Animals
Animal studies were conducted with the approval of the appropriate Institutional Animal Care and Use Committee (IACUC) at Beth Israel Deaconess Medical Center, AlphaGenesis (19-04, 18-14), and Bioqual (17-014.2, 17-015.2). All animals used in these studies were outbred, young adult (3–10 years old) Indian-origin rhesus macaques (Macaca mulatta) of both sexes housed at either AlphaGenesis (Yemasee, SC) or Bioqual (Rockville, MD). Animals were randomly allocated for each experiment. For the single challenge experiments, animals were intravenously infused with PGDM1400 (10 mg/kg) one day prior to challenge. The animals were then intrarectally challenged with 500 MID50 of PGDM1400-sensitive SHIV-325C (n = 4) or 500 MID50 SHIV-325C combined with a sub-infectious dose of 0.5 MID50 PGDM1400-resistant SHIV-SF162P3 (n = 4). Control animals were intrarectally challenged with either 500 MID50 SHIV-325C (n = 3), 500 MID50 SHIV-SF162P3 (n = 4), or 0.5 SHIV-SF162P3 (n = 4) to show infectivity of the virus stocks at those doses.
In the repeat challenge experiments, macaques were infused with PGDM1400 one day prior to each challenge. These animals (n = 5) were challenged five times with 500 MID50 SHIV-325C combined with a lower sub-infectious dose of 0.05 MID 50 SHIV-SF162P3. Control animals were challenged intrarectally every two weeks until the animals became viremic with either 500 MID50 SHIV-325C (n = 4), 5 MID50 SHIV-SF162P3 (n = 4), or the sub-infectious dose of 0.05 MID50 SHIV-325C (n = 4). In both studies, immunological and virological assays were performed blinded.
PGDM1400 pharmacokinetics
Serum levels of human PGDM1400 mAb were monitored using a previously described human IgG specific enzyme-linked immunosorbent assay (ELISA)15. In brief, ELISA plates were coated overnight at 4 °C with 1 μg/mL of goat anti-human IgG (H + L) secondary antibody (monkey pre-adsorbed) (Novus Biologicals) and then blocked for 2 h. Serum samples were assayed at 3-fold dilutions starting at a 1:3 dilution in Blocker Casein in PBS (Thermo Fisher Scientific) diluent. Samples were incubated for 1 h at ambient temperature and then removed, and plates were washed. Wells then were incubated for 1 h with horseradish peroxidase (HRP)-conjugated goat anti-human IgG (monkey pre-adsorbed) (Southern Biotech) at a 1:4,000 dilution. Wells were washed and then incubated with SureBlue Reserve TMB Microwell Peroxidase Substrate (100 μL/well; Seracare) for 3 min followed by TMB Stop Solution (100 μL/well; Seracare) to stop the reaction. Microplates were read at 450 nm. The concentration of PGDM1400 mAb was interpolated from the linear range of concurrently run purified human IgG (Sigma) standard curves using Prism software, version 10.3.1 (GraphPad).
Viral RNA analyses
Viral RNA was extracted from plasma with the QIAmp Viral RNA Kit (Qiagen) and then converted to cDNA with Superscript III VILO (Invitrogen). The cDNA was quantified via qPCR with QuantStudio 1.7.1 as previously described16 using the following primers (Integrated DNA Technologies) and probe (Applied Biosystems) with TaqMan Fast Advanced Master Mix (Applied Biosystems): Fwd, 5′- GTCTGCGTCATCTGGTGCATTC -3′; Rev, 5′- CACTAGGTGTCTCTGCACTATCTGTTTTG -3′; and Probe 5′-(FAM) CTTCCTCAGTGTGTTTCACTTTCTCTTCTGCG-(BHQ)-3′. Human viral loads were run either the Hologic Aptima HIV-1 Quant or Abbott Real-Time HIV-1 assays as described previously7.
Serum cytokine analysis
Serum cytokines and chemokines were measured via multiplex ELISA U-PLEX assays on Multiplex ELISA (Meso Scale Discovery) according to manufacturer’s instructions by the Metabolism and Mitochondrial Research Core (Beth Israel Deaconess Medical Center, Boston, MA).
Flow cytometry
PBMCs were isolated from blood at day 42, frozen, and stored in a liquid nitrogen freezer. Cells were later defrosted, washed, and then stained with monoclonal antibodies at concentrations suggested by the manufacturer (Becton Dickinson unless noted) against CD3 (SP34; Alexa Fluor 700), CD4 (OKT4; BV510, Biolegend), CD8 (SK1; APC-Cy7), CD14 (M5E2; BUV737), CD16 (3G8; BV650), CD25 (PE-Cy7; M-A251), CD28 (L293; PerCP-Cy5.5), CD38 (APC; HB-7), CD56 (NCAM16; BV786), CD69 (TP1.55.3; PE-TexasRed; Beckman Coulter), CD95 (DX2; BV711), CCR5 (3A9; PE), CCR7 (3D12; BV421), HLA-DR (BUV-395; G46-6), Ki67 (B56; FITC), and PD-1 (EH21.1; BV605). Samples were run on an LSR II and analyzed in FlowJo (see Supplementary Fig. 6 for gating strategy).
Single genome amplification (SGA) and pseudovirus construction
SGA was performed on macaque samples as previously described17. Briefly, RNA isolated from plasma samples at one to two timepoints within detection of positive viral loads with the QIAmp Viral RNA Mini Kit (QIAgen) was reverse transcribed with Superscript IV reverse transcriptase (Invitrogen) and the primer 5’- TGTAATAAATCCCTTCCAGTCCCCCC-3’. All primers for this assay were sourced from Integrated DNA Technologies (IDT). Limiting dilution PCR with the outer primers (Fwd: 5’- CCTCCCCCTCCAGGACTAGC-3’; Rev: 5’- TGTAATAAATCCCTTCCAGTCCCCCC-3’) was then performed with Platinum SuperFi II PCR Master Mix (Invitrogen) and the following PCR conditions: 1’ at 98 °C; 35 cycles of 10” of 98 °C, 10” of 60 °C, and 2’15” of 72 °C; 5’ at 72 °C; hold at 4 °C. The inner PCR reaction, using 1.5 uL of the outer PCR product, was then performed with Platinum SuperFi II PCR Master Mix (Invitrogen), a second set of primers (Fwd: 5’- ATAGACATGGAGACACCCTTGAGGGAGC-3’; Rev: 5’- ATGAGACATRTCTATTGCCAATTTGTA-3’), and slightly altered PCR conditions (45 cycles and a 55 °C annealing temperature). Amplicons were considered to be the result of a single cDNA molecule according to the Poisson distribution were processed for sequencing if one third or fewer wells were positive (p < 0.05). Unique Env sequences with complete open reading frames were reamplified with Platinum PCR SuperMix High Fidelity (Invitrogen) to generate 3’-A overhangs with an altered set of inner PCR primers (Fwd: 5’- CACCTTAGGCATCTCCTATGGCAGGAAGAAG-3’; Rev: same as previous) and the following conditions: 1’ at 94 °C; 45 cycles of 30” at 94 °C, 30” at 55°, and 3’30” at 68 °C; 5’ at 68 °C; hold at 4 °C. HIV-specific primers and PCR conditions were used as previously described for the human samples7. The resultant sequences were inserted into CMV-based plasmid expression vectors with the pcDNA 3.1/V5-His TOPO TA Expression Kit (Invitrogen) and used for pseudovirus generation as described previously18. Briefly, 293T cells were transfected with 3.3 ug of the Env expression plasmid and 10 ug of env-deficient HIV-1 backbone plasmid pSG3ΔEnv using Fugene 6 (Promega). Pseudovirus was collected after two days and underwent 0.45 μm filtration prior to storage at −80 °C.
Sequence analysis
Whole envelope sequencing was performed for the viruses from the single-challenge experiment by the MGH CCIB DNA Core. Partial envelope sequencing was performed for relevant regions of virus from the macaques infected with SHIV-325C in the absence of PGDM1400 and the macaques from the repeat challenge experiments. Alignments and neighbor-joining consensus trees were generated with Geneious Prime (Version 2024.0.2). Highlighter plots were generated using the Los Alamos National Laboratories (LANL) Highlighter webtool (https://www.hiv.lanl.gov/content/sequence/HIGHLIGHT/highlighter_top.html). Potential N-linked glycosylation sites (PNGS) were detected using the LANL N-GlycoSite webtool (https://www.hiv.lanl.gov/content/sequence/GLYCOSITE/glycosite.html) via any amino acid sequence matches to the N-linked glycosylation recognition sequence N-X-[ST], where X may be any amino acid except proline.
Neutralization assay
Pseudoviruses and the SHIV-SF162P3 and SHIV-325C stocks were analyzed for sensitivity to PGDM1400 neutralization via the TZM-bl luciferase reporter neutralization as previously described19. Briefly, reduction in Tat-regulated luciferase (Luc) reporter gene expression in TZM-bl cells was used to measure the neutralization of Env pseudoviruses by PGDM1400 using a fivefold dilution series starting with a concentration of 50 μg/mL in duplicate wells. IC50, IC80, IC90, and IC99 were then calculated.
Deep sequencing of SHIV-325C Env
RNA isolated from our SHIV-325C stock was reverse transcribed with Superscript IV reverse transcriptase (Invitrogen), and the resultant cDNA was amplified with Platnium SuperFi II PCR Master Mix (Invitrogen) and our Env-specific outer primers as described above. Specific regions of Env were then amplified with Platnium SuperFi II PCR Master Mix and sent out for library preparation and Illumina MiSeq 150 bp paired-end sequencing at the MGH CCIB DNA Core. Primer sets (Integrated DNA Technologies) are described in Supplementary Table 2. Pre-processing of deep sequencing data to remove adaptor sequences and low-quality bases was performed by the MGH CCIB DNA Core. Using Geneious Prime (Version 2024.0.2), the processed Illumina sequences were aligned to the SHIV-325C Env consensus sequence, and single-nucleotide polymorphisms were detected using the program’s built-in tool to detect single-nucleotide polymorphisms and variants and calculate approximate p values for those variants.
Statistical analyses
Virological and immunological data analysis (excluding modeling) was performed using GraphPad Prism Version 10 (GraphPad Software). Two-sided Mann–Whitney tests were used to compare groups.
Partial least squares discriminant analysis (PLS-DA)
To investigate the association between predictor variables (cytokines) and the two challenge groups, we performed Partial Least Squares Discriminant Analysis (PLS-DA) using R (version R.4.4.2). PLS-DA is a supervised dimensionality reduction technique that models the relationship between the feature matrix (cytokines) and a categorical response variable (groups), aiming to find latent components that maximize the separation between groups.
The analysis was conducted using the mixOmics 6.28.0 package. To visualize the results, we generated a biplot using the plotIndiv() and plotVar() functions. The biplot displays both the samples and the features in the space defined by the first two latent components. Each sample is represented as a point, colored by its group membership, while each feature is represented as an arrow. The direction of each arrow indicates the orientation of the variable in the component space, and the length of the arrow reflects the strength of the association between the feature and the components. Features with longer arrows are more strongly associated with the variation that separates the groups. Arrows pointing in the same direction as a group cluster suggest a positive association between that feature and the group. All visualizations were refined using base R plotting or ggplot2 to ensure clear interpretation and high-quality presentation.
Machine learning analysis of serum cytokines
To identify markers that significantly contribute to the classification of samples into “Mixed” and “NotMixed” groups at the D3 and D7 timepoints, a machine learning-based Random Forest approach was employed to determine the relative importance of each protein marker and to evaluate their association with the respective groups. The dataset underwent a series of preprocessing steps to ensure its suitability for machine learning analysis. First, samples corresponding to the D3 or D7 timepoint were filtered for analysis. Non-numeric columns, such as sample identifiers and time labels, were removed as they were not relevant to the classification task. Rows containing missing values were excluded to maintain the completeness and integrity of the data used for model training.
To evaluate the discriminative power of protein markers, a Random Forest classifier was used to classify samples into the “Mixed” and “NotMixed” groups. The Category column, which indicated group membership, served as the dependent variable, while the protein marker measurements acted as independent variables. The model was configured to generate 500 decision trees to ensure robust predictions and reliable feature importance metrics. Feature importance was assessed using the Mean Decrease in Gini impurity, which measures the contribution of each protein marker to reducing classification uncertainty.
To further interpret the importance of individual protein markers, their levels were compared between the “Mixed” and “NotMixed” groups. The mean expression difference for each protein marker was calculated, and features were categorized based on whether their higher expression was associated with the “Mixed” or “NotMixed” group. This allowed for the identification of proteins with stronger ties to either group.
The feature importance scores were visualized using a dot plot. In the plot, protein markers were arranged along the y-axis in descending order of their importance scores. The x-axis displayed the feature importance values derived from the Random Forest model. A red gradient color scale was used to highlight the magnitude of importance, with darker shades representing higher importance. The dot size was proportional to the feature importance score, further emphasizing the most critical markers.
To assess the statistical significance of the association of a group of cytokines instead of individual cytokines with the mixed challenge, we applied the Rotation Gene Set Testing (ROAST) method from the limma R package. Specifically, we are interested in determining whether a group of cytokines, exhibit significant changes in expression associated with the Mixed challenge compared to the Not Mixed challenge groups. ROAST works by applying a rotation-based resampling technique to simulate the null distribution, rather than relying on traditional permutation tests, which makes it particularly powerful for smaller sample sizes. It adjusts for correlations between genes, providing a more accurate assessment of whether the cytokines in the set are collectively upregulated, downregulated, or exhibit mixed regulation.
K-means clustering analysis
To assess how well the cytokine profiles naturally group the samples, independent of group labels, we used K-means clustering on the log-transformed cytokine data after filtering out cytokines with zero variance and removing highly collinear features. The clustering results were compared to the actual group labels to evaluate accuracy. The confusion matrix showed the distribution of samples across predicted clusters versus true groups, and the overall clustering accuracy was calculated as the proportion of correctly assigned samples. The results were visualized with a cluster plot, showing separation of samples colored by their assigned cluster. The following R packages were used for data analysis and visualization: “ggplot2”, “randomForest”, “mixOmics”, “FactoMineR”, “factoextra”, and “cluster”.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.