Neoantigen specific CD4⁺ T cells in human melanoma have diverse differentiation states and correlate with CD8⁺ T cell, macrophage, and B cell function

Summary:

CD4⁺ T cells that recognize tumor antigens are required for immune checkpoint inhibitor efficacy in murine models but their contributions in human cancer are unclear. We used single cell RNA sequencing and T cell receptor sequences to identify signatures and functional correlates of tumor specific CD4⁺ T cells infiltrating human melanoma. Conventional CD4⁺ T cells that recognize tumor neoantigens express CXCL13 and are subdivided into clusters expressing memory and T follicular helper markers, and those expressing cytolytic markers, inhibitory receptors, and IFN-γ. The frequency of CXCL13⁺ CD4⁺ T cells in the tumor correlated with the transcriptional states of CD8⁺ T cells and macrophages, maturation of B cells, and patient survival. Similar correlations were observed in a breast cancer cohort. These results identify phenotypes and functional correlates of tumor specific CD4⁺ T cells in melanoma and suggest the possibility of using such cells to modify the tumor microenvironment.

Graphical Abstract

Introduction

Clinical responses to immune checkpoint blockade (ICB) are thought to be mediated in part by T cell recognition of neoantigens that are derived from cancer specific mutations and presented on major histocompatibility complex (MHC) molecules (Gubin et al., 2014; Le et al., 2015; Rizvi et al., 2015). Prior work characterizing the antigen specificity and transcriptional states of T cells in human tumors has primarily focused on class I MHC restricted CD8⁺ cytotoxic T cells. Tumor infiltrates contain heterogeneous populations of CD8⁺ T cells, some of which are specific for tumor antigens and exist in varying states of functional exhaustion (Caushi et al., 2021; Oliveira et al., 2021)(Siddiqui et al., 2019), and many phenotypically distinct “bystander” CD8⁺ T cells specific for viral or other antigens (Duhen et al., 2018; Scheper et al., 2019; Simoni et al., 2018). CD4⁺ T cells specific for neoantigens including oncogenic driver mutations presented by class II MHC have also been identified in cancer patients (Linnemann et al., 2015; Ott et al., 2017; Sahin et al., 2017) (Cafri et al., 2019; Veatch et al., 2019; Veatch et al., 2018; Yossef et al., 2018). In murine models, CD4⁺ neoantigen-specific T cells residing within tumors are required for response to ICB (Alspach et al., 2019), and may mediate tumor rejection through direct destruction of tumor cells (Quezada et al., 2010), activation of innate immune cells (Mumberg et al., 1999; Tveita et al., 2016), and stimulation of CD8⁺ T cells (Ossendorp et al., 1998). CD4⁺ T cells expanded from bladder cancers are capable of lysing autologous tumor (Oh et al., 2020), and anecdotal reports in patients have shown that the adoptive transfer of CD4⁺ T cells alone can have marked antitumor effects (Hunder et al., 2008; Tran et al., 2014) suggesting an important role of this subset in immunity to human cancer.

Single cell sequencing of CD4⁺ T cells from primary human tumors has identified multiple subpopulations, including regulatory T cells, T cells expressing cytolytic proteins, and cells sharing markers with T follicular helper cells (T_FH) (Bassez et al., 2021; Bonnal et al., 2021; Oh et al., 2020). Tertiary lymphoid structures containing B cells and CD4⁺ T cells of unknown specificity, and a B cell transcriptional signature have been associated with response to ICB in melanoma, suggesting functional contributions of some subsets of CD4⁺ T cells or B cells (Cabrita et al., 2020; Helmink et al., 2020; Voabil et al., 2021). However, the phenotype, heterogeneity, and functional roles of the CD4⁺ T cells that infiltrate tumors and are specific for tumor antigens have not been defined beyond enrichment in fractions of cells that express PD-1 or CD39 (Balança et al., 2021; Kortekaas et al., 2020; Yossef et al., 2018). Here, we identified the specificity of CD4⁺ T cells for tumor antigens in patients with melanoma and then used single cell sequencing and matching of TCR clonotypes to characterize their transcriptional signatures in tumor infiltrates. Our data show that CXCL13 expression identifies three distinct cell states of tumor reactive CD4⁺ T cells in melanoma. The frequency of tumor infiltrating CXCL13⁺ CD4⁺ T cells correlates with differences in frequency and functional states of CD8⁺ T cells, myeloid cells and B cells in the tumor microenvironment.

Results:

Identification of neoantigen-specific CD4⁺ T cell TCRVb clonotypes from melanoma tumors.

Matching of information obtained by scRNA sequencing data to recognition of tumor antigens by individual T cells infiltrating tumors requires identification of clonal TCRαβ sequences and their antigen specificity. To this end we performed whole exome sequencing on tumor and normal cells and RNA sequencing of tumor from 4 patients to identify and rank prevalent nonsynonymous mutations that could serve as tumor neoantigens. We then selected the 30 to 45 highest expressed mutations in each tumor to screen for recognition by the patients’ CD4⁺ T cells. For each candidate neoantigen, 20-mer peptides with the variant amino acid at either position 7 or 13 were synthesized for screening (Table S1) and purified 27-mer peptides with the mutation at position 14 were used for confirmation.

Because isolation of tumor-specific T cells from TIL or blood is challenging, we employed multiple approaches to identify and validate TCRVb sequences of neoantigen-specific CD4⁺ T cells (Figure S1A). PD1^high CD4⁺ T cells, which have been shown to be enriched for tumor reactivity (Yossef et al., 2018), were sorted from tumor infiltrating lymphocytes (TIL) of patients X205 and X422 (Figure 1A), expanded in limiting dilution cultures to isolate individual clones, and screened for IFN-γ secretion in response to stimulation with pools of peptides comprising candidate neoantigens arranged in an analytic grid where each peptide was represented in two pools (Figure S1B–C). T cell cultures with reactivity for mutant peptides were identified from each patient (Figure S1D–E), specificity for the mutant versus wildtype peptide was confirmed (Figure 1B–C, Figure S1 F–G), and TCRVb sequencing performed to identify unique clonal TCRs (Table S2). Additional tumor-reactive CD4⁺ T cells were isolated from X205 TIL by stimulating bulk ex-vivo expanded PD-1^high T cells in duplicate with peptide pools comprised of tumor associated self-antigens (Tyrosinase, Mart1, TRP2, GP100, SSX2, NY-ESO, MAGE A1 and MAGE A3) and with candidate neoantigens and corresponding wild type sequences, and then sorting CD4⁺ T cells that expressed CD40L after antigen restimulation and performing TCRVb sequencing (Figure 1D). TCRVb clonotypes that were present in CD40L sorted T cells from both replicates were inferred to be antigen reactive. This approach identified 5/9 previously identified TCR clonotypes and an additional 11 TCR clonotypes (Figure 1E, Table S2). Upregulation of CD40L was not observed after stimulation of PD-1^intermediate and PD-1^low CD4⁺ T cells from patient X205 with the neoantigen pool, but these T cells upregulated CD40L after stimulation with a panel of viral antigens and 23 viral specific TCRVb clonotypes were identified to be enriched in these cultures (Figure S2A–B). Stimulation of the PD-1^high subset of CD4⁺ T cells in patient X422 with neoantigens but not viral antigens induced IFN-γ in a subset of T cells and the converse was observed after stimulation of the PD-1^low subset (Figure S2C). These observations confirmed restriction of tumor antigen specific CD4⁺ T cells in these patients to the PD-1^high fraction.

Figure 1. Identification of neoantigen specific TCRVb clonotypes from melanoma patients.

A. FACS sorting of CD3⁺ CD4⁺ T cells based on PD-1 expression in patients X205 and X422. B-C. IFN-γ secretion by T cell lines expanded in limiting dilution cultures from intratumoral PD-1^high CD4⁺ T cells from patient X205 (B) and X422 (C) after incubation with 1μM 27-mer peptide containing the mutant or wildtype amino acid at position ⁺14. The dominant TCR Vb clonotype present in each culture was identified by TCR Vb sequencing. Horizontal lines denotes the median. D-E. T cell cultures expanded in bulk from intratumoral PD-1^high CD4⁺ T cells from patient X205 were incubated with indicated pools of peptides and stained for expression of CD40L and CD4 (D). CD40L⁺ CD4⁺ cells were isolated by FACS, TCRVb genes in sorted CD40L positive T cells were identified by sequencing, and the percentage of selected clonotypes (rows) found in replicate assays relative to the total across all peptide stimulations (columns) are shown (WT= wildtype peptide) (E). F. IFN-γ release by T cells lentivirally transduced with synthetic TCR sequences corresponding to the clonotypes that expanded with BRAF peptide stimulation after incubation with the indicated concentrations of mutant and wildtype 27-mer BRAF peptide. Data shown as the mean plus or minus the standard error of the mean. G. PBMC from patient X198 was stimulated with a pool of mutant peptides and restimulated on day ⁺13 with individual 27-mer peptides containing the wildtype or mutated amino acid at position ⁺14 and the frequency of IFN-γ secreting CD4⁺ T cells was measured by Elispot. H. Duplicate parallel cultures of PBMC from patient X198 were stimulated with individual mutant peptides and cultured for 13 days followed by TCRVb sequencing of enriched CD4⁺ T cells. The percentage of TCRVb templates (rows) seen in replicate cultures stimulated with a single mutant peptide (columns) relative to other stimulations are shown. I. Summary of the total number of TCRVb clonotypes specific for neoantigens, tumor self-antigens and viral antigens that were identified from each of the 4 patients. N/A indicates no such clones were identified. N/E indicates not evaluated.

See also Figures S1 and S2 and Tables S1 and S2.

Patients X197 and X198 had scarce primary tumor samples, but donated leukapheresis products following adoptive immunotherapy with autologous TILs. Because tumor-reactive T cells are often enriched in blood after TIL infusion (Rosenberg et al., 2011) PBMCs obtained from the apheresis after the TIL infusion were stimulated with pools of candidate neoantigen and self antigen peptides (Figure S1A). We previously isolated CD4⁺ T cells from PBMC of patient X197 that recognized the BRAF V600E mutation (Veatch et al., 2018), and identified 3 different TCR clonotypes that were expanded following stimulation with BRAF V600E peptide in duplicate cultures, while cultures stimulated with pools of self antigens did not expand any clones in duplicate cultures (Figure S2D). Each of these 3 TCRa/b clonotypes were synthesized as a synthetic TCR in a lentiviral construct and conferred specificity to BRAF V600E following gene transfer (Veatch et al., 2018) (Figure 1F). None of the 3 BRAF V600E-specific clonotypes were detectable following peptide stimulation of PBMC isolated prior to TIL treatment (Veatch et al., 2018). This validation supports the use of TCR sequencing of T cells responding in duplicate to neoantigen peptide stimulations as a reliable method of identifying antigen specific TCR clonotypes in patients following TIL treatment. CD4⁺ T cells expanded by stimulation of PBMC from patient X198 with candidate neoantigens reacted with 7 different mutated peptides and did not recognize the corresponding wild-type peptides (Figure 1G, Figure S2E). The TCRVb clonotypes responsible for recognition of each neoantigen in X198 were inferred by identifying TCRVb sequences that selectively expanded in replicate cultures of PBMC stimulated with individual purified mutated peptides (Figure 1H) (Danilova et al., 2018).

To assist with analysis of potential bystander T cells that might be present in TIL, PBMC from patients X197 and X198 were also stimulated with viral antigens and expanded TCRVb clonotypes in these cultures were identified by sequencing (Figure S2D, S2F). TCRB sequencing of tumor and pre TIL infusion blood samples from X197 and X198 showed that TCR sequences corresponding to self and neoantigen specific T cells but not virus specific T cells were enriched in tumor relative to blood (Figure S2G–H). Virus specific clonotypes were not isolated from patient X422 and no self antigen specific TCRVb clonotypes were identified from patients X198, X197 or X422. In total, we identified 66 different TCRVb clonotypes specific for 14 tumor neoantigens from the 4 patients, 15 TCRVb clonotypes specific for nonmutated tumor antigens from patient X205, and 326 TCRVb clonotypes specific for viral antigens from patients X197, X198 and X205 (Figure 1I). Immunogenic neoantigens were expressed at higher levels than non-immunogenic candidate antigens that were screened, but did not have higher clonality within tumors (Figure S2I–J).

Phenotype of tumor antigen specific CD4⁺ T cells in melanoma

Single cell suspensions from tumor samples of the 4 patients were stained with fluorescent antibodies to CD4 and CD3 and a panel of 34 oligonucleotide labelled antibodies, and CD3⁺ CD4⁺ T cells were sorted and analyzed by single cell targeted mRNA sequencing of 400 immune response genes and TCR VDJ rearrangements (Ma et al., 2021). Tumor samples from X197 and X198 prior to TIL infusion were used for this analysis. BRAF V600E specific T cells were rare in the tumor of patient X197 and all clones were Vbeta 3.1, therefore we enriched 50% of the CD4⁺ T cells for Vbeta 3.1 in this patient prior to sequencing to provide greater coverage of this tumor specific population. In total, 10,186 tumor infiltrating CD4⁺ T cells were sequenced from the 4 patients. Unsupervised clustering of these cells by mRNA expression defined a cluster of FoxP3⁺ regulatory T cells (T_REG), 3 clusters of CXCL13⁺ non- T_REG conventional T cells (T_CONV) that also expressed surface PD-1, and 2 clusters of CXCL13⁻ T_CONV that expressed high levels of IL7R mRNA (Figure 2A–C). 259 cells (17–146 per patient) expressed 40 different TCR clonotypes (2–18 per patient) specific for neoantigens, 108 cells from patient X205 expressed 14 TCR clonotypes specific for self antigens, and 9 cells from patients X197 and X198 expressed 5 TCR clonotypes specific for viral antigens (Figure 2D). Strikingly, alignment of TCRVb sequences placed 97% of neoantigen specific (range 93–100%) and 99% of self antigen specific T cells in the CXCL13⁺ clusters, and all 9 virus specific T cells from 2 patients in the CXCL13⁻ T_CONV cluster (Figure 2E–F, Figure S3A). All 91 cells that expressed one of the 14 neoantigen specific TCRs identified from T cell clones or verified to be neoantigen specific by TCR gene transfer in 3 of the 4 patients showed a CXCL13⁺ signature. TCR clonotypes present in two or more cells in the tumor or enriched in tumor relative to blood were more common in the CXCL13⁺ T_CONV and T_REG clusters than in the CXCL13⁻ T_CONV cluster (Figure 2G–H).

Figure 2. Tumor antigen specific conventional CD4⁺ T cells exhibit transcriptional signatures distinct from Treg and bystander conventional CD4⁺ T cells.

A-B. Unsupervised clustering of 10,186 CD4⁺ T cells from primary tumor samples of 4 patients visualized by UMAP with cells labelled by patient (A) and phenotypic clusters (B). C. Expression of FOXP3, CXCL13, LAG3, TCF7, GZMA, TYMS, IL7R and IFNG mRNA, and PD-1/CD279, Tim3, CXCR5, CD103, CD127 and CD39 surface protein in tumor infiltrating CD4⁺ T cells is shown with color reflecting log₂ of expression. D. Summary of antigen specific CD4⁺ T cells identified in the tumor infiltrate by TCRVb sequence. E. Cells with TCRVb clonotypes specific for neoantigens and self-antigens are shown. F. Proportion of neoantigen, self-antigen and viral antigen specific cells in CXCL13⁺, CXCL13⁻, and FOXP3⁺ clusters. G-H. The number of cells with each clonotype (G) and the relative enrichment of TCR clonotypes detected in tumor relative to peripheral blood(H). I. mRNA expression of immune related genes defining each cluster of CD4⁺ T cells in the tumor.

See also Figure S3 and Tables S3 and S4.

In total, 8.8% of the CXCL13⁺ PD-1^high subset of tumor infiltrating CD4⁺ cells sequenced in these 4 patients expressed a TCR identified to be tumor specific. The PD-1^low CXCL13⁻ subset of Tconv cells contained viral specific T cells from 2 patients, and viral reactivity in the other 2 patients was limited to CD4⁺ T cells expanded from the PD-1^low fraction (Figure S2A–C). Previous work has shown that tumor infiltrating T_REG cells also have specificity for tumor antigens but these cells would not be detected using our methods which require IFN-γ secretion (Ahmadzadeh et al., 2019). Together, the data shows that conventional CD4⁺ T cells infiltrating tumors can be divided into populations of clonally expanded, tumor enriched, PD-1^high, CXCL13⁺ cells that contain tumor antigen specific cells, and clonally diverse PD-1^low cells containing “bystander” cells specific for non-tumor antigens.

CXCL13⁺ CD4⁺ T cells in melanoma contain three distinct cell states

Unsupervised clustering of CD4⁺ T cells showed neoantigen specific cells within each of three CXCL13⁺ clusters (Figure 2A–F, Figure S3A). One cluster was defined by increased expression of memory genes (TCF7, IL7R) as well as genes (BCL6 and CD200) and surface markers (CXCR5) associated with T follicular helper (T_FH) cells (Figure 2C,I). A second TCF7⁻ population expressed coinhibitory (Tim-3 surface expression and LAG3 mRNA), inflammatory (CCL3, CCL4, IFNG mRNA), cytolytic (GZMA/K, PRF1 mRNA) markers, and the tissue resident memory marker CD103 (Figure 2C,I Table S3). A third CXCL13⁺ population expressed genes involved in proliferation (TYMS, TOP2A, MCM2/4 mRNA) (Figure 2C,2I). Neoantigen specific T cells were present in all 3 CXCL13⁺ clusters in 3 out of 4 patients (Table S4). There was overlap of TCRVb clonotypes between each of these CXCL13⁺ clusters (Figure S3B) and individual neoantigen specific clonotypes were represented in all 3 CXCL13⁺ clusters (Figure S3C), reminiscent of TCF7⁺ and TCF7⁻ subsets of CD8⁺ T cells that respond differently to ICB therapy in murine tumors (Miller et al., 2019; Siddiqui et al., 2019; Wu et al., 2016).

We next asked whether the transcriptional signatures associated with tumor specific CD4⁺ T cells were present in a larger melanoma patient cohort (Table S5), and whether the presence of these CD4⁺ T cells related to the number and phenotype of other immune cells in the melanoma tumor microenvironment (TME). We used an extended panel of 53 surface antibodies and targeted mRNA sequencing of 405 genes to re-analyze tumors from the initial 4 patients and to study 16 additional patients. Sequencing of between 422 and 3185 CD4⁺ T cells from each of the 20 patients followed by unsupervised clustering revealed FOXP3⁺ T_REG, CXCL13⁻ T_CONV, and 3 clusters of CXCL13⁺ T_CONV that expressed T_FH genes, cytolytic genes, and cell proliferation genes (Figure 3A–C). The proportion of CD4+ T cells in each cluster varied between patients with the total frequency of CXCL13⁺ CD4⁺ T cells correlating with expression of class II MHC on tumor cells, but not BRAF mutational status or tumor mutational burden in the subset of patients who underwent whole exome sequencing (Figure S3D–E). Flow cytometry of CD4⁺ T cells from 3 patients confirmed that CXCL13 producing cells were FOXP3⁻ and PD-1^high (Figure S3F). Bulk RNA sequencing of CD25^low conventional T cells sorted on PD-1 and CXCL6 expression also indicated CXCL13 was expressed predominantly in PD-1^high cells (>250 fold, FDR < 10⁻⁹), and that TCF7 mRNA was expressed at higher levels in the PD-1^high CXCR6⁻ subset (>16 fold, FDR 2×10⁻⁵ Figure S3G). In 4 patients whom material was available, the CD25^low PD-1^high CXCR6 CXCR6⁻ population proliferated better in response to stimulation than the CD25^low PD-1^high CXCR6⁺ (CXCL13⁺ TCF7⁻) population (Figure S3H). Thus the transcriptional signatures observed in CXCL13⁺ CD4⁺ T cells that contained all of the tumor specific CD4⁺ T cells identified in the initial 4 patients were present across a larger set of melanoma patients, suggesting this subset identifies or is enriched for T cells containing tumor reactive TCRs. To support this, we cloned 6 TCRs that were present in at least 2 cells in the CXCL13⁺ subset from patient X231, who had the lowest fraction of CXCL13⁺ CD4⁺ T cells (7.5% of CD4⁺ T cells) in the cohort of 20 patients. We used lentiviral transduction to express the 6 TCRs in donor T cells and then screened the transduced T cells for reactivity to candidate neoantigens expressed in the patient’s tumor. Two of the 6 TCRs conferred specificity to a single neoantigen created by a missense mutation in SEC31A (Figure 3D).

Figure 3: The fraction of CXCL13⁺ CD4⁺ T cells correlates with overall survival in melanoma.

A-B. Data from 26773 sorted CD4⁺ T cells from 20 patients were visualized by UMAP with (A) cells clustered by phenotype and (B) log₂ expression of the indicated mRNA or surface markers. C. Heatmap showing log₂ expression of mRNA that define phenotypic clusters of CD4⁺ T cells. D. Donor T cells transduced with TCR sequences from CXCL13⁺ T cells from patient X231 were incubated with wildtype or mutant SEC31A peptide and IFN-γ production was measured by ICS. E. Overall survival of 20 melanoma patients stratified by greater or less than the median of conventional CD4⁺ T cells belonging to CXCL13⁺ clusters. F. Overall survival of 471 patients in the cancer genome atlas based on the median of CXCL13 expression normalized to CD4 expression. G. Overall survival of 471 patients based on the median predicted fraction of T_FH cells using the CIBERSORT algorithm.

See also figures S3 and S4 and tables S5 and S6.

The fraction of CXCL13⁺ CD4⁺ T cells is predictive of overall survival in melanoma

We then asked whether the frequency of CXCL13⁺ CD4⁺ T cells infiltrating melanoma might be associated with a more effective immune response. The fraction of CXCL13⁺ CD4⁺ T_CONV varied between 7% and 55% of total tumor infiltrating CD4⁺ T cells in this patient cohort, and those patients with greater than the median of CXCL13⁺ CD4⁺ T cells as a fraction of CD4⁺ T_CONV cells exhibited improved survival (Figure 3E). Of the CXCL13⁺ CD4⁺ subsets, the TCF7⁻ subset best correlated with survival (Figure S4A). Total CD8⁺ T cells also correlated with survival whereas total CD4⁺ T cells or T_REG cells did not correlate with survival in this cohort (Figure S4B, Table S6).

To support an association between increased CXCL13⁺ CD4⁺ T cells and improved survival in melanoma, we identified genes in the single cell RNA seq data that could act as surrogates for the presence of CXCL13⁺ CD4⁺ T cells and analyzed their expression levels in publicly available bulk RNA sequencing datasets. Expression of CXCL13 and BTLA were largely restricted to CXCL13⁺ CD4⁺ T cell subsets as well as less common CD8⁺ T cell subsets and rare cDC1 cells in the case of BTLA (Figure S4C–D). IL21 expression was largely confined to CXCL13⁺ CD4⁺ T cell subsets (Figure S4E). Analysis of TCGA data from 471 melanoma patients showed that increased expression of CXCL13 and BTLA were associated with greater overall survival (corrected p=0.003, 0.017 respectively) (Figure S4F–G) These genes can correlate with total immune infiltrate, however CXCL13 and BTLA expression also predicted greater survival when normalized to CD4 expression (corrected p=0.001, 0.038 respectively, Figure 3F, Figure S4H). IL21 was expressed in a minority of melanoma patients in this dataset and was associated with a nonsignificant trend toward improved survival (corrected p=0.12, Figure S4I). As an alternate method of estimating the number of CXCL13+ CD4+ T cells within tumor samples, we utilized the CIBERSORT algorithm (Chen et al., 2018) and found that the estimated fraction of T_FH cells also correlated with survival (p=.001, Figure 3G).

Phenotype of CD8⁺ T cells in the tumor microenvironment

The expression of CXCL13 in subsets of tumor infiltrating CD4⁺ and CD8⁺ T cells led us to examine other parallels in the phenotypes of these cells. We analyzed 223 to 2248 CD8⁺ T cells by targeted scRNA seq in tumor samples from each of the 20 patients. This analysis included two patients (X197 and X198) in whom CD8⁺ T cells specific for tumor antigens were identified by culturing tumor fragments in high dose IL-2 and testing for reactivity to peptides corresponding to highly expressed mutations and lineage specific antigens. Patient X197 exhibited CD8⁺ T cell reactivity to tyrosinase, TRP2, and Mart1 and the cancer testes antigen Mage A3. TCRVb sequencing of cells sorted based on IFN-γ secretion following incubation with each of these antigens identified 7 clonotypes (Veatch et al., 2018). We did not identify CD8⁺ T cells specific for self-antigens in patient X198 but identified 3 TCR clonotypes reactive to mutated neoantigens (Figure S5A–E, Figure S6A–C).

Unsupervised clustering of CD8⁺ T cells from the two patients in our study showed that a majority expressed CXCL13, PD-1, and CD39 consistent with previous work (Duhen et al., 2018; Litchfield et al., 2021; Simoni et al., 2018; Thommen et al., 2018). Within the CXCL13⁺ cells, distinct clusters that co-expressed CXCL13 and TCF7, or CXCL13, TYMS and other markers of proliferation were identified (Figure 4A, Figure S6D). Tumor antigen specific cells from patients X197 and X198 were found in the PD-1⁺ CXCL13⁺ clusters, but not in the cluster lacking expression of CXCL13 and PD-1 (Figure 4B). Unsupervised clustering of the scRNA seq data from the entire cohort of 20 patients also identified CD8⁺ T cells that expressed PD-1, CD103, CD39, TOX and CXCL13. As observed in the first 2 patients, a subset of CXCL13⁺ CD8⁺ T cells expressed TYMS and other markers of cell proliferation, and a minor subset expressed TCF7, consistent with less differentiated cells described by others (Figure 4C–D, Figure S6E–F) (Miller et al., 2019; Siddiqui et al., 2019; Wu et al., 2016). The fraction of CXCL13⁺ CD8⁺ T cells of total CD8⁺ T cells showed a nonsignificant trend toward improved survival in our cohort (Figure S6G).

Figure 4. Tumor infiltrating CD8⁺ T cells share gene expression signatures and their frequency and proliferative state correlate with tumor antigen specific CD4⁺ T cells.

A. 2585 CD8⁺ T cells from 2 melanoma patients were visualized by UMAP and clustered by phenotype. B. Presence of tumor antigen specific TCRVb in the UMAP plots. C. 15332 CD8⁺ T cells from 20 patients visualized by UMAP showing mRNA expression of CXCL13, TYMS, IL7R, TCF7, TOX, and GZMB, and surface expression of CD279/PD-1, CD39, Tim3 and CD103. D. Phenotypic categories of CD8⁺ T cells are indicated. E. Jaccard similarity index expressed as a percentage of mRNA expressed genes upregulated at least 2-fold in phenotypic categories of CD4⁺ and CD8⁺ T cells. F. Spearman correlation of different CD4⁺ T cell populations as a fraction of CD4⁺ T cells compared to CD8⁺ T cell populations as a fraction of CD45⁺ cells across the 20-patient cohort. G. Spearman correlation of CXCL13 expression with CD8a expression in the 471 patients with melanoma. H. Spearman correlations of CXCL13, BTLA and IL21 expression to CD8A expression, with and without normalization to CD4 expression. * FDR <.05 ** FDR<0.01 *** p <10⁻⁴. p<10⁻⁶ for all correlations in H, false discovery rates for spearman correlations determined using Benjamini Hockberg correction.

See also figures S5 and S6.

The sets of genes upregulated greater than 2-fold in individual phenotypic clusters of CD8⁺ and CD4⁺ T cells were compared for overlap using a Jaccard index. There was a pattern of overlap between the CXCL13⁺TCF7⁺ CD4⁺ and CD8⁺ populations with multiple T_FH associated genes such as BCL6 and CD200 also upregulated in the CD8⁺ population (Figure 4E). There was also overlap between the CXCL13⁺ TCF7⁻ populations, including cytolytic, inflammatory and coinhibitory molecules. As expected, there was also increased overlap in gene expression within the CXCL13⁻ conventional T cell populations.

CXCL13⁺ CD4⁺ T cells correlate with CD8⁺ T cell infiltration and activation

CD4⁺ T cells can coordinate the functions of other immune cells in the TME. We therefore sought to examine if the fraction of CXCL13⁺ CD4⁺ T cells was related to the presence and phenotype of CD8⁺ T cells, myeloid cells, and B cells. As CD4⁺ T cells could potentially affect both the number and phenotype of CD8⁺ T cells and B cells, we compared CD4⁺ T cell subsets as a fraction of CD4⁺ T cells to CD8⁺ and B cell subsets as a fraction of CD45⁺ hematopoietic cells. Across the 20 patient cohort, the fraction of CXCL13⁺ CD4⁺ T cells correlated with CD8⁺ T cells (Spearman correlation coefficient R= 0.52, corrected p=0.027) and more closely with CXCL13⁺ CD8⁺ T cells (R=0.62, p=0.007) and proliferating TYMS⁺ CXCL13⁺ CD8⁺ T cells (R=0.65, p=0.005). The TCF7⁻ effector (R=0.69, p= 0.003) and proliferating CXCL13⁺ CD4⁺ subsets correlated more closely with proliferating CD8⁺ T cells (R=0.82 p=0.00013) than with the T_FH subset (R=0.30, p=0.14, Figure 4F).

There are no public melanoma scRNA seq datasets with sufficient patient numbers to independently validate these observations, therefore we interrogated the TCGA data, using expression of CXCL13, BTLA and IL21 as surrogates of the presence of CXCL13⁺ CD4⁺ T cells. Each of these markers correlated with CD8a expression in melanoma (Figure 4G–H, p < 1e-54), and these correlations remained after normalization of CD4 expression (p < 1e-9). These observations show the fraction of CD4⁺ T cells with a signature of tumor antigen specificity correlates with infiltration and activation of CD8⁺ T cells, consistent with such CD4⁺ T cells providing helper functions to tumor antigen specific CD8⁺ T cells.

CXCL13⁺ CD4⁺ T cells correlate with macrophage activation

We then examined correlations between CXCL13⁺ CD4⁺ subsets and myeloid cell populations and phenotypes. Sequencing of between 147 and 2448 CD45⁺ CD3⁻ CD19⁻ cells from each of the 20 tumors identified CD123⁺ CD4⁺ plasmacytic dendritic cells (pDC); BTLA⁺ LAMP3⁺ type 1 conventional dendritic cells (cDC1), FCER⁺ CD1C⁺ type 2 conventional dendritic cells (cDC2) and a large number of CD14⁺ macrophages which varied in their expression of C1Qa/b and FCN1 (Figure 5A–B, Figure S7A) (Xiong et al., 2020). There were no correlations between the various myeloid subsets and CXCL13⁺ CD4⁺ T cell fractions at an FDR of <5% (Figure S7B), although our power to detect these correlations is limited by the small numbers of dendritic cell subsets in our samples.

Figure 5. The presence of CXCL13⁺ CD4⁺ T cells correlates with an immune stimulatory phenotype in macrophages.

A-B. 12977 CD45⁺ CD19⁻ CD3⁻ cells from 20 melanoma patients were separated into phenotypic clusters with differentially expressed genes shown (A) and phenotypic clusters visualized by UMAP (B). C. Spearman correlations of expression of the indicated mRNA in macrophages in the 20-patient cohort with the fraction of CD4⁺ T cells in CXCL13⁺ clusters. D. Correlation of the fraction of CXCL9⁺ macrophages and the fraction of CXCL13⁺ CD4⁺ cells of total CD4⁺ T cells. E-F. Expression of CXCL13 compared to CXCL9 (E) or the indicated chemokines (F) in 471 melanoma patients compared to expression of CXCL13, BTLA, and IL21, with or without normalization of these to CD4, with Spearman correlations shown. *significant with FDR <0.05 ** significant with FDR <0.01 by spearman correlation with the Benjamini-Hockberg correction. All correlations in (F) p < 10⁻¹⁰ for correlations with CXCL9, CXCL10, CXCL11 and IL15. G. Receptor ligand interactions between CD4⁺ T cell subsets and B cells, CD8⁺ T cells and myeloid cells generated by cellphoneDB.

See also figure S7.

Macrophages can modulate antitumor immunity by producing proinflammatory or suppressive chemokines and cytokines. CXCL9, CXCL10 and CXCL11 produced in response to IFN-γ mediate CXCR3 dependent lymphocyte recruitment (House et al., 2020), whereas CXCL1 and CXCL3 mediate CXCR2 dependent recruitment of immunosuppressive myeloid cells (Alfaro et al., 2016). Tumor associated macrophages (TAMs) can also produce cytokines such as IL-15 that can support T cell function and IL-10 that is suppressive (Carrero et al., 2019; Ruffell et al., 2014). To evaluate potential correlations with CXCL13⁺ CD4⁺ T cells, we focused on the TAMs that make up the bulk of the myeloid cells across these tumors. We investigated 25 immune regulatory genes whose expression varied across TAMs in our sample and found that the fraction of CXCL13⁺ CD4⁺ T cells positively correlated with the inflammatory chemokines CXCL9, CXCL10 and CXCL11, as well as IL15 and the activation marker CD40, and negatively correlated with CXCL1, CXCL3 and IL10 with an FDR of less than 5% (Figure 5C–D).

The association of macrophage activation with the presence of CXCL13⁺ CD4⁺ T cells was corroborated in the larger TCGA melanoma data set. CXCL13, BTLA and IL21 expression, which serve as a proxy for the presence of tumor antigen specific CD4⁺ T cells, correlated with IL15, CXCL9, CXCL10 and CXCL11 expression in the TCGA data, but did not correlate with CXCL1 or CXCL3 expression (Figure 5E–F). These relationships were maintained when expression of CXCL13, BTLA and IL21 were normalized to CD4 expression indicating they were not general features of lymphocyte infiltration. Global analysis of receptor ligand interactions between cell populations showed increased interactions between CXCL13⁺ CD4⁺ T cells and CXCR5⁺ T cells and B cells, and the TCF7⁻ subset showed increased interactions with myeloid and other cell types mediated by the inflammatory chemokines CCL3, CCL4, and CCL5 (Figure 5G).

CXCL13⁺ CD4⁺ T cells correlate with B cell maturation and colocalize with B lineage cells

CD19⁺ B cells in TIL were sorted for single cell sequencing to evaluate B cell maturation stages and relationship to CXCL13⁺ CD4⁺ T cell infiltration. Naive B cells expressing IGHD and membrane IGHM, memory cells expressing surface IGHG1, and secreted IGHG1⁺ CD20⁻ plasma cells were identified (Figure 6A–B, Figure S7C) (Horns et al., 2020). The fraction of CXCL13⁺ CD4⁺ T cells did not correlate with the total number of tumor infiltrating B cells as a fraction of hematopoietic cells, but there was a significant association with memory B cells (Spearman R = 0.70, corrected p = .01) and trends toward correlation between the frequency of TCF7⁺ CXCR5⁺ T_FH cells and the total number of B cells as a fraction of CD45⁺ hematopoietic cells (Spearman R = 0.57, corrected p = 0.06), and the number of naïve B cells (Spearman R = 0.54, corrected p = 0.08) (Figure 6C–D). There was no correlation between the TCF7⁻ subset of CXCL13⁺ CD4⁺ T cells and any B cell subset. Consistent with a role for the T_FH subset of CXCL13⁺ CD4⁺ T cells in B cell stimulation, T_FH cells but not effector cells isolated from one patient stimulated B cell proliferation in vitro (Figure S7D). B cell receptor clonotypes were expanded in the memory and plasma cell compartments and individual clonotypes were found in both subsets (Figure S7E,F). Using CXCL13, BTLA and IL21 expression as surrogates for T_FH frequency, we found robust correlation between the expression of all 3 genes and markers of B cells (CD19), naïve B cells (IGHD) and memory and plasma B cells (IGHG1) in the TCGA data. These correlations persisted when controlled for CD4 expression (Figure 6E–F).

Figure 6. TFH-like CD4⁺ T cells correlate with B cell infiltration and co-localize with B cells in melanoma.

A-B. 9428 CD19⁺ B lineage cells from 20 patients were visualized by UMAP with expression of individual genes shown (A) as well as phenotypic clusters (B). C-D. Correlation of memory B cells as a percentage of CD45⁺ cells and CXCL13⁺, TCF7⁺, CD4⁺ T cells as a percentage of CD4⁺ T cells (C) and Spearman correlation with B cells subsets with CXCL13⁺ CD4⁺ T cell subsets in 20 melanoma patients (D) E-F. Spearman correlations of CXCL13 versus IGHG1 (E) or CXCL13, BTLA, and IL21 expression versus B cell markers in 471 melanoma patients (F). G. Multiplex immunohistochemistry of a tertiary lymphoid structure in patient X198, with CD4⁺ T cells inferred from cells staining for CD3 but not CD8. H-I. Proximity analysis of the number of CD79a⁺ B lineage cells within 100 microns of each CD4⁺ or CD8⁺ T cell based on CXCL13 staining in patient X197 (H) and X198 (I). Boxplots show minimum, 25th percentile, median, 75^th percentile, and maximum. J. Representative image showing distribution of PD-1^high Tcf7⁺ (red) and PD-1^high Tcf7⁻ (blue) CD4⁺ T cells overlayed on classifier showing B cell rich (light blue) versus poor (yellow) regions. K. Fraction of either CD4⁺ or CD4⁺ PD1⁺ T cells in B cell rich zones in n=8 melanoma cases with significant B cell infiltrates. * FDR<0.05 ** p<0.0001.

See also figures S7 and S8.

In two patients from who we had identified tumor antigen specific CD4⁺ and CD8⁺ T cells and observed tertiary lymphoid structures (TLS), we used multiplex immunohistochemistry to analyze the localization of CXCL13⁺ CD4⁺ and CD8⁺ T cells relative to TLS (Figure 6G, Figure S8A,B). Analysis of over 10,000 individual CD8⁺ and CD4⁺ T cells across the tumors of these patients revealed that both CD4⁺ CXCL13⁺ and CD8⁺ CXCL13⁺ T cells were more likely to be in close proximity to CD79a⁺ B lineage cells than their CXCL13⁻ counterparts (Figure 6 H–I). To confirm and extend these observations we analyzed tumors from 6 additional patients who had B cell tumor infiltrates and had not been previously treated with PD-1 inhibitors and identified CD4⁺ T cell subsets through staining for PD-1, FoxP3 and Tcf1. Consistent with the prior findings, we found that PD-1^high Tcf1⁺ FoxP3⁻ CD4+ T cells preferentially localized to B cell rich areas of tumor (Figure 6J–K, Figure S8C,D), suggesting that CXCR5/CXCL13 interactions might mediate localization and that a population of tumor antigen specific CD4⁺ T cells may localize to a distinct intratumoral niche as observed with CD8⁺ T cells (Jansen et al., 2019).

Infiltrating CD4⁺ T cells in breast cancer treated with immune checkpoint inhibitors exhibit similar phenotypes and correlate to activation of CD8⁺ T cells and myeloid cells.

An important question is whether the population structures of CD4⁺ T cells observed in the single cell sequencing data in melanoma and the associated correlations are present in a distinct cancer type. To address this, we utilized single cell RNA seq data from a cohort of 31 breast cancer patients treated with anti PD-1 (Bassez et al., 2021). CD4⁺ T cells in this cohort showed similar patterns of gene expression, with clusters of CXCL13⁻, PD-1⁻ cells expressing high IL7R, FOXP3⁺ regulatory T cells, and CXCL13⁺ cells comprised of three distinct populations that expressed higher relative levels of TYMS, T_FH (BCL6) and memory markers (TCF7 and IL7R), or HAVCR2 (the gene encoding Tim-3), IFNG and GZMA (Figure 7A). These CXCL13⁺ populations were clonally expanded in tumors and expanded with immune checkpoint inhibitor treatment (Bassez et al., 2021).

Figure 7. Infiltrating CD4⁺ T cells in breast cancer patients treated with immune checkpoint inhibitors exhibit similar phenotypes and correlate to activation of CD8⁺ T cells and myeloid cells.

A. Single cell RNA sequencing of CD4⁺ T cells from 31 patients with breast cancer visualized by UMAP for the expression of the indicated genes, with cells clustered by phenotype (right lower panel). B. Correlation between CXCL13⁺ cell populations as a fraction of CD4⁺ T cells and proliferating CD8⁺ T cells as a fraction of all cells, C. Spearman correlations of the indicated cell subsets. D-F. Correlation between CXCL13⁺ CD4⁺ T cells as a fraction of CD4⁺ T cells and the fraction of myeloid cells expressing CXCL9 (D), IL15 (E), and a panel of genes (F).

In this breast cancer dataset CXCL13⁺ CD4⁺ T cell populations as a fraction of CD4⁺ T cells correlated with CD8⁺ T cells (Spearman R = 0.68, p <10⁻⁴) and more closely with CXCL13⁺ CD8⁺ T cells (R = 0.84, p<10⁻⁸) and proliferating CD8⁺ T cells (R=0. p<10⁻⁸) 83, as a fraction of tumor infiltrating cells (Figure 7B–C). Mirroring our findings in melanoma, CXCL13⁺ CD4⁺ T cells as a fraction of CD4⁺ T cells was also correlated positively with expression of CXCL9, CXCL10, CXCL11, IL15 and CD40 (Spearman R > 0.66, p < 10⁻⁴), but not with expression of CXCL2, CXCL3, or IL10 (Figure 7D–F). The TCF7⁺ CXCL13⁺ subset as a fraction of CD4⁺ T cells correlated modestly with the fraction of total B cells, naive B cells, and memory B cells as a fraction of total cells (Spearman R=0.45, p<0.01), but TCF7⁻ CXCL13⁺ CD4⁺ T cells did not, as observed in melanoma. Taken together, these data show that similar populations of CD4⁺ T cells exist in a second tumor type with distinct biology from melanoma, and that the fraction of CXCL13⁺ CD4⁺ T cells correlates with activation in the CD8⁺ T cell, myeloid and B cell components of the TME. Looking broadly across cancer types in the TCGA, CXCL13 expression normalized to CD4 expression above the median was associated with better overall survival at FDR < 5% in cutaneous melanoma, breast cancer, and head and neck cancer, but worse survival in glioblastoma, uveal melanoma, and papillary and clear cell renal carcinoma (Figure S8D).

Discussion

The phenotype and functions of CD4⁺ T cells that infiltrate tumors and recognize tumor antigens remain poorly understood. We identified 350 neoantigen and tumor associated antigen specific CD4⁺ T cells from primary tumor samples of 4 melanoma patients and show that their phenotypes and transcriptional signatures are distinct from other infiltrating CD4⁺ T cell populations. This represents a large increase in the number of tumor specific CD4⁺ T cells compared to previous studies (Balança et al., 2021; Cachot et al., 2021). Unlike clonally diverse CD4⁺ T cells that include virus-specific cells, CD4⁺ T cell clusters containing tumor antigen specific cells are frequently clonally expanded and characterized by the expression of CXCL13, PD-1 and TOX. CXCL13⁺ CD4⁺ T cells could be subdivided into three distinct clusters defined by markers of proliferation, a second with expression of TCF7 and T_FH markers, and a third TCF7⁻, Tim3⁺, CD103⁺, GZMA⁺, IFNG⁺ population. While our finding of the presence of tumor antigen specific T cells exclusively within the PD-1^high CXCL13^high subset in TIL argues that this is a signature of tumor antigen specificity, we cannot exclude the possibility that some tumor reactive T cells reside in the clonally diverse PD-1 negative T_CONV compartment or that some CXCL13^high CD4⁺ T cells are not tumor antigen specific.

CXCL13⁺ T_FH-like CD4⁺ T cells and more generally PD-1^high conventional CD4⁺ T cells have been identified in non-small cell lung, head and neck, colorectal and triple negative breast cancers, however their antigen specificity has not been defined and it has been controversial whether such cells are stimulatory or suppressive of antitumor immunity (Bonnal et al., 2021; Cui et al., 2020; Gu-Trantien et al., 2013; Hollern et al., 2019; Ruffin et al., 2021; Singh et al., 2020; Zappasodi et al., 2018). Our finding that the presence of CXCL13⁺ CD4⁺ T cells and that CXCL13 expression more broadly, correlated with overall survival in our cohort despite diverse subsequent therapies argue for a potential beneficial role of these cells in antitumor immunity. This is supported by the recent observations that CXCL13 expression predicts response to immune checkpoint inhibition independent of CD8⁺ T cell infiltration in a meta-analysis of multiple cancer types (Litchfield et al., 2021). CXCL13⁺ CD4⁺ T cells also predicted response to checkpoint inhibition in breast cancer (Zhang et al., 2021)as did the secretion of CXCL13 by cultured tumor fragments in other human tumors (Voabil et al., 2021). CXCL13 is also produced by tumor antigen specific CD8⁺ T cell subsets, but our data and that of others suggest that CD4⁺ T cells are a major source of CXCL13 within tumors (Gu-Trantien et al., 2013). Furthermore, greater numbers of PD-1^high CD4⁺ T cells within tumors predicted immune checkpoint inhibitor responsiveness better than overall immune infiltration or infiltration of PD-1^high CD8⁺ T cells (Voabil et al., 2021).

Although CD4⁺ T cells can mediate direct lysis of class II MHC positive tumor cells (Oh et al., 2020), only 5 of the 20 patients in our cohort had class II MHC expression on a significant fraction of tumor cells, consistent with prior work showing the majority of melanoma cells are class II MHC negative (Rodig et al., 2018). CD4⁺ T cells can also be activated by tumor antigens cross presented by antigen presenting cells and coordinate antitumor immunity through several indirect mechanisms. These include local support of CD8⁺ T cells (Alspach et al., 2019), activation of macrophages (Mumberg et al., 1999; Tveita et al., 2016), and activation of B cells (Hollern et al., 2019). Recent work in mice showed that CD4⁺ T cells activated by tumor antigen specific B cells stimulate antitumor CD8⁺ T cell responses through IL-21 secretion, illustrating one such mechanism (Cui et al., 2021; Zander et al., 2019). Our data in 20 melanoma patients demonstrating that the abundance of CD4⁺ T cells with the signature of tumor antigen reactivity correlated with local activation of CD8⁺ T cells, activation of an immune stimulatory phenotype in macrophages, and the presence and differentiation of B cells, supports diverse functional roles for tumor antigen-specific CD4⁺ T cells in shaping the TME. Correlations of the associated markers CXCL13, BTLA, and IL21 with markers of CD8⁺ T cell and B cell infiltration and macrophage activation in a larger cohort of melanoma patients, and by analysis of data from an independent breast cancer cohort provide additional support for this conclusion.

Our study is limited by the inability to define the antigen specificity of all of the CD4⁺ T cells in the CXCL13⁺ and putative bystander subsets in tumor infiltrates, potentially because our criteria for screening potential neoantigens included only a subset of expressed mutations. Our assay to detect tumor antigen specificity also relied on T cell proliferation in culture and production of IFN-γ, which precluded analysis of the antigen specificity of regulatory T cells. Finally, correlations between the presence of CXCL13⁺ CD4⁺ T cells and characteristics of other immune cells in the TME does not prove that CXCL13⁺ CD4⁺ T cells instruct the changes in the TME. The ultimate test of whether tumor antigen specific CD4⁺ T cells have a causal role in modifying the TME in the direction of successful antitumor immunity will be whether therapeutic interventions that enhance the frequency and function of these cells, such as by the adoptive transfer of tumor antigen specific CD4⁺ T cells, are capable of mediating these changes. The ability to prospectively identify tumor antigen specific CD4⁺ T cells by phenotype could allow the isolation of such cells or their tumor reactive TCRs for use in adoptive transfer to activate the TME and potentially improve responses to ICB. The existence of such cells across different tumor types suggests that such a therapeutic approach, if successful, could be broadly applicable in immunogenic solid tumors.

STAR Methods

Resource Availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Josh Veatch (jveatch@fhcrc.org).

Materials Availability

This study did not generate new unique reagents

Data and Code Availability

Raw and processed Single-cell RNA-seq datais publicly available at the GEO database with accession number GSE198265. This paper also analyzes existing, publicly available data from a breast cancer cohort (Bassez et al., 2021) which is available at the EGA database with accession number EGAD00001006608. Raw and processed bulk TCRB sequencing is available from the adaptive website at https://clients.adaptivebiotech.com/pub/veatch-2022-cc. Sharing of raw exome sequencing and RNA sequencing data was not allowed by the informed consent signed by patients enrolled in this study.

Microscopy data reported in this paper will be shared by the lead contact upon request.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Analysis code and scripts to generate figures are available at https://github.com/cshasha/veatch 2022.

Experimental model and subject details

Patient cohort

Twenty patients were enrolled in a clinical protocol approved by the Institutional Review Board of Fred Hutchinson Cancer Research Center (FHCRC 2643; NCT01807182) to provide tumor samples for research and for production of therapeutic tumor infiltrating lymphocyte (TIL) products. Melanoma patients with stage IV or stage III disease unlikely to be cured by surgery, >18 years of age, with an ECOG </=1, and with a site of metastatic disease that could be safely resected or biopsied, were eligible for enrollment. Leukapheresis products were obtained under a separate IRB reviewed clinical protocol from patients X197 and X198 after the patients received TIL infusion for progressive disease. All patients provided informed consent for enrollment on these protocols.

Tumor single cell suspensions were dissociated from surgical tumor resection samples using the human tumor dissociation kit (Milltenyi) and a Gentlemacs dissociator fitted with C-tubes (Milltenyi) following the manufacturer’s instructions and cryopreserved. For production of clinical tumor infiltrating lymphocyte (TIL) products used to detect CD8⁺ T cell responses from patients X197 and X198, TIL were expanded from tumor fragments in 6,000 IU/ml recombinant IL-2 (Proleukin; Novartis), using methodologies developed at the Surgery Branch of the National Cancer Institute(Dudley et al., 2001). TIL cultures were selected based on cell growth and autologous tumor reactivity as determined by IFNg secretion after co-culture with autologous tumor cells. The TIL were cryopreserved until needed for use, then thawed and further expanded using a rapid expansion protocol, as previously-described(Riddell and Greenberg, 1990).

Antigen presenting cells

Autologous B cells were isolated from fresh or thawed PBMC by positive selection using human CD19 microbeads (Miltenyi, cat# 130-050-301) according to the manufacturer’s instructions. Isolated B cells were cultured with irradiated (5000Gy) NIH 3T3 cells expressing human CD40L for 7 days in B cell medium supplemented with 200U/ml human IL-4 (Peprotech) as described.(Tran et al., 2014) B cells were subsequently harvested and restimulated with 3T3 CD40L and fresh medium every 3 days. B cells were pulsed with peptides and used in assays at day +3 of stimulation 2 or 3.

T cell culture

For isolation of CD4⁺ T cells from tumor samples, cryopreserved tumor single cell suspensions were made from tumor tissue that remained after TIL generation in X205 and X422 were thawed and incubated overnight in RPMI media with L-glutamine and HEPES (Gibco) supplemented with 10% human serum, 50 μM betamercaptoethanol, penicillin and streptomycin, 4 mM L-glutamine (termed CTL media) and 2 ng/ml recombinant human IL-7 (Peprotech). The cell suspensions were then stained with fluorescently labelled antibodies to human CD4, CD3 and PD-1 and the top tertile of PD-1 expressing CD3⁺ CD4⁺ T cells were sorted, and plated at 10 cells per well in a 96 well plates with 100,000 irradiated allogeneic PBMC in CTL media supplemented with 3,000 U/ml of recombinant human IL-2 and 50ng/ml anti-CD3 (OKT3).(Yossef et al., 2018) Cytokines were added every 7 days and colonies of cells were screened for reactivity against mutant peptides at day 28 of culture. Between 3 and 10% of wells formed T cell colonies large enough to screen. These T cell lines from patients X422 and X205 were screened for antigen reactivity by incubation with pools of all 20-mer crude mutant peptide pairs (2ug/ml/peptide) from that patient in the presence of brefeldin A, followed by staining with antibodies for human CD4 and CD19, fixing and permeabilizing using the BD Cytofix/Cytoperm kit (BD cat#554714) using the manufacturer’s instructions, and staining of intracellular IFN-γ (#B27). Quantitation of IFN-γ secreting cells by FACS was compared to an unstimulated aliquot of T cells. Cell lines with >50% IFN-γ secretion were then expanded using a rapid expansion protocol and cryopreserved(Riddell and Greenberg, 1990). Cryopreserved T cells were thawed and rested overnight in CTL media supplemented with 10 U/ml IL-2 prior to ELISA after stimulation with overlapping peptide pools to determine the individual target mutation, and then stimulation with purified 27-mer peptides with either the wildtype or mutant amino acid at position +14 to confirm antigen specificities.

Ten thousand of the top tertile of PD-1 expressing CD4⁺ CD3⁺ from patient X205 were also polyclonally expanded in 4ml CTL in a 12 well plate by stimulation with anti-CD3 (OKT3, 50ng/ml) in cultures supplemented with 6×10⁶ irradiated (5000 rad) allogeneic PBMC and 3,000 U/ml IL-2. After 21 days, the T cells were incubated at a 1:1 ratio with autologous B cells and neoantigen peptide pools (2ug/ml each peptide) in the presence of mononesin, anti-CD40, anti-CD28, anti-CD49a, and fluorescently labelled antiCD40L for 16 hours as described(Chattopadhyay et al., 2006). CD40L expressing CD4⁺ T cells were sorted by FACS and DNA was prepared for TCRVb sequencing to identify additional neoantigen reactive TCRVb clonotypes.

Cryopreserved PBMC from the two patients that received TIL infusion were thawed and rested overnight in CTL and 2 ng/ml recombinant human IL-7. The following morning PBMC were washed, and 10 × 10⁶ cells were plated in individual wells of a 6 well plate in 5 ml CTL media containing a pool of 1μg/ml of each candidate peptide supplemented with IL-21 (30 ng/ml). Recombinant IL-7 (5 ng/ml), IL-15 (1 ng/ml) and IL-2 10 U/ml (Peprotech) was added on day +3, and half media changes with supplemental IL-2, IL-7 and IL-15 were performed on days +3, +6, and +9. On day +13, cells from individual wells were harvested and T cells assayed by ELISA, ELISpot, or cytokine staining assays. Cultures that were evaluated by TCRVb sequencing for expansion of TCRVb clones were enriched for CD4⁺ T cells by negative magnetic selection with the human CD4 isolation kit (StemCell) prior to DNA isolation and TCRVb sequencing.

Method details

Nucleic acid preparation

Tumor single cell suspensions for whole exome sequencing and bulk RNA sequencing were thawed and depleted of hematopoietic cells using the EasySep^™ Human CD45 Depletion Kit (StemCell) and DNA and RNA were extracted using the AllPrep DNA/RNA Mini Kit (Qiagen). Patient X198 tumor DNA and RNA was isolated without depletion of hematopoietic cells. DNA extraction from T cells for TCRVb sequencing, and from peripheral blood mononuclear cells (PBMC) for TCRVb sequencing or exome sequencing was performed using the DNEasy kit (Qiagen) or the QIAamp DNA Micro Kit (Qiagen) if cellular input was below 500,000 cells. DNA following stimulation of PBMC was extracted following enrichment of CD4⁺ T cells using EasySep^™ Human CD4⁺ T Cell Isolation Kit (StemCell).

Whole exome sequencing:

Exome sequencing libraries were prepared using the Agilent SureSelectXT Reagent Kit and exon targets isolated using the Agilent All Human Exon v6 (Agilent Technologies, Santa Clara, CA, USA). 200 ng of genomic DNA was fragmented using a Covaris LE220 focused-ultrasonicator (Covaris, Inc., Woburn, MA, USA) and libraries prepared and captured on a Sciclone NGSx Workstation (PerkinElmer, Waltham, MA, USA). Library size distributions were validated using an Agilent 2200 TapeStation. Additional library QC, blending of pooled indexed libraries, and cluster optimization was performed using Life Technologies’ Invitrogen Qubit® 2.0 Fluorometer.

The resulting libraries were sequenced on an Illumina HiSeq 2500 using a paired-end 100bp (PE100) strategy. Image analysis and base calling was performed using Illumina’s Real Time Analysis v1.18 software, followed by “demultiplexing” of indexed reads and generation of FASTQ files using Illumina’s bcl2fastq Conversion Software v1.8.4 (http://support.illumina.com/downloads/bcl2fastq_conversion_software_184.html). Read pairs passing standard Illumina quality filters were retained for further analysis. Paired reads were aligned to the human genome reference (GRCh37/hg19) with the BWA-MEM short-read aligner(Li, 2013; Li and Durbin, 2009). The resulting alignment files, in standard BAM format, were processed by Picard 2.0.1 and GATK 3.5(McKenna et al., 2010) for quality score recalibration, indel realignment, and duplicate removal according to recommended best practices(Van der Auwera et al., 2013).

To call somatic mutations from the analysis-ready tumor and normal BAM files, we used three independent software packages: MuTect 1.1.7(Cibulskis et al., 2013) and Strelka 1.0.14(Saunders et al., 2012). Variant calls from both tools, in VCF format, were annotated with Oncotator(Ramos et al., 2015). Annotated missense somatic variants were combined into a single summary for each sample as follows. First, any mutation annotated as “somatic” but present in dbSNP was removed if it was not also present in COSMIC or its minor allele frequency was greater than 1% (according to the UCSC Genome Browser snp150Common table). Variants supported by both variant callers were retained, and those supported by only one variant caller were subject to manual inspection.

Characteristics of exome sequencing from 9 sequenced cases are shown in Table S7.

RNA-Seq data processing:

An RNA-seq library was prepared from total RNA using the TruSeq RNA Sample Prep v2 Kit (Illumina, Inc., San Diego, CA, USA) and a Sciclone NGSx Workstation (PerkinElmer, Waltham, MA, USA). Library size distributions were validated using an Agilent 2200 TapeStation (Agilent Technologies, Santa Clara, CA, USA). Additional library QC, blending of pooled indexed libraries, and cluster optimization was performed using Life Technologies’ Invitrogen Qubit® 2.0 Fluorometer (Life Technologies-Invitrogen, Carlsbad, CA, USA). The library was sequenced on an Illumina HiSeq 2500 to generate 61M read pairs (two 50nt reads per pair). Reads were aligned to a human RefSeq derived reference transcriptome with RSEM 1.2.19(Li and Dewey, 2011) to derive abundances for each gene in transcript-per-million (TPM) units.

For bulk RNA sequencing of sorted intratumoral CD4 subsets, tumor cells from patients X431, X330 and X368 were thawed, washed and stained with fluorescently labelled antibodies for anti-human CD3 BV711, CD4 BUV395, PD-1 FITC, CD25 PE-Cy7, CD127 BV421, CD200 PE, CXCR6 APC. Then we sorted the CD4 subsets i.e., CD25^low PD-1^high CXCR6⁻ T_FH cells, CD25^low PD-1^high CXCR6⁺ effector cells, CD127⁻ CD25⁺ Treg cells and CD25_low PD-1_low CD127^high bystander cells from each tumor sample. Next, we prepared cDNA from sorted cells using the SMART-Seq HT Kit (Takara bio, cat# 634438) followed library preparation for RNA-sequencing using Illumina Nextera XT DNA kit (cat# 15032354) with IDT for Illumina UDI primers (cat# 20026930) based on manufacturer’s protocol. Library size distributions were validated using an Agilent Tapestation High Sensitivity DNA 5000. Additional library quality control and blending of pooled indexed libraries were performed using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific). RNA-seq libraries were sequenced using an Illumina NextSeq2000 using a paired-end, 50-base read length sequencing strategy.

Selection of mutations for screening:

Screening of mutations from patient X197 for T cell reactivity was described previously.(Veatch et al., 2018) Patient X198 had a large number of mutations (>1100) and variants called by both MuTect and Strelka were filtered for variant allele frequency (VAF) of greater than 20% and only the top 46 mutations by total number variant reads in the RNAseq were selected for screening. Patient X205 had 164 SNVs identified. Variants called by both MuTect and Strelka were filtered for VAF greater than 30% and the top 30 mutations were selected with a mRNA expression level of >14TPM. Patient X422 had 299 identified SNVs, which were filtered for a VAF greater than 25% and the top 33 mutations with mRNA expression above 20 TPM were selected for screening. There were few insertions or deletions resulting in frameshifts that would need be subject to nonsense mediated decay and while these were examined manually, none were selected for screening.

For identifying CD8 epitopes, SNVs from patient X198 mutations that were called by both MuTect and Strelka were filtered for VAF > 0.1 and detection of the variant in at least 5 reads from RNA sequencing. We then used NetMHC version 4.0 to select 169 different 9 and 10 amino acid peptides predicted to bind to HLA-A0101, HLA-A2301 and HLA- B0801 which were then used in screening.

Peptides

Two peptides spanning each mutation with the mutated residue at position +7 or +13 of the 20 amino acid sequence were synthesized by Elim Biopharma and used for initial stimulation and screening to detect CD4+ T cell responses. Subsequent experiments to confirm T cell reactivity only to the mutant and not the wild type peptides were performed with >80% purity 27-mer peptides with the mutant or wildtype amino acid at position 14.

T-cell receptor vector construction

For TCR construction, the vector PRRL(Jones et al., 2009) was modified with six point mutations in the start codon and putative promoter regions of the woodchuck hepatitis virus X protein(Lim and Brown, 2016).The 5’ to 3’ order of the TCR construct was composed of the TCR beta gene, a P2A translational skip sequence, then the TCR alpha gene. To facilitate TCR chain pairing, cysteine residues were introduced as previously described(Kuball et al., 2007). A linear DNA fragment (Life Sciences) was synthesized containing codon-optimized DNA fragments with TRBV, CDR3, and TRBJ which were followed by TRBC sequence containing a cysteine substitution at residue 57, followed by a P2A skip sequence, TRAV, CDR3, TRAJ then TRAC. The DNA fragment was cloned using the NEBuilder cloning kit (New England Biolabs) into PRRL-SIN cut with enzymes PstI and AscI (Thermo Fisher). Sequences were verified using Sanger sequencing. Transduction efficiency into T cells was determined one week later by measuring percent CD3-positive cells in each transduced sample compared to CRISPR-no transduction control cells.

CRISPR-Cas9–mediated gene deletion

CRISPR-Cas9 ribonucleoprotein targeting the first exon of the TCR alpha and TCR beta constant regions were created as previously described (Veatch et al., 2019) using equal volumes of 80 μmol/L TracRNA (IDT), 80 μmol/L TrbcRNA (IDT) with 80 μmol/L of the gRNA AGAGTCTCTCAGCTGGTACA in duplex buffer (IDT), mixing well and incubating in a heating block at 95°C for 5 minutes and allowing the mixture to slowly cool. The resulting 40 μmol/L duplexed RNA was mixed with an equal volume of 24 μmol/L Cas9 protein (IDT) and 1/20 volume of 400 μmol/L Cas9 electroporation enhancer (IDT) and incubated at room temperature for 15 minutes prior to electroporation.

On day 0, CD4⁺ T cells from a patient, who provided informed consent on an IRB-approved protocol, were thawed from liquid nitrogen stocks and stimulated with anti-CD3/anti-CD28 microbeads at a 3:1 bead:cell ratio (Dynabeads, Invitrogen) in CTL supplemented with IL2 (50 U/mL) and IL7 (5 ng/mL) for 2 days. Also, on day 0, Lenti-X cells (Clontech) were transiently transfected with each TCR vector, as well as psPAX2 (Addgene plasmid no. 12260) and pCMV-VSV-G (Addgene plasmid no. 8454) packaging plasmids. On day +2, magnetic beads were removed, and cells were nucleofected using a Lonza 4D nucleofector in 100 μL of buffer P3 using program EH-115. Cells were allowed to rest for 4 hours in media prior to lentiviral transduction. Lentiviral supernatant was harvested from Lenti-X cultures, filtered using 0.45-μm polyethersulfone (PES) syringe filters (Millipore), and added to activated T cells in a 48-well tissue culture plate. Polybrene (Millipore) was added to a final concentration of 4.4 μg/mL, and cells were centrifuged at 800 × g and 32°C for 90 minutes. Sixteen hours later, viral supernatant was replaced with fresh CTL containing IL2 (50 IU/mL) and IL7 (5 ng/mL). Half-media changes were then performed every 48 to 72 hours using CTL containing IL2 and IL7. Transduced T cells were assessed for transduction efficiency using antibodies specific to CD3 and grown in a rapid expansion protocol described previously 10 to 13 days prior to conducting immune assays.

Antigen specificity assays

ELISpot assays were performed by incubating 50,000 T cells from cultures with 100,000 peptide pulsed (10mg/ml or otherwise indicated) and control autologous B cells as APC overnight using the human interferon gamma ELISpot kit (Mabtech) following the manufacturer’s instructions. ELISA was performed by incubating 50,000 T cells with 100,000 peptide pulsed and control autologous B cells and IFN-g release in the supernatant was quantitated using the human interferon gamma ELISA kit (Thermo fisher).

Intracellular stain for CXCL13 and FoxP3

Tumor samples were thawed and stained immediately for surface markers followed by staining for intracellular markers. Briefly, cells were washed with plain PBS and stained with Fixable Live/Dead Aqua followed by staining with surface antibodies for anti-human CD4, PD-1, and CXCR6. Thereafter, cells were fixed/permeabilized using FOXP3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific, #00-5523-00) and stained with CXCL13 and FoxP3 antibodies.

B cell proliferation/TFH assay

For B cell stimulation assays, allogeneic B cells were isolated from the peripheral blood of a healthy donor using the human B cell positive selection release kit (StemCell) and rested overnight in B cell medium prior to labelling with cell trace violet (Thermofisher Scientific) and coculture of with 1:1 rested CD4⁺ T cells. B cell proliferation was measured at day 5 by flow cytometry.

Tumor single cell suspensions were thawed and labelled with LIVE/DEAD^™ Fixable Aqua Dead Cell Stain Kit (Thermofisher Scientific, cat# L34957) as per manufacturer’s protocol. We incubated the cells in Fc block at RT for 5mins, then stained them with fluorescently labelled antibodies CD3 BV711, CD4 BUV395, PD-1 FITC, CD25 PE-Cy7, CD127 BV421, CD200 PE, CXCR6 APC at 4degC for 30mins. Next, CD3⁺ CD4⁺ T cells were sorted into a CD127⁻ CD25⁺ Treg population, and conventional T cells were separated into PD-1 high and PD-1 low CD127⁺ high populations (termed “bystanders”) and PD-1 high cells were separated into CXCR6⁻ TFH cells and CXCR6⁺ “effector” cells using a symphony S6 cell sorter. Sorted cells were rested overnight in CTL+ 50U/mL IL-2 and 5ng/ml IL-7 prior to assays. On the same day, B cells were isolated from thawed PBMCs of a healthy donor using. Isolated B cells were rested overnight in IMDM supplemented with 10% human serum, penicillin-streptomycin, and L-glutamine (termed BCM). After overnight culture, B cells were labelled with CellTrace Violet Cell Proliferation Kit (Thermofisher Scientific, cat# C34571) per manufacturer’s protocol, then cocultured with each of the different T cell subsets in the presence of SEB 100ng/ml. Proliferation was analyzed after 5 days by flow cytometry.

T cell proliferation assay

Tumor single cell suspensions from were thawed and CD45⁺ cells were enriched using EasySep^™ Release Human CD45 Positive Selection Kit (StemCell Technologies, cat#100–0105) based on manufacturer’s protocol. We then added Fc block to the isolated cells and incubated RT for 5mins, then stained them with fluorescently labelled antibodies antibodies CD3, CD4, PD-1, CD25, CD127, CD200, CXCR6 at 4 °C for 30mins. Next, we sorted the different CD4 T cell subsets (Effectors, TFH, Bystanders) using BD FACSymphony S6 System and cultured the sorted cells overnight in CTL media supplemented with IL-2 50U/ml and IL-7 5ng/ml. After resting the cells overnight, T cells were labelled using CellTrace Violet Cell Proliferation Kit (Thermofisher Scientific, cat# C34571) per manufacturer’s recommendations. Next, we added CD3/CD28 Dynabeads to the labelled cells at 3:1 bead to cell ratio in CTL, and cultured cells at 37 °C. After 24hours, IL-2 10U/ml was added to the cells and cultured for 4 days. On day 4, we removed the Dynabeads and stained cells with antibodies CD3 BV711, CD4 BUV395, CD19 PE and Proliferation was analyzed using flow cytometry. Relative proliferation of cell fractions was compared between patients using a ratio paired t-test.

TCRVb sequencing

DNA from tumors, sorted T cells, PBMC, or T cell cultures was analyzed for TCRVb repertoire using the human TCRB kit (Adaptive Biotechnologies) and analyzed using company software. A single PCR reaction was used for analyzing oligoclonal cell lines and small populations of sorted cells, 2 PCR reactions (survey depth) were used for PBMC cultures, and 4 PCR reactions per sample were used for analysis of PBMC and primary tumor samples. TCRVb clonotypes identified as antigen specific had to meet the following criteria: 1) detected with at least 2 templates in both replicates of a specific stimulation and 2) both replicates of the specific stimulation had to be greater than the sum of all other stimulations. All primary TCRVb sequencing data will be made available at the Adaptive Biotechnologies website https://clients.adaptivebiotech.com/immuneaccess.

Single-cell capture and cDNA library preparation for single cell RNA sequencing

Single-cell libraries were prepared using the BD Rhapsody Express system (BD Biosciences, #633707) and Targeted mRNA and AbSeq Reagent Kit-4 pack (BD Biosciences, #633771) according to the manufacturer’s protocol (BD Biosciences). Briefly, tumor samples from each donor were thawed and labelled with sample tags using BD Single-Cell Multiplexing Kit (BD Biosciences, #633781). Cells from each donor were labelled with a unique sample tag. Cells were then washed in BD Stain buffer (BD Biosciences, #554656), pooled together and incubated in Fc block followed by labelling cells with BD AbSeq Ab-Oligos master mix (3uL per Ab-Oligo). For flow sorting various lymphoid and myeloid cell populations, cells were stained with a mixture of fluorescently labelled antibodies (BD Biosciences or BioLegend): anti-human CD3 (#OKT3), CD4 (#RPA-T4), CD8 (#RPA-T8), CD19 (#HIB19), CD45 (#2D1), for 30 mins at 4°C. After sorting, cells were counted and pooled such that there were 20,000 cells resuspended in 620 mL of BD Sample buffer. The BD Rhapsody cartridges were primed, and the pooled cells were loaded onto the cartridges and incubated at room temperature. Next, Cell Capture beads were washed & loaded onto the cartridge. After incubation, cartridges were washed twice with BD Sample Buffer. This was followed by lysis of cells and retrieval of beads. After washing the retrieved Cell Capture beads, we proceeded immediately with reverse transcription. For the subsequent steps, we switched to BD Rhapsody system VDJ CDR3, Sample Tag, and BD AbSeq library protocol (BD Biosciences) to generate TCR and BCR libraries in addition to mRNA, AbSeq, Sample Tag libraries. Poly-T Template Switching Oligo (TSO) was added during reverse transcription reaction to allow identification of VDJ recombination events in B and T cells. This was followed by denaturation, hybridization, Klenow extension (New England Biolabs, #M0212L) and treatment with Exonuclease-I.

cDNA Libraries were prepared in a two-step nested PCR reaction followed by Index PCR using BD Rhapsody Targeted mRNA and AbSeq Amplification Kit (BD Biosciences, #633774) and BD Rhapsody^™ Immune Response Panel Hs (BD Biosciences, #633750). We included BD Rhapsody Supplemental Panel (BD Biosciences, #633742) to look at additional genes GATA3, MAF, CCR6, CD40L, TOX, IL6R, IL9R and IL10. After the first PCR, longer PCR products (mRNA, TCR, BCR) were separated from shorter products (AbSeq & Sample Tag) based on double-sided size selection using Agencourt AMPure XP beads (Beckman Coulter, #A63880). The concentration of libraries was estimated using Qubit^™ dsDNA HS Assay Kit (Thermo Fisher Scientific, #Q32851). The products from second PCR were diluted as per protocol’s recommendations for Index PCR. The quality check and quantification of Index libraries was performed using Agilent 2200 TapeStation with High Sensitivity D5000 ScreenTape (Agilent).

Genes in the targeted panel and oligonucleotide labelled antibodies used are shown in supplementary table S6.

Sequencing was performed on a Novaseq SP flow cell at 75 × 225 and after demultiplexing, sample tags, mRNA, and antibody sequencing reads were trimmed to 75nt and TCR and BCR VDJ libraries were kept at 225nt and FASTQ files were uploaded to the BD Rhapsody analysis pipeline (https://www.sevenbridges.com/bdgenomics/) with refined cell calling disabled. For the initial analysis of 4 patients, CD4⁺ T cells were sorted and sequenced. For the analysis of the full cohort of 20 patients, CD19⁺ and CD3⁺ cell populations were labelled and sequenced separately from CD45⁺ CD19⁻ CD3⁻ myeloid populations to facilitate greater cell detection and sequencing depth of the lymphoid cells.

Immunohistochemistry

Formalin-fixed paraffin-embedded tissues were sectioned at 4 microns onto positively-charged slides and baked for 1 hour at 60°C. The slides were then dewaxed and stained on a Leica BOND Rx stainer (Leica, Buffalo Grove, IL) using Leica Bond reagents for dewaxing (Dewax Solution), antigen retrieval/antibody stripping (Epitope Retrieval Solution 2), and rinsing after each step (Bond Wash Solution). Antigen retrieval and antibody stripping steps were performed at 100°C with all other steps at ambient temperature.

Endogenous peroxidase was blocked with 3% H₂O₂ for 5 minutes followed by protein blocking with TCT buffer (0.05M Tris, 0.15M NaCl, 0.25% Casein, 0.1% Tween 20, 0.05% ProClin300 pH 7.6) for 10 minutes. The first primary antibody (position 1) was applied for 60 minutes followed by the secondary application for 10 minutes and the application of the tertiary TSA-amplification reagent (PerkinElmer OPAL fluor) for 10 minutes. A high stringency wash was performed after the secondary and tertiary applications using high-salt TBST solution (0.05M Tris, 0.3M NaCl, and 0.1% Tween-20, pH 7.2–7.6). Species specific HRP polymer was used for all secondary applications.

The primary and secondary antibodies were stripped with retrieval solution for 20 minutes before repeating the process with the second primary antibody (position 2) starting with a new application of 3% H2O2. The process was repeated until seven positions were completed. For the eighth position, following the secondary antibody application, Opal TSA-DIG was applied for 10 minutes, followed by the 20 minute stripping step in retrieval solution and application of Opal 780 fluor for 10 minutes with high stringency washes performed after the secondary, TSA-DIG, and Opal 780 fluor applications. The stripping step was not performed after the final position.

Slides were removed from the stainer and stained with DAPI for 5 minutes, rinsed for 5 minutes, and coverslipped with Prolong Gold Antifade reagent (Invitrogen/Life Technologies, Grand Island, NY).

Slides were cured overnight at room temperature, then whole slide images were acquired on the Vectra Polaris Quantitative Pathology Imaging System (Akoya Biosciences, Marlborough, MA). The entire tissue was selected for imaging using Phenochart and multispectral image tiles were acquired using the Polaris. Images were spectrally unmixed using Phenoptics inForm sotware and exported as multi-image TIF files, which were analyzed with HALO image analysis software (Indica Labs, Cooales, NM). Cellular analysis of the images was performed by first identifying cells based on nuclear recognition (DAPI stain), then measuring fluorescence intensity of the estimated cytoplasmic areas of each cell. A mean intensity threshold above background was used to determine positivity for each fluorochrome within the cytoplasm, thereby, defining cells as either positive or negative for each marker. The positive cell data was then used to define colocalized populations.

Single-cell data preprocessing and analysis

Data preprocessing and analysis was performed primarily using the Scanpy toolkit(Wolf et al., 2018) in Python. Low quality cells with fewer than 100 total counts and 10 expressed genes were removed. Doublets were first removed with Scrublet,(Wolock et al., 2019) using its automatic doublet detection threshold. CPM normalization was performed with the Scanpy ‘normalize_total’ function, and the data was then log-transformed. Sample tags were extracted from BD Rhapsody analysis pipeline (https://www.sevenbridges.com/bdgenomics/), and cells that had been identified as multiplets or that were labeled as ‘undetermined’ were removed. Batch correction to correct for variability among patients was performed on mRNA data using ComBat(Johnson et al., 2007; Tangherloni et al., 2021) implemented in Scanpy.

Dimension reduction was first performed by principal component analysis and then by UMAP after construction of a nearest neighbors graph using the first 20 principal components. This was performed on mRNA data alone. Unsupervised clustering was performed using the Leiden algorithm, implemented in Scanpy. The resulting clusters were manually labeled according to top differentially expressed genes in each cluster; clusters containing tumor cells or evident doublets not captured by Scrublet were discarded. Differential expression between clusters was assessed using MAST(Finak et al., 2015) with a threshold of FDR of 0.05. Heatmaps were generated from the top ten differentially expressed genes from each cluster as calculated with MAST.

TCRVb analysis was performed using some functionality from Scirpy(Sturm et al., 2020) to assess clonal expansion and calculate clonotype size. The Scirpy ‘repertoire_overlap’ function was used to calculate the Jaccard similarity coefficients based on clonotype representation between clusters.

TCGA analysis.

Gene expression data was extracted from the TCGA-SKCM (Skin Cutaneous Melanoma) data set using the TCGAbiolinks package (Colaprico et al., 2015). Relevant genes were extracted and normalized to CD4. Spearman correlations and corresponding p-values were calculated between genes of interest, with false discovery rates estimated using the Benjamini-Hockberg multiple testing correction. Correlation plots were generated using Gene Expression Profiling Interactive Analysis (GEPIA, http://gepia.cancer-pku.cn/) (Tang et al., 2017)

Breast cancer data analysis

Single cell RNA-seq data was downloaded from cohort 1 of (Bassez et al., 2021). The provided cell metadata was used to identify B cells, CD4 and CD8 T cells, and myeloid cells. For each broad cell type, data was processed with the standard Scanpy (Wolf et al., 2018) pipeline (log-normalization, highly variable genes extraction, PCA, and UMAP transform). Leiden clustering was performed, and differential expression heatmaps (generated with the ‘rank_genes_groups’ function in Scanpy) were used to identify cell subtypes from the Leiden clusters.

Proximity analysis of immunohistochemistry

Images were analyzed using digital image analysis software (HALO v3.1, Indica Labs, Corrales, NM). Briefly, specific area(s) within the image were annotated for analysis; for X197 lymph node metastasis section, a layer with total classified area of 29.53mm² was annotated based on Mel-Sox10 staining, and for X198 tumor section, a randomly selected layer with total classified area of 42.77mm² was annotated for analysis. For analysis (algorithm: Indica Labs- HighPlex FL v3.2.1), settings for each stain/marker were tuned to identify cell phenotypes defined using one or more markers (e.g., CD4+ CXCL13+ defined as CD3+ CD8− CXCL13+). We then screened randomly selected regions across the image for positive cells for a given phenotype using realtime tuning. Finally, the analysis algorithm was run on the annotation layer to obtain the number of cells corresponding to each of the defined phenotypes. For proximity analysis, we used the Spatial Analysis Module in HALO to quantify the number of cells within 100um distance of another cell type.

For the second panel, slides were stained with CD3, CD8, CD79A, Mel-Sox10, FoxP3, PD-1, Tcf7, DAPI. Areas rich in B cells were identified & marked based on CD79A staining using the tissue classifier feature. Annotations were analyzed to compare the distribution of different phenotypes within the B cell rich versus non-Bcell rich regions. The enrichment of different cell populations within B-cell rich regions in tumors containing these regions was measured across the patient cohort using a ratio paired t-test.

Quantitation and statistical analysis

Differential expression of genes and surface proteins from single cell sequencing were determined using MAST as above. Other statistical testing was done using Prism software (Graphpad). Survival analysis was conducted using the log rank test. Correlation analyses were performed using spearman correlation and for multiple testing a false discovery rate was determined by the Benjamini-Hochberg procedure. T cell proliferation between different cell populations from different patients was performed with a paired T test. Overlap of gene sets from CD4⁺ and CD8⁺ T cell populations was determined using a fishers exact test, and comparisons between proximity of T cell populations to B cells were done with the Kruskal-Wallis test with Dunn’s correction.

Supplementary Material

2

Table S1: Exome and RNA sequencing of 5 patients with screened peptides (related to figure 1)

3

Table S3: Summary of differentially expressed mRNA genes and surface markers between clusters (Related to figure 2)

4

Table S2: Summary file of TCRVb sequencing clonotypes (related to figure 1)

6

Table S7: Gene and antibody panels (related to STAR methods)

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-human CD3	Biolegend	cat#317328; RRID:AB_2562907
Anti-human CD4	Biolegend	cat#564724; RRID:AB_2738917
Anti-human CD8	Biolegend	cat#301014; RRID: AB_314132
Anti-human CD19	Biolegend	cat#302208; RRID:AB_314238
Anti-human CD25	Biolegend	cat#302612; RRID:AB_314282
Anti-human CD38	Biolegend	cat#356617; AB_2566230
Anti-human CD45	Biolegend	cat#304014; RRID:AB_314402
Anti-human CD56	Biolegend	cat# 362545; RRID:AB_2565963
Anti-human PD-1	Biolegend	cat#329904; RRID:AB_940479
Anti-human CD127	Biolegend	cat#351309; RRID:AB_10898326
Anti-human CD154	Biolegend	cat#310805; RRID:AB_314828
Anti-human CD200	Biolegend	cat#329306; RRID:AB_2074200
Anti-human CXCR6	Biolegend	cat#356006; RRID:AB_2562223
Anti-human IFN-γ	Biolegend	cat#502511; RRID:AB_315236
Anti-human CXCL13	R&D Systems	cat#IC801P; RRID:AB_2086047
Anti-human FoxP3	Thermo Fisher	cat#48477642; RRID:AB_1834364
TCR V beta 3.1	Thermo Fisher	cat#TCR2740; RRID:AB_223625
Anti-FLAG Tag	Biolegend	cat#637310; RRID:AB_2563148
Cell Trace Violet stain kit	Thermo Fisher	cat#C34557
LIVE/DEAD™ Fixable Aqua Dead Cell Stain Kit	Thermo Fisher	cat# L34957
Biological samples
Human tumor infiltrating lymphocytes	This paper	See Table S5
Human melanoma tumor tissues	This paper	See Table S5
Human peripheral blood mononuclear cells	This paper	See Table S5
Human primary B cells	This paper	See Table S5
Chemicals, peptides, and recombinant proteins
Human T-Activator CD3/28 Dynabeads	Thermo Fisher	cat #11132D
Anti-CD3 (OKT3)	Thermo Fisher	cat #48-0037-42
Recombinant human IL-2	Peprotech	cat# 200–02
Recombinant human IL-4	Peprotech	cat# 200–04
Recombinant human IL-7	Peprotech	cat#200–07
Recombinant human IL-15	Peprotech	cat#200–15
Custom patient specific peptides	Elim Biopharma	custom
Peptivator NY-ESO	Milltenyi	130-095-380
Peptivator MAGE-A1	Milltenyi	130-095-383
Peptivator MAGE-A3	Milltenyi	130-095-384
Peptivator Tyrosinase	Milltenyi	130-094-446
Peptivator GP100	Milltenyi	130-094-449
Peptivator MART-1	Milltenyi	130-094-597
Peptivator MAGE-A4	Milltenyi	130-095-387
CPI viral positive control solution	Immunospot	CTL-CPI-001
Pepmix human TRP2	JPT	PM-TRP-2
Golgi Stop	BD Biosciences	cat# 554724; RRID:AB_2869012
Golgi Plug	BD Biosciences	cat# 555029; RRID:AB_2869014
anti-CD40	Milltenyi	Cat#130-094-133
anti-CD28	BD Biosciences	Cat# 555725
anti-CD49a	BD Biosciences	Cat# 556634
Critical commercial assays
Human TCRB kit	Adaptive Biotechnology	cat#isk10101
EasySep Human CD45 Depletion Kit	StemCell	cat#100–0105
EasySep Human CD4+ T Cell Isolation Kit	StemCell	cat#17952
AllPrep DNA/RNA Mini Kit	Qiagen	cat#80284
DNEasy Blood & Tissue kit	Qiagen	cat# 69504
QIAamp DNA Micro Kit	Qiagen	cat# 56304
Agilent SureSelectXT Reagent Kit	Agilent	cat# G9611B
TruSeq RNA Sample Prep v2 Kit	Illumina, Inc., San Diego, CA, USA	cat# RS-122–2001
BD Cytofix/Cytoperm kit	BD Biosciences	cat# 554714
FOXP3/Transcription Factor Staining Buffer Set	Thermo Fisher Scientific	cat#00-5523-00
Human IFN-g ELISA kit	Thermo fisher	cat#EHIFNG
Human IFN-g ELISpot kit	Mabtech	cat# 3420–4APT-2
Qubit™ dsDNA HS Assay Kit	Thermo Fisher Scientific	cat# Q32851
Human CD19 microbeads	Miltenyi	cat# 130-050-301
Deposited data
TCRVb sequencing data	This paper	https://clients.adaptivebiotech.com/pub/veatch-2022-cc.
Single cell RNAseq data	This paper	GEO: GSE198265
Analysis code and scripts to generate figures	This paper	https://github.com/cshasha/veatch_2022
Single cell RNAseq data	Bassez et al. Nature Medicine 27.5 (2021): 820–832	https://ega-archive.org/datasets/EGAD00001006608
Experimental models: Cell lines
NIH 3T3-CD40L	Brian Till
Lenti-X 293T	Clontech	Cat #632180
Plat-E	Cell Biolabs	RRID: CVCL_B488
Oligonucleotides
Poly-T Template Switching Oligo (TSO)	IDT	(5’ TTTTTTTTTTTTT TTTTTTTTTTTTrGrGrG 3’)
Recombinant DNA
psPax2	Addgene	Addgene Plasmid #12260
pMD2.G (VSVg)	Addgene	Addgene Plasmid #12259
Software and algorithms
MuTect 1.1.7	Cibulskis et al.	https://software.broadinstitute.org/cancer/cga/mutect
Strelka 1.0.14	Saunders et al.	https://github.com/Illumina/strelka
Scanpy toolkit	Wolf et al.	https://scanpy.readthedocs.io/en/stable/
FlowJo v10	TreeStar	https://www.flowjo.com/solutions/flowjo/downloads
Prism	GraphPad	https://www.graphpad.com/scientific-software/prism/
HALO Image Analysis Software	Indica Labs	http://www.indicalab.com/halo/
Library preparation and sequencing reagents
BD Rhapsody Express system	BD Biosciences	cat#633707
Targeted mRNA and AbSeq Reagent Kit-4 pack	BD Biosciences	cat#633771
BD Single-Cell Multiplexing Kit	BD Biosciences	cat# 633781
BD Rhapsody Supplemental Panel	BD Biosciences	Cat# 633742
Agencourt AMPure XP beads	Beckman Coulter	Cat# A63880
BD Stain buffer	BD Biosciences	cat# 554656
Klenow fragment	New England Biolabs	cat# M0212L
SMART-Seq HT Kit	Takara Bio	Cat#634438
Illumina Nextera XT DNA kit	Illumina	Cat#15032354
IDT for Illumina UDI primers	Illumina	cat# 20026930

Highlights:

Tumor antigen specific CD4⁺ T cells express CXCL13 and PD-1 in human melanoma

A proliferative T_FH-like subset is colocalized with B cells in the tumor microenvironment

A second CXCL13⁺ subset expresses Tim-3, cytolytic markers and interferon gamma

CXCL13⁺ CD4⁺ T cells correlate with survival and macrophage, CD8⁺ T and B cell activation

Veatch et al. identify the phenotype of tumor-specific conventional CD4⁺ T cells infiltrating human melanoma, and find these cells form distinct functional states resembling T follicular helper cells or cytolytic cells. The frequency of tumor-specific CD4⁺ T cells correlates with survival and activation of intratumoral macrophages, CD8⁺ T cells and B cells.