Immunomodulatory effects of PI3Kδ inhibition in solid tumors – evaluation in a randomized phase II trial

Phosphoinositide 3-kinase (PI3Kδ) this PI3K have been approved for hematological malignancies. While studies in hematological and solid tumor models in mice have demonstrated that PI3Kδ inhibitors (PI3Kδi) can induce anti-tumor immunity, the impact of PI3Kδi on solid tumors in humans remains unclear. Here, we assessed the effects of the PI3Kδi AMG319 in patients with resectable head and neck cancer in a neoadjuvant, double-blind, placebo-controlled randomised phase-II trial. We find that PI3Kδ inhibition decreases tumor-infiltrating immunosuppressive T REG cells and causes heightened cytotoxic potential of tumor-infiltrating CD8 + and CD4 + T cells. Loss of intratumoral T REG cells and an increase in the frequency of activated T REG cells in the blood post-treatment are indicative of systemic effects on T REG tissue retention and maintenance. At the tested AMG319 doses, immune-related adverse events caused treatment discontinuation in 12/21 of AMG319-treated patients, further suggestive of systemic effects on T REG cells. Consistent with this notion, in a murine syngeneic tumor model, PI3Kδi decreased T REG cells in both tumor and non-malignant tissues and affected T REG subtype composition, maintenance and functionality. Our data demonstrate the cancer-immunotherapy potential of PI3Kδ inhibition in humans, but its modulation will need to be carefully balanced to harness its anti-tumor capacity while minimizing immune related toxicity. placebo-controlled randomized phase II trial. AMG319 had an immunomodulatory impact that is remarkably consistent with preclinical data in mice and also aligns with immunomodulatory effects observed in human B-cell malignancies. Our data show that PI3Kδ inhibition in human tumor tissue modulates T cell activities in a manner consistent with enhanced anti-tumor immune responses. Surprisingly however, toxicity profiles (onset and frequency of irAEs) differed from a previous study of AMG319 in B cell malignancies 25 , suggesting that alternative dosing regimens will be required to effectively and safely exploit the immunomodulatory impact of PI3Kδ inhibition in human solid cancers. these T REG cells expressed higher levels of ICOS 42 , PD-1 43 , GITR 44 and 4-1BB 45 , markers indicating recent cell activation following antigen encounter, we hypothesize that PI3Kδ inhibition may displace activated tissue T REG cells into the circulation and leave the normal tissues vulnerable to inflammation, presumably triggered by the unrestrained activity of effector T cells. These observations are consistent with previous studies, implying that alterations of the PI3Kδ-Foxo1 pathway affect the trafficking of T REG cells by altering the expression of lymphoid homing molecules 7 . In support of this hypothesis, we find that PI3Kδ inhibition in mice causes systemic changes in the T REG cell compartment and differentially affects T REG subsets in different organ systems. Notably, we find substantial changes in the transcriptional features and composition of colonic T REG cell subsets, which indicate that PI3Kδ inhibition impacts T REG functionality, survival and tissue retention, thus altering T REG cell frequencies or T REG subtype compositions in both tumor and non-malignant tissues. Overall, our complementary studies in mice and humans highlight potential mechanisms that drive irAEs and anti-tumor activity of PI3Kδi. CD3 (SK7), CD127 (eBioRDR5), CD45 (HI30), CD14 (HCD14), CD20 (2H7); anti-mouse Cd3 (145-2C11), Cd4 (RM4-5), Cd8 (53-6.7), Ki67 (B56), Tox (REA473), Cd19 (6D5), Cd45 (30-F11), FoxP3 (FJK-16s), GzmB (QA16A02). All samples were acquired on a BD FACS Fortessa or sorted on a BD FACS Fusion (both BD Biosciences) and analyzed using FlowJo 10.4.1 for subsequent single-cell RNA-seq analysis.


Introduction
Among the PI3K isoforms, PI3Kδ is highly expressed in all leukocytes. PI3Kδ is critical for B-cell antigen receptor (BCR) signaling, facilitating B-cell activation, survival and proliferation, through activation of downstream kinases such as Protein Kinase B (AKT) and the mechanistic target of rapamycin (mTOR) complex 1 (reviewed in Ref. 1 ). PI3Kδ inhibitors (PI3Kδi), such as idelalisib, have been approved for the treatment of B-cell malignancies chronic lymphocytic leukemia (CLL) and follicular lymphoma, where the malignant cells are reliant on BCR signaling [2][3][4] . PI3Kδ also plays a role in T-cell differentiation, migration and function 5,6 by modulating T-cell antigen receptor (TCR) signaling and repression of FOXO transcription factors 1 , which in turn regulate genes involved in cell trafficking such as Klf2, S1pr1 and Ccr7 7 . Interestingly, T regulatory (TREG) cells seem more sensitive to PI3Kδ inhibition than other T cell subsets, such as cytotoxic CD8 + T cells, which appear largely unaffected in preclinical studies 8-10 and in idelalisib-treated CLL patients 11,12 . In B-cell malignancies, indirect evidence suggests that PI3Kδi-mediated immune-related adverse events (irAEs) such as colitis, hepatotoxicity and pneumonitis 12,13 may be related to the effects of PI3Kδ inhibition on TREG cells (reviewed in Ref. 4,14 ). Accordingly, an extensive body of preclinical and clinical studies has shown that disruptions of the PI3Kδ-Foxo1 pathway primarily affect TREG cells 5,10,11,15 by exerting effects that impact TREG cell proliferation 10 , survival 16 , suppressive capacity 17,18 and tissue residency 7 . These findings imply that the preferential inhibition of TREG cells by PI3Kδi can rebalance the host immune system to induce
To start evaluating the potential for PI3Kδ inhibitors as immunotherapeutic agents in human solid cancers, we administered the PI3Kδi AMG319 to treatment-naive patients with resectable head and neck squamous cell carcinoma (HNSCC) in a neoadjuvant, double-blind, placebo-controlled randomized phase II trial. AMG319 had an immunomodulatory impact that is remarkably consistent with preclinical data in mice and also aligns with immunomodulatory effects observed in human B-cell malignancies. Our data show that PI3Kδ inhibition in human tumor tissue modulates T cell activities in a manner consistent with enhanced anti-tumor immune responses. Surprisingly however, toxicity profiles (onset and frequency of irAEs) differed from a previous study of AMG319 in B cell malignancies 25 , suggesting that alternative dosing regimens will be required to effectively and safely exploit the immunomodulatory impact of PI3Kδ inhibition in human solid cancers.

PI3Kδ inhibition drives systemic TREG cell depletion
We initiated our studies by testing the impact of a PI3Kδi in a mouse solid tumour model, performing a more extensive T cell profiling than in any other study to date. C57BL/6 wild-type mice were inoculated with the syngeneic B16F10-OVA melanoma tumor cell line and treated with the previously described PI3Kδi PI-3065 7 . Consistent with previous studies 8, 9 , we found a significant decrease in tumor volume (Extended Data Fig. 1a), even if PI3Kδi treatment was initiated in a therapeutic setting once tumors had become palpable (Fig. 1a).
The reduction in tumor volume was accompanied by a significant decrease in intratumoral TREG cells (Fig. 1b). At the same time, intratumoral CD8 + T cells were significantly increased in frequency, expressed higher levels of PD-1 and exhibited higher proliferative capacity and cytotoxic potential (Fig. 1c-f). Tox, a transcription factor recently identified as critical for adaptation and survival of CD8 + T cells in the tumor microenvironment (TME) 26 , was also increased post-PI3Kδi (Fig. 1g). Notably and contrary to previous reports 27, 28 , we found that the expression of both GzmB and Ki-67 was almost exclusively limited to Tox + CD8 + T cells ( Fig. 1h), demonstrating that these cells, despite showing high expression of PD-1 and Tox, are not functionally exhausted in this tumor model. Thus, the PI3Kδi-induced reduction in intratumoral TREG cells correlated with augmented anti-tumor CD8 + T cell responses. To assess whether PI3Kδ inhibition acts locally within the tumor tissue or systemically i.e., also affecting other non-malignant organs, we performed flow-cytometric analyses of TREG cells in 5 spleen, tumor and colonic tissue of tumor-bearing FoxP3-RFP reporter mice. Importantly, in PI3Kδi-treated mice, but not placebo-treated control mice, we found a significant decrease in TREG cells in all assessed tissues, indicative of systemic effects of PI3Kδi on TREG maintenance or survival (Extended Data Fig. 1b).

Colonic TREG cells are sensitive to PI3Kδ inhibition
Since colitis is one of the major irAEs in patients receiving PI3Kδi 4,12,13 , we hypothesized that TREG cells present in colonic tissue may be especially sensitive to PI3Kδi.
To test this hypothesis in an unbiased manner, we performed single-cell RNA-sequencing of TREG cells isolated from tumor, spleen (lymphoid organ) and colonic tissue of PI3Kδi-and placebo-treated B16F10-OVA tumor-bearing FoxP3-RFP reporter mice. Unbiased clustering analysis depicted by uniform manifold approximation and projection (UMAP) analysis identified 10 TREG cell clusters, implying substantial TREG cell heterogeneity and tissue adaptations ( Fig. 2a-c). In agreement with previous studies 29,30 and based on their unique transcriptomic signatures, tissue TREG cells (tumor, colon) clustered distinctly from one another and from lymphoid organ TREG cells (spleen), suggesting the existence of several distinct TREG subtypes in different locations ( Fig. 2a and Extended Data Fig. 2a,b). Colonic TREG cells exhibited the most pronounced differences between PI3Kδi and placebo treatment, with 869 differentially expressed genes, while splenic and tumor TREG cells exhibited fewer differences ( Fig. 2d-f and Extended Data Table 1). We also found major changes in the composition of colonic TREG subsets, but not splenic or intratumoral TREG subsets (Fig. 2b,c and Extended Data Fig. 2c). Two of the colonic TREG subsets (clusters 2 and 8) were depleted in PI3Kδitreated mice (Fig. 2b,c and Extended Data Fig. 2c). Cluster 2 colonic TREG cells were enriched for the expression of Ctla4 and genes encoding chemokine receptors (Ccr1, Ccr2, Ccr4), critical for their suppressive 31,32 and migratory 33 capacity, respectively (Fig. 2g, Fig.2d and Extended Data Table 2). Cluster 8 colonic TREG cells resembled the recently described tissue-resident ST2 TREG cells [34][35][36] , which are critical for the protection against chronic inflammation and facilitation of tissue repair. Accordingly, we found enrichment 6 in the expression of the ST2 TREG signature genes Il1rl1 (encoding for IL-33R), Gata3 and Id2, as well as of several genes associated with highly suppressive effector TREG cells (Klrg1, Cd44, Cd69, Pdcd1, Areg, Nr4a1, Il10 and Tgfb1) ( Fig. 2g and Extended Data Table 2). While colonic TREG cells in cluster 0 and cluster 8 shared this ST-2 signature, cells in cluster 8 were also enriched for transcripts linked to cellular activation (Cd44, Icos and Klrg1) and superior suppressive capacity (Ctla4, Il10 and Gzmb) (Fig. 2h). While the TREG cell clusters (2 and 8) with highly suppressive properties were depleted in PI3Kδi-treated mice, cluster 5 TREG cells, which were enriched in PI3Kδi-treated mice, showed higher expression of several interferonrelated response genes (Stat1, Stat3, Ifrd1) 37,38 , suggestive of a pro-inflammatory environment ( Fig. 2g and Extended Data Table 2). Further corroborating the diverse effects of PI3Kδ inhibition on TREG cells in different tissues, we observed a significant increase of CD8 + T cells in colonic, but not splenic tissue (Extended Data Fig. 2e,f). Moreover, these colonic CD8 + T cells expressed higher levels of PD-1 and ICOS upon PI3Kδ inhibition (Extended Data Fig.   2g,h), implying treatment-related changes to cell activation, presumably mediated by the TREG cell alterations. Taken together, these findings suggest a heightened sensitivity of certain colonic TREG subsets to PI3Kδi, potentially related to the high incidence of colitis observed in patients treated with PI3Kδ inhibitors such as idelalisib.

Systemic effects of PI3Kδ inhibition in HNSCC patients are associated with the occurrence of irAEs
In order to explore how our findings above translate to humans, we conducted a multicenter, placebo-controlled phase II neoadjuvant trial with the PI3Kδi AMG319 in resectable HNSCC, with primary study endpoints of safety and changes in the density of CD8 + T cells ( Fig. 3a and https://www.clinicaltrialsregister.eu/ctr-search/trial/2014-004388-20/results). We focused on Human Papilloma Virus (HPV)-negative HNSCC, as this cancer type is more prevalent, and because patients with this cancer type have poorer outcomes when compared to HPV-positive HNSCC, likely due to overall lower tumor infiltrating lymphocyte (TIL) infiltration [39][40][41] . Patients were recruited after initial diagnosis and before definitive surgical treatment (Extended Data Table 3); drug treatment or placebo was given for up to 24 or 28 days respectively, prior to resection of tumor.
In a previous phase I dose escalation study of heavily-pretreated patients with either CLL or non-Hodgkin lymphoma, AMG319 doses of up to 400mg were explored without reaching a maximally-tolerated dose, and exhibited PK dynamics with a mean half-life of 3.8-6.6h in plasma 25 . In that phase 1 study, daily dosing with 400mg AMG319 led to near complete target inhibition (BCR-induced pAKT in ex vivo IgD-stimulated CLL samples) and >50% nodal regression 25 , while immune related adverse events (irAE) at grade 3 or above according to the common toxicity criteria (CTC) occurred after days 40 and 60. We thus reasoned that high grade irAEs were unlikely to occur during the shorter treatment duration in the neoadjuvant setting (Extended Data Fig.3a), and therefore selected 400 mg/day as the starting dose.
The intended time from initiating treatment with AMG319 or placebo to surgical resection of tumor was up to 4 weeks, with weekly blood draws. Thirty-three patients were randomized in a 2:1 ratio (AMG319:Placebo) to the trial and 30 patients received at least one dose of AMG319 or placebo. Fifteen patients received 400 mg daily of AMG319 (range of 7-24 days per patient). Unexpectedly, at the 400 mg dose, 9/15 patients experienced irAEs that lead to withdrawal of treatment (Extended Data Table 3). After a formal safety review, 6 additional patients were recruited and treated at a reduced dose of 300 mg per day. Again, 3/6 patients had irAEs that led to discontinuation of treatment (Extended Data Table 3) although the planned sample size of 54 randomized patients had not been reached. One patient experienced grade 4 colitis after completion of 24 daily doses of AMG319, and eventually required colectomy; the study was discontinued at this point. Overall, 9 patients were randomized to the placebo arm (Extended Data Fig. 3a), of these, one experienced a serious adverse event (grade 4 post-surgical infection). We measured target inhibition (pAKT levels in B cells; Fig. 3b) and drug levels ( Fig. 3c) to verify drug administration (Extended Data Table 3). Pharmacokinetic (PK) analysis indicated peaking drug levels on day 8; 8/10 patients who discontinued treatment did so between day 7 and 9 (Extended Data Table 3), which resulted in loss of detectable drug in the PK analysis on day 15 (Fig. 3c). As expected, 8 AMG319 treatment caused a significant decrease in the pAKT levels in B cells 1 4h posttreatment on day 1 and day 15 ( Fig. 3b and Extended Data Table 3). Interestingly, modulation of PI3Kδ did not appear to affect humoral IgG responses to vaccination with the recall antigen tetanus toxoid (Fig. 3d), while we observed few tetanus-specific T cells in either the AMG319 or placebo treated patients (Extended Data Table 3), pointing to an overall degree of functional immunocompromise in our patient cohort.
Clinically, and most likely reflecting the brief treatment period, we did not observe any significant differences in the measured tumor volumes between the study arms in the 23 patients in whom this was evaluable. Two partial responses (PR) and one pathological complete response occurred in AMG319-treated patients (Extended Data Fig. 3b); one PR was observed in the control group. The full evaluation of radiological measurements has previously been reported at https://www.clinicaltrialsregister.eu/ctr-search/trial/2014-004388-20/results to the EU Clinical Trials Register in compliance with regulatory requirements.

PI3Kδ inhibition displaces activated TREG cells into circulation
To investigate whether PI3Kδi, akin to our findings in preclinical models ( Fig. 2 and Extended Data Fig.2), affects tissue-residency and functionality of TREG cells, we assessed the frequency and phenotype of circulating TREG cells over the course of the treatment. We observed a significant treatment-related increase in the percentage of circulating TREG cells (CD4 + CD25 + CD127 lo ) (Fig. 4a), while the proportion of TREG cells in the placebo group remained stable (Extended Data Fig. 3c). In addition, we observed a significant increase in the expression levels of ICOS 42 , PD-1 43 , GITR 44 and 4-1BB 45 , markers of cell activation and recent antigen encounter, in circulating TREG cells (Fig. 4b). Together, these data indicate that PI3Kδi treatment can displace activated TREG cells from tissues, including tumor tissues, into the circulation, presumably by altering the expression of tissue homing factors like KLF2 and S1PR1, direct targets of FOXO1. In line with these findings, we also observed frequent, and in a few patients, severe grade 3/4 irAEs in the AMG319-treated patients (Extended Data Table 3). Of the 91 total irAEs we observed, 85 were in the AMG319 treatment group and 6 in the placebo group; all 23 grade 3 and 1 grade 4 irAEs could be attributed to AMG319 treatment (Fig. 4c). The most prevalent irAEs were skin rashes (29%; 25% observed in the treatment group and 4% in placebo group), diarrhea (29%; 28% in observed treatment group and 1% in placebo group) and transaminitis (14% all of which were observed in the treatment group), consistent with a loss of TREG cells or TREG cell functionality in multiple tissues causing immunopathology ( Fig. 2c and Extended Data Table 3). Unlike the phase 1 dose-escalation study of AMG319 in heavily pretreated patients with either CLL or non-Hodgkin lymphoma 25 , the onset of irAEs was surprisingly rapid and led to treatment discontinuation in 12/21 AMG319-treated patients.

Intratumoral CD8 + T cells exhibit signs of heightened cytotoxic potential
To assess the impact of PI3Kδi treatment on anti-tumor immune responses, we first performed bulk RNA-seq in pre-and post-treatment tumor samples. Differential gene expression analysis revealed substantial differences in the AMG319-treatment group (93 differentially-expressed genes (DEGs)), but not for the placebo group (3 DEGs) (Fig. 5a, b and Extended Data Table 4). PI3Kδ-inhibition disrupted expression of genes in the PI3K  Table 4). The significant downregulation of angiogenesis and vascular development-related genes suggests a potential of PI3Kδ inhibition to reduce neovascularization in tumors and thus curb tumor growth, potentially mediated by the depletion of intratumoral TREG cells, which has been shown to promote angiogenesis 46 . In the AMG319-treatment group, we also found significantly higher expression of GZMA, and a trend (p=0.06) towards higher expression of PRF1 (encoding for Perforin-1) in post-treatment tumor samples (Extended Data Fig. 4e,f), implying an increase in the number or activity of cytotoxic cells. We corroborated these results with bulk-RNA-seq analysis of sorted tumor-infiltrating CD8 + T cells from 2 patients in whom paired tumor samples were available pre and post PI3Kδi treatment (Fig. 5c). We found higher expression of IFNG, GZMB and PRF1 in post-treatment samples, which indicated enhanced cytotoxic potential of tumor-infiltrating CD8 + T cells following PI3Kδi treatment, consistent with our murine data (Fig. 1e).
To contextualize our data, we compared expression levels of CD8A as a measure of TIL infiltration with data from a previous HNSCC patient cohort (Extended data table 4) 40 .
Our analysis revealed low baseline TIL infiltration in the majority of the samples in our trial ( Fig. 5d), consistent with the notion that HPV-negative cancers are predominantly immunecold tumors. These data, as well as the intermittent dosing of AMG319 in a neoadjuvant setting, also contribute to our understanding as to why a primary efficacy endpoint of the study, a doubling in CD8 + TILs assessed by IHC, was not met. AMG-319 nonetheless further reduced the TREG cell frequency in these tumors (Extended Data Fig. 4b,c) and increased expression of markers associated with cytotoxic function, consistent with our murine data (Fig.   1e).

PI3Kδ inhibition drives oligoclonal T cell expansion
To assess the consequences of PI3Kδ-inhibition on anti-tumor immunity in more detail, we next performed combined single-cell RNA-seq and TCR-seq analysis of tumor-infiltrating CD3 + T cells from 6 patients with available pre-and post-treatment tumor samples. UMAP analysis demonstrated separation of CD4 + from CD8 + T cells (Fig. 6a). Single-cell differential gene expression analysis of the CD4 + and CD8 + clusters showed a treatment-related increase in expression of cytotoxicity genes (e.g. GZMB and PRF1) (Extended Data Table 5, Fig. 5b), in line with our previous results ( Fig. 1e and Fig. 5c) and implying that tumor-infiltrating CD4 + and CD8 + T cells exhibit enhanced cytotoxic properties following PI3Kδi treatment.
Accordingly, Ingenuity Pathway Analysis (IPA) of the differentially-expressed genes in CD8 + T cells identified CSF2 (GM-CSF), a pro-inflammatory mediator and indicator of tissue inflammation that can enhance the generation of CTLs 47,48 , IL-2 and TCR as the top 3 upstream regulators (Extended Data Table 6), indicative of pro-survival IL-2 signalling and TCR activation. Crucially, and in line with these findings, our single-cell TCR-seq data demonstrate substantial clonal expansion of CD8 + T cells, and to a lesser degree of CD4 + T cells, post-treatment (Fig. 6c,d). Several, but not all, of these clonally-expanded CD8 + T cells were also present in pre-treatment samples, suggesting that PI3Kδ inhibition drives the expansion of both new and pre-existing T cells. Together, these data indicate that PI3Kδ inhibition causes profound changes in the TME, characterized by enhanced CD4 + and CD8 + T cell activation, oligoclonal T cell expansion and increased cytolytic activity, consistent with a decrease in intratumoral TREG cells, ensuing T cell activation.

Discussion
PI3K inhibitors were initially considered to mainly target cancer cell-intrinsic PI3K activity, which was the underlying rationale to test inhibitors against the leukocyte-enriched PI3Kδ in hematological malignancies. However, subsequent studies have shown that PI3Kδ inhibition also has clear immunomodulatory activities, largely T-cell mediated, which were under-appreciated at the time of the early trials in B-cell malignancies, causing irAEs that have hampered clinical progress and utility. Several lines of evidence suggest that PI3Kδi preferentially inhibit TREG cells over other T cell subsets 8,9,11,12,14 but to date, no trials have been performed to explicitly explore this concept in humans. Our current study provides the first in-depth investigation on the immune impact of PI3Kδ inhibition in patients with solid tumors.
We find that in the tumor tissue, PI3Kδ inhibition leads to substantial changes in the cell composition of the TME by reducing TREG cells and activating intratumoral CD4 + and CD8 + T cells, which clonally expand and display heightened cytotoxic and cytolytic features.
However, in the circulation, paradoxically, we found increased number of TREG cells. Because these TREG cells expressed higher levels of ICOS 42 , PD-1 43

Author contribution and acknowledgements
SE: study design, experimental work, data interpretation, paper writing; CR-S: bioinformatic evaluation, data interpretation, paper review; EK: study design, patient recruitment, paper review; LC: data collation, data interpretation, paper writing and review; JT: bioinformatic evaluation, data interpretation, paper review; OW: experimental work, data interpretation, paper review; AvW: experimental work, data interpretation, paper review; DJ: experimental work, data interpretation, paper review; KMC: experimental work, data interpretation, paper review; HS, MM, AW: experimental work, paper review; EL-G: experimental work, provision of study materials, data interpretation, paper review YL: experimental work, paper review; NAD, LE, FK: study conduct, safety data review, and monitoring, data review and verification for sponsor GA: study design, safety review, data review GH: generation and provision of placebo and IMP JJS, AGS, RS, JAMC, CP, JHD, PB, RPS: patient recruitment, paper review PL: data generation and interpretation, paper review WW: study design, statistical review for sponsor AH: study design, statistical review GJT: histopathological evaluation, data generation and interpretation, paper review TMJ: study development, patient recruitment, paper review FA, bioinformatic analyses and supervision, data review, paper review KO: study design, paper writing and review BV: study design, paper writing and review *PV: study design, data generation and review, paper writing and review.

Mice
C57BL/6J mice were obtained from the Jackson labs. FoxP3 RFP mice were a gift from Stephen Schoenberger (La Jolla Institute for Immunology). All mice were between 6-12-week-old at the beginning of the experiment;. the animal work was approved by the relevant La Jolla institute for Immunology Animal Ethics Committee. B16F10-OVA cells were a gift from Prof. Linden (La Jolla Institute for Immunology). Cell lines tested negative for mycoplasma infection and were treated with Plasmocin to prevent contamination.
Mice were put on either a control diet or a diet containing a PI3Kδ inhibitor PI-3065 on day 1 or day 5 after tumor inoculation. Diets were prepared using powdered 2018 global rodent diet (Envigo) mixed with or without PI-3065 at 0.5 g/kg, which corresponds to a daily dose of 75 mg/kg as used in our previous study 2 . To pellet the food, 50% v/w water was added to the diet and dough thoroughly mixed, compressed, moulded and dried before use. Tumor size was monitored every other day, and tumor harvested at indicated time points for analysis of tumorinfiltrating lymphocytes. Tumor volume was calculated as ½ x D x d 2 , where D is the major axis and d is the minor axis, as described previously 49 .

Flow cytometry
Cells dispersed from cryopreserved tumor tissue or PBMCs isolated from peripheral blood were prepared in staining buffer (PBS with 2% FBS and 2 mM EDTA), FcR blocked (clone 2.4G2, BD Biosciences) and stained with antibodies as indicated below for 30 min at 4°C. Cell viability was determined using fixable viability dye (ThermoFisher).

Single-cell transcriptome analysis
For human tumor, single-cell RNA-seq was performed by Smart-seq2 as previously described 51 . Reads were mapped with our in-house pipeline as above. Good quality cells were defined as those with at least 200 genes, at least 60 percent of mapping reads, mitochondrial counts of at most 20 percent, at least 50,000 total counts (reported by STAR excluding tRNA and rRNA), and a 5' to 3' bias of at most 2. Filtered cells were analyzed using the package Seurat (v3.1.5). In order to separate CD4 and CD8 more effectively, we performed Differential gene expression analysis between single-positive cells using CD4 and CD8B genes. Cells were clustered using 178 significant genes (adjusted P-value < 0.5).
For murine single-cell RNA-seq, we used the 10x platform (10x Genomics, Pleasanton, CA, USA) according to the manufacturer's instructions. Reads were mapped with Cell Ranger followed by our in-house QC pipeline (https://github.com/vijaybioinfo/quality_control) and demultiplexed with bcl2fastq using default parameters. Cell Ranger aggr routine was used and CITE-seq data was processed using our custom pipeline (https://github.com/vijaybioinfo/ab_capture). Briefly, raw output from Cell Ranger is taken and cell barcodes with less than 100 UMI counts as their top feature are discarded, the remaining barcodes are classified by MULTIseqDemux from Seurat. Finally, cell barcodes where the assigned feature doesn't have the highest UMI count are fixed, and cells with a fold change of less than 3 between the top two features are reclassified as doublets. Before clustering, cells were filtered for at least 300 at most 5000 genes, at least 500 and at most 10000 UMI counts, and at most 5 percent of mitochondrial counts. Cell types were identified using Seurat's FindAllMarkers function. Differential expression was calculated with MAST 52 (v1.10.0) DESeq2 (v1.24.0) as previously described 51 and genes with an adjusted P-value < 0.05 and a fold change of > log2 0.5 were defined as significant. GSEA scores were estimated with fgsea (v1.10.1) in R using signal-to-noise ratio as the metric (minSize = 3 and maxSize = 500). Enrichment scores were shown as GSEA plots. Signature scores were computed using Seurat's AddModuleScore function with default parameters. In short, the score is defined for each cell by subtracting the mean expression of an aggregate of control gene lists from the 31 mean of the signature gene list. Control gene lists were randomly selected (same size as the signature list) from bins delimited based on the level of expression of the signature list.

T-cell receptor analysis
TCR were reconstructed from single-cell RNA-seq reads using MiXCR with default parameters. Then, shared TCR were defined by having the same CDR3 sequence in both the alpha and beta chains and coming from the same donor. Enriched TCR were defined as those with a frequency higher or equal to two. Lastly, TCR network plots were generated using the Python package graphviz.

Accession codes
Sequencing data has been uploaded onto the Gene Expression Omnibus (accession code GSE166150). Access for reviewers can be found using the password utqpmasoplopdwz

Quantification and statistical analysis
The number of subjects, samples or mice/group, replication in independent experiments, and