Antigen-boosted CD4CAR T cells fail to expand or control viremia in multiple nonhuman primate models of HIV

Key Points

• •Despite promise in mouse models, HIV-specific CAR T-cell activity across multiple immunocompetent macaque models is minimal.
• •Inactivation of candidate immune checkpoint genes is insufficient to reinvigorate HIV-specific CAR-T cells in vivo.

Visual Abstract

Abstract

Chimeric antigen receptor T (CAR T)-cell therapy has demonstrated curative potential in B-cell malignancies; yet, translating this success to chronic infections such as HIV remains a major challenge. In people living with HIV who are receiving suppressive antiretroviral therapy (ART), low-antigen levels limit CAR T-cell expansion and persistence. We previously reported data from a pilot study suggesting that HIV-targeted CD4CAR T cells could overcome this barrier through exogenous antigen supplementation, leading to robust in vivo expansion. Here we sought to comprehensively confirm and expand on those findings. We tested a broad array of strategies to enhance CD4CAR T-cell efficacy, including CRISPR-Cas9-mediated gene editing of immune checkpoint and HIV-associated genes, single and pooled competitive infusions of engineered CAR T cells, distinct CAR constructs incorporating either CD28 or 4-1BB costimulatory domains, and exogenous antigen boosting. We also developed highly sensitive droplet digital polymerase chain reaction assays to quantify CAR T-cell frequency and to corroborate the flow cytometry–based quantification of CD4CAR T-cell expansion. We evaluated these new approaches across multiple nonhuman primate (NHP) models of HIV, including both simian immunodeficiency virus– and simian-human immunodeficiency virus–infected, ART-suppressed NHPs. Although CD4CAR T-cell products exhibited antigen-specific proliferation and cytotoxicity ex vivo, they failed to expand, persist, or control viremia in vivo. We were also unable to confirm the previously observed CD4CAR T-cell expansions from our earlier studies, which have been retracted. Together, these data highlight the need for alternative strategies to potentiate anti-HIV CD4CAR T cells in the immunocompetent setting.

Introduction

Adoptive transfer of chimeric antigen receptor (CAR) T cells has revolutionized the treatment of hematologic malignancies and has emerged as a promising strategy for addressing chronic viral infections, such as HIV. However, HIV presents unique barriers to successful CAR T-cell therapy that differ substantially from those in the cancer setting. In people living with HIV receiving durable antiretroviral therapy (ART), viral antigen levels are extremely low, thereby limiting CAR T-cell expansion and persistence.¹ In addition, HIV persists in difficult-to-access tissue reservoirs, such as gastrointestinal-associated lymphoid tissue and the central nervous system,² posing additional hurdles for effective CAR T-cell trafficking and clearance. The high mutability of HIV also presents a challenge for targeting viral epitopes.³ Our group and others have developed HIV-targeted CAR T cells, termed CD4CAR T cells, that incorporate the extracellular domain of CD4 as the targeting moiety, thereby leveraging the conserved nature of the CD4 binding site to minimize susceptibility to viral escape mutations.4, 5, 6 Preclinical studies in humanized mouse models of HIV-1 infection have demonstrated that CD4CAR T cells are able to expand in response to antigen, provide CD4⁺ T cells with partial protection against infection, and reduce viremia after ART cessation.^5,6

Previously, we reported that exogenous antigen supplementation using K562 cells engineered to overexpress the HIV envelope glycoprotein (K562-Env) could overcome the low-antigen environment conferred by ART and enable robust CD4CAR T-cell proliferation in simian-HIV (SHIV)-infected, ART-suppressed nonhuman primates (NHPs).⁴ This strategy was conceptually inspired by analogous efforts in oncology, such as the use of lipid nanoparticle-delivered claudin-6 antigen to boost anti–claudin-6 CAR T-cell responses.⁷ In our previous study, all 4 CAR-treated, ART-suppressed NHPs received irradiated K562-Env cells and showed expansion of CD4CAR T cells up to 40% in peripheral CD3⁺ T cells, with 2 animals exhibiting transient control of viral replication following ART withdrawal. In the 2 animals without initial viral suppression, subsequent administration of a rhesusized anti–programmed cell death protein 1 (anti–PD-1) antibody (nivolumab) induced CAR T-cell reexpansion and a temporary reduction in viral load. To our knowledge, this was the first demonstration of meaningful expansion of HIV-targeted CAR T cells in an NHP model of HIV. In this study, we sought to reproduce and extend these findings through systematic engineering of CD4CAR T cells and evaluation in multiple NHP models of HIV. It is important to note that reevaluation of samples from our previous study failed to confirm the expansion of CD4CAR T cells reported at that time.

Methods

CD4CAR T-cell design

The CD4CAR transgene was adopted directly from our previous studies, which included a long EF1α promoter that drives CD4CAR expression and the incorporation of domains 1-4 of the CD4 ectodomain and a CD8 hinge and transmembrane domain.^4,5 For phase 1A and 1B (SHIV) studies, the CD4CAR incorporated a 4-1BB costimulatory domain, whereas in phase 2 simian immunodeficiency virus (SIV) studies, we co-transduced cells with dual CD4CAR constructs that incorporated either a 4-1BB or a CD28 costimulatory domain.

In vivo NHP CAR T-cell studies

Rhesus macaques were housed in accordance with the National Institutes of Health standards as outlined in the Guide for the Care and Use of Laboratory Animals (National Research Council, 1996),⁸ the Institute for Laboratory Animal Research (ILAR) recommendations, and the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) accreditation guidelines, as previously described.⁴ All animal procedures were approved by the institutional animal care and use committees of Fred Hutchinson Cancer Research Center/University of Washington (protocol no. 3235–04).

Statistical analysis

Statistical comparisons were performed using paired 2-tailed t tests.

Results

CD4CAR T cells failed to expand or control viremia in SHIV-infected, ART-suppressed NHPs

We first evaluated CD4CAR T cells in a study design that was analogous to our previous work⁴ that used SHIV-infected, ART-suppressed NHPs. When compared in figures, we refer to data from this previous study as Rust et al.⁴ The first cohort in our follow-up study, termed phase 1A (SHIV), was specifically designed to assess the independent contributions of CCR5 gene editing and exogenous antigen boosting to CAR T-cell expansion and antiviral activity. Animals were infected with SHIV-1157ipd3N4 for 7 weeks, treated with ART for 37 to 45 weeks, and subsequently administered CD4CAR T cells and boosted with K562-Env cells, followed by ART withdrawal (Figure 1A). We hypothesized that administering CD4CAR T cells and the K562-Env boost during ART suppression would prime the CAR T cells in a controlled antigen environment, thereby enabling them to respond rapidly to recrudescent virus after ART withdrawal and limit further viral dissemination. The animals were divided into 3 treatment cohorts: (1) CD4CAR T cells without antigen boosting (CAR-T, n = 4), (2) CD4CAR T cells with K562-Env antigen boosting (CAR-T + boost, n = 2), and (3) CCR5-knockout CD4CAR T cells with K562-Env boosting (ΔCCR5 CAR-T + boost, n = 3). We performed tissue collections at key time points throughout the study, including before CAR T-cell infusion, after CAR T-cell infusion before the boost, after the boost, and after ART withdrawal. An overview of the animal characteristics, including SHIV strain, animal identification (ID), age, and CAR T-cell design, is provided in supplemental Table 1 (available on the Blood website).

Figure 1.

Overview of phase 1A (SHIV) NHP study. (A) Phase 1A (SHIV) study design overview. Rhesus macaques were infected with SHIV-1157ipd3N4, treated with ART, and administered CD4CAR T cells with or without K562-Env boosting, followed by ART withdrawal. Stars (∗) indicate tissue collection time points. (B) Manufacturing schema for CD4CAR T cells, including separate isolation and activation of CD4⁺ and CD8⁺ T cells, lentiviral transduction with CD4CAR vectors, CRISPR-mediated CCR5 editing, and G-REX expansion.

CD4CAR T cells were manufactured using established protocols,^4,9 including separate isolation of CD4⁺ and CD8⁺ T cells, activation with artificial antigen-presenting cells (aAPCs), lentiviral transduction, CRISPR–ribonucleoprotein–mediated CCR5 editing, and expansion in G-REX flasks (Wilson Wolf, Saint Paul, MN) (Figure 1B). On average, 2.35 × 10⁸ total cells per kg, including 9.144 × 10⁷ CAR⁺ cells per kg, were infused into each animal (Figure 2A). The transduction efficiencies ranged from 32.7% to 47.15% (average, 40.6%), and CCR5 editing efficiency, as measured by amplicon-based next-generation sequencing of the CCR5 locus followed by bioinformatic analysis by CRISPResso2,¹⁰ in the ΔCCR5 CAR-T group, ranged from 76.11% to 96.24% (average, 87.39%) (Figure 2B). Following CAR T-cell infusion, K562-Env boosting, and ART withdrawal, we did not detect CAR T-cell expansion in postinfusion peripheral blood mononuclear cells (PBMCs; Figure 2C). Consistent with this lack of CAR signal in peripheral blood, CAR T cells were undetectable in tissues that were collected longitudinally throughout the study, including samples collected from preinfusion, at 2-, 4-, and 6-weeks postinfusion, and at necropsy (supplemental Figure 1). To further investigate potential factors that contributed to the differences from our previous study, we compared plasma from the Rust et al⁴ cohort with that from our phase 1A (SHIV) cohort at matched time points. We hypothesized that differences in the peripheral immune environment could explain the observed discrepancies in CAR T-cell expansion. Across the 37 analytes measured, we detected no significant differences in cytokines, chemokines, or growth factors (supplemental Figures 2-4). In addition, we assayed plasma for anti-Env antibodies based on the hypothesis that differences in immunologic responses to the K562-Env antigen boost in phase 1A (SHIV) animals may have contributed to the lack of in vivo CAR T-cell expansion observed in this cohort. However, although all boosted animals across our previous study and the phase 1A study showed increases in plasma anti-Env antibodies following K562-Env boost (supplemental Figure 5A), we did not detect any significant differences in the level of anti-Env antibodies at any time-matched time points between the phase 1A (SHIV) and the Rust et al⁴ plasma samples (supplemental Figure 5B).

Figure 2.

CD4CAR T cells failed to expand or control viremia in SHIV-infected, ART-suppressed NHPs in phase 1A (SHIV). (A) Absolute number of infused T cells and CD4CAR⁺ T cells per kg body weight across treatment groups. (B) CD4CAR transduction efficiency and CCR5 editing efficiency in infusion products. (C) Frequency of CD4CAR⁺ T cells in PBMCs following infusion, K562-Env boosting, and ART withdrawal. (D) Longitudinal plasma viral load measurements following at SHIV infection and ART initiation. (E) Plasma viral loads following CAR T-cell infusion and ART withdrawal. Each data point represents an individual NHP. Flow cytometry was used to quantify the percentage of CD3⁺CD4CAR⁺ cells in panels A to C. LOQ, limit of quantification.

Following ART interruption, viral rebound occurred in a subset of animals across all groups, whereas others remained suppressed through necropsy (Figure 2D-E). However, the lack of CAR T-cell expansion in these animals suggests that viral control was not CAR mediated but instead reflected the high rate of spontaneous control observed in the SHIV model, which is a known limitation of this system.¹¹ Consistent with this, many animals in our cohort exhibited declining viremia or spontaneous viral control before ART initiation (Figure 2D). We also evaluated an additional cohort of ART-suppressed, SHIV-infected NHPs, referred to as phase 1B (SHIV). Animals either received no CAR T cells (no CAR; n = 3), thereby serving as a control for assessing viral rebound after ART withdrawal in the absence of CD4CAR T-cell therapy and K562-Env boosting, or received CCR5-edited CD4CAR T cells that were manufactured using a revised protocol and K562-Env regimen (ΔCCR5 CAR-T + boost [MOD]; n = 2; supplemental Table 1). In this modified (MOD) arm, we reduced the soluble interleukin-15 (IL-15) supplementation from 5 ng/mL to 0.5 ng/mL during CAR–T-cell manufacturing and activated T cells with a modified aAPC-based approach. In this approach, aAPCs were cultured in soluble IL-15 and were engineered to overexpress the IL-15 receptor alpha (IL-15Rα), thereby providing superagonist-like IL-15 signaling during T-cell activation.¹² CAR T cells were manufactured from PBMCs that were collected both before and after SHIV infection based on the hypotheses that (1) SHIV-specific T cells might synergistically enhance HIV/SHIV-specific CAR-T potency and (2) using postinfection PBMCs better model real-world scenarios in which preinfection samples are unavailable. These manufacturing changes are summarized in supplemental Table 2. Finally, we adjusted the treatment timeline as follows: K562-Env boosting and ART withdrawal were both performed 7 days after CAR T-cell infusion. Despite these protocol changes, ΔCCR5 CAR-T + boost (MOD) animals showed no CAR T-cell expansion and no measurable impact on viral load when compared with the no CAR controls (supplemental Figure 6). Taken together, these findings suggest that CD4CAR T-cell expansion in SHIV-infected NHP models may be limited.

Enhanced CD4CAR T-cell manufacturing in a high viral antigenemia model

To address the limitations observed in our phase 1A and phase 1B (SHIV) cohorts in which low antigenicity and/or weak/delayed viral rebound may have impaired CAR T-cell activation and expansion, we transitioned to a SIV-infected NHP model for phase 2 [hereafter referred to as phase 2 (SIV)]. When compared with SHIV, SIV infection induces more consistent viral rebound after ART withdrawal and has a lower rate of spontaneous viral control, thereby providing a more robust platform for evaluating CAR T-cell function in vivo.^13,14 Building on our previous study, which suggested that CAR T cells could be reinvigorated through immune checkpoint blockade via administration of an anti–PD-1 antibody,⁴ phase 2 was designed to assess whether targeted disruption of additional immune checkpoint and HIV-associated genes could further enhance CAR T-cell expansion and persistence. Eight genes were selected for knockout, namely 6 immune checkpoint regulators (PD-1,¹⁵ TIGIT,¹⁶ CTLA-4,¹⁷ DGKα,¹⁸ LAG-3,¹⁹ TIM-3²⁰), CCR5,²¹ and FLI-1, a gene linked to T-cell effector (T_EFF) function identified from in vivo CRISPR screens.²² We evaluated 3 cohorts of SIV-infected, ART-suppressed NHPs (supplemental Table 3; Figure 3A). The first received CCR5-knockout CD4CAR T cells combined with K562-Env boosting (ΔCCR5 CAR-T, n = 3). The second received a pooled infusion product that contained single-plex, gene-edited CD4CAR T cells targeting 1 of the 8 genes listed above in combination with K562-Env antigen boosting (CRISPR array CAR-T, n = 3). The third received no CAR T-cell infusion (no CAR, n = 3). For the CRISPR array group, single-plex gene-edited CAR T cells were manufactured individually in an arrayed format following the same standard workflow as for ΔCCR5 CAR T cells, pooled in equal proportions, and infused according to our established competitive repopulation study design.²³ Longitudinal PBMC samples were collected pre- and postinfusion, and editing at each locus was quantified over time to determine whether specific knockouts became enriched or depleted relative to baseline. This approach provided a surrogate measure of the fitness advantage or disadvantage conferred by each gene edit (Figure 3B).

Figure 3.

Overview of phase 2 (SIV) study design and pooled CRISPR CAR-T screening approach. (A) Schematic of the phase 2 (SIV) in vivo study. Rhesus macaques were infected with SIVmac239M, treated with ART, and infused with CD4CAR T cells, followed by K562-Env boosting and ART withdrawal. Stars (∗) indicate tissue collection time points. (B) Design of the pooled CRISPR array screen. CD4CAR T cells gene-edited at distinct immunoregulatory or HIV susceptibility genes were manufactured ex vivo, pooled, and infused into single SIV-infected, ART-suppressed NHPs. On-target editing was quantified by next-generation sequencing at longitudinal time points to detect enrichment or depletion of specific edits, indicating potential in vivo selective advantages or disadvantages. (C) Summary of CAR T-cell manufacturing, including CD4⁺ and CD8⁺ T-cell isolation, CRISPR-mediated gene editing, activation with aAPCs, lentiviral transduction with CD4CAR, and expansion in G-REX flasks.

In addition to changes in the HIV model and study design, several refinements were introduced into the CAR T-cell manufacturing process and dosing scheme (Figure 3C; supplemental Table 2) to optimize the CAR T-cell fitness, improve in vivo persistence, and better simulate clinically relevant CAR T-cell manufacturing processes. As in phase 1B (SHIV), CAR T cells were activated with aAPCs that were engineered to overexpress IL-15Rα. We also removed soluble IL-15 from the T-cell media to test whether aAPC-expressed IL-15Rα was sufficient to support CAR T-cell function. CAR T cells were manufactured from PBMCs that were collected both before and after SIV infection. Here, we refer to SIV⁻ and SIV⁺ fractions as input T cells that were collected before SIV infection (SIV⁻) or after SIV infection and ART suppression (SIV⁺); notably, the SIV⁺ fraction was not entirely composed of infected cells but rather may have contained a subset of SIV-infected and SIV-specific T cells. Residual ART in SIV⁺ cells should minimize recrudescent virus replication during CAR manufacturing, but it also limits the potency of our lentiviral vectors.²⁴ To mitigate the risk for chromosomal loss associated with CRISPR electroporation following T-cell activation,²⁵ the workflow was adjusted such that ribonucleoprotein electroporation was performed on day 0, before aAPC stimulation. Finally, based on previous reports that suggested enhanced persistence and expansion with mixed costimulatory domains,⁶ CAR T cells were cotransduced with dual lentiviral vectors that encoded CD4CAR constructs that containted either a 4-1BB or a CD28 costimulatory domain.

Gene-edited CD4CAR T cells demonstrate antigen-specific cytotoxicity and proliferation ex vivo

We manufactured the CD4CAR T cells using our revised protocol (supplemental Table 2) and infused either pooled or single gene-edited CAR T-cell products at an average dose of 1.45 × 10⁸ total cells per kg, including 3.315 × 10⁷ CAR⁺ cells per kg (Figure 4A). In the CRISPR array cohort, pooled infusions were designed to balance each of the 8 distinct single-plex gene-edited products across the SIV⁻ and SIV⁺ CAR T-cell fractions (Figure 4B). CAR expression among gene-edited products ranged from 6.6% to 45% (Figure 4C-D) with higher transduction efficiency in SIV⁻ fractions (average, 24.8%) than in SIV⁺ fractions (average, 14.4%), likely as a consequence of ART-mediated inhibition of lentiviral transduction. Amplicon-based next-generation sequencing showed robust on-target editing across all 8 loci (Figure 4E-F) with an average editing efficiency of 91.2% in SIV⁻ CAR T cells and 75.8% in SIV⁺ CAR T cells. Given these high editing efficiencies, we expected the majority of CD4CAR T cells to show biallelic editing. We hypothesized that the lower editing rates observed in SIV⁺ CAR T cells may reflect impaired cellular fitness because of SIV infection.

Figure 4.

Composition and characterization of phase 2 (SIV) CD4CAR T-cell products. (A) Total CAR T-cell dose and CD3⁺CD4CAR⁺ cell dose infused into animals in the ΔCCR5 and CRISPR array CAR T-cell cohorts. (B) Relative composition of the CRISPR array CAR T-cell products for each indicated animal, based on the frequency of CD3⁺CD4CAR⁺ cells corresponding to each individual edit, as measured by flow cytometry. (C-D) CAR transduction efficiency in ΔCCR5 CAR T-cell products (C) and individual gene-edited products within the CRISPR array CAR T-cell cohort (D), shown separately for cell products manufactured from T cells collected before (SIV^˗) or after (SIV⁺) SIV infection and ART suppression. (E-F) On-target editing efficiency measured by next-generation sequencing in ΔCCR5 CAR T cells (E) and individual gene-edited CAR T cells within the CRISPR array cohort (F), separated by SIV status. In panels C-F, the filled and open circles represent SIV^˗ and SIV⁺ CAR T-cell fractions, respectively.

We rigorously evaluated the function of gene-edited CAR T cells ex vivo before infusion into SIV-infected, ART-suppressed NHPs. To assess cytotoxicity, we cocultured gene-edited CAR T cells with either antigen-expressing (LLCMK2-Env) or antigen-negative (LLCMK2-GFP) target cells and measured target cell killing using the xCELLigence Real-Time Cell Analysis platform. In both the ΔCCR5 and CRISPR array CAR T-cell groups, gene-edited CAR T cells exhibited robust antigen-specific cytotoxicity against antigen-positive targets with minimal nonspecific killing of antigen-negative cells (Figure 5A-D). We further assessed the proliferative capacity using a CellTrace Violet-based proliferation assay (CellTrace Violet; Life Technologies, Carlsbad, CA). Upon coculture with antigen-positive K562-Env target cells, all gene-edited CAR T cells demonstrated antigen-specific proliferation, indicated by the dilution of CellTrace Violet within the CAR⁺ T-cell fraction (Figure 5E-H). No evidence of nonspecific CAR signaling was observed in the absence of antigen stimulation. In summary, all CRISPR-edited CAR T cells retained core antigen-specific killing and proliferation, and we did not observe significant differences among the individual gene-knockout products in these short-term assays. Longer-term repeated antigen stimulation assays may be required to reveal functional differences imparted by the distinct gene knockouts.²³

Figure 5.

Ex vivo antigen-specific cytotoxicity and proliferation of gene-edited CD4CAR T-cell products from phase 2 (SIV). (A-D) Cytotoxicity was measured using real-time impedance-based cytolysis assays with Env-positive (LLCMK2-Env) or control (LLCMK2-GFP) target cells. Killing activity was reported as area under the curve (AUC) over 48 hours. (A) Cytolytic activity of ΔCCR5 CAR T cells, shown for SIV^˗ and SIV⁺ fractions. (B) Summary AUC values for ΔCCR5 CAR T cells, averaged across paired SIV^˗ and SIV⁺ fractions for each NHP. (C) Cytolytic activity of individual gene-edited CAR T-cell products from the CRISPR array cohort. (D) Summary AUC values for CRISPR array CAR T cells, averaged across SIV^˗ and SIV⁺ fractions for each edited product (8 edits per NHP; n = 3 NHPs). (E-H) CAR T cells were labeled with CellTrace Violet and co-cultured with Env-expressing (K562-Env) or control (wild-type K562) cells. Proliferation was calculated as the ratio of CellTrace Violet (mean fluorescence intensity) between stimulated and control conditions (see Methods). (E) Proliferation of ΔCCR5 CAR T cells, shown separately for CAR⁺ and CAR⁻ subsets across SIV^˗ and SIV⁺ fractions. (F) Summary proliferation ratios for ΔCCR5 CAR T cells, averaged across SIV^˗ and SIV⁺ fractions for each NHP. (G) Proliferation of individual gene-edited CAR T-cell products from the CRISPR array cohort, shown for CAR⁺ and CAR^˗ subsets across SIV^˗ and SIV⁺ fractions. (H) Summary proliferation ratios for CRISPR array CAR T cells, averaged across SIV^˗ and SIV⁺ fractions for each edited product (8 edits per NHP; n = 3 NHPs). In panels A,C,E,G, the filled and open circles represent SIV^˗ and SIV⁺ CAR T-cell fractions, respectively. Statistical comparisons in panels B,D,F,H were performed using paired 2-tailed t tests. ∗P < .05; ∗∗∗∗P < .00005. GFP, green fluorescent protein.

Gene-edited CD4CAR T cells failed to expand or control viremia in SIV-infected, ART-suppressed NHPs

We observed robust plasma viremia following SIVmac239M infection of NHPs, and the initiation of ART effectively suppressed viral replication (Figure 6A). However, after CAR T-cell infusion followed by K562-Env boosting and ART withdrawal, we detected minimal CAR T-cell expansion by flow cytometry across both groups; the ΔCCR5 CAR T cells expanded only up to 5.56% of the CD3⁺ T cells at 10 days after infusion, and the CRISPR array CAR T cells showed minimal persistence above 2% of the total CD3⁺ T cells throughout the study duration (Figure 6B). Similarly, we detected minimal (<1%) CD4CAR T cells in peripheral tissues at multiple time points after infusion (supplemental Figure 7). Given the low CAR T-cell abundance observed and the potential for technical artifacts with flow cytometry, such as subjective gating and high background signal, we quantified CAR T cells at the DNA level using highly sensitive digital droplet polymerase chain reaction (ddPCR) assays. Analysis of the infusion products confirmed CAR-specific amplification and high vector copy numbers (VCNs) that ranged from 0.70 to 1.63 (Figure 6C). In postinfusion PBMC samples, ddPCR detected very low levels of CAR DNA with peak expansion occurring at day 7 (Figure 6D). The average VCN measured by ddPCR in CD3⁺ T cells was 0.019 in the ΔCCR5 cohort and 0.005 in the CRISPR array cohort, corresponding to ∼0.5% to 1.9% CAR marking in peripheral CD3⁺ T cells across the 2 groups, concordant with the observed flow cytometry data. Viral rebound occurred between 10- and 20-days following ART withdrawal in all animals (Figure 6E). Because ddPCR cannot specifically detect individual gene-edited CAR T cells but instead detects total CAR DNA, we assessed the editing frequencies across the 8 target loci in postinfusion PBMCs to determine whether any specific gene-edited T-cell populations preferentially expanded in vivo. In the ΔCCR5 CAR T-cell cohort, the peak editing levels at day 7 ranged from 1.2% to 4.14% of the total PBMCs, declining to 1.16% to 2.44% by day 28 (Figure 6F). In the CRISPR array cohort, editing was generally lower, with most edits detected at frequencies below 1% at all the time points assessed (Figure 6G-I). Together data indicate that gene-edited CD4CAR T cells failed to expand, persist, or exert measurable antiviral activity in SIV-infected, ART-suppressed NHPs, and specific knockout of inhibitory molecules failed to improve CAR T-cell activity.

Figure 6.

Gene-edited CD4CAR T cells failed to expand or control viremia in SIV-infected ART-suppressed NHPs. (A) Longitudinal plasma viral load measurements following SIVmac239M infection and ART treatment. (B) Frequency of CAR T cells in PBMCs following infusion, K562-Env boosting, and ART withdrawal. (C) VCN in CD3⁺ T cells from ΔCCR5 and CRISPR array CAR T-cell infusion products. (D) VCN in CD3⁺ T cells from PBMCs collected longitudinally before and after CAR T-cell infusion. (E) Plasma viral load measurements following CAR T-cell infusion and ART withdrawal. (F) Editing efficiency at the CCR5 locus in PBMCs from animals treated with ΔCCR5 CAR T cells. (G-I) Editing efficiency across CRISPR array targets (PD-1, TIGIT, CTLA-4, LAG-3, DGKα, TIM-3, FLI-1, and CCR5) in PBMCs from animals treated with CRISPR array CAR T cells: A20031 (G), A20028 (H), and A20032 (I). Each data point in panels A-F represents an individual NHP; each data point in panels G-I represents gene-editing measurements across the 8 edited loci for the specified NHP.

No CD4CAR T-cell expansion detected in reevaluation of previous data

Given the lack of CAR T-cell expansion observed in phase 1A (SHIV), phase 1B (SHIV), and phase 2 (SIV) studies across a total of 17 CD4CAR T-cell–treated NHPs, we reevaluated our previous findings.⁴ Postinfusion PBMC samples from Rust et al⁴ that corresponded to time points of peak CD4CAR T-cell expansion were analyzed using single-cell RNA sequencing,²⁶ but bioinformatic analysis failed to detect CD4CAR⁺ cells in any samples analyzed (data not shown). We additionally analyzed postinfusion PBMC samples with our established ddPCR assay and found no evidence of CAR T-cell expansion, particularly at 30 to 40 days after infusion, which corresponds to the time frame of peak expansion previously reported (Figure 7A). Instead, ddPCR measurements closely mirrored those from our phase 2 (SIV) cohorts. As a positive control, we applied a comparable ddPCR workflow to postinfusion PBMCs from a separate study (Bui et al) in which NHPs received CD20-targeted CAR T cells.⁹ These samples showed robust CAR T-cell expansion and corroborated previously reported flow cytometry data, with VCNs reaching a peak of up to 1.976 at 7 to 10 days after infusion (Figure 7B). Prompted by these findings, we reanalyzed the original flow cytometry data that had formed the basis for the reported 15% to 40% CAR T-cell expansion in our previous study.⁴ Although the original gating strategy was established with appropriate negative controls (eg, preinfusion PBMC samples or fluorescence-minus-one [FMO] controls) to distinguish CAR⁺ from CAR⁻ events within ex vivo–manufactured CAR T-cell infusion products, in vivo blood samples from CAR T-cell–treated animals in our previous study⁴ were not routinely run with these controls. To address this limitation, we restained available postinfusion PBMC samples from the reported expansion phase in our previous study and included both negative and positive controls to establish well-defined flow cytometry gates. When gated on preinfusion PBMCs, a true biologic negative control (supplemental Figure 8), these samples showed no appreciable CD4CAR signal above the background (supplemental Figure 8C). When gated on a representative FMO control (supplemental Figure 9A), these samples showed CD4CAR marking of between 0.56% and 3.3% (supplemental Figure 9C), although preinfusion PBMCs also showed a high background of 2.16% (supplemental Figure 9D), and ex vivo–manufactured CAR T cells that were stained in parallel showed clear overlap of the applied CD4CAR gate into the CD4CAR⁻ population (supplemental Figure 9B). A summary of these findings, including experimental distinctions between the 2 negative control samples (ie, preinfusion PBMCs and FMO sample) that were used to establish gates in supplemental Figures 8 and 9, is provided in supplemental Table 4. These 2 analytical approaches emphasize both the importance of including FMO controls to aid in the quantification of low frequency events²⁷ and the critical need to corroborate flow cytometry findings from CAR T-cell studies with more quantitative and sensitive methods, including ddPCR. Although we were unable to restain all original PBMC time points by flow cytometry due to limited sample availability, our extensive ddPCR analysis, together with repeated flow cytometry data, collectively suggest that minimal CAR T-cell expansion occurred in our prior study.

Figure 7.

Quantification of CAR T cells in postinfusion PBMCs by ddPCR across NHP studies. (A) VCN in CD3⁺ T cells from postinfusion PBMCs from Rust et al⁴ and phase 2 (SIV) animals treated with ΔCCR5 or CRISPR array CD4CAR T cells. (B) VCN in CD3⁺ T cells from postinfusion PBMCs from Rust et al⁴ (as in panel A), and Bui et al,⁹ which used a CD20-targeted CAR T cell. The gray shaded regions indicate the reported expansion phase in Rust et al.⁴

HIV-1 gp120 binding does not alter anti-CD4 immunolabeling

We tested several biologic hypotheses to reconcile our PCR data (Figure 5) with the flow cytometry data reported in our previous study. We focused on Env-dependent alterations to the anti-CD4 antibody affinity,^28,29 because these properties were central to our flow cytometry–based CAR immunolabeling approach. Briefly, Bachelder et al reported that protein-protein interactions between CD4 and HIV-1 Env alter CD4 structure and augment the binding of specific anti-CD4 antibody clones.²⁸ Our current and previous studies used a CAR that contained the human D1 to D4 domains of CD4, thereby enabling us to use a biotin-conjugated anti-CD4 clone RPA-T4 that is specific for the human CD4 sequence within our CAR. We have previously shown and routinely confirmed in ex vivo assays that this approach specifically labels human CD4-based CAR T cells with little/no background from endogenous macaque CD4³⁰ and devised a similar approach for tissue-based immunohistochemistry assays.³¹ Nevertheless, our conflicting ddPCR data (Figure 7) compelled us to revisit our flow cytometry methodology. Building on Bachelder et al,²⁸ we hypothesized that the binding of Env proteins (from K562-Env administration and/or recrudescent SHIV replication) to macaque CD4 in our previous study⁴ may have altered the CD4 conformation and thus the binding of the human CD4-specific RPA-T4 antibody clone. Hence, the flow cytometry signal that we interpreted as CD4CAR protein may have actually represented Env-bound NHP CD4. To test this hypothesis, we devised an analogous experiment to that of Bachelder et al²⁸ to investigate CD4CAR-independent labeling of NHP CD4 by anti-CD4 clone RPA-T4. First, we preincubated PBMCs from 3 naïve rhesus macaque donors with recombinant HIV-1 SF162 gp120 Env, HXBc2 gp120 Env, or irradiated K562-Env cells and then labeled with anti-CD4 antibodies that were either directly conjugated to clone RPA-T4 (supplemental Figure 10) or an RPA-T4–biotin conjugate, followed by streptavidin-BV650 secondary antibodies, as used in both Rust et al⁴ and the current study (supplemental Figures 11 and 12). We set threshold flow cytometry gates using FMO controls: no RPA-T4-BV650 (direct conjugate experiment) or RPA-T4-biotin (indirect labeling experiment). We also compared CD4 surface staining with cells that were fixed and permeabilized to capture surface and internalized CD4 proteins. When the cells were surface stained without permeabilization, we observed trends consistent with an Env-dependent augmentation of RPA-T4 binding to rhesus macaque CD4 (supplemental Figure 10B). However, critical inconsistencies, including variability observed between repeated experiments, limited the interpretation of these results (supplemental Figure 10A vs B). Cells that were fixed and permeabilized for surface plus intracellular staining showed that Env preincubation had minimal impact (supplemental Figures 10C and 11C). Despite some suggestive findings, we conclude that the collective data shown in supplemental Figures 10 and 11 do not sufficiently explain the spike in anti-CD4 signal that we observed by flow cytometry in our previous study.

Discussion

HIV cure strategies face major challenges because of the persistence of long-lived viral reservoirs and the lack of therapeutic approaches that are both safe and effective. Although a few cases of sustained HIV remission have been achieved through hematopoietic stem cell transplantation with CCR5-null donor cells, thereby effectively reconstituting the immune system with HIV-resistant cells, these transplants were performed primarily to treat the underlying hematologic malignancies and HIV cure was a secondary outcome.^32,33 The intensive preconditioning regimens required for hematopoietic stem cell transplantation carry significant toxicity and mortality risks, rendering this approach unsuitable for otherwise healthy people living with HIV.³⁴ In addition, HIV persists in latent reservoirs within tissues, such as the gut and central nervous system, which are difficult to access with conventional therapies. CAR T-cell therapy as an HIV cure represents a promising alternative approach with the potential to deliver small numbers of highly potent effector cells that can persist in vivo, traffic to reservoir sites,^9,35 and selectively eliminate infected cells.

Despite efforts to improve CAR T-cell design, optimize manufacturing, and enhance trafficking and HIV resistance via gene editing, we observed no evidence of in vivo CAR T-cell expansion. Relatedly, we did not observe trafficking to key secondary tissues, including lymph nodes or gastrointestinal-associated lymphoid tissue where latently infected CAR T-cell targets are known to reside. A retrospective re-analysis of archived samples from the original study reported in Rust et al⁴ likewise failed to detect CAR T-cell expansion. Given the lack of CD4CAR T cells in reanalyzed PBMC samples from Rust et al,⁴ the transient viral control previously observed was likely independent of CAR T-cell function, reflecting inherent variability and spontaneous viral control previously observed in the SHIV model.^36,37 Collectively, our current studies and retrospective analysis do not support evidence of CD4CAR T-cell expansion in NHP models of HIV infection.

The observation of CD4CAR T-cell expansion measured by flow cytometry in our previous study, despite a lack of cell-associated CAR DNA in our current study, indicates technical and/or intractable biologic artifacts. Previous reports demonstrated that Env binding can alter CD4 conformation in ways that affect the binding of specific anti-CD4 monoclonal antibodies.^28,29,38,39 Based on this, we considered whether Env gp120 binding might induce conformational changes in macaque CD4 that modify recognition by our D1 domain–specific RPA-T4 antibody clone used to detect CD4CAR T cells. In this scenario, Env proteins from either the K562-Env boost or from reactivated SHIV replication could have bound macaque CD4 and influenced RPA-T4 binding, thereby producing a CD4⁺ signal artifact. However, assays designed to test this hypothesis did not support this mechanism. Given the negative PCR data, we must acknowledge that the CD4⁺ flow cytometry signal that we originally measured and reported in Rust et al⁴ may have been technical in nature.

The lack of curative end points in CD4CAR T-cell–treated NHP models aligns with more recent NHP data⁴⁰ and current early-stage clinical trials^41,42 but contrasts with promising results in immunocompromised mouse models.^6,43,44 Beyond the challenge of antigen scarcity, the biologic properties of the HIV Env may further limit its suitability as a CAR T-cell target. Env expression on infected cells is often sparse and short-lived, with rapid internalization and shedding reducing its stability as a target.45, 46, 47, 48 Structurally, Env is heavily glycosylated and conformationally flexible, shifting between closed and open states that may alternately expose or shield key epitopes.49, 50, 51 These features may collectively impair the ability of CAR T cells to consistently recognize and eliminate Env-expressing cells in an immunocompetent setting, even when using binding domains that are directed at conserved regions. Despite these limitations, some Env-targeting approaches, particularly those that involve broadly neutralizing antibodies (bNAbs), have recently reported promising clinical results. bNAbs can engage both infected cells and free virions, which may provide broader antiviral coverage.52, 53, 54 CD4-based CAR T cells, in contrast, may be functionally limited by binding to circulating Env-containing virions, which could act as decoys and reduce their ability to engage with infected cells. Recent trials, including the RIO study of dual bNAbs,⁵⁵ 2 studies that combined bNAbs with TLR7 agonists,^56,57 and a phase 2 study of long-acting lenacapavir with bNAbs,⁵⁸ have shown partial or sustained virologic control in subsets of participants. These promising recent reports highlight the potential of antibody-based and combinatorial immunotherapies to overcome the structural and biologic barriers that may limit the effectiveness of CD4CAR T-cell therapies and suggest a path forward for achieving durable HIV remission.

Conflict-of-interest disclosure: M.R.B. reports serving as a consultant for Interius BioTherapeutics and Capstan, Inc. J.L.R. reports being a scientific co-founder of and holding equity in BlueWhale Bio; receiving research funding from Kite Pharma, a Gilead company; and being eligible to receive milestone-based payments from Kite Pharma, a Gilead company. H.-P.K. reports serving as a consultant to and having ownership interests in Rocket Pharmaceuticals, Homology Medicines, Vor Biopharma, and Ensoma, Inc, and serving as a member of the scientific advisory board at Umoja Biopharma. The remaining authors declare no competing financial interests.