Pooled screening of CAR T cells identifies diverse immune signaling domains for next-generation immunotherapies

Abstract

Chimeric antigen receptors (CARs) repurpose natural signaling components to retarget T cells to refractory cancers but have shown limited efficacy in persistent, recurrent malignancies. Here, we introduce "CAR Pooling," a multiplexed approach to rapidly identify CAR designs with clinical potential. Forty CARs with signaling domains derived from a range of immune cell lineages were evaluated in pooled assays for their ability to stimulate critical T cell effector functions during repetitive stimulation that mimics long-term tumor antigen exposure. Several domains were identified from the tumor necrosis factor (TNF) receptor family that have been primarily associated with B cells. CD40 enhanced proliferation, whereas B cell-activating factor receptor (BAFF-R) and transmembrane activator and CAML interactor (TACI) promoted cytotoxicity. These functions were enhanced relative to clinical benchmarks after prolonged antigen stimulation, and CAR T cell signaling through these domains fell into distinct states of memory, cytotoxicity, and metabolism. BAFF-R CAR T cells were enriched for a highly cytotoxic transcriptional signature previously associated with positive clinical outcomes. We also observed that replacing the 4-1BB intracellular signaling domain with the BAFF-R signaling domain in a clinically validated B cell maturation antigen (BCMA)-specific CAR resulted in enhanced activity in a xenotransplant model of multiple myeloma. Together, these results show that CAR Pooling is a general approach for rapid exploration of CAR architecture and activity to improve the efficacy of CAR T cell therapies.

Fig. 1.. Generation and screening of a pooled library of CARs with diverse signaling domains.

(A) Our CARs combine an αCD19 ScFv (FMC63), a CD8 hinge and transmembrane domain (TMD), an intracellular signaling domain, and a CD3ζ domain (left). Forty domains from across the human proteome were codon-optimized, synthesized, pooled into a plasmid library, and packaged to generate lentivirus (right). These spanned protein families such as immunoglobulin superfamily (IgSF), T cell/transmembrane, immunoglobulin, and mucin (TIM), TNF receptor super family (TNFRSF), and triggering receptors expressed on myeloid cells (TREM). (B) A Venn diagram showing natural expression of library members across immune cell types. Expression patterns are listed in table S1. NK, natural killer cell; NKT, natural killer T cell; DC, dendritic cell. (C) Primary human CD4 and CD8 T cells were separately isolated from PBMCs for two human donors and lentivirally transduced with the library. The cells were FACS purified using T2A-GFP fluorescence within one log of mean expression to reduce variability. (D) The pooled library was repeatedly stimulated 1:1 with CD19+ or CD19- irradiated K562 tumor cells to quantify antigen specific activation (CD69), cytokine production (IFN-γ, IL-2), and proliferation (CTV: CellTrace Violet). The library was also sequenced at the specified timepoints to measure relative expansion of individual constructs. (E) Percentage of CD8+ T cells expressing different numbers of exhaustion markers (PD1, TIM3, LAG3, CD39) after a repeat stimulation assay with CD19+ irradiated K562 tumor cells. T cells expressing CD28, 4-1BB, or CD3ζ-only CARs are compared to untransduced (Untrans.) cells. Grey boxes correspond to timepoints in which no live cells remained. Significance was assessed in CD8 T cells for 2 donors using a Repeated Measures ANOVA model. FDR-corrected p < 0.05:*, p<0.001:***; ns, not significant. The second donor and data for CD4+ T cells are shown in fig. S1 A and B. (F) We used FlowSeq, a FACS and next generation sequencing (NGS)-based pooled quantification workflow, to quantify enrichment by sorting the library into bins of fluorescent signal corresponding to a functional readout, as shown by the colored histograms. We then amplified the costimulatory domain by genomic DNA extraction and PCR, and performed amplicon sequencing on each bin to estimate the phenotype for each library member.

Fig. 2.. Comparison of proliferation, expansion, and cytokine secretion identifies differences in costimulatory activity at different time scales and in different T cell types.

(A) FlowSeq measurement of proliferation are shown for CD4 (left) and CD8 (right) T cells containing different signaling domains, separately stimulated in vitro with irradiated CD19+ K562 tumor cells for 3 or 4 days (see Fig. 1D). Percentage of cells with different numbers of divisions were calculated from amplicon sequencing of sorted CTV bins. CARs are ranked from left to right by the average number of cell divisions in CD4 or CD8 cells from lowest (left) to highest (right). 4-1BB and CD28 CARs are highlighted in blue shades. On the right, two donors are shown separately along with 4-1BB and CD28 rank. (B) FlowSeq measurements of intracellular cytokine production are shown for CD4 (yellow) and CD8 (orange) T cells averaged across three independent donors, 18 hours after addition of CD19+ irradiated K562 cells. Mean of each domain is indicated by an open circle. Dashed lines indicate the cytokine production of the average library member, normalized for each donor and T cell subset. Significance was determined using a Wilcoxon rank-sum test. FC, fold-change. (C) Relative expansion over time is shown for CD4 and CD8 T cells, based on the average fold-change in library abundance from baseline before stimulation. Mean of 3 replicates from 2 donors are shown. The 6 domains with the most and least relative expansion are labeled in green and pink, respectively, with CD28 and 4-1BB labeled in blue shades. Significance is based on FDR-corrected p values derived from a linear mixed effects model (see Methods). (D) Comparison of CD4 versus CD8 proliferation and expansion after CD19+ K562 stimulation. Larger dots correspond to constructs that have significantly better performance in either CD4 T cells or CD8 T cells. The x-axis is the mean fold-change across CD4 and CD8 T cells, and the y-axis is the CD8:CD4 expansion ratio. Significance was determined using a Wald test. For all panels, FDR < 0.05:*, < 0.01:**, <0.001:***, <0.0001:****.

Fig. 3.. Multidimensional comparison of signaling domains across multiple weeks of chronic antigen stimulation identifies a subset with potent stimulatory activity.

(A) Hierarchical clustering of the CAR signaling domain library in CD4 and CD8 T cells stimulated with CD19+ K562 tumor cells. All 40 domains were clustered based on their z-scores in each assay. 8 stimulatory domains were identified on the right (green bar). Gray boxes are excluded domains where < 500 cells were detected for the assay. TNF receptor family members are marked with black dots. KLRG1 (pink), CD40 (dark purple), CD30 (light purple), 4-1BB (light blue), BAFF-R (light green), CD28 (dark blue), and TACI (dark green) are highlighted. Significance is indicated by a black border and an asterisk(Wilcoxon rank-sum test, FDR < 0.05). (B) Principal components analysis (PCA) of pooled library screen cytokine, proliferation, expansion, and activation data is shown for CD4 T cells, CD8 T cells, CD19+ and CD19− stimulation conditions, and all donors and timepoints. Chosen CARs are larger, with shapes and color indicating known function and protein family. (C) Fold-change of the proportion of selected CARs within the library at each timepoint (x-axis) over 24 days of repeated stimulation with irradiated CD19+ K562 cells as compared to the average CAR in the pooled library. CARs were measured in CD4 and CD8 primary human T cells individually in 2 to 3 biological replicates. Significant p values were derived from a linear mixed effects model; FDR < 0.05:*, < 0.01:**, <0.001:***, <0.0001:****.

Fig. 4.. Distinct signaling domains differentially affects proliferation, long-term expansion, markers of memory, and exhaustion.

(A) The timeline for arrayed proliferation assays in (B and C) is shown. Primary human CD4 and CD8 T cells were separately transduced with the 8 CARs in (B and C) and stimulated 1:1 with irradiated CD19+ or CD19- K562 cells every three days. Proliferation was assessed by CTV dilution every 9 days. (B) Relative proliferation of each CAR was quantified by the relative decrease in mean fluorescence intensity (MFI), representing the dilution of CTV dye, between two donors of CD4 or CD8 T cells. The color legend ranges from less proliferative (dark blue) to more proliferative (yellow). The x-axis indicates the day the cells were stained. White boxes are representative of CAR T cells that dropped out of culture. (C) Histograms of CTV staining of CD4 or CD8 CAR T cells in a representative donor on selected days are shown. Data summarized in (B). Both donors are shown in fig. S4A and B. AU, arbitrary units. (D) Quantification of the cumulative expansion of CD4 T cells engineered with either CD40 (purple), 4-1BB (light blue), or CD28 (dark blue) CARs and stimulated as described in (A). Co-cultures were measured every three days starting on day 3 by flow cytometry and counting beads. The y-axis measures cumulative fold-expansion every three days. (Significance was derived from a linear mixed effects model, for CD40 comparisons, all p < .001). (E) Cells were transduced and stimulated as described in (A). Every 9 days in culture, cells were rested for 6 days without additional stimulation and assessed for surface expression of PD1, LAG3, and TIM3. (F) Percentage of CAR T cells expressing 0 to 3 of the exhaustion markers PD1, TIM3, LAG3 after day 6 and day 15 as described in (E). All CARs, markers, and time points are shown in fig. S4D and E. (G) CD27 surface expression on CD8 CAR T cells was measured over 33 days, as in (E). Percentage of CD27-high cells is shown for each CAR and day on the right. All CARs and time points are shown in fig. S4G. Significance in (F and G) was assessed using a repeated measures (RM) ANOVA model, (FDR < 0.05:*, < 0.01:**, < 0.001:***, < 0.0001:****).

Fig. 5.. Comparison of signaling domains across measures of cytokine secretion, T cell signaling reporters, in vitro cytotoxicity, and in vivo solid tumor clearance.

(A) Timeline for in vitro cytotoxicity and cytokine production assays. CD4 or CD8 T cells were transduced and stimulated as in Fig. 4A. Once weekly, a portion were stained for cytokine production as described in (B). For cytotoxicity assays, a portion of T cells (Ts) were sorted from the same co-culture by FACS, rested overnight, then cultured 1:1 with mKate+ CD19+ K562 cells and imaged every 60 minutes by Incucyte for the next 3 to 5 days (C and D). The color legend is shown on the right. (B) Mean cytokine production by CD4 T cells at 1, 4, and 10 days, measured across two donors by intracellular cytokine staining. Percentage of cytokine-positive cells was averaged between two donors. (C) Cytotoxicity of CD4 or CD8 CAR T cells sorted at the indicated days was quantified at 80 and 32 of co-culture respectively, by calculating the percentage of tumor cells at each time point relative to no T cells. CARs are ranked from least to most cytotoxic for each day. Error bars indicate the standard error calculated across donors. (D) Representative plots of cytotoxicity are shown for two donors’ CAR T cells sorted at day 8 and day 22 for CD4 T cells and day 0 and day 29 for CD8 T cells, plotting the percentage of mKate+ tumor cells remaining relative to a well with no T cells (gray). Vertical dashed lines indicate the time points analyzed in (C). (E) Overall significance is shown for the cytotoxicity of CD4 and CD8 T cells from two donors for all days indicated above. Data were analyzed using a Wilcoxon signed-rank test and FDR-corrected; p<0.05:*, p<0.01:**, p<0.001:***, p<0.0001:****; ns, not significant. (F) NFκB transcriptional activity was determined using a reporter Jurkat cell line transduced with each CAR and stimulated with either CD19- or CD19+ K562s. Samples were assessed by flow cytometry. The y-axis is relative (Rel.) to untransduced cells.

Fig. 6.. Single-cell RNA-seq and CITE-seq characterize functional differences between CAR costimulatory domains.

(A) Weighted-nearest neighbor (WNN) Uniform Manifold Approximation and Projection (UMAP) embedding of scRNA-seq and CITE-seq data from stimulated and resting CAR T cells are shown for samples from two donors. UMAP separates into CD4 and CD8 lobes (left and right sides). Cells are colored by eight CD4 and nine CD8 phenotypic clusters. Bottom inset: Cells colored by cell cycle phase, donor identity, and stimulated versus resting cells. fig. S6A shows the UMAP faceted for each CAR and stimulation condition. (B) Heatmap of differentially expressed (DE) genes. For each cluster, the top 50 DE genes are ordered by hierarchical clustering of the pseudo-bulk expression z-scores for all clusters and donors. Genes in multiple clusters are only included for the cluster with the highest score. For each, four genes that are representative of the overall phenotype of the cluster are highlighted. (C) UMAP plots show relative CITE-seq expression for the surface expression of six markers associated with T cell differentiation and activation. HLA, human leukocyte antigen; IL2RA, IL-2 receptor subunit α. (D) Mean z-scores are shown for MSigDB gene modules associated with various aspects of T cell activation, metabolism, and signaling among the three major activated phenotypic clusters in CD4 and CD8 T cells. TCA, tricarboxylic acid cycle; HIF1a, hypoxia inducible factor 1 subunit alpha; OXPHOS, oxidative phosphorylation; AICD, activation-induced cell death. (E) Enrichment of stimulated CAR T cells containing different signaling domains within each phenotypic cluster. The size of each dot corresponds to the percentage of stimulated CAR T cells in a cluster and with a costimulatory domain. The color of each dot corresponds to the log2 enrichment or depletion of that CAR relative to others. Clusters are arranged with the most activated at the center to correspond to the (A) UMAP. Similar plots for resting cells and a per-donor breakdown are in fig. S6C and E, respectively. (F) Correlation of T cell gene signatures indicative of enhanced CAR T engraftment (top) and melanoma survival (bottom) with phenotypic clusters in CD4 and CD8 CAR T cells (middle column) or with CARs containing different costimulatory domains (right column). Cluster and CAR colors match those in (E). The two dots per group correspond to two separate donors. Error bars indicate 99% confidence intervals for the z-scores.

Fig. 7.. BAFF-R demonstrates potent anti-tumor in vivo activity in solid and hematological cancers.

(A) Tumors were untreated, treated with untransduced T cells, or treated with engineered CAR T cells. Tumor size was monitored over 49 days post tumor injection. CAR T cells containing the potent costimulatory domains are shown, compared to untransduced T cells and no T cell controls. Data are representative of 5 in vivo experiments using 2 human donors. (B) Tumor size as in (A), showing a CD3ζ-only control and a KLRG1 inhibitory CAR, compared to untransduced T cells and no T cells. Error bars in (A and B) indicate standard error of the mean for tumor volumes across mice with the same treatment. (C) MM1S cancer radiance after luciferin injection was plotted over time for individual mice across CAR T cells generated from two donors. Tumors were measured by BLI every 7 days for a total of over 30 days. This was repeated independently in two donors. For panels (A to C), statistical analysis was done by t test on the normalized tumor volume area under the curve (AUC). (D) Survival curves are shown mice with MM1S cancer using T cells from a representative donor; mice were followed past 100 days (Mantel Cox test). For all panels, p<0.05:*, p<0.01:**, p<0.0001:****; ns, not significant.