,PAR_multiwell[,j],j)
    }
  }
}

Appendix 2: calc_THRES_KS_130711_V02 R Script

library(rapmad)

INDEX<-read.table("sample.xls",header=F,sep = "\t", strip.white=T,skip=1)
INDEX.dat<-INDEX[INDEX[,1]=="dat",]
if(ncol(INDEX.dat)==4){
       LIST.1 <- c(as.character(INDEX.dat[,2]),as.character(INDEX.dat[,4]))
       }else{
       LIST.1 <- c(as.character(INDEX.dat[,2]))
       }
LIST.1<-LIST.1[LIST.1!=""]

FILENAME<-"OUTFILE_THRESH.Rdat"

DAT<-NULL
for (cnt1 in 1:length(LIST.1)){

       DAT.TMP<-read.table(paste(LIST.1[cnt1],sep=""),header=TRUE,sep = "\t", strip.white=T)
       DAT<-cbind(DAT,DAT.TMP$MMC2)
       }
#INDEX<-read.table(paste("INDEX_2070_NEW.txt",sep=""),header=TRUE,sep = "\t", strip.white=T)

CN<-substr(LIST.1,1,nchar(LIST.1)-4)
colnames(DAT)<-CN
DAT<-cbind(DAT.TMP[,c(1:3)],DAT)
save(DAT,file=FILENAME)

####Plot of Signal Distribution for Raw Data
TAB.OUT<-NULL
png(filename = paste("01_thresholds_raw_data.png",sep=""), width = 3200, height =
1600*round((length(LIST.1)+.1)/2),
       units = "px", pointsize = 40, bg = "white", res = NA,
       restoreConsole = TRUE)

       par(mfrow=c(round((length(LIST.1)+.1)/2),2))
       par(mai=c(2.5,2.5,1,.5))

for (c1 in (4:ncol(DAT))){
       FIT.TMP<-signal.gmm.generic(DAT[,c1],logScale=F)
       y1<-rnorm(10000,mean = FIT.TMP[[2]][1], sd = FIT.TMP[[3]][1])
       y2<-rnorm(10000,mean = FIT.TMP[[2]][2], sd = FIT.TMP[[3]][2])
       D<-density(DAT[,c1])
       plot(density(DAT[,c1]),lwd=4,main=colnames(DAT)[c1],xlim=c(FIT.TMP[[2]][1]-
(FIT.TMP[[3]][1]*3),FIT.TMP[[2]][1]+(FIT.TMP[[3]][1]*15)))
       lines(density(y1),col="red",lwd=4)
       lines(density(y2),col="blue",lwd=4)
       abline(v=FIT.TMP[[5]],lwd=3,col="red",lty=2)
       abline(v=FIT.TMP[[2]][1]+(FIT.TMP[[3]][1]*5),lwd=3,col="blue",lty=2)
       DAT.STAT<-cbind(DAT[,c1],FIT.TMP[[1]])
       DAT.STAT.p0.01<-DAT.STAT[DAT.STAT[,2]<0.01,]
       abline(v=min(DAT.STAT.p0.01[,1]),lwd=3,col="green",lty=2)
       DAT.STAT.p0<-DAT.STAT[DAT.STAT[,2]==0,]
       abline(v=min(DAT.STAT.p0[,1]),lwd=3,col="magenta",lty=2)
       legend(min(DAT.STAT.p0[,1]),max(D$y),c("Data","FIT-Noise","Fit-Signal","FDR
cutoff","SD.noise*5","P<0.01","P=0"),lwd=4,
                                                          lty=c(1,1,1,2,2,2,2),
       col=c("black","red","blue","red","blue","green","magenta"))
       TAB.OUT<-rbind(TAB.OUT,cbind(colnames(DAT)[c1],FIT.TMP[[5]],FIT.TMP[[2]][1]+(FIT.TMP[[3]][1]*5),min(DAT.ST
AT.p0.01[,1]),min(DAT.STAT.p0[,1])))
       }
dev.off()

##make DAT-CTRL
DAT.COR<-NULL
DAT.ID.KEEP<-NULL
if (ncol(INDEX.dat)==4){
       INDEX.dat.corr<-INDEX.dat[INDEX.dat[,4]!="",]
       for (c1 in (1:nrow(INDEX.dat.corr))){
       DAT.ID.TMP<-
substr(as.character(INDEX.dat.corr[c1,2]),1,nchar(as.character(INDEX.dat.corr[c1,2]))-4)
       DAT.ID.KEEP<-c(DAT.ID.KEEP,DAT.ID.TMP)
       CTR.ID.TMP<-
substr(as.character(INDEX.dat.corr[c1,4]),1,nchar(as.character(INDEX.dat.corr[c1,4]))-4)
       DAT.COR<-cbind(DAT.COR,DAT[[DAT.ID.TMP]]-DAT[[CTR.ID.TMP]])
       }
       colnames(DAT.COR)<-paste(DAT.ID.KEEP,"COR",sep="_")
       DAT.COR<-cbind(DAT[,1:3],DAT.COR)
####Plot of Signal Distribution for Corrected Data

png(filename = paste("02_thresholds_corrected_data.png",sep=""), width = 3200, height =
1600*round((length(DAT.ID.KEEP)+.1)/2),
       units = "px", pointsize = 40, bg = "white", res = NA,
       restoreConsole = TRUE)

       par(mfrow=c(round((length(DAT.ID.KEEP)+.1)/2),2))
       par(mai=c(2.5,2.5,1,.5))

for (c1 in (4:ncol(DAT.COR))){
       FIT.TMP<-signal.gmm.generic(DAT.COR[,c1],logScale=F)
       y1<-rnorm(10000,mean = FIT.TMP[[2]][1], sd = FIT.TMP[[3]][1])
       y2<-rnorm(10000,mean = FIT.TMP[[2]][2], sd = FIT.TMP[[3]][2])
       D<-density(DAT.COR[,c1])
       plot(density(DAT.COR[,c1]),lwd=4,main=colnames(DAT.COR)[c1],xlim=c(FIT.TMP[[2]][1]-
(FIT.TMP[[3]][1]*3),FIT.TMP[[2]][1]+(FIT.TMP[[3]][1]*15)))
       lines(density(y1),col="red",lwd=4)
       lines(density(y2),col="blue",lwd=4)
       abline(v=FIT.TMP[[5]],lwd=3,col="red",lty=2)
       abline(v=FIT.TMP[[2]][1]+(FIT.TMP[[3]][1]*5),lwd=3,col="blue",lty=2)
       DAT.STAT<-cbind(DAT.COR[,c1],FIT.TMP[[1]])
       DAT.STAT.p0.01<-DAT.STAT[DAT.STAT[,2]<0.01,]
       abline(v=min(DAT.STAT.p0.01[,1]),lwd=3,col="green",lty=2)
       DAT.STAT.p0<-DAT.STAT[DAT.STAT[,2]==0,]
       abline(v=min(DAT.STAT.p0[,1]),lwd=3,col="magenta",lty=2)
       legend(min(DAT.STAT.p0[,1]),max(D$y),c("Data","FIT-Noise","Fit-Signal","FDR
cutoff","SD.noise*5","P<0.01","P=0"),lwd=4,
                                                          lty=c(1,1,1,2,2,2,2),
       col=c("black","red","blue","red","blue","green","magenta"))
       TAB.OUT<-
rbind(TAB.OUT,cbind(colnames(DAT.COR)[c1],FIT.TMP[[5]],FIT.TMP[[2]][1]+(FIT.TMP[[3]][1]*5),min(DA T.STAT.p0.01[,1]),min(DAT.STAT.p0[,1])))
       }
dev.off()
}
colnames(TAB.OUT)<-c("Experiment","FDR-cutoff","SD.noise*5","P<0.01","P=0")
write.table(TAB.OUT,file="thresholds.xls",quote=F,row.names=F,col.names=T,sep="\t")

Appendix 3: Table_select_V01 R script

readDAT = function(name) {
  return(read.table(name, header=TRUE, sep ="\t", strip.white=TRUE, colClasses="character"))
}

filterDAT = function(dat) {
  tmp <- NULL
  exclude =
c("AA","empty","GGSGGGSDYKDDDDK","GGSDYKDDDDK","DYKDDDDK","printingbuffer","mouse_IgG","human_IgG
","IgG-mouse","IgG-human")
  for(i in 1:nrow(dat)) {
    if( length(unlist(strsplit(dat$Annotation[i],"|",fixed=TRUE))) == 4 & (!(dat$ID[i] %in%
exclude)) ) {
      tmp <- c(tmp,TRUE)
    } else {
      tmp <- c(tmp,FALSE)
    }
  }
  return(dat[tmp,])
}

#################################################################################################
######################

VAL <- "MMC2"

files <- read.table("samples.xls", header=FALSE, sep ="\t", strip.white=TRUE,
colClasses="character")
TAB <- NULL
for(i in 1:nrow(files)) {
  print(i)
  print(files[[1]][i])
  DAT <- filterDAT(readDAT(files[[1]][i]))
  TAB <- cbind(TAB, as.numeric(DAT[[VAL]]))
}
colnames(TAB) <- files[[2]]
rownames(TAB) <- sprintf("%s - %s", DAT$ID, DAT$Name)

TAB <- TAB[order(as.numeric(substr(rownames(TAB),19,22))),]
TAB <- apply(TAB,c(1,2),round)

OUT <- rownames(TAB)
OUT <- strsplit(OUT, " - ", fixed=T)
OUT <- do.call("rbind",OUT)
OUT <- cbind(OUT, do.call("rbind",strsplit(substr(OUT[,2],6,14),"",fixed=T))) # Protein
OUT <- cbind(OUT, as.numeric(substr(OUT[,2],16,19))+1 ) # Position+1
OUT <- cbind(OUT, do.call("rbind",strsplit(substr(OUT[,2],21,182),"",fixed=T))) # Clade
PROT <- toupper(c("env", "gag", "nef", "pol", "rev", "tat", "vif", "vpr", "vpu"))
CLADE <- c("A","A1","A2","B","C","D","F1","F2","G","H","J","K","01_AE","02_AG","03_AB","04_cpx","05_DF","0
6_cpx","07_BC","08_BC","09_cpx","10_CD","11_cpx","12_BF","13_cpx","14_BG","15_01B","16_A2D","17_B
F","18_cpx","19_cpx","20_BG","21_A2D","22_01A1","23_BG","24_BG","25_cpx","26_AU","27_cpx","28_BF"
,"29_BF","31_BC","32_06A1","33_01B","34_01B","35_AD","36_cpx","37_cpx","38_BF1","39_BF","40_BF","
42_BF","43_02G","44_BF","45_cpx","46_BF","47_BF","49_cpx","0102A","01A1","01ADF2","01AF2U","01B",
"01BC","01C","01DU","0206","0209","0213","0225","02A","02A1","02A1U","02AG","02B","02C","02D","02
GK","02U","06A1","0708","09A","13U","1819","26C","26CU","A1A2CD","A1A2D","A1C","A1CD","A1CDGKU","
A1CG","A1D","A1DHK","A1F2","A1G","A1GHU","A1GJ","A1GU","A1H","A1U","A2C","A2CD","A2D","AC","ACD",
"AD","ADGU","ADU","AF2","AF2G","AGKU","AGU","AHJU","AKU","AU","BC","BCF1","BCU","BF","BF1","BG","
CD","CF1","CF1U","CU","DF","DF1G","DG","DO","DU","F2KU","GKU","JKU","P","A3","0107","01GHJKU","11
A1","23A1","A1B","A1DK","A2G","ACFG","AG","AGH","BFG","DF1","GH","F","01F2","01U","02H","07B","11
C","CGU","x","01A1G","02F2","KU","A1A2","GJ")
colnames(OUT) <- c("Sequence","Name",PROT,"Position+1",CLADE)

OUT <- cbind(OUT, TAB)
write.table(OUT, file="table_select.xls", quote=TRUE, sep="\t", row.names=F, col.names=TRUE)