Tumor-intrinsic IFNγ signaling and niche adaptation drive early colonization in ovarian cancer metastasis

Abstract

The mechanisms enabling disseminating tumor cells to survive and colonize peritoneal niches during ovarian cancer metastasis remain incompletely understood. Here, we present MetTag, a single-cell barcoding and transcriptome profiling approach with time-stamped batch identifiers to resolve temporal dynamics of dissemination. Sequencing of MetTag barcodes reveals enrichment of early-disseminated clones across metastatic sites, and targeted depletion of pioneers diminishes the outgrowth of later arriving cells. MetTag-coupled single-cell RNA sequencing uncovers a distinct interferon-gamma-centric transcriptional trajectory enriched among pioneer clones. CRISPR/Cas9 screening and subsequent interferon-gamma receptor 1 knockout in pioneer cells significantly reduces metastatic burden, highlighting a critical window where active interferon-gamma signaling shapes the post-seeding metastatic niche and outgrowth. Mechanistically, the tumor-intrinsic interferon-gamma response and peritoneal macrophages cooperatively shield disseminating tumor cells from anoikis. Our study defines the temporal clonal architecture of peritoneal metastasis, revealing a first come, first served dissemination principle, where pioneer fitness determines the success of subsequent colonizers.

The mechanisms enabling disseminating tumour cells (DTCs) to survive and colonise peritoneal niches in ovarian cancer metastasis remain poorly understood. Using single-cell barcoding, transcriptome profiling, and in vivo CRISPR screening, the authors identify a niche-adaptive IFNγ response signature that supports pioneer DTC outgrowth.

Introduction

Cancer metastasis encompasses continuous metastatic dissemination from the primary tumor, accompanied by passive clonal selection, and active epigenetic adaptation in response to the metastatic niche¹. Resolving the clonal, temporal, and microenvironmental determinants of this process remains a significant challenge in the field. Unlike many cancer types where hematogenous metastasis predominates, high-grade serous ovarian carcinoma (HGSOC) exhibits a unique pattern of transcoelomic, local spread within the peritoneal cavity². In HGSOC, malignant ascites functions as a dynamic reservoir of free-floating cancer cells and cell clusters – disseminating tumor cells (DTCs) detached from the primary tumor – initially seeding locally and colonizing the greater omentum and other peritoneal organs³. Different from immunologically cold tumor microenvironments, malignant ascites is heavily enriched with immune cells and acellular factors known to influence tumor progression. Consequently, ovarian cancer DTCs - continuously shed from the primary tumor - must overcome immune surveillance and adapt to the unique intraperitoneal microenvironment to establish colonization^4–8. In contrast to the pre-metastatic niche (PMN), which describes systemic conditioning of distant sites by the primary tumor prior to DTC arrival^9,10, pioneer and subsequent DTC colonization actively remodel the metastatic niche, forming a progressively evolving post-seeding metastatic niche (PSMN) that influences the ultimate outcome of metastasis. Clinical studies in multiple tumor types, including HGSOC, have corroborated the presence of DTCs in bone marrow and distant organs of patients well before reaching advanced (stage III & IV) tumor stages^11–17. Despite early metastatic dissemination observations, the underlying molecular mechanisms that modulate reciprocal interplay between DTC seeding waves and PSMN evolution driving metastasis outgrowth remain largely unknown.

In this work, we employ clonal tracing, single-cell RNA sequencing (scRNA-seq), in vivo CRISPR screening, and Cre-based niche-tracing analysis to dissect the complex interplay between continuously arriving DTCs and ovarian cancer PSMN. We identify a critical pioneer DTC dependence on IFNγ response signature and enrichment of early pioneer DTCs during outgrowth. These IFNγ-responsive pioneers dictate the success of later arriving DTCs. We further demonstrate that DTC-PSMN co-conditioning favors the enrichment of a macrophage population capable of protecting DTCs from anoikis-mediated death via paracrine signaling. Together, our data reveal a first- come, first-served principle where pioneer DTC clones dictate the fate of later clones within a relatively short and defined temporal window.

Results

MetTag captures spatiotemporal clonal dissemination patterns in ovarian metastasis

To capture the transcriptional and clonal dynamics of DTCs, we employed an RNA-level barcoding system based on the LARRY (Lineage and RNA RecoverY) barcoding construct, which enables single-cell transcriptome-level lineage tracing¹⁸. Critically, we added a unique batch identifier BC.ID as a time stamp for distinguishing time points (TPs) representing early or later-arriving DTCs. Within the U6-driven barcode sequence, a Chromium 10x capture sequence 1 (CS1) allows for simultaneous capture of the LARRY barcode transcript with mRNA transcripts (Fig. 1a and Supplementary Fig. 1a). LARRY barcode diversity in each resulting MetTag CROP.seq LARRY-BC.ID library batch, hereafter referred to as MetTag-LARRY-GFP, was comparable to the original LARRY library (Supplementary Fig. 1b–d). Next, we generated six unique isogenic ID8 p53^-/- murine ovarian cancer pools with MetTag-LARRY-GFP lentiviral libraries harboring unique batch identifiers from BC.ID1 to BC.ID6 (Fig. 1a) and verified that the six lines had comparable in vitro proliferation rates (Supplementary Fig. 1e, f).

Fig. 1. MetTag captures spatiotemporal clonal dissemination in ovarian cancer metastasis.

a Schematic depicting MetTag CROP.seq LARRY-BC.ID/MetTag-LARRY-GFP barcoding construct. b Staggered injection and tissue collection timeline for clonal tracing using barcoded ID8 p53^-/- cells, created in BioRender. Aleksandrovic, E. (2026) https://BioRender.com/qxshldk.) DNA-level barcode distributions in omenta under WT (n = 4), NRG host (n = 4), and MAC depletion (control; n = 4, clodrosome; n = 3). d, e Barcode distributions in ascites and fat pads under WT (n = 4) and NRG hosts (n = 4). In (c–e) panels, BC.ID1 = TP-1, BC.ID2 = TP-2, etc., for clarity. Data in (c-e) is presented as mean with SD. (f) Injection schema and timeline for comparing one-time (400,000 total, n = 10) and fractionated (2,400,000 total, n = 10) doses. g Representative BLI images at endpoint (left) and total emission in each group (n = 10/group) at endpoint (right). Significance was determined using a two-tailed Student’s t test, and the data is presented as mean with SD. h Timeline schematic depicting in vivo DT-mediated ablation of DTR-expressing early DTCs, followed by analysis of BC.ID distributions in omenta 14 days post initial injection (n = 3 naïve, n = 4 for PBS/DT for 1i-1k). i Quantification of omental weights at a 2-week timepoint. Data was analyzed via a two-tailed Student’s t test and graphed as mean with standard deviation (SD). j BC.ID distribution frequencies in DT and PBS-treated groups. (k) Contribution of the Mid and Late BC.IDs in PBS and DT-treated groups. Data was compared using a two-tailed unpaired Welch’s t test. The center line on the box plot indicates the median, bounds represent the 25^th and 75^th percentiles, and whiskers extend to minima and maxima. l Timeline schematic for early timepoint whole tissue omental imaging. m, n Representative whole-tissue images of cleared omenta harvested 24 h post GFP+ ID8 p53^-/- injection, 38x magnification (scale bars, 40 μm). Images were taken in three representative locations/omenta (n = 4 mice total), without (m) or with (n) prior RFP+ ID8 p53^-/- preconditioning. 3D data was visualized in Napari. Source data are provided as a Source Data file.

To model continuous dissemination in vivo, each individual MetTag-LARRY-GFP line was injected intraperitoneally (i.p.) in a fractionated manner at low doses (400,000 cells/dose, every 48 hours) across six different time points to model early/pioneer, mid, and late waves of DTCs (Fig. 1b). Four weeks after the final DTC exposure (~6 weeks after initial), ascitic fluid (Asc), omenta (Om), and peritoneal pelvic fat pads (FPs) were extracted. This experimental scheme was performed in several mouse models, including wild-type C57BL/6 J mice, immunocompromised NOD.Cg-Rag1^tm1MomIL2rg^tm1Wjl/SzJ (NRG) mice (lacking functional T, B, and NK cells), and under conditions of macrophage (MAC) depletion via clodrosome administration. Temporally spaced DTC exposure resulted in robust metastasis formation in fully immunocompetent mice, evidenced by ascites accumulation and positive GFP signal in the omentum and pelvic adipose tissue (Supplementary Fig. 1g top). Quantification of overall metastasis burden in a separate cohort of ID8 p53^-/- Luciferase cells injected into NRG mice yielded reduced omental burden compared to immunocompetent female C57BL/6 (B6) mice (Supplementary Fig. 1g, h). Moreover, clodrosome-mediated macrophage depletion in B6 mice resulted in reduced GFP+ cancer cell burden in ascites under the fractionated injection schema (Supplementary Fig. 1i, j), suggesting a potential pro-metastatic role of both macrophages and lymphoid (T, B, and NK) cells in ovarian cancer.

We used MetTag-LARRY-GFP amplicon DNA sequencing to quantify the relative abundance of BC.IDs in final overt metastases: early timepoints (BC.ID1 = TP-1, BC.ID2 = TP-2), mid (BC.ID3 = TP-3, BC.ID4 = TP-4), and late (BC.ID5 = TP-5, BC.ID6 = TP-6) arrivers (Fig. 1b, pie chart). Here, the term late DTCs refers to DTCs introduced at later timepoints into a system that has been preconditioned by prior waves of mid and early DTCs, and importantly, is not interchangeable with the clinical terms for early and late dissemination. Interestingly, in overt metastases collected from three distinct peritoneal niches (omentum, ascites, and fat pad), we observed a consistent pattern of early DTC enrichment, suggesting a relative dominance of early arrivers in shaping metastasis outcomes. We observed a nearly identical BC.ID distribution across all three anatomical locations in both wild-type and NRG mice (Figs. 1c–e), as well as macrophage-depleted hosts in omenta (Fig. 1c), suggesting a negligible role of niche-specific tumor immune microenvironment (TIME) in the overall dominance of early arriving DTCs, despite their apparent role in modulating overall metastasis burden.

Early arriving DTCs are sufficient to drive omental colonization

To examine if later arriving cancer cells are dispensable for overall metastasis burden, we evaluated tumor burden in mice with a single injection of tumor cells (TP-1 only) compared to mice with fractionated dosing (TP-1-6) as described in Fig. 1b in female B6 mice with luciferase-tagged ID8 p53^-/- cells. At the experimental endpoint of four weeks post final cancer cell injection, tumor burden levels were comparable (Fig. 1f, g), as measured by in vivo bioluminescent imaging (BLI). These data indicate that early DTCs are not dependent on later DTCs for metastatic colonization and outgrowth. Next, we specifically depleted early DTCs via diphtheria toxin receptor (DTR)-mediated internalization of diphtheria toxin (DT). Upon validating DTR expression in MetTag-LARRY-GFP BC.ID1 and BC.ID2 lines and confirming DT sensitivity in vitro (Supplementary Fig. 1k, l), we conducted a similar staggered injection experiment as described in Fig. 1b (400,000 cancer cells/dose) with the addition of either PBS or DT treatment and collected omenta at an early week 2 timepoint (Fig. 1h). Depletion of early DTCs via DT treatment significantly reduced omental metastatic burden (Fig. 1i), with proportional dominance of mid and late BC.IDs (Fig. 1j, k). Importantly, given that this BC.ID compositional shift did not result in aggressive outgrowth of later-arriving DTCs (Fig. 1i), these results indicate that early DTCs are functionally essential and largely contribute to full metastatic outgrowth. Furthermore, the sequential seeding experimental scheme prompted us to ask whether observed early DTC dominance reflects an advantage of seeding kinetics rather than mechanisms residing in DTC-intrinsic adaptation and/or PSMN co-evolution. To determine whether 24 h is sufficient for omental colonization and whether prior seeding of DTCs constrains subsequent seeding and colonization, we performed a short-term (72 h) in vivo sequential injection experiment using RFP-labeled cells followed by GFP-labeled ID8 p53^-/- cells (Fig. 1l). Whole-tissue immunofluorescence light-sheet microscopy performed 24 h after GFP-cell injection showed that GFP+ cells (TP-2) were able to colonize the omentum at similarly detectable levels regardless of prior RFP+ cell seeding (Fig. 1m, n). These data indicate that: (1) prior seeding by leading DTCs (RFP+ ) does not appreciably compromise the ability of subsequently arriving DTCs (GFP+ ) to colonize the omentum over the first 72 h, and (2) the observed early DTC dominance is not a result of simple competition for available niche space, but rather a gradual adaptive shift in niche hospitability.

Clonal analysis reveals shared evolutionary path towards an inflammatory signature

To gain a better understanding of the transcriptional programs necessary for early DTCs to colonize and to assess their clonal expansion status, we performed CITE-seq (scRNA-seq with HTO sample multiplexing) coupled with direct capture of U6-driven MetTag-LARRY-GFP transcripts via capture sequence 1 (CS1) (Fig. 2a). Sequential injections were performed in the same manner as in Fig. 1b with a reverse barcode injection schema to ensure no unintended bias towards certain barcodes gained over the cell selection period. Single-cell analysis of MetTag-LARRY-GFP BC.IDs from metastasis-bearing tissues revealed that early DTCs (TP1-2) were enriched at endpoint (Fig. 2a, b). The prominent trend towards early DTC dominance was consistent with gDNA sequencing data in Fig. 1c, d. We next asked whether the dominance of early DTCs in overt metastases represents a widespread niche-adaptive and PSMN conditioning feature of DTCs. We applied three MetTag-LARRY-GFP BC.IDs in a triple negative breast cancer (TNBC) mouse model of lung metastasis (Lung.Met): BC.ID1 representing early TP-1 and TP2, BC.ID2 representing mid TP-3 and TP4, and BC.ID6 represents the late TP-5 and TP-6. In this model, we identified almost exclusively early TP DTC dominance (91.9%) in overt lung metastases (Fig. 2c). Collectively, these data suggest that early-arriving DTCs preferentially dominate endpoint metastases in both hematogenous (TNBC) and transcoelomic (ovarian) metastases.

Fig. 2. Clonal analysis reveals shared evolutionary path towards inflammatory signature.

a Schematic of scRNA-seq-based MetTag-LARRY-GFP and transcriptome detection method where LARRY barcodes are readily captured by Chromium 10 x Feature Barcoding (top). Pie charts of RNA-level BC.ID frequencies in each location (bottom) grouped as early (BC.ID5, BC.ID6), mid (BC.ID4, BC.ID3), and late (BC.ID2, BC.ID1). This experiment used a reversed BC.ID injection order compared to Fig. 1. OmMet; HTO n’s = 2, Ascites; HTO n’s = 4. b Heatmap of barcode frequencies, per individual sample/HTO: BC.ID6 = TP-1, BC.ID5 = TP-2, BC.ID4 = TP-3, BC.ID3 = TP-4, BC.ID2 = TP-5, BC.ID1 = TP-6. c Timeline schematic (top) and MetTag-LARRY-GFP BC.ID distribution (bottom) in independent TNBC lung metastasis model depicting enrichment of scRNA-seq derived early TP DTCs in metastatic lungs (HTO n’s = 3) at endpoint. d UMAP of reclustered cancer cells with transcriptional program labels obtained from integrated OmMet (n = 2 mice, 2029 cells) and Ascites samples (n = 4 mice, 373 cells); these biological replicate numbers apply to panels (d–k) (top). Stacked bar chart depicting the frequency of Seurat subclusters (bottom). e Dot plot depicting Hallmark pathways across subclusters used to define each cluster subset signature in (d). Significance was determined via two-sided Wilcoxon rank-sum test with Bonferroni correction. f Schematic of possible outcomes based on enriched clone transcriptome patterns. g Heatmap of normalized clonal LARRY barcode counts across the two anatomical locations and mice (HTOs). h UMAP overlay of top 3 expanded LARRY clones in the omentum (2402 cells; n = 6 mice), of which 39 cells belong to the top three clones. i Subcluster proportions in top 3 barcode clones. j PAGA velocity graph with trajectory arrows and inferred cluster lineage. k scVelo and CellRank-based simulated path from start to endpoint clusters. Source data are provided as a Source Data file.

As most DTCs do not survive and only rare clones successfully adapt to distant niches¹⁹, we examined whether tumor cells disseminated over time evolve along diverse trajectories or converge on a common path characterized by a limited set of adaptive transcriptome features. We first annotated cell clusters with higher levels of keratins and absence of CD45 transcript expression (Krt18 + , Krt8 + , Ptprc-) as the tumor cell-enriched cluster on the UMAP (Supplementary Fig. 2a, b). Tumor cell cluster states were determined through a combination of cluster marker genes and pathway-level analysis using the escape R package²⁰ (Supplementary Fig. 2c–e). Ascites-derived cancer cells were strongly enriched for interferon alpha (IFNα) and gamma (IFNγ) signaling (cluster 3), as opposed to four distinct heterogeneous transcriptional states in OmMet, including proliferative, epithelial-mesenchymal transitioning (EMT), matrisome, and a hypoxic, inflammatory subcluster (Fig. 2d, e).

Cellular barcoding approaches have been widely used to reveal clonal dynamics driven by both genetic predisposition and epigenetic changes in response to environmental cues²¹. As even isogenic cell lines consist of preexisting epigenetically diverse cell populations²², we asked whether the most expanded clones would (1) cluster together based on transcriptional status, indicating strong pre-determined cell-autonomous programs driving metastasis, or (2) disperse into heterogeneous transcriptional states (clusters) reflective of diverse adaptations to different metastatic niche cues (Fig. 2f). First, we observed minimal overlap between dominant outgrowing clones across each mouse (HTO sample) in both anatomical locations (Fig. 2g). While one clone, BC-1, was shared across ascites and omenta, most of the highly expanded clones are distinct between niches. Although these results do not exclude genetic contributions, they suggest that successful clonal outgrowth requires unique adaptations reflective of the specific microenvironment.

Given that omental metastases showed diverse single-cell transcriptional states (Fig. 2d, e), we projected the clonal identity of the top three expanded clones from omenta onto the cancer cell UMAP and examined similarities between clones. Importantly, the top three identified clones, BC-1, BC-4, and BC-49, were not significantly overrepresented in the six MetTag-LARRY-GFP lines before injection (Supplementary Fig. 2f). At both the cluster and pathway levels in vivo, dominant clones were enriched for several defined signatures (Fig. 2h, i and Supplementary Fig. 2g). Apart from the proliferative subcluster 4, expanded clones consistently occupied the hypoxic, inflammatory subcluster 0 (Fig. 2i), suggesting diverse, yet directed evolutionary trajectories within the same clonal population. To understand the relationships among different subclusters, we applied RNA velocity and pseudotime analysis using scVelo and CellRank²³. This approach revealed a clear predicted trajectory originating from the IFN signaling enriched ascites compartment subcluster 3, leading to proliferative status, then further converging to subcluster 0, a terminal inflammatory state in the omentum (Fig. 2j, k and Supplementary Fig. 2h). IFNγ and IFNα signature scores across the six barcode-defined DTC waves are shown in Supplementary Fig. 2i. Similarly to the ovarian cancer dataset, we observed a heterogeneous clonal distribution pattern between each mouse in the TNBC lung metastasis MetTag-LARRY-GFP scRNA-seq dataset, suggesting non-genetic, transcriptomic drivers of successful colonization (Supplementary Fig. 2j, k). To examine the generality of the enriched IFN signature in metastases, we performed a separate scRNA-seq experiment comparing TNBC Lung.Met and E0771 cells in vitro (Supplementary Fig. 3a). Compared to cultured cells, in vivo Lung.Met tumor expressed higher IFNγ and IFNα signatures (Supplementary Fig. 3b–e). Collectively, the ovarian and lung metastasis data suggests that transcriptional adaptation towards IFN-signaling in vivo is a common trait of successfully colonized DTCs. In ovarian metastasis specifically, successful metastatic clones in the omentum occupy heterogeneous transcriptional states. scRNA-seq and trajectory analyses further suggests that these omental states may arise from an IFN-signaling-rich ascites population that precedes later niche-specific adaptations.

In vivo CRISPR screen identifies functional essentiality of IFNγ signaling in omental colonization

We next asked whether transcriptional adaptation of cancer cells to the ascites environment is essential for subsequent intraperitoneal seeding and successful omental metastasis. We hypothesized that the ascites gene signature adopted by early DTCs – characterized by limited nutrient availability and low attachment – is (1) induced by initial detachment from a primary tumor and (2) necessary for subsequent peritoneal metastasis. To test this hypothesis, we performed differential gene expression (DEG) analysis and over-representation pathway analysis (ORA) with the clusterProfiler R package²⁴ to compare gene expression in metastatic tumor cells residing in the two distinct niche compartments (Asc vs Om). Compared to OmMet, DEGs of AscMet consisted of hemoglobin genes (Hbb-bs, Hba-a1, Hba-a2, Hbb-bt) and interferon-stimulated genes (Ifi27l2a and Irf7). In contrast, OmMets were enriched in immediate-early response genes (Atf3, Egr1, Fos, Fosb, Jun)^25,26, adaptive immune immunoglobulin components (Igkc, Igha, Jchain)²⁷, and angiogenesis (Vegfa) (Fig. 3a). Next, IFNγ/IFNα module scoring and pathway analysis confirmed a dominant interferon-rich inflammatory signature in AscMet, as opposed to more heterogeneous programs in OmMet (Fig. 3b and Supplementary Fig. 3f–h).

Fig. 3. Early DTC activation of IFNγ response is necessary for omental colonization.

a Volcano of DEGs between OmMet (n = 2, 2029 cells for 3a-3b) and AscMet (n = 4, 373 cells for 3a-3b) (Seurat FindMarkers(), Bonferroni-corrected two-sided Wilcoxon rank-sum test). b IFN signature comparison between AscMet and OmMet. Box plots: center line indicates median, bounds are 25^th−75^th percentile, and whiskers extend to 1.5 × interquartile range. c CRISPR screening schematic, created in BioRender. Aleksandrovic, E. (2026) https://BioRender.com/qxshldk. d Gene-level log2FC in OmMet (n = 5, 3d-3f) and AscMet (n = 4, 3d-3f) relative to Day 0. e Distribution of individual sgRNAs targeting depleted genes. Bars indicate depleted guides (n = 10) with z-score < 0. f Top 10 depleted pathways associated with depleted genes. g RT-qPCR gene expression graphed as fold change. Data points represent biological replicates (n = 3), each an average of technical replicates. Data was analyzed via two-tailed Student’s t test and plotted as mean with SD. h Time course western blot probing Stat1/pStat1 and GBP1. The blot shown is representative of two independent experiments. i Metastasis study schematic. Two independent groups were pooled; cohort 1 sgNTC; n = 8, sgIfngr1; n = 7; and cohort 2 sgNTC; n = 5, sgIfngr1; n = 5. j log10-transformed emission/flux values with mean, 95% CI. Week 2; p = 0.2679, week 3; p = 0.0103, week 4; p = 0.010, week 5; p = 0.0076, week 6; p = 0.0022 (left). Analyzed by two-tailed Student’s t test; week 6 data plotted as median (right). k Kaplan-Meier survival curve for sgNTC (n = 8) and sgIfngr1 (n = 7). Computed by two-sided log-rank test (p = 0.0044). l Schematic for staggered sgIfngr1 (n = 4 mice) or sgNTC1 (n = 3 mice) followed by sgNTC2-cell injection experiment. m Representative BLI images (left) and BLI quantification (right; n = 4 and n = 3 mice per group) analyzed using a two-tailed Student’s t test and plotted as mean with SD. n gDNA-derived heatmap depicting enrichment of the 1^st DTC wave compared to the 2^nd wave. Source data are provided as a Source Data file.

To assess the functional significance of AscMet and OmMet DEGs, we designed a custom in vivo CRISPR/Cas9 screen targeting 100 DEGs (Fig. 3c). SpCas9 was introduced into ID8 p53^-/- cancer cells and the functionality of the CRISPR system was validated in vitro (Supplementary Fig. 3i, j). 2.4 million cells (~ 2000 × representation of sgRNA library) were i.p. injected to a whole-body Cas9-expressing host (Fig. 3c). This host has central tolerance to Cas9, decreasing immunogenicity towards Cas9-expressing tumor cells²⁸. Six weeks post-injection, metastasis-bearing omenta and ascites were harvested for genomic DNA (gDNA) extraction and NGS sequencing, followed by MAGeCK²⁹ analysis (Fig. 3c). The positive controls for the screen, Pten gRNAs, were significantly enriched in metastases derived from both anatomical locations (Fig. 3d). Compared to the initial injected cell pool sgRNA distribution, sgRNAs targeting eight genes in OmMet and ten genes in AscMet were significantly depleted (p < 0.05) (Fig. 3d, e). The top depleted (dropout) genes – CD300lb, Ms4a8a, Marco, Gbp2b (OmMet), and Slfn1, Marco (AscMet) – were upregulated DEGs associated with AscMet, suggesting a necessity of initial transcriptome adaptation to the ascites niche - the first niche DTCs encounter - for subsequent metastasis. Furthermore, GSEA analysis of dropout genes using the GO Biological Function (GOBP) gene sets revealed a shared depletion of apoptotic cell clearance processes in both Asc and Om compartments and drop out of programs related to IFNγ in omentum (Fig. 3f).

To experimentally validate whether IFNγ is required for the transcriptional upregulation of CRISPR screen hits in cancer cells, we treated ID8 p53^-/- cells in vitro with recombinant IFNγ. We tested the inducibility of Marco, Gbp2b, and Slfn1 by IFNγ via RT-qPCR, where Stat1 and Irf1 transcription factors served as positive controls for IFNγ response³⁰ (Fig. 3g). By 24 h of treatment, all three genes were significantly induced by IFNγ (Fig. 3g). The protein level of Gbp2b, an interferon-inducible GTPase, was also significantly induced by IFNγ (Fig. 3h). Collectively, despite the transcriptional heterogeneity revealed by scRNA-seq, consistent enrichment of apoptotic pathway and the sequential activation of an IFNγ response (first in ascites, then required for omental metastasis) suggest two functionally essential adaptive mechanisms employed by DTCs to survive in vivo clonal selection.

Loss of IFNγ-response in DTC significantly decreases peritoneal metastases and extends overall survival

As most of the top hits from the CRISPR screen were IFNγ-inducible, we reasoned that IFNγ may serve as a primary regulator that exerts broader pro-metastatic adaptive cellular programs in DTCs. Thus, instead of focusing on a single downstream gene target, we sought to disable the entire cascade using an IFNγ receptor 1 (Ifngr1) knockout. We first confirmed robust expression of Ifngr1 in both AscMet and OmMet in vivo (Supplementary Fig. 3k). Next, we genetically knocked out Ifngr1 in ID8 p53^-/- cells using three individual guides (sg#1, sg#4, and sg#6). Successful knockout of Ifngr1 was validated via flow cytometry (Supplementary Fig. 3l), and ablation of the IFNγ transcriptional downstream targets was validated via western blot probing for Stat1 and Gbp2b for sg#1 (Supplementary Fig. 3m). We next i.p. injected ID8.Luciferase-sgNTC or ID8.Luciferase-sgIfngr1(sg#1) to Cas9-expressing mice. Tumor burden was monitored by BLI intensity throughout the 6-week experimental window (Fig. 3i). sgIfngr1 groups showed a significant reduction of overall metastasis burden compared with sgNTC group (Fig. 3j and Supplementary Fig. 3n). In contrast to sgNTC group, ~ 80% of mice in sgIfngr1 groups remained alive at the end of the survival study (p = 0.004) (Fig. 3k).

The above results demonstrated that DTCs’ intrinsic IFNγ signaling is essential for the overall metastatic outcome. Given that early DTCs preferentially dominate endpoint metastases (Fig. 2a–c), we next investigated whether the IFNγ response in early DTCs influences the success of subsequent waves of DTC colonization. We specifically perturbed IFNγ response (sgIfngr1) in the first batch of early DTCs (G2), but not the immediate second wave of sgNTC2 DTCs (Fig. 3l). Interestingly, loss of Ifngr1 in the first batch of DTCs significantly reduced overall metastasis burden at experiment endpoint (Fig. 3m). In the residual metastases in the sgIfngr1 group (G2), control (sgNTC2) cells outcompete the sgIfngr1 cells, as shown with sgRNA amplicon sequencing (Fig. 3n). Taken together, these results demonstrated that a robust IFNγ response in pioneer DTCs is required for overall metastasis success (Fig. 3l–n).

IFNγ response mediates resistance to anoikis by activating pro-survival programs

We next sought to examine how IFNγ signaling promotes metastasis. First, we observed that Ifngr1 knockout did not alter cancer cell fitness under standard adherent culture conditions in the presence or absence of IFNγ treatment, evidenced by both CellTiter-Glo luminescent cell viability assay (Fig. 4a, left) and colony count formation (Supplementary Fig. 4a). Interestingly, IFNγ treatment had a protective effect and increased cell viability in control (sgNTC) cells under ultra-low attachment (ULA) conditions. This protective effect was ablated in the sgIfngr1 groups (Fig. 4a).

Fig. 4. IFNγ Response mediates resistance to anoikis through activation of anti-apoptotic programs.

a Cell-Titer Glo (CTG) assay luminescence graphed as fold change for regular culture (left) and ultra-low attachment (ULA) conditions (right), with and without IFNγ treatment, comparing sgNTC and sgIfngr1 lines. Fold change per biological replicate (n = 5) was normalized to the regular attachment baseline. Data were analyzed by a two-tailed Student’s t test and graphed as mean with SD. b Volcano plots of top DEGs between sgNTC cells under adherent conditions (left), under ULA with (n = 3) and without (n = 3) IFNγ treatment (right); n = 3 biological replicates per group for panels (b–e). Bulk RNA-seq was analyzed with DESeq2 and visualized as variance-stabilized counts; DEGs were evaluated via a two-sided Wald test with Benjamini-Hochberg adjustment. c Venn diagram showing 25 overlapping and 30 unique upregulated genes in cells treated with IFNγ under ULA. d Heatmap of vst-normalized gene counts for ULA-specific upregulated genes, ordered by treatment and group. e Top 15 Reactome pathways upregulated in ULA + IFNγ compared to regular culture. Pathway significance was determined via a one-sided hypergeometric test with Benjamini-Hochberg adjustment. CTG flux for Stat1-KO (f) and Irf1-KO (g) under ULA, with (n = 5) or without (n = 5) IFNγ treatment. Data in (f) and (g) was graphed as median and analyzed using a two-tailed Student’s t test. (h) scRNA-seq based gating strategy used to identify Irf1^high (714 cells) and Irf1^low (186 cells) for module scoring (cells pooled from n = 6 mice). i Selected module expression in Irf1^high and Irf1^low (cells pooled from n = 6 mice). The violin plot center line indicates the median, bounds represent the 25^th and 75^th percentile, and whiskers extend to 1.5 x of the interquartile range. j Violin plots comparing expression level of Parp14 (left), Mcl-1 (middle), and Bcl2 (right) between Irf1^high and Irf1^low (cells pooled from n = 6 mice). Transcript levels in (i) and (j) were compared via two-sided wilcox.test() in Seurat. Source data are provided as a Source Data file.

To investigate the underlying mechanism, we performed bulk RNA-seq (Supplementary Fig. 4b) with wild type (sgNTC) and single clone knockout line (sgIfngr1, sg#1). First, principal component analysis (PCA) and MA dispersion plot showed that ULA culture alone induced significant transcriptomic shifts in both sgNTC and sgIfngr1 cells (Supplementary Fig. 4c, d). Wild-type cells that survived under ULA conditions exhibited an overall reduced S/G2 signature (Supplementary Fig. 4e). Next, we compared the transcriptomic changes in response to IFNγ. PCA of all groups revealed variance between wild-type cells treated with PBS and IFNγ. This IFNγ induced variance was ablated in the sgIfngr1 knockout line as expected, independent of culture conditions (Supplementary Fig. 4c). We analyzed DEGs between cells with or without IFNγ treatment under both regular and ULA conditions (Fig. 4b). Interestingly, despite lower proliferation, cells grown under ULA condition exhibited transcriptionally diverse responses (Fig. 4b and Supplementary Fig. 4f) and were overall more responsive to IFNγ than under regular condition, rendering more DEGs detected (Fig. 4c). Besides 25 common IFNγ responsive genes, we discovered 30 IFNγ responsive genes uniquely up-regulated only under ULA condition (Fig. 4c). We next grouped the unique 30 upregulated genes into broad classes. As expected, most genes represented classical interferon stimulated genes (ISGs) with known roles in viral defense and antigen presentation (Fig. 4d). In addition, we discovered several non-canonical ISGs, such as the immunoproteasome 20S subunit beta 10 (Psmb10), poly(ADP-ribose) polymerase family member 14 (Parp14), as well as PD-L1 (Cd274) (Fig. 4d). Pathway level comparison between IFNγ-treated cells under ULA versus adherent (Reg) conditions revealed Reactome pathways related to metabolic rewiring and glycosaminoglycan remodeling (Fig. 4e). Importantly, Psmb10, Parp14, and PD-L1 all have reported context dependent, cell autonomous pro-survival functions^31–34. Supporting the functional role of downstream responsive genes, knockout of IFNγ transcription factors Stat1 and Irf1 phenocopied Ifngr1 knockout, leading to loss of IFNγ-induced survival phenotype in suspended cultures (Fig. 4f, g and Supplementary Fig. 4g).

To validate in vivo relevance of these programs, we analyzed above DEGs detected in bulk RNA-seq using our single-cell RNA-seq data derived from ovarian metastases. In vivo AscMet exhibited higher levels of immunoproteasome genes (Psmb10, Psmb9) as well as poly(ADP-ribose) polymerase family members (Parp12, Parp14, and Parp9) compared with OmMet (Supplementary Fig. 4h). Unexpectedly, Cd274 was more highly expressed in OmMet compared to AscMet (Supplementary Fig. 4h). Next, we performed RNA-based gating and defined Irf1^high and Irf1^low cancer cells to represent IFNγ downstream responsive levels of each cell (Fig. 4h). We curated custom module scores that reflected common pathways cancer cells utilize to resist anoikis-mediated death: autophagy, stress, and pro-survival/anti-apoptosis³⁵ (Fig. 4i). Irf1^high cells most significantly upregulated the pro-survival/anti-apoptosis module (Fig. 4i, right panel). On an individual gene level, Irf1^high cells upregulated Parp14, Mcl-1 and Bcl2, but not Cd274 and Psmb10 (Fig. 4j and Supplementary Fig. 4i). Finally, we assessed the IFNγ inducibility of Parp14, a known inhibitor of apoptosis in multiple myeloma³³. We observed robust inducibility of Parp14 transcript and protein under ULA conditions in vitro (Supplementary Fig. 4j, k), validating our bulk RNA-seq results, and suggesting active pro-survival programs in detached DTCs under treatment.

Early DTC niche pre-conditioning leads to an LPM and NK-rich microenvironment

Given that early pioneer DTCs possess a unique survival advantage through IFNγ signaling, we next explored if initial dissemination events remodel the metastatic niche. We performed flow cytometry analysis on ascites harvested from the two-week early time point DT treatment experiment (Fig. 1h). In the absence of early DTC depletion (PBS group), metastasis-bearing ascites exhibited significant shifts in both myeloid and lymphoid cell populations relative to the early-DTC-depleted (DT-treated) group. Within the myeloid compartment, depletion of early DTCs significantly reduced the frequency of large peritoneal macrophages (LPMs) in ascites compared with the PBS control group, whereas small peritoneal macrophages (SPMs) and neutrophils were largely unchanged (Fig. 5a). Within the lymphoid compartment, NK cells were significantly reduced in the DT-treated group, whereas total T cells and the CD8/CD4 T cell ratio remained comparable between the two groups (Fig. 5b).

Fig. 5. Pioneer DTC colonization shapes niche-specific macrophage states.

Flow cytometry assessment of (a) myeloid and (b) lymphoid cell populations in naïve (n = 3), PBS-treated (n = 4), and DT treated (n = 4) mice two weeks post initial cell injection. Schematic for the DTR-mediated early DTC wave ablation experiment provided in Fig. 1h. Data were graphed as median and analyzed by two-tailed Student’s t test. c CellChat-generated circos plots of interaction strengths between Broad cell ID categories in metastatic omenta (left) and ascites (right), derived from scRNA-seq. d Stacked bar chart of cell type proportions across the four experimental groups: lavage (n = 4 mice), ascites (n = 4 mice), naïve omenta (n = 4 mice pooled into one HTO), metastatic omenta (n = 2 mice); these numbers apply to panels (d–g). e UMAPs of reclustered myeloid cells, split by experimental condition: lavage (n = 4 mice; 122 cells), ascites (n = 4 mice; 776 cells), naïve omenta (n = 4 mice pooled into one HTO; 321 cells), metastatic omenta (n = 2 mice; 1,139 cells). f UMAP coordinates of cells with high LPM and SPM signature expression (2358 cells; n = 4 lavage, 4 ascites, 4 naïve omenta, 2 metastatic omenta). g Stacked bar chart of relative LPM/SPM ratios in each group. h UMAP of subclustered myeloid cells overlayed with scVelo RNA velocity arrows (1915 cells pooled from n = 6 mice). i DEG volcano plot comparing ascites LPMs (n = 4 mice; 373 cells) and omentum LPMs (n = 2 mice; 423 cells). Two-sided Wilcoxon test in Seurat, with Bonferroni correction. j, k Reactome pathways enrichment analysis of omental (n = 2 mice; 653 cells) and ascites-derived TAMs (n = 4 mice; 476 cells) subsets using a hypergeometric test with Benjamini-Hochberg adjustment. l Stacked bar chart of subsetted T/NK cells comparing cell type proportions in ascites (n = 4 mice; 225 cells) and omentum (n = 2 mice; 492 cells). Source data are provided as a Source Data file.

Next, we assessed major interaction hubs in vivo at metastatic endpoint using CellChat³⁶ analysis on major cell types derived from scRNA-seq Seurat clusters (Supplementary Fig. 5a, b). Although there was a general increase in both the number and strength of cell-cell interactions between cells from naïve and metastasis-derived samples (Supplementary Fig. 5c), tumor cells primarily interacted with myeloid cells in both ascites in omentum (Fig. 5c). Stromal interactions were enriched specifically in the metastatic omentum (Fig. 5c). This suggests that interaction between IFNγ-responsive tumor cells and macrophages may be a key axis of communication across niches. We found a proportional increase in myeloid cells in both metastatic sites compared to their naïve counterparts (Fig. 5d) with the major contribution to the proportional increase arising from subclusters 0,1, 5, and 6 (Fig. 5e and Supplementary Fig. 5d). These macrophage subclusters were identified as tumor-associated macrophages (TAMs) based on high expression of Adgre1, Cd68, Csf1r, Itgam, Mertk, and Trem2 (Supplementary Fig. 5e).

The omental niche drives a pro-inflammatory TAM program

We further classified TAMs into SPMs (expression of Ly6c2, Ccr2, Cd14, Fcgr1) and LPMs (expression of Adgre1, Timd4, Gata6, Icam2)^37,38 (Fig. 5f). Of note, LPMs represented the dominant macrophage population in naïve peritoneal lavage (Fig. 5g) and are subsequently more likely to be the first population encountered by DTCs in the peritoneal cavity. To better understand similarities and differences between LPMs and SPMs at different niches, we performed RNA velocity analysis. We observed a dynamic transition from ascites-derived SPMs to omental SPMs (cluster 1 to 0), and ascites-derived LPMs to omental LPMs (cluster 5 to 6) (Fig. 5h). We then assessed the niche-specific functional states of resident LPM in ascites (Asc.LPMs) and metastatic omenta (Om.LPMs) (Fig. 5i). Asc.LPMs were enriched in tissue residency markers (Cmklr1³⁹) and components of MHC class I (B2m, H2-T23) while Om.LPMs activated programs related to NF-κB signaling (Nfkbiz, Tnfaip3), costimulation (Cd86), and chemokines (Cxcl2) (Fig. 5i).

Given the apparent macrophage niche-dependent differentiation states, we next asked whether omental TAMs reflect a polarized, terminal state within the metastatic niche. If so, increased TAM polarization would likely shape the T cell compartment^40,41. To assess this, we first subsetted macrophages (Cd68^high/Adgre1^high, Supplementary Fig. 5f) and scored them against M1 and M2 gene modules. Compared to ascites-derived TAMs, omental TAMs exhibited higher hybrid functional state scores (Supplementary Fig. 5g), reflecting an immunomodulatory TAM cellular state⁴² that cannot be captured by binning into a strictly immune-promoting or suppressing category. Unlike the immunomodulatory and ECM-related omental signatures, ascites-derived TAMs were enriched in pathways associated with cellular stress responses and NRF2-dependent redox metabolism (Fig. 5j, k). On the gene level, ascites TAM signatures were reflected in elevated expression of key stress-related genes – including Anxa1 and Hmox1 – as well as components of mitochondrial respiration, Ndufb8 and Cox5a (Supplementary Fig. 5h). Indeed, while the ascites-derived T/NK compartment was dominated by a mixture of naïve CD4 + and CD8 + T cells, omental metastases and niche tissue contained a broader and more complex repertoire of activated, exhausted, and regulatory T cells (Fig. 5l) enriched in pathways related to regulation of immune cell activation (Supplementary Fig. 5i). Flow cytometry analysis at metastatic endpoint corroborated higher CD8 T cell infiltration in metastatic omenta compared to ascites (Supplementary Fig. 5j). These findings indicate that ascites TAMs may represent an earlier, less polarized yet metabolically active state, while omental TAMs have transitioned into an immunomodulatory, polarized phenotype.

Peritoneal macrophages are direct DTC pro-survival interactors

We first tested whether macrophages are functionally important in promoting ovarian cancer metastasis. Macrophage depletion via direct i.p. liposomal clodronate (clodrosome) injections (Fig. 6a) demonstrated that peritoneal TAMs are pro-metastatic: depletion significantly reduced omental burden, as evidenced by reduced omental weight, tumor area across tissue sections, and reduced CD45- cancer cell frequencies in ascites at experimental endpoint (Fig. 6a–d and Supplementary Fig. 6a gating strategy). Although clodrosome treatment targets broad macrophage populations, we noted a preferred depletion of ascites LPMs, leading to a 51% reduction of this population at the metastatic endpoint (Supplementary Fig. 6b).

Fig. 6. Peritoneal macrophages sustain DTC fitness through TNF-induced pro-survival signaling.

a Macrophage depletion strategy schematic with representative tissue images. Omental tumor burden quantification (b) and ascites flow quantification (c) (n = 7/group). Significance in (b), (c) was determined via two-tailed Student’s t test and presented as mean with SD. d Representative H&E images (left) and tumor area quantification (right) for control (n = 3) and clodrosome-treated mice (n = 3). Welch’s two-tailed t test was used to compare groups. Data is presented as mean with SD. e Cre-based NicheTracing schematic (top; scale bar, 5 mm). Representative gating strategy for identifying cells tagged with ZsG (bottom). f Quantification of F4/80 and TREM2 expression in contacted, ZsG^high cells (n = 4 tamoxifen-treated). Data are mean with SD, compared by two-tailed unpaired Student’s t test. g Trem2-KO mouse model schematic. h Log10-transformed flux value comparison for Trem2-fl/fl (wild type; n = 5) and Lys2-Cre/Trem2-fl/fl (knockout; n = 5) mice at endpoint. Significance was determined via a two-tailed Student’s t test and data presented as median. i Macrophage isolation and coculture schematic. j Cancer cell viability post-coculture in three cell lines, n = 3 samples per group. Data was analyzed using a two-tailed Student’s t test and plotted as medians. k Cell proportions in human scRNA-seq dataset (Mendeley Data⁶⁹). Maintype-2 annotation provided by the publisher. l Top NicheNet-predicted ligands in ascites after filtering for robust expression and overlapping with top ligands from the murine dataset. m Bulk RNA-seq heatmap comparing enrichment of selected receptors in ULA-cultured cells treated with IFNγ to attached cells. n CTG luminescence fold change (ULA +/− cytokines; n = 6; two-tailed Student’s t test, mean with SD). o Bcl2a1a RT-qPCR (PBS and TNF, n = 3 per group), where dots represent the average of technical replicates. Fold change was analyzed using a two-tailed Student’s t test and graphed as mean with SD.αα Source data are provided as a Source Data file.

To identify cells that have been in close contact with ovarian metastases, we developed a non-SP-TATK Cre-based iteration of the NicheTracing system⁴³ to genetically trace microenvironmental contacts in vivo over a prolonged timeframe. In this system, cancer cells expressing Cre.ERT2 were injected into Ai6 (RCL-zsGreen) Cre reporter mice. Upon intercellular-cytoplasmic transfer and Tamoxifen-mediated nuclear entry of Cre into neighboring cells, Cre-mediated recombination turns on the zsGreen cassette in host cells (Fig. 6e). After validating Cre.ERT2 expression in transduced cell lines (Supplementary Fig. 7a), we benchmarked this system using in vitro coculture (Supplementary Fig. 7b–f) and in an independent model of TNBC lung metastasis established using the E0771 cell line (Supplementary Fig. 7g, h). In the ovarian metastasis model, we readily detected zsGreen^high (ZsG^high) niche cells in both metastasis-bearing omenta and ascites at the experimental endpoint six weeks post cancer cell injection (Fig. 6e). We performed flow cytometry analysis on ascites and gated cells into ZsG^low and ZsG^high (Fig. 6f). Unlike lung metastases where only ~ 30% ZsG^high cells were F4/80^high macrophages (Supplementary Fig. 7h, right panel), over 90% of ZsG^high host cells were F4/80^high LPMs (Fig. 6f). Collectively, the above data suggest that LPMs interact with tumor cells in ascites and have pro-metastatic functions.

We next explored two mechanistic angles to determine how TAMs interact with DTCs. First, we observed that a significant portion of ZsG^high F4/80^high/Ly6G^low TAMs expressed the efferocytotic marker Trem2 (Fig. 6f and Supplementary Fig. 7i, j). As Trem2 macrophages are known to play lipid-sensing, anti-inflammatory and immunosuppressive tumor promoting roles^44–46, we specifically analyzed Trem2^high and Trem2^low TAM populations in our scRNA-seq dataset (Supplementary Fig. 7k). Trem2^high cells showed high expression of broad chemotaxis, ERK pathway, and adhesion pathways (Supplementary Fig. 7l–n). In contrast, significantly enriched pathways in Trem2^low cells were related to epithelial cell proliferation, mesenchymal transition, and TGFβ pathways (Supplementary Fig. 7n), indicating distinct roles. After induction of experimental ovarian metastasis in macrophage-specific (Lyz2-driven) Trem2 KO mice (Fig. 6g and Supplementary Fig. 7o, p), unexpectedly, we observed no significant effect of Trem2 KO on ovarian metastasis burden (Fig. 6h). As ablating Trem2 signaling in macrophages was not sufficient to change the overall ovarian cancer metastasis progression, this suggests that peritoneal macrophages could promote metastasis independently of known Trem2-mediated efferocytotic and anti-inflammatory programs.

Given that the predominant niche contacts in vivo are established with LPMs, the enriched cell type after early DTC priming, and having ruled out the Trem2 efferocytotic program, we hypothesized that peritoneal macrophages instead directly interact with cancer cells to promote initial colonization events. Although previous studies have reported that M2-polarized TAMs could promote tumor cell spheroid formation⁴⁷, it remains unclear if and how naïve peritoneal macrophages could promote early DTC fitness. We performed a macrophage-cancer cell coculture under ultra-low attachment (ULA) conditions (Fig. 6i). Naïve macrophages from peritoneal lavage were enriched using F4/80 magnetic beads (Supplementary Fig. 6c, flow gating strategy) and cocultured with three different murine ovarian cancer cell lines; ID8 p53^-/-, ID8 p53^-/- BRCA2^-/-, and PPNM⁴⁸ (Fig. 6j and Supplementary Fig. 6d flow gating strategy). After a 48-hour coculture period, flow cytometry analysis revealed that cancer cell viability, as measured by live/dead staining (ZombieAqua), decreased under ULA conditions compared to regular attachment plates. Most importantly, survival was partially rescued with macrophage coculture across all three cell lines (Fig. 6j). These results indicate that peritoneal macrophages directly support cancer cell survival under low attachment conditions.

To determine whether this pro-survival macrophage-DTC interaction is conserved in human ovarian cancer, we analyzed a human-derived single-cell dataset containing over 200,000 single cells from 14 ovarian cancer patients collected from five locations: peripheral blood (PB), peripheral lymph nodes (PLN), malignant ascites (Asc), primary tumor (Pri.OT), and matched metastatic tumor (Met.Ome)⁴⁹. As observed in our murine scRNA-seq dataset, cancer cell-containing compartments, including Asc/Pri.OT/ Met.Ome, had increased proportions of macrophages compared to PB and PLN (Fig. 6k and Supplementary Fig. 8a). DEG analysis between TAMs at metastatic sites (Supplementary Fig. 8b) and subsequent GO pathway analysis conducted on the DEG list revealed immune-modulatory pathways (cytokine, chemokine, allograft rejection) were upregulated in Om.TAMs while cargo receptor activity related genes were enriched in Asc.TAMs (Supplementary Fig. 8c).

Macrophage-derived cytokines rescue ascites DTCs from anoikis

We applied NicheNetR analysis to further characterize ligand-receptor interactions between macrophages (ligand provider) and tumor cells (signal receiver through receptor) (Fig. 6k). We implemented this strategy in both our murine dataset (Supplementary Fig. 8d) as well as the patient dataset (Supplementary Fig. 8e). After defining myeloid cells as the sender cell type in the murine dataset and macrophage cluster as the sender in the human dataset, we compared the top predicted tumor-interacting ligands in each dataset. We identified four overlapping sender ligands across species - TGFB1, IL10, TNF, and IL-1β - that were the most significant ligands predicted to interact with cancer cell receptors (Fig. 6l and Supplementary Fig. 8d, e). These ligands were expressed at high levels in patient macrophages (Supplementary Fig. 8f). Next, we examined the top predicted downstream target genes in cancer cells affected by the top TAM ligands. Pathway analysis of the predicted downstream genes yielded pathways involved in cell-cell adhesion, response to oxygen levels, and apoptosis regulation (Supplementary Fig. 8g). Importantly, both SPMs and LPMs in our in vivo murine scRNA-seq dataset expressed the four common ligands (TGFB1, IL10, TNF, and IL-1β) (Supplementary Fig. 8h). We next derived a list of possible corresponding receptors to these four ligands based on prior interaction potential in the NicheNet analysis. Interestingly, cross-examination of corresponding common receptor expression in the bulk RNA-seq data derived from the ULA culture condition (Fig. 4 RNA-seq panels) revealed a subset of up-regulated receptors specific to the ULA culture condition (e.g., Tgfbr1/2, Il1r1, and Il10rb). Such ULA-dependent receptor up-regulation was largely independent of IFNγ treatment (Fig. 6m). In parallel to cancer cell intrinsic IFNγ programs, these data suggest that TAMs provide a cocktail of ligands that could potentially amplify pro-survival signals in IFNγ-primed DTCs.

We next assessed the potential of TNF, IL10, and IL-1β cytokines to promote anchorage independent survival in vitro. TNF and IL-1β, but not IL10 treatment, significantly increased CTG flux in an anchorage-dependent manner (Fig. 6n and Supplementary Fig. 8i). TNF has been reported to promote survival pathways in contexts where the TNF receptor remains sequestered on the cell surface upon TNF binding, and surface sequestration can occur through receptor sialylation via the St6gal1 sialyltransferase⁵⁰. In line with that notion, we observed increased expression of the anti-apoptotic gene Bcl2a1a after TNF treatment (Fig. 6o), and ULA culture primed cancer cells to express higher levels of St6gal1 in vitro (Supplementary Fig. 8j). Mechanistically, TNF treatment induced the activation of canonical pro-survival pathways, including NF-κB (measured via p-Iκbα) in three cell lines and Akt (measured via p-Akt) in two cell lines (Supplementary Fig. 8k).

A combined DTC pro-survival signature predicts clinical outcome in ovarian cancer

In this study, we identified two convergent pro-survival programs that enable DTC fitness in the peritoneal cavity: a tumor-intrinsic IFNγ response and a macrophage-derived paracrine signaling collectively protect DTCs from anoikis. To assess whether these pathways carry clinical relevance, we constructed a composite gene signature derived from our findings (25 genes related to DTC IFNγ signature & 4 genes related to macrophage-DTC NicheNet analysis) for prognostic potential using multivariate Cox proportional hazard modeling across multiple ovarian cancer patient cohorts (Supplementary Fig. 9a).

For overall survival (OS), Ifngr1 was the largest contributor to the risk score, consistent with our data demonstrating that loss of IFNγ receptor signaling significantly reduced metastatic burden and extended survival in the mouse model (Supplementary Fig. 9a). In addition, St6gal1, a TNF-responsive gene implicated its pro-survival signaling, ranked as the second highest contributor to OS risk (Supplementary Fig. 9a). For recurrence-free survival (RFS), IFNγ-inducible Gbp2b/GBP1 has the highest risk score alongside with Marco and Stat1 (Supplementary Fig. 9a). We next stratified patients using our 29 gene composite gene signature median risk score. In both discovery cohorts and independent validation cohorts, ovarian patient tumor gene expression with a high-risk score showed a consistent reduced OS and RFS (Supplementary Fig. 9b, c). These results indicate that the collective contribution of DTC-intrinsic and macrophage-derived pro-survival programs characterized in this study jointly carry prognostic significance in HGSOC.

Discussion

DTC-niche interactions at each stage of the metastatic cascade fundamentally determine disease progression and patient outcomes. Following the initial wave of metastatic dissemination, early DTCs inevitably reshape the metastatic niche microenvironment, irrespective of whether they ultimately survive or perish upon arrival. While the pre-metastatic niche (PMN) has been extensively characterized^9,10, the dynamic processes following DTC seeding represent a less explored system, which we term the Post-Seeding Metastatic Niche (PSMN). The multicellular dynamics resulting from progressive DTC accumulation at newly colonized PSMN mirrors a well-established ecological concept: the Allee effect, which describes how population growth rates depend on population density through competition or cooperation. Although the Allee effect has been demonstrated in metastatic contexts using in vitro models⁵¹, how continuously arriving DTCs dynamically alter PSMN properties and influence the fate of later-arriving tumor cells within the intact physiology of living systems remains unstudied.

Leveraging our MetTag barcoding and staggered injection model, the data presented here consistently demonstrated a skewed overrepresentation of early DTCs: TP-1 and TP-2 DTCs overwhelmingly dominate the metastatic burden at day 40, whereas cells introduced at later days (TP-6) showed relatively reduced contributions to final overt metastases. This pattern persists across immunocompetent, immunodeficient (NRG), and macrophage-depleted hosts (Fig. 1). This skewness in early DTC dominance cannot simply be attributed to inherent, longer chronological growth times of early DTCs, as there remains an ample growth window for TP-6 cells (from Day 11 to Day 40). In addition, these results are consistent with early cancer cell dissemination suggested by clinical tumor genetics studies in ovarian and breast cancer metastasis^11–17. There are several non-mutually exclusive hypotheses that could explain the early DTC dominance observed in our study. First, the omentum might contain a limited number of permissive micro-niches that favor DTC colonization. Pioneer DTCs might rapidly saturate these micro-niches and monopolize spatial resources – a relative first come, first served model operating on a shorter time scale of tumor evolution. Second, the immune-independent nature of skewed distribution of temporally disseminated DTCs suggests a role of non-immune stromal elements in dictating colonization success. Early DTCs may actively remodel the stromal niche, rendering it less favorable for later DTCs. In addition to these inhibitory mechanisms, early DTCs may in parallel, build a niche to support themselves and, to a lesser extent, later arrivers, as their complete absence abolishes later metastatic outgrowth (Fig. 3l, m). Comprehensive spatiotemporally resolved imaging and time course analyses are required to further explore these hypotheses.

Beyond the seed and soil hypothesis, it has been increasingly appreciated that metastasis is not only not merely passive fitness-based selection of DTC clones at the niche, but also an active epigenetic DTC-niche mutual adaptation process. Our single-cell transcriptome analysis, coupled with barcoding clonal analysis, provided us with clonal resolution of the underlying genetic and epigenetic determinants governing DTC behavior. Interestingly, instead of occupying unique transcriptome clusters - which would likely suggest distinctive genetic mutational features - the expanded clones form similar coordinated adaptive mini-ecosystems encompassing multifaceted transcriptome features that converge toward an inflammatory/IFNγ response (Fig. 2h, i). These data align with single-cell whole-genome sequencing (scWGS) of treatment naïve ovarian cancer patients, which details the existence of (1) significant clonal heterogeneity within the same patient and (2) distinct IFNα/IFNγ expressing clones⁵². Our CRISPR/Cas9 in vivo screen results (Fig. 3) further support the notion that cancer cell fitness in the peritoneal cavity and subsequent omental metastasis depend on rapid, transcriptional-level adaptation to IFNγ-responsive programs. As previously noted, our data indicate that the early DTCs IFNγ response is required for subsequent metastatic seeding (Fig. 3l–n). Knocking out Ifngr1 in the early batch of DTCs significantly reduced the metastatic potential of later DTCs with intact Ifngr1 (as evidenced by total metastatic burden), indicating a critical temporal window in which early IFNγ responsiveness and active PSMN conditioning by initial colonized DTCs jointly drive the subsequent metastatic continuum. One important hypothesis to consider in relation to these results is the influence of classical concomitant tumor immunity. This framework describes a phenomenon where a primary tumor restricts distant metastases in an antigenically matched, T cell-dependent, manner^53,54. While the failure of Ifngr1-KO pioneer DTCs to efficiently colonize may result in an increase of available antigens for host immune cell priming, supporting this framework, the persistence of early DTCs in immunodeficient NRG mice (Fig. 1) suggests a more fundamental principle for the observed seeding bottleneck. Our findings expand the conventional view of clonal fitness by revealing its temporal and niche-dependent nature, where transient adaptive states at critical windows determine metastatic success. In this first come, first served model, early metastatic DTC–niche crosstalk through IFNγ sensing-mediated metabolic and epigenetic adaptation determines metastasis success by enabling the earliest colonizers to occupy and remodel permissive micro-niches while restraining later DTCs’ entry.

Although the findings presented here support a tumor cell-centric pro-metastatic role of IFNγ, IFNγ in metastatic niches can act in complex ways. In some contexts, it reduces metastasis through activation of effector T and NK cells, thereby enhancing the effectiveness of immunotherapy^55,56, while in other contexts, IFNγ can induce apoptosis in regulatory T cells and inhibit angiogenesis⁵⁷. Specifically in ovarian cancer, both anti-metastatic and pro-metastatic roles have been reported^58,59. IFNγ has been shown to synergize with beta-glucan treatment to reduce peritoneal metastasis and Ifngr1 knockout in host cells, leading to a significant increase in metastasis⁶⁰. For pro-metastatic roles, IFNγ has been previously reported to upregulate PD-L1 and IL-8 in ovarian cancer cells, leading to increased cell migration and proliferation⁵⁹. We, however, did not observe any proliferation differences via colony formation assay in vitro (Supplementary Fig. 4a) after a 7-day IFNγ exposure. Given that Cd274 and Psmb10 transcript levels were not increased in Irf1^high cells in vivo (Fig. 4j), we focused on Parp14 induction as an active pro-survival cellular adaptation maintained in ascites over time, as opposed to an immediate physiological response to IFNγ signaling. Based on the data presented here with Ifngr1 gene knockout, we believe that, although both effector immune cells and cancer cells could be impacted by IFNγ, IFNγ response in cancer cells (intrinsically) leads to an overall pro-metastatic effect. This effect is an important mechanism driving metastasis, and it should be carefully considered when designing future anti-metastasis treatment. Further functional and mechanistic exploration of Parp14 and CRISPR screen drop out hits, such as Gbp2b and Marco in the context of anoikis resistance is of additional interest for future studies. Importantly, we do not propose that Ifngr1 downstream signaling in cancer cells itself directly reshapes the omental niche. However, we do observe that the loss of attachment conferred by the ascites microenvironment: (1) affects downstream interferon signaling, which impacts survival, and (2) impacts surface expression of receptors responsible for tumor-macrophage interactions.

Cells of the myeloid lineage are characterized by significant plasticity, especially in disease^61,62. However, recent studies point to niche-regulated plasticity restriction⁶³. Upon infiltration of inflammatory monocytes into metastatic sites, they often exhibit commonly described reprogrammed states such as increased checkpoint expression or efferocytotic receptor upregulation due to cancer cell engulfment. In addition to these classical reprogramming events, myeloid cells exhibit tissue-specific reprogramming after prolonged residence at a specific niche⁶³. Thus, although we observe Trem2^high SPMs and LPMs in metastatic omenta and ascites, depleting Trem2 signaling did not significantly impact the metastasis phenotype (Fig. 6h), pointing the role of other programs within the broad LPM population that could provide support to cancer cells. Although IFNγ treatment is sufficient to trigger expression of anti-apoptotic genes, and loss of attachment triggers IL-1β, TNF, TGFβ, and IL10 receptor transcription (Fig. 6m), our data suggests that macrophages (extrinsically) provide multiple pro-survival signals to these primed cancer cells. Specifically, TNF and IL-1β are known amplifiers of NF-κB^64,65, a master-regulator of anti-apoptotic and anti-necrotic gene expression⁶⁶. Given our in vitro findings with TNF and IL-1β treatment (Fig. 6n), we hypothesize that initial IFNγ exposure and loss of attachment prime DTCs to respond strongly to macrophage-derived pro-survival cues and collectively amplify pro-survival programs in DTCs. Importantly, these findings build upon prior knowledge on the role of EGF/EGFR paracrine signaling between tumor cells and TAMs in promoting adhesion of cancer cells through integrins, and thereby spheroid formation⁴⁷.

While our in vivo studies using murine ID8 p53^-/- cells allowed us to explore cancer cell-immune interactions in a fully immunocompetent background, we acknowledge its differences from human HGSOC histologic features, disease heterogeneity, and complexity of genomic alterations. Further studies would benefit from including other experimental models of ovarian cancer. In addition, at advanced stages of ovarian cancer modeled in this study, a significant contribution to the free-floating ascites cancer cell fraction likely arises from secondary seeding from established omental implants, which we cannot distinguish using our barcoding strategy or via bioluminescent imaging, as these approaches only capture a snapshot of the entire peritoneal space. Furthermore, the CRISPR screen gene list was restricted to only the top DEGs due to in vivo coverage constraints and served to dissect the main pathway modules, as opposed to identifying the most significant single gene driver. Dropout-based CRISPR screening approaches are inherently biased towards genes that are necessary for both cell survival and later proliferative expansion. IFNγ responsive genes may facilitate survival in ascites without creating a sufficiently large magnitude of proliferation differences to yield strong dropout hits. Consistent with our interpretation, several IFNγ responsive genes (Marco and Gbp2b) had higher false discovery rate (FDR) values in ascites, which we interpreted as a technical limitation of dropout screening in slower cycling cells. Importantly, although we focused on the role of IFNγ signaling in anoikis resistance, pro-metastatic effects of IFNγ likely extend beyond these survival pathways. Finally, while we utilized NicheTracing to assess cancer cell and immune cell interactions in ascites, our study lacks (1) comparative tracing results in the omental niche and (2) real-time imaging or tissue-level immunofluorescent staining of DTC interactions at different time points.

In summary, our study dissected the complex landscape of ovarian cancer metastasis through a systematic characterization of both cancer cells and local immune factors using MetTag temporal, clonal, and scRNA-seq analysis. Our findings highlight the importance of ascites as a functional reservoir for pro-survival factors, and that early acquisition of IFNγ response and interaction with macrophages are necessary deterministic adaptation steps for DTC survival and colonization.

Methods

Ethical regulations

All animal experiments were conducted according to IACUC protocols, which were pre-approved by the UT Southwestern Medical Center IACUC committee (2022-103328) and the University of Notre Dame’s IACUC committee (18-05-468).

Mice

Mice were housed with unlimited access to food and water in a temperature and humidity-controlled holding room (22 ± 2 °C, 50 ± 10% humidity) with a standard 12 h light/dark cycle. Health checks were performed daily during in vivo experiments. Wild-type C57BL/6 J (B6) mice were 8–10 weeks old at the experiment onset. Control and treated transgenic mice were age-matched, greater than 8 weeks of age. All experiments were carried out exclusively on female mice. Male mice were not considered in the study design and analysis, due to the specific cancer types being studied. C57BL/6 J (B6, 000664), NOD.Cg-Rag1^tm1MomIl2rg^tm1Wjl/SzJ (NRG, 007799), B6.Cg-Gt(ROSA)^{26Sortm6(CAG-ZsGreen1)Hze}/J (Ai6, 007906), and B6J.129(Cg)-Gt(ROSA)26Sor^{tm1.1(CAG-cas9*,-EGFP)Fezh}/J (Cas9 026179) mice were obtained from Jackson Laboratory (Ben Harbor, Me). We generated Lyz2-Cre(+/-)/Trem2(fl/fl) double transgenic mice by breeding Lyz2-Cre heterozygous (+/-) mice with Trem2(fl/fl). These lines were generous gifts from the Zhong Lab at UT Southwestern Medical Center. Specific numbers of mice used in each experiment are indicated in the corresponding figure legends.

Cell lines

The TNBC E0771 cell line (CH3 Biosystems, 940001-Vial) was maintained in RPMI-1640 media supplemented with 10%FBS and 1% Penicilin/Streptomycin (Pen/Strep). ID8 p53^-/-, ID8 p53^-/- BRCA2^-/-, and PPNM ovarian cancer cell lines, provided by the Stack lab at the University of Notre Dame, were maintained in DMEM low glucose media with 5% FBS, 1% Pen/Strep, and 1% Insulin-Transferrin-Selenium (ITS). PPNM cell media was additionally supplemented with 100 ng/mL Cholera toxin and 2 ng/mL epidermal growth factor (EGF). HEK293T cells were cultured in DMEM high glucose supplemented with 10% FBS, 1% Pen/Strep, and 1% GlutaMAX (Gibco, 35050061). Cell cultures were grown at 37 °C, 5% CO₂, and 95% humidity. All cell lines were tested and were negative for mycoplasma contamination using a routine mycoplasma PCR kit (ABM). Murine cell lines were not independently authenticated in this study, however experiments were carried out at low passage numbers after obtaining from the original source.

Establishment and monitoring of experimental metastasis

For the single-injection OvCa model, 4 × 10⁵ ID8-LARRY.GFP or ID8-Luciferase were injected into the peritoneal cavity. To ensure adequate library coverage, 2.4 × 10⁶ ID8-Cas9.Lib tagged cells were injected for CRISPR screening experiments (> 1000 x coverage). In the repeated injection model, mice received 4 × 10⁵ cells per injection, administered six times over the course of 12 days. Mice were typically sacrificed 6 weeks after the initial dose of cancer cells. 6 × 10⁵ E0771-LARRY.GFP or E0771-Luciferase tagged cells were injected into the retro-orbital sinus of mice for a one-time injection lung metastasis model or 1.5 × 10⁵ cells in six separate injections for the repeated injection model. Mice were sacrificed between 14–18 days in both experimental models. For monitoring metastasis in luciferase-expressing models, 100 μL of D-luciferin (15 mg/mL stock, GoldBio cat#: LUCK-500) was administered intraperitoneally and imaged using an AMI-HTX spectral imaging instrument. The total emission (flux) data for each region of interest (ROI) was obtained after image analysis using the Aura software. For experimental metastasis studies, the maximal permitted burden was defined as body weight change exceeding 20% within one week, from tumor-associated weight loss or ascites accumulation, and no animals exceeded this limit. For fixed-timepoint experiments, mice were euthanized before significant symptom onset. For survival studies, animals were monitored daily for lethargy, decreased mobility, respiratory distress, excessive abdominal distension due to ascites accumulation, or deterioration in body condition score. All animals, including lung metastasis and ovarian metastasis experiments, were euthanized via CO2 asphyxiation followed by cervical dislocation immediately upon reaching humane endpoints. Death was not used as an experimental endpoint.

Macrophage depletion and histological analysis

Macrophage depletion was achieved using encapsulated liposomal clodronate (Encapsula Nano Sciences, SKU# CLD-8901). Mice received a pre-treatment with 200 μL clodrosome or empty liposome nanoparticles i.p. 24 h before cancer cell injection. Following pre-treatment, 100 μL of clodrosome was administered twice weekly throughout the study duration. Omental blocks were collected at the endpoint and fixed in 10% neutral buffered formalin (NBF). Fixed tissues were embedded into paraffin blocks and serially sectioned at five non-consecutive intervals separated by 100 μm. Tissue sections were then stained with hematoxylin and eosin (H&E), and images were captured on a Hamamatsu Nanozoomer at the UTSW Whole Brain Microscopy Facility (RRID:SCR_017949) and analyzed in NDPView2 software. For the quantification of tumor area, metastatic tumor regions were manually segmented to calculate the total tumor area for each sample. Any contaminating pancreatic tissue was excluded from quantification.

Individual gene knockout guide cloning

The CROP-seq-V2 plasmid (Addgene # 127458) was first modified to include a P2A-mCherry sequence downstream of the puromycin resistance gene (PuroR) through a NEBuilder HiFi DNA Assembly (NEB, E5520S) reaction. Then, single-guide RNA (sgRNA) sequences were designed using CRISPick (Mouse GRCm38 reference genome), oligos synthesized by IDT with BsmBI overhangs, annealed, and cloned into the modified CROP-seq-V2 plasmid using the NEBridge Golden Gate Assembly Kit (NEB, E1602S). The resulting plasmids were first transformed into competent E. coli, plated on LB-ampicillin (AMP) plates, and grown at 30 °C, before plasmid extraction. Below are the sgRNA sequences used for gene targeting, including the necessary BsmBI overhangs:

sgNTC-1 Top oligo

5’- CAC CGA CCT GAT ACG TCG TCG CGT A −3’

sgNTC-1 Bottom oligo

5’- AAA CTA CGC GAC GAC GTA TCA GGT C −3’

sgNTC-2 Top oligo

5’- CAC CGG TAT TAC TGA TAT TGG TGG G −3’

sgNTC-2 Bottom oligo

5’- AAA CCC CAC CAA TAT CAG TAA TAC C −3’

sgIrf1 guide 1 Top oligo

5’- CAC CGT GGA AGC ACG CTG CTA AGC A −3’

sgIrf1 guide 1 Bottom oligo

5’- AAA CTG CTT AGC AGC GTG CTT CCA C −3’

sgIrf1 guide 2 Top oligo

5’- CAC CGC TGT AGG TTA TAC AGA TCA G −3’

sgIrf1 guide 2 Bottom oligo

5’- AAA CCT GAT CTG TAT AAC CTA CAG C −3’

sgStat1 guide 1 Top oligo

5’- CAC CGT ATC CTG TGG TAC AAC ATG C −3’

sgStat1 guide 1 Bottom oligo

5’- AAA CGC ATG TTG TAC CAC AGG ATA C −3’

sgStat1 guide 2 Top oligo

5’- CAC CGG CAA GCG TAA TCT CCA GGT A −3’

sgStat1 guide 2 Bottom oligo

5’- AAA CTA CCT GGA GAT TAC GCT TGC C −3’

sgIfngr1 guide 1 Top oligo

5’- CAC CGG GCT CGG AGA GAT TAC CCG A −3’

sgIfngr1 guide 1 Bottom oligo

5’- AAA CTC GGG TAA TCT CTC CGA GCC C −3’

sgIfngr1 guide 4 Top oligo

5’- CAC CGA CGC ACT CAC CTC CCC ACT C −3’

sgIfngr1 guide 4 Bottom oligo

5’- AAA CGA GTG GGG AGG TGA GTG CGT C −3’

sgIfngr1 guide 6 Top oligo

5’- CAC CGC TGG GCC AGA GTT AAA GCT A − 3’

sgIfngr1 guide 6 Bottom oligo

5’- AAA CTA GCT TTA ACT CTG GCC CAG C − 3’

Library cloning

Six MetTag-LARRY-BC.ID libraries were generated via overhang PCR by amplifying the region within the original pLARRY-eGFP plasmid containing random barcodes. Primers were designed with overhangs containing: (1) a unique 6-nucleotide library identifier (Id), (2) a chromium 10x capture sequence, and (3) BsmBI restriction sites flanking both ends of the barcode (synthesized by Sigma-Aldrich). Amplified barcode fragments were introduced into the pRG212 third-generation lentiviral plasmid using the NEB Golden Gate Assembly Kit (NEB, E1602S) at a 2:1 insert-to-vector molar ratio. The cloning reaction was transformed into Endura electrocompetent E. coli cells (Biosearch Technologies, 60242-1) and plated overnight on LB-AMP plates. 3 × 10⁶ − 2 × 10⁷ CFU colonies were harvested by flushing prewarmed LB broth and cultured for 2 h in liquid LB. sgRNA oligo pools were synthesized as oPools (IDT), containing BsmBI cloning compatible overhangs. The oligos were then annealed and introduced into the CROP-seq-V2.mCherry lentiviral plasmid using the NEB Golden Gate Assembly Kit as described above. Approximately 5 × 10⁷ CFUs for the OvCa library and 2 × 10⁷ colonies for the MetTag library were harvested for maxiprep. For both the LARRY barcode and CRISPR libraries, plasmids were purified using the ZymoPURE II Plasmid Maxiprep kit (Zymo Research, D4202).

Plasmid library amplification and diversity validation

Plasmid DNA was amplified using a two-step PCR to first amplify and then add indexing sequences to libraries. First round PCR was conducted using custom P5 and P7 adapters with overhangs compatible with Nextera indexing primers, followed by cleanup with Agencourt AMPure XP beads (Beckman Colter, NC9933872) at a 1.8 × bead ratio to remove primers. Index PCR reaction was conducted as outlined in the Illumina 16S Metagenomic Sequencing Library Preparation manual (Illumina, Doc #15044223, page 11-12) using KAPA HiFi HotStart ReadyMix (Roche, KK2601) and Nextera XT Index kit v2 primers (Illumina, FC-131-1001). Final indexed libraries were quantified via the Qubit HS assay kit. LARRY libraries were run on a Bioanalyzer instrument and sequenced on an Illumina MiSeq instrument. CRISPR libraries were run on TapeStation using the High Sensitivity D1000 assay (Agilent, 5067-5585, 5067-5594) and sequenced on a NovaSeq-X instrument. Demultiplexed fastq files were filtered for high-quality reads, and the number of unique reads was quantified. For the CRISPR libraries, a customized script was used to estimate the percentage coverage based on the original guide RNA whitelist.

Lentivirus production and target cell transduction

MetTag-LARRY-GFP, CROP-seq-V2.mCherry guide, or Lenti-Luciferase.P2A.Neo (Addgene # 105621) and second-generation lentiviral packaging plasmids were transfected into HEK293T cells along with the CalFectin transfection reagent (SignaGen, SL100478). Lentivirus was harvested after 72 h and concentrated using Lenti-X Concentrator (Takara Bio, 631231). Concentrated lentivirus was added to target cells along with 8 μg/mL polybrene. To achieve low MOI for CRISPR screening experiments, ID8 p53^-/- ovarian cancer cells were infected at a low multiplicity of infection (MOI) of 0.15, as calculated by the percentage of mCherry-positive cells 48 h post transduction. 48 hours post transfection, cells were selected with either G418 (MetTag-LARRY-GFP lines; Sigma-Aldrich, 4727878001) or puromycin (CROP-seq-V2.mCherry lines; Invivogen, ant-pr-1) for a minimum of 7 days. For clonal and time stamp (BC.ID) tracking, all barcoded pools of cells were infected simultaneously using the same cell passage, same lentivirus titer as measured by a Lenti-X™ GoStix™ Plus kit (Takara Bio, 631280) and were thawed in a staggered manner for injection. This approach ensured equal and short culture time to maintain clonal diversity.

Genomic DNA amplicon library preparation

Genomic DNA was extracted from cultured cells and digested tissues using either the Quick-DNA Midiprep Kit (Zymo Research, D4075) or the Quick-DNA Miniprep Plus Kit (Zymo Research, D4068), depending on the starting cell number. Extracted DNA yield was quantified via a NanoDrop spectrophotometer, and samples were normalized to the same, lowest input concentration prior to PCR. Two rounds of PCR amplification were performed on ≥ 200 ng starting gDNA material. The first round was utilized for amplicon amplification with nested primers for 15–20 cycles using a high-fidelity hot start polymerase, followed by purification with a 1.8 × bead-to-sample AMPure XP bead ratio. Eluted amplicon DNA was quantified using the Qubit high sensitivity (HS) assay kit (Thermo Fisher Scientific, Q33231) and concentrations normalized prior to indexing. Approximately 5 ng of DNA was carried into the index PCR rection, which was conducted similarly to the plasmid library construction protocol described above. Index products were purified using the same AMPure XP beads ratio to remove excess primers. Library quantity was assessed via the Qubit HS assay kit and analyzed on a TapeStation instrument using the High Sensitivity D1000 assay to confirm expected library size and distribution. LARRY gDNA libraries were sequenced in-house on an Illumina MiniSeq instrument (Mid output kit, 300 cycles) while CRISPR gDNA libraries (101 genes × 10 guides per gene + 20 nontargeting guides = 1030 guides) were sequenced on a NovaSeq-X instrument. High-quality reads were filtered and analyzed via two parallel approaches: LARRY barcode counts were extracted to quantify library batch identity (ID), while CRISPR libraries were processed using the MAGeCK pipeline to identify enriched or depleted guide RNAs. Primers used for amplification were as follows:

LARRY Round 1 PCR REV, Tm: 55.9:

5’ – GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACA GTG GGG AAT TTC TAC TCT TGT AGA TCG – 3’

LARRY Round 1 PCR FWD Stagger, Tm: 59.0:

5’-TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG CTT GCT AGG ACC GGC CTT AAA GC −3’

CRISPR Round 1 PCR REV, Tm: 51.0:

5’ – GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACA GTT CTT GGC TTT ATA TAT CTT GTG G– 3’

CRISPR Round 1 PCR FWD Stagger, Tm: 53.0:

5’ – TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG CTC AAG TTG ATA ACG GAC TAG C– 3’

PiggyBac transposon transfection

SpCas9, CreERT2, and DTR coding sequences were cloned into PiggyBac transposon plasmids containing a hygromycin resistance cassette via Gateway cloning (ThermoFisher Scientific, 11791020 & 11789020). Target cells were plated in 6-well plates 24 h prior to transfection. SpCas9, CreERT2, or DTR PiggyBac plasmid was co-transfected with Super PiggyBac Transposase (System Biosciences, PB210PA-1), along with the Calfectin transfection reagent. The transfection mixture was added to cells in a dropwise manner. 12–16 h post-transfection, complete growth media was replenished. Forty-eight hours post- transfection, cells were selected with hygromycin B (Corning, 30-240-CR) for a minimum of 10 days to enrich for successful transgene integrated cells.

Cell Preparation for Single Cell RNA Sequencing

Naïve and tumor-bearing mice were euthanized via CO2 asphyxiation followed by cervical dislocation six weeks following initial cancer cell injection. To collect peritoneal cells, 5 mL of cold 1 × HBSS was injected intraperitoneally using a 25 G needle, followed by lavage and peritoneal fluid retrieval using a 20 G needle. Omenta were excised from the peritoneal cavity, minced, and digested for 30 min- at 37 °C in RPMI-1640 medium containing 0.4% collagenase type II (Gibco, 17101015), 50 μg/mL DNAseI, and 0.1% BSA. Following digestion, cell suspensions were passed through a 70 μM strainer and washed with cold 1 × PBS. Lavage, ascites, and omental samples were resuspended in 1 × RBC lysis buffer (BioLegend, 420301) to remove red blood cells. To prepare high viability (> 80%) cell suspensions for single-cell analysis, omental samples were stained with ZombieRed (BioLegend, 423109) to mark dead cells and live cells were sorted via FACS. Sorted cells were then prepared for 10X Genomics Chromium Single Cell Gene Expression analysis. For TNBC lung sample scRNA-seq, mice were perfused with 5 mL of ice-cold 1 × PBS, lungs were collected and imaged using the GFP channel on a dissecting scope. Lungs with detectable tumor burden were cut into small pieces and processed into single cell suspensions by digestion with 1 × collagenase/hyaluronidase (STEMCELL Technologies, #07912) and 0.15 mg/mL DNase I in serum-free RPMI 1640 medium. After passing the digested samples through a 70 μM strainer and washing with serum-free RPMI, samples were sorted for GFP + , RBCs were lysed and processed similarly to OvCa samples downstream. A detailed protocol for sample multiplexing via hashtags can be found in the CITE-seq/cell hashing protocol on the CITE-seq website (https://cite-seq.com/wp-content/uploads/2019/02/cite-seq_and_hashing_protocol_190213.pdf). In brief, samples were first blocked with FcR Blocking Reagent (Miltenyi Biotec, 130-092-575) in 50 μl of cell staining buffer for 20 min on ice. Samples were then stained with TotalSeq™-A hashing antibodies obtained from BioLegend, at a concentration of 1.0 μg/test, including:

M-HTO-1 (M1/42; 30-F11, 155801),

M-HTO-2 (M1/42; 30-F11, 155803),

M-HTO-3 (M1/42; 30-F11, 155805),

M-HTO-4 (M1/42; 30-F11, 155807),

M-HTO-5 (M1/42; 30-F11, 155809),

M-HTO-6 (M1/42; 30-F11, 155811),

M-HTO-7 (M1/42; 30-F11, 155813),

M-HTO-8 (M1/42; 30-F11, 155815).

Staining was performed for 25 min on ice, followed by three washes with staining buffer (decreasing EDTA concentration per wash) and loading onto a 10 x Genomics Chromium Controller.

scRNA-seq library preparation and illumina sequencing

Single-cell transcriptomic libraries were prepared as outlined in the Chromium Next GEM Single Cell 3’ Reagents Kits v3.1 with Feature Barcoding technology kit manual. Kits necessary for transcriptome library preparation were: Chromium Next GEM Single Cell 3ʹ GEM, Library & Gel Bead Kit v3.1, 4 rxns (10x Genomics, PN-1000128),

Chromium Single Cell 3ʹ Feature Barcode Library Kit, 16 rxns (10x Genomics, PN-1000079), Chromium Next GEM Chip G Single Cell Kit, 16 rxns (10x Genomics, PN-1000127), and Single Index Kit T Set A, 96 rxns (10x Genomics, PN-1000213). Hashtag libraries were prepared using the standard CITE-seq protocol (https://cite-seq.com/wp-content/uploads/2019/02/cite-seq_and_hashing_protocol_190213.pdf). LARRY-BC.ID barcode libraries were constructed using a custom nested PCR protocol followed by indexing. Primers required for LARRY-BC.ID library preparation included:

Round 1 PCR FWD, Tm: 61.0

5’ – GCA GCG TCA GAT GTG TAT AAG AGA CAG – 3’

Round 1 PCR REV: 45.0

5’ – CCT TGG CAC CCG AGA ATT CCA AAT TTC TAC TCT TGT AGA T – 3’

Index PCR FWD, Tm: 61.0

5’ – AAT GAT ACG GCG ACC ACC GAG ATC TAC ACT CGT CGG CAG CGT CAG ATG TGT ATA AGA GAC AG – 3’

Index PCR REV (ADT-RPI-1), Tm 61.0:

5’ – CAA GCA GAA GAC GGC ATA CGA GAT CGT GAT GTG ACT GGA GTT CCT TGG CAC CCG AGA ATT CCA – 3’

The quality and quantity of the three individual libraries (transcriptome, HTO, and LARRY) were assessed via Qubit and TapeStation prior to sequencing. Libraries were pooled for sequencing on an Illumina NovaSeq X Plus platform with paired end reads (PE150).

scRNA-seq general data analysis

Raw fastq files generated from the sequencing runs (mRNA, HTO, LARRY outputs) were processed through the 10x Genomics CellRanger pipeline v.7.2 to generate count matrices. Gene expression (mRNA) reads were aligned to a custom mouse reference genome containing fluorescent marker transcripts (mm10_3.0.0_FPs). Data from two different chromium 10x lanes, ascites lane and omentum lane, were merged using the SCT-normalized dataset integration method. Details related to integration steps can be found at: https://satijalab.org/seurat/articles/. In brief, individual lane single cell data was normalized for HTO (HTOdemux), followed by SCTransform prior to integration. Downstream analyses, including dimensional reduction via principal component analysis (PCA), clustering and UMAP embedding, and differential gene expression (DEG) analysis, were performed in RStudio under the Seurat package (v4.0)⁶⁷. Cell type identities were assigned based on cluster marker genes. Subcluster level labels indicating different cellular statuses were determined through a combination of cluster marker genes and examining enriched pathways. For each HTO sample, LARRY barcode counts were converted to relative frequencies. This transformation allowed for barcode abundance visualization using the same enrichment scale on a heatmap and for comparisons across biological repeats (HTOs). RNA velocity analysis was performed in Python using scVelo and CellRank. Loom files were generated for each lane and merged, followed by Seurat object conversion to H5AD format. Metadata containing clustering and HTO information was integrated into the SCANPY anndata object. More details on the CellRank method can be found at: https://cellrank.readthedocs.io/en/latest/index.html. NicheNetR and CellChat were performed in RStudio. Sender and receiver cell types were selected prior to NicheNetR analysis.

scRNA-seq pathway level analysis

Over-representation analysis (ORA) was performed on a filtered list of upregulated or downregulated genes from DEG analysis and the ReactomePA and clusterProfiler RStudio packages. org.Mm.eg.db package was used to convert genes to Entrez ID symbols. For Single Cell Gene Set Enrichment Analysis (ssGSEA), we utilized the escape R package²⁰ (https://github.com/BorchLab/escape). In brief, murine Hallmark gene sets were obtained directly within escape using the function getGeneSets (species = Mus musculus, library = H). Per-cell enrichment scores were calculated with runEscape using the UCell method, with a minimum gene set size of 5, and stored as a separate assay (escape.UCell_h) within the Seurat object. The parameters sets.size = 2000 and sets.n = 5 were used as specified in the script. The resulting enrichment score matrix was used for downstream visualization and clustering (UMAP embedding using Feature Plot). To identify pathways enriched in individual clusters, differential pathway analysis was performed using Seurat FindAllMarkers on the enrichment assay with the Wilcoxon rank-sum test. Only positively enriched pathways were retained for downstream visualization (minimum detection fraction = 0.1, log2 FC threshold = 0.1). Top enriched pathways were ranked by average log2 FC and visualized using violin plots.

Flow cytometry

Samples were extracted and pre-processed similarly as the scRNA-seq samples. Following tissue digestion of lung or omentum tissues and RBC lysis, digested samples were washed in 1 × Flow Staining Buffer (Tonbo Biosciences, TNB-4222-L500). For viability assessment, cells were incubated with ZombieAqua Fixable Viability Kit (BioLegend, 423102) at a 1:750 dilution in 1 × PBS for 20 min at room temperature. Following viability staining, cells were blocked with 1 × FcR Blocking Reagent (Miltenyi Biotec, 130-092-575) to minimize nonspecific antibody staining. Cells were then stained at the recommended concentration of 0.25 μg/test (100 μL volume, per 1 million cells) with a combination of the following fluorochrome-conjugated surface antibodies: Pacific Blue™ anti-mouse CD45 (BioLegend, 157211, S18009F), APC/Fire™ 750 anti-mouse CD3 (BioLegend, 100247, 17A2) or PE/Cyanine7 anti-mouse CD3 (BioLegend, 100219, 17A2), PE anti-TREM-2 (BioLegend, 824805, 6E9), APC anti-mouse CCR5 (BioLegend, 107011, HM-CCR5), Brilliant Violet 605™ anti-mouse/human CD11b (BioLegend, 101237, M1/70), APC anti-mouse I-A-I-E (BioLegend, 107613, M5/114.15.2), Brilliant Violet 711™ anti-mouse CD4 (BioLegend, 100447, GK1.5), PE/Dazzle™ 594 anti-mouse NK-1.1 (BioLegend, 156517, S17016D) or PE anti-mouse NK-1.1 (BioLegend, 156503, S17016D), Brilliant Violet 421™ anti-mouse PD-L1 (BioLegend, 124315, 10 F.9G2), PE/Cyanine7 anti-mouse CD8a (BioLegend, 100721, 53-6.7) or APC/Cyanine7 anti-mouse CD8a (BioLegend, 100713, 53-6.7), Brilliant Violet 785™ anti-mouse F4/80 (BioLegend, 123141, BM8). Cancer cells harvested from culture were similarly stained with ZombieAqua, FcR blocked, and then incubated with IFNGR1 Recombinant Rabbit Monoclonal Antibody FITC (Invitrogen, MA5-46853, 062) with the corresponding Rabbit IgG Isotype Control FITC, eBioscience™ (Invitrogen, 11-4614-80). Cells were then washed with staining buffer, followed by analysis on a Cytek Northern Lights flow cytometer. FCS files were exported and analyzed using FlowJo (BD Biosciences).

Western blotting

Cells were plated at a density of 250,000-500,000 cells per 10 cm dish for regular attachment plate experiments. For anoikis assays, three × six wells per sample were seeded with 300,000 cells each. The next day, cells were either treated with vehicle (1 × PBS) or 100 ng/mL mouse recombinant IFNγ (PeproTech, AF-315-05-100UG). After treatment, cells were harvested, washed with 1 × PBS, and lysed in M2 lysis buffer (50 nM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100, 0.5 mM EDTA) with 1 × Protease/Phosphatase Inhibitor Cocktail (CST, 5872). Cell suspensions intended for PARP14 blots were treated with RIPA lysis buffer (Thermo Scientific, #89901) and Pierce Universal Nuclease (Thermo Scientific, #88700). Cells were lysed for 1 h on ice before centrifugation to collect protein lysates. Protein quantity was determined using the Pierce^TM BCA Protein Assay Kit (Thermo Scientific, 23225). Protein lysate samples utilized for GBP-1 and PARP14 detection were resolved using 4–20% Mini-PROTEAN TGX Precast Protein Gels (Bio-Rad, 4561094). Other immunoblots were performed on fixed concentration using 8–10% SDS-polyacrylamide gels. All gel transfers were performed onto nitrocellulose membranes (Fisher Scientific, 45-004-075). Nitrocellulose membranes were blocked in 5% BSA/1 × TBST for 1 hour at room temperature, followed by primary antibody incubation overnight on a 4 °C shaker. The following day, blots were washed with 1 × TBST followed by incubation with rabbit or mouse-specific horseradish peroxidase (HRP)-conjugated secondary antibodies. Signals were developed with SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific, #34577) or SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, #34094). Primary antibodies used for western blotting were:

Beta-Actin (13E5) Rabbit mAb, CST #4970, 1:5000 dilution

GBP-1 Rabbit Polyclonal Ab, Thermo, PA5-23509, 1:1000 dilution

Cas9 (S. pyogenes) (7A9-3A3) Mouse mAb, CST #14697, 1:1000 dilution

IRF-1 (D5E4) XP® Rabbit mAb, CST #8478, 1:1000 dilution

Cre Recombinase (D7L7L) XP® Rabbit mAb, CST #15036, 1:1000 dilution

GAPDH (14C10) Rabbit mAb, CST #2118, 1:5000 dilution

Stat1 (D1K9Y) Rabbit mAb, CST #14994, 1:1000 dilution

p-Stat1 (Tyr701) (58D6) Rabbit mAb, CST #9167, 1:1000 dilution

PARP-14 (C-1) Mouse mAb, Santa Cruz sc-377150, 1:2000 dilution

p-Iκbα (Ser32)(14D4) Rabbit mAb, CST #2859, 1:1000 dilution

Iκbα (L35A5) Mouse mAb (amino-terminal antigen), CST #4814, 1:1000 dilution

Akt (pan) (C67E7) Rabbit mAb, CST#4691, 1:1000 dilution

p-Akt (Ser473) (D9E) Rabbit mAb, CST #4060, 1:1000 dilution

Vinculin (E1E9V) XP® Rabbit mAb, CST #13901, 1:5000 dilution

Anti-rabbit IgG, HRP-linked Ab, CST #7074, 1:5000 dilution

Anti-mouse IgG, HRP-linked Ab, CST #7076, 1:5000 dilution

Clinical bioinformatics

To investigate the prognostic value of genes identified in the present study, a total of 29 genes related to interferon response (25 genes) and NicheNet receptor-ligands (4 genes) were used as a gene pool to construct a prognostic signature. Clinical outcome data, including overall survival (OS) and progression-free survival (PFS), as well as related bulk RNA-seq data, were retrieved from the curatedOvarianData R package⁶⁸. To develop the prognostic model, genes with a multivariate Cox regression p-value < 0.1 were first prioritized. The remaining candidate genes were then ranked by the absolute value of their Cox regression coefficients in descending order and iteratively added to the gene panel until up to 80% of all candidate genes were included. Cox regression models were subsequently trained on the discovery cohort for OS and PFS separately using the final selected gene set, and performance was validated in an independent cohort. In each cohort, patients were stratified into low-risk and high-risk groups according to the median linear risk score derived from multivariate Cox regression. Kaplan–Meier curves were used to illustrate survival outcomes between the two risk groups, and log-rank tests were applied to evaluate the prognostic discrimination of the selected gene signature. Patient scRNA-seq RDS file was obtained from Mendeley Data⁶⁹ (https://doi.org/10.17632/rc47y6m9mp.1) and contained 14 patient samples, including PBMCs, lymph nodes, primary tumor, metastasis, and malignant ascites. The RNA data was re-normalized and analyzed using the Seurat package in RStudio using a similar approach to the murine scRNA-seq dataset described above.

Quantitative reverse transcriptase (RT) PCR

The IFNγ treatment time course was performed in 12-well plates. After seeding 1.0 × 10⁵ cells per well, wells were treated with 100 ng/mL IFNγ for 0 h (PBS control), 4 h, 8 h, or 24 h. Total cell RNA was extracted from cells using the Direct-zol RNA Miniprep kit with TriReagent (Zymo Research, R2051-A) following the manufacturer's protocol. One microgram of total RNA was used for reverse transcription to synthesize single stranded cDNA using the iScript^TM Reverse Transcription Supermix for RT-qPCR (Bio-Rad, 1708840). qPCR reactions were performed on a Bio-Rad CFX Opus instrument with SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, 1725270). Data analysis was performed using the CFX Maestro Software. Cq values were used to calculate relative gene expression levels. In brief, calculations were based on the 2^-ΔΔCT method with housekeeping genes serving as internal controls and fold change was calculated by normalizing to the zero-hour treatment timepoint Cq values. RT-qPCR primers used were as follows:

Beta-actin (Mouse) Forward primer TM: 57.6

5’- GGC TGT ATT CCC CTC CAT CG −3’

Beta-actin (Mouse) Reverse primer TM: 55.9

5’- CCA GTT GGT AAC AAT GCC ATG T-3’

GAPDH (Mouse) Forward primer TM: 59.8

5’- CAT CAC TGC CAC CCA GAA GAC TG −3’

GAPDH (Mouse) Reverse primer TM: 61.6

5’- ATG CCA GTG AGC TTC CCG TTC AG −3’

Gbp2b (Mouse) Forward primer TM: 58.5

5’- ATC TGC TCA TTG CTC AGA CTT ACT GG −3’

Gbp2b (Mouse) Reverse primer TM: 57.4

5’- AAG TAT TTT CTC AGC ATC ACT TCA GCC −3’

Irf1 (Mouse) Forward primer TM: 60.4

5’- ACC AAA TCC CAG GGC TGA TCT GG −3’

Irf1 (Mouse) Reverse primer TM: 58.3

5’- GAA GTT TGC CTT CCA TGT CTT GGG-3’

Stat1 (Mouse) Forward primer TM: 55.1

5’- GCT GCC TAT GAT GTC TCG TTT −3’

Stat1 (Mouse) Reverse primer TM: 55.2

5’- TGC TTT TCC GTA TGT TGT GCT −3’

Marco (Mouse) Forward primer TM: 61.1

5’- ATC CAC CTG ATC CTG CTC ACG G −3’

Marco (Mouse) Reverse primer TM: 60.8

5’- CAC TGC AGC GAG AAG AAG GGC −3’

Slfn1 (Mouse) Forward primer TM: 60.9

5’- AGG GAA AGG TCA AGG CTC ACA TTA AAA ATC C −3’

Slfn1 (Mouse) Reverse primer TM: 60.1

5’- AGG CTT CCC AAC AGA GAC ATC TGG −3’

Parp14 (Mouse) Forward primer TM: 59.8

5’- AAG CCT CAG CCT CTA AGA AGT CAA GG −3’

Parp14 (Mouse) Reverse primer TM: 60.6

5’- CAG GTT TTC AAA TGC CAC CAT GGA GG −3’

St6gal1 (Mouse) Forward primer TM: 58.4

5’- CTG CAG GAT CTC TGA AGA ACT CCC −3’

St6gal1 (Mouse) Reverse primer TM: 59.0

5’- TGT ACA AAC TGT CCT TCA GGA AGC G −3’

Bcl2a1a (Mouse) Forward primer TM: 57.9

5’- CAG ATT GCC CTG GAT GTA TGT GC −3’

Bcl2a1a (Mouse) Reverse primer TM: 59.5

5’- TCT GCA GAA AAG TCA GCC AGC C −3’

Anoikis assays

Cancer cells (1.0 × 10⁴) were seeded into 96-well ultra-low attachment (ULA) plates (Corning, CLS7007) to mimic anchorage-independent conditions in the peritoneal cavity environment. After 24 h, 100 ng/mL recombinant mouse IFNγ or vehicle control (1 × PBS) was added to the appropriate wells. Twenty-four hours after treatment (i.e., 48 h after start of culture), CellTiter-Glo Luminescent Cell Viability Assay reagent (Promega, G7570) was applied to every assay well, and plates were incubated for 20 min at room temperature. Luminescence, indicating cellular ATP levels, was measured using a Tecan plate reader instrument. For macrophage coculture experiments, 1.0 × 10⁴ cancer cells were seeded together with 1.0 × 10⁴ macrophages per well in ULA plates. Macrophages were enriched from the peritoneal cavity of wild-type female mice using the MojoSort Mouse F4/80 Positive Selection Kit (BioLegend, 480170), following the manufacturer’s protocol. After 48 h of coculture, cell suspensions were harvested, pelleted, and incubated in TrypLE (Thermo Fisher Scientific, 12605028) for 10 min at 37 °C to dissociate cell aggregates. Cell suspensions were passed through a 40 μM strainer before flow cytometric staining as described earlier.

Colony formation assays

Two hundred ID8 p53^-/- cells were plated onto 6-well plates at the start of the experiment. For wells treated with IFN-γ, 100 ng/mL IFN-γ was maintained throughout the 7-day colony formation assay period. Wells were washed with 1 × PBS, fixed with methanol, and stained with crystal violet solution (0.5% w/v) at room temperature for 20 min. Wells were then washed with deionized water and dried overnight. Per well colony area was obtained in ImageJ.

Cre-based NicheTracing

HEK293T cells were transfected with a loxp-dsRED-loxp-eGFP Cre reporter construct (Addgene # 141148) and cocultured in a 1:1 ratio with Cre.ERT2 expressing ID8 or E0771 cell lines in the presence or absence of 2 μM 4-hydroxytamoxifen (4-OHT, MedChemExpress, Cat # HY-16950). Seventy-two hours after plating, wells were assessed for GFP+ switched cells using a Nikon Ti2 microscope. For independent in vivo validation, E0771-Cre.ERT2 cells were injected into the retro-orbital sinus of Ai6 reporter mice. One week post injection, three consecutive tamoxifen or corn oil control treatments were performed (75 mg/kg/dose) i.p.. Two weeks post cancer cell injection, mice were euthanized, perfused intracardially, and lungs were extracted for imaging and flow cytometry analysis. For the ovarian metastasis model, tamoxifen injections were administered in five consecutive doses, two weeks post ID8-Cre.ERT2 cell i.p. injection.

Bulk RNA-seq

3.0 × 10⁵ Ifngr1-KO or sgNTC ID8 p53^-/- cells were plated onto 6-well ULA and regular treatment plates. Recombinant IFNγ (100 ng/mL) or vehicle (1 × PBS) was added to corresponding wells 24 hours post initial seeding. Cells were harvested 24 h post-IFNγ treatment. ULA samples were subjected to dead cell removal (Miltenyi Biotec, 130-090-101), and total RNA was extracted using the Direct-zol RNA Miniprep kit with TriReagent (Zymo Research, R2051-A) following the manufacturer's protocol. Purified RNA was subjected to library preparation. In brief, mRNA was purified using poly-T oligo magnetic beads, fragmented, and reverse transcribed. The resulting cDNA was subjected to amplification and indexing. Final libraries were sequenced on a NovaSeq-X instrument. Raw reads were trimmed and aligned to the mouse reference genome (mm10). Following transcript level quantification, subsequent count matrices were analyzed in RStudio. DESeq2 was used for downstream analysis, and variance-stabilized counts (VSTs) were used for PCA and visualization. Pathway analysis (Hallmark, Reactome, Go Biological Processes) was conducted through the clusterProfiler R package.

Diphtheria toxin-mediated DTC ablation

DTR or vector (stuffer) control MetTag-LARRY-GFP cells were injected i.p. in a staggered manner into female C57BL/6 J mice according to the timeline established in Fig. 1. Two hours post-first wave of DTR-expressing cancer cell injection, mice received the first dose of either DT (1 ug DT/100ul PBS; Sigma, D0564) or PBS vehicle i.p. to abolish DTCs before colonizing. DT administration was continued every other day for the duration of the cancer cell injection. Animals were monitored daily for DT-induced side effects and sacrificed only two days post-final cancer cell injection to limit the duration of DT exposure.

Light sheet fluorescent microscopy of cleared omental tissue

4 × 10⁵ ID8 p53^-/- cancer cells expressing an RFP tag or vehicle (1 × PBS) were injected intraperitoneally into C57BL/6 background mice. 48 hours later, both groups received a second dose of 4 × 10⁵ ID8 p53^-/- cells labeled with GFP. Omenta were harvested 24 h post ID8 p53^-/- GFP injection. Tissues were surgically harvested, fixed overnight in 4% paraformaldehyde (PFA) at 4 °C with gentle rotation. Tissues were then washed in 1 × PBS containing 0.02% sodium azide to remove residual fixative. After fixation, specimens were decolorized overnight with 25% Quadrol at 37 °C and then permeabilized overnight with blocking buffer (0.5%NP40,10% DMSO, 5% donkey serum, 0.5% Triton X-100 in 1 × PBS) at RT, before immunolabeling. Omental samples were incubated with primary antibodies; RFP preabsorbed antibody (Rockland Immunochemicals INC., 600-401-379) at a 1:200 dilution and anti-GFP antibody (Aves Labs, AB_2307313) at a 1:100 dilution under gentle rotation for 5 days at RT, washed with wash buffer (0.5%NP40, 10% DMSO in 1 × PBS) for 6 h, refreshing every 2 h. To reduce fluorescent aggregate formation, samples were post-fixed with 4% PFA and labeled with secondary antibodies: donkey anti-chicken AF488 (Invitrogen, a78948) at a 1:1000 dilution and donkey anti-rabbit CF555 (Biotium, 20038) at a 1:2,000 dilution for 4 days at RT with gentle rotation. Secondary antibodies were removed with wash buffer for 6 hours, refreshing every 2 h and refreshing next day. After immunolabeling, tissues were incubated in SYTOX deep red (Invitrogen, S11381) overnight for DNA staining. Following immunostaining, tissues were dehydrated through a graded methanol series (25%, 50%, 75% and 2 × 100%). Samples were then delipidated with dichloromethane until the samples sank into the tube. At the end, tissues were cleared by refractive index matching in a benzyl alcohol/benzyl benzoate solution (1:2). This solvent-based clearing workflow reduced light scattering and enabled volumetric imaging of intact or minimally sectioned specimens. Tissues were imaged on a multiscale cleared tissue axially swept light-sheet microscopy (MCT-ASLM) platform. The datasets analyzed were acquired using the high-magnification nanoscale configuration in the 38x imaging format with a pixel size of 0.167 × 0.167 × 0.200 microns. Three-dimensional (3 d) fluorescence image stacks were processed to reduce punctate aggregate-like signal before downstream visualization. For each volume, a smooth background estimate was generated with anisotropic Gaussian filtering, and a background-subtracted response image was used to emphasize small, bright structures relative to the surrounding tissue signal. Candidate aggregates were detected from the extreme upper tail of this response distribution using robust statistics, together with an intensity-to-background ratio criterion, so that unusually bright objects could be identified even in heterogeneous fields. Detected seeds were grouped into connected 3D components and expanded locally to capture the immediate aggregate neighborhood while limiting growth into adjacent tissue. Masked regions were then replaced with a local intensity estimate derived from the surrounding image and lightly smoothed to reduce abrupt transitions. Aggregate detection and replacement were repeated iteratively until the residual burden of high-intensity outliers was substantially reduced or no longer detected by the same decision rules. In addition to the cleaned image, the workflow records a cumulative aggregate mask, per-iteration summaries, and preview renderings to facilitate quality control. 3D fluorescence data were visualized interactively in Napari. Individual fluorescence channels were displayed as separate image layers with additive blending and attenuated maximum-intensity rendering to support inspection of spatial overlap in 3D. Channel identities were rendered with consistent lookup tables, while contrast windows were defined on a per-dataset basis to account for sample and region-dependent fluctuations in brightness. Voxel spacing was applied during rendering to preserve the native anisotropy of the image volumes.

Statistics & reproducibility

Sample sizes were determined based on standard practices established in similar studies for the field, and no statistical method was used to predetermine sample size. No data were excluded, with the exception of statistically significant outliers (ROUT method, Grubbs’ test). The Investigators were not blinded to allocation during experiments and outcome assessment, with the exception of H&E tumor area quantification.

Statistical analyses were performed in GraphPad Prism and RStudio (v4.0). Differential gene expression was assessed using the Wilcoxon rank-sum test (scRNA-seq) or DESeq2 (bulk RNA-seq). Pathway ORA utilized Fisher’s exact test. Log fold changes of > or < than 0.25 with p-values < 0.05 were considered statistically significant. Multivariate Cox regression models were used to conduct clinical bioinformatics and establish prognostic signatures. Differences between independent experimental groups (both in vivo or in vitro) were analyzed using a two-tailed Student’s t-test or Welch’s t test where appropriate. In vitro experiments, including western blotting, were repeated independently at least twice with similar outcomes.

Materials availability

Plasmids produced in this study have been deposited on Addgene with the following identifiers: pENTR-SpCas9 (Addgene ID: 242844), pPB-SpCas9 (Addgene ID: 242845), CROPseq-V2-Puro.P2A.mCherry (Addgene ID: 242847), pENTR-CreERT2 (Addgene ID: 242848), and pPB-CreERT2 (Addgene ID: 242849).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization and experiment design, E.A., M.S.S., J. Lea, and S.Z.; Data analysis, E.A., S.Z., X.L., Z. Zhao, H.M.B., A.T., K.M.D., L.Y., and L.X.; Experiments and methodology, E.A., S.Z., S.R.F., S.M.G., J. Lopez, W.M., N.D., T.C.R., M.Z., Z. Zhong, H.M.B., and K.M.D.; Manuscript writing and revision, E.A., S.M.G., and S.Z.; Study supervision, S.Z. We attest that all coauthors reviewed and pre-approved the manuscript before submission.

Peer review

Peer review information

Nature Communications thanks Chunguang Liang, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This study was supported in part by NIH training grant T32GM075762 (to Aleksandrovic); NIH grants R01 CA194697-01, R01 CA255064-01A1, and R21 CA263798-01 (to Zhang); NIH grant R35GM142654 (to Zhong); NIH grants RM1GM145399 and U54CA268072 (to Dean); Cancer Prevention and Research Institute of Texas (CPRIT) Scholar Award RR220024 (to Zhang); CPRIT Individual Investigator Research Award RP230261 (to Zhong); and CPRIT Core Facility Grant RP250571 (to Dean). We additionally acknowledge support from the Dee Family endowment (to Zhang).

Data availability

scRNA-seq datasets containing LARRY barcode and HTO multiplexing information generated in this study have been deposited in the GEO database under accession codes GSE302619 (ovarian; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE302619) and GSE309120 (lung; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE309120). Raw bulk RNA-seq data has been deposited in the GEO database under accession code GSE337368. The ovarian cancer publicly available scRNA-seq dataset used in this study is available on the Mendeley Data database (https://doi.org/10.17632/rc47y6m9mp.1)⁶⁹. Processed scRNA-seq and RNA velocity data objects related to GSE302619 are available on the Zenodo database (https://doi.org/10.5281/zenodo.21245349)⁷⁰. Raw, unaligned bulk DNA-seq data are available under restricted access due to a lack of suitable standardized repositories specific for in-house clonal and multiplexed barcode screening formats, and due to exceptionally large file sizes. Access to these raw data files will be granted to academics and non-commercial researchers and can be obtained by contacting the corresponding author (Dr. Siyuan Zhang, siyuan.zhang@utsouthwestern.edu). Requests will be processed within 14 days, and raw data will be stored securely for a minimum of five years following this publication. Remaining data, including the processed bulk DNA-seq data related to barcode and sgRNA, are available within the Article, Supplementary Information, or Source Data file. Source data are provided in this paper.

Code availability

Codes used for scRNA-seq and barcode analysis can be found on Github (https://github.com/S-Zhang-Lab/MetTag_OVA_public.githttps://github.com/S-Zhang-Lab/MetTag_BC_TNBC_public). Additional code required to reanalyze the data reported here can be requested from the lead contact.

Competing interests

We do not have any competing interests to declare.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77222-8.