Case of complete response to immunotherapy in MMR-deficient prostate cancer associated with NK-like and CD4+CD8+ T cells Authors: Alexander K Tsai, John R Lozada, Philippa R Kennedy, David Moline, Rachana Pandey, Riley C Lyons, Christine Luo, Rulin Wang, Ali T Arafa, Elise L Femino, Sarah Zipkowitz, Alexis Figueroa, Patrick J McCann, Matthew C Dallos, Andrew Elliott, Paari Murugan, Martin Felices, Nicholas A Zorko, Badrinath R Konety, Scott M Dehm, Jeffrey S Miller, Steven S Shen, Elizabeth A Thompson, Laura A Sena, Srinivasan Yegnasubramanian, Justin Hwang, Emmanuel S Antonarakis Published: 2026 ABSTRACT SummaryMismatch repair deficiency (dMMR) and microsatellite instability (MSI-H) are rare in prostate cancer, occurring in 2%–4% of cases. These defects result in increased genomic instability and elevated tumor mutational burden (TMB), which can support responses to immune checkpoint inhibitors (ICIs). Here, we report a patient with locally advanced Gleason 5 + 5 = 10 prostatic adenocarcinoma harboring MSH2 and MSH6 genomic deletions with ultrahigh TMB (>250 mutations/megabase) in whom pembrolizumab resulted in a striking complete radiographic, pathologic, and molecular response. Using digital-spatial microscopy, single-cell RNA/T cell receptor (TCR) sequencing, and multiplex cytometry, we identify atypical tumor-infiltrating T cells with natural killer-like phenotypes and CD4+CD8+ (double-positive) lymphocytes. These clonal T cell populations expand preferentially following ICI and adopt terminally differentiated and cytotoxic profiles that may drive clinical response. Similar T cells are also present in diverse cancers and expand exclusively in ICI-responsive patients. These findings inform on the cellular mechanisms by which immunotherapies may mediate profound responses in patients with dMMR solid tumors.Keywords: prostate cancer, dMMR, MSI-high, TMB, pembrolizumab, complete pathologic response, T cell, cytotoxic T lymphocyte, immunotherapy, immune checkpoint inhibitor Introduction Although the prognosis of early-stage prostate cancer (PC) is generally favorable, high-grade prostatic malignancies are associated with significantly poorer outcomes.1,2 Despite advances in local and systemic therapies, recurrences following curative-intent treatment are frequent in Gleason 5 + 5 = 10 PC.1,2 Immune checkpoint inhibitors (ICIs) have been investigated extensively for metastatic PC and are generally ineffective in molecularly unselected patients.3 However, rare PC patients with mismatch repair deficiency (dMMR)/microsatellite instability (MSI-H) and/or high tumor mutational burden (TMB) occasionally respond favorably to ICI.4,5 The mechanisms governing ICI sensitivity continue to evolve, and whether patients with dMMR/MSI-H malignancies exhibit unique immunological responses to ICI remains unknown. Here, we report a remarkable case of a patient with locally advanced, Gleason 5 + 5 = 10, dMMR/MSI-H PC with ultrahigh TMB, who achieved a complete pathological and molecular response with pembrolizumab that was associated with clonal expansion of natural killer (NK)-like and double-positive (DP) CD4+CD8+ T cells. These NK-like and DP T cells expanded exclusively in ICI-responsive patients when examining larger PC and non-prostatic dMMR/MSI-H cohorts. Results Clinical case history, genomic analysis, and histologic evaluation of tumor-infiltrating immune cells A 67-year-old man presented with pelvic pain, hematuria, and hematochezia. MRI and prostate-specific membrane antigen (PSMA)-positron emission tomography (PET)/computed tomography (CT) imaging revealed an 8-cm prostatic mass with local invasion into the bladder and rectum, but without distant metastases (Figures 1A and 1B). Prostate biopsy demonstrated sheets of malignant cells with near-complete replacement of normal glands, indicative of Gleason 5 + 5 = 10 acinar adenocarcinoma, involving 100% of all cores (Figure 1C). This represented stage IIIC (cT4N0M0) disease. Clinical case history, genomic analysis, and histologic evaluation of tumor-infiltrating immune cells (A) T2-weighted transverse view of prostatic MRI performed prior to treatment during diagnostic workup. Measurements of the large prostatic mass are shown along with rectal (∗) and bladder (+) invasion. (B) 18F-PSMA-PET/CT imaging performed prior to treatment during diagnostic workup. (C) Representative H&E section of Gleason 5 + 5 = 10 (Grade Group 5) prostatic acinar adenocarcinoma from a pre-treatment prostate biopsy. Scale bar, 100 μm. (D) Serum prostate-specific antigen (PSA) measurements from the patient following diagnosis. Therapies administered are shown (D, degarelix). Serial peripheral blood mononuclear cell (PBMC) collections analyzed in downstream assays are also depicted. (E) T2-weighted transverse view of prostatic MRI performed after four cycles of pembrolizumab. (F) 68Ga-PSMA-PET/CT imaging performed after four cycles of pembrolizumab. (G) Representative H&E section of the post-ICI prostatectomy sample with chronic inflammation and fibrosis along with atrophic glands. Prostatic adenocarcinoma was not detected. Scale bar, 50 μm. (H) Histogram of tumor mutational burden (TMB) from 584 dMMR (assessed by IHC) and/or MSI-H (assessed by DNA sequencing) prostate cancer patients from the Caris Life Sciences CODEai database. ∗, median TMB (3 mutations/megabase) from all PC patients (including pMMR/MSS patients) within this database. +, patient TMB (266 mutations/megabase). (I) Pathogenic and likely pathogenic DNA alterations detected from whole-exome NGS analysis performed on diagnostic PC biopsies with variant allele frequencies (VAFs). Likely pathogenic variants and a variant of unknown significant (VUS) in POLE are indicated; all other variants were classified as pathogenic. (J) Number of mutations matching the four Catalog of Somatic Mutations in Cancer (COSMIC) single-base-substitution (SBS) mutational signatures identified: SBS5, SBS6, SBS21, and SBS42. SBS6 and SBS21 are associated with dMMR/MSI-H defects. SBS5 has unknown etiology, whereas SBS42 is associated with exposure to haloalkanes. (K) Number of somatic mutations per megabase based on COSMIC SBS mutational signature. (L) Characteristic substitutions associated with each signature, and their prevalence among all variants. No variants were associated with DNA polymerase deficiency COSMIC signatures. See also Figure S1 and Table S1. Somatic whole-exome next-generation sequencing (NGS) of the tumor revealed dMMR/MSI-H disease harboring homozygous genomic deletions of MSH2 and MSH6 loci with accompanying protein loss. Germline genetic testing was unremarkable, without evidence for Lynch syndrome. Androgen deprivation therapy (ADT) was initiated using degarelix 240 mg followed by leuprolide 45 mg and abiraterone 1,000 mg daily. While prostate-specific antigen (PSA) initially declined, the patient rapidly developed castration-resistant PC after only 3 months (Figure 1D). At that time, pembrolizumab (200 mg intravenously, in 21-day cycles) was administered for six cycles. This resulted in a brisk PSA decline to undetectable levels after two ICI cycles. Repeat MRI after four cycles of pembrolizumab revealed a normal-sized prostate gland measuring 3.3 × 3.2 × 2.4 cm with post-treatment fibrotic changes but without evidence of residual tumor (Figure 1E). No focal radiotracer avidity was detected on repeat PSMA-PET/CT imaging after four cycles of ICI, indicative of a complete radiographic response (Figure 1F). Radical prostatectomy with pelvic lymphadenectomy was subsequently performed 1 month after the final pembrolizumab dose. Intraoperatively, significant fibrosis was noted between the prostate and rectum. Strikingly, histopathological examination revealed only glandular prostatic atrophy, fibrosis, and chronic inflammation without residual adenocarcinoma (Figure 1G). Five pelvic nodes were also negative for carcinoma. Repeat whole-exome NGS from the post-ICI prostatectomy specimen, which detects single-base substitutions (SBSs) derived from samples with 1%–2% tumor involvement, did not identify any somatic DNA alterations detected in the pre-treatment tumor biopsy. Thus, the patient was deemed to have a complete radiographic, pathologic, and molecular response. No additional adjuvant therapies were administered. The patient remains recurrence free, biochemically and radiographically, more than 18 months following prostatectomy with non-castrate testosterone levels. Genomic, transcriptomic, cytometric, and spatial imaging approaches were used to serially profile immunological evolution in response to ICI. Whole-exome sequencing analysis (Table S1) revealed a TMB of 266 mutations/megabase—approximately 10-fold higher than the median TMB identified in a large cohort of 584 dMMR/MSI-H PCs (Figure 1H). In addition to MSH2/MSH6 deletions, somatic pathogenic variants were present in several other genes involved in DNA damage/response including ATM, KMT2C, MSH3, RAD51C, PRKDC, and SETD2 (Figure 1I), which may have contributed to the markedly elevated TMB. Mutational signature analysis was performed given that a variant of unknown significance (VUS) was also identified in DNA polymerase epsilon (POLE), which could have further influenced mutation accumulation. Catalog of Somatic Mutations in Cancer (COSMIC) SBS6 and SBS21 mutational signatures, associated with dMMR/MSI-H, accounted for 61% of the genomic alterations identified by NGS on pre-treatment biopsies (Figures 1J–1L). Polymerase deficiency mutational signatures were not detected, suggesting that the POLE VUS represented a passenger mutation. Digital spatial microscopy with co-detection by indexing (CODEX) was performed on pre-treatment (tumor biopsy) and post-ICI (prostatectomy) specimens to interrogate microenvironment changes. Immune infiltration patterns varied distinctly in the two specimens: while diffuse infiltration characterized pre-treatment biopsies, immune cells in the prostatectomy specimen were confined to glands and were virtually absent in fibrotic zones that developed following ICI-induced tumor eradication (Figures S1A and S1B). Cycling CD45+ leukocytes were observed exclusively within biopsy tissue, and ∼40% of tumor cells were also actively proliferating (Figure S1C). PD-L1 was detected only in biopsies and expressed primarily on CD11c+ myeloid cells (Figure S1D). Regarding lymphocytes, while tissue-infiltrating T cells were comparable in both samples, antigen-experienced CD45RO and PD-1-expressing CD4+ and CD8+ T cell subsets were enriched >2-fold in pre-treatment biopsies, as were proliferating Ki-67+ T cells (Figures S1D and S1E), suggesting the development of tumor-specific T cells that contracted following therapy. Additionally, atypical T cell phenotypes were identified in tumor tissues. Specifically, CD56, a marker of NK and NKT cells, was detected on “NK-like” CD8+ T cells (Figure S1F). Unconventional co-expression of CD4+ and CD8+ was also observed in a subset of DP T cells (Figure S1G). These unique T cells expressed PD-1 in pre-treatment biopsies, suggestive of tumor specificity. Clonal NK-like and CD4+CD8+ double-positive T cells adopted signatures of effector functionality Single-cell RNA sequencing (scRNA-seq) was performed on peripheral blood mononuclear cells (PBMCs) from pre-treatment (PBMC Pre) and initial post-ICI (PBMC Post #1) time points (Figures 1D, S2A, and S2B), given an expectation that pivotal immunological changes would occur shortly following ICI initiation. NK cells and T cells segregated into 11 clusters and were interrogated more deeply due to their frequent association with ICI response (Figure 2A). Most CD8+ T cell clusters (C1, C2, C5, and C9) resembled effector memory (TEM) subsets that curiously expressed various NK-associated genes (Figures 2A and S2C). Signatures of T cell function6 were similar between CD8+ TEM and non-anergic NK cell clusters (Figure 2B), reinforcing that TEM populations adopted gene expression patterns similar to NK cells. Aligning with CODEX findings, NCAM1+ (encoding CD56) NK like T cells and CD4+CD8A+ co-expressing DP T cells were identified (Figure 2C). These unique T cell subsets exhibited preferential expansion following ICI treatment compared with the overall CD8+ T cell population (Figure 2D). The NK-like and DP T cells predominantly populated clusters C2 and C5, respectively (Figure 2E). Additionally, signatures of T cell function, including glycolysis, chemokine receptor signaling, T cell receptor (TCR) signaling, effector function, and cytotoxicity, were augmented in the NK-like (C2) and DP (C5) T cells following ICI administration (Figure 2F). Overall, effector memory NK-like and DP T cells expanded after ICI, suggesting that these T cell subsets might contribute to anti-tumor immunity. Clonal NK-like and CD4+CD8+ double-positive T cells adopted signatures of effector functionality (A) Single-cell RNA sequencing showing hierarchical NK cell and T cell clusters by uniform manifold approximation and projection (UMAP) from pre-ICI and post-ICI time point #1 (see Figure 1D). Cell clusters were annotated as CD8 effector memory T cells (TEM; C1, C2, C5, and C9), CD8 anergic T cells (C4), naive/central memory CD4 T cells (TNAIVE/CM; C0), cycling CD4 T cells (TCYCLING; C10), regulatory CD4 T cells (TREG; C7), cytotoxic NK cells (NKCYTOTOXIC; C3), adaptive NK cells (NKADAPTIVE; C6), and anergic NK cells (NKANERGIC; C8). Defining genes for each cluster are listed. (B) T cell function gene signature scores (see Chu et al.6) for NK cell and T cell clusters (pre- and post-ICI time points are combined). (C) CD4 (left), CD8A (middle), and NCAM1 (right) gene expression (pre- and post-ICI time points are combined). (D) Frequencies and fold change of CD8+ T cells (left), CD8A+NCAM1+ NK-like T cells (middle), and CD4+CD8A+ DP T cells (right) of total PBMCs at pre-ICI and post-ICI time points. (E) T cell clusters represented among CD8A+NCAM1+ NK-like T cells (upper) or CD4+CD8A+ DP T cells (lower). (F) Fold change in T cell signature scores within NK-like T cell cluster C2 and DP T cell cluster C5 between post-ICI and pre-ICI time points. (G) T cell receptor (TCR) clonotypes from single-cell TCR sequencing represented among cluster C2 (left) and cluster C5 (right) at pre-ICI and post-ICI time points. The designation “Other/NA” includes all other TCR clonotypes and cells lacking TCR data. (H and I) Distribution of CATSPGQYEQYF (NK-like T cells), CASSLAAGANEQFF (DP T cells), and other TCR clones among NK/T cell clusters (H) with quantifications at pre-ICI and post-ICI time points (I). (J) CATSPGQYEQYF (NK-like T cell) clonotype count at all four time points by bulk TCR sequencing. (K) Cytotoxicity signature scores for CATSPGQYEQYF (NK-like) and CASSLAAGANEQFF (DP) clones compared with all other clones at pre-ICI and post-ICI time points. Wilcoxon tests were used for comparisons and statistical significance denoted as ∗∗∗∗p < 0.0001. See also Figures S1–S3. Single-cell TCR sequencing was next used to evaluate T cell clones within NK-like and DP T cells. Notably, Vα24/Vβ18- and Vα7.2-based TCR chains were not detected, confirming that C2 and C5 cells represented αβ T cells as opposed to NKT or mucosal-associated innate T cells. Strikingly, two unique TCR clones comprised >50% of T cells within clusters C2 and C5 (Figure 2G). These two TCR clones, CATSPGQYEQYF in C2 and CASSLAAGANEQFF in C5, expanded following ICI and remained confined to their respective clusters (Figures 2H and 2I), suggesting terminal differentiation. The dominant NK-like and DP clones were comparatively rare in clusters other than C2 and C5 (Figure S3A). By orthogonal bulk TCR sequencing, both clones were among only eight TCRs that were shared between pre-treatment tumor biopsies and circulating T cells (Figure S3B), indicating that NK-like and DP T cells infiltrated the patient’s tumor. DP T cells expressing the CASSLAAGANEQFF TCR formed the tenth most prevalent intratumoral T cell clone, whereas NK-like T cells harboring the CATSPGQYEQYF TCR were the second most dominant intratumoral clone (Figure S3C). Notably, the CATSPGQYEQYF clone mapping to NK-like T cells markedly expanded in circulation by 47.9-fold after ICI (Figure 3J), emerging as the second most abundant circulating TCR clone at all three post-ICI time points (Figure S3B). Thus, ICI was associated with a dramatic and durable expansion of NK-like T cells. Terminally differentiated NK-like and CD4+CD8+ double-positive T cells exhibited cytotoxic potential (A) Schematic of PBMC collections in relation to pembrolizumab treatment. PBMCs from all four time points were interrogated using 29-parameter spectral flow cytometry and 42-parameter cytometry by time-of-flight (CyTOF) analyses. PBMCs from healthy donors (3 for flow cytometry [age and sex matched] and 11 for CyTOF) served as healthy control (HC) samples. (B) Principal component analysis comparing patient PBMCs at each time point to HC PBMCs using spectral flow cytometry. (C) UMAP dimensional analysis of concatenated patient PBMCs from all four time points assessed by spectral flow cytometry, with manual annotation of the indicated cell populations. (D) Fold difference in frequencies of the indicated populations within the T cell compartment when comparing patient T cell subsets (averaged between all four time points) with HC T cell subsets (averaged between the three healthy controls) assessed using spectral flow cytometry. (E and F) Representative plots (E) of frequencies of TNAIVE (CCR7+CD45RA+), TCM (CCR7+CD45RA–), TEM (CCR7–CD45RA–), and TEMRA (CCR7–CD45RA+) within the CD8+ T cell compartment of patient and HC PBMCs (F) assessed by spectral flow cytometry and CyTOF. (G) Histograms of CD56 on CD8+ T cells from spectral flow cytometry showing a representative HC and the patient at all four time points. (H) CD56+ frequencies of CD8+ T cells from patient and HC PBMCs assessed using flow cytometry and CyTOF. (I and J) Heatmaps of CD56 geometric mean fluorescent intensity (gMFI; I) with CD56+ frequencies based on TNAIVE, TCM, TEM, and TEMRA CD8+ T cell subsets (combined across all four time points; J) assessed using spectral flow cytometry and CyTOF. ANOVA with Tukey tests were used for comparisons with statistical significance denoted as ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. (K and L) Representative plots (K) of perforin+granzyme B+ (GzmB+) frequencies of TNAIVE, TCM, TEM, and TEMRA CD8+ T cell subsets combined across all four time points (L) assessed using CyTOF. ANOVA with Tukey tests were used for comparisons, with statistical significance denoted as ∗∗∗∗p < 0.0001. (M and N) Representative plots (gated on CD45+CD3+ T cells; M) of frequencies of CD4+, CD8+, CD4LoCD8Hi, and CD4HiCD8Lo T cells within patient and HC PBMCs (N) assessed using spectral flow cytometry. (O and P) Representative plots (O) of frequencies of TEMRA within either CD4LoCD8Hi or CD4HiCD8Lo T cell subsets of patient (combined across all four time points) and HC PBMCs (P) assessed using spectral flow cytometry. Unpaired t tests were used for comparisons with statistical significance denoted as ∗p < 0.05 and ∗∗∗p < 0.001. (Q and R) Representative plots (Q) of frequencies of effector T cell-associated CD57+KLRG1+ cells within CD4+, CD8+, CD4LoCD8Hi, and CD4HiCD8Lo patient-derived T cells combined across all four time points (R) assessed using CyTOF. ANOVA with Tukey tests were used for comparisons with statistical significance denoted as ∗∗p < 0.01 and ∗∗∗∗p < 0.0001. (S and T) Representative plots (S) of frequencies of cytotoxicity-associated perforin+granzyme B+ cells within CD4+, CD8+, CD4LoCD8Hi, and CD4HiCD8Lo patient-derived T cells combined across all four time points (T) assessed using CyTOF. ANOVA with Tukey tests were used for comparisons with statistical significance denoted as ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. (U) Frequencies of proliferating Ki-67+ cells within CD56–CD4+, CD56+CD4+, CD56–CD8+, CD56+CD8+, and CD4HiCD8Lo T cell subsets across all four time points assessed using CyTOF. All panels show mean ± SEM. See also Figures S4 and S5. Overall clonal diversity within circulating T cells contracted following ICI initiation, consistent with evolving dominance of select tumor-specific clones (Figure S3D). The most prevalent NK-like and DP T cell clones also exhibited increased signatures of cytotoxicity compared with all other CD8+ T cell clones (Figure 2K), further supporting possible tumor specificity. Altogether, clonal NK-like and DP T cells expanded in circulation following ICI and adopted effector-associated functions that were further enhanced by ICI. As these atypical cells were also detected in pre-treatment tumor tissue and exhibit anti-tumor potential,7,8,9 they likely represent relevant tumor-specific populations. Terminally differentiated NK-like and CD4+CD8+ double-positive T cells exhibited cytotoxic potential NK-like and DP T cells at all four PBMC time points (Figure 3A) were further evaluated using orthogonal multiparameter spectral flow cytometry and cytometry by time of flight (CyTOF), and compared with healthy controls (HCs). Patient and age- and sex-matched HC PBMCs diverged by principal component analysis (Figure 3B), and NK-like and DP T cells segregated using unbiased dimensional analysis (Figure 3C). Two distinct DP populations were identified—CD4HiCD8Lo and CD4LoCD8Hi—aligning with foundational descriptions of DP T cells.10 CD4HiCD8Lo DP cells, along with NK-like CD56+CD4+ and CD56+CD8+ T cell populations were 3- to 9-fold more prevalent within the T cell compartment of the patient compared with HCs (Figures 3D and S4A), suggesting that these populations expanded in response to PC development. Canonical T cell subsets including naive, central memory (TCM), TEM, and terminally differentiated CD45RA+ TEMRA were examined within both CD4+ and CD8+ T cells. A marked increase in TEMRA frequency was observed within both the CD4+ (>10-fold) and CD8+ (∼2-fold) compartments compared with HCs (Figures 3E, 3F, S4B, and S4C). CD56 was expressed predominantly within TEMRA, and these cells were increased 2- to 5-fold compared with HC lymphocytes (Figures 3G–3J and S4D–S4G). Patient TEMRA cells were also enriched for cytotoxic perforin and granzyme B compared with all other T cell subsets (Figures 3K, 3L, S4H, and S4I). Thus, NK-like T cells formed terminally differentiated effector-like cells in circulation. Regarding DP T cells, although two populations were identified, only CD4HiCD8Lo T cells expanded following ICI treatment (Figures 3M and 3N). Similar to NK-like T cells, CD4HiCD8Lo DP T cells exclusively adopted TEMRA phenotypes, which differed distinctly from differentiation patterns adopted by rare CD4HiCD8Lo cells in HCs (Figures 3O and 3P). Additionally, patient-derived CD4HiCD8Lo DP T cells lacked memory-associated markers (CD27, CD127; Figures S4J and S4K), acquired markers associated with effector T cells and NK cells (CD56, CD57, CD161, and KLRG1; Figures 3Q, 3R, S4L, and S4M), and produced robust levels of cytotoxic perforin and granzyme B (Figures 3S and 3T). While NK-like and DP T cells adopted phenotypes consistent with terminally differentiated TEMRA, PD-1 was not detected (Figure S5A), suggestive of lack of recent antigen exposure and making functional exhaustion of these circulating cells unlikely. Notably, Ki-67+ cycling CD4+CD56+, CD8+CD56+, and CD4HiCD8Lo cells increased following ICI, and CD56+ T cells exhibited greater proliferation relative to their CD56– counterparts (Figure 3U). Collectively, clonal CD56+ NK-like and CD4+CD8+ DP T cells were more prevalent at baseline compared with HCs, infiltrated tumors, expanded/proliferated after ICI, and adopted terminally differentiated phenotypes poised to exert cytotoxic activity. Cytotoxic NK-like and CD4+CD8+ DP T cells expand in cancer patients responding to immunotherapy To extend our findings, we evaluated additional cancer patient cohorts treated with ICI. To the best of our knowledge, PC cohorts with dMMR/MSI-H defects do not exist due to the rarity of these defects in PC. While ICI use is also rare in PC, a recent clinical trial of patients treated with neoadjuvant Fc-enhanced anti-CTLA4 (aCTLA4) immunotherapy by Ager et al.11 allowed for evaluation of PC patients treated with ICI. CD8+ T cells from CyTOF performed on intratumoral samples from patients treated with combination ADT plus aCTLA4 were analyzed by uniform manifold approximation and projection and compared with untreated patients. We first reaffirmed previously reported findings that a population of tumor-infiltrating 41BB+CD39+ CD8+ T cells exclusively formed in patients treated with aCTLA4 (Figures 4A–4C).11 Most of these cells also co-expressed CD56 and resembled our previously characterized NK-like T cells. Tumor-infiltrating CD56+CD8+ T cells arising uniquely in aCTLA4-treated patients also expressed PD-1, reinforcing possible tumor reactivity. While total CD8+ T cells were comparable in untreated and aCTLA4-treated patients (Figure 4D), 41BB+CD39+CD56+ CD8+ T cell frequencies were significantly enriched in patients treated with aCTLA4 and nearly undetectable in untreated patients (Figure 4E). A CD4+CD8+ DP T cell population that similarly expressed 41BB, CD39, and CD56 was also significantly enriched in aCTLA4-treated patients (Figures 4F–4H). As 41BB is induced in activated T cells12 and CD39 expression is linked to tumor-antigen specificity,13 ICIs may promote development of activated tumor-specific NK-like and DP T cells in PC. Cytotoxic NK-like and CD4+CD8+ DP T cells expand in cancer patients responding to immunotherapy (A) UMAP analysis of concatenated intratumoral CD45+ leukocytes from CyTOF performed on PC patients treated with or without neoadjuvant ADT plus aCTLA4 (see Ager et al.11) with CD4+ and CD8+ T cells annotated. (B) UMAP analysis of concatenated CD8+ T cells, with cells from each treatment group annotated. (C) Heatmaps of concatenated CD8+ T cells depicting geometric mean intensity (gMI) of 41BB, CD39, CD56, and PD-1. (D) CD8+ T cell frequency of CD45+ leukocytes. ns, not significant by Mann-Whitney test. (E) 41BB+CD39+CD56+ frequency of CD8+ T cells. Mann-Whitney test was used for comparison with statistical significance denoted as ∗∗p < 0.01. (F) Concatenated CD3+ T cells with CD4+CD8+ DP gated. (G) UMAP analysis of concatenated CD4+CD8+ DP T cells and gMI heatmaps of 41BB, CD39, CD56, and PD-1. (H) 41BB+CD39+CD56+CD4+CD8+ frequency of CD3+ T cells. Mann-Whitney test was used for comparison with statistical significance denoted as ∗∗p < 0.01. (I) UMAP analysis of T cells from scRNA-seq of dMMR/MSI-H CRC patients treated with anti-PD-1 ICI (see Chen et al.14) with CD4+ and CD8+ T cells annotated. (J) CD4, CD8A, and NCAM1 gene expression in T cells from CRC patients. (K) Cytotoxicity signature scores for NK-like and CD4+CD8A+ cells compared with all other T cells at pre-ICI and post-ICI time points in CRC patients. Wilcoxon tests were used for comparisons with statistical significance denoted as ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗∗p < 0.0001. ns, not significant. (L) Fold change in CD8+ T cells and CD4+CD8A+ DP T cells following ICI treatment in CRC patients. (M) UMAP analysis of T cells from scRNA-seq of dMMR/MSI-H endometrial carcinoma patients treated with pembrolizumab (see Chow et al.15). (N) CD4, CD8A, and NCAM1 gene expression in T cells from endometrial carcinoma patients. (O) Fold change in T cells following pembrolizumab treatment in endometrial carcinoma patients segregated by nonresponders (NR) and responders (R). ns, not significant by Mann-Whitney test. (P) Fold change in CD4+CD8A+ DP T cells following pembrolizumab treatment in endometrial carcinoma patients segregated by NR and R. Mann-Whitney test was used for comparison with statistical significance denoted as ∗p < 0.05. Dots in box (median and interquartile range) and whisker (minimum and maximum) plots in (D, E, H, L, O, and P) represent individual patients. See also Figure S6. To further evaluate if NK-like and/or DP T cells were present in patients with dMMR/MSI-H cancers, we analyzed scRNA-seq from Chen et al.,14 in which tumor-infiltrating T cells were tracked in 12 patients with dMMR/MSI-H colorectal cancer (CRC) that responded to neoadjuvant treatment with anti-PD-1 ICIs. CD8A+NCAM1+ NK-like T cells exhibited significantly increased cytotoxicity signature scores at both pre-ICI and post-ICI time points compared with other T cells (Figures 4I–4K). While CD4+CD8A+ DP T cell cytotoxicity signatures were similar to those of other T cells, scores significantly increased following ICI. Moreover, DP T cells expanded at least to some degree in 60% of patients, whereas overall CD8+ T cell populations were largely unchanged after ICI treatment (Figure 4L). Therefore, NK-like T cells may exhibit increased anti-tumor activity compared with other T cell subsets in CRC patients with dMMR/MSI-H tumors, and DP T cells selectively expand following ICI. Most patients with dMMR/MSI-H CRC respond favorably to ICI, restricting our ability to perform comparisons between responders and nonresponders. Consequently, we performed similar evaluations of circulating T cells in dMMR/MSI-H endometrial carcinoma patients by scRNA-seq from Chow et al.,15 where ICI responses were more variable. While CD4+CD8A+ DP T cells were identified, NK-like T cells were rarely detected (Figures 4M and 4N), potentially due to differences in cancer type, intratumoral versus circulating T cell frequencies, or technical limitations in sequencing. Although total circulating T cell frequencies were largely unchanged after treatment in both ICI responders and nonresponders, DP T cells exclusively expanded in responders (Figures 4O and 4P). To expand our findings to MMR-proficient (pMMR) cancers, we evaluated tumor-infiltrating CD4+CD8A+ DP T cells from triple-negative breast16 and head/neck squamous cell carcinoma17 patients treated with ICIs. Similar to findings in dMMR/MSI-H cancers, DP T cells expanded only in ICI responders (Figures S6A–S6F). Accordingly, DP expansion may correlate with ICI response in diverse pMMR and dMMR cancers. Discussion Although T cells targeting neoantigens generated by hypermutated malignancies are foundational for responses to ICI, how these cells acquire anti-tumor potential remains unclear. Herein, we describe a patient with dMMR/MSI-H, ultra-mutated PC in whom pembrolizumab promoted clonal expansion of NK-like and CD4+CD8+ DP T cells. These unique cells may have been essential in rendering a durable, complete pathological and molecular response to ICI in a patient with Gleason 10 locally advanced PC—a striking clinical course not previously described.18 T cells acquiring NK-associated phenotypes are implicated in numerous diseases.7,8 These unique lymphocytes can elicit potent cytotoxic activity against tumor cells, often using MHC-independent innate-like mechanisms.7,8 Although NK-like T cells appear protective in certain hematologic malignancies,19,20,21,22,23 inverse associations have been observed in several solid tumors.24,25,26,27,28 In our patient, NK-like T cells were highly enriched in circulation compared with PBMCs from HCs and infiltrated tumor tissues. Additionally, the cells acquired terminally differentiated phenotypes and produced high levels of effector molecules—features consistent with established hallmarks of NK-like T cells.7,8 Remarkably, a single NK-like T cell clone with cytotoxic potential expanded nearly 50-fold following initiation of ICI, supporting a possible role in complete eradication of this high-grade, castration-resistant PC. NK-like cells exhibiting phenotypes associated with activation and tumor specificity were also independently identified in patients treated with aCTLA4 ICI, reinforcing that immunotherapy may drive formation of NK-like T cells in PC. These cells may also be conserved in dMMR/MSI-H cancers, as they adopted cytotoxic profiles in CRC patients with such defects. Regarding DP T cells, emerging evidence supports that they represent a mature lymphocyte population capable of MHC-I and/or MHC-II-restricted tumor-specific potential.9 Akin to NK-like T cells, DP T cells have been reported to possess pro- or anti-tumor functions and frequently express markers associated with effector cytotoxic functions.9 The origins of these cells have recently become clearer, with some groups hypothesizing that tumor-reactive DP T cells arise from CD4 or CD8 single-positive precursors and acquire shared functions of each canonical T cell type.29,30,31,32 In the case herein, similar to NK-like T cells, CD4HiCD8Lo DP T cells infiltrated the patient’s tumor, were terminally differentiated, and displayed cytotoxic potential. Yet, as these cells were present prior to ICI treatment and only moderately increased after treatment, it remains unclear whether they exerted significant anti-tumor effects. Nevertheless, DP T cells appear to selectively expand in ICI responders across diverse cancers and, therefore, may represent a potential biomarker of ICI response. In conclusion, our findings support NK-like and DP T cells as promising components of anti-tumor immunity. Additional investigations on the ontogeny of these cells and their potential to serve as immunotherapy targets, including approaches with purified CD56+ and CD4+CD8+ T cells with tumor-specific receptors, are warranted. This case further reinforces the remarkable potential of ICIs in dMMR/MSI-H cancers with ultrahigh TMB, even in tumors not typically considered responsive to immunotherapies, and suggests that some MMR-deficient malignancies may be cured with nonoperative management.18,33 Limitations of the study Although NK-like and DP T cells in our patient adopted several attributes suggestive of tumor specificity, attempts to infer TCR specificity were hampered by the patient’s ultrahigh TMB and thousands of potential neoantigens. Emerging improvements in predictive in silico technologies may aid in identifying T cell specificity in similarly challenging situations. Advances in spatial and single-cell technologies will also continue to allow for validation of our unique findings while clarifying the contexts under which NK-like and DP T cells contribute to anti-tumor immunity. Our findings reflect a single-patient case study, albeit with deep molecular and immunologic characterization, and may not be generalizable to all dMMR/MSI-H and/or ultra-mutated cancers. To partially alleviate this limitation, we expanded our findings in prostate, dMMR/MSI-H colorectal, endometrial, breast, and head/neck cancer patient cohorts treated with ICIs. While several concordant trends were observed, these cohorts were too small to rigorously control for confounding variables that might influence NK-like and DP T cells. Additional larger and prospective studies are necessary to more accurately quantify associations between these cells and patient outcomes. Resource availability Lead contact Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Emmanuel S. Antonarakis (anton401@umn.edu). Materials availability This study did not generate new unique reagents. Data and code availability •Local law prohibits depositing raw genomic datasets derived from human samples. Single-cell RNA-seq data have been deposited at GEO: GSE332846 and are publicly available as of the date of publication. This paper analyzes existing, publicly available data, accessible at DOIs: https://doi.org/10.1016/j.xcrm.2026.102638, https://doi.org/10.1016/j.ccell.2024.06.009, https://doi.org/10.1158/2159-8290.CD-22-0686, https://doi.org/10.1016/j.ccell.2021.09.010, and https://doi.org/10.1016/j.cell.2022.06.018. Local law prohibits depositing raw genomic datasets derived from human samples. Single-cell RNA-seq data have been deposited at GEO: GSE332846 and are publicly available as of the date of publication. This paper analyzes existing, publicly available data, accessible at DOIs: https://doi.org/10.1016/j.xcrm.2026.102638, https://doi.org/10.1016/j.ccell.2024.06.009, https://doi.org/10.1158/2159-8290.CD-22-0686, https://doi.org/10.1016/j.ccell.2021.09.010, and https://doi.org/10.1016/j.cell.2022.06.018. •This paper does not report original code. This paper does not report original code. •Any additional information required to reanalyze the data reported in this paper is available from 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. Acknowledgments We thank Kellie N. Smith, PhD, and Li Zhang at Johns Hopkins University for conducting bulk TCR sequencing. We thank the Johns Hopkins Experimental and Computational Genomics Core, supported by NCI P30CA006973, for assistance with the single-cell RNA and coupled TCR sequencing. We also thank Megan Larson and Rose Wangen at the University of Minnesota for performing CyTOF analysis. For multiplex imaging, we thank Liam McLaughlin, Thomas Pengo, and Myat Mo at the University of Minnesota, the Clinical and Translational Science Institute’s Histology Core (University of Minnesota), and the Immune Monitoring Core at Mayo Clinic (Rochester, Minnesota) for analytical support. This work was supported by an NIH General Medical Sciences T32GM008244 grant (to J.R.L.); NCI grants F30CA294723 (to J.R.L.), R35CA283892 (to J.S.M. and M.F.), R37CA288972 (to J.H.), U54CA274370 (to S.Y.), P50CA272391 (to S.Y.), P01CA065493 (to P.R.K.), P01CA111412 (to P.R.K.), and P30CA077598 (to E.S.A.); DOD grant W81XWH-22-2-0025 (to E.S.A.); the Commonwealth and V Foundations (to S.Y.); and the Conquer Cancer Foundation (to J.R.L.). Author contributions Conceptualization, A.K.T., J.R.L., J.H., and E.S.A.; methodology, A.K.T., J.R.L., P.R.K., D.M., R.P., C.L., M.F., S.S.S., E.A.T., L.A.S., S.Y., J.H., and E.S.A.; investigation, A.K.T., J.R.L., P.R.K., D.M., R.P., R.C.L., C.L., R.W., E.L.F., S.Z., A.F., P.J.M., M.C.D., A.E., P.M., M.F., B.R.K., S.S.S., E.A.T., L.A.S., S.Y., and J.H.; writing – original draft, A.K.T.; writing – review and editing, J.R.L., P.R.K., D.M., R.P., A.T.A., A.E., P.M., N.A.Z., B.R.K., S.M.D., J.S.M., S.S.S., L.A.S., S.Y., J.H., and E.S.A.; funding acquisition, P.R.K., S.Y., J.H., and E.S.A.; resources, P.R.K., M.F., S.S.S., L.A.S., S.Y., J.H., and E.S.A.; supervision, J.H. and E.S.A. Declaration of interests D.M. consults for Tempus. M.C.D. receives research support from Johnson & Johnson and Bristol Myers Squibb and personal fees from DAVA Oncology, Curio, Bayer, Bristol Myers Squibb, Astra Zeneca, Xilio, and Johnson & Johnson. J.S.M. has been a paid consultant for Fate Therapeutics and has received research funds and stock options from this relationship. J.S.M. and M.F. receive research support and, with the University of Minnesota, are shared owners of the TriKE technology licensed by the University to GT Biopharma, Inc. J.S.M. and M.F. also consult for and hold stock options in GT Biopharma, Inc. Fate and GT Biopharma may commercially benefit from the results of this research project. J.S.M. serves on the Scientific Advisory Board of Sanofi, Vycellix, and Simcha. N.A.Z. has received honoraria from the Association of Community Cancer Centers (ACCC) and Department of Defense Prostate Cancer Research Program, Mosaic Research Management, Slingshot Insights, and Bayer (Institutional); research funding to his institution from Janssen Research & Development, Johnson & Johnson, Janux Therapeutics, Amgen, Lava Therapeutics, GT Biopharma, ArsenalBio, and Takeda; and expenses from Telix Pharmaceuticals, Caris Life Sciences, Bayer, Janux Therapeutics, and DAVA Oncology and serves in an advisory capacity for Bayer and Takeda. L.A.S. has received funding to her institution from Panbela Therapeutics. J.H. consults for Tempus and Astrin Biosciences and is co-founder of EMRGNSE, LLC. E.S.A. receives grants and personal fees from Janssen, Johnson & Johnson, Sanofi, Bayer, Bristol Myers Squibb, Convergent Therapeutics, Curium, MacroGenics, Merck, Pfizer, and AstraZeneca; personal fees from Aadi Bioscience, Abeona Therapeutics, Aikido Pharma, Astellas, Amgen, Blue Earth, Boundless Bio, Corcept Therapeutics, Duality Bio, Exact Sciences, Hookipa Pharma, Invitae, Eli Lilly, Foundation Medicine, Menarini-Silicon Biosystems, Tango Therapeutics, Tempus, Tolmar Scientific, VIR Biotechnology, and Z-alpha; and grants from Novartis, Celgene, and Orion and has a patent for an AR-V7 biomarker technology that has been licensed to Qiagen. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the authors did not use any generative AI or AI-assisted technologies. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesCD34 Atto550Akoya BiosciencesCat# 4250057; RRID: AB_3676526CD56 AF647Akoya BiosciencesCat# STP7000HLA-DR AF750Akoya BiosciencesCat# 4450095; RRID: AB_3094500CD45RO Atto550Akoya BiosciencesCat# 4250023; RRID: AB_2895053FOXP3 AF647Akoya BiosciencesCat# 4550071; RRID: AB_2927679CD20 AF750Akoya BiosciencesCat# 4450094; RRID: AB_3094498CD44 Atto550Akoya BiosciencesCat# 4450041; RRID: AB_2936081IDO1 AF647Akoya BiosciencesCat# 4550123; RRID: AB_3476035Ki-67 AF750Akoya BiosciencesCat# 4450096; RRID: AB_3094497IFNG Atto550Akoya BiosciencesCat# 4250062; RRID: AB_3476455TCF-1 AF647Akoya BiosciencesCat# 4550068; RRID: AB_3717468Pan-Cytokeratin AF750Akoya BiosciencesCat# 4450093; RRID: AB_3662772iNOS Atto550Akoya BiosciencesCat# 4250073; RRID: AB_3676531CD4 AF647Akoya BiosciencesCat# 4550112; RRID: AB_3094499Granzyme-B Atto550Akoya BiosciencesCat# 4250055; RRID: AB_3472025HIF1A AF647Akoya BiosciencesCat# 4550069; RRID: AB_3082972CD8 Atto550Akoya BiosciencesCat# 4250012; RRID: AB_2915960PD-1 AF647Akoya BiosciencesCat# 4550038; RRID: AB_3096407PD-L1 AF647Akoya BiosciencesCat# 4550128; RRID: AB_3676534CD57 Atto550Akoya BiosciencesCat# 4250108; RRID: AB_3717470CD11c AF647Akoya BiosciencesCat# 4550135; RRID: AB_3678450CD163 Atto550Akoya BiosciencesCat# STP7000CD31 Atto550Akoya BiosciencesCat# 4250104; RRID: AB_3662762CD45 AF647Akoya BiosciencesCat# 4550121; RRID: AB_3674468CD66 AF647Akoya BiosciencesCat# 4550001; RRID: AB_3475664CD68 AF647Akoya BiosciencesCat# 4550113; RRID: AB_2935894CD3ε AF647Akoya BiosciencesCat# 4550125; RRID: AB_3094503CD7AbcamCat# ab230834; RRID: AB_2889384CD3 Spark UV 387BioLegendCat# 323066CD4 AF680InvitrogenCat# 606-0049-42; RRID: AB_2896239CD8 APC/Fire-810BioLegendCat# 344764; RRID: AB_2860890CD11c PEBD BiosciencesCat# 566730; RRID: AB_2869833CD14 BV605BioLegendCat# 301834; RRID: AB_2563798CD16 BV711BioLegendCat# 302044; RRID: AB_2563802CD19 APC-H7BD BiosciencesCat# 560727; RRID: AB_1727437CD20 BV786BD BiosciencesCat# 568713; RRID: AB_3684489CD21 BUV496BD BiosciencesCat# 750614; RRID: AB_2874746CD24 BUV395BD BiosciencesCat# 563818; RRID: AB_2632389CD27 PE-CF594BD BiosciencesCat# 562297; RRID: AB_11154596CD36 BV480BD BiosciencesCat# 746612; RRID: AB_2871651CD38 BUV661BD BiosciencesCat# 612969; RRID: AB_2870242CD40 PE-Cy7BioLegendCat# 334322; RRID: AB_10645472CD45RA BUV563BD BiosciencesCat# 612926; RRID: AB_2870211CD56 PE/Fire-810BioLegendCat# 392435; RRID: AB_2927902CD86 BUV737BD BiosciencesCat# 612784; RRID: AB_2814790CD123 PacBlueBioLegendCat# 306044; RRID: AB_2750165CD138 PE/Fire-640BioLegendCat# 356550CCR7 PE/Fire-744BioLegendCat# 353276CXCR3 APCBioLegendCat# 353707; RRID: AB_10962949CXCR5 PE-Cy5BioLegendCat# 356952HLA-DR BV570BioLegendCat# 307638; RRID: AB_2650882IgD BUV805BD BiosciencesCat# 742039; RRID: AB_2871332IgM AF700BioLegendCat# 314538; RRID: AB_2566615LOX-1 BV421BioLegendCat# 358610; RRID: AB_2728343PD-1 BUV615BD BiosciencesCat# 612991; RRID: AB_2870262CD45 89YStandardBioToolsCat# 201349CD196/CCR6 141PrStandardBioToolsCat# 201349CD123 143NDStandardBioToolsCat# 201349CD19 144NDStandardBioToolsCat# 201349CD4 145NDStandardBioToolsCat# 201349CD8a 146NDStandardBioToolsCat# 201349CD11c 147SmStandardBioToolsCat# 201349CD16 148NDStandardBioToolsCat# 201349CD45RO 149SmStandardBioToolsCat# 201349CD45RA 150NDStandardBioToolsCat# 201349CD161 151EuStandardBioToolsCat# 201349CD194/CCR4 152SmStandardBioToolsCat# 201349CD25 153EuStandardBioToolsCat# 201349CD27 154SmStandardBioToolsCat# 201349CD57 155GdStandardBioToolsCat# 201349CD183/CXCR3 156GdStandardBioToolsCat# 201349CD185/CXCR5 158GdStandardBioToolsCat# 201349CD28 160GdStandardBioToolsCat# 201349CD38 161DyStandardBioToolsCat# 201349CD56/NCAM 163DyStandardBioToolsCat# 201349TCRgd 164DyStandardBioToolsCat# 201349CD294 166ErStandardBioToolsCat# 201349CD197/CCR7 167ErStandardBioToolsCat# 201349CD14 168ErStandardBioToolsCat# 201349CD3 170ErStandardBioToolsCat# 201349CD20 171YbStandardBioToolsCat# 201349CD66b 172YbStandardBioToolsCat# 201349HLA-DR 173YbStandardBioToolsCat# 201349IgD 174YbStandardBioToolsCat# 201349CD127 176YbStandardBioToolsCat# 201349CD69 113CdStandardBioToolsCat# 3113002BKLRG1 114Cd∗BiolegendCat# 367702; RRID: AB_2632728CD62L 116Cd∗BiolegendCat# 304835; RRID: AB_2563758FoxP3 159TbStandardBioToolsCat# 3159039AKi-67 162DyStandardBioToolsCat# 3162012BCD279/PD-1 165HoStandardBioToolsCat# 3165042BTCF-1 169TmBiolegendCat# 655202; RRID: AB_2562103CD278/ICOS 175LuStandardBioToolsCat# 3175039BPerforin 196PtStandardBioToolsCat# 3196002CGranzyme-B 198PtStandardBioToolsCat# 3198002CTIGIT 209BiStandardBioToolsCat# 3209013BDeposited dataRaw and analyzed dataThis paperGEO: GSE332846Software and algorithmsnf-core/sarek v3.4.4Menard et al.29https://github.com/nf-core/sarek/tree/3.4.4Burrows-Wheeler Aligner v0.7.18Li30https://github.com/lh3/bwaGenome Analysis Toolkit v4.5.0.0Auwera and O’Connor31https://github.com/broadinstitute/gatkMutect2 v4.2.6.1Auwera and O’Connor31https://github.com/TRON-Bioinformatics/tronflow-mutect2GATK HaplotypeCaller v4.2.6.1Auwera and O’Connor31https://github.com/broadinstitute/gatkEnsembl Variant Effect Predictor v110McLaren et al.32https://github.com/Ensembl/ensembl-vepBcftools v1.21Danecek et al.33https://github.com/samtools/bcftoolsSigProfilerAssignment v1.0.3Díaz-Gay et al.34https://github.com/AlexandrovLab/SigProfilerAssignmentQuPath v0.4.4Bankhead et al.35https://qupath.github.io/Stardist v0.9.1Weigert and Schmidt36https://github.com/stardist/stardistcytoMAP v1.4.21Stoltzfus et al.37https://gitlab.com/gernerlab/cytomapCell Ranger v9.0.0Zheng et al.38https://github.com/10XGenomics/cellrangerSeurat v4.3.0Hao et al.39https://satijalab.org/seurat/Harmony v1.2.0Korsunsky et al.40https://github.com/immunogenomics/harmonySingleR v1.4.1Aran et al.41https://github.com/dviraran/SingleRAUCell v1.24.0Aibar et al.42https://github.com/aertslab/AUCellscRepertoire v2.3.4Yang et al.43https://github.com/BorchLab/scRepertoireImmunarch R package v0.10.3Popov et al.44https://github.com/immunomind/immunarchedgeRRobinson et al.45https://github.com/OliverVoogd/edgeRPheatmap v1.0.13N/Ahttps://github.com/raivokolde/pheatmapOMIQDotmatricshttps://www.omiq.ai/JMP v19JMPhttps://www.jmp.com/en/homeFlowJo v9BD Bioscienceshttps://www.flowjo.com/Maxpar PathsetterStandard Biotoolshttps://www.standardbio.com/products/software/maxpar-pathsetterGraphPad Prism v10Dotmaticshttps://www.graphpad.com/features Experimental model and study participant details The 67-year-old otherwise healthy Caucasian male patient provided written informed consent for collection and biobanking of blood samples along with interrogation of tissue samples according to an internal review board (IRB)-approved protocol (IRB #00003639). The patient’s age and gender did not influence the results of the study. Method details Tumor mutational burden (TMB) quantification and single-base-substitution (SBS) signature analysis Total DNA was isolated from formalin-fixed, paraffin-embedded (FFPE) tumor tissue that was microdissected for tumor purity, as well as paired peripheral blood. Samples were sequenced using the Novaseq 6000 whole-exome sequencer alongside a panel designed to enrich for >20,000 genes to detect amplifications and deletions. Paired-end reads for each sample were processed using the nf-core/sarek workflow (v3.4.4).34 Reads were aligned to the human GRCh38 reference genome using the Burrows-Wheeler Aligner (v0.7.18).35 The Genome Analysis Toolkit (GATK, v4.5.0.0)36 was used to identify the number of unique, duplicate, and unmapped read pairs from the aligned file. Mutect2 (v4.2.6.1)36 and GATK HaplotypeCaller (v4.2.6.1)36 were used to detect somatic and germline variants, respectively, in the tumor-normal pair. Ensembl Variant Effect Predictor (VEP, v110)37 was then used to predict the consequences of the variants on gene and protein sequences. Bcftools (v1.21)38 was implemented to filter somatic variants to those with variant allele frequencies (VAFs) ≥0.01, total read depth ≥50 reads, and allelic depth ≥3 reads. Similar thresholds were implemented for germline variants, except for a VAF cutoff. All variants were filtered for a variant effect predictor (VEP) consequence of ‘missense’ or greater severity and variants without a gene symbol were omitted. To estimate tumor mutational burden (TMB), the total number of variants were divided by 25 megabases, given the total exon region size for the whole-exome sequencing panel. Lastly, SigProfilerAssignment39 was applied to assign and visualize Catalog of Somatic Mutations in Cancer (COSMIC) signatures for the tumor-normal pair using the somatic and germline variants detected. A de-identified dataset of real-world patients’ samples that underwent comprehensive molecular profiling at a CLIA/CAP-certified lab (Caris Life Sciences) was queried for TMB values in prostate cancer patients. Next-generation sequencing (NGS) was performed on genomic DNA isolated from FFPE tumor samples using NextSeq or NovaSeq 6000 platforms. TMB values were compiled from patients with histology-confirmed prostate cancer and were stratified by MMR and MSI status. Digital-spatial microscopy (CODEX) Five-micron histological sections were cut from the FFPE biopsy and prostatectomy samples, mounted onto a slide and stained according to the manufacturer’s protocol for imaging on the Phenocycler Fusion 2.0 (Akoya Biosciences). The sections were deparaffinized and rehydrated, followed by antigen retrieval with citrate buffer (Fisher Scientific; Cat# 00-4955-58). All antibodies (see key resources table) were purchased conjugated to DNA barcodes except anti-human CD7 (Abcam; Cat# ab230834), which was conjugated to a DNA barcode (Akoya Biosciences; Cat# 5550024) using an antibody conjugation kit (Akoya Biosciences; Cat# 7000009). After the slide was stained with primary antibodies, autofluorescence removal was performed in a light bath consisting of 4.5% hydrogen peroxide and 27 mM sodium hydroxide in phosphate-buffered saline (PBS) for two rounds of 45 min. Imaging was performed at 20× on a Phenocycler Fusion 2.0 (Akoya Biosciences). Complementary barcodes conjugated to Atto550, Alexa Fluor 647 and Alexa Fluor 750 were used in cyclical hybridization, buffer exchange, imaging and stripping cycles. Image stitching, deconvolution, alignment and cycle concatenation were performed in Akoya Biosciences software. Images were examined and further analyzed in QuPATH software (v0.4.4).40 Antibodies had previously been validated on a human tonsil sample, but any markers that produced a low signal-to-noise ratio or untenable staining pattern at this stage were rejected (CD16, CD45RA, CD49a, CD69, CD11b). Tissue regions that exhibited poor imaging quality were excluded from subsequent analysis. Nuclei were segmented using Stardist41 and a neural network classifier was trained to identify positive cells for each channel. Unsupervised clustering of regions was performed in cytoMAP42 on cell objects extracted from QuPATH. Raster scanning was used to define neighborhoods and these were clustered into self-organizing maps with region number defined by the Davies-Bouldin criterion, normalizing mean fluorescence intensity to max mean fluorescence intensity for all cells, inputting the nearest neighbor self-organizing map, and clustering by markers of interest including the mean membrane intensity for CD45, CD3ε, CD4, CD8, CD56, CD7, CD11c, PD-1, and nuclear intensity for Ki-67. For image quantification, T cells are defined as CD45+CD3ε+ cells and further subdivided by other markers (CD4, CD8, CD45RO, PD-1, Ki-67). Single-cell RNA- and TCR-sequencing and analysis Cryopreserved peripheral blood mononuclear cells (PBMCs) were thawed and adjusted to a final concentration of 1,000–2,000 viable cells/μL. Approximately 17,000 cells were loaded onto the Chromium X (10× Genomics) to capture a target of 10,000 cells per sample. Libraries were prepared using the Chromium Next GEM Single Cell V(D)J Reagent Kits v1.1, following the manufacturer’s instructions. The 5′ digital gene expression (DGE) libraries were sequenced on an Illumina NovaSeq X system to a target depth of 50,000 reads per cell, while full-length V(D)J segments were enriched from barcoded cDNA to enable pairing of TCRα and TCRβ chains with the corresponding transcriptome. The 5′ V(D)J libraries were sequenced on an Illumina NovaSeq X system to a depth of approximately 5,000 reads per cell. For data pre-processing, Cell Ranger (v.9.0.0)43 was used to align reads to the human GRCh38 human reference genome, assign cell barcodes, and generate count matrices. Low-quality cells were filtered based on total number of unique genes detected (<200) and proportion of mitochondrial genes (>20%). Data was normalized using the SCTransform function in the Seurat pipeline (v4.3.0).46 The top 2,000 highly variable genes (HVGs) were used for principal component analysis and Harmony (v1.2.0)44 was implemented to correct batch effects. Seurat was used to apply Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction and unsupervised clustering. SingleR (v1.4.1)45 was used in conjunction with Blueprint/ENCODE reference transcriptomes to annotate immune cell types within the samples. To comprehensively annotate T/NK cell subsets, CD8+ T cells, CD4+ T cells, and NK cells were isolated and differentially expressed genes (DEGs) were calculated across clusters. DEGs were then manually inspected for markers of T cell and NK cell phenotypes. To infer the functional potential of each distinct T/NK cell subset, we employed AUCell (v1.24.0)47 which applies rank-based scoring to infer the enrichment of gene sets within single cells. Functional gene sets were based on previously published signatures.6 scV(D)J-seq data was analyzed using the Cell Ranger VDJ pipeline which aligned V(D)J-seq reads to GRCh38 to assemble TCR sequences. scRepertoire (v2.3.4)48 was then implemented to combine TCR information with the single-cell transcriptomes. For dMMR/MSI-H colorectal,14 dMMR/MSI-H endometrial carcinoma,15 triple-negative breast cancer,16 and head/neck squamous cell carcinoma17 validation datasets, count matrices aligned and mapped to the GRCh38 human reference genome were acquired. Similarly, low quality cells were filtered, and data was normalized using the SCTransform function in the Seurat pipeline. Harmony was implemented to correct batch effects and UMAP was applied for dimensionality reduction and unsupervised clustering. T cells were annotated using SingleR. Fold expansion was calculated using the percentage of T cell subtype out of all cells, comparing post-versus pre-ICI proportions. TCR-sequencing and analysis TCR-sequencing was performed on FFPE prostate biopsies and PBMCs across four timepoints by the Johns Hopkins Sidney Kimmel Comprehensive Cancer Center FEST and TCR Immunogenomics Core as previously described.49 Briefly, DNA was extracted from PBMCs using the DNeasy Blood and Tissue Kit (Qiagen). The TCR-beta locus was amplified and sequenced using the AmpliSeq for Illumina TCR beta-short read assay. Data preprocessing was performed to eliminate non-productive TCR sequences and to align and trim the nucleotide sequences to obtain only the complementarity-determining region 3 (CDR3). Sequences not beginning with C or ending with F or W and having less than seven amino acids in the CDR3 were eliminated. Processed data was analyzed using the Immunarch R package50 for computing basic clonotype statistics and visualizing the processed TCR-seq data, in which raw clonotype counts were normalized by library size in counts per million (CPM) with edgeR.51 Individual clonotypes with CPM ≥5 were retained as this threshold efficiently excluded outliers without affecting the mean distribution. Within each sample, clonotypes were ranked by CPM and top 40 most expanded clonotypes were selected. Sample-specific top clonotype lists were merged based on the CDR3 amino acid sequences to create a clonotype-by-sample matrix. Missing values in any sample were assigned a CPM value of 0 to indicate absence. CPM values were then log-transformed, and the final matrix was visualized as a heatmap using the pheatmap package (v1.0.13). Spectral flow cytometry Immunophenotyping was performed on cryopreserved PBMCs using spectral flow cytometry. Briefly, PBMCs were thawed using a CryoThaw device (Medax; Cat# 1-9X-NIH1) in a thawing medium composed of RPMI-1640, 10% heat-inactivated FBS, and 0.02 mg/mL DNase I (Stemcell; Cat# 07900). The thawed cells were centrifuged at 300g for 8 min, and the resulting cell pellet was washed with thawing medium and then with PBS. To assess cell viability and prevent non-specific antibody binding, the cells were resuspended in PBS and incubated with LIVE/DEAD Fixable Blue Dead Cell Stain (Invitrogen; Cat# L23105; 1:1000 dilution) and Human Fc Block (BD Biosciences; Cat# 564219; 1:20 dilution) for 20 min at room temperature, protected from light. Following a wash with PBS, the cells were stained with a pre-titrated surface antibody cocktail (see key resources table) diluted in a staining buffer consisting of PBS with 50% v/v BD Horizon Brilliant Stain Buffer (BD Biosciences; Cat# 563794). This incubation was performed for 20 min at room temperature, protected from light. After staining, the samples were washed twice with PBS and subsequently fixed in 1% paraformaldehyde (PFA) in PBS for 15 min at room temperature. Finally, the cells were washed and resuspended in PBS for data acquisition on a 4-laser Cytek Aurora spectral flow cytometer. Unmixed FCS files were then analyzed using OMIQ (Dotmatics) or FlowJo (BD Biosciences; v10). Principal component analysis (PCA) plots were produced using JMP (v19). For UMAP dimensional analysis, 35,000 events from each timepoint were concatenated and analyzed using the UMAP plugin (v4.1.1)52 within FlowJo. Cytometry by time of flight (CYTOF) Cryopreserved human PBMCs were thawed and rested overnight at 37°C prior to staining with the Maxpar Direct Immune Profiling Assay (MDIPA; Standard BioTools; Cat# 201349). Up to 3 × 106 cells per timepoint were prepared for staining via incubation with an Fc receptor blocking solution (Biolegend; Cat# 422301) and then transferred to an MDIPA tube for surface staining. Additional surface antibodies (see key resources table), including Live/dead 103Rh (StandardBioTools; Cat# 201349) were added by hand. After surface staining, samples were washed and processed following the MDIPA protocol for PBMCs with minimal edits for intracellular staining (see key resources table). Samples were prepared for intracellular staining using the FoxP3/transcription factor staining buffer set (eBioscience; Cat# 00–5523). After intracellular staining, samples were fixed with 1.6% fresh formaldehyde and held overnight at 4°C in Maxpar Fix/Perm with DNA intercalator. Stained samples were acquired on the HeliosTM mass cytometer with EQ four element calibration beads (Standard BioTools; Cat# 201078). The resulting files were normalized and cleaned to isolate single-cell events. Basic analysis was completed within the Maxpar Pathsetter analysis platform to identify lymphocyte subsets or via Flowjo (BD Biosciences; v10). Metal-tagged antibodies not commercially available were conjugated using the Maxpar antibody labeling kit (Standard Biotools) and stored in antibody stabilizer or HRP protector (Boca Scientific; Cat# 131 000 or 222 000). For PC validation datasets,11 patient CyTOF samples with ≥10,000 CD45+ events were concatenated and analyzed using the UMAP plugin within FlowJo. Quantification and statistical analysis Variance normality in all datasets was tested using Shapiro-Wilk tests. Parametric ANOVA with Tukey or nonparametric Kruskal-Wallis tests were used when comparing >2 groups. Parametric unpaired t-tests or nonparametric Mann-Whitney tests were used when comparing two groups. Wilcoxen tests were used for T cell signature score comparisons from scRNA-seq results. Statistical significance for all tests was defined as ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Statistics were performed in Prism (v10). Published: June 26, 2026 Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.xcrm.2026.102889. Supplemental information Document S1. Figures S1–S6 Table S1. DNA variants detected by whole-exome NGS analysis performed on pre-treatment prostate cancer biopsies, related to Figure 1 Somatic genetic variants detected by the whole-exome MI Profile NGS assay conducted at Caris Life Sciences. Variant classifications and variant allele frequencies (VAFs) are shown. The assay also reports MSI status and TMB, which was initially reported as >150 mutations/megabase (see STAR Methods for details on TMB calculation, which was 266 mutations/megabase). Document S2. Article plus supplemental information References Associated Data Supplementary Materials Document S1. Figures S1–S6 Table S1. DNA variants detected by whole-exome NGS analysis performed on pre-treatment prostate cancer biopsies, related to Figure 1 Somatic genetic variants detected by the whole-exome MI Profile NGS assay conducted at Caris Life Sciences. Variant classifications and variant allele frequencies (VAFs) are shown. The assay also reports MSI status and TMB, which was initially reported as >150 mutations/megabase (see STAR Methods for details on TMB calculation, which was 266 mutations/megabase). Document S2. Article plus supplemental information Data Availability Statement •Local law prohibits depositing raw genomic datasets derived from human samples. Single-cell RNA-seq data have been deposited at GEO: GSE332846 and are publicly available as of the date of publication. This paper analyzes existing, publicly available data, accessible at DOIs: https://doi.org/10.1016/j.xcrm.2026.102638, https://doi.org/10.1016/j.ccell.2024.06.009, https://doi.org/10.1158/2159-8290.CD-22-0686, https://doi.org/10.1016/j.ccell.2021.09.010, and https://doi.org/10.1016/j.cell.2022.06.018. Local law prohibits depositing raw genomic datasets derived from human samples. Single-cell RNA-seq data have been deposited at GEO: GSE332846 and are publicly available as of the date of publication. This paper analyzes existing, publicly available data, accessible at DOIs: https://doi.org/10.1016/j.xcrm.2026.102638, https://doi.org/10.1016/j.ccell.2024.06.009, https://doi.org/10.1158/2159-8290.CD-22-0686, https://doi.org/10.1016/j.ccell.2021.09.010, and https://doi.org/10.1016/j.cell.2022.06.018. •This paper does not report original code. This paper does not report original code. •Any additional information required to reanalyze the data reported in this paper is available from 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.