Neoadjuvant PARPi or chemotherapy in ovarian cancer informs targeting effector Treg cells for homologous-recombination-deﬁcient tumors

cancer, sensitizing tumor cells to poly (ADP-ribose) polymerase (PARP) inhibition. However, the impact of HRD and related therapies on the tumor microenvironment (TME) remains elusive. Our study generates single-cell gene expression and T cell receptor proﬁles, along with validatory multimodal datasets from >100 high-grade serous ovarian cancer (HGSOC) samples, primarily from a phase II clinical trial (NCT04507841). Neoadjuvant monotherapy with the PARP inhibitor (PARPi) niraparib achieves impressive 62.5% and 73.6% response rates per RECIST v.1.1 and GCIG CA125, respectively. We identify effector regulatory T cells (eTregs) as key responders to HRD and neoadjuvant therapies, co-occurring with other tumor-reactive T cells, particularly terminally exhausted CD8 + T cells (Tex). TME-wide interferon signaling correlates with cancer cells upre-gulating MHC class II and co-inhibitory ligands, potentially driving Treg and Tex fates. Depleting eTregs in HRD mouse models, with or without PARP inhibition, signiﬁcantly suppresses tumor growth without observable toxicities, underscoring the potential of eTreg-focused therapeutics for HGSOC and other HRD-related tumors.


In brief
Examination of tumor microenvironment dynamics following neoadjuvant PARP inhibition or chemotherapy in newly diagnosed high-grade serous ovarian cancers (HGSOCs) reveals effector Treg cells (eTregs) as a major mediator of homologous recombination deficiency (HRD) and treatment effects.Depletion of eTregs is shown to be effective in suppressing HRD tumors in multiple mouse models.

INTRODUCTION
2][3] It has attracted wide interest because of synthetic lethality, wherein this defect renders tumors vulnerable to poly (ADP-ribose) polymerase (PARP) inhibition. 4,5Multiple PARP inhibitors have been approved for the treatment of ovarian, breast, prostate, and pancreatic cancers, and many more are being developed for or tested in various solid and hematological malignancies. 6,7Although recent studies have started to shed light on the interplay between HRD and the tumor microenvironment (TME) of human cancers, [8][9][10] including ovarian cancer, 11 it remains a top priority to decode the interaction effects of HRD and PARP inhibition on the TME.Practical hurdles to such characterizations, including prior treatment exposures and heterogeneous patient populations, impede a deeper understanding of TME plasticity and the development of novel, effective cancer therapies.
Because high-grade serous ovarian cancer (HGSOC) has the highest prevalence of HRD and often requires neoadjuvant therapies to achieve optimal debulking, 12,13 it exemplifies an archetypal disease for characterizing the effect of HRD on treatment-naive TME and subsequent changes induced by PARP inhibition.Efforts to address this knowledge gap would have crucial implications for understanding why in HGSOC evidence of therapeutic benefits by immune checkpoint inhibitors (ICIs) is lacking. 14Across phase III clinical trials, the addition of PD-1/PD-L1-targeting agents, including nivolumab, 15 avelumab, 16,17 and atezolizumab, 18,19 to chemotherapy and/or targeted therapy (e.g., bevacizumab) failed to confer a survival benefit in a variety of disease settings.Combining nivolumab with CTLA-4-targeting ipilimumab resulted in significantly improved progression-free survival than nivolumab alone but in a smaller cohort. 20Importantly, PD-L1 positive score as a classical ICI biomarker did not show consistent trends with the response rate (RR) in the trials mentioned above.These clinical observations suggest a complex immunomodulatory landscape in HGSOC beyond the PD-1/PD-L1 signaling, calling for a comprehensive search of critical immune factors.
Here, we leveraged a large number of samples from a clinical study to evaluate neoadjuvant PARP inhibition in HRD HGSOC (NCT04507841). 21In parallel, we collected samples from patients receiving neoadjuvant chemotherapy (NACT).Single-cell and bulk multimodal profiling of these samples yielded valuable data that enabled us to delineate the TME divergence between HRD and homologous recombination-proficient (HRP) tumors as well as their phenotypic evolution following the neoadjuvant therapies.

A single-cell atlas of the HGSOC TME under neoadjuvant treatment
To investigate the effects of HRD, neoadjuvant therapies, and their interactions on the TME, we recruited treatment-naive patients with newly diagnosed unresectable HGSOC in our phase II clinical trial (NCT04507841) (Figure 1A; Table S1).We determined the response to neoadjuvant niraparib monotherapy (NANT) or NACT for each patient according to the Gynecologic Cancer InterGroup (GCIG) CA125 response criteria 22,23 (Figure 1B; Table S1).Dynamically and effectively monitoring overall tumor burden within a short period, especially for patients with non-measurable lesions, GCIG CA125 response was suited for longitudinal quantitative analysis of TME correlates.Among 53 HRD patients who received NANT, 14 (26.4%)achieved complete response (CR) and 25 (47.2%)reached partial response (PR), resulting in a GCIG CA125 RR of 73.6% (62.5% per RECIST v.1.1)(Figure 1C).Six HRD patients chose to receive NACT and exhibited an RR of 100.0%, including three patients with CR and three patients with PR (Figure 1C).In parallel, 21 HRP patients received NACT.Six patients yielded CR, and 12 patients acquired PR, resulting in an RR of 85.7% (Figure 1C).In total, 80 patients who underwent either neoadjuvant treatment displayed a GCIG CA125 RR of 78.8% (Figure 1C).Overall, the safety profile of NANT was manageable, and no new safety signal was observed, with hematologic toxicities as the most common treatment-related adverse events.These results indicate that NANT is an effective treatment option for controlling disease progression for HRD patients with HGSOC.
From 34 patients in the trial cohort, we obtained 67 tumor samples (39 pre-treatment and 28 post-treatment), including 24 sitematched pre-and post-treatment pairs as our main cohort for single-cell RNA sequencing (scRNA-seq) (Figure 1D; Table S1).These patients had a balanced distribution of HRD status and genomic aberrations, as surveyed by our HRD assay (Figure 1E; Table S2).The scRNA-seq data were processed through a rigorous quality control (QC) and preprocessing pipeline (Figures S1A-S1D).We observed consistent high quality across samples, with low ambient contamination (Figure S1A), high percell unique molecular identifier (UMI) and gene count, and low mitochondrial DNA UMI and ribosomal UMI proportions (D) Cross-site sampling of tumor and blood samples for single-cell and bulk multi-omic profiling from ovarian cancer patients.Numbers in parentheses indicate patient count and sample count, respectively.(E) Genomic alterations of a panel of homologous recombination repair (HRR) genes across 34 patients in the main cohort.Top, genomic instability scores (GISs); dotted black line indicates a GIS threshold of 42 for HRD approved by the US Food and Drug Administration.Right, sample count with a mutated gene of interest.Germline mutations are highlighted with the ''+'' sign.Silent mutations with clear functional consequences are highlighted with the ''*'' sign.(F) Uniform manifold approximation and projection (UMAP) embedding of all cells in the main cohort with major cell types annotated.(G) Major cell type composition and clinical metadata for each sample.FIGO, The International Federation of Gynecology and Obstetrics; HRD, homologous recombination deficiency; HRP, homologous recombination proficiency; BRCAm, mutated BRCA1/2; BRCAwt, wild-type BRCA1/2; GIS, genomic instability score; ISD, individualized starting dose; TC, paclitaxel and carboplatin; CR, complete response; PR, partial response; NR, non-response; RR, response rate; GCIG, the Gynecologic Cancer InterGroup; CA125, cancer antigen 125.See also Figure S1 and Tables S1 and S2.(legend continued on next page) (Figure S1E).Finally, we computationally integrated >620,000 post-QC cells and assigned them to 10 major cell types based on the expression patterns of well-established marker genes (Figures 1F and S1F-S1H).All samples contained the 10 cell types and showed a balanced distribution of cell counts, indicating a comprehensive and robust sampling of the HGSOC TME (Figure 1G).In addition to this main cohort, we performed single-cell T cell receptor sequencing (scTCR-seq), multiplex immunohistochemistry (mIHC), flow cytometry, and bulk TCRseq for additional tumors and patient-matched peripheral blood samples (Figure 1D; Tables S3, S4, S5, and S6).Together, this rich, multimodal, and high-quality dataset represents a unique resource to elucidate how tumor-intrinsic HRD status and realworld therapies affect the TME of ovarian cancer.

HRD-tumor-enriched eTregs underlie the TME response to neoadjuvant therapies
Intertumoral TME heterogeneity established before therapeutic interventions often portends diverse patient outcomes. 24,25At the level of major cell types, we observed a trend toward an increase of lymphocytes and endothelial cells while a decrease of epithelial cells and myeloid cells in pre-treatment metastatic tumors (Figure S2A), in line with previous studies, 11,26,27 albeit not reaching statistical significance except for B cells.Similarly, no major cell types showed a significant difference between pre-treatment HRD and HRP tumors (Figure S2B).For the effects of NACT and niraparib, we identified mast cells and endothelial cells to be significantly enriched in post-treatment tumors, respectively (Figures S2C and S2D), which merits further investigation.
These results prompted us to examine the impact of HRD and neoadjuvant therapies at a finer scale.For this purpose, we first built a cell-state hierarchy that covered all major cell types through an extensive search of the literature for transcriptional evidence supporting the identities of 97 cell states (Figure S3; STAR Methods).For example, our annotation revealed 15 and 12 states in the CD4 + and the CD8 + T cell compartments, respectively (Figures 2A and 2B).We then compared cell-state proportions between HRD subtypes and treatment phases, identifying a wide range of cell-state shifts (Figure 2C).Interestingly, many cell states showed concordant abundance changes in chemotherapy and niraparib treatment (Figure 2D).
To focus on the most convergent signals, we identified four overlapped cell states out of the three comparisons, namely interferon (IFN)-responding myofibroblastic cancer-associated fibroblasts (myCAFs), effector regulatory T cells (eTregs), and proliferative CD4 + /CD8 + T cells (Figures 2C, 2E, and S2E-S2G), all of which had higher abundance in the treatment-naive HRD TME while decreased in post-treatment HRD tumors.The Treg is the most abundant T cell population in our dataset, and its eTreg subset is well known for immunosuppressive functions. 28,29We thus focused on validating the related observations.First, we analyzed a published scRNA-seq cohort, 30 under a matched T cell annotation scheme (Figure S4), to recapitulate a dramatic decrease of eTregs but not resting Tregs in post-NACT HGSOC tumors (Figure 2F).Second, we observed through a signature-based analysis of the TCGA HGSOC cohort that eTregs had a significantly higher infiltration level in HRD tumors (Figure 2G), which, when extended to contrast with other non-Treg T cell-state marker genes, further revealed the unique enrichment of eTregs (Figure 2H).Third, we performed mIHC of 85 HGSOC tumors (Table S3) and confirmed total Tregs and eTregs showing a significantly higher abundance in HRD tumors while markedly reduced after niraparib treatment (Figures 2I  and 2J).Importantly, cancer cells (panCK + ) and eTregs resided nearby in HRD tumors, implying a more frequent functional interaction (Figures 2I and 2K).Lastly, we quantified the proportion of total Tregs and eTregs in another 35 treatment-naive tumors using flow cytometry (Table S4) and confirmed their stronger presence in the HRD TME (Figures 2L and 2M).These results show that the marked reduction of HRD-enriched eTregs represents the most robust change upon neoadjuvant therapies.(M) Difference in the proportion of CD25 + CD127 low and CD25 + CD127 low CCR8 + cells among CD4 + T cells between HRP and HRD samples according to flow cytometry results (n = 8 HRP samples, n = 21 HRD samples for CD25 + CD127 low cells; n = 5 HRP samples, n = 19 samples for CD25 + CD127 low CCR8 + cells).FC, fold change; R s , Spearman rank correlation coefficient.FPKM, fragments per kilobase of transcript per million mapped reads.For boxplots, the middle line in the box is the mean, the bottom and top of the box are the first and third quartiles, and the whiskers extend to the 1.53 interquartile range of the lower and the upper quartiles, respectively.p values were calculated using a two-sided paired Wilcoxon rank-sum test for (F) and a two-sided unpaired t test for (G), (H), (J), and (M).*p < 0.05, **p < 0.01, ***p < 0.001.See also Figures S2, S3, and S4 and Tables S3, S4, and S7.populations, including the eTreg (Figure 2E), suggesting a dominant role of eTreg-centric T cell-state shifts mediating HRD and treatment effects.In further investigating this phenomenon, we identified a cluster of CD4 + and CD8 + T cell states co-occurring with eTregs and recapitulated the pattern in an independent HGSOC cohort (Figure 3A).Notably, almost all of these T cell populations were enriched in HRD tumors and showed elevated JAK-STAT signaling (Figure 3A).Given that HGSOC tumors with HRD are known to harbor higher tumor mutational burden (TMB) 3,31 and potentially produce more neoantigens, 32 this observation suggested a commonly activated state among the T cells responding to accrued tumor antigens.Indeed, in addition to eTregs, 33,34 this cluster contained several populations known to be enriched for tumor-reactive T cells, such as exhausted CD8 + T cells (Texs) 35,36 and CXCL13 + PD-1 + CXCR5 À peripheral T helper-like (''Tph-like'') cells (Figure 3A). 37,38Accordingly, using mIHC, we validated that eTregs and Texs co-occurred and both had a stronger presence in HRD tumors (Figure 3B).

Tregs and
Next, we provided multiple lines of evidence of their tumor reactivity, specifically in our HGSOC context.First, most of the reactive populations showed a uniquely heightened gene expression signature 36 derived from T cells bearing validated neoantigen-specific TCRs (Figure 3C).Second, proliferative CD4 + and CD8 + T cells expressed Treg and Tex programs, respectively, at a level almost on par with non-proliferative Tregs and Texs (Figure 3D), suggesting that Tregs and Texs bore remarkable proliferative potential.Third, using our paired scRNA-seq/TCR-seq data (Table S5) of enriched tumor-infiltrating T cells, we identified Treg, Tex, Tph-like, and proliferative T cells as the most expanded populations (Figure 3E).Importantly, these expansions were universally strong across all surveyed samples (Figure 3F), demonstrating their robust activation in the HGSOC TME.Finally, Tregs and Texs showed intertumoral co-expansion, again pointing to functional coordination potentially orchestrated by varying tumor antigen burdens (Figure 3G).
Although the co-existence of immunosuppressive eTregs with other activated, reactive T cell populations in HRD tumors presented a well-regulated balance of anti-and pro-tumor forces, the neoadjuvant therapies specifically counteracted eTregs, Treg-like proliferative CD4 + T cells, and Tex-like proliferative CD8 + T cells (Figure 2C), securing an anti-tumor net effect.To further study the treatment effect from a T cell clonal expansion standpoint, which is an axis of tumor reactivity independent of transcriptional phenotypes, we turned back to our treatment-experienced main scRNA-seq cohort to delineate the longitudinal TCR repertoire dynamics using the TCR repertoire utilities for solid tissue 4 (TRUST4) algorithm (Figure S5A). 39Supporting the validity of the inferred data, we found that the TCRs (1) were exclusively enriched in transcriptome-defined T cells (Figure S5B), (2) formed clones with extremely high privacy (Figures S5C and S5D), and (3) recapitulated Treg dominance in the CD4 + population and overall stronger CD8 + expansion, with Tex among the top (Figures S5E and S5F).With these data, the only observed treatmentinduced differential expansions after niraparib treatment happened in eTregs and CD8 + effector memory re-expressing CD45RA (Temra) cells, with the former suppressed while the latter elevated.By contrast, Tex expansion was not changed by niraparib treatment (Figure 3H).We thus reasoned that the suppression of eTregs by NANT in the HRD TME, both abundance-wise and clonal-expansion-wise, should lead to even more heightened tumor reactivity.Indeed, longitudinally sequenced bulk TCR repertoires of 60 tumors and matched peripheral blood mononuclear cells (PBMCs) from 23 patients in the NANT trial revealed a tumor-restricted significant decrease in TCR diversity and an increase in clonality in most patients (Figure 3I).Together, these results unraveled the complex kinetics of HRD-dependent and therapy-perturbed tumor-reactivity landscape of tumor-infiltrating T cells wherein the key immunoregulatory role of eTregs was highlighted.(A) Abundance co-variation between CD4 + and CD8 + T cell states across treatment-naive samples for in-house data (lower triangle) and data from GSE165897 (upper triangle).A cluster of highly correlated cell states containing eTregs is highlighted.Circles on both sides of the main heatmap denote in size the proportion of cells within a cell state with a positive JAK-STAT activity score and in color gradient the mean JAK-STAT activity score.The leftmost squares denote in color gradient the log 2 FC of cell-state abundance in HRD over HRP tumors.Squares with a statistically significant difference are marked by an asterisk.FC, fold change.(B) Correlation between the proportion of CD4 + FOXP3 + TNFRSF9 + cells among CD4 + cells and that of CD8 + PD-1 + cells over CD8 + cells across treatment-naive samples based on mIHC (n = 18 HRP samples, n = 34 HRD samples).Representative cell identity maps highlight the co-occurrence of CD4 + FOXP3 + TNFRSF9 + cells and CD8 + PD- (H) Difference in TCR clonotype expansion proportions of eTreg (left), CD8 + Temra (middle), and terminal Tex (right) between pre-and post-niraparib samples (n = 12 pre-nira samples, n = 6 post-nira samples for eTreg; n = 6 pre-nira samples, n = 7 post-nira samples for CD8 + Temra; n = 8 pre-nira samples, n = 4 post-nira samples for terminal Tex).(I) Difference in total clonal diversity between pre-and post-niraparib blood or tumor TCR repertoire, as measured by clonality, Gini coefficient, D50 index, and normalized Shannon-Wiener index (n = 12 patients).For boxplots, the middle line in the box is the mean, the bottom and top of the box are the first and third quartiles, and the whiskers extend to the 1.53 interquartile range of the lower and the upper quartiles, respectively.R s , Spearman rank correlation coefficient.p values were calculated using a two-sided Mann-Whitney U test for (A), a two-sided unpaired t test for (H), and a two-sided paired Wilcoxon rank-sum test for (I).ns, non-significant, *p < 0.05, **p < 0.01, ***p < 0.001.See also Figure S5 and Tables S5 and S7.
Intra-Treg/Tex cell-state shifts driven by HRD and neoadjuvant therapeutic responses Analysis of tumor-infiltrating Tregs [40][41][42] and Texs [43][44][45] has revealed high intra-population heterogeneity for both.However, it remains unclear to what extent Texs behave similarly to and potentially coordinate with their terminally differentiated CD4 + counterparts, eTregs, on a finer scale.Our scRNA-seq data captured >20,000 Tregs and $10,000 Texs, thereby allowing a deep interrogation into the Treg and Tex sub-states (Figures 4A-4P).We were able to further classify eTregs into 9 sub-states, and along with resting Tregs, we identified 10 substates of Tregs in total (Figure 4A); in parallel, we identified 5 sub-states of Texs (Figure 4H).For Tregs, we observed a clear differentiation gradient supported by RNA velocity, with the resting population as the starting point and the terminal effector one as the endpoint (Figure 4B).Texs showed a similar progenitor-to-terminal trajectory by RNA velocity, although with a greater level of complexity (Figure 4I).In Tregs, resting cells highly expressed known T cell naiveness/memory markers such as CCR7, LEF1, and KLF2, while the terminal cells were enriched for Treg markers like TNFRSF18, CCR8, and LAG3 (Figure 4C).Similarly, Texs showed a state separation based on enriched expressions of progenitor markers such as GZMK or CMC1 and terminal exhaustion markers such as ENTPD1 or LAYN (Figure 4J).Additionally, both terminally differentiated Tregs and Texs showed the strongest suppression/exhaustion and tumor-reactivity signatures (Figures 4D and 4K).Gene regulatory networks constructed through single-cell regulatory network inference and clustering (SCENIC) 46 revealed statespecific transcription factor activities in Tregs and Texs.Regulons of known lineage drivers, such as TCF7 and NR3C1 for T cell memory and BATF and NFKB1 for Treg maturation, were specifically overrepresented in cells of matched Treg states (Figure 4E).Likewise, Texs exhibited TCF21 and IFZF1 activities in the progenitor state, while NFKB1, REL, CEBPB, and JUN were present in the terminal state (Figure 4L).Beyond their transcriptional states, we recapitulated a clear resting/progenitor-to-terminal trajectory by integrating RNA velocity and clonal expansion in our scRNA-seq/TCR-seq cohort, with the terminal domains enriched for the most expanded clones (Fig- ures 4F and 4M).These state transitions were again confirmed by differential expression patterns of canonical marker genes (Figures 4G and 4N).
Armed with these finest cell-state maps, we next investigated how the interaction between HRD and therapies was manifested through cell-state perturbations within Tregs and Texs.Graphbased differential abundance testing using Milo 47 pinpointed the terminal states of Tregs and Texs as significantly enriched in HRD tumors while depleted after treatment (Figures 4O and  4P).We thus revealed an intriguing consequence of neoadjuvant therapies to be the reversal of Treg and Tex differentiation gradients.

Differential intratumoral and peripheral clonal origins of Tregs and Texs
To further understand how these differentiated and reactive populations were instituted in the TME under HRD and later perturbed by treatment, we interrogated their clonal origins using our scRNA/TCR-seq data.Associating the cell states by their TCR repertoire overlaps, we found CD8 + states to be more extensively exchanged among each other, while CD4 + state exchanges were largely confined to a few cases (Figure 5A).Interestingly, effector Tph-like cells and Tregs had many shared clones, and both showed little overlap with the other CD4 + states.The exception was with the proliferative population, again indicating a tumor-reactive convergence constantly replenished by a proliferative pool (Figures 3A, 3C, and 5A).By contrast, the Tex repertoire in the CD8 + compartment had large overlaps with all effectors as well as the resident memory and the proliferative populations (Figure 5A).There was limited sharing of Texs with central memory cells and Temras, suggesting a differentiation path diverging from the rest of the lineage (Figure 5A).
We then traced Treg and Tex clonal evolution across all individuals by assessing the extent these clones could be found in other populations in the private lineage history of each patient.Tregs showed moderate clonal kinship with proliferative CD4 + T cells in most samples and with Tph-like cells in several samples (Figure 5B).For Texs, besides their strong connection with GZMK + progenitors, tissue-resident memory (Trm) and several effector subsets provided a possible source as well (Figure 5B).Notably, a TCR repertoire overlap does not necessarily indicate a directional evolutionary relationship between two populations.But overall, our clonal tracing results implicated intratumoral non-Treg and non-Tex populations as the predecessors in the Treg and Tex differentiation trajectories, respectively.See also Table S7.
Several studies have suggested, outside of the induction from conventional CD4 + T cells, a possibility of peripheral replenishment of tumor-infiltrating Tregs, albeit in a context-specific manner. 42,48To gauge how frequently tumor and blood TCR repertoire engaged in crosstalk in HGSOC, we first calculated the blood-tumor repertoire overlap across patients using our bulk TCR-seq data (Table S6).The proportion of shared clones varied greatly across individuals and was independent of niraparib treatment (Figures 5C-5E).Importantly, for most individuals, overlapped clones showed a strong co-expansion pattern, indicating an effective clonal supplement from the circulation (Figures 5F and 5G).
Since bulk TCR-seq data lacked a cell-state resolution, we directly searched for the peripheral presence of intratumoral Treg clones by profiling bulk TCR repertoires of purified blood CD4 + T cells from the same patients in our scRNA-seq/TCRseq cohort (Table S6).Surprisingly, this analysis revealed negligible clonal exchanges between the two compartments (Figure 5H).Few CD4 + cell states had >5% of their TCR clones found in the periphery blood, and Tregs showed the weakest signals with an average overlap of <1% (Figure 5I).Thus, Tregs seemed to possess a strong but tumor-restricted self-expansion capability through a constant exchange with the CD4 + proliferative pool.By contrast, highly expanded Texs could converge from multiple effector CD8 + subsets.

IFN-induced MHC class II expression in HRD tumor cells potentially contributes to eTreg enrichment
To understand where terminal eTregs or terminal Texs, two populations of the utmost interest based on our previous analyses, were positioned in a hierarchical and coordinated TME response, we examined TME-wide the cell states they co-occur with.Interestingly, such cell states were commonly marked by the activation of IFN signaling (Figures 6A and 6D).These included M1-like macrophages, IFN-responding myCAFs, IFNresponding cancer cells, and THY1 + pericytes (Figures 6A and  6D).Importantly, the direction and magnitude of these co-variations were strictly aligned with the Treg/Tex differentiation gradients (Figures 6B and 6E).Additionally, they remained significant when we switched the calculation of cell-state proportion to one level above, namely eTregs over CD4 + T cells and Texs over CD8 + T cells (Figures S6A-S6F).
Behaviors of cancer cells are fundamental driving forces of the TME and are often the direct targets of chemotherapy and targeted therapies such as niraparib. 49Among the three major cancer cell populations in our dataset, the IFN-responding population strongly co-occurred with terminal eTregs (Figure 6C) and, to a lesser extent, with terminal Texs (Figure 6F).To understand how IFN-responding cancer cells would specifically contribute to Treg enrichment, we first evaluated the differential antigen-dependent Treg interactions among cancer cell states by comparing their antigen presentation machinery expressions.Interestingly, IFN-responding cancer cells, enriched in HRD tumors (Figure 6G), had significantly upregulated major histocompatibility complex (MHC) class II expressions.MHC class I genes already had a strong presence at baseline in non-IFN-responding states and were of even higher abundance in the IFN-responding population (Figure 6H).This reflected a well-established role of IFN in upregulating both MHC class I and MHC class II gene expression through activating the master regulators, CITA and CIITA. 50,51pithelial cells, especially those lining the intestine, have long been known to possess MHC class II-dependent Treg-promoting capabilities. 52,53However, only recently have several lines of evidence emerged to support a similar model in tumor contexts. 54,55In our data, IFN-responding cancer cells did not express CD80/86 genes at all and had minimal expressions of most alternative co-stimulatory molecules (Figure 6H).This was in sharp contrast to canonical antigen-presenting cells such as dendritic cells and macrophages, which not only possessed a full set of MHC class I/MHC class II complexes but had a strong presence of co-stimulatory genes (Figures 6H and S6G).We further observed that even in the other cancer cell populations not typified by enhanced IFN signaling, HRD had a strong positive contribution to elevated MHC class II gene expressions, based on a transcriptome-wide pseudobulk analysis (Figure 6I).This close connection between HRD and MHC class II activation is well aligned with HRD-driven genomic instability being a major activator of the cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) DNA sensing pathway. 56,57Conversely, the expression level of MHC class I genes showed no difference between HRD and HRP cancer cells in all three states (Figures S6H and  S6I).To validate these findings, we experimentally demonstrated the MHC class II responsiveness to IFN treatment or BRCA1/2 deficiency in multiple ovarian or breast cancer cell lines (Figure 6J).Therefore, interacting with MHC class II-positive cancer cells for CD4 + T cells may lead to a Treg phenotype through aberrant stimulation of the TCR signaling without essential subsequent co-stimulatory signals. 58,59esides passively sequestering T cells into regulatory and dysfunctional states, cancer cells may actively propagate suppressive signaling through the upregulation of co-inhibitory molecules. 58,60Indeed, our sample-level analysis showed  S5 and S6.strongly increased expressions of many co-inhibitory molecules in IFN-responding cancer cells (Figure 6K), including factors with established roles in favoring Texs and/or Tregs, such as PD-L1, 61,62 Galectin-9, 63,64 indoleamine 2,3-dioxygenase (IDO), 65,66 and ICOSL. 67,68eg depletion as an effective therapeutic strategy to suppress HRD tumors Our systematic analyses highlighted the abundant enrichment of immunosuppressive eTregs in HRD tumors and their marked post-treatment reduction, which may be a key contributor to the HRD-dependent clinical benefits of PARP inhibition in HGSOC patients.Indeed, eTregs appeared as one of the peripheral immune subsets significantly reduced by maintenance PARPi in epithelial ovarian cancer patients in a recent translational study.69 In our NANT trial, the post-treatment CA125 level across patients showed a strong positive correlation with the proportion of tumor-infiltrating terminal eTregs (Figure 7A).This link between patient response and remaining eTregs prompted us to hypothesize that directly targeting eTregs in combination with niraparib would achieve an improved tumor-suppressing effect in the HRD context.
1][72] To evaluate its efficacy, we first constructed an orthotopic HGSOC mouse model through intrabursal injection of ID8-Luc cells with CRISPR-edited Trp53 and Brca1 into CCR8-humanized (hCCR8) mice.These HRD Trp53 À/À tumors harbored a significantly higher proportion of eTregs and terminal Texs than their HRP counterparts, demonstrating human HGSOC relevance (Figure 7B).We then administered niraparib and/or CCR8 mAb to tumor-bearing mice after about 2 weeks of transplantation (Figure 7C).Although tumor progression was attenuated by niraparib or CCR8 mAb alone, their combination indeed showed a significantly more pronounced inhibitory effect (Figures 7D-7F).Importantly, the development of hemorrhagic ascites, a lethal hallmark of ovarian cancer, 73,74 was dramatically curbed by the combination ther-apy (Figure 7G).PD-1 + Treg frequency was significantly reduced by all therapies, with the largest reduction seen in the combination treatment group, as measured independently by flow cytometry (Figure 7H) and IHC (Figures 7I and 7J).In addition to CCR8 targeting, we explored another Treg-depleting therapy and observed similar anti-tumor effects in conjunction with PARP inhibition.Specifically, niraparib and CD25 mAb together (Figure 7K) significantly slowed tumor progression (Figures 7L  and 7M), prevented hemorrhagic ascites (Figure 7N), and dampened PD-1 + Tregs (Figure 7O) in the same orthotopic model but without the humanized CCR8.Exhibiting broader applicability, the combination regimen had its superior anti-tumor effect recapitulated in two orthotopic breast cancer mouse models, namely hCCR8 mice bearing Brca2-deficient EO771 cells (Figures 7P-7R) or wild-type (WT) mice bearing Brca1-deficient EO771 cells (Figures 7S-7U), both at the inguinal mammary fat pad.Notably, the non-response to single-agent niraparib in the Brca2 À/À breast cancer model stood in contrast to responses in the Brca1 À/À models, likely because BRCA1 and BRCA2 mutations differentially reshaped the TME in different cancer contexts.Finally, we evaluated the potential toxicities induced by different therapies with routine blood and biochemical examinations in mice.Notably, there was no significant myelosuppression (Figures S7A and S7B) or tissue damage (Figures S7C and  S7D) in mice treated with the combination therapy compared with those receiving the monotherapy or control.Thus, the depletion of eTregs by combining niraparib and CCR8/CD25 mAbs represents a promising intervention strategy for tumors harboring HRD.Further investigations are merited to determine whether the combination therapy can provide long-term memory responses and survival benefits.

DISCUSSION
Through comprehensive profiling of a neoadjuvant trial cohort, our study was endowed with a unique ''window of opportunity'' 75 to investigate for the first time the interaction effects of HRD and PARP inhibition on the TME of treatment-naive HGSOC tumors.Although previous in vitro and preclinical studies showed that PARP-1 deletion or inhibition promotes Treg differentiation and (K) Difference in pseudobulk gene expression Z scores of co-inhibitory molecules among cancer cell states (n = 67 samples).Lines connect data points of the same sample.Labels in parentheses are common protein names.DC, dendritic cell; MFI, median fluorescence intensity; R s , Spearman rank correlation coefficient.p values were calculated using a two-sided unpaired t test for (G) and (J) and a two-sided paired t test for (K).ns, non-significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.See also Figure S6.suppressive functions in non-cancer contexts, [76][77][78][79] our observation of eTreg depletion by PARPi is the net effect of a series of complex TME events, succeeding a collective emergence of eTregs and Texs in response to heightened TMB induced by HRD.In searching for an underlying mechanism, we identified a group of HRD-enriched IFN-responding cancer cells bearing aberrantly high expressions of MHC molecules and co-inhibitory ligands while lacking co-stimulatory molecules, thus potentially reinforcing the presence of Tregs/Texs.1][82][83][84] The net effect of IFN on the TME may heavily rely on the location, timing, and duration of exposure.Depending on the tissue context and progression status of a tumor, these parameters can drastically vary among cancer patients.Considering this, chronic IFN signaling underlying adaptive immunosuppression is particularly relevant in HGSOC because most HGSOC tumors are diagnosed at an advanced stage with widespread affected areas. 85,86A long latency period as such allows HGSOC tumors with genomic instability to produce IFN persistently.
In our data, rampant IFN signaling leads to widespread immunoediting within the TME, manifested through the delicate balance where the emergence of immunosuppressive eTregs goes hand in hand with the surge of tumor-reactive effector T cells.Echoing a recent study by Oliveira et al. that established tumor-reactive Tregs to be a major mediator of immune evasion in high-TMB melanoma, 54 our study hints at, in the absence of direct evidence supporting Treg tumor reactivity, a similar maneuver adopted by HGSOC and potentially HRD tumors in general.Accordingly, when the anti-CCR8 mAb was used to directly and highly specifically deplete eTregs, HRD tumors were effectively suppressed in multiple syngeneic mouse models.Our strategy highlighted a focus on removing the dominant immunosuppressive force to secure a net gain of anti-tumor effects.However, HRD is a unique genomic feature whose definition and utility are largely restricted to a few cancer types (e.g., ovarian cancer, breast cancer, and prostate cancer).The extent to which the effect of HRD on the TME can be generalized to other cancer types with different kinds of genomic instability is a key question in future research.
In summary, our in-depth analysis of patient samples from a well-controlled trial cohort together with mechanistic interrogations in highly relevant preclinical models (1) revealed cancer-cell-extrinsic mechanisms underlying HRD as a valuable biomarker; (2) confirmed the potential of niraparib as a candidate for chemo-free neoadjuvant therapeutics; and (3) unraveled a rational realm of eTreg-centered immunotherapeutic opportunities in HGSOC.Future efforts are warranted to take this potential to transform the current treatment landscape where well-established ICIs have failed to generate promising responses.

Limitations of the study
Our study has several limitations regarding data collection, analysis, and interpretation.First, we heavily relied on scRNA-seq data to derive cell type/state proportions, which may obscure the measurements of certain populations due to sampling and processing bias.However, for our major focus, the T lymphocytes, mIHC and flow cytometry analyses on a large number of tumor samples validated the findings regarding Treg enrichment in HRD and reduction after treatment.Second, previous studies have demonstrated intratumoral or site-specific TME heterogeneity by profiling multiple tumor sites within a patient.Our analysis did not consider this factor, mainly due to ensuring and maximizing patient well-being in obtaining pre-treatment diagnostic laparoscopic biopsies to avoid additional trauma and bleeding.Also, maximizing the patient count per cohort ensured (K) Schematic illustration of Trp53 À/À Brca1 À/À ID8 cells transplanted into wild-type (WT) mice via intrabursal injection 2 weeks before treatment with niraparib (i.g.) and/or CD25 mAb (i.p.) (n = 7-8 mice/group).(L-O) Same as (D) and (F)-(H) but for niraparib (i.g.) and/or CD25 mAb (i.p.) in wild-type (WT) mice (M, n = 7 mice/group; O, n = 6-7 mice/group).(P) Schematic illustration of Brca2 À/À EO771 cells transplanted into the inguinal mammary fat pad of hCCR8 mice and the administration of niraparib (i.g.) and/or CCR8 mAb (i.p.) (n = 6 mice/group) after the tumor volume reached $50 mm 3 .(Q) Change in tumor volumes at multiple time points among treatment groups (n = 6 mice/group, two-way analysis of variance, *p < 0.05).(R) Images of surgically resected Brca2 À/À EO771 tumors at endpoint among treatment groups.Scale bars, 1 cm.(S-U) Same as (P)-(R) but for niraparib (i.g.) and/or CD25 mAb (i.p.) in WT mice (T, n = 6 mice/group; U, scale bars, 1 cm).i.g., intragastric administration; i.p., intraperitoneal administration.Data are shown as mean values ± SEM. p values were calculated using a two-sided unpaired t test except for (Q) and (T) where two-way analysis of variance (ANOVA) was used.*p < 0.05, **p < 0.01, ***p < 0.001.See also Figure S7.
sufficient statistical power for inter-tumor comparisons, which is a focus of our study.Third, we used computational methods to infer key biological processes from our scRNA-seq data, such as longitudinal TCR repertoire dynamics and Tex differentiation trajectory.Although these inference data cannot replace direct measurements, subsequent multifaceted analyses validated good inference quality and led to important discoveries that would otherwise be masked.

Materials availability
All unique/stable reagents generated in this study are available from the lead contact with a completed Materials Transfer Agreement.
Data and code availability d Raw sequencing data have been deposited in the Genome Sequence Archive for Human at the National Genomics Data Center with accession number HRA007180 for 10x scRNA-seq, HRA007216 for Singleron scRNA-seq, HRA007181 for Singleron scTCR-seq, and HRA007230 for bulk TCR-seq.All records are under project PRJCA016620 (https://ngdc.cncb.ac.cn/ bioproject/browse/PRJCA016620), which will be provided for scientific research upon request complying with the law due to human patient privacy concerns.Processed, de-identified scRNA-seq, scTCR-seq, and bulk TCR-seq profiles are publicly available at the Gene Expression Omnibus (GEO) with accession number GSE222557.This study also utilized existing, publicly available data, the accession numbers for which are listed in the key resources table.d All original code has been deposited at Zenodo in the form of Jupyter Notebooks or R scripts (https://zenodo.org/doi/10.5281/zenodo.10990093).d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS Patient recruitment and sample collection
Niraparib-based neoadjuvant therapy in HGSOC patients with HRD is a multicenter, prospective, single-arm, open-label, phase II clinical study, and the comprehensive study design was described in our previous publication. 21We consecutively recruited patients who were newly diagnosed with unresectable HGSOC (Fagotti score R 8 or upper abdominal computed tomography [CT] score R 3) and would receive neoadjuvant therapies followed by interval debulking surgery (IDS) to be the candidate population between March 5, 2021, and July 18, 2023.
HRD testing was performed on tumor biopsy samples (tumor content R 30%) using a customized and validated HRD assay 131 that profiles germline and somatic mutations in a panel of 36 DNA damage repair genes and calculates a genomic instability score (GIS) in seven business days.Tumor samples with GIS R 42 or with pathogenic or likely pathogenic BRCA mutations were defined as HRD, and the remaining were designated HRP.Niraparib was administered to eligible HRD patients at a dose of 200 or 300 mg (body weight R 77 kg and platelet count R 150000/UL) for 2 cycles, each lasting 28 days.Patients who were not interested in the clinical study (HRP or subjective rejection) were given carboplatin AUC = 5 and paclitaxel 175mg/m 2 every 3 weeks for 2-3 cycles.The responses to neoadjuvant therapies were determined according to Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1 and the Gynecologic Cancer InterGroup (GCIG) CA125 response criteria, 22,23 as tumor burden changes and were categorized as complete response (CR), partial response (PR), or non-response (NR).Patients with CR or PR would subsequently receive IDS.The incidence of adverse events (AEs) at any grade during the treatment, surgery, and chemotherapy was estimated by Common Terminology Criteria for Adverse Events (CTCAE version 5.0).
This study was approved by the Ethics Committee of Tongji Medical College, Huazhong University of Science and Technology ([S122]-5) and strictly complied with the ethical principles in the Declaration of Helsinki, Good Clinical Practice, and applicable regulatory requirements.All patients provided informed consent for the collection of tissue and blood samples for research and genomic profiling.34 patients were enrolled in the main scRNA-seq cohort, 17 of which received niraparib treatment, and 17 received chemotherapy.For these patients, we obtained a total of 67 tumor samples, including 24 matched pre-and post-treatment specimen pairs (Table S1).Additional tumor and blood specimens were collected for mIHC (Table S3), flow cytometry (Table S4), scRNA/TCR-seq (Table S5), and bulk TCR-seq (Table S6).Pre-treatment biopsies were collected via laparoscopic sampling before clinical enrollment.The site-matched post-niraparib tumor samples were collected via open laparotomy IDS when the patients completed the 2 cycles of niraparib, and an objective response assessment was performed.For patients with chemotherapy, the post-treatment samples were obtained through surgical resection, according to NCCN guidelines.All the tumor samples were collected strictly according to clinical practice guidelines.

METHOD DETAILS Sample collection and processing
Peripheral blood was collected from donors using Streck Cyto-Chex BCT (Streck, La Vista, NE, USA).PBMCs were isolated from the peripheral blood with human PBMC separation liquid (Ficoll, TBD Science) and were re-suspended in 1 3 PBS for further immunostaining and T-cell sorting.Pre-and post-treatment tumor samples were site-matched and obtained as described above.Fresh pre-treatment tumor samples were shipped in Mylteni Biotecâ Tissue Storage Solution and maintained at 4 C until processing.An 8 cm 3 (2 3 2 3 2 cm) tumor tissue derived from each patient was cut into three pieces: one was sent to Precision Scientific laboratory for HRD testing, another was dissociated into a single-cell suspension for scRNA-seq, and the third was used to generate paraffin-embedded samples for IHC staining.The post-treatment tumor samples were divided into two pieces for scRNA-seq and IHC staining.To generate single-cell suspensions, tumor tissues were dissociated immediately after they arrived in the lab using the gentleMACSÔ Dissociator with Heaters and the human tumor dissociation kit (Mylteni Biotec), as per the manufacturer's instructions.Following dissociation, single-cell suspensions were filtered using MACS SmartStrainers (70 mm and 40 mm).They were then treated with Red Blood Cell Lysis Solution and one or two rounds of dead cell removal using the Dead Cell Removal Kit, depending on the number of dead cells in each sample.Finally, cell pellets were re-suspended in 1 3 PBS for further scRNA-seq or scTCR-seq.

HRD assay
Blood, formalin-fixed paraffin-embedded (FFPE) tissue specimens, and fresh tumor samples were collected for each participant.Genomic DNA was extracted from blood samples using the DNeasy Blood & Tissue Kit (QIAGEN).Blood samples were lysed in Proteinase K-supplemented lysis buffer.Buffering conditions were adjusted to provide optimal DNA-binding conditions, and the lysate was loaded onto the DNeasy Mini spin column or the DNeasy 96 plate.During centrifugation, the DNA was selectively bound to the DNeasy membrane, whereas contaminants passed through.The remaining contaminants and enzyme inhibitors were further removed in two wash steps, and the DNA was then eluted in water or buffer.The collected FFPE tumor specimens were stained with hematoxylin and eosin (H&E) to evaluate tumor purity.The process of HRD testing was only performed for samples with an estimated tumor purity of > 30%.Genomic DNA was extracted from fresh tumor samples using the DNeasy Blood &Tissue Kit and the above-mentioned protocol for blood samples.The purified DNA was quantified on a Qubit 4.0 (Life) to confirm a sufficient amount for downstream experiments.The extracted DNA from blood and fresh tumor samples was captured using the Precision Human HRD Assay (Precision Scientific (Beijing) Co., Ltd.), which is a customized panel of probes against sites of nucleotide polymorphisms to quantify GIS and genomic regions of 36 DNA damage repair genes for germline and somatic mutation calling.The captured DNA was sequenced on an Illumina NovaSeq 6000 NGS instrument.

scRNA-seq library preparation and sequencing
The scRNA-seq library was prepared with Chromium Next GEM Single Cell 3' Reagent Kits v.3.1 from 10x Genomics according to the manufacturer's instructions.In brief, the single-cell suspensions dissociated from tumors (derived from P01-P34) were centrifuged and washed twice with 1 3 PBS, and then the single-cell pellets were re-suspended within 1 3 PBS supplemented with 0.04% BSA at a final concentration of 1 3 10 6 cells/mL.About 10 mL of cell suspensions (z10000 cells) were mixed with barcoded Single Cell 3ʹ v.3.1 Gel Beads and Partitioning Oil to generate nanoliter-scale Gel beads in EMulsion (GEMs).Next, the GEMs were subjected to reverse transcription PCR using a T100 Thermal Cycler (Bio-Rad) in a 125 mL volume (program: 53 C for 45 min, 85 C for 5 min, hold at 4 C) to obtain full-length cDNA with barcoding.After this, the first-strand cDNAs were incubated with Dynabeads Cleanup Mix to purify them from the PCR mix and then amplified using the Amplification Reaction Mix on a T100 Thermal Cycler (Bio-Rad) (program: 98 C for 3 min, 98 C for 15 seconds, 63 C for 20 seconds, 72 C for 1 min [step 2 to step 4 for total 12 of cycles], and hold at 4 C).Subsequently, the amplified cDNA was used to generate a single-cell 3ʹ gene expression dual index library using Chromium Next GEM Single Cell 3' Library Kit v.3.1 (10x Genomics).Finally, the well-prepared single cell 3ʹ gene expression libraries were sequenced on an Illumina NovaSeq 6000 sequencer.

scTCR-seq library preparation and sequencing
The scTCR-seq library was generated using the sCircleâ Single Cell Full-Length Immunoreceptor Library Kit (Singleron) according to the manufacturer's instructions.In brief, the sorted CD3 + T cells from 10 patients were resuspended in cold PBS at a concentration of $ 3 3 10 5 cells/mL.Approximately 100 mL of homogenous cell suspension was injected into a prepared microchip via the inlet port slowly and let stand for 5 min at room temperature, allowing the suspension to fully enter the microwells.Following cell loading, 60 mL of the barcode beads suspension was loaded onto the microchip and incubated for 1 min.After three washes with PBS, 100 mL of prepared Lysis Mix was injected into the microchip to lyse the cells, leading to the released mRNA binding to the barcode beads.Immediately following mRNA capture, barcode bead retrieval was performed with a Singleron Magnetic Rack and pre-chilled wash buffer A (supplied by the Library Kit) to obtain the barcode beads with bound mRNA.The mRNA was reverse transcribed on a thermal mixer preheated to 42 C at 1300 rpm for 90 min, followed by cDNA amplification using a T100 Thermal Cycler (Bio-Rad).A portion of the purified cDNA was subjected to scRNA-seq library preparation.The remaining cDNA was used for scTCR-seq library preparation and programmed at cDNA circularization, digestion, and purification initially.After three rounds of TCR enrichment, the immune receptor enriched library was prepared through fragmentation, adapter ligation, purification of the adapter-ligated cDNA, enriched library amplification, and purification and size selection of the amplified enriched libraries.Lastly, the prepared libraries were sequenced on an Illumina NovaSeq 6000 Sequencer.
Bulk RNA isolation and TCR sequencing DNA from CD4 + T cells was extracted using the QIAamp DNA Mini Kit (QIAGEN).To generate the template library for the T-cell repertoire, multiplex PCR amplification of CDR3 of the TCR b chain (TRB) was conducted, including PCR1 and PCR2 inclusively and semiquantitatively.Primers were designed to acquire maximum coverage of a heterogeneous set of target sequences of V and J families with a minimal PCR bias.Primer sequences were filed as part of a Chinese patent (CN105087789A).During the first round, 10 PCR cycles were used to amplify the CDR3 fragments using 32 forward primers for V genes and 13 reverse primers for J genes with a Multiplex PCR Kit (QIAGEN).Then, PCR1 was performed with 1 cycle of 95 C for 15 min, 10 cycles of denaturation at 94 C for 30 sec, and 10 cycles of annealing at 60 C for 90 sec and extension for 30 sec at 72 C.After a final extension for 5 min at 72 C, the samples were held at 4 C.The target fragment of the multiplex PCR products was purified with Agencourt magnetic beads (Beckman Coulter).In the second round, PCR2 was performed using universal primers with a KAPA HiFi HotStart ReadyMix PCR Kit (KAPA BioSystems) as follows: one cycle at 98 C for 1 min; 20 cycles of denaturation at 98 C for 20 sec, annealing at 60 C for 30 sec, and extension at 72 C for 30 sec; and a final extension at 72 C for 5 min.The samples were then held at 4 C. Final PCR products were purified again.The targeted fragments were retrieved using an E-Gel Power Snap Electrophoresis System (InvitrogenÔ).The library was assessed on an Agilent 2100 Bioanalyzer and the DNA HS kit (Agilent Technologies) to estimate fragment size distribution, while the library concentration was measured using a Qubit fluorometer and the Qubit dsDNA HS (High Sensitivity) Assay Kit (Invitrogen).Paired-end sequencing of libraries was carried out with a read length of 100 bp using the DNBSEQ-T7RS sequencer (MGI) platform.
DNA from patient-matched PBMC and tumor samples of ovarian cancer patients was extracted using the QIAamp DNA Mini Kit (QIAGEN) following the manufacturer's instructions.Multiplex PCR primers from Zhang et al. 132 were used to amplify the rearranged CDR3 regions of TCRb, including 30 forward V primers and 13 reverse J primers.The reaction mix was prepared as follows: QIAGEN Multiplex PCR Master Mix (QIAGEN), 0.5 3 Q solution, 0.2 mM VbF pool, and JbR pool were combined in a 50 mL volume.The cycling conditions were: 95 C for 15 min, followed by 25-30 cycles at 94 C for 30 sec, 60 C for 90 sec, and 72 C for 30 sec, with a final extension at 72 C for 5 min.After multiplexed PCR, single-stranded PCR products were produced via denaturation and then underwent the circularization program.The replication of single-stranded circular DNA molecules was achieved through rolling cycle amplification, resulting in the generation of a DNA nanoball (DNB) containing multiple copies of DNA.High-quality DNBs were then loaded into patterned nanoarrays using MGISEQ 2000.

Flow cytometry for Treg quantification in human tumors
To determine Treg ratios in clinical specimens, fresh tumors derived from patients enrolled in the clinical trial were cut into pieces on ice and dissociated into single-cell suspensions using a gentleMACSÔ Dissociator with Heaters and a Tumor Dissociation Kit, human (Miltenyi Biotec), as per the manufacturer's instructions.These single-cell suspensions were preincubated for $30 min with Fixable Viability Dye (BioLegend) to stain dead cells and then labeled with surface antibodies, including CD45-PE (BioLegend), CD3-PC5.5 (BioLegend), CD4-BV785 (BioLegend), CD8-APC (BioLegend), CD25-BV510 (BD Biosciences), CD127-BV605 (BD Biosciences), and CCR8-BV421 (BD Biosciences).The fluorescence intensity was determined using a Beckman Coulter Cytoflex.FlowJo was used for data analysis.

Multiplex immunohistochemistry (mIHC)
For each clinical sample, a part of fresh tumor tissue was washed with PBS, fixed with formaldehyde, and embedded in paraffin.Multiplex IHC was performed on serial tissue slides cut from these paraffine-embed samples following the manufacturer's instruction (Opal 6-plex Manual Detection Kit, NEL81100KT, Akoya Biosciences).Slides were stained with CD4 (Abcam), CD8 (Abcam), PD-1 (Cell Signaling Technology), TNFRSF9 (Cell Signaling Technology), FOXP3 (Abcam), pan-cytokeratin (Abcam), and DAPI (Thermo Fisher Scientific) in the order mentioned.All primary antibodies were incubated overnight in a cryogenic laboratory for more specific antigen-antibody binding.Then, the slides were blocked with an antifade mounting medium and scanned by PHENO IMAGER HT (Akoya Biosciences).

Cell culture and treatment
Human breast cancer cell lines (BT-549 and MDA-MB-231), ovarian cancer cell lines (OV90 and SKOV3), and HEK293T cells were purchased from ATCC.Human ovarian cancer cell line A2780 was purchased from Sigma-Aldrich and OVCAR8 was obtained from MDACC's Characterized Cell Line Core.MDA-MB-231 and HEK293T cells were cultured in DMEM (Thermo Fisher Scientific) supplemented with 10% FBS (Gibco) and penicillin-streptomycin solution (Thermo Fisher Scientific).BT-549, A2780, and OVCAR8 cells were cultured in RPMI 1640 medium (Thermo Fisher Scientific) with 10% FBS (Gibco) and penicillin-streptomycin solution (Thermo Fisher Scientific).OV90 cells were cultured in MCDB 105 medium (Sigma-Aldrich) and 199 medium (Thermo Fisher Scientific) with 10% FBS (Gibco) and penicillin-streptomycin solution (Thermo Fisher Scientific).SKOV3 cells were cultured in McCoy's 5A medium (Thermo Fisher Scientific) supplemented with 10% FBS (Gibco) and penicillin-streptomycin solution (Thermo Fisher Scientific).Mouse breast cancer cell line EO771 was purchased from ATCC and mouse ovarian cancer cell line ID8 was obtained from Sigma-Aldrich.EO771 and ID8 cells were cultured in DMEM with 10% FBS (Gibco) and penicillin-streptomycin solution (Thermo Fisher Scientific).
All cells were seeded in 12-well culture plates at a density of 1310 5 cells per well.Recombinant human IFN-a (Acro biosystems), IFN-b (PeproTech), and IFN-g (PeproTech) were added to the wells at different final concentrations ranging from 0.1 ng/mL to 20 ng/ mL to stimulate the tumor cells for 48 h.Subsequently, these treated cells were harvested and subjected to flow cytometry.
Transfection of small interfering RNA (siRNA) Control siRNA, BRCA1 siRNA, and BRCA2 siRNA were purchased from Ribo Life Science (custom service).These siRNAs were transfected into tumor cells using LipofectamineÔ 3000 (Thermo Fisher Scientific) according to the manufacturer's instructions.The sequences are as follows: Total RNA extraction and real-time PCR Total RNA from tumor cells was extracted using TRIzol reagent (Sigma-Aldrich) and reverse transcribed using a Reverse Transcription Kit (Vazyme) according to the manufacturer's protocol.Real-time PCR was performed using SYBR Green (Bio-Rad Laboratories) on a Bio-Rad CFX Connect instrument.The expressions were quantified using the comparative Ct (2 -DDCt ) method.The Ct value for each sample was normalized to that of the GAPDH gene.The primer sequences for GAPDH, BRCA1, and BRCA2 are as follows: The guide RNAs (gRNA) targeting Trp53 exon5, Brca1 exon 10, and Brca2 exon 3 were designed according to a previous publication. 133After being annealed in a thermocycler, these oligos were cloned into the LentiCRISPRv2 plasmid (Addgene).All plasmids were sequenced to confirm successful ligation and the gRNA sequences are as follows: Trp53 exon 5, TGAGCGCTGCTCCGATGGTGA and AGTGAAGCCCTCCGAGTGTC Brca1 exon 10, GACTCCTTCCCAGGACAACT and TGTCTACATTGAACTAGGTA Brca2 exon 3, TGCTTTGGCATATTATACGG and TAGGACCGATAAGCCTCAAT For virus packaging, HEK293T cells were plated on 6 cm dishes at 3,600,000 cells per dish in antibiotic-free DMEM overnight.Cells were co-transfected the following day with 6 mg of lentiviral transfer plasmid, 4.5 mg of psPAX2 (Addgene), and 3 mg of pMD2.G (Addgene) packaging plasmids using HP DNA Transfection Reagent (Roche).The culture medium was refreshed 18 h after transfection and harvested 48 h later.
For the generation of Trp53-or Brca1/2-deficient cell lines, EO771 and ID8 cells were transfected with lentiviruses using polybrene (2.5 mg/mL), selected under puromycin (2.5 mg/mL) for 72 h, and sorted into single wells for the single-cell clonal expansion.The expanded single-cell colonies were used for mouse model generation after verifications with DNA sequencing.
To validate the status of mutation in these engineered cells, total DNA was extracted using the QIAamp DNA Mini Kit (QIAGEN) according to the manufacturer's instructions.PCR was performed using PlatinumÔ SuperFi II DNA (Thermo Fisher Scientific) to amplify the DNA strands spanning potential sites of deletion.These PCR products were cloned into Blunt Simple Cloning Vector (TRANS) and subjected to Sanger sequencing to select clones with large fragment deletion or frameshift mutations in all alleles.PCR primer sequences are as follows:

Mouse models Mouse ovarian cancer model
To generate mouse orthotopic ovarian cancer models, mouse Trp53 and Brca1 double genetic knockout (KO) ID8 cells (3310 6 / mouse) were resuspended in 30 mL PBS mixed with Matrigel at a ratio of 1:1 and transplanted into hCCR8 or WT C57BL/6 mice via intrabursal injection.After about 2 weeks of transplantation, the two types of tumor-bearing mice were randomly divided into 4 treatment cohorts, respectively: control, niraparib, anti-CCR8 or anti-CD25 monoclonal antibody (mAb), and niraparib combined with anti-CCR8 or anti-CD25 mAb.hCCR8 C57BL/6 mice received niraparib (40 mg/kg, oral administration) 5 times a week and anti-CCR8 mAb (3 mg/kg, intraperitoneal administration) twice a week.WT C57BL/6 mice received niraparib (40 mg/kg, oral administration) 5 times a week and anti-CD25 mAb (200 mg/mouse, intraperitoneal administration) once a week.The tumor volumes were monitored every week and calculated as luminescence intensity per sec using an IVISâ Lumina III In Vivo Imaging System (PerkinElmer).After 4 or 5 weeks of treatment, the presence of ascites and tumor weights were recorded, and the immune status in tumors was accessed using flow cytometry and IHC staining.

Mouse breast cancer model
To generate mouse orthotopic breast cancer models, mouse Brca2 KO EO771 cells (5310 5 /mouse) were resuspended in 100 mL PBS mixed with Matrigel at a ratio of 1:1 and transplanted into 4 th inguinal mammary fat pad of hCCR8 C57BL/6 mice.Mouse Brca1 KO EO771 cells (5310 5 /mouse) were also transplanted into the 4 th inguinal mammary fat pad of WT C57BL/6 mice.The tumor volumes were measured every day or every other day and calculated as L 3 W 3 W 31/2, where L is length and W is width.Once the tumor volumes reached $50 mm 3 , the tumor-bearing hCCR8 mice were treated with niraparib (40 mg/kg, oral administration) 5 times a week and anti-CCR8 mAb (3 mg/kg, intraperitoneal administration) twice a week, Tumor-bearing WT mice were treated with niraparib (40 mg/kg, oral administration) 5 times a week and anti-CD25 mAb (200 mg/mouse, intraperitoneal administration) once a week.The EO771 tumors were harvested, and their immune status was evaluated using flow cytometry and IHC staining.

Dual-target immunofluorescence for Treg quantification in mouse tumors
Treg density in mouse tumors was evaluated in FFPE tumor slide specimens using dual-target immunofluorescence.Briefly, antigen retrieval, nonspecific antigen blocking, and incubation of primary and fluorescence-labeled secondary antibodies and DAPI were sequentially performed.The following antibody combinations were used: anti-mouse PD-1 (Cell Signal Technology) and anti-human/mouse FOXP3 (Abcam).Whole slide scans were acquired using the Pannoramic SCAN (3DHISTECH) digital pathology scanning system.

HRD and mutation analysis
After mapping sequenced reads to the human reference genome (GRCh37), we first calculated the GIS using an in-house algorithm. 131This algorithm estimated three genomic features, loss of heterozygosity, telomeric allelic imbalance, and large-scale transition.The GIS was the unweighted sum of the three values.We then called germline and somatic mutations using GATK HaplotypeCaller and Mutect2, 108 respectively.Non-silent somatic mutations (or somatic and germline mutations for BRCA1/2 and TP53) with a prevalence of < 0.01 in cohorts of 1000 Genomes, ExAC, or gnomAD resources are summarized in Table S2 and were visualized using oncoplot. 113Mutations were further manually curated to be pathogenic or likely pathogenic according to the recommendations of the American College of Medical Genetics and Genomics (ACMG) 134 and the AMP/ASCO/CAP Somatic Variants Guideline 135 in the interpretation of sequence variations.Finally, the HRD status of a sample was defined as positive if the GIS was R 42 or if there were pathogenic/likely-pathogenic BRCA1/2 germline or somatic mutations.
scRNA-seq data preprocessing and quality control FASTQ files were processed using kallisto-bustools 112 to obtain gene-level UMI counts for each droplet, with the GENCODE v37 human genome as the reference.The kallisto-bustools workflow was set to ''standard'' so that only exonic reads would be counted towards the final UMI count.We then devised a biology-aware multi-layer workflow to process and quality-control these raw count matrices to ensure a robust balance between the removal of low-quality droplets and the preservation of small-RNA-content cells.This was an essential step tailored to scRNA-seq data generated from tumor samples where cell sizes of different populations can be drastically different, and a universal hard-set threshold would thus not be suitable.
First, we focused on assessing the overall data quality of each sample and leniently removing empty droplets by visualizing a panel of key droplet-level statistics.EmptyDrops in the DropletUtils 104 package was used to calculate the probability of each droplet being an empty droplet.The inflection point was used as a cutoff below which all droplets were deemed empty and would form an ambient pool for the remaining droplets to be compared against.Then, plots depicting multiple key metrics and the interactions thereof were inspected.These include (i) a scatter plot showing the relationship between UMI count and gene count in each droplet to evaluate the level of sequencing saturation and detect clear outliers; (ii) a knee plot showing the log rank of droplets by UMI count to judge the level of separation between empty droplets and cell-containing droplets; (iii) a knee plot showing the log rank of droplets by UMI count, the same as above; (iv) a density plot showing the distribution of mitochondrial DNA (mtDNA) UMI count proportion to help assess the magnitude of mitochondrial contamination; (v) a scatter plot showing the relationship between gene count and mtDNA UMI count proportion to detect the segregation between a low-mtDNA high-transcriptome-diversity (i.e., with a high gene count) cluster and a high-mtDNA low-transcriptome-diversity cluster; (vi) a scatter plot showing the relationship between UMI count and -log(empty drop probability) to visualize the estimation by EmptyDrops of the probability of each droplet being a real cell-containing data point.The samples that passed these criteria were retained, and within each sample, droplets that passed the EmptyDrops FDR threshold of 1% were retained.
Second, we took into consideration the clustering information that would inform the collectively abnormal behavior of a group of cells to further perform quality control on a reduced-dimension space.Scrublet 123 was employed to calculate a doublet score for each retained cell.We then followed the downstream analysis workflow implemented in Scanpy. 120Specifically, we (i) applied the log1pCP10K normalization to the raw counts, (ii) selected highly variable genes using a dispersion-based method (the maximum and minimum average expression of a gene were set to be 3 and 0.0125, respectively, and the lower expression dispersion cutoff was set to 0.5) or through Triku, 128 (iii) regressed out the effects of the total count per cell and the percentage of mitochondrial gene count, (iv) scaled the normalized and regressed counts so that the data had unit variance and zero mean (any values above 10 were clipped), (v) calculated the first 50 principal components, (vi) reduced the data dimension through UMAP, and (vii) clustered the single cells using an unsupervised graph-based clustering algorithm, Leiden.By overlaying the UMI count, mitochondrial and ribosomal UMI proportions, doublet score, and EmptyDrops FDR values on top of the UMAP space, we identified clusters enriched for low-quality cells to discard.For example, a cluster where most cells had a low UMI count, high mitochondrial UMI fraction, and high empty droplet probability would most likely be apoptotic cancer cells and were filtered out.Importantly, this observation is usually not readily noticeable in the first layer of QC.
Third, we relied on prior biological knowledge to further investigate any aberrant entities in the dataset.Specifically, we built upon the UMAP space in the second layer of QC, identified cluster-specific marker genes using Student's t-test, and annotated the clusters based on the expression patterns of literature-derived marker genes along with the annotations from the original studies.This step helped us understand the presence/absence of biological identities that are common across cancer types or specific to ovarian cancer and thus served as a further control for data quality.
Notably, these three QC phases were not conducted in a strictly sequential manner.Instead, every piece of new information manifested in each stage could facilitate the clarification of unsolved situations in the other stages.For example, we may see a small cluster in stage 2 that shows a significantly lower UMI count than the rest of the dataset and mark it as a potential ambient droplet cluster.However, when QC stage 3 reveals that this cluster is annotated as T cells and that this sample is highly enriched in tumor cells, we would then discard the mark and instead choose to retain these cells.This is because we would then understand that this cluster being an outlier in terms of low UMI count reflects an authentic RNA content difference between immune cells and cancer cells.

scRNA-seq data integration and annotation
After the preprocessing and QC of each sample as described above, we started over with the count matrices and followed the same workflow in QC stages 2 and 3. Additionally, we applied Harmony 109 to remove sample-level batch effects after calculating the principal components and before conducting UMAP-based dimension reduction.We evaluated the effectiveness of data integration by (i) a qualitative visual inspection of the UMAP plots before and after integration to confirm that the cells were clustered by cell type instead of the sample source; (ii) a quantitative assessment of batch effect based on a sample-level silhouette coefficient across cell types where a lower value after integration indicated a reduction in sample-driven cell clustering.
For cell type annotation, we first identified 10 major cell types based on the expression patterns of well-established marker genes.These include MS4A1 and CD79A for B cells, EPCAM and KRT19 for epithelial cells, PLVAP and PECAM1 for endothelial cells, COL1A1 and BGN for fibroblasts, TPSB2 and CPA3 for mast cells, AIF1 and LST1 for myeloid cells, MZB1 and DERL3 for plasma cells, RGS5 and COL18A1 for pericytes, CD2 and CD3D for T cells and NK cells, and IRF7 and LILRA4 for plasmacytoid dendritic cells.Next, for each major cell type, we concatenated and reprocessed the count matrices using the same workflow and parameters as described above.We then identified the subordinate cell states through an extensive search of the literature and combined a set of complementary approaches to assign these states to each subpopulation (Figure S3), including (i) examining the expression distribution of known gene markers across cell clusters.For example, the area within the CD4 + T cell space that is positive for both FOXP3 and TNFRSF18 would be annotated as eTregs.(ii) Calculating for every single cell a score of a specific gene signature that was derived from the literature to mark a relevant cell state.For example, we annotated a cluster of CD8 + T cells as effector exhausted T cells because of their high scoring of the ''CD8 dysfunction'' signature from a study by Li et al. 136 We used UCell 129 for signature scoring.All public cell state gene signatures used to score single cells are listed in Table S7.(iii) Projecting the cells onto reference atlases that represent a comprehensive coverage of known cell states within a major cell type and have high-quality annotation information.For example, we identified mesothelial antigen-presenting cancer-associated fibroblasts based on a projected identity of these cells being ''CALB2 + FIB'' in a reference by Qian et al. 91 Symphony 127 was used for reference construction and application.All public scRNA-seq datasets used to build reference atlases are cataloged in Table S7.
Because cancer cells uniquely have their cell states largely driven by underlying patient-specific somatic alterations, we employed a different strategy to first distinguish malignant cells from normal epithelial cells and then their states.We first applied inferCNV 137 to compute a single-cell copy number variation (CNV) profile for all epithelial cells within each sample.Then, we used the CNV profiles rather than the original gene expression profiles as the basis for dimension reduction and clustering.We found a significant agreement between high-CNV-based cancer cell assignment and reference-projection-based secretory epithelial cell characterization (known to be the cells of origin for most ovarian cancers 138 ), enabling a robust separation between normal and malignant epithelial cells.Finally, we used consensus non-negative matrix factorization (cNMF) 99 to identify recurrent cancer cell states across patient samples.This analysis led to a three-state annotation of cancer cells into an IFN-responding cluster, a proliferative cluster, and a non-proliferative cluster.

scRNA-seq data analysis of T cells in an external cohort
For independent validation, we downloaded single-cell gene expression profiles of ovarian cancer samples from a recent study 30 as UMI counts from the Gene Expression Omnibus (GEO) under the accession number GSE165897.This published cohort contains 11 HGSOC patients who received NACT and had paired pre-and post-treatment samples collected for scRNA-seq.No QC was further applied beyond what was already conducted by the original study.We first extracted T cells and then employed the same processing workflow for this dataset as for our in-house dataset.To ensure a direct comparative analysis between the two datasets, we used the same set of marker genes, signatures, and reference atlases to annotate the T cell states in this cohort under the same nomenclature scheme.We further interrogated the all-cluster-by-all-cluster overlap of cell state marker genes to demonstrate a good concordance between the two annotated T cell spaces.
Comparative and correlative analysis of cell state proportions Annotated cell states were organized in a hierarchy supported by the literature.Differential abundance testing and correlation testing were performed on the proportion of a specific cell state (the numerator) among its immediate preceding cell state or cell type (the denominator).For example, the number of CD4 + T cells served as the denominator when testing for Treg abundance shift, while the number of Tregs was the denominator when testing for terminal eTreg abundance shift.To avoid spurious testing results, we removed samples in which the cell count of a cell state serving as the denominator was < 50.

Graph-based differential abundance testing
To detect cell-state abundance changes between different conditions in an annotation-free manner, in addition to the direct comparison based on population fractions, we performed differential abundance testing analysis using the milopy package. 47Specifically, cell neighborhoods were first defined on a k-nearest neighbor (kNN) graph.Differential abundance testing was then performed for each neighborhood using a negative binomial general linear model framework.

Inference of eTreg infiltration in bulk RNA-seq data
To obtain an eTreg signature, we combined the top 10 differentially expressed genes in eTregs derived from our scRNA-seq data.To account for the infiltration level of total CD4 + T cells in a bulk tumor, we further calculated the ratio of the geometric FPKM average of the eTreg signature over the FPKM of the CD4 gene.As FPKM values are of a linear scale, such ratios served as a proxy of the relative abundance of eTregs to all CD4 + T cells in a bulk tumor profiled by RNA-seq, allowing us to compare it between HRD and HRP HGSOC tumors in the Cancer Genome Atlas (TCGA). 139

RNA velocity analysis
For scRNA-seq data generated through the 10x platform, spliced and unspliced UMI counts for each gene in each cell were counted using kallisto-bustools with the workflow set to ''nucleus'' and using an intron-containing index.For scRNA-seq data generated through the Singleron platform (for paired scRNA/TCR-seq), the same count tables were obtained using Velocyto 130 with intermediate bam files as the input.Subsequent analyses were performed using Scanpy and scVelo. 124Specifically, library size normalization and log transformation were applied to the count matrices.The top 2,000 genes with the highest expression dispersion were then selected for subsequent velocity computation.Next, for each cell, the moments (means and uncentered variances) of normalized spliced/unspliced counts were computed using the 30 nearest neighbors.RNA velocity estimation was then conducted with the mode set to ''deterministic.''We then used the gene expression graph used for clustering and annotation to embed RNA velocities in the UMAP space in the form of streamlines to facilitate a direct interpretation of velocities on top of the original annotated space.
Gene regulatory network analysis Activated regulons in each subpopulation of Treg and Tex cells were analyzed using SCENIC 46 with a raw count matrix as input.Briefly, the co-expression network was calculated by GRNBoost2, and the regulons were identified by RcisTarget.Next, the regulon activity for each cell was scored by AUCell.The differentially activated regulons in each Treg or Tex subset were identified by a Wilcoxon test using cells from the other subsets as control.The Benjamini-Hochberg test was used to correct multiple hypotheses.
Paired scRNA/TCR-seq data analysis For scRNA-seq data, FASTQ files were processed using the CeleScope (https://github.com/singleron-RD/CeleScope)''rna'' workflow, designed for transcriptome sequencing data generated through the Singleron platform, to obtain gene-level UMI counts for each droplet, with GENCODE v37 human genome as the reference.Briefly, after using fastqc (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and cutadapt 140 to QC and filter the data, reads were compared with the reference genome using STAR. 141Finally, featureCounts 142 was used to generate the gene count matrices.The same preprocessing, QC, and downstream analysis workflows for the main cohort were applied again.For paired same-cell scTCR-seq data, FASTQ files were processed using CeleScope ''flv_CR'' workflow, designed for full-length VDJ enrichment sequencing data generated through the Singleron platform, to obtain TCR ⍺and b-chain CDR3 nucleotide sequences.Only the assembled chains that were productive, high confidence, full length, and had a valid cell barcode and an unambiguous chain type assignment were retained.If a cell had > 2 qualified chains of the same type, only the chain with the highest UMI count was qualified and retained.Single-cell CDR3 sequences were used for downstream analysis with Scirpy. 122Clonotypes were determined based on identical CDR3 nucleotide sequences for both arms.Expanded clones were determined by a clonal size of > 1.The magnitude of TCR repertoire overlap between different T cell subpopulations was calculated using the Jaccard index.
TCR clonotype reconstruction from non-VDJ-enriched scRNA-seq data TCR CDR3 sequences were extracted from the scRNA-seq FASTQ files using TRUST4, 39 and sample cells with at least one productive chain (either ⍺or b-chain) were retained.Samples with > 25 cells matching this criterion were retained for subsequent analysis with Scirpy.Due to data sparsity of the inferred CDR3 profiles, we employed a relaxed definition of TCR clone by only requiring either of the two chains to have identical CDR3 nucleotide sequences.We called expanded TCR clones based on them containing at least two cells.TCR clonotype overlap was measured by the Jaccard index.The quality of CDR3 inference by TRUST4 was evaluated by examining (i) the agreement between the enrichment of T cell gene marker expressions and that of TCR UMI count, (ii) the proportion of expanded clones that only contained cells from the same patient, thus meeting the standard of being a ''private'' clone, and (iii) the concordance between the relative expansion magnitudes across different T cell subpopulations based on the inferred CDR3 profiles and those based on VDJ-enriched bona fide CDR3 profiles.
Bulk TCR-seq data processing and analysis Sequences were processed and analyzed using MiXCR. 115PBMC bulk TCR-seq clonotypes were associated with patient-matched intratumoral single-cell TCR clonotypes by matching their TCR b-chains based on identical nucleotide sequences.

Pseudobulk analysis
To suppress false positive discoveries in differential expression analyses using single cells as data points for statistical testing, we employed a pseudo-bulk alternative. 143Briefly, for cells of a specific combination of a cell state and an HRD status/treatment group, we first aggregated reads across biological replicates, transforming a genes-by-cells matrix to a genes-by-replicates matrix using matrix multiplication.Then, we ran DESeq2, 102 which used a Wald test of the negative binomial model coefficients to compute the statistical significance.
mIHC Image analysis For each slide, 8 fields of view (FoVs) were obtained, and cell segmentation was performed using the adaptive cell segmentation setting in image analysis software (InForm v.2.6.0,Akoya Biosciences/PerkinElmer).Then, the cytoplasmic (CD4, CD8, PD-1, TNFRSF9, panCK) and nuclear (FOXP3) optical signals in the spectral unmixing image were isolated and quantified.The data was consolidated with the phenoptr v.0.3.2 package for R software and was analyzed with reference to described analysis in a previous study. 144Specifically, thresholds for both the mean pixel intensity and the total pixel intensity for all channels were manually thresholded in each slide and evaluated by 4 independent researchers.The specificity and sensitivity of the positive threshold were further adjusted carefully in all FoVs using a visualization strategy where the positive cells of each marker are mapped onto the spectral unmixing image of the corresponding channel.Each cell phenotype was evaluated as the average percentage of CD4 + or CD8 + T cells.The inter-cellular distance between nearest neighbors was calculated using the phenoptr package.Briefly, for each cell in a FoV, the distance matrix was used to find the nearest neighbor cell in each of the provided phenotypes, and the cell ID and distance to the nearest neighbor cell were reported.After that, the median distance from the panCK + cells to the target cells was computed among all cells of the desired marker combination.

QUANTIFICATION AND STATISTICAL ANALYSIS
All details related to statistical analysis can be found in figure legends and in the corresponding methods.No statistical methods were used to predetermine sample sizes.Data met the assumptions of the statistical tests used and were tested for normality.Statistical significance was set at p < 0.05.

Figure 1 .
Figure 1.Study design and dataset overview (A) Flowchart of patient enrollment and treatment.(B) Percentage change in GCIG CA125 level from baseline.Dashed black line represents a À50% change in GCIG CA125 level from baseline, indicating a GCIG CA125 response.A CA125 value below the reference level further indicates CR.Otherwise, PR is documented.(C) Patient response according to GCIG CA125 response criteria stratified by HRD status and treatment cohort.(D) Cross-site sampling of tumor and blood samples for single-cell and bulk multi-omic profiling from ovarian cancer patients.Numbers in parentheses indicate patient count and sample count, respectively.(E) Genomic alterations of a panel of homologous recombination repair (HRR) genes across 34 patients in the main cohort.Top, genomic instability scores (GISs); dotted black line indicates a GIS threshold of 42 for HRD approved by the US Food and Drug Administration.Right, sample count with a mutated gene of interest.Germline mutations are highlighted with the ''+'' sign.Silent mutations with clear functional consequences are highlighted with the ''*'' sign.(F) Uniform manifold approximation and projection (UMAP) embedding of all cells in the main cohort with major cell types annotated.(G) Major cell type composition and clinical metadata for each sample.FIGO, The International Federation of Gynecology and Obstetrics; HRD, homologous recombination deficiency; HRP, homologous recombination proficiency; BRCAm, mutated BRCA1/2; BRCAwt, wild-type BRCA1/2; GIS, genomic instability score; ISD, individualized starting dose; TC, paclitaxel and carboplatin; CR, complete response; PR, partial response; NR, non-response; RR, response rate; GCIG, the Gynecologic Cancer InterGroup; CA125, cancer antigen 125.See also FigureS1and TablesS1 and S2.

Figure 2 .
Figure 2. Tregs are enriched in pre-treatment HRD tumors and suppressed by neoadjuvant treatment (A and B) Cell-state annotations of CD4 + (A) and CD8 + (B) T lymphocytes.Zoom-in boxes indicate marker genes, signatures, or reference projections.(C) Differential abundance testing of 97 cell states across 10 major cell types.Left, dendrogram showing cell-state transcriptional similarities.Middle, scaled log 2 FC of cell-state abundance in each comparison (two-sided Mann-Whitney U test for the HRD status comparison; two-sided paired Wilcoxon rank-sum test for the treatment comparisons).Differences with a raw p > 0.05 are masked in the heatmap.Right, cell count of each state.Four cell states with significant abundance changes across comparisons are highlighted.
Texs among tumor-reactive T cell populations coordinated by HRD Our systematic analysis across nearly 100 cell states revealed that three out of the final four top candidates were T cell (D) Correlation of cell state log 2 FC between niraparib therapy and chemotherapy for HRD samples only.(E) Overlap between top hits of the three differential abundance tests.(F) Difference in eTreg proportion among all CD4 + T cells between pre-and post-chemotherapy tumors based on data from GSE165897 (n = 11 patients).(G) Difference in log 2 ratio of the geometric FPKM average of an eTreg signature to CD4 FPKM between HRP and HRD ovarian cancer samples from TCGA (n = 61 HRP samples, n = 293 HRD samples).(H) Log 2 FC of either expression levels or expression level ratios in FPKM across T cell marker genes.Entries with a statistically significant difference are marked by an asterisk sign.(I) Representative fields of view for pseudocolor (top), cell identity (middle), or the proximity of PanCK + -CD4 + FOXP3 + phenotype pairs (bottom) for one HRP tissue and one HRD tissue.Scale bars, 200 mm.(J) Difference in proportion of FOXP3 + , FOXP3 + TNFRSF9 + , and FOXP3 + PD-1 + cells among CD4 + T cells between HRP and HRD samples (left, n = 18 HRP samples, n = 34 HRD samples) or between pre-nira and post-nira samples (right, n = 21 pre-nira samples, n = 15 post-nira samples) based on mIHC measurements.(K) Difference in distribution of panCK + -FOXP3 + TNFRSF9 + (left) or panCK + -FOXP3 + PD-1 + (right) nearest-neighbor distance between HRP and HRD samples.Vertical lines indicate the median nearest-neighbor distance.(L) Representative flow cytometry plots of a CD25 + CD127 low population in all CD4 + T cells and a CCR8 + population in all Tregs.

Figure 3 .
Figure 3. HRD drives co-occurring hyper-expanded tumor-reactive T cell populations 1 + cells.Scale bars, 200 mm.(C) UMAP embeddings of CD4 + (left) and CD8 + (right) T cells showing tumor-reactivity score.Cell states with uniquely high scores are labeled.(D) Average gene expression levels of state marker genes across CD4 + (left) and CD8 + (right) T cell states.The circle size denotes the proportion of cells within a cell state where a gene expression is non-zero.(E) UMAP embeddings of CD4 + (left) and CD8 + (right) T cells with colors denoting cell-state annotations (top) or color grade representing log-transformed TCR clone size (bottom).(F) Expansion magnitude of CD4 + (top) and CD8 + (bottom) T cell states across 10 samples.(G) Correlation between the proportion of expanded Tregs and that of expanded Texs across 10 samples.

Figure 4 .
Figure 4.The dynamics of fine-grained Treg and Tex subsets under HRD and neoadjuvant treatment (A) UMAP embedding of Tregs with colors indicating cell-state annotations.(B) UMAP embedding of Tregs of the same color annotations in (A) overlayed with streamlines denoting RNA velocity.(C) UMAP embeddings of Tregs showing gene expression patterns of cell-state markers.(D) Difference in gene signature scores across Treg cell states.(E) SCENIC results showing the activities of transcription factor regulons in each cell.Colored bars on top of the heatmap indicate cell-state annotations.The names of cell-state-specific representative transcription factors are highlighted.(F) UMAP embeddings of Tregs with color grade representing TCR clone size and streamlines indicating RNA velocity.(G) UMAP embeddings of Tregs showing expression patterns of cell-state gene markers.(H-N) Same as (A)-(G) but for Texs.(O) UMAP embedding of Tregs with Milo differential abundance testing results overlayed in color grade.Nodes are neighborhoods, colored by the abundance log 2 FC between HRD and HRP samples (left) or between pre-and post-treatment samples under chemotherapy (middle) or niraparib therapy (right).Node sizes correspond to the number of cells in each neighborhood.Areas with strong enrichment signals are highlighted with a black dashed circle.FC, fold change.(P) Same as (O) but for Texs.See also TableS7.

Figure 5 .
Figure 5. Tracing Treg/Tex origins in the TME and the circulation (A) TCR repertoire overlaps between CD4 + (top) and CD8 + (bottom) T cell states.(B) Fraction of Treg clonotypes found in non-Treg cells (top) or of Tex clonotypes found in non-Tex cells (bottom) across samples.(C and D) Number of TCR clonotypes shared between matched PBMC and tumor samples for pre-(C) or post-niraparib (D) patients based on bulk TCR-seq.(E) Difference in blood and tumor repertoire overlap between pre-and post-niraparib samples (n = 12 patients, two-sided paired Wilcoxon rank-sum test, ns, nonsignificant).(F and G) TCR repertoire co-expansion between PBMC and patient-matched tumor across pre-(F) and post-niraparib (G) samples.(H) Number of CD4 + TCR clonotypes shared between matched PBMC and tumor samples for treatment-naive patients based on bulk TCR-seq and scTCR-seq, respectively.(I) Fraction of intratumoral TCR clonotypes observed in sample-matched PBMC across CD4 + T cell states.See also TablesS5 and S6.

Figure 6 .
Figure 6.IFN-responding cancer cells contribute to Treg induction through MHC class II (A) Ranking of cell types and cell states by abundance co-variation with terminal eTregs (among all Tregs) across all samples.(B) Abundance co-variation between top IFN-responding cell states and all Treg cell states (among all Tregs).(C) Abundance co-variation between IFN-responding cancer cells (among all cancer cells) and terminal eTregs (among all Tregs) (n = 48 samples).(D-F) Same as (A)-(C) but for terminal Texs (F, n = 38 samples).(G) Difference in cancer cell-state proportion between pre-treatment HRD and HRP samples (n = 12 HRP samples, n = 25 HRD samples).For boxplots, the middle line in the box is the mean, the bottom and top of the box are the first and third quartiles, and the whiskers extend to the 1.53 interquartile range of the lower and the upper quartiles, respectively.(H) Average expression levels of MHC class I and MHC class II genes (left) or of primary and secondary co-stimulatory T cell ligand genes (right) across cancer (top) and DC (bottom) cell states.Circle size denotes the proportion of cells within a cell state where a gene expression is non-zero.Labels in parentheses are common protein names.(I) Pseudobulk (sample-average) expression levels of MHC class II genes in proliferative (left) and non-proliferative cancer cells (right).Heatmaps on the top show the HRD status of each pre-treatment sample.Colored dots on the right indicate pseudobulk DESeq2 results where the color denotes log 2 fold change, and the size denotes statistical significance.(J) Change of log-transformed MHC class II MFI in IFN-treated (left) or BRCA1/2-deficient (right) cells (n = 3 replicates).(K)Difference in pseudobulk gene expression Z scores of co-inhibitory molecules among cancer cell states (n = 67 samples).Lines connect data points of the same sample.Labels in parentheses are common protein names.DC, dendritic cell; MFI, median fluorescence intensity; R s , Spearman rank correlation coefficient.p values were calculated using a two-sided unpaired t test for (G) and (J) and a two-sided paired t test for (K).ns, non-significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.See also FigureS6.
legend on next page)

Figure S2 .
Figure S2.Differential abundance testing of major cell types and minor cell states, related to Figure2

Figure
Figure S4.T cell-focused annotation of an independent scRNA-seq dataset, related to Figure 2

(
A and B) UMAP embeddings of CD4 + (A) and CD8 + (B) T cells extracted from a public dataset (GSE165897) colored by T cell state.(C and D) Overlap of cell state-specific marker genes between GSE165897 and the in-house dataset for CD4 + (C) and CD8 + (D) T cell states, measured by Jaccard index.

Figure S5 .
Figure S5.Quality control of TCR clonotypes inferred from scRNA-seq data, related to Figure 3 (A) Number of TCR clonotypes with a full pair of ⍺and b-chains, only a b-chain, or only an ⍺-chain.(B) UMAP embeddings of all cells overlaid with color grades in log-transformed TCR UMI count (left), CD3D expression level (middle), and TYROBP expression level (right).(C and D) Sample composition of each CD4 + (C) or CD8 + (D) TCR clonotype.The circle size denotes the number of cells within each clone.(E and F) Fraction of expanded clones in each CD4 + (E) or CD8 + (F) T cell state across samples.See also TableS6.

Figure S6 .
Figure S6.IFN-responding cancer cells contribute to Treg induction through MHC class II, related to Figure 6

(
A) Ranking of cell types and cell states by abundance correlation with eTregs (out of all CD4 + T cells) across all samples.(B) Abundance correlation between IFN-responding cancer cells (out of all cancer cells) and eTregs (out of all CD4 + T cells).R s , Spearman rank correlation coefficient.(C and D) Same as (A) and (B) but for Texs out of all CD8 + T cells.R s , Spearman rank correlation coefficient.(E) Abundance correlation between top IFN-responding cell states and CD4 + T cell states (each out of all CD4 + T cells).R s , Spearman rank correlation coefficient.(F) Same as (E) but for CD8 + T cells.(G) Average expression levels of MHC class I and MHC class II genes (left) or of primary and secondary co-stimulatory T cell ligand genes (right) across MoMac cell states.The circle size denotes the proportion of cells within a cell state where a gene expression is non-zero.Labels in parentheses are common protein names.(H) Pseudobulk (sample-average) expression levels of MHC class I genes in proliferative (left) and non-proliferative cancer cells (right).Heatmaps on the top show the HRD status of each pre-treatment sample.

Figure S7 .
Figure S7.Niraparib combined with CCR8 mAb or CD25 mAb caused no serious side effects in mice, related to Figure7

(
A) Difference in level of hemoglobin, lymphocytes, platelets, erythrocytes, and leukocytes in the blood of hCCR8 mice treated with control, niraparib, CCR8 mAb, or their combination.(B) Same as (A) but for WT mice treated with control, niraparib, CD25 mAb, or their combination.(C) Difference in level of ALT, AST, BUN, CREA, CK, and LDH in the serum of hCCR8 mice treated with control, niraparib, CCR8 mAb, or their combination.(D) Same as (C) but for WT mice treated with control, niraparib, CD25 mAb, or their combination.ALT, alanine aminotransferase; AST, aspartate aminotransferase; BUN, blood urea nitrogen; CREA, creatinine; CK, creatine kinase; LDH, lactate dehydrogenase.