We performed single-cell RNA-sequencing and T cell receptor clonotype tracking of breast and ovarian cancer patients treated with the CDK4/6 inhibitor ribociclib and PD-1 blockade. We highlight evidence of two orthogonal treatment-associated phenomena: expansion of T cell effector populations and promotion of T cell memory formation. Augmentation of the antitumor memory pool by ribociclib boosts the efficacy of subsequent PD-1 blockade in mouse models of melanoma and breast cancer, pointing toward sequential therapy as a potentially safe and synergistic strategy in patients.

Inhibitors of Cyclin-Dependent Kinases 4 and 6 (CDK4/6) have recently become keystone agents in the treatment of advanced hormone-receptor-positive breast cancer (Fassl et al., 2022). Beyond their effects on cell-cycle progression, they have been shown to function as potent modulators of T cell immunity (Deng et al., 2018; Goel et al., 2017; Schaer et al., 2018; Zhang et al., 2018). This insight has led to efforts exploring whether these agents can synergize with established T cell–targeting immunotherapies, such as blockers of the immune checkpoint Programmed Cell Death Protein 1 (PD-1; Tolba et al., 2021; Yuan et al., 2021).

Immune checkpoint blockade (ICB) fundamentally centers around the relief of effector T cell inhibition with antibodies targeting multiple inhibitory pathways now approved for clinical use (Burnell et al., 2022; Schildberg et al., 2016). Prior to the advent of ICB, diseases such as metastatic melanoma were uniformly lethal, whereas today ICB can induce a positive response in nearly half of treated patients (Larkin et al., 2019). However, the limited lifespan of an invigorated effector T cell cannot account for “tail of the curve” survivors, some of whom remain free of disease well beyond 10 yr (Wherry et al., 2003; Wolchok et al., 2022). This remarkable phenomenon depends on the persistence of long-lived antitumor memory T cells. CD8 T cell fate decisions are affected in the early stages of priming, with lineage commitment occurring prior to the first cell division (Clancy-Thompson et al., 2018; Kakaradov et al., 2017). Several factors have been shown to skew toward the memory cell fate including low antigen density, increased reliance on oxidative phosphorylation, and low Myc activity (Guo et al., 2022; Henrickson et al., 2013; Kaech and Ahmed, 2001; Pearce et al., 2009; van der Windt et al., 2012; Verbist et al., 2016). Despite the critical importance of memory cells, no therapies have been definitively shown to target or induce their formation. CDK4/6 inhibitors have emerged as promising candidates, with our group and others demonstrating that they enhance memory formation in mouse models of melanoma and in breast cancer patients (Heckler et al., 2021; Lelliott et al., 2021). We previously showed that exposure to CDK4/6 inhibitors at the time of T cell priming decreased activity of Myc and increased persistence of adoptively transferred tumor-specific T cells in mice. In humans, recently activated CD8 T cells from peripheral blood of patients starting on CDK4/6 inhibitors showed clonotype skewing toward memory precursors and decreased MYC activity, consistent with mechanistic studies performed in mice (Heckler et al., 2021).

High-grade toxicities often encountered upon combination treatment with a CDK4/6 inhibitor and PD-1 blockade have made it difficult to assess long-term clinical outcomes (Dougan et al., 2021; Pujol et al., 2021). Consequently, whether CDK4/6 inhibition acts cooperatively with PD-1 blockade in antitumor immunity has remained unclear. Here, we demonstrate by single-cell RNA-sequencing (scRNA-seq) of patient blood and tumor samples that dual treatment does affect distinct aspects of T cell activation, with PD-1 blockade invigorating short-term effector responses while CDK4/6 inhibition promotes the expression of markers of stemness and memory. Based on our findings, we proposed that sequential therapy may be a more optimal dosing strategy, whereby early treatment with a CDK4/6 inhibitor promotes the formation of a more durable memory T cell pool that can later serve as a substrate for PD-1 blockade, and we validate this hypothesis in a mouse model of melanoma of neoadjuvant treatment of melanoma and breast cancer.

We sequenced circulating T cells and tumor-infiltrating lymphocytes (TILs) from six patients with metastatic breast or ovarian cancer enrolled in a phase I clinical trial (NCT03294694) investigating the combination of the CDK4/6 inhibitor ribociclib with spartalizumab, a monoclonal antibody against PD-1 (Fig. 1 A and Table 1). Samples were obtained before treatment and after one cycle of treatment. Single cells were also subject to TCR-seq, allowing us to group T cells expressing the same TCR into clonotypes and to track their fates across time and tissue site (Luoma et al., 2020).

In total, we sequenced 30,088 T cells and identified 22,433 distinct TCR clonotypes in the patients’ blood samples (Fig. 1, B and C). Cells were visualized by the similarity of their transcriptional profile on a Uniform Manifold Approximation and Projection (UMAP) plot and were found to cluster into 14 naive, effector, memory, and invariant subpopulations with cluster defining genes consistent with each of these cell types (Fig. S1 A). Cells from all patients were represented uniformly in most clusters (Fig. S1 B), with the exception of clusters Eff-CD320 and Eff-KIR3DL2, which were predominantly composed of patient-specific clonally expanded cells. Cluster Eff-CD320, in particular, consisted of a terminally differentiated effector CD4 population marked by high expression of cytotoxic genes and was almost entirely composed of an impressively expanded clonotype isolated from patient P1, putatively reactive to the tumor neoantigen NY-ESO-1 by sequence homology (Chen et al., 2021). Although the baseline T cell repertoire of the patients varied greatly in richness and clonal expansion (Fig. S2, A and B), most patients showed increased clonal expansion post-treatment (Fig. S2 C).

We first set out to determine whether we could detect treatment-induced effector cell expansion, the canonical effect of PD-1 blockade, which is known to be measurable not only in the tumor but also in peripheral circulation (Kamphorst et al., 2017; Zhang et al., 2020). Indeed, we found that many clonotypes from each of the six patients expanded in post-treatment blood samples relative to their pre-treatment frequency (Fig. 1 D), with the majority comprising CD8 T cells. Importantly, 39 of the 374 (10.4%) expanded clonotypes were also found in the tumor. The treatment-associated increase in clonal expansion disproportionately favored a subset of the sampled circulating repertoire. This pattern of increased clonal dominance, which occurred in five of the six patients, has previously been noted as an early signature of peripheral T cell dynamics in the setting of immune checkpoint blockade (Fig. S2 C; Valpione et al., 2020).

Differential expression analysis revealed that the source cells of treatment-expanded clonotypes expressed higher levels of PDCD1 (coding for PD-1) and key cytotoxic genes (GZMB, PRF1, GNLY) relative to other pre-treatment cells, consistent with the preferential targeting of PD-1+ cytotoxic T cells by anti–PD-1 therapy (Fig. 1 E). We then examined the flux of treatment-expanded clonotypes from pre-treatment to post-treatment clusters, and we found that the majority of these clonotypes originated from the effector clusters Eff-GZMB, Eff-GZMK, and Eff-CD320 (Fig. 1 F), in keeping with differential expression analysis. The most frequent transcriptional states after treatment corresponded to the same set of effector clusters, with a notable transition of some clones into the Cycling-MKI67 cluster, indicative of ongoing clonal proliferation at the time of sampling. Similar results were observed when analyzing each patient individually, including the one patient in our cohort with ovarian cancer (Fig. S2 D), though, interestingly, the group of patients with briefly stable disease displayed more pronounced flux into the cycling cluster than patients with progressive disease (Fig. S2 E).

Interestingly, most treatment-expanded clonotypes were not detected in pre-treatment samples (Fig. 1 D). The emergence of novel expanded (NE) clonotypes, a phenomenon variously termed “clonal replacement” or “clonal revival,” has previously been associated with PD-1 blockade in several cancer settings (Liu et al., 2021; Yost et al., 2019). Conjecturing that some NE clonotypes are liable to be false-positive artifacts of the inevitable undersampling of the T cell repertoire, we compared our findings to a published dataset of longitudinal peripheral scTCR-seq in metastatic breast cancer patients receiving standard of care chemotherapy with or without the PD-L1 blocking antibody atezolizumab (Zhang et al., 2021). We found that the rate of emergence of NE clonotypes was significantly elevated over the background rate observed in chemotherapy-only controls (Fig. 1 G). TCR-seq depth was similar across the three compared treatment groups (Fig. 1 H).

We proceeded to examine the transcriptomes of NE clonotypes, reasoning that they would be enriched with recently naive T cells that had newly encountered their antigen during the study interval. Our group and others have previously demonstrated that exposing recently activated T cells to a CDK4/6 inhibitor, whether in vitro or in vivo, skews them toward a memory fate (Heckler et al., 2021; Lelliott et al., 2021). Thus, we hypothesized that we could detect evidence of this phenomenon within NE clonotypes. Differential expression analysis of NE clones versus pre-existing expanded (PE) clones showed strong upregulation of genes associated with memory and stem-like phenotypes (TCF7, IL7R, CD27, LEF1, and SATB1), whereas genes associated with exhaustion (TOX; Alfei et al., 2019; Khan et al., 2019; Scott et al., 2019) and terminal effector function (such as NKG7 and GNLY) were relatively downregulated (Fig. 2 A). To determine whether this effect usually occurs with blockade of the PD-(L)1 axis or whether the memory skewing effect is a result of the ribociclib combination therapy, we extracted the top 40 differentially expressed genes between NE and PE clones and measured their expression in the Zhang et al. dataset (Zhang et al., 2021). We confirmed that the memory-skewed markers of NE were uniquely upregulated in patients treated with anti–PD-1 and ribociclib, while treatment with anti–PD-L1 strongly favored upregulation the effector-rich markers of PE cells (Fig. 2, B and C). This distinct transcriptional profile was reflected in the cluster distribution of NE clones, which were found more frequently in memory clusters than PE clones (Fig. 2 D), an effect that was preserved regardless of the patient’s clinical outcome (Fig. S2 G).

We next analyzed baseline tumor biopsies and identified several transcriptionally defined clusters of T cells including naive, memory/effector memory, progenitor exhausted, and terminal exhausted, similar to previous reports (Lowery et al., 2022; Oliveira et al., 2021; Fig. 3 A). We cross-compared all TCR clonotypes from tumor and blood to a database of viral reactive clonotypes and found that 3.4% of clonotypes had putative viral specificities (Bagaev et al., 2020; Fig. S2 H). Consistent with previous findings, these possibly viral reactive clonotypes were predominantly found in the memory/effector memory T cell tumor-infiltrating clusters (Caushi et al., 2021; Oliveira et al., 2021; Fig. 3 B).

We next sought to investigate whether only NE clonotypes experienced an upregulation of memory markers or whether all non-singleton clonotypes exhibited some degree of memory skewing among their member cells. To this end, we took advantage of a transcriptional signature that differentiated memory from effector precursors among recently activated T cells in our previously published dataset (Fig. S3). Each clonotype was assigned a memory score before and after treatment. Most PE clonotypes (80%) scored modestly higher at the latter timepoint, consistent with these clonotypes mostly being terminally differentiated cells (Fig. 3 C). This effect was reversed in the Zhang et al. dataset, indicating that PD-L1 plus chemotherapy is associated with a decreased memory signature (Fig. 3 C). Scoring with an irrelevant gene signature showed no significant change between pre- and post-treatment samples (Fig. 3 C). Importantly, after cross-matching TCR sequences to TILs screening out putative viral-reactive clonotypes, clonotypes matching those in blood with increasing memory scores were predominantly found in the terminal exhausted T cell cluster, suggesting that at least some of the T cell clonotypes with increased memory skewing are tumor-specific (Fig. 3 B; Oliveira et al., 2021).

Motivated by the finding that ribociclib may promote memory formation among tumor-infiltrating clones in patients, we wondered if early treatment with a CDK4/6 inhibitor could “sensitize” patients to later therapy with PD-1 blockade by providing a superior pool of antitumor memory cells. Although sequential therapy has been tested in mice and humans, none of the previous studies separated drug dosing by enough time to allow the formation of memory cells. To determine whether memory cells formed in the presence of CDK4/6 inhibition could be reinvigorated by PD-1 blockade, we used a murine model of melanoma. CD45.1+ TRP1high CD8 T cells, whose TCR recognizes the melanoma antigen TRP1 with high affinity (Dougan et al., 2013), were adoptively transferred into CD45.2+ mice after being activated in vitro in the presence or absence of ribociclib. The mice were rested for 36 d, during which time the frequency of CD45.1+ T cells in their blood was measured weekly. The cells that had been activated in the presence of ribociclib expanded and persisted to a greater degree (Fig. 4 A). On day 37, the mice were challenged with tumors, and half of each group was randomly selected to receive anti–PD-1 therapy. Only those mice that had received ribociclib-treated cells experienced a significant survival benefit from PD-1 blockade (median survival of 24 vs. 17 d, P = 0.0003, Fig. 4 A).

To test whether oral dosing of ribociclib to a whole mouse could similarly affect endogenously generated anti-tumor T cells, we used a 4T1 murine breast cancer cell line that had been transduced to express GFP. On the BALB/c background, GFP is presented by H-2Kd and can induce a robust CD8 T cell response (Agudo et al., 2015; Krall et al., 2018). We implanted BALB/c mice with 4T1-GFP tumor cells and treated the mice for 1 wk with daily ribociclib by oral gavage. This short course of ribociclib reduced tumor size but did not clear the tumors. All mice were then surgically cured and allowed to rest for 35 d to recover from surgery and allow for memory cell development. Mice were then rechallenged with twice the dose of 4T1-GFP tumor cells on the opposite flank and randomized to receive either anti–PD-1 or isotype control. As shown in Fig. 4 B, mice receiving either ribociclib or PD-1 blockade alone showed increased survival compared to mice receiving no active therapy. However, mice that had received an early course of ribociclib and later received PD-1 blockade were fully protected, with significantly increased survival compared to PD-1 blockade alone. These data show that, in a mouse model of breast cancer, neoadjuvant ribociclib alone induced immunologic memory, and that outcomes were further enhanced with sequential checkpoint blockade.

Prior to this study, attempts at dual immunotherapy with a CDK4/6 inhibitor and PD-1 blockade in metastatic cancer did not show promise, largely due to frequently treatment-limiting toxicities (Pujol et al., 2021; Rugo et al., 2020). By retrospectively analyzing circulating and tumor-infiltrating T cells from treated patients, taking advantage of single-cell-resolution sequencing to track clonal dynamics and transcriptomes over time, we were able to gain deep insights into the cooperative but nonoverlapping effects of each treatment on reshaping patients’ antitumor immune response. We verified that cell-cycle inhibition by ribociclib does not prevent the desirable effect of PD-1 blockade on cytotoxic effector cell expansion. We also replicated the finding of memory fate skewing under the influence of CDK4/6 inhibition in this setting, and extended the significance of this observation by matching the affected circulating cells to exhausted T cell populations in the tumor.

PD-(L)1 blockade has been shown to reinvigorate a stem-like progenitor population of exhausted CD8 T cells in both mice and humans (Beltra et al., 2020; Collier et al., 2021; Eberhardt et al., 2021; Hudson et al., 2019). The origin of these stem-like progenitor cells likely involves newly primed T cells as new T cell priming and replacement of clonotypes is simultaneously occurring with PD-(L)1 blockade (Liu et al., 2021; Yost et al., 2019). Single-cell analysis from breast cancer patients has shown that intratumoral CD103+ resident memory T cells correspond with improved prognosis, although treatment with PD-1 blockade appears to reinvigorate a different population of CD8 effector memory cells characterized by expression of GZMK and EOMES (Bassez et al., 2021; Savas et al., 2018). Whereas PD-(L)1 blockade pushes T cells toward a more differentiated effector cell fate, we have shown that the addition of ribociclib skews newly primed CD8 T cells toward a memory-like fate, thereby replenishing the stem-like progenitor cell pool.

A major strength of our methodology was the deployment of detailed scTCR clonotype tracking, an innovative expansion of scRNA-seq technology. Global increases in the human T cell memory pool do not occur with CDK4/6 inhibitors (Peuker et al., 2022), precluding use of bulk assays. Our work highlights the power of this platform, which enabled us to isolate specific subgroups of clonally expanded populations (PE vs. NE clones) that were predicted to be differentially affected by combination treatment. Parsing out distinct transcriptional signatures of treatment would have been impossible with traditional single-cell analyses in which cells are clustered and compared on the basis of their transcriptional similarity.

Our analysis was limited by low recovery of TILs, which likely caused us to miss some tumor-matched circulating clonotypes. While we speculate that memory skewed T cells in the blood would migrate into the tumor, we were limited by lack of adequate paired on-treatment biopsies to determine the fate of treatment-expanded clonotypes in the tumor itself. Additionally, some clonotypes that we label as potentially tumor-reactive may not be, given that definitive epitope determination is not possible by TCR sequencing alone. Furthermore, it is important to note that we limited our analytical focus to pre-existing knowledge of the signature effects of each treatment. Our dataset originated in an early, safety-oriented clinical trial which did not compare combination therapy to single-agent therapy controls. As such, we are unable to assess novel or unexpected effects that may arise in the combination treatment setting. Although our cohort included one patient with ovarian cancer who did not receive fluvestrant, we recognize that this is insufficient to exclude potential confounding effects of cancer type or administration of selective estrogen receptor degraders.

Despite these limitations, we believe that our observations motivate further study of combined but sequential therapy with CDK4/6 inhibitors and PD-1 blockade. While the introduction of immune checkpoint blockade has been a revolutionary step in the treatment of many cancers, only a minority of patients benefit. Our findings suggest that early exposure to CDK4/6 inhibitors may generate a critical pool of memory T cells to act as the foundation of subsequent anti–PD-1 responses. Since none of our patients achieved clinical benefit, we speculate that these differences in T cell biology may be even more pronounced in patients responding to therapy. Recent work supports the plausibility of this paradigm, as persistent neoantigen-reactive T cells that arise after administration of tumor vaccines appear to robustly reinvigorate with PD-1 blockade (Hu et al., 2021). Simultaneous combination therapy of CDK4/6 inhibitors with PD-1 or PD-L1 blockade have unacceptably high rates of hepatotoxicity; however staggering use of these regimens in sequential fashion offers a chance to avoid toxicities while still retaining much of the immune-potentiating effects of CDK4/6 inhibitors. Therefore, early sequential application of CDK4/6 inhibitors, perhaps in the neoadjuvant setting, followed by checkpoint inhibition could become a key strategy in increasing the fraction of the population responsive to PD-1 blockade.

Disclosures: A.C. Garrido-Castro reported “other” from Novartis during the conduct of the study; institutional research funding from AstraZeneca, Merck, and Gilead/Immunomedics; and non-financial support from Roche/Genentech outside the submitted work. M. Dougan reported personal fees from SQZ Biotech, Moderna, AzurRx, Palleon, Foghorn Therapeutics, Neoleukin Therapeutics, Cerberus, Veravas, Aditum, Genentech, Mallinckrodt, and Partner Therapeutics and grants from Eli Lilly outside the submitted work. In addition, M. Dougan receives research funding from Novartis, has received consulting fees from Tillotts Pharma and ORIC Pharmaceuticals, and is a member of the SAB for Neoleukin Therapeutics. S. Goel reported grants from Lilly during the conduct of the study; grants from Lilly, personal fees from Novartis, personal fees from Pfizer, and grants from G1 Therapeutics outside the submitted work. G.I. Shapiro reported grants from Pfizer and Eli Lilly during the conduct of the study; grants from Merck KGaA/EMD-Serono, Tango, Bristol Myers Squibb, and Merck & Co. and personal fees from Bicycle Therapeutics, Cybrexa Therapeutics, Bayer, Boehringer Ingelheim, ImmunoMet, Artios, Concarlo Holdings, Syros, Zentalis, CytomX Therapeutics, Blueprint Medicines, Kymera Therapeutics, Janssen, and Xinthera outside the submitted work. In addition, G.I. Shapiro had a patent to dosage regimen for sapacitabine and seliciclib issued “Cyclacel Pharmaceuticals and Geoffrey Shapiro” and a patent to compositions and methods for predicting response and resistance to CDK4/6 inhibition pending “Liam Cornell and Geoffrey Shapiro.” S.M. Tolaney reported personal fees from Genentech/Roche, Pfizer, Merck, SeaGen, Novartis, AstraZeneca, DSI, Eli Lilly, Eisai, CytomX, Blueprint Medicines, Mersana, Gilead, Ellipses Pharma, 4D Pharma, OncoSec Medical Inc, BeyondSpring Pharma, OncXerna, Zymeworks, Zentalis, Reveal Genomics, ARC Therapeutics, Infinity Therapeutics, Chugai Pharmaceuticals, Myovant, Zetagen, Umoja Biopharma, Menarini/Stemline, Aadi biopharma, Bayer, BMS, and Sanofi and grants from Genentech, Pfizer, Eli Lilly, Exelixis, Novartis, BMS, Sanofi, Merck, Gilead, Eisai, AstraZeneca, Cyclacel, SeaGen, Nanostring, and Nektar outside the submitted work. S.M. Tolaney also receives institutional research funding from Odonate and has served as an advisor/consultant to Nektar, Nanostring, Puma, Sanofi, Puma, Silverback Therapeutics, G1 Therapeutics, Athenex, OncoPep, Kyowa Kirin Pharmaceuticals, Samsung Bioepsis Inc., and Certara. S.K. Dougan reported grants from Novartis, BMS, grants from Eli Lilly, grants from Genocea, and personal fees from Kojin outside the submitted work. In addition, S.K. Dougan is a founder, science advisory board member (SAB) and equity holder in Kojin. No other disclosures were reported.