CDK4/6 Inhibition Promotes Antitumor Immunity through the Induction of T-cell Memory

Aberrant cell cycling via disruption of the cyclin-dependent kinases 4 and 6 (CDK4/6)/retinoblastoma (RB) tumor suppressor pathway is observed in many cancers and has spurred the clinical development of inhibitors that specifically target these kinases (1, 2). CDK4/6 inhibitors (CDK4/6i) are now an approved treatment for hormone receptor–positive breast cancer and under investigation in clinical trials as a therapeutic for a wide array of cancer types (3). In addition to inducing tumor cell cytostasis, recent studies have highlighted a role for CDK4/6i in promoting antitumor immunity (4–8), an effect that has been attributed to both tumor-intrinsic mechanisms, such as increasing tumor expression of MHC class I (4), as well as direct effects on the immune system, such as enhancing T-cell activation (6). Indeed, CDK4/6i have shown remarkable synergy with blockade of the PD-1/PD-L1 axis in preclinical models (4, 6–8), prompting evaluation of this combination therapy in a number of clinical trials (e.g., NCT04075604, NCT02778685, NCT04118036).

T-cell dysfunction in the tumor microenvironment is a major barrier to effective antitumor immunity. The advent of modern technologies, such as single-cell transcriptomics, has provided unprecedented molecular insights into the heterogeneous nature of tumor-infiltrating T lymphocytes (TIL; refs. 9–11), revealing a diverse spectrum of phenotypic states with varying effector capacity. Notably, such technology has enabled the identification of T cells with stem- or memory-like properties as the primary responders to immune checkpoint blockade (ICB) and the key mediators of long-term antitumor immunity (11–17). These intratumoral memory-like subsets demonstrate superior capacity for self-renewal and longevity and can reconstitute and maintain a phenotypically diverse T-cell compartment (15). Indeed, strategies to preserve stem-like properties of autologous T cells and chimeric antigen receptor (CAR) T cells have resulted in superior antitumor efficacy after adoptive transfer (14, 18–20). Such cell states are governed by diverse transcriptional and epigenetic regulatory networks (21–23), and targeting these networks to direct immune cell fate is of great clinical interest.

Here, using integrated single-cell multi-omics, whole-genome CRISPR/Cas9 screening and functional-based in vitro and in vivo analyses, we identify CDK4/6 as a critical regulator of T-cell fate. Specifically, short-term pharmacologic inhibition of CDK4/6 led to direct T cell–intrinsic RB-dependent induction of immunologic memory, which correlated with long-term endogenous antitumor immune control after therapy removal. Further, preconditioning mouse or human CAR T cells with CDK4/6i enhanced their persistence and antitumor efficacy in murine models. Finally, we showed that the T-cell–intrinsic gene signature induced by CDK4/6 inhibition is associated with favorable responses to ICB in patients with melanoma and validated these findings through serial sample collection from a patient with melanoma over the course of treatment with CDK4/6i and subsequent ICB. Together, our results comprehensively and mechanistically detail the influence of pharmacologic CDK4/6 inhibition on the phenotype and function of T lymphocytes—the key mediators of antitumor immunity. These findings provide essential insight into the utility of these inhibitors as immunomodulatory agents and as adjuvants to a range of T cell–directed immunotherapies.

To comprehensively evaluate the effect of CDK4/6 inhibition on antitumor T-cell immunity, we used multimodal single-cell sequencing (CITE-seq; refs. 24, 25) to analyze the intratumoral T-cell population in MC38-OVA tumors from mice treated with the CDK4/6i palbociclib or a vehicle control (Fig. 1A). This approach allowed us to simultaneously measure surface protein expression and cellular transcripts while demultiplexing pooled treatment conditions (Fig. 1A). Dimensional analysis revealed 10 distinct T-cell clusters within MC38-OVA tumors (Fig. 1B), and antibody-derived tags (ADT) delineated discrete populations of CD4⁺ and CD8⁺ TILs (Supplementary Fig. S1A). Notably, tumors treated with CDK4/6i had a lower frequency of CD4⁺ regulatory T cells (Foxp3⁺ Tregs), consistent with previous reports (4, 26), and a higher frequency of CD8⁺ memory-like T cells, marked by high expression of Tcf7, Sell, Bcl2, and Cd7 (Fig. 1B–D; Supplementary Fig. S1B). Further analyses of gene expression dynamics of the CD8⁺ T-cell pool across a pseudotime trajectory revealed that CD8⁺ TILs from CDK4/6i-treated mice were skewed toward an earlier, and more stem-like, differentiation state (Fig. 1E and F; Supplementary Fig. S1C). Consistent with this, cells earlier on the pseudotime trajectory expressed higher levels of memory-associated genes (Tcf7, Sell, Il7r), and lower levels of effector-associated genes (Prf1 and Gzmb) and exhaustion markers (PD-1 and TIM3; Fig. 1G; Supplementary Fig. S1D). Similarly, inferred transcription factor (TF) activity analysis (Single-Cell Regulatory Network Inference and Clustering; SCENIC) revealed that memory-associated TFs, including TCF1 and FOXP1, were highly active in “earlier” T-cell clusters, whereas the effector-associated TFs, IRF8 and NFATc1, were most active in cells further along the pseudotime trajectory (Supplementary Fig. S1E–S1G). Published gene signatures of memory versus effector cells derived from lymphocytic choriomeningitis virus (LCMV) infection (27) also correlated with early and late pseudotime, respectively (Supplementary Fig. S1H). Accordingly, cells within the CD8⁺ memory T-cell cluster, which was increased after CDK4/6i treatment, were enriched for a memory versus effector gene signature (Supplementary Fig. S1I and S1J). We further characterized the CD8⁺ T-cell population by ex vivo bulk 3′ RNA sequencing (RNA-seq) and, in accordance with our single-cell analyses, found enrichment of memory versus effector signatures after CDK4/6i treatment (Supplementary Fig. S2A and S2B), with increased expression of several memory-associated genes (Sell, Il7r, and Tcf7; Supplementary Fig. S2C). To validate these findings with a second pharmacologic inhibitor of CDK4/6, we reanalyzed previously published single-cell RNA-seq data on TILs from lung tumors of mice treated with trilaciclib (6). Here, we also found an increase in memory-associated genes (Supplementary Fig. S2D), which was not discovered or reported in the previous study. In support of our transcriptional analyses, further investigation by flow cytometry demonstrated a significantly higher proportion of CD8⁺ TILs with a central memory (Tcm) phenotype (CD44⁺CD62L⁺; Fig. 1H), which was consistent across various tumor models (ref. 28; Supplementary Fig. S2E). CD8⁺ TILs from CDK4/6i-treated mice also had higher expression of memory markers TCF1 and CD127, and lower expression of exhaustion markers PD-1 and TIM3 (Supplementary Fig. S2F and S2G). Notably, this CDK4/6i-mediated induction of memory was also observed in the tumor-specific OVA-reactive CD8⁺ T-cell population (Supplementary Fig. S2H).

We next assessed whether these phenotypic and transcriptional changes were associated with stable epigenetic alterations using assay for transposase-accessible chromatin sequencing (ATAC-seq) to assess chromatin accessibility in CD8⁺ T cells from vehicle- or CDK4/6i-treated MC38-OVA tumors. Indeed, ex vivo bulk ATAC-seq revealed a global reduction in chromatic accessibility in the CD8⁺ T-cell population (Supplementary Fig. S2I and S2J), consistent with the more quiescent nature of memory T cells (29). Furthermore, gene set enrichment analysis (GSEA) of genes associated with regions of increased chromatin accessibility confirmed significant enrichment for T-cell memory signatures (Supplementary Fig. S2K and S2L), suggesting that CDK4/6i-driven phenotypic and transcriptional changes stem from stable epigenetic underpinnings.

To examine the long-term functional effect of CDK4/6 inhibition on antitumor T-cell immunity, we next treated MC38-OVA tumor–bearing mice with CDK4/6i over a short window (days 3–13, when the antitumor T-cell response is most active), followed by therapy withdrawal. At the time of therapy cessation, tumor volumes were equivalent in both vehicle- and CDK4/6i-treated mice (Fig. 1I and J). Strikingly, after therapy withdrawal, prior treatment with CDK4/6i resulted in complete tumor clearance in 21/22 mice compared with 9/22 for vehicle-treated mice (Fig. 1K). Notably, depletion of CD4⁺ and CD8⁺ T cells abrogated the efficacy of CDK4/6i, indicating that the response was T cell–mediated (Supplementary Fig. S2M). This was despite CDK4/6i-treated mice having an overall reduction in TILs before therapy withdrawal (Supplementary Fig. S2N). Together, these findings suggested that T cells present in CDK4/6i-treated tumors exhibited a greater capacity for ongoing antitumor activity than those from vehicle-treated tumors. Collectively, these data demonstrate that pharmacologic inhibition of CDK4/6 promotes T-cell memory and generates an intratumoral T-cell compartment capable of driving a potent and sustained antitumor response.

Lymphocyte differentiation is tightly regulated during cell proliferation (30–32), although it is unknown whether acute manipulation of cell-cycle machinery has a directional influence on cell fate. To determine whether the emergence of T-cell memory observed in CDK4/6i-treated tumors was due to the direct inhibition of CDK4/6 within these cells, we exposed activated primary mouse CD8⁺ T cells to CDK4/6i in vitro and monitored Tcm acquisition and division number. CDK4/6i exposure 0, 24, and 72 hours after activation decreased the mean division number and concurrently increased the proportion of Tcm cells in a dose-dependent manner, occurring in as little as 24 hours after drug exposure (Fig. 2A; Supplementary Fig. S3A and S3B). Notably, exposing cells to CDK4/6i 72 hours after activation not only prevented further differentiation of Tcm into effector cells (Tem), but redirected differentiation of Tem into Tcm (Supplementary Fig. S3C), with Tcm acquisition occurring independently of cell division (Supplementary Fig. S3D). This indicated that CDK4/6i was not simply restraining differentiation but was rather directly promoting memory formation, demonstrating a directional relationship between cell-cycle machinery and differentiation. Further analysis by 3′ RNA-seq revealed an increase in memory-associated transcripts after CDK4/6i, and GSEA analysis confirmed enrichment of memory-associated signatures coupled with a decrease in cell cycle–associated signatures, including E2F targets (Fig. 2B–E; Supplementary Fig. S3E). Paradoxically, CDK4/6 inhibition also induced effector-associated transcripts, including Gzma, Gzmc, and Pdcd1 (Fig. 2B and C), consistent with previous reports that CDK4/6i promotes T-cell activation (6). Indeed, CDK4/6i-treated OT-I T cells expressed higher levels of T-cell activation markers (CD25, CD69) and demonstrated enhanced cytotoxicity when cocultured with OVA-expressing tumor targets (Supplementary Fig. S3F–S3H). Consistent with this, upon restimulation with CD3/28, OT-I cells that had been pretreated with CDK4/6i maintained their capacity to differentiate into effector cells and again demonstrated significantly enhanced effector function (measured by cytokine secretion) compared with untreated cells (Supplementary Fig. S3I). Notably, continuous pharmacologic inhibition of CDK4/6 throughout restimulation did not diminish the capacity of cells to differentiate into functional effectors (Supplementary Fig. S3I). To determine whether acute in vitro treatment with CDK4/6i resulted in epigenetic changes that may account for these effects, we performed chromatin immunoprecipitation sequencing (ChIP-seq) for H3K27 acetylation. After acute (6 hours) treatment, we observed a dramatic reduction in H3K27 acetylation at cell cycle–related gene regions, including E2F TFs (Supplementary Fig. S4A and S4B). Accordingly, motif analysis identified significant enrichment in E2F TF motifs at regions associated with decreased acetylation and an increase in acetylation at motifs associated with drivers of T-cell memory, including forkhead-box–containing factors (ref. 33; Fig. 2F). To determine whether these changes were followed by long-term epigenetic commitment to a distinct cellular state, we performed ATAC-seq on in vitro–activated T cells after CDK4/6 inhibition. Concordant with our ex vivo ATAC-seq (Supplementary Fig. S2I and S2J), we observed a global reduction in chromatin accessibility, with an increase in accessibility at regions of genes involved in T-cell memory, including Tcf7, Ccr7, and Sell (Fig. 2G; Supplementary Fig. S4C).

To unravel the apparent dichotomy of both memory and effector features in response to CDK4/6 inhibition, we performed single-cell RNA-seq on in vitro–activated T cells after 24-hour exposure to CDK4/6i. This analysis revealed the existence of distinct T-cell subpopulations that were altered upon CDK4/6i treatment, with increases in clusters 2, 4, 5, and 8; decreases in cluster 0; and a global G₁ cell-cycle arrest (Fig. 2H and I; Supplementary Fig. S4D and S4E). We next generated memory and effector gene signatures and, using AUCell to compare enrichment scores, identified distinct memory-like and effector-like clusters within the cellular pool (Fig. 2J; Supplementary Fig. S4F and S4G). Interestingly, of those clusters that increased in frequency after CDK4/6i, clusters 2 and 5 were memory-like, and cluster 8 was effector-like (Fig. 2I), demonstrating that CDK4/6 inhibition promotes memory differentiation and enhances effector function in phenotypically distinct subsets in a heterogeneous T-cell pool.

To determine the molecular drivers of CDK4/6i-induced memory, we performed a genome-wide CRISPR/Cas9 screen for genes that regulate the memory marker CD62L (SELL) in response to CDK4/6 inhibition. Cas9-expressing Jurkat T cells (which upregulate CD62L upon CDK4/6 inhibition) were transduced with a genome-wide single-guide RNA (sgRNA) library, followed by treatment with CDK4/6i and FACS for cells that failed to upregulate CD62L (Fig. 3A). Sequencing of the sorted population revealed that sgRNAs targeting RB transcriptional corepressor 1 (RB1) were significantly enriched only in the treated condition (Fig. 3B; Supplementary Table S1), suggesting that RB is required for CDK4/6i-induced CD62L upregulation. To identify potential screen hits that are direct phosphorylation substrates of CDK4/6 kinase activity, we performed global phosphoproteomics on both Jurkat cells and activated primary mouse CD8⁺ T cells upon acute (2 hour) exposure to CDK4/6i (Fig. 3C). We detected a significant loss of specific phosphorylation sites on multiple peptides in both Jurkat and mouse T cells, including the canonical CDK4/6 targets, RB and RBL1/2 (Fig. 3D–F; Supplementary Fig. S5A–S5D). Notably, RB was an overlapping hit across the phosphoproteomic analysis in both Jurkat and mouse T cells and was the top hit in our genome-wide CRISPR/Cas9 screen (Fig. 3B and G; Supplementary Fig. S5C and S5D), indicating that CDK4/6i-mediated memory formation is mediated through RB. Accordingly, targeted deletion of RB1 using CRISPR/Cas9 in Jurkat cells abrogated the induction of CD62L and dampened the antiproliferative effects of CDK4/6i (Fig. 3H; Supplementary Fig. S6A and S6B). Further analyses by 3′ RNA-seq showed that RB loss likewise abrogated the transcriptional response to CDK4/6i, with global CDK4/6i-induced transcriptional changes, including SELL, largely attenuated in the absence of RB (Fig. 3I; Supplementary Fig. S6C). Consistent with this, CRISPR/Cas9-targeted deletion of Rb1 also abrogated Tcm formation in activated primary mouse CD8⁺ T cells (Fig. 3J; Supplementary Fig. S6D). To determine whether CDK4/6i-induced memory formation occurs through an RB-mediated block in G₁ of the cell cycle, we treated cells with the G₁ blocking agent thymidine, which induces G₁ arrest independently of RB, by preventing DNA synthesis and entry into S phase. Indeed, both CDK4/6 inhibition and thymidine induced G₁ arrest in mouse CD8⁺ T cells, whereas only CDK4/6 inhibition reduced phosphorylation of RB (Supplementary Fig. S6E and S6F). Thymidine treatment of these cells phenocopied CDK4/6 inhibition, transcriptionally and phenotypically, and largely rescued Tcm formation in Rb1 knockout (KO) cells (Fig. 3J; Supplementary Fig. S6G–S6I), suggesting that RB-mediated G₁ arrest itself contributes to CDK4/6i-induced memory differentiation. Further analysis of Rb1 KO cells by 3′ RNA-seq revealed downregulation of memory-associated genes and an increase in effector signatures (Fig. 3K and L), which failed to be reversed upon CDK4/6 inhibition (Fig. 3M and N). Across a panel of cell-cycle inhibitors, CDK4/6i was the most robust inducer of a memory phenotype and was also the most potent mediator of G₁ arrest in these cells (Supplementary Fig. S6J and S6K). Notably, the other cell-cycle inhibitors did not induce a memory phenotype at concentrations at which cell-cycle inhibition was observed, indicating that cell-cycle arrest itself is required but not sufficient to drive memory formation in response to CDK4/6i. Taken together, these data suggest that CDK4/6i-mediated transcriptional reprogramming and memory formation occur through RB-dependent G₁ cell-cycle arrest and concurrent inhibition of CDK4/6 signaling events.

The capacity for long-term survival is a fundamental characteristic of T-cell memory (34). To determine whether CDK4/6i-treated cells demonstrated superior survival in vivo, untreated and CDK4/6i-pretreated in vitro–activated CD45.1 OT-I T cells were transferred into congenic CD45.2 mice, and their persistence and phenotype were evaluated over time by flow cytometry (Fig. 4A). A significantly higher frequency of pretreated OT-I T cells was detected in both the blood and the spleen over 30 days (Fig. 4B; Supplementary Fig. S7A and S7B). After 30 days, untreated and pretreated OT-I T cells that persisted in the blood and spleen were phenotypically similar (Supplementary Fig. S7C–S7F), with characteristics of memory precursors (MPEC; KLRG1-CD127⁺; Fig. 4C). Upon antigen reencounter, bona fide MPECs rapidly proliferate and differentiate into KLRG1⁺ short-lived effectors (SLEC; ref. 35). To evaluate the recall capacity of CDK4/6i-treated cells, untreated or pretreated in vitro–activated CD45.1 OT-I T cells were again transferred into congenic CD45.2 hosts. After 30 days, OT-I T cells were isolated and retransferred at equal numbers into new hosts that were simultaneously infected with Listeria-OVA (LM-OVA; Fig. 4D). Both untreated and pretreated OT-I T cells expanded and differentiated in response to LM-OVA infection, rapidly acquiring a functional, cytokine-producing SLEC phenotype (KLRG1⁺CD127⁻; Fig. 4E and F; Supplementary Fig. S8A–S8F). This confirmed that CDK4/6 inhibition drives both the phenotypic and functional acquisition of T-cell memory.

We next used single-cell RNA-seq to characterize the transcriptional profile of those T cells that persisted in the spleen 30 days after transfer (Fig. 4G). Consistent with our flow cytometry data, the transferred T cells from both untreated and pretreated conditions were phenotypically similar (Supplementary Fig. S9A–S9C) and were predominately resting in G₁ (Supplementary Fig. S9D). Using pseudo–bulk RNA-seq analysis and gene comparison with our pretransfer in vitro cultures, we generated a gene signature for persisting cells, which was marked by high expression of memory-associated genes (Sell, Foxp1, Tcf7, Cd7, Il7r, Klf2, and Ccl5) and low expression of effector-associated genes (Gzmd, Ifng, Prf1, Gzma, and Gzmb; Fig. 4H and I). Further analysis of our previous in vitro single-cell data (Fig. 2H–J) revealed an increase in the frequency of cells with this persistence signature, from 7.5% to 26.8%, after CDK4/6i treatment (Fig. 4J and K; Supplementary Fig. S9E). Together, these data indicate that CDK4/6 inhibition promotes differentiation of T cells toward a bona fide memory phenotype, characteristic of functional long-term persistence.

We next tested whether treatment with CDK4/6i would induce a memory phenotype in CAR T cells and overcome two major impediments to successful CAR T-cell therapy: T-cell exhaustion and lack of persistence (36). Using clinical protocols, we generated human CAR T cells targeting the Lewis Y (LeY) antigen (ref. 37; Fig. 5A) and found that exposure to CDK4/6i yielded an enrichment of CD4⁺ and CD8⁺ stem cell memory (Tscm) CAR T cells (Fig. 5B; Supplementary Fig. S10A and S10B). These stem-like CAR T cells are reported to have superior survival and self-renewal capacities (38, 39). 3′ RNA-seq of the CAR T cells revealed significant changes in gene expression after CDK4/6 inhibition, with GSEA analyses confirming enrichment of memory versus effector signatures, and an expected downregulation of E2F target genes (Fig. 5C; Supplementary Fig. S10C and S10D). To examine in vivo persistence, LeY CAR T cells, generated using peripheral blood mononuclear cells (PBMC) from two independent donors, were pretreated with CDK4/6i and transferred into NOD/SCID gamma (NSG) mice (Fig. 5D). Compared with untreated controls, we observed a significant increase in both the frequency and number of CD8⁺ and CD4⁺ pretreated CAR T cells in the blood in the weeks after transfer (Fig. 5E and F; Supplementary Fig. S10E and S10F). To assess the functional recall capacity of these CAR T cells, we challenged mice 30 days after CAR T-cell transfer with the LeY⁺ ovarian cancer cell line OVCAR-3 (Fig. 5D) and observed a significant increase in the expansion of pretreated CAR T cells in the blood of mice compared with the untreated CAR group (Fig. 5G). This translated to a significant and striking enhancement in tumor control in mice that had previously received pretreated CAR T cells (Fig. 5H and I; Supplementary Fig. S10G). Notably, there was a significant negative correlation between tumor size at day 30 and the number of CD3⁺ T cells in the blood before tumor challenge (Fig. 5J), suggesting that the enhanced persistence of CDK4/6i-treated CAR T cells was a critical factor driving tumor control. Indeed, mice that received pretreated CAR T cells had significantly higher numbers of tumor-infiltrating CD4⁺ and CD8⁺ CAR T cells 40 days after tumor inoculation (Fig. 5K). To validate these findings, we examined a second CAR T-cell model using mouse T cells transduced with a CAR targeting HER2. Consistent with our previous observations, CDK4/6 inhibition induced Tcm formation in these CAR T cells, leading to increased persistence in vivo (Supplementary Fig. S10H–S10N). We next examined the effects of CDK4/6i pretreatment on the acute efficacy of CAR T cells, by transferring untreated or pretreated anti-LeY CAR T cells into OVCAR-3 tumor–bearing mice. Pretreating CAR T cells with CDK4/6i significantly enhanced their antitumor activity (Fig. 5L and M). Consistent with this, several weeks after transfer, we found significantly higher numbers of CD4⁺ and CD8⁺ cells in the blood and CD8⁺ cells in the tumors of mice that received pretreated cells (Fig. 5N and O). Together, these data demonstrate that in vitro treatment with CDK4/6i is a robust strategy to enhance the phenotype and long-term efficacy of CAR T cells.

CDK4/6i have shown remarkable synergy with ICB in preclinical mouse models (4, 6–8). However, the mechanisms underpinning the synergy of these therapies are incompletely understood. Given that recent studies highlight stem-like T cells in the tumor microenvironment as the key mediators of ICB responses (12, 16, 17), we hypothesized that CDK4/6i-mediated induction of T-cell memory generates a more favorable T-cell pool for immune checkpoint targeting and will thus contribute to the efficacy of this combination. To explore this, we analyzed single-cell RNA-seq data on the tumor-infiltrating CD8⁺ T-cell compartment from patients with melanoma who either responded or did not respond to ICB (9). Using our ex vivo transcriptomic datasets on CD8⁺ TILs from mice treated with CDK4/6i, we generated a T cell–intrinsic CDK4/6i response signature and used AUCell to determine the enrichment of this signature in patient samples. Indeed, this CDK4/6i response signature was enriched in CD8⁺ TILs from patients who responded to ICB, accurately stratifying responders and nonresponders at both single-cell and patient levels (Fig. 6A–D; Supplementary Fig. S11A and S11B). Together, these findings demonstrated that CDK4/6i induces a T cell–intrinsic gene signature that is associated with favorable responses to ICB in patients.

Although CDK4/6i enhance ICB efficacy in preclinical mouse models, this combination has led to severe toxicities in clinical trials (40). As such, the future utility of CDK4/6i as immune adjuvants is uncertain. Our findings suggest that short-term exposure to CDK4/6i is sufficient to reprogram T-cell differentiation into a phenotype more amenable to immune checkpoint targeting. We hence questioned whether CDK4/6i would enhance ICB when used as a short-term priming tool, rather than co-administered on a continuous schedule, which is associated with high toxicity in patients. To test this, we examined the efficacy of PD-1 ICB in MC38 tumor–bearing mice that were primed with a CDK4/6i or vehicle control. To prime the mice, a CDK4/6i was administered for 2 days only, 1 week before the start of anti–PD-1 therapy. Tumors in both vehicle-primed and CDK4/6i-primed groups reached an equivalent size at the time PD-1 therapy commenced. At this size, vehicle-primed tumors were largely refractory to anti–PD-1, whereas CDK4/6i-primed tumors demonstrated significantly enhanced sensitivity, with anti–PD-1 clearing tumors in 7 of 10 of these mice (Fig. 6E). Consistent with this response, we observed a corresponding increase in the number of effector cells in the blood of CDK4/6i-primed mice after anti–PD-1 therapy (Supplementary Fig. S12A).

To investigate this further, we collected and analyzed serial blood samples from a melanoma patient over the course of treatment, during which the patient received CDK4/6i in combination with a targeted MEK inhibitor (MEKi), followed by treatment with combination PD-1 and CTLA4 ICB (Fig. 6F), and achieved a complete response. Notably, in primary mouse CD8⁺ T cells, MEK inhibition blocked ERK phosphorylation but did not alter phosphorylation of RB or cell proliferation (Supplementary Fig. S12B and S12C). Consistent with this, MEK inhibition did not alter T-cell memory formation, alone or in combination with CDK4/6i (Supplementary Fig. S12D), suggesting this inhibitor does not confound the CDK4/6i-mediated phenotype. Using CITE-seq technology to evaluate the series of patient samples, we observed a time-dependent increase in the frequency of the CD8⁺ T-cell memory population after CDK4/6i + MEKi therapy, characterized by high expression of SELL, IL7R, and TCF7 (Fig. 6G–I; Supplementary Fig. S12E–S12G), consistent with our preclinical analyses. The identity of these CD8⁺ clusters was further confirmed using SCENIC analysis (Supplementary Fig. S12H and S12I). Importantly, upon ICB, the patient achieved a clinical and immunologic response in which this CD8⁺ memory cluster diminished, suggesting that these cells mounted a response by differentiating into functional effectors (Fig. 6H and I), which is consistent with our mouse data and with the reported role of stem-like T-cell subsets after ICB (12). Indeed, pseudotime analyses revealed a differentiation trajectory from memory-like populations through to effector cells, with CDK4/6i restraining CD8⁺ T cells early in this trajectory and subsequent ICB therapy accelerating differentiation (Fig. 6J; Supplementary Fig. S12J–S12L). T-cell receptor (TCR) clonotype tracking across this time series also revealed that CDK4/6 inhibition increased the frequency of rare T-cell clones, predominantly existing within the memory population, followed by amplification of these clones upon ICB (Fig. 6K; Supplementary Fig. S12M). This suggested that ICB therapy liberates the use of a greater variety of TCRs in peripheral blood, consistent with published data (41). Collectively, these data suggest that CDK4/6i may be used as a priming tool to promote a more favorable T-cell phenotype before the administration of ICB.

In this study, we demonstrated that pharmacologic inhibition of CDK4/6 in cytotoxic T cells induces RB-mediated G₁ arrest and promotes the acquisition of a memory phenotype that significantly potentiates the long-term antitumor activity of these cells (Fig. 7). Although CDK4/6i are emerging as a promising new cancer therapeutic, currently the majority of clinical trials incorporating CDK4/6i are restricted to cancer types in which RB is functional, as CDK4/6i induction of cell-cycle arrest requires RB-mediated inactivation of E2F TFs (42, 43). Indeed, the clinical success of CDK4/6i is primarily attributed to the inhibition of tumor cell proliferation, coupled with immunomodulation that occurs downstream of tumor-intrinsic cytostasis (3, 4, 8). Unexpectedly, however, CDK4/6i recently demonstrated efficacy in triple-negative breast cancer (where RB loss is common) as part of a phase II trial where CDK4/6i was incorporated only as a tool to transiently arrest cycling of healthy cells to protect from chemotherapy-induced myelotoxicity (44). Such an example highlights the potential of CDK4/6i beyond cancers with targetable tumor-intrinsic pathways, but the mechanisms underpinning their clinical efficacy in this setting remain unclear. Here, our discovery that CDK4/6i directly influence cytotoxic T-cell differentiation sheds new light on their clinical activity and substantially broadens the prospective utility of these therapeutic agents, which may be used strategically as clinical tools to augment antitumor immunity in a wide array of cancer types.

Despite a well-established role for CDK4/6 in cellular proliferation, previous reports have indicated that CDK4/6i does not compromise expansion of tumor-specific CD8⁺ T cells (4, 6). In contrast, through comprehensive analysis of the direct T cell–intrinsic effects of CDK4/6i, we demonstrated clear CDK4/6i-mediated antiproliferative activity in these cells. Importantly, while rapid clonal expansion of antigen-specific lymphocytes is often an assumed requirement for optimal T-cell immunity, here we demonstrated that controlled regulation of the cell cycle may in fact promote a more robust and long-term antitumor immune response. Interestingly, in addition to memory formation, we observed enhanced T-cell effector function in response to CDK4/6i, consistent with previous reports that CDK4/6i mediates the derepression of NFAT (6). Using single-cell RNA-seq, we demonstrated that this enhancement in both memory and effector functions occurred in transcriptionally distinct subsets. Thus, CDK4/6i appears to serve a dual function in a heterogeneous T-cell population, enhancing acute cytotoxic effector function while promoting differentiation of memory subsets that are capable of maintaining ongoing antitumor immunity.

Strategies to preserve or induce stem-like properties of T cells to promote sustained immunity are of significant clinical interest. Thus, the capacity for CDK4/6i to promote T-cell memory makes them an attractive clinical tool, both as immunomodulatory agents in their own right and as adjuvants to T cell–directed immunotherapies. Indeed, in this study we demonstrated that short-term priming with CDK4/6i enhanced endogenous antitumor T-cell immunity, significantly improved the persistence and efficacy of adoptively transferred CAR T cells, and induced a T cell–intrinsic gene signature that correlated with favorable responses to ICB in melanoma patients. In support of this, preconditioning tumor-bearing mice with a CDK4/6i significantly improved the efficacy of PD-1 ICB. CDK4/6i have already demonstrated synergy with blockade of the PD-1/PD-L1 axis in a number of preclinical models (4, 6–8) through mechanisms that are incompletely understood. Although the efficacy of this combination is largely attributed to tumor-intrinsic immunomodulatory activity of CDK4/6i, our data suggest a novel mechanism whereby CDK4/6i promote a more favorable T-cell pool for immune checkpoint targeting. This suggests that CDK4/6 inhibition may be used to prime the T-cell pool before the application of ICB. Such a strategy has enormous implications for the utility of CDK4/6i as immunotherapy adjuvants, as so far clinical trials coadministering CDK4/6 inhibition with ICB have been halted owing to severe toxicity (40). Using CDK4/6i as a preconditioning tool, rather than a combination therapy, may mitigate this toxicity risk. Furthermore, as CDK4/6 inhibition compromises expansion of CD8⁺ T cells, using CDK4/6 inhibition as a priming tool may also be a more efficacious strategy, as it removes the potential complication of restricted T-cell expansion after ICB administration, which is likely to occur if CDK4/6 inhibition is continuously coadministered. Our results, as well as recent reports, demonstrate that transient CDK4/6 inhibition is sufficient to promote antitumor immunity (45, 46), and here we demonstrate a novel strategy for using CDK4/6i to enhance the efficacy of ICB, without the toxicity risk associated with combination approaches.

In summary, using integrated multi-omics approaches, we uncover a novel mechanism of action of CDK4/6i in promoting long-term antitumor immunity through the induction of T-cell memory. We show that direct pharmacologic inhibition of CDK4/6 in T cells redirects cell fate and may be clinically leveraged to improve a variety of T cell–directed antitumor immunotherapies, including CAR T-cell therapy and ICB. As CDK4/6i undergo investigation in hundreds of clinical trials, our findings have direct and widespread implications, with potential to inform current and future design of trial protocols incorporating CDK4/6i, particularly in combination with immunotherapy, for the treatment of cancer.

Animal work was conducted with approval from the Peter MacCallum Animal Experimentation Ethics Committee in accordance with the National Health and Medical Research Council Australian code for the use of animals for research purposes 8th edition (2013). C57BL/6 mice were purchased from Walter Eliza Hall Institute, and C57BL/6-Tg(OT-I) and NSG mice were bred in-house. All tumors were injected subcutaneously on the right flank, and mice were randomized and treated with 80 mg/kg palbociclib daily by oral gavage, starting 3 days after tumor inoculation. Palbociclib (6-acetyl-8-cyclopentyl-5-methyl-2-((5-(piperazin-1-yl)pyridin-2-yl)amino)pyrido[2, 3-d]pyrimidin-7(8H)-one) was provided by Pfizer. Mice were treated for 7 or 10 days for tumor harvest or survival studies, respectively. CD8⁺ (YTS 169.4) or CD4⁺ (GK1.5) depletion antibodies were administered at 250 μg/mouse on days –1 and 0 and 150 μg/mouse on days 4 and 8 and weekly ongoing, with day 0 being the day of tumor inoculation. For survival studies, mice were euthanized when tumors reached an ethical endpoint of 1,200 mm³ or no tumor could be seen 100 days after inoculation. For transfer studies (excluding LM-OVA), mice received 5 × 10⁶ OT-I T cells or 10 × 10⁶ LeY/HER2 CAR T cells by intravenous administration. For LM-OVA studies, spleens were harvested from mice that had 30 days previously received 1 × 10⁶ OT-I T cells intravenously, and then splenocytes containing 5 × 10⁵ transferred OT-I cells were retransferred into recipient mice simultaneously with 10⁵ CFU LM-OVA.

Tumors were digested with Collagenase IV (1.6 mg/mL) and DNase (2 U/mL) in DMEM for 45 minutes at 37°C with agitation and filtered through a 70-μm filter. Spleens were processed through a 70-μm filter and lysed with red cell lysis buffer (150 mmol/L NH₄Cl, 10 mmol/L KHCO₃, and 0.1 mmol/L Na2-EDTA). Lymphocytes were isolated from blood using Histopaque Density Gradient Media (Sigma-Aldrich).

Mouse T cells were isolated from tumors by FACS sorting on CD3⁺TCRB⁺CD90.2⁺CD11b⁻ cells, plus CD8a⁺ where indicated. Human T cells were isolated from PBMCs by FACS sorting CD3⁺CD14⁻ CD16⁻ CD19⁻ cells. Fixable yellow (Invitrogen, L34959) or propidium iodide (PI) was used to stain live/dead cells. Anti-mouse antibodies used were CD3 (17A2), TCRB (H57-597), CD11b (M1/70), CD90.2 (53-2.1), CD8α (clone 53-6.7), CD4 (GK1.5), CD44 (clone IM7), CD62L (clone MEL14), CD45.1 (clone A20), CD45.2 (clone 104), KLRG1 (clone 2F1), CD127 (clone A7R34), TCF1 (C63D9), PD-1 (29F.1A12), TIM3 (B8.2C12), CD25 (clone PC61), CD69 (H1.2F3), Bim (2819; Cell Signaling Technology), Bcl2 (3F11), CD27 (LG.3A10), CXCR3 (CXCR3-173), IRF4 (IRF4.3E4), Ki-67 (B56), EOMES (W17001), TBET (4B10), GZMB (GRB04), CD107a (1D4B), IFNγ (XMG1.2), TNFα (MP6-XT22), and IL-2 (JES6-5H4). Anti-human antibodies used were CD62L (SK11), CD45RA (HI100), CD45RO (UCHL1), CCR7 (G043H7), CD8 (RPA-T8), CD4 (OKT4), CD3 (OKT3), CD14 (HCD14), CD16 (3G8), and CD19 (HIB19). Cell sorting was conducted using BD FACSAria Fusion, and analysis was performed on BD LSR Fortessa X-20 or BD FACSymphony flow cytometer (BD Biosciences). Data were analyzed using FlowJo software.

MC38-OVA and B16-OVA cell lines were cultured in DMEM containing 10% FBS at 37°C in 10% CO₂. YOVAL1.1 cells were cultured in RPMI 1640 plus 20 mmol/L HEPES containing 10% FBS, 1% GlutaMAX, 1 mmol/L sodium pyruvate, 1% MEM nonessential amino acids, and 0.1% 2-mercaptoethanol at 37°C in 5% CO₂. Jurkat and OVCAR-3 cells were cultured in RPMI 1640 containing 10% FBS and 1% GlutaMAX. All cell lines were confirmed negative for Mycoplasma by PCR before frozen stocks were prepared. Cells were then used within 3 weeks of being thawed.

Naïve CD8⁺ T cells were isolated from C57BL/6 mouse spleens using EasySep Mouse CD8⁺ T Cell Isolation Kit (Stem Cell Technologies) and labeled with division tracking dye Cell Trace Violet. Purity of the T-cell population was verified as >95% CD8⁺ CD4⁻ by flow cytometry. Labeled CD8⁺ T cells were stimulated with plate-bound αCD3 antibody, αCD28 antibody (clone 37.51; 2 μg/mL; WEHI), and mIL2 (100 U/mL; WEHI). To obtain OT-I T cells, splenocytes from C57BL/6.OT-I transgenic mice were cultured with 10  nmol/L SIINFEKL plus 100 IU/mL IL2 for 72 hours. All primary mouse cells were cultured in RPMI 1640 plus 20 mmol/L HEPES containing 10% FBS, 1% GlutaMAX, 1 mmol/L sodium pyruvate, 1% MEM nonessential amino acids, 0.1% 2-mercaptoethanol, and antibiotic–antimitotic at 37°C in 5% CO₂ and supplemented with 100 IU/mL IL2.

For all assays, primary mouse cells were activated for 72 hours followed by 24-hour treatment with 1 μmol/L palbociclib unless otherwise specified. Absolute cell numbers were determined with the addition of 1 × 10⁴ calibration beads directly to cells before analysis; 0.2 μmol/L PI was used to identify dead cells by exclusion. The ratio of live cells to beads was measured by flow cytometry to determine the absolute live cell number. Mean division number calculations were performed as previously described (47).

Cells were “stained” with Cell Hashing antibodies and CITE-seq antibodies as previously described (24, 25). Stained and washed cells were counted, brought to ∼1,000 cells/μL, and loaded onto the 10x Chromium instrument (10x Genomics) to generate single-cell Gel Beads-in-Emulsion and capture/barcode cells. All samples followed the 10x Genomics Single Cell 3′ v3 according to the manufacturer's instructions up until the cDNA amplification step (10x Genomics). Two picomoles of Hashtag oligonucleotides (HTO) and ADT additive oligonucleotides were spiked into the cDNA amplification PCR, and cDNA was amplified according to the 10x Single Cell 3′ v3 protocol (10x Genomics). After cDNA amplification, 0.6× Solid Phase Reversible Immobilization (SPRI) was used to separate the large cDNA fraction derived from cellular mRNAs (retained on beads) from the ADT- and Cell Hashtag (HTO)–containing fraction (in supernatant). The cDNA fraction was processed according to the 10x Genomics Single Cell 3′ v3 protocol to generate the transcriptome library; indexing was done using a Chromium i7 Multiplex Kit. An additional 1.4× reaction volume of SPRI beads was added to the ADT/HTO fraction to bring the ratio up to 2.0×. The beads were washed with 80% ethanol and eluted in water, and an additional round of 2.0× SPRI was performed to remove excess single-stranded oligonucleotides from cDNA amplification. After final elution, separate PCRs were set up to generate the CITE-seq ADT library (SI-PCR and RPI-x primers) and the HTO library (SI-PCR and D7xx_s). For 5′ multi-omics including gene expression, TCR sequencing, and Hashing/CITE-seq, we used the 10x Single Cell V(D)J kit with Feature Barcoding (enabled for 5′ Gene Expression, TCR, and Feature Barcoding for Cell Surface Protein) protocol following the manufacturer's instructions. TCR libraries were prepared using the 10x Chromium Single Cell V(D)J Enrichment Kit, Human (or mouse) T Cell. 5′ HTO/ADT (1 library) was prepared using Chromium Single Cell 5′ Feature Barcode Library Kit and indexed using the Chromium i7 Multiplex Kit N, Set A. Reads were aligned to the mm10 or hg19 reference genome, cellular barcodes were demultiplexed, and unique molecular identifiers and antibody capture (ADTs and HTOs) were quantified using 10x Genomics Cell Ranger software (version 3.1.0). Cell barcodes containing RNA or antibody counts from cells from >1 sample (intersample doublets) were identified using Seurat's HTODemux function. Barcodes containing counts from >1 cell within the same sample (intrasample doublets) were identified using the Scrublet python package (version 0.2.1; ref. 48). A cutoff of >1 median absolute deviation (MAD) value above the median Scrublet score was chosen. Cells identified as either type of doublet were removed from the analysis. Relative TF activity in each cell was estimated using the SCENIC method (49). Area-under-curve (AUC) scores for TFs were calculated based on inferred gene regulatory networks using the Pyscenic Python package (version 0.9.19). Gene expression and antibody count matrices were processed in R (version 3.6.1) using the Seurat R package (version 3.1.0; ref. 50). RNA transcript counts for barcodes identified as cells by Cell Ranger were normalized using sctransform via Seurat's SCTransform function. The counts were first transformed with no covariates in the sctransform model, and then cell-cycle phase scores were estimated using Seurat's CellCycleScoring function, with mouse homologs of the cell-cycle gene sets provided by Seurat. The sctransform normalization was then rerun with the cell-cycle phase scores and the percentage of raw RNA counts belonging to mitochondrial genes for each cell as variables to be regressed out in the model. ADT counts were normalized using centered log ratio transformation. Principal component analysis was then performed on the sctransform-scaled RNA expression values for genes with residual variance in the sctransform model >1.3. A shared-nearest-neighbors (SNN) network was calculated using the top 10 principal components using the FindNeighbors function with k-nearest neighbors set to 50 and cosine distance metric. The SNN network was then used to identify cell populations using the FindClusters function using the Louvain algorithm with resolution parameter 0.6. Uniform Manifold Approximation and Projection (UMAP) values were also calculated using the RunUMAP function with the top 10 principal components as input and parameters n.neighbors = 50 and metric = cosine. Monocle3 was used for pseudotime analysis using default settings.

Jurkat cells were transduced with mCherry-Cas9 using UCas9Cherry and cell sorted for mCherry⁺ cells. mCherry-Cas9–expressing cells were then transduced with lentivirus containing a custom cloned genome-wide sgRNA library. Forty-eight hours after transduction, successfully transduced cells were selected with puromycin (1 mg/mL; Millipore) for 5 days. After selection at the commencement of screen, a reference sample at time point 0, T0, was snap frozen and stored at −80°C. For the screen, library-containing cells (20 × 10⁶) were left untreated or treated with palbociclib (1 μmol/L) for 3 days. Cells were then stained with anti-human CD62L and FACS sorted on the lowest 10% percentile of CD62L expression. Cells were put back in culture for 5 days, and the process was repeated. Genomic DNA was extracted from screen and control samples using the DNeasy Blood & Tissue Kit (Qiagen) according to the manufacturer's instructions. sgRNA sequences were then amplified for next-generation sequencing by PCR using specific adaptor sequences. PCR products were pooled and cleaned up using the AMPure XP-PCR purification system (Beckman Coulter) according to the manufacturer's instructions. Samples were subsequently multiplexed and sequenced on the NextSeq 500 (Illumina) in-house at the Peter MacCallum Molecular Genomics Core Facility generating 75-bp single-end reads. After demultiplexing with CASAVA (version 1.8), adaptor sequences were removed using Cutadapt (version 1.7), leaving the 20-bp sgRNA sequence. Model-based Analysis of Genome-wide CRISPR/Cas9 Knockout, version 0.5.9 (MAGeCK count and MAGeCK test), software was subsequently used to count the reads and perform sgRNA enrichment and statistical analyses between treated/sorted and control samples. Screen data were visualized using the R package, ggplot2, version 2.2.1.

Targeted gene deletion was performed by electroporation of Cas9 nuclease/synthetic guide RNA (Synthego) into target cells using Amaxa 4D-Nucleofector electroporation system (Lonza, AAF-1002B) as per manufacturer's protocol. Primary mouse naïve CD8⁺ T cells were isolated and electroporated with Cas9/sgRNA using P3 Primary Cell 4D-Nucleofector X kit (Lonza, V4XP-3024) before activation with CD3/CD28. Jurkat cells were electroporated using SE Cell Line 4D-Nucleofector X Kit (Lonza, 4XC-1032).

DNA encoding the anti-LeY chimeric receptor was generated with standard molecular biology techniques, using a scFv (51) generated from the humanized mAb Hu3S193 (52). A truncated human CD34 was inserted as an extracellular domain in the current second-generation construct (anti-LeY scFv-CD3ζ-CD28) as previously described (37). Human CD34 was cloned from cDNA (53), and a truncated (extracellular domain) form was amplified using PCR and cloned into the LeY retroviral plasmid using appropriate restriction digest and ligations.

Human PBMCs were isolated from normal donor buffy coats and retrovirally transduced with an anti-human LeY chimeric receptor. PBMCs were stimulated with anti-human CD3 (OKT3 30 ng/ml, Ortho Biotech) and IL2 (600 U/mL), followed 3 and 4 days later by incubation with supernatant from the PG13 retroviral producer cell line on a Retronectin (Takara Bio) matrix as per the manufacturer's instructions. Briefly, Retronectin was coated at a concentration of 6 mg/cm² in non–tissue culture 6-well plates (Becton Dickinson) and incubated with 5 mL of retroviral supernatant. After 4 hours, supernatant was removed, and 2.5 × 10⁶ T cells in 5 ml of complete medium were added for 12 hours. The transduced T cells were expanded for ≤8 days in culture with IL2 (600 U/mL) ± 1 μmol/L palbociclib.

Splenocytes from C57BL/6 mice were activated with anti-CD3 (0.5 mg/mL) and anti-CD28 (0.5 mg/mL) in the presence of IL2 (100 IU/mL) and IL7 (2 ng/mL), followed by transduction with retrovirus encoding the CAR (comprised of an extracellular scFv specific for human HER2, a CD8 hinge region, and transmembrane CD28 and CD3z domains) as previously described (54, 55).

Cell pellets were lysed in 2% SDS buffer (0.5 mmol/L EDTA and 20 mmol/L HEPES), boiled (95°C, 5 minutes), and quantified using the DCTM protein assay (Bio-Rad) as per manufacturer's protocol. Equal amounts of total protein in 5× SDS sample buffer [313 mmol/L Tris HCl, pH 6.8, 50% (v/v) Glycerol, 10% (v/v) β-mercaptoethanol, 10% (w/v) SDS, and 0.05% (w/v) Bromophenol blue] were boiled (95°C, 5 minutes), loaded into precast gels (Bio-Rad), and resolved via SDS-PAGE with running buffer containing 25 mmol/L Tris, 190 mmol/L glycine, and 0.1% (w/v) SDS. Precision-plus protein dual color standard (Bio-Rad) was used as a molecular weight marker. Proteins were transferred onto methanol-activated polyvinylidene difluoride membranes (Millipore) using the Trans-Blot Turbo semi-dry transfer system (Bio-Rad) for 20 to 30 minutes with Tris-glycine transfer buffer [50 mmol/L Tris, 40 mmol/L glycine, 0.375% (w/v) SDS, and 20% (v/v) methanol]. Membranes were blocked in 5% skim milk made up in TBS containing 0.1% Tween-20 (TBS-T) for 1 hour before probing with primary antibody overnight at 4°C, followed by corresponding horseradish peroxidase–conjugated secondary antibody for 1 hour at room temperature. Immunoblots were washed three times with TBS-T (10 minutes each) after each antibody incubation, and proteins were detected using ECL western blotting substrate (Amersham GE Healthcare). Antibody details: anti-RB1 (G3–245), anti-Tubulin (DM1A).

Cells were collected and washed once with ice-cold PBS before resuspension in TRIzol (ThermoFisher Scientific, 15596026). RNA was isolated using the Direct-zol RNA MiniPrep kit (Zymo Research, R2052) according to the manufacturer's instructions. Sequencing libraries were prepared using the QuantSeq 3′-mRNA Seq Library Prep Kit for Ilumina (Lexogen). Libraries were sequenced on the Illumina NextSeq 500 to obtain 75-bp single-end reads. Sequencing files were demultiplexed using Bcl2fastq (version 2.17.1.14) to generate FASTQ files on which quality control was performed using FASTQC (version 0.11.5). Sequencing reads were trimmed using cutadapt (version 1.7) and aligned to the mouse reference genome (mm10) or human reference genome (Hg19) using HISAT2 (version 2.1.0). Read counting across genomic features was performed using FeatureCounts (from the Subread package, version 1.5.0) before differential gene expression analysis using Voom/Limma. GSEAs were performed using GSEA software (Broad Institute), and barcode plots were plotted using ReplotGSEA.R (part of the Rtoolbox R package).

Cells were washed once in ice-cold PBS and lysed in ATAC lysis buffer (0.1% Tween-20, 0.1% NP-40, 3 mmol/L MgCl₂, 10 mmol/L NaCl, and 10 mmol/L Tris HCl, pH 7.4). Tagmentation was performed with Tn5 transposase and 2× TD Buffer (Nextera DNA Library Prep Kit, Illumina) for 30 minutes at 37°C. Tagmented DNA was purified using a MinElute colum (Qiagen, 28004) and amplified for 12 cycles using 2 × KAPA HiFi HotStart ReadyMix (Kapa Biosystems, KK2602). The amplified libraries were purified using MinElute columns (Qiagen) and sequenced on an Ilumina NextSeq 500 with 75-bp single-end reads. Library quality control and quantification were performed using D1000 high-sensitivity screen tape with 4200 TapeStation Instrument (Agilent Technologies) and size selected for 200 to 700 bp using a Pippin Prep system (Sage Science).

T cells were cross-linked with 2% formaldehyde for 10 minutes at room temperature and quenched with addition of 1.25 mol/L glycine. Cells were washed twice (5% BSA, 800 × g, 5 minutes, 4°C) and lysed in nuclear extraction buffer 3 times. Nuclei were resuspended in sonication buffer and sonicated with a Covaris S220 sonicator (peak power, 105; duty factor, 20; cycle/burst, 200; duration, 600 seconds). Samples were cleared by centrifugation at 12,000 × g for 20 minutes, and 1 volume of dilution buffer was added to cleared chromatin; 1% chromatin was taken as input. Immunoprecipitation was performed overnight at 4°C with rotation with 2 μg anti-H3K27ac antibody (Abcam). Immunoprecipitated samples were captured by incubation with 20 μL Dynabeads Protein G (Life Technologies) blocked with 0.1% BSA for 2 hours. Beads were then washed twice each with wash buffer 1, wash buffer 2, wash buffer 3, and TE buffer. DNA was eluted with 100 μL elution buffer for 30 minutes twice and reverse cross-linked. DNA product was purified using Zymo ChIP DNA Clean and Concentrator Kit. Libraries were generated using the NEBNext Ultra II DNA library prep kit from NEB. For sequencing, libraries were pooled and sequenced on a NextSeq 500 (Illumina), and 15 to 20 million single-end 50-bp reads were generated per sample.

Bcl2fastq version 2.17.1.14 was used for demultiplexing. The Fastq files generated by sequencing were aligned to the mouse reference genome (GRCm38/mm10) using bowtie (version 2.2.3). Samtools (version 1.8) was used for manipulation of SAM and BAM files, after which MACS (version 2.0.10) was used for peak calling. Browser-viewable TDF files were generated using IGVTools (version 2.3.72), and ChIP-seq tracks were visualized using IGV (version 2.3.55). Differentially accessible regions were quantitatively analyzed using Rsubread featureCounts on merged reference bed file containing all peaks identified across treatment conditions, after which the limma-voom package was used for statistical analysis of differentially accessible regions. Subsequently, HOMER (version 4.8.3) was used for motif analysis on MACS2 peak summits using FindMotifGenome.pl for differentially accessible regions, using all summits as a background set with the –bg option. ChIP-seq and ATAC-seq peaks were annotated to genes using the AnnotatePeaks.pl function, after which R was used for visualization.

Patient peripheral blood samples were obtained via Peter MacCallum Cancer Centre with written informed consent from the patient and approval from the Peter MacCallum Cancer Centre institutional human research ethics committee (HREC #11/105).

Statistical tests were performed using GraphPad Prism. All error bars show  ± SEM. Significance was determined as *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

All sequencing datasets are available on the Gene Expression Omnibus (GEO) under accession number GSE182664.

500 g of each cell pellet was digested using the USP protocol. Cell pellets were dissolved in 5% SDS, 10 mmol/L Tris (2-carboxy-ethyl)-phosphin-HCl, 40 mmol/L 2-CAA, Tris-HCl (pH 7.5) by heating to 95°C for 10 minutes. DNA was hydrolyzed using 1% trifluoroacetic acid (TFA) before the sample was neutralized using 3 mol/L Tris (final concentration 160 mmol/L). 20 μL of carboxy-agarose magnetic beads (CUBE Biotech) were added to each sample before addition of acetonitrile (ACN; 70% v/v). Samples were mixed and incubated for 20 minutes at room temperature. Samples were placed against a magnetic rack and washed twice with 70% ethanol and once with 100% ACN. Samples were lyophilized to dryness to remove residual ACN. 50 L of lysis buffer [10% trifluoroethanol (v/v), 100 mmol/L ammonium bicarbonate)] containing 1:25 enzyme:substrate ratio of trypsin (Promega) plus LysC (Wako). Samples were placed in an ultrasonic water bath for 2 minutes before incubation in a Thermomixer (Eppendorf) at 37°C, 1,200 rpm for 12 hours. Samples were placed on a magnetic rack and supernatant-containing peptides collected. Beads were washed in 50 L of MilliQ water, placed on a magnetic rack, and supernatant collected and pooled with first collection. Peptide solution was acidified using TFA (1% v/v final). Solution was centrifuged at 20,000 × g to remove precipitate and supernatant collected. ACN-containing TFA was added such that the final concentration was 80% ACN, 0.1% TFA. 50 L of FeIII-NTA beads (CUBE Biotech) were added to each sample and incubated with gentle shaking for 30 minutes at room temperature. Samples were placed against a magnetic rack and washed vigorously three times using wash buffer (80% ACN, 0.1% TFA). Beads were transferred with 40 μL of wash buffer into the top of prewetted C8 stage-tips (2 plugs, prewet with 50 μL ACN before use). C8 stage-tips were centrifuged at 500 × g to remove wash buffer. 30 μL of 10% TFA was added to the bottom of a collection tube. Phosphopeptides were eluted using 2X 20 μL Elution buffer (50% ACN, 2.5% ammonium hydroxide, pH 10). Samples were lyophilized until near to dryness. Peptide solutions were resuspended in 50 μL 5% formic acid (FA) and transferred into C18 stage-tips (2X plugs, wet with 20 μL 100% IPA, 60% ACN, 5% FA, reequilibrated using 5% FA before use). Stage-tips were washed with 2X 50 μL 5% FA and eluted using 50 μL 60% ACN, 5% FA. Samples were lyophilized to dryness and stored at −80°C. Samples were analyzed on a M-class UPLC (Waters Corp.) coupled to a timsTOF Pro (Bruker) equipped with a CaptiveSpray source. Peptides were resuspended in 2% ACN, 1% FA separated on a 25 cm × 75 μm analytic column, 1.6 μm C18 beads with a packed emitter tip (Aurora Series, IonOpticks). The column temperature was maintained at 50°C using an integrated column oven (Sonation GmbH). The column was equilibrated using five column volumes before loading the sample in 100% buffer A (99.9% MilliQ water, 0.1% FA). Samples were separated at 400 nL/minute using a linear gradient from 2% to 17% buffer B (99.9% ACN, 0.1% FA; 55 minutes), 17% to 25% buffer B (21 minutes) before ramping to 35% buffer B (13 minutes), ramp to 85% buffer B (3 minutes), and sustained for 10 minutes. The timsTOF Pro (Bruker) was operated in PASEF mode using Compass Hystar 5.0.36.0. The settings were as follows: mass range, 100 to 1,700 m/z, 1/K0, start 0.6 V⋅s/cm2, end, 1.6 V⋅s/cm2, ramp time 109.9 ms, lock duty cycle to 100%, capillary voltage 1,600 V, dry gas 3 L/minute, dry temp 180°C; PASEF settings: 10 MS/MS scans (total cycle time 1.26 seconds), charge range 0–5, active exclusion for 0.4 minutes, scheduling target intensity 20,000, intensity threshold 2,500, CID collision energy 48 eV. Raw files were analyzed using MaxQuant (version 1.6.14). The database search was performed using the Uniprot Homo sapiens database plus common contaminants with strict trypsin specificity allowing up to two missed cleavages. The minimum peptide length was 7 amino acids. Carbamidomethylation of cysteine was a fixed modification while N-acetylation of proteins N-termini, oxidation of methionine, and phosphorylation of serine/threonine/tyrosine were set as variable modifications. During the MaxQuant main search, precursor ion mass error tolerance was set to 6 ppm. PSM and protein identifications were filtered using a target-decoy approach at a false discovery rate of 1% with the match between runs option enabled. Further analysis was performed using a custom pipeline developed in R, which utilizes the MaxQuant output file evidence.txt. A feature was defined as the combination of peptide sequence, charge, and modification. Features not found in at least 50% of the replicates in one group were removed. To correct for injection volume variability, feature intensities were normalized by converting to base 2 logarithms and then multiplying each value by the ratio of maximum median intensity of all replicates over median replicate intensity. Missing values were imputed using a random normal distribution of values with the mean set at mean of the real distribution of values minus 1.8 SD, and an SD of 0.3 times the SD of the distribution of the measured intensities. The probability of differential peptide expression between groups was calculated using the Limma R package2. Probability values were corrected for multiple testing using Benjamini–Hochberg method. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD026539.

E.J. Lelliott reports PeterMac Postgraduate Scholarship; Melbourne University Research Scholarship (58616); Cancer Therapeutics CRC PhD Top Up Scholarship; and CASS Foundation - Medicine/Science grant (9870). P.A. Beavis reports National Breast Cancer Foundation fellowship (ECF-17-005) and grants from AstraZeneca and Gilead Sciences outside the submitted work. P.J. Neeson reports National Health and Medical Research Council grant and grants from BMS, Roche Genentech, Compugen, Allergan, and Crispr Therapeutics outside the submitted work. G.A. McArthur reports National Health and Medical Research Council project grant (1100189) and other support from Roche-Genentech and Array Biopharma/Pfizer outside the submitted work. R.W. Johnstone reports National Health and Medical Research Council project grant, program grant (454569), and fellowship (SPRF); Cancer Council Victoria project grant; Kids' Cancer Project; grants from Astra Zenec, Roche, and BMS; and grants, personal fees, and other support from MecRx during the conduct of the study; and is a scientific consultant and shareholder in MecRx. J. Oliaro reports grants from National Health and Medical Research Council Australia (1139626), National Breast Cancer Foundation (IIRS-18-151), Peter MacCallum Cancer Foundation, and CASS Foundation during the conduct of the study and outside the submitted work. K.E. Sheppard reports grants from the National Health and Medical Research Council (1100189) during the conduct of the study and nonfinancial support from GlaxoSmithKline and grants from Victorian Cancer Agency outside the submitted work. E.D. Hawkins reports National Health and Medical Research Council project grants (1140187, 1165591) and fellowship (CDF2, 159488); and Leukemia & Lymphoma Society grant (6552-18). S.J. Vervoort reports National Health and Medical Research Council fellowship (EL1, GNT1178339); Peter Mac Foundation grant; Netherlands Organization for Scientific Research (Rubicon Fellowship, NWO, 019.161LW.017); and Kids' Cancer Project. C.J. Kearney reports National Health and Medical Research Council fellowship (ECF). D. Meyran reports Plan Cancer 2014–2019, l'Institut Servier and Fondation Nuovo-Soldati, and Pfizer Oncology. The McArthur laboratory receives non-financial support from Pfizer Oncology for supply of palbociclib. No disclosures were reported by the other authors.

E.J. Lelliott: Conceptualization, data curation, formal analysis, investigation, writing–original draft. I.Y. Kong: Data curation, validation, investigation. M. Zethoven: Data curation, software, visualization. K.M. Ramsbottom: Investigation. L.G. Martelotto: Data curation, validation, investigation. D. Meyran: Investigation. J. Jiang Zhu: Investigation, methodology. M. Costacurta: Investigation, visualization, methodology. L. Kirby: Investigation, methodology. J.J. Sandow: Investigation, visualization. L. Lim: Investigation, methodology. P.M. Dominguez: Validation, investigation, methodology. I. Todorovski: Visualization, methodology. N.M. Haynes: Investigation, methodology. P.A. Beavis: Investigation, methodology. P.J. Neeson: Investigation, methodology. E.D. Hawkins: Investigation, methodology. G.A. McArthur: Resources, methodology. I.A. Parish: Investigation, methodology. R.W. Johnstone: Resources, funding acquisition, methodology, project administration, writing–review and editing. J. Oliaro: Resources, methodology, writing–review and editing. K.E. Sheppard: Resources, supervision, funding acquisition, methodology. C.J. Kearney: Conceptualization, data curation, formal analysis, supervision, visualization, writing–review and editing. S.J. Vervoort: Conceptualization, formal analysis, supervision, investigation, visualization, methodology, writing–review and editing.

This work was supported by the National Health and Medical Research Council project grants [1100189 (to G.A. McArthur and K.E. Sheppard); 1139626 (to J. Oliaro and R.W. Johnstone); and 1140187 and 1165591 (to E.D. Hawkins)], program grants (454569, to R.W. Johnstone and P.J. Neeson), and fellowships [EL1, GNT1178339, to S.J. Vervoort; 159488, to C.J. Kearney (ECF); R.W. Johnstone (SPRF); to E.D. Hawkins (CDF2)]; Peter Mac Foundation grant (to S.J. Vervoort); Cancer Council Victoria project grants (to R.W. Johnstone); National Breast Cancer Foundation grants (IIRS-18-151, to J. Oliaro) and fellowships (ECF-17-005, to P.A. Beavis); The CASS Foundation Medicine/Science Grant (9870, to E.J. Lelliott); Netherlands Organization for Scientific Research (Rubicon Fellowship, NWO, 019.161LW.017, to S.J. Vervoort); PeterMac Postgraduate Scholarship (to E.J. Lelliott); Melbourne University Research Scholarship (58616, to E.J. Lelliott); Cancer Therapeutics CRC (CTx) PhD Top Up Scholarship (to E.J. Lelliott); The Kids' Cancer Project (to R.W. Johnstone and S.J. Vervoort); Leukemia & Lymphoma Society grant (6552-18, to E.D. Hawkins); Plan Cancer 2014–2019, l'Institut Servier and Fondation Nuovo-Soldati (to D. Meyran) and Pfizer Oncology.

We thank Shahneen Sandhu and the Peter MacCallum Cancer Centre Melanoma Biomarker Project for provision of patient samples, and the Peter MacCallum Cancer Centre Translation Research Laboratory (Carleen Cullinane, Susan Jackson, Kerry Warren, and Jeannette Schreuders) for technical support with animal work.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.