Granzyme F: Exhaustion marker and modulator of CAR T cell-mediated cytotoxicity

Abstract

Granzymes are a family of proteases used by CD8 T cells to mediate cytotoxicity and other less-defined activities. The substrate and mechanism-of-action of many granzymes are unknown, though they diverge among the family members. Here we show that mouse CD8+ tumor-infiltrating lymphocytes (TIL) express a unique array of granzymes relative to CD8 T cells outside the tumor microenvironment in multiple tumor models. Granzyme F was one of the most highly upregulated genes in TIL and was exclusively detected in PD1/TIM3 double-positive CD8 TILs. To determine the function of granzyme F and to improve the cytotoxic response to leukemia, we constructed novel chimeric antigen receptor (CAR)-T cells to over-express a single granzyme, granzyme F or the better-characterized granzymes A or B. Using these doubly recombinant T cells, we demonstrated that granzyme F expression improved T cell-mediated cytotoxicity against target leukemia cells and induced an alternative form of cell death than CAR T cells expressing only endogenous granzymes or exogenous granzymes A or B. However, increasing expression of granzyme F also had a detrimental impact on viability of the host T cells, decreasing their persistence in circulation in vivo. These results suggest a unique role for granzyme F as a marker of terminally differentiated CD8 T cells with increased cytotoxicity, but also increased self-directed cytotoxicity, suggesting a potential mechanism for the end of the terminal exhaustion pathway.

Introduction

Granzyme exocytosis by T cells and NK cells is a critical pathway to induce cancer cell death. In brief, these cytotoxic lymphocytes recognize target cells and induce cell death through the release of secretory granules. The substrates, functions, and mechanisms of cytotoxicity are not fully characterized for many granzymes [reviewed in (1)]. Apoptosis is the most common and regulated form of programmed cell death [reviewed in (2)] characterized by cellular blebbing and the release of cellular contents in an immunologically ‘quiet’ manner (3–5). However, alternative forms of cell death, such as necroptosis, induce potent immune responses through cellular membrane rupture and release of intracellular contents, which may be harnessed to enhance immunotherapies (6, 7). While the mechanisms of some granzymes have been previously described (1) the lytic capacity of most of these effector molecules has not yet been leveraged for the improvement of immunotherapies.

Granzyme B induces apoptotic cell death by cleaving and activating caspase-3 or the protein BID (8–12). Gzmb is one of the most highly expressed granzyme genes in cytotoxic T cells. Gzma induces pyroptosis through activating gasdermin proteins (13), which may modulate the tumor microenvironment (TME) by increasing the inflammatory milieu. Gzmf expression is low in T cells under most conditions but induces a necroptotic-like form of cell death, resulting in rupture of the cellular membrane (14). Since necroptosis is an immunogenic form of cell death, Gzmf expression may be a marker of a more tumoricidal T cell capable of eliciting epitope spreading. Despite protective mechanisms that shield T cells from granzymes, the impact of granzyme expression on the T cells themselves is still uncertain [reviewed in (15)]. A previous study showed that granzyme B induces apoptosis in the T cells producing it (16). Thus, modulating granzyme expression, such that TIL contribute to an immunogenic TME without causing self-elimination, could be a valuable contribution to immunotherapies. In particular, the low expression level of Gzmf under most conditions makes this a desirable target for improving T cell cytotoxicity.

T cell exhaustion is a differentiation pathway in response to chronic antigen exposure, in which T cells become hypofunctional and cannot proliferate, secrete cytokines or produce effector molecules [reviewed in (17)]. T cell exhaustion in the TME results in ineffective control of tumor growth and is a substantial hurdle for cancer immunotherapies. While Gzmb is known to be regulated differentially in specific T cell exhausted states (18), expression of the other granzymes have not been fully characterized in exhausted T cells. Given granzymes are highly cytotoxic molecules and critical effector molecules for T cell activity, their expression requires precise regulation. Therefore, defining the functional capacity of T cells based on effector molecule production provides additional framing of T cell exhaustion and differentiation state.

Chimeric antigen receptor (CAR)-T cell therapy takes advantage of the cytotoxic pathways of granzyme-producing T cells to target cancers. When the recombinant antigen-binding receptor made from the variable region of a single-chain antibody binds to its antigen, a signal is transduced through the downstream activation domains and granzymes are produced and secreted by the T cells (19, 20). CAR T cells targeting CD19 have shown remarkable clinical activity in patients with highly refractory B-lineage leukemias and lymphomas; however, less than 50% of these patients will achieve a long-term remission (21–26). The heterogeneity of antigen expression is likely a barrier to CAR T cell success and strategies which can induce epitope spreading may improve the effectiveness of CAR T cells for such patients (27).

In this study, we determined the expression pattern of the granzyme protease family in T cells within the TME and periphery and found that Gzmf expression is restricted to exhausted T cells. We further utilized CAR T cells in a murine model of acute lymphocytic leukemia (ALL) and modulated the terminal phase of T cell engagement with target cells by overexpression of specific granzymes. Gzma displayed evidence of killing both the tumor and the transduced T cells. In contrast, Gzmb overexpression had minimal impact on the CAR T cell response, likely because Gzmb is highly expressed in the absence of exogenous protein. As predicted, overexpression of Gzmf in CAR-T cells increased targeted cytotoxicity and improved the tumoricidal capacity of CAR T cells in vitro. However, relative to other granzyme overexpressing CAR T cells, Gzmf also increased the amount of cell death in the CAR T cells expressing it. In summary, Gzmf expression was actively restricted to exhausted T cells, where it acted to eliminate more target cells but also induced T cell suicide.

Materials and Methods

Mice:

C57BL/6J (B6) mice were purchased from The Jackson Laboratory and BALB/cAnNCr (BALB/c) mice were purchased from Charles River Laboratories. The Institutional Animal Care and Use Committee at the University of Colorado School of Medicine reviewed and approved all animal protocols.

Cell Lines:

E2aPbx (E2a) and E2a-GFP are murine pre-B ALL cell generated by the transduction of the human E2A-PBX (TCF3-PBX1) transgene into the bone marrow of CD3ε^−/− B6 mice (28). CT26 is a murine colon carcinoma cell line from BALB/c mice purchased from ATCC (29). A223 cells were derived from a head and neck squamous cell carcinoma (HNSCC) harboring the Kras^G12D mutation and Smad4 deletion in B6 and CD8^−/− mice (30). Cells were tested for mycoplasma contamination by PCR at the Barbara Davis Center Bioresource Core prior to freezing aliquots, which were used for no more than one month. Cells were cultured as described (31).

Tumor challenge, TIL and spleen processing:

1×10⁵ CT26 tumor cells in 100 uL PBS (Life Technologies) were injected subcutaneously in both hind flanks and TIL were harvested on day 14 as described (31). Splenocytes were mechanically dissociated, and red blood cells were lysed as described (32). A223 tumor samples and matching spleens were similarly prepared as described (33, 34).

Publicly available data analysis:

Microarray data were extracted from Gene Expression Omnibus (GEO) https://www.ncbi.nlm.nih.gov/geo/. Using datasets that examined CD8 T cells from the TME with matching CD8 T cell expression data from outside the TME, data were probed for expression patterns of granzyme genes with the Geo2R software. We examined data from the CT26 (GSE79858) (35), the MC38 (GSE111492) (36), and the B16-F10 (GSE53388) (37) tumor models and the lymphocytic choriomeningitis virus (LCMV) model (GSE131847) (38). Quantile normalized log2 microarray data from CD8+ T effector (Teff) and resident memory T cells (Trm) from lung, liver, and small intestine were pulled from GSE65045 (39). IDs were aligned to their respective genes using Platform ID GPL11202, and Student’s t-tests were performed for all granzymes found in the dataset for differential expression analysis. Experimental details are in the original publications. LCMV single-cell data were concatenated into an AnnData (v.0.8.0) object in Python (v.3.8.16), and analyzed using the ScanPy package (v.1.9.1). Cells expressing less than 200 genes, genes present in less than 5 cells, cells expressing more than 7000 genes or cells with greater than 8% mitochondrial genes, were removed. We regressed effects of total counts per cell and the percentage of mitochondrial genes expressed. Each gene was scaled to unit variance, and values exceeding a standard deviation of 10 were clipped. CD8+ T cell RNAseq data under normal and hypoxic conditions were pulled from GSE155192 (40) and log2 normalized. Student’s t-tests were performed between normal and hypoxic conditions for all granzymes found in the dataset. Finally, single cell RNAseq data of TIL from CT26 tumors are publicly available at GEO accession number GSE212980 (31).

qPCR:

CD8 T cells were enriched by magnetic negative-selection using EasySep Mouse T cell Isolation Kit per manufacturer’s instructions (Stemcell Technologies #19851A). RNA was isolated from 1×10⁶ CD8+ T cells using RNeasy Mini-RNA Purification Kit per manufacturer’s instructions (Qiagen #74104). cDNA was synthesized using iScript cDNA Synthesis Kit (Bio-Rad #1708891) and cDNA was used as templates in real time quantitative PCR (qPCR) using SYBR green (ThermoFisher #4368706) per manufacturer’s instructions. Primers used are listed in Table 1. Values for a given gene were normalized to Ct values of control genes (Gapdh, Rn18s, and Ubc) that are less variable.

Table 1.

Primers designed for quantification of indicated genes by qPCR.

Target Gene	Forward Primer	Reverse Primer
Gzma	AGACCGTATATGGCTCTACT	CCCTCACGTGTATATTCATC
Gzmb	ATCAAGGATCAGCAGCCTGA	TGATGTCATTGGAGAATGTCT
Gzmc	GGAGATAATCGGAGGCAATGAG	TTCCCACCAACTTTCAGAAACT
Gzmd	TGCGAGAGGTTCAACTGGTC	CCCCGAGTCACCCTTGAAA
Gzmf	TGAGGTTTGTGAAAGATAATG	TCACTGGTGTTGTCCTTATC
Gzmm	ATGGAGGTCTGCTGGTCC	TCCAGGAGCTGTAGGGGG
Gzmk	TGTCCGTGCTCCTTAACACC	GCAGCAGTGCGAAGCTTTAT
Prf1	GTACAACTTTAATAGCGACACAGTA	AGTCAAGGTGGAGTGGAGGT
Fasl	GCAGAAGGAACTGGCAGAAC	TTAAATGGGCCACACTCCTC
Gapdh	CATCACTGCCACCCAGAAGACTG	ATGCCAGTGAGCTTCCCGTTCAG
Rn18s	GGCTACCACATCCAAGGA	GCTGGCACCAGACTTGC
Ubc	AGCCCAGTGTTACCACCAAG	ACCCAAGAACAAGCACAAGG

Granzyme B ELISA:

CAR T cells were cultured overnight with target cells at a 1:2 effector to target ratio or in the absence of target cells overnight. Supernatant from the cell culture was taken and an ELISA probing for release of granzyme B (abcam #ab238265) was performed following manufacturer’s instructions.

In vitro stimulation:

CD8 T cells from splenocytes were magnetically isolated and resuspended at 1×10⁶ cells/mL in 25 uL of CD3/CD28 beads (Gibco #11453D). Cells were cultured for 3 days in CD3/CD28 beads and 40 U/mL IL-2. Samples were also treated with either 10 ng/mL IFNγ (BioLegend #575302), 50% CT26 tumor-conditioned medium (supernatant from CT26 cells cultured for 2 days in complete medium (31)) or 2 ng/mL TGF-beta (BioLegend #763102) as indicated. On day 3, 1×10⁶ cells were removed for RNA isolation for qPCR, CD3/CD28 beads were removed from samples, and cells were resuspended at 1×10⁶ cells. On day 5, RNA was isolated from 1×10⁶ cells. Control RNA was prepared from unstimulated cells.

Flow cytometry and PrimeFlow:

For flow cytometry, 2×10⁶ cells were washed in PBS and stained with fixable amine-reactive viability dye (see Tables 2, 3, and 4) per manufacturer’s instructions. Amine-reactive dyes were also used for evaluation of cell membrane permeabilization. Samples were then washed with flow buffer [1x PBS, 2% FBS, and 0.1% sodium azide (41)] and stained for surface markers. Indicated antibodies were added to each sample and incubated for 30 minutes. In experiments staining for annexin V, samples were surface stained, washed with 1x Annexin V Staining Buffer, and stained with 5 uL of annexin V per 100 uL of sample for 15 minutes at room temperature. For PrimeFlow analyses, target probes were designed to interrogate expression of Gzma, Gzmb, Gzmf and a control probe to beta-actin. 5×10⁶ cells were prepared for all PrimeFlow samples per manufacturer’s instructions (ThermoFisher). Single color controls, UltraComp eBeads (ThermoFisher #01-2222-41) or cells used in experimentation stained with a single fluorophore, were used to compensate the flow cytometry studies. All flow cytometry data were analyzed using FlowJo software (Tree Star).

Table 2.

Flow cytometry panel for PrimeFlow experiments using 5-laser Cytek Aurora.

Antigen	Clone	Fluor	Channel	Company	Catalogue #
Fixable viability dye		“Blue”	UV6	Thermo Fisher	L23105
CD45	30-F11	Pac Blue	V3	Biolegend	103125
CD3	17A2	BV785	V15	Biolegend	100232
CD8b	YTS156.7.7	BV711	V13	Biolegend	126633
CD137	17B5	PE/Cy7	YG9	Thermo Fisher	50-112-3331
PD-1	29F.1A12	BV510	V7	Biolegend	135241
TIM3	B8.2C12	BV421	V1	Biolegend	134019
Slamf6	13G3	BV605	V10	BD	745250
GzmF probe		AF647	R2	Thermo Fisher	Custom*
GzmB probe		AF568	YG3	Thermo Fisher	Custom*
GzmA probe		AF488	B2	Thermo Fisher	Custom*
Actin-beta (Control probe)		AF750	R7	Thermo Fisher	Custom*

^*ThermoFisher designed PrimeFlow probes by determining unique regions of the granzyme genes to target.

Table 3.

Flow cytometry panel for in vitro CAR cytotoxicity experiments using 5-laser Cytek Aurora.

Antigen	Clone	Fluor	Channel	Company	Catalogue #
Fixable viability dye		“Blue”	UV6	Thermo Fisher	L23105
Flag Tag	M2	APC	R1	Abcam	AD0059F
CD8	53–6.7	PEcy7	YG9	Invitrogen	53–6.7
CD4	RM4–5	Pac blue	V3	Biolegend	100531
GFP		GFP	B2
Annexin V		APC/Fire-750	R7	Biolegend	640953
CD69	H1.2F3	BV650	V11	Biolegend	104541
Calreticulin	D3E6	PE	YG1	Cell Signaling Technology	19780S

Table 4.

Flow cytometry panel for CAR transduction efficiency using 5-laser Cytek Aurora.

Antigen	Clone	Fluor	Channel	Company	Catalogue #
Fixable viability dye		“Blue”	UV6	Thermo Fisher	L23105
Flag Tag	M2	APC	R1	Abcam	AD0059F
CD8	53–6.7	PEcy7	YG9	Invitrogen	53–6.7
CD4	RM4–5	Pac blue	V3	Biolegend	100531
CD69	H1.2F3	BV650	V11	Biolegend	104541

Generation of murine CD19 CAR-T cells:

Granzyme-expressing CAR constructs were generated in the pMSCV-Flag-mCD19-CD28-ZI2-tEGFR gamma-retroviral transfer plasmid by VectorBuilder. Granzyme genes were inserted in place of the tEGFR sequence, following the CAR sequence and P2a linker. Retroviral vectors encoding each CAR were produced by transfection of Platinum-E cells, which stably express gag, pol, and ecotropic env genes, using Lipofectamine 3000 (Life Technologies #L3000015). Viral supernatants were harvested 48 hours after transfection and frozen in aliquots. Splenic T cells from a B6 mouse were activated as above using 25 μL CD3/CD28 beads (Life Technologies #11456D) per 1×10⁶ T cells and beads were removed 3 days after transduction. T cells were cultured at a starting concentration of 1×10⁶ cells/mL with IL-2 (40 U/mL) and IL-7 (10 ng/ml) (Peprotech #217–17). 2 mL of activated T cells were transduced per well of a 6-well plate, in retronectin-coated 6-well plates (Takara) with 2–3 mL of viral supernatant and on days 2 and 3 post-T cell purification. A single mouse was used to produce each different type of CAR T cell described for each biological replicate shown. Transduction efficiency was evaluated by flow cytometry based on expression of the flag-tagged CAR molecule, on day 5. cell numbers were adjusted based on transduction efficiency for downstream use, such that the same number of transduced cells were used for each CAR T cell group.

CAR-T in vivo studies:

C57BL/6 mice were injected with 1×10⁶ E2a cells. Up to three days later, mice were lymphodepleted with 500 cGy using ¹³⁷Cs source. Two days following irradiation, mice were treated with 1×10⁵ anti-CD19-CAR T cells by retro-orbital injection. The number of CAR+ T cells was equal across cohorts, adjusting for transduction efficiency. CAR T cells were detected by flow cytometry on blood draws or bone marrow (Table 3). Flow cytometric analyses were performed every 7–10 days until completion of experiments. Bone marrow was harvested from femurs, processed into single cell suspensions, stained with antibodies, and analyzed by flow cytometry.

Data Sharing Statement:

For original data, please contact Jill.Slansky@CUAnschutz.edu.

Results

Granzyme F expression is increased in CD8 TIL relative to T cells outside the TME

Studies of antigen-specific CT26 CD8 TIL show that Gzmf expression is the most significantly upregulated gene relative to splenic antigen-specific T cells (35). To determine if the differential expression of Gzmf is unique compared to other granzymes in T cells, we further analyzed these data (Figure 1a) (35). While other granzymes, many of which were in the Gzmb locus, had increased expression in the TME, Gzmf was the most highly upregulated. To determine if the increased expression is unique to CT26, we examined publicly available microarray data of MC38 colon carcinoma CD8 TIL (Figure 1b) (36) and the B16-F10 melanoma CD8 TIL (Figure 1c) (37). These results demonstrated that Gzmf is highly upregulated in CD8 TIL in multiple models. We confirmed these findings by qPCR in the CT26 tumor (Figure 1d) and the head and neck squamous cell carcinoma A223 (Figure 1e) (30, 34, 42) relative to matched splenic CD8 T cells. Gzmf was highly upregulated in CD8 TIL, where it had one of the highest expressions when normalized to housekeeping gene expression, relative to CD8 T cells in the spleen, where it was lowly expressed. To determine if the effect found on Gzmf expression was tumor specific or associated to the memory and tissue residency status of CD8 T cells, we examined microarray data from GSE65045 in which tissue resident memory CD8 T cells (Trm) from the lung, liver and small intestine (SI) were compared to effector CD8 T cells (Teff) from the blood. We analyzed this microarray data for expression of Gzma, Gzmb, Gzmc, Gzmf and Gzmk (Figure 1f). While there was a significant difference between Trm and Teff in the lung for Gzma, Gzmb and Gzmk, we found no significant difference in these Trm populations for Gzmf or Gzmc expression. Differences in Gzmf expression between Trm and Teff cells was minimal in comparison to the change in expression observed in the TME, suggesting that the TME specifically contributes to Gzmf expression.

Figure 1: Expression of Gzmf in CD8 T cells is higher in the TME than other tissues in multiple tumor models.

Microarray data from GEO were examined for expression of granzyme encoding genes in CD8 TIL relative to CD8 T cells from either matched spleens or lymph nodes. T cells were analyzed from a) the CT26 tumor (n=3) (35), b) the MC38 tumor (n=2) (36), and c) the B16 tumor (n=3) (37). Microarray data were analyzed using the Geo2r software, and only significantly altered genes were shown as determined by a corrected p value <0.05. Further explanation of the data is in the original publications. d-e) qPCR was performed for the indicated genes in CD8 TIL and CD8 T cells from the spleen. Relative expression of these genes (left), and transcripts per 100,000 housekeeping transcripts in the spleen (middle) and tumor (right), are shown in d) the CT26 tumor model (n=5) and e) the A223 tumor model (n=5). f) Microarray data from GEO (GSE65045) (39) was used to compare lung, liver or small intestine (SI) Trm to Teff cells for their expression of granzyme genes and a t test used to determine statistical significance (p ≤ 0.05*, ≤ 0.01**, ≤ 0.001***, ≤ 0.0001****). For qPCR data a one-sample t test was used to compare the fold change for each gene to no change in expression and for genes with a p value <0.05, a single star indicates significance.

The influence of T cell activation on granzyme expression

To elucidate pathways influencing granzyme expression, we stimulated CD8 T cells in vitro with anti-CD3 and anti-CD28 (CD3/CD28) and determined granzyme expression by qPCR. Relative to unstimulated CD8 T cells, CD3/CD28 stimulation significantly increased the expression of Gzma by day 5 and Gzmb by day 3, which stayed high until day 5 (Figure 2a). We evaluated Gzmf, Gzmk, and Gzmm similarly (Figure 2a). CD3/CD28 significantly altered neither Gzmf nor Gzmm; though Gzmk expression increased by day 5 like Gzma, located on the same locus. Expression of multiple granzymes, perforin and FasL, was compared 5 days post-CD3/CD28 stimulation and revealed that Gzmb was the most significantly upregulated gene of all examined (Figure 2b).

Figure 2: Expression of granzyme genes is differentially regulated after TCR stimulation.

a) T cells from the spleens of naïve mice were stimulated with anti-CD3 and anti-CD28 beads (CD3/CD28) and gene expression was measured on days 3 and 5 by qPCR for the indicated genes relative to unstimulated controls. Statistical significance was determined by a paired t test. b) The fold change of the indicated gene expression 5 days post stimulation, relative to matched unstimulated CD8 splenocytes, is shown. A one-sample t test was used to compare fold-change values to no change in expression. The expression of c) Gzma, d) Gzmb, e) Gzmf, f) Gzmk, g) Gzmm, were determined under different in vitro conditions, stimulated with either CD3/CD28 alone, Interferon-gamma (IFNγ), TGF-beta, or tumor-conditioned medium (TCM). All samples received CD3/CD28 for the first 3 days and 40 U/mL IL-2 while in culture. Statistical significance between the fold change in gene expression between the different treatment groups was determined by ANOVA. n=6 for all samples except for TGF-beta-treated cells where n=4. h) scRNAseq data from circulating CD8 T cells in LCMV infected mice were analyzed for expression of the indicated genes from 0 to 90 days post infection (38). Only granzymes that had detectable expression are shown (Gzmf was undetectable). Gene expression was normalized to counts per 10,000 counts. i) RNAseq data from GSE155192 (40), in which CD8 T cells were stimulated in normoxic or hypoxic culture conditions with acute antigen exposure (left) or chronic antigen exposure (right), was used to investigate granzyme expression under these conditions. p ≤ 0.05*, ≤ 0.01**, ≤ 0.001***, ≤ 0.0001****.

We also tested 3 other in vitro culture conditions to determine if alternative T cell stimulation in combination with CD3/CD28 induced Gzmf expression (Figures 2c–g). We tested interferon-gamma (IFNγ) because the expression of multiple interferon-inducible genes correlated with Gzmf expression (Table 5). TGF-beta was tested because TIL transcriptomes reflect strong TGF-beta signaling (35) and the Gzmf promoter has seventeen G/C-rich sites predicted to bind the Sp1/KLF-like family of transcription factors, some of which are regulated by TGF-beta (43). Finally, we tested tumor-conditioned medium (TCM) because soluble factors secreted into the medium of growing tumors may be responsible for the expression of different genes. None of these conditions significantly impacted Gzmf expression relative to unstimulated cells or to CD3/CD28 stimulation alone. These analyses, in combination with Figure 1, showed that Gzmf expression was low outside the TME and increased in TIL due to unknown factors or differentiation pathways specific to the TME. Only Gzmk responded to any of these stimulation conditions, and had reduced expression in TCM relative to all other stimulation groups, evidencing a unique expression pattern that is inhibited in the presence of soluble tumor factors.

Table 5.

Genes correlated with Gzmf expression from scRNAseq of CT26 CD8 TIL (31).

Gene	R value‡	BH§
Irf8	0.185203	8.85E-53
Ifitm3	0.189345	3.14E-55
Prf1	0.193603	8.29E-58
Gm44040	0.204955	4.84E-65
Lgals3	0.225798	1.59E-79
Serpinb9b	0.244256	1.22E-93
Spp1	0.267869	0
Gzmb	0.275442	0
Ifitm1	0.295241	0
Ifitm2	0.333623	0
Gzme	0.348327	0
Gzmd	0.377166	0
Gzmc	0.487932	0
Gzmf	1	0

^‡Pearson’s R-values were determined for the relationship of Gzmf with every other significantly expressed gene. The genes with the highest and significant R-value were displayed.

^§Benjamini-Hochberg (BH) corrected P-values were calculated to address the false discovery rate.

To determine Gzm-expression changes in conditions of viral exhaustion, we examined publicly available single-cell RNA sequencing (scRNAseq) data of an LCMV infection model (Figure 2h) (38). Inhibitory receptor Pdcd1 and Havcr2 expression was detected as expected, but at lower normalized counts than Gzma, Gzmb and Gzmk. Similar to CD3/CD28 treatment alone (Figure 2a), Gzmb was rapidly upregulated by day 3, whereas Gzma expression continued increasing until day 7. Gzmf, Gzmc, Gzmd, Gzme, Gzmg, and Gzmn were not detected under the conditions used, highlighting the differential regulation of granzyme expression.

To determine if other features of the TME were sufficient to induce Gzmf expression, we analyzed data from GSE155192 (40), in which CD8 T cells were stimulated with chronic or acute antigen exposures under hypoxic or normoxic conditions in vitro (Figure 2i). We found that hypoxia had no significant impact on Gzmf expression with acute antigen exposure. However, under chronic antigen exposure Gzmf expression was increased modestly relative to the periphery. Gzmk was the only granzyme examined that had decreased expression under hypoxic conditions, in both chronic and acute antigen stimulation.

Gzmf expression increased in PD1+/TIM3+ CD8 TIL

We used PrimeFlow, a flow cytometry-based technique for detecting RNA transcripts, to simultaneously interrogate the Gzmf transcript and cell surface proteins from CT26 CD8 TIL. These analyses revealed an increased frequency of Gzmf expressing TIL relative to splenic CD8 T cells (Figure 3a). Gzmf expressing cells were predominantly 4–1BB+, a marker for antigen experience (Figure 3b) (44). They also expressed the inhibitory receptors PD1+/TIM3+, indicative of T cell exhaustion (45), at a significantly greater frequency than Gzmf-negative cells (Figure 3c). In addition, only a subpopulation of PD1+/TIM3+ cells expressed Gzmf (Figure 3d), indicating that these cells are a subpopulation of exhausted TIL. Representative flow plots of inhibitory receptors show that Gzmf-positive cells expressed the exhaustion markers PD1 and Tim3 (Figure 3e).

Figure 3: Gzmf is expressed more frequently in exhausted PD1+/TIM3+ CD8 TIL than CD8 splenocytes or PD1- TIL in the CT26 tumor model.

a) Using PrimeFlow analyses, the frequency of total CD8 T cells that express Gzmf from the CT26 TME or from the spleen was determined (n=4). Gzmf+ and Gzmf- cells were examined for expression of b) the antigen-experienced marker 4–1BB (n=6) and c) co-expression of PD1 and TIM3 (n=6). d) PrimeFlow analysis shows a subset of PD1 and TIM3 double-positive cells expressed Gzmf (n=6). e) Representative flow plots show PD1 and TIM3 expression of Gzmf+ CD8 TIL (green) and of Gzmf⁻ CD8 TIL (gray). f) UMAPs of scRNAseq data with the general immunophenotypes of groups of clusters (GSE212980) (31) and with g) Gzma, h) Gzmb and i) Gzmf expression indicated within each cell in blue are shown. A t test was used to determine statistical significance between splenocyte and TIL expression of Gzmf, and between Gzmf+ and Gzmf- TIL expression of 4–1BB. A repeated measures ANOVA was used to determine statistical significance in the expression of PD1 and TIM3 in Gzmf+ cells (p ≤ 0.05*, ≤ 0.01**, ≤ 0.001***, ≤ 0.0001****).

We re-analyzed scRNAseq data from CT26 antigen-specific CD8 TIL (31) for granzyme gene expression. Overarching transcriptional profiles of these data were labeled on a uniform manifold approximation and projections (UMAP) (Figure 3f). We determined that Gzma was restricted to “progenitor-like” clusters (Figure 3g), Gzmb was highly expressed in all but one antigen-specific subpopulation (Figure 3h), and Gzmf was expressed in more PD1+/TIM3+ clusters (Figure 3i). Tetramer-negative TIL did not express these granzymes. We examined the correlation between Gzmf expression and the other genes in the dataset. Genes that most significantly correlated with Gzmf expression were other granzymes located on the Gzmb locus, Prf1, some interferon-inducible genes, and Serpinb9b (Table 5). Some serpins protect T cells from granzyme cytotoxicity (15, 46).

GZMF-overexpressing CAR T cells have increased cytotoxic activity and induce a different cell death pathway

To understand the function of granzyme F, we expressed it with a CD19-CAR in B6 mice with the acute lymphocytic leukemia (ALL), E2a (28). Gzmf and the better characterized Gzma and Gzmb were inserted downstream of the CAR sequence and a P2a linker, then transduced into T cells to produce F-CAR, A-CAR, and B-CAR, respectively. As a positive control, we transduced T cells with CAR and no exogenous granzyme (Plain-CAR) (47–51), and as a negative control, T cells were treated similarly but were not genetically engineered (Untransduced). Endogenous granzyme expression was present in these T cells. Granzyme gene expression was measured in these CAR products by qPCR and fold change was determined relative to unstimulated T cells (Figure 4a). Each CAR construct overexpressed the specific granzyme gene relative to the Untransduced T cells, resulting in increased Gzma and Gzmb expression, as expected. Gzmf in F-CAR T cells was highly increased because endogenous levels were low. We ensured that granzyme protein expression increased in the B-CAR T cells relative to the Plain-CAR T cells by co-culturing with the E2a cell line expressing green fluorescent protein (E2a-GFP) or in the absence of target cells. When we examined supernatant by ELISA (Figure 4b), higher concentrations of granzyme B were found from the CAR-B T cells incubated with target cells.

Figure 4: F-CARs have improved tumoricidal activity and induce an alternative form of cell death in target cancer cells.

a) Expression of Gzma, Gzmb and Gzmf was determined in A-CAR, B-CAR and F-CAR samples respectively, and in untransduced controls, by qPCR and fold change (log₂) relative to unstimulated cells is shown, where a change of 1 indicates a doubling of relative expression. E2a-GFP cells were co-cultured with CAR T cells at a 1:2 effector to target (E:T) ratio overnight. b) Supernatant was taken after this coculture, or from CAR T cells in the absence of target cells (E:T=1:0), and an ELISA probing for released granzyme B was performed. c) The ratio shown of E2a-GFP cells to CAR T cells after the co-culture was determined by flow cytometry. d) The frequency of viable E2a-GFP is shown. The percentage of dying E2a-GFP cells, which were e) positive for annexin V and negative for the amine-reactive permeable membrane dye, or f) positive for the permeable membrane dye is shown. A repeated measures ANOVA was used to determine statistical significance between experimental groups. n=10 for all experimental groups and was repeated 5 times, p ≤ 0.05*, ≤ 0.01**, ≤ 0.001***, ≤ 0.0001****.

Using these constructs, CAR T cells were co-cultured overnight with E2a-GFP to determine the activity of the different granzymes. The ratio of E2a-GFP cells to CAR T cells was determined before and after co-culture by flow cytometry. The greatest decrease in E2a-GFP cells was in the F-CAR samples (Figure 4c). In addition, the granzyme-CAR samples had significantly reduced frequency of viable E2a-GFP cells after co-culture (Figure 4d). While F-CAR cells were significantly more cytotoxic than Plain-CARs, the differences in cytotoxicity observed between the granzyme overexpressing groups was not statistically significant.

F-CARs had reduced induction of apoptosis in target E2a-GFP cells, as determined by annexin V staining (Figure 4e), and increased induction of membrane permeabilization, as determined by staining with an amine reactive dye, relative to the other groups (Figure 4f). The unique cytotoxic activity of Gzmf, inducing membrane permeabilization rather than externalization of phosphotidyl serine, could have downstream effects in vivo. Cell membrane rupture increases release of damage-associated molecular patterns (DAMPs) and subsequent invigoration of the immune response (52, 53). These results suggest that Gzmf overexpression induces a unique form of cytotoxicity, and that overexpression alone may alter cell death induced in target cells.

Viability and CD69 expression in F-CARs is decreased

To investigate cell-intrinsic effects of granzymes on T cells, we examined granzyme overexpressing CAR T cells after co-culture with E2a-GFP cells overnight. F-CARs had the lowest viability and significantly more cell death than Plain-CARs and A-CARs, indicating that GZMF expression is detrimental to the T cell producing it (Figure 5a). We further investigated cell death pathways in cultured T cells by quantitating permeable cell membranes and annexin V (Figure 5b). All CAR T cells had permeabilized membranes, but this bias was significantly increased in granzyme-overexpressing CAR T cells. The membrane permeabilization bias indicates either the type of cell death occurring or the rate of cell death, given apoptotic cells eventually lose membrane integrity in vitro(54). The decrease in frequency of annexin V single-positive cells in granzyme overexpressing groups reflects a physiological change in T cell viability.

Figure 5: F-CARs have reduced viability and CD69 expression in vitro.

a) After an overnight co-culture of CAR T cells with E2a-GFP cells (E:T = 1:2), the frequency of CAR T cell death was normalized to CAR-negative T cell death in the same culture by taking the percentage of non-viable CAR+ T cells and subtracting the percentage of non-viable CAR-negative T cells to correct for cell death that occurred in samples independent of granzyme overexpression. b) The frequency of dying cells that had a permeable membrane (orange bars) or were only annexin V positive (blue bars) is shown. The gMFI of CD69 expression as determined by flow cytometry after E2a-GFP co-culture is shown for c) CAR+ T cells and d) the difference between CAR expressing and untransduced T cells. A repeated measures ANOVA was used to determine statistical significance. n=10 for all experimental groups and was repeated 5 times, p ≤ 0.05*, ≤ 0.01**, ≤ 0.001***, ≤ 0.0001****.

Given the improved cytotoxicity against cancer cells but reduced viability of granzyme overexpressing CAR T cells, we determined if granzyme overexpression impacted CAR T cell activation. We examined CD69 expression after T cells were co-cultured with E2a-GFP cells. Plain-CAR T cells had significantly higher geometric mean fluorescent intensity (gMFI) than Gzm-CAR T cells (Figure 5c). Similar results were obtained after normalizing CD69 gMFI expression to CAR-negative T cells in the same culture (corrects for CD69 expression from non-antigen-specific T cells) (Figure 5d). Notably, all samples had increased CD69 expression on CAR T cells relative to other T cells in the same culture (Figure 5d), consistent with changes in CD69 expression resulting from antigen-specific interactions. These results reveal that granzyme overexpression, and particularly GZMF overexpression, reduces the viability and activation of CARs in vitro.

Plain- and B-CARs control cancer growth but A- and F-CARs do not

CAR T cells were transferred into E2A leukemia-bearing mice to determine the effect of granzyme expression on controlling leukemia and CAR persistence in vivo. We monitored CAR persistence in the bone marrow, where these leukemic cells concentrate, 55 days post-adoptive CAR T cell transfer as a fraction of total T cells (Figure 6a), CD4 T cells (Figure 6b), and CD8 T cells (Figure 6c). We found no difference in CAR T cell persistence between CARs as a percentage of bulk T cells or as a percentage of CD4 T cells, but there were more Plain-CARs as a percentage of CD8 T cells relative to granzyme-overexpressing groups. The frequency of CD8 CAR T cells was less than CD4 CAR T cells in granzyme-overexpressing samples. A-CAR cells seemed to have the worst persistence in the bone marrow at the timepoint examined. While this was not significantly different from B-CAR and F-CAR results, it is worth noting that Gzma is located on a different locus and has pro-inflammatory activity (55, 56) which may contribute to this decrease in bone marrow persistence.

Figure 6: Plain-CARs and B-CARs control antigen-expressing tumor cell growth better than F-CARs and A-CARs.

The frequency of CARs in the bone marrow 55 days after CAR transfer as a percentage of a) total T cells, b) CD4 T cells, and c) CD8 T cells. d) The frequency of CAR T cells in circulation 5 days post CAR transfer is shown as a percentage of T cells. ANOVA was used to determine statistical significance. e) CD19+ cell-free survival, defined as less than 1% of viable lymphocytes expressing CD19 and B220, after E2a cancer cell challenge is shown in a Kaplan-Meyer survival plot and evaluated for statistical significance by a log rank test; a Bonferonni-corrected alpha value was used to correct for multiple comparison bias, n=10 mice per group.

Our data suggest that CD4 CARs were unaffected by granzyme overexpression, indicative of a biological difference between CD4 and CD8 T cells in their susceptibility to granzyme self-directed cytotoxicity. We determined the frequency of CAR T cells as a percentage of T cells in blood 5 days after CAR T cell adoptive transfer (Figure 6d) and determined that Plain-CARs persist better than Gzm-CARs in the blood at this timepoint.

To determine how the granzyme overexpressing CAR T cells would influence target cell growth, we measured the fraction of viable CD19+B220+ lymphocytes in the blood over time (Figure 6e). Plain-CARs and B-CARs controlled CD19 expansion the best. Untransduced T cells, A-CARs, and F-CARs did not control CD19+ cell expansion. These results suggest that different granzymes have different roles in the anticancer response: granzyme B is well tolerated, while expression of granzymes A and F may be toxic to the T cells which produce them. These conclusions are consistent with F-CAR impairing its persistence and control of target cell expansion over time, even if in vitro granzyme overexpression increases the cytotoxic activity of CARs against target cells. Given that Gzmf is expressed by exhausted cells (Figure 3c), its capacity to eliminate the cell producing it may not be an artifact of its increased cytotoxic activity but instead, a mechanism by which exhausted T cells are actively self-eliminating.

Discussion

Granzyme expression is highly regulated: specifically, Gzmf is one of the most upregulated genes in CD8 TIL(35–37). Here we demonstrate that T cell activation influences expression of granzymes differently, and granzymes have distinct expression patterns. Our results illustrate a dynamic system in which T cells alter their expression of granzymes in response to specific stimuli. We show that Gzma, Gzmb and Gzmk expression is upregulated in response to TCR stimulation, though to different extents. Expression of other granzymes tested did not change in response to TCR stimulation alone but likely required other stimuli. The expression patterns position granzymes as useful barometers of T cell functional states given their expression is already a critical metric in distinguishing functional and ineffectual T cell responses.

Other studies found that T cells taken from progressing tumors expressed significantly more Gzmf than regressing tumors, suggesting that Gzmf may be a marker of suboptimal T cell responses (42). We determined that Gzmf is expressed specifically in exhausted CD8 TIL. The increase in expression of a cytotoxic mediator during exhaustion provides insight into T cell dysfunction and positions Gzmf as a marker of hypofunctional T cells. Furthermore, Gzmf expression in exhausted CD8 T cells may implicate Gzmf as having an active role in T cell dysfunction.

Historically, the study of granzymes has focused on Gzma or Gzmb. However, Gzmk, another often overlooked granzyme, has been implicated by multiple groups as a marker of unique differentiation pathways in TIL (57, 58). We found that treatment of T cells with tumor-conditioned media reduced Gzmk expression in response to TCR activation, suggesting that further investigation into this molecule’s expression are necessary to completely understand T cell differentiation in the TME. The granzyme family members as markers of T cell function, given their regulated expression patterns in the TME, may become a critical tool in predicting antitumor responses.

Through the overexpression of granzymes in CAR T cells, we determined that Gzmf increases the cytotoxic activity of T cells against target tumor cells; but we also found that Gzmf overexpression induced increased cell death in the CAR T cells producing it. Furthermore, Gzmf induced an alternative cell death pathway that increases membrane permeabilization in dying cells, consistent with another study using recombinant granzyme F (14). A-CARs and B-CARs induced significantly more apoptotic cell death than F-CARs; A-CARs had less self-directed cytotoxicity than F-CARs. In vivo, we established that overexpression of Gzma and Gzmf impaired CAR T cell control of leukemia to the same level as untreated controls, but Gzmb, the most ubiquitously expressed granzyme, had no significant impact on controlling leukemia relative to Plain-CARs.

Mechanisms exist to protect T cells from their own granzyme-mediated cytotoxicity, such as serpins (15, 59–61). However, our data shows that granzyme A and F are not as well protected against as granzyme B. Better protection against granzyme B expression in T cells is likely evolutionarily advantageous, as granzyme B is highly expressed. The lack of protection against GZMF, usually expressed at low levels, may benefit an organism if T cells need to self-eliminate.

In summary, these findings suggest that Gzmf-expressing effector T cells are a short-lived population with enhanced cytotoxic activity but which are susceptible to the same effector molecules they use to eliminate target cells. Self-targeted granzyme-mediated destruction of exhausted T cells could expand the exhaustion paradigm to define the final steps of terminal differentiation by providing a mechanism by which exhausted T cells are terminated.

Key Points:

Granzyme F expression is highly upregulated in exhausted CD8 TIL.

Granzyme F is cytotoxic to target cells and to the T cells producing it.

List of Acronyms

TIL

Tumor infiltrating lymphocytes

TME

Tumor microenvironment

CAR-T

Chimeric Antigen Receptor T Cells

A-CAR

Granzyme A overexpressing CAR T cells

B-CAR

Granzyme B overexpressing CAR T cells

F-CAR

Granzyme F overexpressing CAR T cells

Teff

Effector T cell

Trm

Memory T cell

TCM

Tumor conditioned media

ALL

Acute lymphocytic leukemia

HNSCC

Head and neck squamous cell carcinoma

LCMV

lymphocytic choriomeningitis virus

scRNAseq

Single cell RNA sequencing

UMAP

Uniform manifold approximation and projections