Peripheral blood mononuclear cells (PBMCs) from each human and rhesus macaque (RM) were analyzed for their effector functionality ( Figure 1 ). A flow-based GranToxiLux (GTL) assay (44) was used to determine the ADCC magnitude of human or RM PBMCs. Target cells were coated with the recombinant SHIV1157 QNE Y173H gp120 protein for the GTL assay. To test ADCC potency, we used a combination of four HIV-specific human or RM IgG1 mAbs each at 1µg/ml (C1C2: JR4 (45, 46) and DH677.3 (45, 46); V1V2: DH827 (45, 47) and DH614.2 (45), which previously demonstrated binding to non-overlapping epitopes on gp120, and mediated ADCC against SHIV1157 QNE Y173H gp120-coated cells and SHIV1157 QNE Y173H-infected cells individually and in corresponding human and rhesus combinations (45, 48). We observed similar total ADCC activity across species (P > 0.05 by two-sided Mann-Whitney U test, Figure 1A ). To assess the contribution of NK cells and monocytes to the total observed ADCC activity (% NK ADCC), we performed area scaling analysis (ASA, Figure S1A ) (49). The contribution of NK cells to the total ADCC was significantly higher in humans compared to RMs (P < 0.001 by two-sided Mann-Whitney U test, Figure 1B ) whereas NK cells contribution to the total ADCC was similar or lower than the monocyte contribution in RMs. The frequency of NK cells in PBMC samples was also significantly higher in humans compared to RMs (P < 0.001 by two-sided Mann-Whitney U test, Figures 1C , S1B, C ). However, the frequency of CD16+ NK cells was similar across species ( Figure 1D ). We also analyzed the median fluorescent intensity (MFI) of cell-surface expressed FcγRIII(a) (CD16) within PBMCs, which is essential for ADCC by NK cells and monocytes (50, 51). Intriguingly, the level of cell-surface CD16 expression was significantly lower in humans (P < 0.001 by two-sided Mann-Whitney U test, Figure 1E ). FcγRIII(a) expressed on NK cells can transduce signals through FcRγ and CD3ζ chains, where cells signaling through the alternative CD3ζ chain typically demonstrate enhanced ADCC (28, 52). Thus, we also measured the frequency of NK cells signaling through CD3ζ (phosphorylated CD3ζ; hereafter pCD3ζ) in PBMCs following FcγRIII(a) cross-linking ( Figure S1D ). The frequency of pCD3ζ+ NK cells was significantly higher in humans (P < 0.001 by two-sided Mann-Whitney U test, Figure 1F ). Together, these data demonstrate variation in frequency and function of NK cells between the two species.

Human PBMCs were genotyped for known polymorphisms of FcγRIIIa at amino acid position 158 (Phe or Val, F/V) in the membrane-proximal, IgG-binding domain, which demonstrated different affinity for antibody Fc with subsequent variation in ADCC potency (14). Among 40 human samples, we identified 17 samples with F/F genotype, 18 with F/V genotype and 5 with V/V genotype ( Table S1 ). For RM cells, we were able to sequence FCGR3 for 34 of the 40 animals ( Table S2 ). Among those 34 RMs, 11 animals had no mutations relative to the reference sequence (Mnul_10, NCBI Genome ID: 215). One animal had only an I158V mutation, which is within the extracellular region of the receptor that forms the binding interface with IgG Fc (Tolbert et al., 2022, in press). 19 animals had both V211M and V215I mutations, which are located within the cytoplasmic tail. In addition, 3 animals had I158V, V211M, and V215I mutations ( Table S2 ).

We then compared NK cell effector functions based on FcγRIII(a) polymorphisms to determine whether genetic diversity was predictive of functional diversity ( Figure 2 ). The median ADCC magnitude was similar between genotypes in humans and RMs ( Figures 2A, B ). The contribution of NK cells and monocytes to the total observed ADCC activity was also similar among all genotypes in humans and RMs ( Figures 2C, D ). We observed similarity in cell-surface expression of FcγRIII(a) within PBMCs ( Figures 2E, F ) and the frequency of NK cells signaling through the CD3ζ chain ( Figures 2G, H ). Importantly, none of these measurements had a statistically significant association with FcγRIII(a) polymorphism(s) (all P > 0.05 by Kruskal-Wallis tests in humans and Mann-Whitney U tests in RMs, with comparisons of polymorphisms at each amino acid position being analyzed independently), suggesting that genetic diversity of FcγRIII(a) is insufficient to explain heterogeneity in NK cell function in either humans or macaques.

We observed heterogeneity among humans and RMs in each of the 4 functional measurements ( Figure 1 ), yet this variability cannot be explained by FcγRIII(a) polymorphisms ( Figure 2 ). Hierarchical clustering based on ADCC magnitude, % NK-mediated ADCC, cell-surface FcγRIII(a) expression (CD16 MFI), and % pCD3ζ+ identified 2 distinct groups within each species cohorts, which are separated on a principal component analysis (PCA) biplot for both humans ( Figure 3A ) and macaques ( Figure 3B ). In humans, F/F and V/V genotypes were overrepresented in cluster 1, while the F/V genotype was evenly balanced across clusters. However, the association between polymorphism and cluster was not statistically significant (P = 0.051 by Fisher’s Exact test, Table 1 ). In RMs, all polymorphisms were represented in each cluster and there were no significant associations between polymorphisms and clusters (P > 0.05 by Fisher’s Exact test, Table 1 ). In humans, the clusters separated well by principal component 1 (PC1), which explained 46.0% of the total variation within the cohort. All four parameters contributed similarly to PC1 ( Figure 3C ). The largest contributors to PC2, which explained an additional 26.7% of the total variation, were % NK-mediated ADCC and % pCD3ζ+ NK cells. In RMs, PC1 explained 46.0% of the total variation within the cohort and the magnitude of contributions to PC1 were similar for all four parameters, consistent with results in the human cohort, but the loading value for % NK ADCC was negative while loadings for the other 3 parameters were positive ( Figure 3D ). The largest contributor to PC2, which explained an additional 20.6% of the total variation, was the total ADCC potency.

The functional profiles of each cluster were further characterized by examining the distribution of the variables by cluster ( Figure 4 ). For human samples, we observed significant differences in ADCC potency (P = 0.002, Figure 4A ), % NK ADCC (P < 0.001, Figure 4B ), and cell-surface FcγRIIIa expression CD16 (MFI; P < 0.001, Figure 3C ) between clusters by Mann-Whitney U tests, with medians for all parameters being higher in cluster 2. Median % pCD3ζ+ was also higher in cluster 2, however there was no statistically significant difference between clusters (P > 0.05 by Mann-Whitney U test, Figure 4D ).

For RM samples, there were significant differences between clusters in all four parameters (by Mann-Whitney U tests) where ADCC potency (P = 0.016, Figure 4E ), CD16 MFI (P < 0.001, Figure 4G ), and % pCD3ζ (P = 0.013, Figure 4H ) were higher in cluster 2, while median % NK ADCC was higher in cluster 1. Altogether, these data indicate that human subjects and RMs can be separated into clusters with distinct functional profiles. Clustering of RMs is independent of FcγRIII genetic diversity, and while there is some evidence that clustering of humans could be influenced by FcγRIIIa polymorphisms, the association was not statistically significant.

Spearman’s rank correlation was used to examine relationships among these functional measurements within humans ( Figure S2 ) and RMs ( Figure S3 ). In humans, no significant correlations were observed to frequency of NK cells (P > 0.05, Figures S2A–D ), but positive correlations were observed between total ADCC and CD16 expression (r = 0.465, P = 0.002, Figure S2F ), ADCC and the frequency of pCD3ζ+ NK cells (r = 0.425, P = 0.011, Figure S2G ), and frequency of NK-cell mediated ADCC and CD16 expression (r = 0.414, P = 0.008, Figure S2H ). In RMs, the frequency of NK cells had a negative correlation to total ADCC potency (r = -0.344, P = 0.032, Figure S2A ) and a positive correlation to NK-cell mediated ADCC (r = 0.538, P < 0.001, Figure S3B ). NK-cell mediated ADCC had negative correlation to the total ADCC potency (r = -0.435, P = 0.005, Figure S3E ), CD16 expression (r = -0.383, P = 0.016, Figure S3H ), and the frequency of pCD3ζ+ NK cells (r = -0.411, P = 0.009, Figure S3I ). CD16 expression had positive correlation to the frequency of pCD3ζ+ NK cells (r = 0.387, P = 0.015, Figure S3J ). These weak to moderate correlations, although they reached statistical significance, highlight that no single parameter alone is sufficient to characterize these different aspects of NK cell function.

The human cohort presented here consisted of 24 males and 16 females with CMV status available for 38 individuals ( Table S1 ), while all 40 RM were CMV+ males ( Table S2 ). Based on descriptive statistics we observed little evidence of any NK cell functional differences based on gender or CMV status in this human cohort ( Figure S4 ).

Based on the heterogeneity observed in FcR-related NK cell functional measurements within these cohorts, we sought to investigate whether we could use screening data to assign treatment groups of RMs in pre-clinical trials to ensure balanced groups. We used covariate constrained randomization to assign animals into the potential treatment groups based on the four functional parameters ( Figure 5 ). We observed that potential treatment groups A and B showed similar % total ADCC ( Figure 5A ), contribution of NK to the total ADCC ( Figure 5B ), CD16 expression ( Figure 5C ), and frequency of pCD3ζ+ NK cells ( Figure 5D ) with no significant differences (P > 0.05 by Mann-Whitney U test). Together, these data demonstrate that assigning RM to treatment groups using covariate constrained randomization produced balanced groups with similar FcR-related NK cell functional profiles.

We further sought to investigate whether covariate constrained randomization of animals will introduce bias on non-NK functional parameters, such as antibody-dependent cellular phagocytosis (ADCP). ADCP has been demonstrated to contribute to protection against infection in preclinical (7, 53, 54) and clinical studies (55). We compared phagocytic activity of monocytes, isolated from RM PBMCs, as defined by a phagocytosis score–the mathematical product of the percentage of phagocytic-active cells and the number of phagocytosed particles per cell (56). Although phagocytosis occurs independently of FcγRIII, we analyzed whether FcγRIII polymorphisms or the NK functional clusters identified in Figure 2 were predictive of in vitro phagocytosis. There were no significant differences between FcγRIII polymorphism groups or NK functional clusters for RMs (P > 0.05 by Kruskal-Wallis test, Figures S5A, B ). Importantly, covariate constrained randomization of RMs showed similar ADCP score in groups A and B with similar ranges and no significant difference (P > 0.05 by Mann-Whitney U test, Figure S5C ). These data demonstrate that assigning RMs into treatment groups using covariate constrained randomization produced groups with similar NK cell functional profiles without introducing bias in non-NK functions, such as ADCP.

The studies were reviewed and approved by the Duke University Medical Center Institutional Review Board, and all participants provided written informed consent. All research was performed in accordance with the Duke University School of Medicine guidelines and regulations. Indian origin rhesus macaques (Macaca mulatta) were used in this study. The animals were housed at the New Iberia Research Center, animal experiments were approved by the New Iberia Primate Research Center IACUC (Protocol 2021-010-8823) and conducted in compliance with the principles described in the Guide for the Care and Use of Laboratory Animals (76). Animals with B*17+ and B*08+ alleles were excluded from this cohort due to the protective effect of these alleles (74). Human samples were collected at and experiments with human samples were approved by the Duke IRB (Pro00000873).

FcγRIIIa is encoded by the gene FCGR3A. Genotyping of the donors for SNP rs396991 was determined by TaqMan SNP Genotyping Pre-Validated Assay (Applied Biosystems, Foster City, CA) and an ABI7900 Sequence Detection System (ABI). The genotyping assay was performed according to manufacturer’s guidelines using 10 ng of DNA. Standard TaqMan thermocycling conditions were 10 min at 95°C, then 40 cycles of 15 s at 92°C and 1 min at 60°C. The endpoint fluorescence was read on an ABI 7900 Sequence Analyzer. The allele call was made by using the ABI proprietary software SDS version 2.4.

FCGR3 sequencing was performed using long-read RNA sequencing as previously described (PMID: 34093580). Briefly, RNA was isolated from RM PBMC samples using the AllPrep DNA/RNA isolation kit (Qiagen, Germantown, MD), reverse transcribed using the Qiagen QuantiTect Reverse Transcription Kit (Qiagen), and Fc receptor cDNA was amplified using gene-specific primers designed with PacBio barcodes (Pacific Biosciences, Menlo Park, CA). PCR products were purified and PacBio SMRTbell library preparation was performed in accordance with manufacturer’s recommendations (Pacific Biosciences). Sequencing was performed on a PacBio Sequel II instrument using 2.1 or 3.0 chemistry (Pacific Biosciences). Datasets were loaded into PacBio SMRT Link 7.0.1 software package for demultiplexing of subreads and generating circular consensus sequences (CCS) and data analysis was conducted using a pipeline similar to that previously described (77). FCGR3 sequences were aligned to Mmul_10 reference sequence (NCBI Genome ID: 215) using a long-read sequence alignment tool (78) and variants were identified using the GATK Haplotype Caller (79), and annotated with ANNOVAR (80).

Human IgG1 and RM IgG1 mAbs were produced by transient transfection of heavy and light chain plasmids into Expi293-F cells with Expifectamine (Thermo Fisher Scientific, Waltham, MA) and purified from cell culture supernatants by protein A resin columns as previously described (81–83).

The ADCC-GranToxiLux assay was used to measure the ability of immune effector cells present in rhesus macaque (RM) PBMC to mediate Env gp120-specific ADCC when directed by HIV-specific RM IgG1 monoclonal antibodies. The assay was performed as previously described (44, 49). Target cells were a clonal isolate of the CEM.NKR_CCR5 CD4⁺ T cell line (NIH AIDS Reagent Program, Division of AIDS, NIAID, NIH: from Dr. Alexandra Trkola (84)) coated with SHIV1157 QNE Y173H gp120 protein (7). Human or RM PBMC effector cells were used at an effector cell to target cell ratio of 30:1 or 60:1, respectively. The antibodies used were a combination of the C1C2 cluster A-region specific antibodies JR4 (85) and DH677.3 (46), and V1V2-specific antibodies DH614.2 (86) and DH827 (47). JR4 and DH614.2 were originally isolated from RM after SHIV-infection or HIV-vaccination, respectively. DH677.3 and DH827 were originally isolated from human volunteers in the RV306 clinical trial. All antibodies were recombinantly produced as human and rhesus IgG1 (82). The combination of four human or four rhesus mAb were tested at a concentration of 1 μg/ml each. The influenza specific human or rhesusized CH65-IgG1 (anti-HA) antibodies (87) were used at the equivalent concentration of 4 μg/mL as negative controls. Data were reported as the maximum proportion of cells positive for proteolytically active granzyme B (GzB) out of the total viable target cell population (maximum %GzB activity) after subtracting the background activity observed in wells containing effector and target cells in the absence of antibodies. Area scaling analysis (ASA) of the GzB⁺ cells was used to evaluate antibody-dependent recruitment of NK cells (% NK ADCC) and monocytes as previously described (49).

Flow cytometry was used to assess the frequency of NK cells in human and RM PBMC, and to measure the levels of cell surface CD16. Cryopreserved PBMC were thawed and incubated overnight (18 hours) in RPMI1640 medium supplemented with 10% FBS at 37°C, 5% CO₂. The cells were then washed with PBS and stained with a viability marker (Fixable Aqua Dead Cell Stain Kit, Thermo Fisher Scientific, San Diego, CA) prior to surface staining with fluorescently conjugated monoclonal antibodies using standard techniques. The panels of fluorescently conjugated antibodies for NK cell phenotyping were modified from what previously published (88, 89). Rhesus PBMCs were phenotyped with: CD3 (Pacific Blue, clone SP34.2, BD Biosciences), CD20 (Alexa Fluor 700, clone 2H7, BD Biosciences), CD8 (APC-H7, clone SK1, BD Biosciences), NKG2A/C (CD159a/c; PE, clone Z199, Beckman-Coulter), CD14 (BV650, clone M5E2, BD Biosciences), CD16 (PE-CF594, clone 3G8, BD Biosciences). Human PBMCs were phenotyped with: CD3 (PE-CF594, clone S4.1/7D6, ThermoFisher), CD4 (PerCPCy5.5, clone OKT4, eBioscience), CD8 (BV650, clone RPA-T8, Biolegend), NKG2A/C (CD159a/c; PE, clone Z199, Beckman-Coulter), CD14 (PE-Cy5, clone Tuk4, Life Technologies), CD19 (PE-Cy5, clone SJ25-C, Invitrogen), CD16 (APC-Cy7, clone 3G8, Beckman Coulter), CD56 (PE-Cy7, clone NCAM16.2, BD Biosciences).

Data analyses were performed using FlowJo software (v10.5.3). RM NK cells were defined as live CD3^- CD14^- CD20^- CD8⁺ NKG2A/C⁺, as previously described (90). CD16 expression for human and RM cells is represented by APC-Cy7 or PE-CF594, respectively, median fluorescence intensity (MFI) of the CD16⁺ NK cell population. All samples were analyzed on a BD LSRFortessa flow cytometer (BD Biosciences) that has been optimized and rigorously maintained under quality control procedures described by Perfetto and colleagues (91).

Phosphorylation of CD3ζ chain (pCD3ζ) was determined as previously described (27). BD Perm buffer III (BD Biosciences, La Jolla, CA) was used for phospho-flow analyses according to the manufacturers’ protocols. Briefly, cells were opsonized with anti-human CD16 (BD Biosciences, clone 3G8, La Jolla, CA) followed by cross-linking with secondary goat-anti-mouse F(ab)’2, (Jackson ImmunoResearch Laboratories, West Grove, PA). Reactions were immediately stopped after 2.5 minutes (experimentally determined, data not shown) with equal volumes of Fix Buffer I (BD Biosciences, La Jolla, CA). After washing, unused/unoccupied sites of the secondary antibody were blocked with normal mouse serum (ThermoFisher) for 10 minutes at room temperature. Fixation was followed by surface stain, permeabilization with Perm Buffer III for 30 minutes on ice, followed by intracellular staining at room temperature according to the manufacturer’s protocol.

ADCP was performed as previously described (56, 92). Briefly, quantification of ADCP was performed by covalently binding SHIV1157 QNE Y173H gp120 protein (7) to fluorescent beads and forming immune complexes by incubating in the presence of combination of four rhesus mAbs each at 25 µg/mL. Immune complexes were then incubated in the presence of rhesus monocytes isolated (Miltenyi Biotec catalog #130-091-097) from cryo-preserved rhesus PBMCs rested overnight (18 hours) in RPMI1640 medium supplemented with 10% FBS at 37°C, 5% CO₂ and then washed with PBS. Fluorescence of the cells was detected using flow cytometry (BD Fortessa cytometer). The magnitude of the ADCP immune response was calculated as an ADCP score by multiplying the mean fluorescence intensity (MFI) and frequency of phagocytosis-positive cells and dividing by the MFI and frequency of the bead-positive cells in an antibody-negative (PBS) control well. The influenza specific rhesusized CH65-IgG1 (anti-HA) antibody (87) at the equivalent concentration of 100 μg/mL was used as a negative control. Each sample was tested once, and ADCP score is the mean score from two biological replicates.

Statistical analyses were performed using R statistical software (93). Clusters were identified using Ward’s hierarchical clustering based on Euclidian distance. Clustering was performed using ADCC, % NK ADCC, % pCD3ζ+, and CD16 expression data with and without the available demographic variables gender, age, race, weight and CMV status. However, these demographic variables did not impact the clustering results in humans or RMs. Principal components analysis was performed using the same 4 functional measurements that went into the cluster analysis and visualized using the ‘factoextra’ package (94). Five humans were missing % pCD3ζ+ values and one RM was missing a CD16 expression value; these subjects were excluded from clustering and PCA. Prior to clustering and PCA, CD16 expression values were log-transformed, and then ADCC, % NK ADCC, % pCD3ζ+, and CD16 were each scaled to have mean 0 and standard deviation 1 within each species cohort. To obtain the groups shown in Figure 5 , we used covariate constrained randomization to assign the potential treatment arms based on the four variables discussed. The randomization was done using the R package cvcrand (95), examining 50,000 potential randomization schemes with the potential assignment sampled from the lowest 10% of the resulting inverse variance weighted balance scores (96).

Kruskal-Wallis tests were used when there were more than 2 groups for comparison ( Figure 4 ). Two-sided Mann-Whitney U tests were used for pairwise comparisons ( Figures 1 , 4 , 5 ). Two-sided Fisher’s exact tests were used to test for association of categorical variables ( Table 1 ). All analyses were performed separately by species, except for the direct comparison of humans and macaques shown in Figure 1 . All comparisons were considered significant at the 0.05 level. Due to the exploratory nature of these analyses, no corrections were made for multiple testing.