HVTN 505 trial

The HVTN 505 trial (ClinicalTrials.gov Identifier: NCT00865566, start date May 2009) enrolled 2504 Ad5-seronegative, fully circumcised men or transgender women in the US who have sex with men. Other eligibility criteria included being between the ages of 18 and 50 years old, having a history of unprotected anal intercourse with one or more male or male-to-female transgender partners or anal intercourse with two or more male or male-to-female transgender partners in the 6 months before randomization, being HIV-1 and HIV-2 seronegative, and having an alanine aminotransferase level no more than 2.5 times the upper limit of the normal range. A total of 2496 participants were randomized to receive the DNA/rAd5 vaccine (1251 participants) or placebo (1245 participants) at weeks 0, 4, 8, and 24. The original primary objective was to assess the effect of the DNA/rAd5 vaccine regimen on HIV-1 viral load setpoint. After study initiation, the protocol was updated to also assess the efficacy of the DNA/rAd5 vaccine regimen on prevention of HIV-1 acquisition. The final analysis was completed early in April 2013 due to reaching an interim monitoring guideline for lack of vaccine efficacy.

Ethics statement

All participants gave written informed consent. The study was approved by the institutional review board (IRB) at the Fred Hutchinson Cancer Research Center in Seattle, which acted as a central IRB through agreements with the Annandale, Aurora—University of Colorado Hospital CRS, Atlanta–Hope Clinic, Cleveland, East Midtown, New York–NYBC, New York–Physicians & Surgeons, NYBC–Bronx, Orlando, Philadelphia, San Francisco, and Union Square sites. For the other study sites, the study was approved by local IRBs (the National Institutes of Allergy and Infectious Diseases IRB at the Bethesda site, the Institutional Review Board for Human Use/University of Alabama at Birmingham at the Birmingham site, the Partners Human Research Committee at the Boston (Brigham) site, the Fenway Health Institutional Review Board at the Boston (Fenway) site, the Office for the Protection of Research Subjects at the Chicago site, the University of Texas Southwestern Medical Center Institutional Review Board at the Dallas site, the Fenway Health Institutional Review Board at the Fenway site, the Institutional Review Board of Human Subject Research at the Houston site, the AIDS Research Alliance Institutional Review Board at the Los Angeles site, the Office of the Human Research Protection Program Medical Institutional Review Board at the Los Angeles–Care Center site, the Vanderbilt University Institutional Review Board at the Nashville site, and the University of Rochester Internal Biosafety Committee at the Rochester site).

HIV-1 vaccine and reference sequences

Vaccine insert and reference sequence accession numbers are given in S1 Supplementary Methods. Notably, the env sequence, in the subtype B Ad5 vector only, had a deletion of the region encoding amino acids 134–191 (HXB2 coordinates), encompassing the V1V2 loop [12].

HIV-1 sequencing

HIV testing was performed at immunization months 0, 1, 2, 2.5, 6, 7, and 9 and then every 3 months through 24 months. The earliest RNA-positive sample was used for HIV-1 sequencing. Near full-length or overlapping half genome sequences were obtained from plasma samples by limiting dilution PCR and direct sequencing [24]. Sequences were assembled using Sequencher™, version 5.0 (Gene Codes Corporation, Ann Arbor MI), aligned using Geneious (Auckland, NZ) and manually edited; hypermutated sequences were identified using Hypermut [25] and removed before analysis. Sequences have been deposited in Genbank under accession numbers xxx…The alignments can be explored with an interactive web-based visualization and database tool (http://sieve.fredhutch.org/viz) [26].

Estimated time since the onset of infection

Estimated time since the onset of infection was defined as the mid-point between the last RNA negative and first RNA positive sample, the latter usually being the draw date for HIV-1 sequencing; if the latter was not the draw date for HIV-1 sequencing, the time between the first positive sample and the draw date for HIV-1 sequencing was added onto the estimated time since onset of infection as calculated by the mid-point method described above. Spearman correlation between the estimated time since the onset of infection and mean pairwise diversity adjusted for treatment and multiplicity of founders was computed using the partial.Spearman method of the PResiduals package in R.

Sequence analysis

All nucleotide sequences from an individual were compared to identify the mindist, i.e., the genome sequence closest to the individual’s consensus (near-full-length or two overlapping halves) genome. Nucleotide sequences were translated to amino acid sequences for protein-based analyses. Co-receptor, positive selection evaluation, potential N-linked glycosylation site identification, maximum-likelihood phylogenetic tree reconstruction, and founder multiplicity estimation are described in S1 Supplementary Methods.

CTL epitope predictions

CTL epitopes were predicted in all breakthrough, vaccine insert, and HIV-1 reference sequences based on each individual’s human leukocyte antigen (HLA) type [27]; see S1 Supplementary Methods for further details.

Site scanning methods

For all scanning methods, only sites passing variability and alignment quality filters, defined using treatment-blinded data, were included. A site was deemed sufficiently variable for the site scanning methods if 4 or more participants had a sequence matching the reference strain (vaccine insert or B.Anc) and 4 or more participants had a sequence mismatching the reference strain. An alignment quality filter was used to determine sites of sufficient confidence in the multiple sequence alignment. Because highly variable regions are poorly aligned and common descent cannot be established unequivocally, highly variable regions in the signal peptide and envelope variable loops were excluded. The pseudo F (PF) test statistic was used to test for a difference in mindist sequence distributions between the vaccine and placebo groups without using a reference sequence. For further details, see S1 Supplementary Methods.

Immunologically relevant AA subsets (e.g., 9-mers or mAb footprints, see S1 Supplementary Methods for more details) were scanned as described [24]. For a given subset of interest, the similarity score was the number of AA matches with the reference subset over all sites. A marginal mean model fit by generalized estimating equations (GEE) was used to compare similarity scores between treatment groups using all breakthrough sequences; this method accounts for within-individual correlation among multiple HIV-1 sequences and bias-corrects the GEE standard errors to account for the limited sample size [28].

Machine learning sieve analysis

Supervised learning approaches were applied to predict the treatment assignments of participants based on HIV-1 sequence features corresponding to: (a) mAb contact sites in Env; (b) two Env peptide binding hotspots; (c) the V3 region of Env; (d) the combined CD4 binding site (CD4bs) antibody footprints; (e) the gp145 region of Env corresponding to the vaccine inserts; (f) each of the other insert proteins separately (Gag/Pol/Nef); (g) each of the remaining non-insert regions separately (including the region of Env not included in the insert); and (h) all of the proteins combined, insert and non-insert. Machine learning analysis details are given in S1 Supplementary Methods.

Vaccine efficacy by distance

A “distance-specific hazard ratio” was defined as the ratio (vaccine/placebo) of the incidence of infection with a specific genotype of HIV-1, with genotype defined by the value of a Hamming distance between the aligned mindist sequence and the subtype B vaccine sequence [29]. These ratios were defined for Hamming distances based on three subsets of Env-gp120 sites (see Results). Methods for estimating the distance-specific hazard ratio with a 95% confidence interval, and for a Wald test of whether the distance-specific hazard ratio curve varies with distance, are described in [30].

Multiplicity adjustment

Adjustment for multiple hypothesis testing considered both false discovery rate (FDR) adjustment using the Benjamini-Hochberg procedure [31] and family-wise error rate (FWER) adjustment using the Holm-Bonferroni procedure [32]; multiplicity adjusted P-values were computed using the p.adjust method of the R computing language. FDR adjusted P-values are reported as Q-values. Adjustment was performed for the set of P-values generated by each test. For gene or protein level tests this included one P-value for each HIV-1 gene or protein. For AA site- and set-scanning, multiplicity adjustment was performed separately for each gene/reference or protein/reference combination.

All P-values reported are 2-sided. For hypothesis generation, results with unadjusted P-value < 0.05 are reported, along with Q-values and FWER-adjusted P-values. Results with Q-values < 0.20 were considered statistically significant.

Global sieve effects: Lower intra-host diversity and greater divergence from the Env-gp120 vaccine insert among vaccine recipient sequences

Though not statistically significant, the frequency of infections with multiple founder variants was two-fold lower in the vaccine vs. placebo recipients– 5 of 26 (19%) in the vaccine group compared to 8 of 20 (40%) in the placebo group (Fisher’s exact test, P-value = 0.19; Fig 1). Intra-host amino acid sequence diversity was lower in vaccine recipients than in placebo recipients among Env-gp120, Gag, and Pol sequences (all included in the vaccine) as well as Vif, which was not included in the vaccine (P ≤ 0.04, Q-values ≤ 0.09) (Fig 2). This difference was wider when the two dually-infected vaccine recipients were excluded (P ≤ 0.01) (S3 Table). No significant differences were found in the other HIV-1 proteins, including Env-gp41 and Nef (which were included in the vaccine) and Rev, Tat, Vpr, Vpu. Fig 2 shows that results were comparable whether or not individuals infected with multiple founders were included. No analyses were restricted to the subset of participants with multiple founders as the limited number of subjects and sequences from each founder together with the absence of critical vaccine/placebo patterns segregating the founders precluded formal analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Intra-host mean pairwise diversity among amino acid sequences.

Intra-host mean pairwise diversity values were calculated among all the sequences from a given participant separately for each protein corresponding to a vaccine insert, with vaccine recipients in red and placebo recipients in blue. The upper panel included all participants; the lower panel included only participants with single HIV-1 founder viruses. Two-sided Mann-Whitney p-values comparing the vaccine and placebo groups are shown above the panels.

https://doi.org/10.1371/journal.pone.0185959.g002

We next assessed vaccine pressure on HIV-1 by comparing pairwise distances of breakthrough sequences from the vaccine insert sequence(s) between the treatment groups. Greater divergence from the vaccine Env-gp120 sequences was observed in vaccine recipients. This difference was significant for the subtype B immunogen–the same subtype as the infecting sequences (mean 0.307 in the vaccine group and 0.285 in the placebo group, P = 0.01, Q-value = 0.12)–and non-significant when compared to the more distant subtype A and C Env-gp120 immunogens (S4 Table). After removing variable segments this difference was smaller in magnitude but slightly more significant (mean 0.222 in the vaccine group and 0.210 in the placebo group, P = 0.004, S6 Table, Fig 3). Similar results were obtained for tree-based distances but were only significant for the subtype B vaccine after removing variable segments (S5 and S6 Tables).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Pairwise distances between breakthrough sequences and each Env vaccine sequence or Env Consensus B.

Participant-specific mean distances of breakthrough sequences to the subtype A, B, and C Env vaccine sequences and the Consensus B sequence (labeled VRC-A, VRC-B, VRC-C and Cons.B, respectively). Distances from breakthrough sequences from vaccine recipients are in red and placebo recipients in blue. Distances are based on Env-gp120 sequences with the hypervariable segments deleted (HXB2 positions from signal peptide 10–16; V1 loop 132–152; V2 loop 186–191; V3 loop 309–310; V4 loop 392–413; and V5 loop 460–464). Two-sided Mann-Whitney p-values comparing the vaccine and placebo groups are shown above the panels.

https://doi.org/10.1371/journal.pone.0185959.g003

Local sieve effects: Specific AA sites and linear peptide sequences in breakthrough viruses differentiated treatment groups

Local scanning sieve methods were used to identify the specific AA sites or sets of AA sites underlying the global sieve effect signal found in Env-gp120. First, five AA site-scanning methods [Gilbert, Wu, Jobes (GWJ) [35]; Expected Gilbert, Wu, Jobes (EGWJ) [15]; Model-Based Sieve (MBS) [21]; Physico-Chemical Property (PCP) [36]; Quasi-Earth Mover’s Distance (QEMD) [15] were used to compare each site to the subtype B vaccine insert. Two signature Env-gp120 sites were identified: site 133 was enriched for charged residues among placebo group sequences by PCP (P = 0.00015, Q-value = 0.19) and site 429 showed more mismatches against the vaccine insert in the vaccine group than in the placebo group by MBS (P = 0.0014, Q-value = 0.15) (S7 Table). In addition, analysis of within subject positive selection in Env-gp120 identified eight Env-gp120 sites under positive selection only in the vaccine group: sites 147, 183, 236, 290, 315, 354, 388, and 442 (S8 Table).

A k-mer scanning method was used to identify signature 9-mer peptides in Env-gp120. Twenty-two overlapping 9-mers had significant treatment group differences in their pairwise distance from the subtype B vaccine insert (Table 1). These 9-mers specified six unique linear signature regions corresponding to AA positions 27–37, 86–94, 192–207, 365–373, 425–437, and 464–479 (P < 0.005, Q-values < 0.18) (Table 1; S2, S3, S4, S5, S6 and S7 Figs). Notably, the last four regions overlapped CD4 binding site contact residues [37], all with greater divergences in the vaccine group. Since these 9-mers also corresponded to potential CTL epitopes, we performed sieve analyses of CTL epitope repertoires defined in silico based on each participant’s HLA genotype. No significant sieve effects associated with CTL epitopes were found (S9, S10, S11, S12, S13 and S14 Tables; S8 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. k-mer signature regions.

https://doi.org/10.1371/journal.pone.0185959.t001

Finally, the same k-mer scanning method was used to evaluate 44 sets of known mAb contact sites [15], 19 of which showed significant vaccine/placebo differences (Fig 4; S15 Table). CD4 binding site (CD4bs) antibodies bind to conserved epitopes in gp120 near the CD4-binding site and can compete with the CD4 receptor for binding to gp120 [38], whereas CD4-induced (CD4i) antibodies bind to conserved epitopes on the inner domain of gp120 that are normally hidden but are exposed by CD4 binding [39, 40]. All CD4bs (n = 15) mAb footprints and two of the five CD4i mAb footprints had significantly greater divergences from the subtype B vaccine insert in the vaccine group, as did 2 of the 14 V3 mAb contact site sets (PGT128 and PGT122, Q-value ≤ 0.2). The other 3 CD4i footprints and 12 V3 mAb contact site sets also had greater divergences from the vaccine insert in the vaccine group, although these signatures were not statistically significant. There was no evidence for a sieve effect in the remaining mAb classes (Quaternary-, Env-gp41-, and Glycan-targeted; P-values > 0.3) (S15 Table). Application of the phylogenetic divergence analysis presented above to the mAb contact site sets found significantly larger tree-based distances in the vaccine group for the combined CD4bs contact sites (mean 0.305 in the vaccine group and 0.271 in the placebo group, P = 0.041; P = 0.075 for pairwise distances) and no significant results for other subsets of mAb contact sites (CD4i-, V3-, Quaternary-, Env-gp41-, and Glycan-targeted; P ≥ 0.16). Since a previous analysis determined that antibodies induced by the DNA/rAd5 vaccine regimen are not broadly neutralizing [41], we investigated the antibody-dependent cellular phagocytosis (ADCP) activity of antibodies from participants in the HVTN505 trial. As shown in S9 Fig, the DNA/rAd5 vaccine regimen induced positive ADCP responses to ConSgp140-coated beads in an estimated 92.5% of vaccinees at 4 weeks post-final vaccination.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Overlap between HIV-1 Env-gp120 k-mer, AA site and mAb footprint signatures identified with different scanning methods.

Positive sieve effects, i.e., greater divergences of sequences from vaccine recipients than from placebo recipients from the vaccine strain, were detected for all 34 CD4bs, CD4i and V3 footprints (S15 Table). The figure represents signatures detected with a Q-value ≤ 0.2 across Env-gp120 (gray bands identify the HIV-1 features labeled at the top of the figure): six k-mer regions are shown as orange bands (labeled a-f as in Table 1); two individual sites, 133 and 429, are labeled in bold on the x-axis; and the 19 of 34 CD4bs, CD4i, and V3 mAb footprints with a Q-value ≤ 0.2 are indicated by dark blue points and labeled on the y-axis in bold font. The 15 mAb footprints that are not statistically significant signatures are indicated by light blue points and labeled on the y-axis in gray. Footprint site sets were grouped by mAb class (CD4bs, CD4i, and V3) and ranked within class by P-value. Footprint site sets for mAb classes where no significant effect was identified are not shown (Quarternary, Env-gp41, and Glycan; see S15 Table).

https://doi.org/10.1371/journal.pone.0185959.g004

Site-specific and k-mer scanning methods were also used to test for AA distribution differences between treatment groups in the remainder of the HIV-1 proteome. The reference sequence for comparisons was the subtype B vaccine insert sequence for Gag, Pol, and Nef and the Ancestral B (B.Anc) sequence for proteins not included in the vaccine (Rev, Tat, Vif, Vpr, and Vpu). Three of the site-specific scanning methods identified Pol 238 as a signature site (GWJ, EGWJ, and QEMD; P = 0.0006, 0.0002, and 0.0006, respectively; Q-values = 0.11, 0.05, and 0.11, respectively; S7 Table). The k-mer scanning methods identified 11 9-mers, all in Pol, that significantly differed between the vaccine and placebo groups when compared to the subtype B vaccine insert (Table 1). After accounting for 9-mer overlap, three unique Pol linear signature regions remained: 230–246, 353–361, and 867–877 (P < 0.005, Q-values < 0.17) (Table 1). The presence of the Pol 238 signature site in the 230–246 linear signature region may explain the identification of this linear signature region. When site subsets identified in previous sieve analyses of the Step and RV144 trials were assessed in the HVTN 505 sequences, no sites showed significant treatment group differences (GWJ: P ≥ 0.18; EGWJ: P ≥ 0.10). The pseudo F test found significant sequence differences only for Vif (P = 0.002, FWER-adjusted P = 0.02). Machine learning analyses considering Gag, Pol, and Nef were not able to classify sequences by vaccine vs. placebo group above the pre-set performance threshold [cross-validated area under the receiver operating characteristic (ROC) curve (AUC) value > 0.7] (S16 Table).

Env-gp120 vaccine similarity associated with genotype-specific vaccine efficacy

While the HVTN 505 trial had no overall vaccine efficacy, it is possible that vaccine efficacy calculated separately against different HIV-1 genotypes was positive for some genotypes (vaccine benefit to lower HIV-1 risk) and negative (vaccine detriment to increase HIV-1 risk) for other genotypes. Both global and local sieve effects were found in Env-gp120, raising the hypothesis that the vaccine preferentially blocked infection of HIV-1 Env-gp120 variants more similar to the vaccine sequence. A hazard ratio (vaccine/placebo) of HIV-1 infection was estimated by stratifying breakthrough sequences based on their distance to the subtype B insert calculated for three sets of Env-gp120 sites: i) all Env-gp120 sites aligned with confidence (n = 432, 64.6% of sites); ii) CD4bs antibody contact sites (n = 93); and iii) sites corresponding to the 4 linear signature regions overlapping the CD4bs where significant sieve effects were found (n = 54; Table 1). The hazard ratio significantly increased with the distance to the subtype B insert for each set (Env-gp120: P = 0.0077; CD4bs antibody contact sites, P = 0.012; CD4bs-related linear signature regions, P = 0.00046; Fig 5). The estimated hazard ratio was approximately 0.20 for HIV-1 sequences that were most similar to the subtype B vaccine insert (which corresponds to 80% vaccine efficacy against these sequences), approximately 1.0 for HIV-1 sequences with intermediate distances (zero vaccine efficacy), and exceeded 2.0 (doubling of risk by vaccination) for HIV-1 sequences that were most divergent from the vaccine (Fig 5). If the measured Env-gp120 distances reflect the distances of the transmitted viruses (which cannot be empirically verified from HVTN 505 data as noted in the Introduction), then these findings suggest that the vaccine may have selectively blocked HIV-1 infection when Env-gp120, especially in the CD4bs region, was close to the vaccine sequence.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Distance-specific vaccine efficacy.

Boxplots of hamming distances of mindist sequences to the subtype B vaccine strain by treatment group for three sets of Env-gp120 sites: (A) all Env-gp120 sites that could be aligned with confidence (n = 432); (C) the 93 CD4bs antibody contact sites; and (E) 54 sites corresponding to the 4 k-mers overlapping the CD4bs where significant sieve effects were found (Table 1). The mid-line of the box denotes the median and the ends of the box denote the 25^th and 75^th percentiles. The whiskers that extend from the top and bottom of the box extend to the most extreme data points that are no more than 1.5 times the interquartile range or if no value meets this criterion, to the data extremes. (B, D, F) Distance-specific hazard ratio based on the corresponding distances shown in panels A, C, and E, respectively. The solid black lines show the estimated hazard ratio as a function of distance, and the dashed red lines represent a 95% point-wise confidence interval (CI) around that estimate. The ‘Sieve test’ 2-sided p-value reports the result of the test of [30] for whether the hazard ratio varies with distance.

https://doi.org/10.1371/journal.pone.0185959.g005