Conceived and designed the experiments: MR NF DCN NJ CB DEH JIM. Performed the experiments: MR NF DCN NJ DEH. Analyzed the data: MR NF CB DEH JIM. Contributed reagents/materials/analysis tools: WD TMA. Wrote the paper: MR NF CB JIM.
Current address: Military HIV Research Program, Henry M. Jackson Foundation, Rockville, Maryland, United States of America
Current address: Vaccine and Infectious Disease Institute, Fred Hutchinson Cancer Research Center, Seattle, Washington, United States of America
David E. Heckerman is an employee of Microsoft Research. This does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials, as detailed online in the guide for authors.
Different vaccine approaches cope with HIV-1 diversity, ranging from centralized1–4 to variability-encompassing5–7 antigens. For all these strategies, a concern remains: how does HIV-1 diversity impact epitope recognition by the immune system? We studied the relationship between HIV-1 diversity and CD8+ T Lymphocytes (CTL) targeting of HIV-1 subtype B Nef using 944 peptides (10-mers overlapping by nine amino acids (AA)) that corresponded to consensus peptides and their most common variants in the HIV-1-B virus population. IFN-γ ELISpot assays were performed using freshly isolated PBMC from 26 HIV-1-infected persons. Three hundred and fifty peptides elicited a response in at least one individual. Individuals targeted a median of 7 discrete regions. Overall, 33% of responses were directed against viral variants but not elicited against consensus-based test peptides. However, there was no significant relationship between the frequency of a 10-mer in the viral population and either its frequency of recognition (Spearman's correlation coefficient ρ = 0.24) or the magnitude of the responses (ρ = 0.16). We found that peptides with a single mutation compared to the consensus were likely to be recognized (especially if the change was conservative) and to elicit responses of similar magnitude as the consensus peptide. Our results indicate that cross-reactivity between rare and frequent variants is likely to play a role in the expansion of CTL responses, and that maximizing antigenic diversity in a vaccine may increase the breadth and depth of CTL responses. However, since there are few obvious preferred pathways to virologic escape, the diversity that may be required to block all potential escape pathways may be too large for a realistic vaccine to accommodate. Furthermore, since peptides were not recognized based on their frequency in the population, it remains unclear by which mechanisms variability-inclusive antigens (i.e., constructs enriched with frequent variants) expand CTL recognition.
Despite setbacks in the development of CTL-based HIV vaccines
Post-Step trial suggestions emphasize the potential benefit
of strategies that would improve T cell breadth
Although it is agreed that HIV-1's extensive variability is a major challenge for a successful vaccine strategy, which must control extremely diverse viral strains, the impact of HIV-1 diversity on CTL targeting remains poorly understood. Our knowledge of HIV-1 diversity and CTL recognition is mostly derived from whole genome mapping studies of ELISpot responses using consensus peptides and small studies of specific epitopes in HIV-1-infected individuals with longitudinal follow-up, particularly in the context of CTL escape.
We sought to formally analyze the effect of HIV-1 diversity
on CTL targeting of Nef, as it comprises both
conserved and variable segments and is often
targeted both during acute/early and chronic
infection: several Nef peptides were recognized by
more than 40% of individuals in large cohorts
of Americans or South-Africans infected with HIV-1
subtype B and C, respectively
Here, we identified novel HIV-1 Nef epitopes and showed that there is no general relationship between the frequency of an HIV-1 peptide among circulating sequences and its frequency of recognition in the cohort. These findings challenge immunogen designs that are based on a “frequency-only” criterion for variant inclusion and warrant further studies into the determinants of CTL variant recognition in HIV-1.
Subjects with chronic HIV infection were recruited at three hospitals in the Boston area. All human subject protocols were approved by the Partners Human Research Committee, and all subjects provided written informed consent prior to enrollment.
Nine hundred and forty four 10-mer peptides
overlapping by 9 AA were synthesized and used in
the present study. These sequences cover the
distribution of HIV-1 B viral sequences based on a
previously described
Shannon entropy. Shannon entropy was used to
score the variability at each position of an
alignment of HIV-1 circulating sequences
Population frequency. We derived all unique 10-mers in the 514-sequence-dataset and defined their population frequency, i.e., the percentage of sequences with the precise 10-mer sequence present in the dataset of 514 Nef sequences.
IFN-γ ELISpot assays were performed on freshly isolated peripheral blood mononuclear cells (PBMC) from 26 individuals. All peptides were tested in separate wells of the Elispot plates. Due to the large number of peptides tested, not all blood draws yielded sufficient PBMC to test at 100,000 cells/well, therefore assays were run with 74,000 to 100,000 cells/well (median 100,000 cells/well). To be scored as positive, a response had to be greater than: a) four times the mean background, b) the mean background plus three standard deviations, c) five spots per well and d) 55 spot-forming cells per million (SFC/M).
The HIV-1-B Nef peptides that were reactive in
the ELISpot assays were compared by testing k-mers
(with k = 8, 9, 10 AA) that
had the same HIV-1 HXB2 strain coordinates; in
brief, a consensus k-mer was compared to its
variant k-mer(s), with k being of the same length
for the consensus and variant to avoid the
creation of gaps when aligning the 2 k-mers. Pairs
of peptides were tested when the database
frequency of one of the peptides was at least
ten-fold greater than that of the other k-mer. For
each given 10-mer peptide pair, cross-reactivity
was also assessed for the two pairs of 9-mers and
three pairs of 8-mers nested in the 10-mer. We
computed a cross-reactivity fraction corresponding
to the number of individuals who recognized the
less frequent peptide divided by the number of
individuals who recognized the more common one. AA
substitutions were characterized as being
conservative, semi-conservative and
non-conservative based on the Dayhoff PAM250
matrix
Most Nef peptides are unique or very rare at the
population level. We dissected 514 Nef HIV-1 B
sequences into overlapping 10-mers and found
19,860 unique 10-mers: a small fraction of
relatively frequent peptides and a long tail of
rare 10-mers. More than 2/3 of all of these
peptides were engendered by private mutations,
i.e., 13,574 peptides were found only once in this
dataset (
The 944 Nef peptides were tested by IFN-γ
ELISpot assays using fresh PBMC from 26 HIV-1
infected persons (
While 944 peptides were tested by IFN-γ
ELISpot assays, only 350 elicited a response in at
least 1 individual (
The number of 10-mers that were tested are represented as gray bars. Black bars correspond to the 10-mers that were recognized at least by one individual in our cohort. Red bars represent known epitopes that had previously been reported in the HIVDB. Each bar is placed at the start position of each 10-mer based on their HXB2 coordinates.
On an individual peptide basis, the median number of ELISpot responses per subject was 21 (interquartile range (IQR) 14–41; range 1–123). Due to the multiplicity of peptide variants tested, several responses can be scored for the same 10-AA segment. Additionally, due to the 1AA-offset between the peptides, the same epitope can be found in a suite of immediately adjacent overlapping 10-mers: for example, an 8-AA-long epitope is present in three consecutive 10-mers. In order to count only the ELISpot responses that were reflective of an independent CTL specificity, we scored only one response per 10-AA segment, and, if we found two or three responses to consecutive 10-mers, we scored them as one independent response (four consecutive responses were counted as two independent responses). Hence, the median number of independent epitopes recognized per individual was seven (IQR 4–13; range 1–32).
To assess whether using coverage-optimized HIV-1
subtype B vaccine inserts would engender broader
and deeper CTL responses in individuals, peptides
covering three potential vaccine insert strategies
were compared: COT+, ‘3-Best’ and
consensus. A three-gene length COT+
corresponded to 561 peptides in our test set. The
‘3-Best’ natural HIV-1 Nef strains,
which were selected from the HIVDB to afford the
highest coverage of HIV-1 variability
When we evaluated the number of CTL responses corresponding to the 3 different Nef vaccine strategies in our cohort, we found that more peptides matching the 3-gene-length COT+ were targeted than peptides contained in the other strategies: nCOT+ = 221, n‘3-best’ = 200, nCON04 = 95, nCOT = 96. However, the percentage of COT+ or ‘3-best’ peptides recognized was similar (39 and 38%, respectively) while the fraction of HIV-1-B 2004 consensus or COT peptides recognized was, as expected, higher −48 and 49% of the set, respectively. Thus, if a single protein length peptide set is used, consensus or COT peptides are optimal for detecting responses, yet the breadth of T cell responses is extended with the use of coverage-optimized peptides.
We analyzed how each 10-mer segment of HXB2 was
recognized to evaluate whether recognition was due
to the consensus peptide, one or more variant
peptide(s) or a combination of consensus and
variants. We rank-ordered peptides based on their
frequency in the database calling ‘variant
1’ the peptides corresponding to the most
common epitopic variant (after the consensus)
found in circulating sequences, ‘variant
2’ corresponds to the second most frequent
peptides, and so on.
Colored blocks correspond to the 10-mers that were recognized, while the 10-mers that were tested are outlined in black. Recognition of peptides is figured using a gradient of colors: consensus peptides are figured in burgundy and variant peptides follow a gradient from the most conserved (in orange) to the most rare (in purple), while peptides not found in a database of 514 sequences are in black. Each block represents the start position of the 10-mers based on their HXB2 coordinates. Panel A shows all the peptides recognized, panel B represents for each individual only the most frequent peptide recognized at each position (i.e., if an individual recognized the consensus peptide and variants 1 and 4, only the recognition through the consensus is figured).
Counting all responses | |||||||||
Cons. | Var. 1 | Var. 2 | Var. 3 | Var. 4 | Var. 5 | Var. 6 | Var. 7 | Absent | |
Peptides recognized | 91 | 81 | 65 | 40 | 24 | 14 | 6 | 0 | 16 |
Peptides not recognized | 112 | 115 | 110 | 99 | 68 | 28 | 10 | 4 | 43 |
Peptides tested | 203 | 196 | 175 | 139 | 92 | 42 | 16 | 4 | 59 |
% peptides recognized | 44.83 | 41.33 | 37.14 | 28.78 | 26.09 | 33.33 | 37.50 | 0.00 | 27.12 |
Counting responses to the consensus and additional responses afforded by the addition of variants | |||||||||
Cons. | Var. 1 | Var. 2 | Var. 3 | Var. 4 | Var. 5 | Var. 6 | Var. 7 | Absent | |
Peptides recognized | 91 | 50 | 33 | 12 | 10 | 5 | 1 | 0 | 10 |
Peptides not recognized | 112 | 146 | 142 | 127 | 82 | 37 | 15 | 4 | 49 |
Peptides tested | 203 | 196 | 175 | 139 | 92 | 42 | 16 | 4 | 59 |
% peptides recognized | 44.83 | 25.51 | 18.86 | 8.63 | 10.87 | 11.90 | 6.25 | 0.00 | 16.95 |
To better characterize the coverage enhancement
afforded by inclusion of variant peptides, we
scored responses to a 10-mer only once for each
individual, i.e., if an individual responded to
the consensus we did not count responses to the
other variants tested.
In summary, each additional variant level tested
yielded smaller increases in the overall spectrum
of recognition.
The pie chart represents the 137 HIV-1 segments that were recognized in our cohort and the proportion of these that were recognized through the consensus 10-mers or through any of the variant peptides.
We compared the average magnitude of responses
toward the consensus and variant peptides.
Ninety-one consensus peptides were recognized with
an average magnitude of 292 SFC/M
(median = 180 SFC/M), while
246 variant peptides were recognized with an
average magnitude of 243 SFC/M
(median = 170 SFC/M). Hence,
the consensus and variant epitopes elicited
responses of similar magnitudes
(p = 0.15) (
Next, we analyzed the magnitude of responses for
pairs of consensus plus variant peptides, i.e.,
the average magnitude found for the consensus
peptide was compared to the average magnitude
elicited by the different variant peptides
beginning at the same position (irrespective of
whether the individual was able to elicit
responses against both consensus and variant
forms). There were 73 paired sets of
consensus/variant(s) that were recognized in the
cohort. The average magnitude elicited by the
consensus (318 SFC/M) or the variant peptides (261
SFC/M) were not significantly different:
p = 0.80 (
Responses of decreasing magnitude between the consensus and variant peptide are figured in blue, responses of increasing magnitude are figured in red.
Last, we analyzed the magnitude of responses on an individual basis, focusing on individuals who mounted responses toward both the consensus and one or more of the variant peptides. Of the 91 peptides that were recognized using consensus peptides, variants of 44 of these were recognized by at least one individual (and up to 11 individuals). There was no significant difference in the magnitude of the responses elicited for the consensus or for a variant (p = 0.70).
We compared the sequence of the peptides
recognized by ELISpot to
The graph shows the magnitude of ELISpot responses as a function of the number of mutations between the peptide tested and the individual's autologous HIV-1 consensus sequence. This analysis was performed for the five individuals in our cohort from whom HIV-1 sequences were available.
We next examined the relationship between the
ELISpot recognition and the database variability
of the HIV-1 segments that were recognized. We
calculated the average Shannon Entropy for each
10-AA segment over the Nef protein based on an
alignment of 514 known, independent sequences. The
HIV-1 segments that were recognized were more
conserved (i.e., with lower Shannon Entropy) than
those that were never recognized (
Thus, we chose an alternative way to evaluate the
impact of HIV-1diversity on ELISpot reactivity
using a peptide-specific metric. Based on 514
independent Nef HIV-1 subtype B sequences, we
calculated the population frequency of each unique
peptide derived from this dataset, i.e., the
percentage of sequences with the specific 10-mer.
We found no relationship between the population
frequency of a 10-mer and CTL targeting (
The frequency of each unique 10-mer in a dataset of 514 independent Nef HIV-1 subtype B sequences is shown. Peptide reactivity data generated by IFN-γ ELISpot assays done on 26 HIV-1 infected individuals were used to analyze the relationship between the population frequency of a peptide and either its frequency of recognition in the cohort (7A – left panel) or the magnitude of the IFN-γ ELISpot responses γ ELISpot response it elicited (7B – right panel).
It is particularly striking that peptides that
were rare in the population, i.e., found in less
than 5% of circulating HIV-1 sequences,
also elicited ELISpot responses in a number of
individuals. For example, a 10-mer found in
0.78% of circulating sequences was
recognized by 14 individuals (54% of our
cohort). There was a striking example of a peptide
(PGIRYP
The fraction of peptides that cross-react is represented as a function of the number of AA differences between the two peptides. For a pair of peptides (with the same HXB2 coordinates), the fraction of peptides that cross-react corresponds to the number of individuals who recognized both peptides divided by the number of individuals who recognized the more common one.
Using a variation-encompassing library of 944 10-AA-long
peptides that recapitulated much of the diversity
found among circulating HIV-1 subtype B Nef
peptides, we identified IFN-γ expressing
responses against 350 peptides, including all but
one of the known optimal epitopes in Nef, and
revealed many novel epitope specificities after
testing a set of only 26 HIV-1 infected individuals.
The peptide library consisted of consensus 10-mers
overlapping by 9 AA spanning the entire Nef protein
and multiple variants covering the range of Nef
diversity found in HIV-1-B infected individuals
– with up to 8 peptidic variants for a
consensus 10-mer. Our results suggest that
variability-inclusive vaccine antigens, such as
mosaic or COT+, can expand the breadth and
depth of CTL responses, as shown recently in
macaques
We demonstrated recognition of 297 Nef peptides in addition
to the 53 (optimally-defined or not) epitopes that
had previously been recorded in the HIVDB –
underlining that our understanding of the
determinants of epitope recognition is fairly
limited. It is particularly significant that this is
found for the oft-targeted Nef protein, suggesting
that our knowledge of epitopes in more variable
proteins such as Env is probably even more limited,
as previously indicated
Overall, reactive peptides were more likely to correspond to more conserved portions of the HIV-1-B Nef protein - the 350 peptides that elicited ELISpot responses corresponded to viral regions of significantly lower Shannon entropy than the HIV-1 regions encompassing the 594 peptides that were not recognized (p<0.0001). Although many CTL responses were detected towards segments of lower Shannon entropy, highly variable segments were also immunodominant targets – in such cases only certain variants of a 10-AA-long viral segment could be recognized. Therefore, we used a measure of diversity that was specific for each 10-mer variant corresponding to the frequency of occurrence of each peptide among circulating HIV-1-B sequences based on a dataset of 514 recorded Nef sequences obtained from HIV-1-B infected individuals (using only independent sequences). We found no relationship between the frequency of a peptide and either the frequency of recognition in our cohort or the magnitude of the responses elicited.
We found that using a variability-enhanced peptide set
increased the breadth and depth of CTL responses,
suggesting that variability-inclusive vaccine
strategies could elicit broader recognition of
epitopes. Indeed, use of mosaic antigens was
recently shown to enhance cellular immune responses
in vaccinated monkeys
Unique viral peptides correspond to private mutations and
represent a significant and growing proportion of
the peptides found among circulating sequences, due
to the extensive and continuously expanding
diversity of HIV-1. We found that 61 peptides found
in less than 1% of circulating sequences were
recognized by at least one individual in our cohort,
and 26 of these rare peptides were recognized by
three or more individuals (and up to 14
individuals). For example, the consensus epitope
(PGIRYPLTFG) was not recognized while its variant
PGIRYP
It remains to be understood what role is played by cross-recognition of peptides in the control of viral replication. Knowing HIV-1's propensity to mutate, it is likely that several variants are generated under CTL pressure and a number of cross-reactive responses may be remainders of immune responses against initial or previous viral variants – it is crucial to determine under which conditions those variants cross-react and whether this has an impact on the efficacy of the anti-viral CTL response. If a multiplicity of peptides still induced substantial CTL responses without significantly compromising viral fitness, there may be a high genetic barrier to abolish CTL recognition. Hence protection by such an epitope might be explained by the complex patterns of mutations that are necessary for efficient escape.
Whether cross-reactivity has an effect on the CTL's
ability to control viral replication is an open
question that has important implications for vaccine
design. If cross-reactivity can broaden the CTL
response elicited by a vaccine and also has a
positive impact on the control of viremia, intrinsic
cross-reactive specificities of HIV-1 should be
harnessed to develop a potentially more immunogenic
vaccine candidate as a means to confer broadly
protective immunity against multiple strains.
However, the lack of association between the
population frequency of an HIV-1 peptide and its
recognition by individuals in the cohort reported
here also suggests that an unrealistically large
vaccine antigen size may be required to protect
against the universe of viral strains capable of
establishing an infection. In addition, since a
number of responses were due to cross-reactivity
between rare and frequent peptides, we surmised that
a number of these very rare peptides, which are
unlikely to have been found in the viruses from our
infected subjects, may not induce efficacious
anti-viral responses, but rather represent decoy
responses. These results led us to discard our
variability-inclusive COT+ vaccine strategy
(TIFF)
(TIFF)
(TIFF)
(TIFF)
(XLS)