Rational Peptide Selection To Detect Human Immunodeficiency Virus Type 1-Specific T-Cell Responses under Resource-Limited Conditions

Abstract

Understanding human immunodeficiency virus type 1 (HIV-1)-specific cytotoxic T-lymphocyte responses is important for the development of vaccines and therapies. We describe a novel method for the rational selection of peptides that target stable regions of the HIV-1 genome, rich in epitopes specifically recognized by the study population. This method will be of particular use under resource/sample-limited conditions.

Understanding the fine specificity of cytotoxic T-lymphocyte (CTL) responses is important for vaccine monitoring and design. The most commonly used assay for the detection of antigen-specific CTLs is the enzyme-linked immunospot (ELISPOT) assay. This rapid assay can detect cytokines secreted from virus-specific T cells in response to antigens, usually in the form of synthetic peptides based on consensus or autologous virus sequences. The number of antigens that can be interrogated by this assay depends on the availability of both peripheral blood mononuclear cells (PBMC) and synthetic peptides. Responses to the entire human immunodeficiency virus type 1 (HIV-1) proteome may be mapped, as has been shown, but this requires a large number of both PBMC and peptides (1, 2). An alternative approach for reducing the number of PBMC needed is through the use of pooled peptides. This has been an effective method of reducing the number of PBMC needed, but it remains dependent on a large number of peptides. Moreover, there are situations in which the numbers of viable PBMC are severely limited, such as in the pediatric setting, or the cost of manufacturing large sets of peptides is prohibitive. Under these circumstances, a more targeted approach is needed to provide information on responses using the minimum number of cells and peptides.

Here, we describe a quantitative method for selecting panels of individual peptides to maximize data on the breadth and magnitude of a response while minimizing the number of PBMC and peptides needed and taking into account the underlying HLA structure of the sampled population.

Our method selects a pool of peptides for interrogation that fall under epitopes of higher HLA prevalence in the studied population, weighted (penalized) for cases of high entropy (sequence variation). In so doing, we focus on regions of the HIV-1 proteome that are most likely to elicit immunogenic responses from the greatest fraction of the population. Moreover, our weighting strategy leads to higher peptide scores for peptides of low entropy, even for those epitopes falling under low-prevalence HLA types, ensuring the study of immunogenic peptides even among that fraction of the sampled population that bears low-frequency HLA types.

We targeted regions of the HIV-1 proteome that were rich in CTL epitopes (3, 11, 12). These regions were then weighted according to the diversity of the HLA restrictions of the epitopes and with further biased selection toward HLA types of higher frequency within the study population. Finally, to reduce the chance of false-negative results, as described by Altfeld et al., we incorporated the entropy of each amino acid within the selected region (1). This method is described in greater detail below.

Identification of epitopes.

We based our analysis on the usage of peptides of 15 amino acids in length. To determine the number of epitopes each peptide contained, we mapped major histocompatibility complex class I (MHC-I)-restricted epitopes, obtained from the Los Alamos database (http://hiv-web.lanl.gov/content/immunology/tables/ctl_summary.html), onto the HIV-1 HXB2 proteome. The HLA restrictions were standardized to the two-digit molecular HLA type nomenclature (8). The epitopes were mapped onto the HXB2 proteome as shown in the examples from HIV-1 Gag p17 (Fig. 1A).

FIG. 1.

Generation of peptide scores. (A) Epitopes were mapped onto the HXB2 genome. Epitopes are shown as white boxes under their specific amino acid sequences; their HLA restriction is shown within. Overlapping epitopes were combined to make one continuous epitope region. A hypothetical example is shown in blue (part i); the joined epitope constitutes the HLA-A2-restricted SLYNTVTAL epitope. (B) Generation of Σ_HLA. Each amino acid received a score equal to that of the HLA prevalences for each epitope that covers it. In parts i and ii, examples are given showing the HLA prevalences mapped onto their respective epitopes. For example, the first residue (glutamic acid [E]) is covered by a single HLA-A1-restricted epitope. HLA-A1 has an 8% prevalence within the North American population; therefore, for this amino acid, its Σ_HLA is 0.08. (iii) The Σ_HLA for each amino acid across the whole of p17 is displayed as a heat map above a skyline graph. (C) Entropy is shown for the example sequence (parts i and ii) and mapped across p17 as a heat map above and skyline graph below (part iii). (D) The final peptide score is generated from the division of the sum of the Σ_HLA from each amino acid within the peptide by the sum of the entropy from each amino acid within the peptide. In this example, from peptide 7890, the total Σ_HLA across the peptide equaled 7.00 and the total entropy was 5.98, giving a final peptide score of 1.17.

Amino acid score.

To maximize the recognition of the selected peptides, epitopes were weighted based on the HLA prevalence (as a percentage) of our North American study group. The HLA prevalence for North America was obtained from the dbMHC IHWG Anthropology database (http://www.ncbi.nlm.nih.gov/mhc/MHC.fcgi?cmd=init). Each amino acid within the HIV-1 proteome received a score that equaled the total of the HLA prevalence for each epitope that covered it (Σ_HLA); an example from two 15-mer peptides from p17 is shown (Fig. 1B). When extended across all of HIV-1 Gag p17, the Σ_HLA can be displayed as a heat map (Fig. 1B, part iii). The sum of each amino acid's Σ_HLA contributed to the final peptide score (Fig. 1D).

Entropy.

The variability of HIV-1 antigens is known to interfere with the accurate detection and measurement of HIV-1-specific T-cell responses (1). Differences in sequence between the consensus-based synthetic peptide antigens and an individual's autologous virus often result in false-negative responses. In order to minimize this bias and systematically focus on conserved regions of HIV-1, we computed the amino acid entropy (variability) of sites across the HIV-1 proteome and incorporated this information into our peptide scoring scheme.

In this example, the entropy of each amino acid was determined with full-length subtype B protein sequence alignments of all HIV-1 proteins, downloaded from the Los Alamos database (http://hiv.lanl.gov/content/hiv-db/ALIGN_CURRENT/ALIGN-INDEX.html). Alignments were restricted to one sequence per individual to minimize the sampling bias. Sequences were then realigned by using ClustalW with default gap parameters and the “IUB” DNA weight matrix (10). Manual aligning of variable regions was performed with the Se-Al sequence alignment editor, version 2.0 (http://evolve.zoo.ox.ac.uk/). The HyPhy software package was implemented to compute site-specific Shannon entropy scores for each coding region from the frequency, f, of amino acid A at position i, according to the formula −ΣA f(A_i) ln[f(A_i)] (6, 9). Sites within each protein were aligned with the HIV-1 HXB2 sequence (4) (Fig. 1C, parts i and ii). The sum of each amino acid's entropy within a peptide was used in the final peptide score (Fig. 1D).

Final peptide score.

In this example, the HIV-1 group M consensus (15-mer) peptide sequences, from the NIH ARRRP, were mapped onto the HIV-1 proteome. The final peptide scores were based on the sum of the amino acid's Σ_HLA from across the peptide divided by the sum of the entropy from each amino acid within the peptide (Fig. 1D). Peptides from each HIV-1 protein were then ranked; Table 1 shows an example for HIV-1 Gag p17. To determine the breadth of responses, top-ranking peptides from each protein were selected for use in an ELISPOT assay.

TABLE 1.

Peptide scores for the p17 protein^a

NIH peptide	Peptide sequence	Score
		HLA	Entropy	Peptide
7872	MGARASVLSGGELDR	0.14	4.21	0.03
7873	ASVLSGGELDRWEKI	1.06	4.52	0.23
7874	SGGELDRWEKIRLRP	3.75	3.77	0.99
7875	LDRWEKIRLRPGGKK	6.59	2.79	2.36
7876	EKIRLRPGGKKKYKL	9.22	4.34	2.12
7877	LRPGGKKKYKLKHIV	9.96	4.99	2.00
7878	GKKKYKLKHIVWASR	8.75	5.08	1.72
7879	YKLKHIVWASRELER	6.74	3.06	2.20
7880	HIVWASRELERFAVN	4.61	2.36	1.95
7881	ASRELERFAVNPGLL	2.39	1.55	1.55
7882	LERFAVNPGLLETSE	1.22	2.55	0.48
7883	AVNPGLLETSEGCRQ	0.00	4.49	0.00
7884	GLLETSEGCRQILGQ	0.00	6.01	0.00
7895	EVKDTKEALEKIEEE	0.00	7.54	0.00
7886	CRQILGQLQPSLQTG	0.00	5.89	0.00
7887	LGQLQPSLQTGSEEL	0.65	6.17	0.11
7888	QPSLQTGSEELRSLY	3.08	6.13	0.50
7889	QTGSEELRSLYNTVA	6.31	5.76	1.10
7890	EELRSLYNTVATLYC	7.00	5.98	1.17
7891	SLYNTVATLYCVHQR	7.15	6.39	1.12
7892	TVATLYCVHQRIEVK	7.07	7.38	0.96
7893	LYCVHQRIEVKDTKE	6.36	4.94	1.29
7894	HQRIEVKDTKEALEK	6.34	1.87	3.39
7895	EVKDTKEALEKIEEE	5.04	0.86	5.83
7896	TKEALEKIEEEQNKS	4.36	0.43	10.23
7897	LEKIEEEQNKSKKKA	0.09	5.86	0.01
7898	EEEQNKSKKKAQQAA	0.00	5.52	0.00
7899	NKSKKKAQQAAADTG	0.14	7.79	0.02
7900	KKAQQAAADTGNSSQ	0.85	10.28	0.08
7901	QAAADTGNSSQVSQN	1.76	9.44	0.19
7902	DTGNSSQVSQNYPIV	2.28	8.14	0.28
7903	SSQVSQNYPIVQNLQ	2.26	6.16	0.37
7904	SQNYPIVQNLQGQMV	1.45	2.30	0.63

^aScores were generated with HLAs prevalent in North America.

Proof of principle.

To test this method, we screened gamma interferon (IFN-γ) responses from 10 HIV-1-infected subjects. The HLA types of the individuals were unknown. The North American HLA prevalence was applied to generate two panels of peptides (Table 2). Panel A contained peptides that had the highest peptide scores, and panel B contained peptides with the lowest (but still greater than 0) peptide scores. Based on our hypothesis, we predicted that panel A would show the highest number of responses.

TABLE 2.

Peptide panels^a

Protein	Panel A		Panel B
	Peptide	Score	Peptide	Score
p17	7896	10.23	7897	0.01
	7895	5.83	7899	0.02
p24	7916	12.62	7941	0.75
	7957	12.33	7962	0.27
p2 p7 p2 p6	7977	1.57	7964	0.06
	7978	1.28	7989	0.01
Protease	5499	6.98	5491	0.06
	5498	3.02	5490	0.04
RT	5526	10.28	5596	0.11
	5525	5.88	5534	0.06
Integrase	5682	5.91	5688	0.02
	5684	4.87	5668	0.01
Tat	5117	0.91	5135	0.09
	5118	0.86	5125	0.01
Rev	6008	0.78	6002	0.06
	6014	0.56	6001	0.01
Vif	6058	1.62	6045	0.09
	6057	1.06	6040	0.09
Vpr	6072	0.80	6064	0.02
	6071	0.78	6081	0.14
Vpu	6426	0.38	6428	0.05
	6427	0.33	5982	0.01
Env	8771	5.99	8832	0.01
	8772	3.11	8862	0.01
Nef	5155	3.49	5177	0.02
	5156	2.71	5139	0.01

^aPanel A contains peptides that received the highest scores for each protein; panel B contains the lowest-scoring peptides.

The IFN-γ ELISPOT assay was conducted by the standard protocol, as previously described (5, 7). In this example, approximately 4 million PBMC were used to cover each peptide panel and controls. As predicted, the highest number of responses was directed toward panel A (Fig. 2). An average of 7.4 peptides induced responses in panel A, compared to 0 in panel B (P = 0.0053) (Fig. 2).

FIG. 2.

Proof of principle. Two panels of 21 peptides were generated by the method described in the legend to Fig. 1. Panel A was selected from the 21 highest-scoring peptides, while panel B was selected from the 21 lowest-scoring (but still scoring >0) peptides. IFN-γ responses from nine (panel A) and five (panel B) HIV-1-infected subjects revealed that panel A was significantly better recognized than panel B (P = 0.0053).

In summary, we have described a novel method for the rational selection of peptides for screening HIV-1-specific CTL responses in resource-limited settings. This protocol targets stable regions of the HIV-1 genome that are rich in epitopes that are specifically recognized by the study population where underlying HLA prevalences are known. In the example given, only 4 million PBMC, which can be isolated from as little as 5 ml of blood, were used and only 26 peptides were needed. If these peptides had been pooled, cell numbers could have been reduced further, although this would reduce the information on the breadth of responses.

This method could easily be adapted to other viruses, such as hepatitis C virus and human cytomegalovirus, and/or expanded to incorporate CD4⁺ T-cell epitopes. Due to the low number of resources needed, this approach will be of particular benefit to in-field therapeutic and vaccine studies, as well as studies conducted under resource-limited conditions.