Mosaic vaccine-induced antibody-dependent cellular phagocytosis associated with delayed HIV-1 viral load rebound post treatment interruption

SUMMARY

A heterologous Ad26/MVA vaccine was given prior to an analytic treatment interruption (ATI) in people living with HIV-1 (mainly CRF01_AE) who initiated antiretroviral treatment (ART) during acute HIV-1. We investigate the impact of Ad26/MVA vaccination on antibody (Ab)-mediated immune responses and their effect on time to viral rebound. The vaccine mainly triggers vaccine-matched binding Abs while, upon viral rebound post ATI, infection-specific CRF01_AE binding Abs increase in all participants. Binding Abs are not associated with time to viral rebound. The Ad26/MVA mosaic vaccine profile consists of correlated non-CRF01_AE binding Ab and Fc effector features, with strong Ab-dependent cellular phagocytosis (ADCP) responses. CRF01_AE-specific ADCP responses (measured either prior to or post ATI) are significantly higher in individuals with delayed viral rebound. Our results suggest that vaccines eliciting cross-reactive responses with circulating viruses in a target population could be beneficial and that ADCP responses may play a role in viral control post treatment interruption.

Graphical Abstract

In brief

Mdluli et al. analyze antibody-mediated responses in participants who received the Ad26/MVA mosaic vaccine before an analytic treatment interruption (ATI). Mosaic vaccination elicits antibody binding responses to multiple subtypes but did not optimally boost the infection-specific CRF01_AE responses. High ADCP responses associate with delayed viral rebound upon ATI.

INTRODUCTION

The initial stages of HIV-1 infection yield a peak in viremia, with millions of HIV-1 RNA copies per milliliter of plasma,¹ and result in the establishment of a latent HIV-1 reservoir.² Antiretroviral therapy (ART) blocks HIV-1 replication, leading to undetectable levels of HIV-1 RNA in plasma after a few weeks of treatment in people living with HIV-1 (PLWH).³ While long-term ART suppresses HIV-1 replication, HIV-1 persists in resting CD4⁺ T cells.^4-6 The reservoir is stable, with a half-life estimated at 44 months,^7-9 and its persistence is in part due to the proliferation of infected cells.^10,11 If ART is interrupted, then HIV-1 will replicate exponentially, leading to rebound viremia.^12,13

Analytic treatment interruption (ATI) studies, in which participants interrupt treatment in a closely monitored setting, can help define interventions that would allow viral control without ART.¹⁴ ATI studies have relied on multiple strategies to target the reservoir and boost HIV-1-specific immune responses with the goal of identifying features that potentially induce delayed viral rebound and, ultimately, ART-free remission. The RV254 study in Bangkok, Thailand,¹⁵ enrolled 714 participants (as of December 31, 2022) during acute HIV-1 infection, and they initiated ART within days of diagnosis. After a median of 3 years of viral suppression (range 1.78–6.57 years), 67 of these PLWH participated in one of four ATI studies. Plasma viral loads rebounded in participants at a median of 28 days.^16-20 In a pooled analysis of the four ATI trials, we showed that lower viral loads in acute HIV-1 infection (prior to ART) and faster time to viral suppression upon ART initiation are associated with delayed viral rebound upon ATI, while clinical variables measured immediately prior to the ATI did not associate with time to rebound.²¹ In addition, no difference was observed in time to viral rebound across Fiebig stages, with median times to rebound to ≥1,000 copies/mL ranging between 24 and 30 days.

One of the four ATI studies, RV405, tested a therapeutic HIV-1 adenovirus serotype 26 and modified vaccinia Ankara (Ad26/MVA) mosaic vaccine regimen prior to ATI.¹⁸ A post-hoc analysis that excluded a placebo group controller with HLA-B*57:01:01, an allele known to be associated with improved clinical outcome, showed a modest delay in rebound in vaccine recipients (median 28, range 13–46 days) compared to placebo recipients (median 20, range 13–24 days) (p = 0.003).

We conducted a systems serology analysis using samples collected from 26 RV405 participants (17 Ad26/MVA mosaic vaccine recipients and 9 placebo recipients). Characterizing antibody-specific responses from acute infection until viral rebound post treatment interruption allowed assessment of the impact of Ad26/MVA vaccination. Our objectives were to test whether a vaccine-specific immune profile could be identified and whether specific responses to vaccination were associated with time to viral rebound post ATI. We showed that therapeutic Ad26/MVA mosaic vaccination preferentially elicited mosaic-specific antibody responses and that vaccine-induced ADCP responses were associated with a delay in time to viral rebound.

RESULTS

Study participants

Our study included 26 RV405 participants (17 Ad26/MVA mosaic vaccine recipients and 9 placebo recipients) enrolled in Bangkok, Thailand (one vaccine participant was excluded at the request of the Thai Ministry of Public Health Ethics Committee). HIV-1 sequencing at HIV-1 diagnosis showed that most infections were established with circulating recombinant form CRF01_AE viruses (n = 21) and CRF01_AE/B recombinants (n = 5); there was a CRF01_AE/subtype B co-infection.¹⁸ Considering the envelope (env), two of the participants with CRF01_AE/B recombinants had env corresponding to subtype B (Figure S1). We analyzed 169 samples collected at six time points between acute HIV-1 infection and until after viral rebound post ATI (Figure S2). At HIV-1 diagnosis in acute infection, the median (interquartile range [IQR]) plasma viral load was 5.94 log10 (IQR 5.31, 6.7) copies/mL, and the median CD4 count was 297 (IQR 231, 446) cells/mm³ (Table S1). Samples were then analyzed at week 60 after ART initiation and at a time point that ranged between 2.3 and 6.4 years of viral suppression on ART (median = 3.1), which corresponded to immediately before the first vaccination. The vaccination series consisted of two doses of a recombinant replication-deficient Ad26 vectored vaccine at weeks 0 and 12, followed by two doses of a recombinant live attenuated MVA virus-vectored vaccine at weeks 24 and 48. The transgenes in the Ad26 vector vaccine encoded a mosaic Env sequence (Ad26.Mos1.Env) and two mosaic inserts of Gag and Pol sequences (Ad26.Mos1.Gag-Pol + Ad26.Mos2.Gag-Pol). The transgene in the MVA vector encoded two mosaic Gag, Pol, and Env sequences (mosaic 1 and mosaic 2). Additional samples were analyzed at week 50 after the first vaccination (2 weeks after the last vaccination), which corresponded to peak vaccine immunity; at week 60 post first vaccination, just prior to the initiation of the ATI; and upon viral rebound. Viral rebound after ATI was delayed by 8 days in the vaccine group, with a median peak viral load after rebound of 4.6 log₁₀ (IQR 4.5, 4.9) in the vaccine group compared to 4.9 log₁₀ (IQR 3.7, 5.0) copies/mL in the placebo group. Considering the relatively small difference in time to viral rebound between vaccinees and placebo recipients, we used two groups (early and delayed rebound) with a cutoff of 28 days (corresponding to the median time to rebound in a prior study where data from four ATI studies were integrated²¹). Given this cutoff, 10 vaccine recipients were in the delayed viral rebound group, while 7 vaccine and all placebo participants were in the early viral rebound group (Table S1). When considering only vaccine recipients, the median time to rebound was 20.5 days for the early group vs. 32.5 days for the delayed viral rebound group. When considering all participants, the median time to rebound was 20 days for the early group vs. 32.5 days for the delayed viral rebound group.

Binding Ab responses increased or decreased in the 60 weeks post ART initiation, depending on the Fiebig stage at HIV-1 diagnosis

We characterized HIV-1-specific binding antibodies in 169 plasma samples (Figure S2) using a bead-based immunoassay designed to measure antibody isotype, subclasses, and Fc gamma receptor (FcγR) responses against two Env gp120 and nine Env gp140 antigens corresponding to the mosaic 1 component of the vaccine and subtypes A, B, C, D, and G; CRF01_AE and 02_AG; and group M (gp120 antigens correspond to the inserts in the RV144 vaccine). We focused on Env responses, as Env is the primary target of HIV-1-specific antibody responses. At HIV-1 diagnosis in acute infection (prior to ART initiation), samples from participants in Fiebig I/II had no measurable binding antibodies (Abs), while samples from participants in Fiebig III/IV showed significantly higher binding Ab responses (p = 0.02 for immunoglobulin G [IgG] against CRF01_AE) (Figure 1). Whether measured pre-ART or at week 60 on ART, IgG2-IgG4, IgM, and IgA binding responses showed little signal (Figure 1A). Participants who had lower Ab titers in acute infection pre-ART had the highest increase in titers. For participants who initiated ART during FI-FIII, IgG responses against the different Envs increased significantly between acute infection pre-ART and week 60 on ART (p = 0.02 for IgG against CRF01_AE) (Figures 1A and 1B), with similar trends for IgG1 and FcγR responses (Figure 1A). In contrast, binding Ab responses decreased in three of the five participants who initiated ART during Fiebig IV (Figures 1A and 1B). Since participants who had lower Ab titers in acute infection pre-ART had the highest increase in titers over time on ART, this resulted, at week 60, in similar Ab binding titers across participants despite ART initiation at different Fiebig stages (Figures 1A and 1B). Nonetheless, there was some variability in the Ab binding responses across participants, and this variability was associated with viral loads at HIV-1 diagnosis in acute infection. We calculated the total viral load area under the curve (AUC), using imputation to account for the missed viremia pre diagnosis, as described previously,²¹ and found that participants with the highest viral load (VL) AUC in acute infection had the highest Ab binding responses at week 60 (Figure S3A). Prior to vaccination, binding Abs had generally decreased proportionally to the length of time from week 60 on ART to date of the first vaccination, which ranged from 1 week to 4 years (Figure S3B). Overall, at week 60 on ART or pre vaccination, Ab binding titers were low, and there was no significant difference across participants based on Fiebig stage at ART initiation.

Figure 1. Binding Ab responses between acute infection (pre ART) and week 60 after ART initiation.

(A) Ab binding analyzed by Fiebig stage (FI/FII, n = 10; FIII, n = 11; FIV, n = 5) pre-ART and at week 60 on ART. Each feature is represented along the columns, and each participant is represented along the rows. Env antigens (2 gp120 and 9 gp140, from left to right) are color-coded by subtype. Data were Z scored scaled prior to analysis, setting the signal below assay background (shown as dashed lines in B) to null to avoid scaling noise.

(B) Differential dynamics in IgG gp140 binding Abs between pre-ART and week 60 based on the Fiebig stage at HIV-1 diagnosis. Shown are significant differences between binding Abs pre-ART and at week 60 on ART, using Wilcoxon signed-rank test. Asterisks represent false discovery rate (FDR)-adjusted significant differences: **p ≤ 0.01, *p ≤ 0.05. Dashed lines show the assay background set to 600 mean fluorescence intensity.

Binding Abs were higher in vaccine than in placebo recipients at peak immunity but not upon viral rebound

At peak immunity after vaccination, binding Ab responses were significantly higher in vaccine recipients compared to placebo recipients across all gp120 and gp140 antigens (p < 0.001 for IgG against mosaic 1 vaccine Env); similar trends were seen across IgG1-IgG3 and the FcγRs, while IgG4, IgA, and IgM showed little signal prior to vaccination and no increase after vaccination (Figures 2A and S4). Vaccine-induced responses tended to increase more for people with lower levels pre vaccination (R = −0.33, p = 0.099 for IgG against mosaic 1 vaccine Env). Vaccine-induced Ab binding responses did not differ based on the Fiebig stage at which participants initiated ART (Figure S5A). At the time of ATI (10 weeks after peak immunity), binding Abs had decreased significantly (p < 0.05 for IgG against mosaic 1, CRF01_AE, and group M gp140s) (Figure 2B) but remained significantly higher among vaccinees compared to placebo recipients (p < 0.0001). However, upon viral rebound, there was no significant difference in binding Ab titers between vaccine and placebo recipients (Figure 2A).

Figure 2. Higher binding Abs in vaccinees than in placebo recipients at peak immunity but not upon viral rebound.

(A) IgG binding Ab responses in vaccine (closed circles, n = 17) and placebo (open circles, n = 9) participants.

(B) Binding Ab dynamics for Ad26/MVA Mosaic vaccine recipients as fold over the pre-vaccination levels; week 0 (pre-vaccination), week 50 (peak immunity) and week 60 (ATI). Significant differences between binding Abs at peak immunity and ATI were evaluated using Wilcoxon signed-rank tests.

(C) Binding Ab dynamics are shown for vaccine and placebo recipients as fold over pre-ART baseline in acute HIV-1 infection; week 60 corresponds to week 60 on ART.

(D) Fc effector functions in vaccine and placebo participants, measured using a CRF01_AE antigen close to the CRF01_AE consensus (97.2% identity) and the mosaic 1 antigen. Point asterisks denote the two vaccine recipients with Env corresponding to subtype B. Binding Abs and Fc effector responses were compared between vaccine and placebo groups at pre-vaccination, peak vaccine immunity, ATI, and at peak VL (viral load) post ATI using Mann-Whitney U tests. Asterisks represent significant differences: ****p ≤ 0.0001, ***p ≤ 0.001, **p ≤ 0.01, *p ≤ 0.05.

In addition to the magnitude of the IgG responses, we analyzed their subtype specificity. In acute infection and through the pre-vaccination time point, IgG responses increased the most toward CRF01_AE antigens, reflecting that participants had Env corresponding to CRF01_AE (Figures 1B and 2C). The median fold change between pre-ART and week 60 on ART was 19.8 for CRF01_AE compared to 9.3 and 9.5 for the vaccine-matched mosaic 1 Env and subtype B antigens, respectively. In contrast, vaccination preferentially increased total IgG responses against vaccine-matched mosaic 1 Env and subtype B antigens, with a median fold change between pre-vaccination and peak immunity of 15.4 and 14.1, respectively (Figure S4B). The lowest median fold increases at peak immunity were seen for IgG responses toward CRF01_AE antigens, with a median fold change of 2.8 (measurements against four CRF01_AE antigens were combined). Median fold increases for other clades ranged between 5.5 (subtype C) and 12.2 (subtype G). Upon viral rebound, binding Abs toward CRF01_AE increased with the highest fold change from ATI, reverting to a profile with similar responses in the vaccine and placebo groups, as seen prior to vaccination (Figures 2A and 2C). The fold change for vaccine-induced binding Ab responses was high compared to the changes observed between diagnosis (pre ART baseline) and pre vaccination. Response kinetics were similar across antigens, whether comparing responses over pre-ART or pre-vaccination levels, but vaccine-induced responses toward mosaic 1 Env and subtype B antigens were the strongest (Figure 2C). We note that the two vaccine recipients with Env corresponding to subtype B did not have Ab binding profiles that distinguished them from vaccine recipients with Env corresponding to CRF01_AE (asterisks in Figure 2A). Compared to binding responses measured pre vaccination, both participants showed the lowest fold increase in CRF01_AE responses, particularly upon viral rebound.

We evaluated the functionality of Ab responses with four Fc effector assays: Ab-dependent complement deposition (ADCD), Ab-dependent cellular phagocytosis (ADCP), Ab-dependent natural killer cell activation (ADNKA), and Ab-dependent neutrophil phagocytosis (ADNP). Each assay was conducted with an antigen matching the vaccine insert mosaic 1 and a CRF01_AE antigen corresponding to the CM235 strain, tested in parallel. At peak immunity and at the start of the ATI (10 weeks later), Fc effector responses measured against the mosaic 1 antigen were significantly higher in the vaccine group than in the placebo group for the four functions (p < 0.05). The highest median fold changes were observed for ADCD (14.47 against the mosaic 1 antigen vs. 0.95-fold change against CRF01_AE) and ADNKA (2.98 against mosaic 1 insert vs. 1.24-fold change against the CRF01_AE antigen) (Figure 2D). When looking at responses against the CRF01_AE insert, only ADCP responses were significantly higher in the vaccine group than in the placebo group (p = 0.0035).

Upon viral rebound, mosaic responses from placebo participants increased to the level of those seen in the vaccine group for ADCP and ADNKA, while ADCD and ADNP remained higher in the vaccine group (p = 0.00054, p = 0.0025, respectively). For the CRF01_AE-specific responses, while responses tended to be higher in the vaccine group, only ADCD responses reached significance (p = 0.02) (Figure 2D) post ATI. Vaccine-induced increases in Fc effector functions did not differ by the Fiebig stage at which the participant initiated ART (Figure S5B). Similar to the binding responses, the two vaccine recipients with Env corresponding to subtype B did not present higher mosaic 1-specific Fc effector responses than the other participants with CRF01_AE Env (Figure 2D) (except for mosaic-specific ADCD at peak immunity and ATI and ADNP at viral rebound for the participant who initiated ART at Fiebig IV).

Vaccine profile defined by non-CRF01_AE binding Abs and Fc effector functions

We used random forest (RF) and partial least-squares discriminant analysis (PLSDA) to identify a comprehensive Ab binding and functional profile associated with vaccine recipients at peak immunity. First, RF was used to select Ab binding features; a minimum of seven binding features discriminated vaccinees from placebo recipients with an out-of-bag error rate of 3.85% (Figure 3A) (due to the high number of binding variables [n = 176] compared to the number of functional responses [n = 8], RF was used only on binding variables to avoid overweighting binding features). The selected total IgG and IgG1 responses were directed toward CRF02_AG, group M, subtypes C and D antigens; these responses were strongly correlated to subtypes B and G and the mosaic 1 vaccine insert but only weakly correlated to CRF01_AE (Figure S6). Second, the selected binding features were combined with Fc effector functions in a PLSDA classification model, which discriminated vaccine recipients from placebo recipients (Figure 3B) (cross-validation showed that the PLSDA model was significantly better than a null model; p < 0.0001; Table S2). Differences were largely captured along the first latent variable (LV1), which accounted for 51.6% of the variance across treatment groups. The Fc effector functions ADNKA and ADCD toward the mosaic 1 vaccine insert and ADCP toward CRF01_AE were among the top discriminatory features along LV1 (Figure 3C) (PLSDA performed either on the RF-selected binding features or on the Fc effector variables showed that both models were able to independently discriminate between the groups; Figure S7). A correlation network was used to link the Ab binding features with Fc effector functions (Figure 3D). All binding Ab features that were strongly associated (Spearman rho > 0.9, p < 0.05) with the RF-selected features were included. Although the RF model selected only total IgG and IgG1 responses, these responses were strongly correlated to FcγR2 and FcγR3 responses. The top discriminatory Fc effector functions toward the mosaic 1 vaccine insert, ADNKA and ADCD, showed high network connectivity with Ab binding features. Other Fc effector functions were weakly connected (mosaic 1-specific ADNP) or not associated with Ab binding features (ADCP). CRF01_AE-specific ADCP was also significantly higher in vaccinees (p = 0.0035) and found among the top discriminatory features in the PLSDA model but did not associate with Ab binding features.

Figure 3. Ad26/MVA mosaic vaccine profile consisted of correlated non-CRF01_AE binding Aby and Fc effector features.

(A) Random forest (RF) analysis showing features associated with Ad26/MVA mosaic vaccination.

(B) PLSDA model separated vaccine (blue) and placebo (black) recipients in the scores plot.

(C) PLSDA model features along latent variable 1 (LV1), ranked by their importance and colored by HIV-1 clade.

(D) Correlation network linking binding Abs and Fc effector responses. Binding features included in the network were those selected through RF analysis and those highly correlated (Spearman rho > 0.9, p < 0.05) to RF-selected features.

Increased CRF01_AE-specific ADCP responses associated with delayed viral rebound

To investigate the impact of the vaccine profile on rebound post ATI, we compared Ab binding responses and Fc effector functions between participants who experienced early or delayed viral rebound after ATI, using a cutoff of 28 days (this cutoff corresponded to the median time to rebound in a prior study where data from four ATI studies were integrated²¹). Ab binding feature selection with RF modeling followed by a PLSDA classification model was used to identify the minimum number of Ab features that best distinguished the two groups. Focusing on vaccine recipients (n = 7 early and n = 10 delayed viral rebound) and binding Ab features, models were generated at peak immunity, at the time of the ATI, and at peak VL post ATI. The RF selection identified only two binding Ab features at peak immunity and only the top node at ATI and peak VL post ATI. However, the model performance was low, with out-of-bag errors of 23.5%, 47%, and 43.8% for peak immunity, ATI, and post ATI, respectively. These high error rates indicate that binding Ab responses failed to discriminate between participants who experienced early versus delayed viral rebound (Figures S8A-S8C). This was further confirmed by a volcano plot, which showed that none of the binding Ab features were significantly different between the two groups (Figures S8D-S8F).

We next combined the RF-selected Ab binding features with Fc effector functions in a PLSDA. The models showed a slight separation of vaccine recipients who experienced early vs. delayed viral rebound across LV1 and LV2, which accounted for 56.7%, 39%, and 59.3% of the total variance at peak immunity, ATI, and post ATI, respectively (Figures 4A-4C); cross-validation showed that the PLSDA models were significantly better than null models (p < 0.0001; Table S2). The loading plots confirmed that delayed viral rebound was associated with elevated Ab functionality toward the CRF01_AE antigen CM235, which is close to the CRF01_AE consensus sequence (97.2% amino acid sequence identity) (Figures 4D-4F). Particularly, CRF01_ AE-specific ADCP responses showed significantly higher levels in vaccine recipients with delayed viral rebound compared to the early rebound group, with p = 0.019, p = 0.033, and p = 0.033 at peak immunity, ATI, and post ATI, respectively (Figures 4G-4I). Other Ab functions toward the CRF01_AE antigens also clustered with the delayed rebound group (Figures 4D-4F), although these responses were very low and not boosted by vaccination and, thus, not sufficient to fully distinguish the groups. Similar results were obtained when the analysis included both vaccine and placebo recipients (n = 15 early and n = 10 delayed viral rebound) (Figure S9; Table S2). CRF01_ AE-specific ADCP was consistently associated with participants with delayed rebound in the PLSDA model (Figures S9A-S9F) and was significantly higher in the delayed viral rebound group, with p < 0.001, p = 0.0025, and p = 0.041 at peak immunity, ATI, and post ATI, respectively (Figures S9G-S9I). A PLSDA model generated prior to vaccination did not show a clear separation between the delayed versus early group, and pre-vaccination CRF01_AE-specific ADCP responses were not significantly different (p = 0.76).

Figure 4. Increased CRF01_AE-specific ADCP responses associated with delayed viral rebound in vaccine participants.

(A–C) PLSDA models showing a slight separation between vaccine recipients with early (dark green open circles) vs. delayed (red filled circles) viral rebound in the scores plot at peak immunity (A), ATI (B), and peak VL post ATI (C).

(D–F) PLSDA model features colored by HIV-1 clade where the relative location of individual features associates with the corresponding group (early/delayed rebound) in the scores plot at peak immunity (D), ATI (E), and peak VL post ATI (F).

(G–I) CRF01 AE-specific ADCP responses in vaccine recipients with early versus delayed viral rebound at peak immunity (G), ATI (H), and peak VL post ATI (I).

(J–L) Kaplan-Meier survival curves for CRF01 AE-specific ADCP responses at peak vaccine immunity (J), ATI (K), and peak VL post-ATI (L). Categories were divided using maximally selected rank statistics at each time point.

The CRF01_AE-specific ADCP responses that were associated with delayed viral rebound were induced by Ad26/MVA mosaic vaccination, highest at peak immunity (week 50), and were maintained at the time of ATI (week 60) (Figure 2D). However, ADCP did not appear to correlate with any specific binding Ab responses (Figure 3D). We visualized the association between CRF01_AE-specific ADCP responses and time to rebound using Kaplan-Meier curves. The curves showed that participants with higher ADCP responses (categorized using the maximally selected rank statistics method) had delayed rebound with significant differences at ATI (p = 0.031) when considering vaccinees only (Figures 4J-4L) and at peak vaccine immunity (p = 0.001) and ATI (p = 0.001) when all participants were included (Figure S9J-S9L).

DISCUSSION

We conducted a comprehensive systems serology analysis to characterize Ab-specifc immune responses in 26 participants who primarily acquired CRF01_AE viruses, initiated ART in acute infection, and were randomized to receive either mosaic vaccination or a placebo before undergoing an ATI. Samples were analyzed at six time points spanning from acute infection immediately before ART initiation until viral rebound after the ATI. We showed that (1) mosaic vaccination preferentially increased mosaic-specific Env Ab binding responses rather than CRF01_AE-specific responses and that (2) CRF01_AE-specific ADCP responses induced post vaccination were associated with delayed viral rebound post ATI.

Our first key finding was that Ad26/MVA mosaic vaccination elicited Env Ab binding responses that were detectable using antigens corresponding to multiple HIV-1 subtypes but were not preferentially boosting CRF01_AE-specific responses. The focus on mosaic or subtype B and C binding responses is logical, given that the two mosaic inserts corresponded to central HIV-1 subtype B and C sequences, respectively (Figure S1). Thus, there is a large distance between mosaic inserts and viruses found in participants, as most participants in this study had acquired CRF01_AE viruses, with Env from 24 of the 26 participants corresponding to CRF01_AE. Hence, CRF01_AE-specific responses were induced following infection and still present pre-vaccination but amplified minimally by vaccination when compared to responses to antigens corresponding to other subtypes. Interestingly, we showed that Ab responses matured over time on ART, with different dynamics depending on Fiebig stage at ART initiation. Participants who initiated ART in Fiebig stage I/II/III showed an increase in Ab binding levels over time, while participants who initiated ART in Fiebig IV saw a decline in binding Abs. These results align with a prior study in this cohort that showed that plasma Ab levels and ADCC increased during the first year following ART initiation in acute HIV-1 infection.²²The degree of Ab maturation prior to vaccination was associated with the quantity and duration of antigen exposure prior to viral suppression. Beyond the first year of ART, Ab levels began to decline, and the extent of this decline was associated with the Ab levels at week 60 on ART.

Beyond Ab binding responses, we also characterized four Fc effector functions (ADCD, ADCP, ADNKA, and ADNP). Classification models revealed a vaccine profile defined by non-CRF01_AE binding Abs and Fc effector functions, suggesting that an important fraction of vaccine-induced responses had diverted from infection-specific CRF01_AE responses. All tested functions toward the vaccine mosaic insert were increased in vaccine recipients, while only ADCP toward CRF01_AE was increased.

Our second key finding was that ADCP responses were associated with delayed viral rebound upon ATI. A robust model discriminated participants by viral rebound group, with the delayed rebound associated with CRF01_AE functions. Specifically, higher CRF01_AE-specific ADCP responses were the most robust predictive markers and were significantly associated with delayed time to viral rebound. This association was particularly pronounced when assessing ADCP responses at peak vaccine immunity and at ATI but less visible by peak VL rebound, indicating that it was the vaccine-elicited responses that affected time to rebound (there was no significant difference prior to vaccination). All eight placebo participants (the B*57:01:01 placebo participant was excluded) were categorized into the early viral rebound group together with seven vaccine recipients who had very low CRF01_AE-specific ADCP responses at the peak immunity time point. Post rebound, viremia rapidly amplified the CRF01_AE-specific ADCP responses in placebo participants, leading to reduced differences between vaccine and placebo participants and, thereby, reduced albeit still significant differences between early and delayed viral rebound groups. These differences, however, did not achieve statistical significance in the Kaplan-Meier analysis, likely due to low power and the limited level of the CRF01_AE-specific ADCP responses induced by the Ad26/MVA mosaic vaccine. However, the PLSDA model demonstrated a discernible separation between participants with early versus delayed viral rebound at peak immunity, ATI, and peak VL post ATI, suggesting that the interplay among Fc-mediated functions could potentially increase the delay in viral rebound. These analyses were done using the median time to viral rebound in the cohort as a cutoff (28 days). In sensitivity analyses that extended this cutoff to 21 and 35 days, CRF01_AE-specific ADCP responses only showed trends of higher responses in the delayed rebound group. Sample sizes were smaller, making comparisons unstable (e.g., n = 3 for the delayed rebound group using a threshold of 35 days). However, the consistency of the PLSDA model showing clustering of participants by rebound group and selection of CRF01_AE-specific ADCP as the most predictive feature of delayed viral rebound across sensitivity thresholds supports our overall conclusions.

The finding that ADCP responses potentially delayed time to rebound post ATI following Ad26/MVA mosaic vaccination is interesting, knowing that ADCP responses have been associated with protection in mosaic vaccination studies in the nonhuman primate model.^23-26 We also found that ADCP responses played a role in vaccine protection in the RV144 vaccine efficacy trial.^27,28 The fact that ADCP responses have been associated with favorable outcomes in the context of both therapeutic and preventive vaccination attests to the biological importance of ADCP. Overall, prior studies together with our results highlight the potentially critical role of ADCP for future vaccine and remission strategies.

An important aspect of our study is the dichotomy it revealed between infection-induced and mosaic vaccine-induced responses. One limitation of our study was that we did not have a Mosaic 2 reagent and that all our binding and Fc effector assays used the mosaic 1 (subtype B-like) reagent. However, by using a panel of different antigens, including the consensus for subtype C, that are more closely related to the mosaic 2 vaccine insert, our Ab profiling is estimated to cover all potential responses elicited by Ad26/MVA mosaic vaccination. Vaccination promoted mosaic-specific or non-CRF01_AE binding Abs, while, upon viral rebound, binding Ab responses were similar for mosaic or CRF01_AE-specific responses. Binding Ab responses induced upon viral rebound did not differ across vaccine and placebo groups, suggesting that non-subtype-matched responses had limited impact post ATI. Our results suggest that Ad26/MVA mosaic vaccination elicited Ab responses to multiple subtypes but did not optimally boost the infection-specific CRF01_AE responses. The limited cross-reactivity emphasizes the need for vaccines that elicit responses corresponding to circulating viruses in a population.

In summary, our comprehensive systems serology analysis in an ATI study following Ad26/MVA mosaic vaccination showed diverse Ab-mediated responses in participants. We found binding Abs to non-CRF01_AE antigens present in the mosaic vaccine and a time-dependent shifting Ab response between infection-specific CRF01_AE responses and vaccine-specific non-CRF01_AE responses. When comparing participants who showed viral rebound before or after 28 days post ATI, we found that CRF01_AE-specific ADCP responses were higher among participants with delayed viral rebound.

Limitations of the study

One limitation of the study is the small sample size. While the comparison of responses between the vaccine (n = 17) and placebo (n = 9) groups may have acceptable power, the power was constrained for the correlates of rebound analysis where treatment confounding had to be controlled by utilizing only the vaccinees (early, n = 7; delayed, n = 10). To address this limitation and ensure the robustness of our predictive models, we conducted 200 randomly sampled evaluations and compared their performance to null models generated by permutating the groups. Additionally, we recognize that the small sample size may have masked some important immune correlates. Future studies are planned to analyze participants pooled from four ATI studies, each with a limited number of participants. Combining these studies will increase the sample size to a total of 53 participants (as described in Mdluli et al.²¹), thereby enhancing power and facilitating correlation of results.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Morgane Rolland (mrolland@hivresearch.org).

Materials availability

This study did not generate new unique reagents.

Data code and availability

Data reported in this paper will be shared by the lead contact upon request.

Code is available at https://www.hivresearch.org/publication-supplements.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Plasma samples were collected from 26 participants who were enrolled in the RV254 prospective acute HIV-1 infection cohort and later recruited in the RV405 cohort in Bangkok, Thailand. The RV254 Acute HIV-1 Infection study (clinicaltrials.gov NCT00796146) is a prospective cohort study designed to follow adults with Fiebig I to V acute HIV-1 infection while RV405 was a randomized controlled trial (double-blinded) where 26 participants received a therapeutic HIV-1 Ad26.Mos.HIV/MVA-Mosaic vaccine regimen (n = 17) or placebo (n = 9) (NCT00796263). Vaccine related analyses were performed with all 26 participants while time to rebound analyses were performed with 25 participants, removing the controller with B*57:01:01. The threshold (28 days) used to categorize participants into viral rebound groups after ATI corresponds to the median time to rebound among the pooled participants from the four ATI studies.²¹ Samples were analyzed at six time points: pre-ART and 60 weeks on ART during the RV254 cohort and pre-vaccination, peak vaccine immunity after receiving Ad26/MVA Mosaic (17 vaccine, 9 placebo), at ATI and at the time of peak viral load post-ATI. Participants had varying time periods between week 60 on ART and being recruited to RV405 with a median of 40.5 days (range 1.2–215); this interval was not significantly different between vaccine and placebo recipients. After ATI, participants reacvarying times with a median of 32.5 days (range 20–58) and median peak viral load of 4.71 log10 (IQR 4.43, 4.92) copies/mL, with no significant differences between vaccine and placebo recipients.

Participants information on sex at birth, age and race was self-reported. Information on gender and socioeconomic status was not reported for this study.

Ethics statement

All participants signed written informed consent and participated in protocols approved by Thai and US (Walter Reed Army Institute of Research) Institute Review Boards. The investigators have adhered to the policies for protection of human subjects as prescribed in AR 70–25.

METHOD DETAILS

Luminex multiplex binding antibody assay

Antibody binding and characterization was performed as previously described,^28,29 modified to assess antigens relevant to the current study. Briefly, heat-inactived human plasma samples were assayed in duplicate at 1:100 and 1:1000 dilutions. Pooled plasma from people living with HIV and without HIV served as positive and negative controls respectively. Samples were incubated with a panel of antigens including 3 gp120 (A244Δ11^C, MNΔ11 ^C, CM235^A) and 12 gp140 (Mos1.Env; Consensus CRF01-AE^B, A^B, B^B, C^B, D ^C, G^B, M^B, CRF02-AG^C; and three CRF01_AE representing naturally isolated sequences 40094v01^B, 40123v03.a^C, 40436v02^C) envelop proteins. Antigens were obtained from the ^ANIH AIDS Reagent Program (Bethesda MD), ^BDuke University Protein Production Facilty (Durham NC) or ^Cproduced in our lab. Each antigen was covalently coupled to uniquely coded cabrboxylated microspheres (Luminex Corp, Austin TX). Following a 2 h incubation with samples, microspheres were were washed then incubated with one of the R-phycoerthrin (PE) conjugated detection reagents. Ig subclass and specificity were assessed using mouse anti-human IgG, IgG1-4 and IgA (Southern Biotech, Birmingham AL). Fc receptor specificity was determined by PE-conjugated recombinant Fcγ receptors 2A(H131 and R131), 2B, 3A(F158 and V158), 3B (NA1, NA2 and SH) and FcγRn. Following a 1 h incubation, excess detection regent was removed by washing and beads were resuspended in 40μL sheath fluid (Luminex Corp). Data was collected as mean fluorescence intensity (MFI) on a Bio-Plex 3D Suspension Array system (Bio-Rad, Hercules CA) running xPONENT v.4.2 (Luminex Corp).

CRF01_AE responses were analyzed as geometric mean fluorescence intensities (MFI) of four antigens, CRF01_AE consensus and three natural CFR01_AE Env (40094v01, 40123v03.a, 40436v02).

Ab-functional profiling

Antibody-dependent neutrophil phagocytosis (ADNP)

Biotinylated gp120 CM235 (Immune Technology, NY, USA) or gp140 Mos1 were incubated with yellow-green streptavidin-fluorescent beads (Molecular Probes, Eugene, OR) for 2 h at 37°C. 10μL of a 100-fold dilution of beads–protein was incubated for 2h at 37°C with 100μL of 100-fold diluted plasma samples before addition of effector cells (50,000 cells/well). Fresh human peripheral blood mononuclear cells were used as effector cells after red blood cell lysis with ACK lysing buffer (ThermoFisher Scientific, Waltham, MA). After 1h incubation at 37°C, the cells were washed, surface stained, fixed with 4% formaldehyde solution (Tousimis, Rockville, MD) and fluorescence was evaluated on an LSRII flow cytometer (BD Bioscience, San Jose, CA). Antibodies used for flow cytometry included anti-human CD3 AF700 (clone UCHT1) and anti-human CD14 APC-Cy7 (clone MφP9) (BD Bioscience, San Jose, CA) as well as anti-human CD66b Pacific Blue (clone G10F5) (Biolegend, San Diego, CA). A phagocytic score was calculated by multiplying the percentage of bead-positive neutrophils (SSC high, CD3⁻ CD14⁻ CD66⁺) by the geometric mean of the fluorescence intensity of bead-positive cells; and dividing by 10,000.

Antibody-dependent cellular phagocytosis (ADCP)

ADCP was measured as previously described.³⁰ Briefly, biotinylated gp120 CM235 or gp140 Mos1 were incubated with yellow-green streptavidin-fluorescent beads (Molecular Probes, Eugene, OR) for 2 h at 37°C. 10μL of a 100-fold dilution of beads–protein was incubated for 2 h at 37°C with 100μL of 100-fold diluted plasma samples before addition of THP-1 cells (20,000 cells per well; Millipore Sigma, Burlington, MA). After 19 h incubation at 37°C, the cells were fixed with 2% formaldehyde solution (Tousimis, Rockville MD) and fluorescence was evaluated on an LSRII flow cytometer (BD Bioscience, San Jose, CA). The phagocytic score was calculated by multiplying the percentage of bead-positive cells by the geometric mean of the fluorescence intensity of bead-positive cells and dividing by 10,000.

Antibody-dependent complement deposition (ADCD)

ADCD was adapted from.³¹ Biotinylated gp120 CM235 or gp140 Mos1 -coated red Neutravidin beads (ThermoFisher) were incubated with 10-fold diluted heat-inactivated plasma samples. Guinea pig complement (Cedarlane Labs, Burlington, NC, USA) diluted in veronal buffer containing calcium and magnesium (Boston Bioproducts, Ashland, MA, USA) was incubated with antibody-bead complexes for 20 min, and C3 deposition onto beads was detected using an anti–guinea pig C3 FITC (polyclonal, ThermoFisher Scientific, Waltham, MA) and fluorescence was measured on an LSRII flow cytometer.

Antibody-dependent NK cell activation (ADNKA)

Enzyme-linked immunosorbent assay (ELISA) plates were coated with gp120 CM235 or gp140 Mos1 (150 ng/well). Plates were then extensively washed with PBS and 50-fold diluted plasma samples were incubated at room temperature (RT) for 2 h. After washing, cryopreserved peripheral blood mononuclear cells (PBMCs) were added to each well and incubated for 6 h at 37°C in presence of anti-CD107a APC (clone H4A3, BD Bioscience), monensin (BD Bioscience), and brefeldin (eBioscience). After stimulation, cells were washed and surface stained at RT for 10 min in the dark. Samples were then washed and fixed using Fix & Perm Medium A (Life Technologies). Intracellular cytokine staining (ICS) was performed in Fix & Perm Medium B (Life Technologies). Commercial monoclonal antibodies used in flow cytometry included: anti-CD3 AF700 (clone UCHT1), anti-CD56 PE-Cy7 (clone B159), anti-CD16 APC-Cy7 (clone 3G8), and anti-IFN-γ V450 (clone B27; all from BD Biosciences) and anti-CD14 V510 (clone M5E2), anti-CD19 V510 (clone HIB19), anti-TNF FITC (clone MAb11; all from BioLegend). Data were acquired on a BD LSRII instrument and analyzed using FlowJo Version 9.8.5 software.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis

Statistical analysis and visualization were performed using R 4.0.0 statistical computing environment (with R packages ggplot2, final-fit, rstatix, correlation and ‘survminer’). Statistical significance was computed using the Mann-Whitney U and Wilcoxon signed-rank tests for unpaired and paired comparisons, respectively. Spearman correlations were calculated with two-tailed p-values. The False Discovery Rate (FDR) was applied to correct for multiple testing to control type-I error (significance levels, *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001). Nominal p-values were used for Fc effector functions comparisons while adjusted p-values were used in binding Abs comparisons. Kaplan-Meier survival curves were plotted and compared by the log-rank test using the ggsurvplot function in the ‘survminer’ R package. Maximally selected rank statistics were used to robustly estimate the optimal cutoff that allows us to divide the continuous variables into groups and visualize differences in the Kaplan-Meier estimates of survival curves. This test selects the cutoff that provides the greatest discrimination power between groups to improve visualization of the curves. The threshold was selected using the surv_cutpoint function and curves were visualized through the R package ‘survminer’. Analyses were performed using the Median Fluorescence Intensity (MFI) values or Fc effector function scores or using the fold change over pre-vaccination levels. Assay negatives were similar and below assay background (600 MFI) across all detections and antigens, therefore the raw MFI was not normalized instead the assay background is illustrated in the figures. However, z-scores were analyzed where necessary by setting signal below assay background to null to avoid scaling background noise as signal. Fold over pre-vaccination were used to capture the actual increase in responses due to vaccination for analyses comparing vaccine and placebo participants at peak vaccine immunity. The threshold for positivity using fold change was set to 3.

Random Forest (RF) modeling

Random Forest models were evaluated to reduce Ab binding features before building a Partial Least Squares Discriminant Analysis (PLSDA) model by integrating selected binding features and functional features to generate a comprehensive profile. Since the groups were unbalanced, the number model was built by down selecting samples from the group with the highest number of samples and the process was repeated 500 times to account for random splits in down selecting samples. The Random Forest algorithm (‘RandomForest’ function in R) was evaluated using default settings and setting the number of trees to 1000. A Random Forest importance score metric was recorded for each feature to rank features according to their importance in discriminating groups. Selected features were defined to have an importance score above 70%.

Partial Least Squares Discriminant Analysis (PLSDA)

A comprehensive PLSDA model was generated with the RF-selected binding features and Fc effector functions to assess and visualize the ability of Ab binding responses to discriminate between groups. The function ‘plsda’ in the mixOmics R package was used to generate each model at each of the time points by setting the number of components to 10 with all other parameters kept at default. Data were standardized before generating the model by centering around zero and scaling to the standard deviation of one for each feature (z-scores). The performance of each of the PLSDA models was assessed using permutation tests. We permuted the treatment groups to create a null model and recalculate the model using the permutated data as described above. The data was permuted 200 times, each time using 75% of the data for training and 25%was used to evaluate the goodness of fit of the model, ensuring that both the training and test set had one representation from each group. A distribution of balanced of accuracy metrics obtained with the non-permutated/true groups was compared with that obtained with the permutated groups (null hypothesis) classification. The assumption was that the classification model created with the permutated data should fail to predict group classes.

Network interactions

A correlation network was used to interrogate correlations among the Random Forest selected binding features and Fc effector functions. We included all Ab binding features that were strongly correlated (Spearman rho >0.9, p < 0.05) with those selected using RF analysis. The network was constructed based on Spearman corrections between antibody binding responses and functional responses. Edges between features were weighted according to correlation coefficient. Edges are visualized for significant correlation after adjusting for multiple comparisons (FDR <0.05). The adjacency network was built and visualized using ‘ggraph’ function in R.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
gp140 RV405 Mosaic1	Janssen	N/A
CRF01_AE gp140 40094v01	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype A1 gp140CF ConA1.env03	Duke Human Vaccine Institute (Durham NC)	N/A
CRF01_AE gp140CF ConAE01.env03	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype B gp140CF ConB.env03	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype C gp140CF ConC.env03	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype G gp140CF ConG.env03	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype M gp140CFI ConM.S	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype B gp70 V1/V2 CaseA2	Duke Human Vaccine Institute (Durham NC)	N/A
Subtype B gp41 HxBc2	Immune Tech (New York NY)	Cat# IT-001–005p
CRF01_AE gp120 A244Δ11	MHRP B Cell Immunology Section	N/A
Subtype B gp120 MNΔ11	MHRP B Cell Immunology Section	N/A
CRF01_AE gp140 40123v03.a	MHRP B Cell Immunology Section	N/A
CRF01_AE gp140 40436v02	MHRP B Cell Immunology Section	N/A
Subtype D gp140 ConD	MHRP B Cell Immunology Section	N/A
CRF02_AG gp140 ConCRF02	MHRP B Cell Immunology Section	N/A
CRF01_AE gp70 V1/V2 A244	MHRP B Cell Immunology Section	N/A
CRF01_AE gp70 V3 A244	MHRP B Cell Immunology Section	N/A
Subtype B gp70 V3 MN	MHRP B Cell Immunology Section	N/A
gp70 MuLV	MHRP B Cell Immunology Section	N/A
CRF01_AE gp120 CM235	NIH-ARP
Deposited data
Source code	This paper	https://www.hivresearch.org/node/961
Software and algorithms
R version 4.0.0	The foundation for statistical computing	https://www.r-project.org/

Highlights.

• The mosaic vaccine triggers vaccine-matched antibody binding responses in PLWH

• Infection-specific CRF01_AE responses are not further stimulated by vaccination

• Binding antibodies are not associated with time to rebound upon ATI

• Vaccine-induced ADCP responses associate with delayed viral rebound