Molecular control of platelet extravasation into tumors and its impact on tumor growth Authors: Hani Lee, Wendolyn Carlos Alcalde, Ricardo Gonzalez Delgado, Ju Young Ahn, Matthew Vasquez, Min Soon Cho, Stephen T C Wong, Anil K Sood, Vahid Afshar-Kharghan Published: 2026 ABSTRACT Key Points•CXCL12-CXCR4, focal adhesion kinase, and platelet endothelial cell adhesion molecule 1 govern platelet extravasation; CXCR4 blockade reduces platelet infiltration and tumor burden.•Dense and α-granule secretions diverge in function: one promotes tumor growth, whereas the other maintains vascular integrity. Introduction Platelets are traditionally viewed as intravascular hemostatic effectors.1, 2, 3 However, evidence from our group4, 5, 6, 7, 8, 9, 10 and others11,12 shows that platelets promote tumor growth and metastasis. Platelets are activated by tumor-derived signals5,13,14 and release bioactive mediators4,7,9,10 that promote tumor growth, angiogenesis, immune suppression, and stromal remodeling.7,15, 16, 17, 18 To act within the tumor microenvironment (TME), platelets must exit the vasculature. We and others have directly observed platelets outside blood vessels within tumors,6,9,19 indicating that extravasated platelets can directly affect cancer cells in the TME. However, mechanisms governing platelet transendothelial migration remain poorly defined. We previously showed that platelet extravasation depends on Gi/G13-mediated cytoskeletal pathways, but not Gq-dependent aggregation, suggesting that migration is driven by motility rather than classical activation,6 with cytoskeletal reorganization and shape change, rather than aggregation, driving platelet transendothelial migration. However, the downstream molecular machinery that translates chemokine sensing and G protein signaling into platelet transendothelial migration remained unknown. Given emerging parallels between platelet trafficking and leukocyte extravasation,20, 21, 22, 23 we investigated whether molecular components central to leukocyte diapedesis, many of which are also expressed in platelets, play analogous roles in platelet migration into tumors.24,25 Here, we identify platelet endothelial cell adhesion molecule 1 (PECAM-1/CD31), focal adhesion kinase 1 (FAK-1), and platelet secretory pathways as core elements of the platelet extravasation program. Our results indicate that platelet extravasation is a regulated process distinct from platelet-mediated effects on tumor progression and provide a rationale for therapeutic strategies that selectively restrict platelet entry or intratumoral activity while preserving systemic hemostatic function. It is important to note that the phenotypes observed in our studies are not attributable to reduced platelet counts in the genetically altered mice. Deficiency of Munc13-4,26 Munc18-2,26 and PECAM-127 has not been associated with significant changes in platelet counts. Materials and methods Reagents 2CP (WuXi AppTec), a synthetic 5-nitrobenzoate small molecule, was dissolved in 10% dimethyl sulfoxide and injected IV twice weekly (3.5 mg/kg) for 3 weeks; 2CP competes with podoplanin (PDPN) for binding to CLEC-2 and selectively blocks PDPN-induced platelet activation with minimal effect on platelet responses to other agonists (eg, thrombin, collagen, ADP).28 Plerixafor (ApexBio) is a bicyclam small-molecule antagonist of the chemokine receptor CXCR429 that blocks CXCL12 binding and inhibits CXCR4-dependent trafficking and is clinically used as a hematopoietic stem cell mobilizer with granulocyte colony-stimulating factor. Plerixafor was dissolved in distilled water and injected subcutaneously every other day (5 mg/kg) for 3 weeks. Ovarian cancer cell lines and generation of CXCL12- and PDPN-deficient ovarian cancer cells ID8 murine ovarian cancer cells were cultured in Dulbecco’s modified Eagle medium supplemented with 10% heat-inactivated fetal bovine serum, 0.1% insulin–transferrin–sodium selenite, and 1% penicillin-streptomycin at 37°C in a humidified incubator with 5% carbon dioxide. CXCL12-deficient ID8 cells were generated using CRISPR-Cas9. Four single guide RNAs (gCXCL12) targeting murine CXCL12 were cloned into the LentiCRISPR version 2 vector and coexpressed with Cas9 to induce double-strand breaks and gene disruption. Control cells were generated using a nontargeting sgRNA (gControl) lacking genomic homology. sgRNA sequences are provided in supplemental Table 1. After transduction, cells were selected with puromycin. Editing efficiency was assessed by T7 endonuclease I (T7E1) assay (supplemental Figure 1A). CRISPR-edited ID8 clones 1 and 2 were further validated by quantitative reverse-transcription polymerase chain reaction and enzyme-linked immunosorbent assay (supplemental Figure 1B-C), pooled, expanded, maintained in puromycin-containing medium, and used for in vivo experiments. ID8 murine ovarian cancer cells were used for experiments with C57BL/6J mice. HeyA8 human ovarian cancer cells were cultured in RPMI-1640 supplemented with 15% heat-inactivated fetal bovine serum and 1% penicillin-streptomycin. PDPN-deficient HeyA8 cells were generated as previously described.30 Briefly, cells were transduced with lentiviral pGIPZ short hairpin RNA (shRNA) constructs targeting PDPN, including shRNA-1 (clone identity [ID]: V2LHS_197169), shRNA-2 (clone ID: V2LHS_196451), shRNA-3 (clone ID: V3LHS_401024), and a combined construct (shRNA-Combo). Control cells received a nonsilencing shRNA vector (scrambled control; clone ID: RHS4346). Stable transductants were selected with puromycin (0.6 μg/mL). All shRNA constructs were obtained from the MD Anderson Cancer Center shRNA and ORFeome Core Facility. Animal models The Institutional Animal Care and Use Committee at MD Anderson Cancer Center approved all animal procedures. Female mice aged 8 to 10 weeks were used in the ovarian orthotopic model. C57BL/6J mice (IMSR_JAX:000664) and PF4-iCre (IMSR_JAX:008535), FAKF/F (IMSR_JAX:031956), and Pecam-1F/F (IMSR_JAX:030758) mice were obtained from The Jackson Laboratory. Munc13-4F/F and Munc18-2F/F mice were provided by Roberto Adachi (MD Anderson). Platelet-specific knockout mice were generated by crossing PF4-iCre with Munc13-4F/F, Munc18-2F/F, FAKF/F, or Pecam-1F/F mice. Athymic nude (Nu/Nu) mice (IMSR_JAX:000819) were also obtained from The Jackson Laboratory. Murine models of ovarian cancer Two million ID8 murine ovarian cancer cells in 100 μL Dulbecco's phosphate-buffered saline (PBS) were injected intraperitoneally into 8- to 10-week-old female C57BL/6J mice. After 4 weeks, tumors were excised, weighed, and embedded in optimal cutting temperature (OCT) compound. For HeyA8 experiments, HeyA8-shPDPN or HeyA8-shControl cells (2 × 106 in 200 μL of Hanks’ balanced salt solution) were injected intraperitoneally into 6- to 8-week-old female athymic nude mice. At 6 weeks, moribund mice were euthanized, and tumors were excised, weighed, and formalin-fixed. Tumor-bearing mice were randomized to receive 2CP (3.5 mg/kg, IV, twice weekly) starting on day 9 for 2 weeks, or plerixafor (5 mg/kg, subcutaneously, every other day) starting on day 9 for 3 weeks. Analysis of single-cell RNA sequencing data Single-cell spatial transcriptomic data for high-grade serous ovarian cancer were obtained from the Broad Institute Single Cell Portal. This data set includes processed profiles from 2.5 million cells across 130 tumors from 94 patients. We downloaded the transcripts per million expression matrix and the metadata containing cell-type annotations. Because the original authors already preprocessed the data, no further cell-level filtering was applied. Analyses and visualizations were performed in R (version 4.5.0) using Seurat (version 5.3.0) and ggrepel (version 0.9.3). Gene expression was visualized in a uniform manifold approximation and projection embedding using Seurat functions and customized ggplot2 overlays. Immunofluorescence imaging OCT-embedded tissues were sectioned at 4 μm, washed with PBS, and blocked with 1% bovine serum albumin. Sections were incubated overnight at 4°C with anti-CD41a (14-0411-82; 1:50; Thermo Fisher Scientific) and anti-CD31 (RB-10333-P1; 1:50; Fisher Scientific) antibodies diluted in blocking buffer. After PBS washes, Alexa Fluor 555 donkey anti–rat immunoglobulin G (A-31572; 1:1000; Thermo Fisher Scientific), Alexa Fluor 488 donkey anti–rabbit immunoglobulin G (711-545-152; 1:1000; Jackson ImmunoResearch), and DAPI (4′,6-diamidino-2-phenylindole; D9542; 1:1000; MilliporeSigma) were applied. Slides were mounted in ProLong Glass Antifade Mountant (P36980; Thermo Fisher Scientific). Deep learning–based platelet segmentation Formalin-fixed, paraffin-embedded tumor sections from murine ovarian cancer models were subjected to whole-slide imaging for quantitative platelet analysis. Each slide was first scanned at low magnification to identify all tissue present, and a region of interest (ROI) encompassing the entire tissue area was defined. Total tissue area was measured using imaging software. To account for tumor size variability, area-based sampling was used. The field of view at 40× magnification with 1.5× digital zoom corresponded to 212 μm × 212 μm (44 944 μm2 per image), and the total tissue area was used to estimate the number of fields required for complete coverage. A minimum threshold of 30% tissue coverage was predefined, and depending on specimen size, imaging ranged from partial sampling (≥30% of the tissue area) to complete whole-slide coverage (100%). For larger specimens, randomly distributed image fields were acquired across the ROI until at least 30% coverage was achieved, whereas for smaller sections, imaging was extended to include the entire tissue area. Images were acquired to reach the intended coverage. High-resolution images were acquired using a 40× objective with 1.5× digital zoom and a stitching function (2 × 2 or 3 × 3 tiled arrays), which were positioned across the ROI in randomly selected locations. As a result, the number of images per slide varied depending on tissue size, typically ranging from ∼36 to >100 images. Image acquisition was performed without regard to platelet distribution, and quantitative analysis was conducted in a blinded manner with respect to experimental group allocation. The same acquisition and analysis pipeline was used across all samples. All acquired images within the defined sampling framework were included in downstream analysis. Platelets were identified and quantified using a deep learning–based segmentation pipeline built on a U-Net architecture applied to high-resolution confocal microscopy images of murine ovarian cancer tissue sections.6,31 Raw images were segmented into uniform 128 × 128 × 3-pixel tiles, and data augmentation (random rotations and horizontal/vertical flips) was applied to enhance diversity, yielding ∼3000 tiles for training and validation. The model used a symmetric encoder-decoder architecture with 4 downsampling and 4 upsampling layers; each block consisted of 3 × 3 convolutions followed by batch normalization and ReLU activation, with 50% dropout to limit overfitting. The encoder extracted multiscale features, and the decoder reconstructed these features for pixelwise classification, with the output layer generating per-pixel probabilities for platelet vs background. The model was trained using binary cross-entropy with logits loss and the Adam optimizer (learning rate 3 × 10−4) for 30 epochs with a batch size of 8 on dual RTX A6000 graphics processing units (NVLink). Platelet identification used multiple specificity criteria. Segmented particles were required to be positive for the platelet marker CD41 and to meet a minimum size threshold of ≥2 μm in diameter, thereby distinguishing intact platelets from platelet-derived extracellular vesicles. In addition, spatial localization relative to the endothelial marker CD31 was used to define extravascular platelets, which were required to be clearly outside CD31+ vascular structures. This approach is consistent with our previously published analyses of platelet localization in tumors.6,31 Representative examples illustrating both segmentation performance and the extent of tissue coverage (30%-100%) are provided in supplemental Figure 2. Statistical analysis All statistical analyses were performed in GraphPad Prism version 10. Differences between groups were analyzed using a 2-tailed Student t test or Welch t test in cases of unequal variance. A P value of < .05 was considered statistically significant. Results Deep learning–based quantification of extravasated platelet density in ovarian tumors We developed a deep learning–based pipeline to quantify intratumoral platelets.6,31 Platelets within tumor sections were identified by immunofluorescence staining with anti-CD41a, followed by quantitative image analysis that combined CD41a positivity and platelet size, based on pixel-defined features. The segmentation process uses pixel-level partitioning to distinguish the platelet signal from the background. Overlay with CD41a staining confirms that algorithm-defined boundaries accurately correspond to platelet-associated structures. Figure 1A shows a representative anti-CD41a image with the ROI indicated. The corresponding segmentation output (Figure 1B) identifies platelet-associated particles. Higher-magnification views (Figure 1C-D) compare the raw signal (Figure 1C) with segmentation (Figure 1D), and the overlay (Figure 1E) confirms concordance with CD41a+ structures. Detection and deep learning–based quantification of platelets in ID8-induced ovarian cancer tumors in wild-type mice. (A) Representative immunofluorescence image of tumor sections stained with anti-CD41a to label platelets. (B) Detection of platelet-sized particles based on size filtering alone, independent of CD41a signal; the yellow box indicates the ROI selected for analysis. (C) Higher-magnification view of the ROI showing raw CD41a fluorescence. (D) Corresponding binary segmentation identifying all platelet-sized particles within the ROI. (E) Overlay of size-based particle segmentation with CD41a signal, used to define true platelets (CD41a+ particles within the expected size range) and exclude CD41a− particles or subthreshold CD41a+ fragments. Scale bars, 20 μm. To specifically quantify extravascular platelets, only CD41a+ particles located outside blood vessels were included, with vessels defined as luminal structures bounded by CD31+ endothelial staining. High-resolution confocal images of murine orthotopic ovarian tumors stained for platelets (CD41a) and endothelial cells (CD31) were used as input (Figure 2A-B). Application of the trained model to whole-slide sections enabled accurate segmentation and quantification of platelets across large tissue areas (Figure 2C-I). ROIs from representative tumor areas showed precise delineation of CD41a+ platelets, including near CD31+ vasculature. Overlay of segmentation masks confirmed strong spatial correspondence with underlying tissue architecture (Figure 2I). The pipeline quantified total platelet counts, analyzed area, and platelet density (counts per square millimeters), enabling reproducible assessment of platelet distribution. (Figure 2J). Deep learning–based quantification of extravasated platelets in ID8-induced ovarian cancer tumors in wild-type mice. (A-B) Immunofluorescence staining of orthotopic ovarian tumor sections was performed using anti-CD41a to label platelets and anti-CD31 to label endothelial cells. Whole-slide images were analyzed with a deep learning–based segmentation pipeline to quantify platelet number and spatial density. (A) Representative whole-slide image showing CD41a+ platelet distribution and (B) corresponding CD31+ endothelial staining. (C-E) Yellow boxes denote ROIs selected for quantitative analysis. (C) Higher-magnification view of the ROI showing CD41a staining used for platelet quantification, (D) the corresponding CD31 staining, and (E) the platelet segmentation mask generated from the CD41a signal. (F-I) Further magnified views of the ROI illustrating CD41a signal (F), CD31 signal (G), segmentation output (H), and overlay of the platelet segmentation on the CD31 image (I). (J) Representative quantitative output table summarizing total platelet counts, analyzed tissue area, and calculated platelet density (counts per square millimeter). CXCR4 blockade decreased platelet extravasation into tumors CXCL12, also known as stromal cell–derived factor 1 (SDF-1), is a chemotactic factor for hematopoietic and lymphoid cells.32,33 Furthermore, CXCL12 guides megakaryocyte migration and regulates platelet function.34 We examined the impact of CXCL12 (SDF-1) blockade on platelet extravasation into tumors using plerixafor, a small-molecule antagonist of CXCR4 (the receptor for SDF-1).35 ID8-induced tumor-bearing B6 mice received intraperitoneal plerixafor at 5 mg/kg or placebo (normal saline solution) every other day, starting 9 days after ID8 cell injection, for 3 weeks. Tumor nodules were resected, weighed, and processed for immunostaining. Mice receiving plerixafor had fewer extravasated platelets within their tumors compared to control mice (2483 ± 763.7 platelets per mm2 vs 1539 ± 342.4 platelets per mm2; P = .0087; Figure 3A). Mice injected with plerixafor developed smaller tumors compared to controls (0.15 ± 0.03 g vs 0.11 ± 0.03 g; P = .008; n = 12 mice in each group; Figure 3B). To assess whether vascular permeability contributed to differences in platelet extravasation, we compared total albumin content in tumors resected from plerixafor- and placebo-treated tumor-bearing mice as an indirect measure of permeability. No significant difference was detected between groups (supplemental Figure 3). CXCR4 blockade suppresses platelet extravasation and tumor growth, whereas tumor cell–derived CXCL12 is dispensable for platelet infiltration but contributes to tumor progression. C57BL/6 mice bearing ID8 ovarian tumors were treated intraperitoneally with plerixafor or saline control. Tumors were excised after 3 weeks and analyzed histologically. (A) Tumor sections stained for CD42b and CD31 revealed fewer platelets outside vascular structures in plerixafor-treated mice (n = 7-10 tumor slide sections per group, ∗∗P < .01). (B) At the end point, tumor weights were lower in the plerixafor group compared with controls. Data are presented as mean ± standard error of the mean (SEM); P values were determined by Welch t test (n = 12 mice per group ∗∗P < .01). ID8 cells were engineered by CRISPR-Cas9 to lack CXCL12 (gCXCL12) or to carry a scrambled guide (gControl), then injected intraperitoneally into C57BL/6 mice. (C) Analysis of tumor sections revealed comparable numbers of extravasated platelets in both groups. (D) In contrast, tumors derived from CXCL12-deficient cells were significantly smaller than those from controls. Data are presented as mean ± SEM; P values were determined by Welch t test (n = 22-37 tumor slide sections per group, ∗∗∗∗P < .0001). To examine CXCL12 expression patterns in the human ovarian TME, single-cell RNA sequencing of human high-grade serous ovarian cancer specimens (130 tumor specimens from 94 patients) was performed to identify sources of CXCL12. (E) Distribution of CXCL12+ cells across major compartments, including malignant epithelial cells, stromal fibroblasts, endothelial cells, and immune subsets. (F) CXCL12 expression levels are visualized across clusters, highlighting stromal and immune populations as the dominant producers. NK, natural killer; ns, not significant; Treg, T regulatory cell. CXCL12 derived from ovarian cancer cells is not the primary source of CXCL12 in tumors Using CRISPR-Cas9 with a CXCL12-targeting guide, we generated CXCL12-deficient ID8 ovarian cancer cells (supplemental Figure 1). CXCL12-deficient (gCXCL12) or control (gControl) ID8 cells were used to induce tumors in B6 mice. After 4 weeks, tumors were resected and processed as described. CXCL12 loss in cancer cells did not affect platelet extravasation (5185 ± 3513 platelets per mm2 vs 4637 ± 1423 platelets per mm2; P = .40; Figure 3C) but significantly reduced tumor growth (0.20 ± 0.04 g vs 0.04 ± 0.02 g; P < .0001; Figure 3D), indicating that tumor cell–derived CXCL12 is not the primary source but contributes to tumor progression. To identify the dominant CXCL12 source, we performed single-cell spatial and perturbational transcriptomic analyses on high-grade serous ovarian cancer specimens (Figure 3E-F). Across >2.5 million cells from 130 tumors from 94 patients, this analysis generated a high-resolution spatial map and identified cancer-associated fibroblasts (CAFs) as the principal CXCL12 source (Figure 3E-F). These data support a model in which CAF-derived CXCL12 engages CXCR4 to promote extravasation; however, this remains indirect, as CAF-specific perturbation was not performed. No CAF-specific promoter is currently available, and global fibroblast CXCL12 knockdown would introduce systemic confounding effects in vivo. FAK is required for platelet extravasation FAK is a nonreceptor tyrosine kinase that regulates focal adhesion turnover, linking integrins bound to the extracellular matrix with the actin cytoskeleton and coordinating cytoskeletal remodeling and directional motility.36,37 In megakaryocytes, FAK regulates platelet production, and in platelets, it controls function.38,39 Platelet FAK is activated upon engagement of integrins (glycoprotein IIb/IIIa [GPIIb/IIIa]) and GPVI with their ligands, regulating cytoskeletal remodeling, shape change, and granule secretion.40 We previously showed that platelet-specific FAK is required for tumor rebound after antiangiogenic therapy in a murine ovarian cancer model.19 We next examined whether FAK regulates platelet extravasation and tumor growth. Platelet-specific FAK-deficient mice (FAKF/F; PF4-Cre) were generated as described,19 and loss of platelet FAK was confirmed by western blot (supplemental Figure 4). Mice with FAK deficiency in platelets and megakaryocytes did not have a lower platelet count compared to controls.39 Platelet-specific FAK deletion significantly reduced platelet extravasation, with tumors from FAKF/F; PF4-Cre (FAKΔ/Δ) mice containing 4203 ± 1558 platelets per mm2 vs 5518 ± 1722 platelets per mm2 in FAKF/F controls (Figure 4A). However, platelet FAK deficiency did not significantly affect tumor growth (0.32 ± 0.10 g vs 0.28 ± 0.07 g; P = .21; Figure 4B). Platelet FAK drives extravasation without affecting tumor growth, whereas platelet PECAM-1 supports trafficking and promotes tumor progression. Platelet-specific FAK-deficient (FAKΔ/Δ) and control (FAKF/F) mice were implanted with ID8 tumors and analyzed after 4 weeks. (A) Tumors from FAKΔ/Δ mice exhibited a lower density of platelets in the extravascular space (n = 33-36 tumor slide sections, ∗∗P < .01). (B) Tumor weights, however, were comparable between groups, indicating that platelet FAK is dispensable for tumor growth. Data are presented as mean ± SEM; P values were determined by Welch t test (n = 8-12 mice per group). The role of platelet PECAM-1 was examined using platelet-specific PECAM-1–deficient (PECAMΔ/Δ) and PECAMF/F mice in the ID8 tumor model. (C) Fewer extravasated platelets were observed in PECAMΔ/Δ tumors compared with controls (n = 25-35 tumor slide sections, ∗∗P < .01). (D) End point tumor weights were also reduced in PECAMΔ/Δ mice, suggesting a contribution of platelet PECAM-1 to both platelet migration and tumor progression. Data are mean presented as ± SEM; P values were determined by Welch t test (n = 12-17 tumor section slides per group, ∗P < .05). ns, not significant. PECAM-1 is required for platelet extravasation into tumors PECAM-1 is expressed on neutrophils, endothelial cells, and platelets. Homophilic PECAM-1 interactions between neutrophils and endothelial cells are critical for adhesion and transendothelial migration.41, 42, 43 Given its expression on platelets,44, 45, 46 we tested whether PECAM-1 similarly regulates platelet extravasation. Platelet-specific PECAM-1-deficient mice (PECAMF/F; PF4-Cre) were generated, and loss of platelet PECAM-1 was confirmed (supplemental Figure 5). In a murine ovarian cancer model, PECAM-1 deficiency reduced platelet extravasation and tumor growth (Figure 4C-D). Tumors from PECAMΔ/Δ mice contained 4348 ± 1044 platelets per mm2 vs 4974 ± 683.4 platelets per mm2 in PECAMF/F controls (P = .0019; Figure 4C). As PECAM-1 engagement inhibits platelet activation,47,48 we speculate that this restraint prevents adhesion at endothelial junctions and subendothelial matrix, facilitating continued extravasation. Tumor weights were also reduced in PECAMΔ/Δ mice compared with PECAMF/F controls (0.30 ± 0.06 g vs 0.36 ± 0.09 g; P = .04; Figure 4D), indicating that platelet PECAM-1 supports both extravasation and tumor progression. The CLEC-2/PDPN axis maintains vascular integrity and limits platelet extravasation PDPN is expressed on high endothelial venules, pericytes, myofibroblasts, and malignant epithelial cells, where it binds platelet CLEC-2 to induce activation. The CLEC-2/PDPN axis is essential for blood–lymphatic vessel separation during embryogenesis49, 50, 51 and contributes to cancer progression, metastasis, and venous thrombosis.13,30,52, 53, 54 We evaluated its role in platelet extravasation using platelet CLEC-2–deficient mice (gift from Lijun Xia, University of Oklahoma Health Sciences Center). All tumor-bearing mice died within 10 days of intraperitoneal tumor cell injection because of intraperitoneal hemorrhage, with tumor implants failing to establish, whereas controls survived for 3 to 4 weeks. Pharmacologic blockade with 2CP, a nitrobenzoate inhibitor of PDPN–CLEC-2 interactions,28 increased platelet extravasation (3143 ± 607.6 platelets per mm2 vs 2051 ± 646.8 platelets per mm2; P < .0001; Figure 5A) but reduced tumor growth (0.26 ± 0.06 g vs 0.36 ± 0.02 g; P = .04; n = 3-4 per group; Figure 5B), indicating loss of CLEC-2–mediated platelet protumor function despite increased extravasation. Disruption of CLEC-2/PDPN interactions modifies platelet behavior and tumor outcome. The CLEC-2–PDPN axis was inhibited pharmacologically with 2CP, a small-molecule inhibitor of PDPN binding to CLEC-2. (A) 2CP-treated tumors showed increased numbers of extravascular platelets compared with controls (n = 20-21 tumor slide sections, ∗∗∗∗P < .0001). (B) Despite this increase, tumor growth was significantly reduced in 2CP-treated mice, indicating that intact CLEC-2/PDPN signaling supports platelet-mediated protumor effects. Data are presented as mean ± SEM; P values were determined by Welch t test (n = 3-4 mice per group, ∗P < .05). (C) To examine the impact of PDPN expressed on ovarian cancer cells on platelet extravasation, HeyA8 human ovarian cancer cells were transduced with shPDPN or with control shRNA and implanted intraperitoneally into Nu/Nu mice. Tumors derived from PDPN-deficient cells contained increased numbers of extravasated platelets compared with controls, as assessed by CD42b/CD31 staining and quantitative image analysis. Data are presented as mean ± SEM; P values were determined by Welch t test (n = 6-12 tumor slide sections per group, ∗∗∗P < .001). shPDPN, shRNA-targeting PDPN. Consistent with prior work showing PDPN expression in ovarian cancer cells and reduced tumor growth upon its knockdown,30 PDPN-deficient HeyA8 tumors in Nu/Nu mice exhibited increased platelet extravasation (413 ± 178.2 platelets per mm2 vs 140 ± 54.58 platelets per mm2; P = .0002; Figure 5C). Overall, these data indicate that CLEC-2–PDPN interactions maintain the integrity of neoangiogenic tumor vessels. Platelet-specific CLEC-2 deficiency leads to fragile, hemorrhagic vasculature, whereas transient CLEC-2 blockade increases extravasation but reduces tumor growth by impairing platelet responsiveness to tumor PDPN. The impact of platelet degranulation on platelet extravasation and tumor growth We investigated the role of platelet degranulation in regulating tumor growth and platelet extravasation. Platelet granule secretion depends on cytoplasmic regulators and membrane-anchored proteins, including members of the SNARE family and their cofactors. Central to this system are Sec1/Munc18 proteins, such as Munc18-2, which control vesicle docking and fusion, and Munc13-4, which primes dense granules for release. Although Munc13-4 in platelets mainly affects dense granule release,26 Munc18-2 is crucial for the release of both α- and dense granules.55 We used platelet-specific Munc18-2–deficient and Munc13-4–deficient mice (both kind gifts from Roberto Adachi, Department of Pulmonary Medicine, MD Anderson Cancer Center, Houston, Texas) in our murine model of ovarian cancer. Munc13-4 deficiency did not affect platelet extravasation (3940 ± 3375 Munc13-4–deficient platelets per mm2 vs 3884 ± 2511 wild-type platelets per mm2; P = .92; Figure 6A). However, mice with Munc13-4–deficient platelets developed significantly smaller tumors than control mice (0.10 ± 0.03 g and 0.17 ± 0.03 g, respectively; P < .0001; Figure 6B). Platelet Munc13-4 is essential for tumor growth but not extravasation, whereas platelet Munc18-2 drives extravasation without influencing tumor growth. The impact of dense granule secretion on platelet trafficking and tumor progression was examined using platelet-specific Munc13-4–deficient mice in the ID8 ovarian tumor model. (A) Quantification of extravasated platelets showed no significant difference between platelet-specific Munc13-4–deficient and wild-type mice (n = 44-61 tumor slide sections per group). (B) In contrast, tumors from mice lacking platelet Munc13-4 were smaller than those from controls, indicating a contribution of platelet dense granule release to tumor growth. Data are presented as mean ± SEM; P values were determined by Welch t test (n = 14-18 mice per group, ∗∗∗∗P < .0001). The role of the exocytotic regulator Munc18-2 in platelet trafficking and tumor progression was evaluated using platelet-specific Munc18-2–deficient mice. (C) Tumors from mice with Munc18-2–deficient platelets exhibited a higher density of platelets in the extravascular space compared with controls (n = 32-45 tumor slide sections, ∗∗∗P < .001). (D) Tumor weights, however, were not different between groups, indicating that loss of Munc18-2–dependent secretion alters platelet migration without affecting tumor growth. Data are presented as mean ± SEM; P values were determined by Welch t test (n = 17-19 tumor section slides per group). ns, not significant. The effect of Munc18-2 on platelet extravasation and tumor growth was opposite to that of Munc13-4. Munc18-2–deficient platelets extravasated more than normal platelets (3786 ± 2236 platelets per mm2 vs 2182 ± 1251 platelets per mm2, respectively; P = .0002; Figure 6C). However, Munc18-2 deficiency did not affect tumor growth (0.19 ± 0.11 g in Munc18-2 platelet-deficient mice vs 0.19 ± 0.08 g in normal mice; P = .888; Figure 6D). These results indicate that dense granule release, mediated by Munc13-4, is critical for tumor growth but dispensable for platelet extravasation. In contrast, α-granule release, regulated by Munc18-2, restrains platelet extravasation without influencing tumor progression. Thus, platelet granule subtypes play divergent roles: dense granules drive tumor-promoting functions, whereas α-granules maintain vascular integrity to limit platelet infiltration. Discussion Our study shows that platelet extravasation into tumors is regulated, is multistep, and parallels leukocyte recruitment21, 22, 23 through platelet-specific adaptations (Figure 7). Classically, platelets are depicted as intravascular, short-lived effectors whose biological relevance ends at vascular injury and hemostatic plug formation.1, 2, 3 Platelet extravasation into tumors. Chemokine signaling, adhesion receptors, and secretory machinery coordinate platelet entry into the TME. Attraction is initiated through the CXCL12-CXCR4 axis, as pharmacologic CXCR4 blockade reduced platelet extravasation and tumor growth, whereas tumor cell–derived CXCL12 was dispensable, with stromal and immune cells identified as the dominant source. Cytoskeletal regulators then govern transendothelial passage: platelet FAK promoted extravasation without influencing tumor size, whereas PECAM-1 supported both platelet migration and tumor growth. At the vessel interface, the CLEC-2/PDPN interaction stabilized tumor vasculature; disruption of this axis, either by pharmacologic CLEC-2 inhibition or tumor cell PDPN knockdown, led to increased platelet leakage but reduced tumor progression. Finally, exocytotic regulators fine-tuned platelet function after extravasation: Munc13-4–dependent dense granule release was essential for tumor growth without affecting migration, whereas Munc18-2 deficiency increased platelet extravasation but did not alter tumor burden. Overall, these findings define molecular machinery of platelet extravasation in which CXCR4, FAK, PECAM-1, and Munc18-2 regulate trafficking across the vasculature, whereas PECAM-1, CLEC-2/PDPN, and CXCL12/CXCR4 interactions, along with Munc13-4–dependent secretion, determine the protumor impact of platelets. A central conceptual advance of this study is the mechanistic separation of platelet trafficking from platelet effector function. Canonical models of platelet activation typically treat aggregation, degranulation, and shape change as a tightly coupled, inseparable program.1,2 Although we did not assess platelet aggregation or degranulation in the genetically altered mouse models used in this study, our data suggest that platelet extravasation can occur independently of these classical activation pathways. In our previous work, we showed that Gq-deficient platelets, despite marked impairment in aggregation6 and the established role of Gq in platelet activation and degranulation,56 extravasate efficiently. Consistent with this, in the current study, mice with platelets impaired in degranulation did not exhibit reduced extravasation and, in some settings, showed increased platelet accumulation within tumors. Together, these findings support a model in which platelet extravasation is largely uncoupled from canonical activation and degranulation pathways. This uncoupling explains how platelets infiltrate tumors without provoking intravascular thrombosis yet still affect tissue growth, vascular remodeling, and immune regulation.10, 11, 12 Our data indicate that platelet extravasation into ovarian cancer is directed by chemokine sensing through the CXCL12-CXCR4 axis. Tumor cell–derived CXCL12 is not required for platelet entry, supporting a model in which platelet trafficking is primarily governed by stromal sources within the TME. CAFs, a major reservoir of CXCL12, are likely sufficient to sustain platelet recruitment and extravasation. This is consistent with the established role of CXCL12-rich stromal niches in regulating immune cell positioning, immune exclusion, and therapeutic resistance32,33 and suggests that platelet trafficking is integrated into these broader chemokine-driven programs. In contrast, the reduction in tumor growth after CXCL12 knockdown in cancer cells reflects the pleiotropic protumorigenic functions of tumor-derived CXCL12. CXCL12 promotes tumor cell proliferation and survival,57, 58, 59 facilitates T-cell exclusion from the TME,60, 61, 62 enhances recruitment of myeloid-derived suppressor cells,63 and supports stromal organization and angiogenesis.64 Platelets undergo sequential steps of adhesion, diapedesis, and abluminal accumulation, paralleling neutrophil trafficking.20, 21, 22, 23 Based on these parallels, we propose that platelet transendothelial migration involves endothelial structures, such as the lateral border recycling compartment, which forms permissive passageways during leukocyte diapedesis.65 In neutrophils, this process is critically dependent on PECAM-1, a core component of the lateral border recycling compartment.24,41, 42, 43 Consistent with this model, we identify platelet PECAM-1 as a critical checkpoint for platelet vascular exit. PECAM-1 deficiency impaired platelet extravasation and reduced tumor growth. Notably, PECAM-1 plays a dual role that is particularly well suited to platelet migration: homophilic PECAM-1 interactions facilitate junctional passage while simultaneously suppressing integrin activation and subendothelial matrix adhesion.45, 46, 47, 48 This is especially relevant in the perivascular tissue and TME, which are enriched in platelet agonists, including fibronectin, collagen, fibrinogen, von Willebrand factor, and tumor-derived ADP,4,5 which would otherwise be expected to trigger platelet activation. This inhibitory signaling likely prevents premature platelet arrest and aggregation during transendothelial transit, allowing platelets to traverse endothelial junctions efficiently. We further identify FAK-1 as a cytoskeletal effector that enables platelet transendothelial migration. FAK integrates integrin- and GPVI-dependent signaling to support adhesion turnover and shape change,36, 37, 38, 39, 40 functions essential for endothelial passage but dispensable for platelet protumor effects. In a previous study,19 we also demonstrated that platelet FAK did not affect tumor growth but reduced growth rebound after discontinuation of antiangiogenic therapy. Platelet engagement with the vasculature diverges from leukocyte trafficking through platelet-specific signaling axes, most notably the CLEC-2/PDPN pathway. PDPN is expressed by fibroblastic reticular cells and pericytes surrounding high endothelial venules and tumor vessels,66,67 and its engagement of platelet CLEC-2 is essential for maintaining vascular integrity during development and inflammation.49, 50, 51,67 This interaction activates platelets, leading to sphingosine-1-phosphate release and reinforcement of VE-cadherin–dependent endothelial junctions.49,68,69 In our studies, permanent loss of platelet CLEC-2 because of genetic manipulation led to intratumoral bleeding. In contrast, pharmacologic disruption of CLEC-2 increased platelet extravasation but reduced tumor growth, possibly because PDPN on cancer cells was unable to activate platelets. These opposing outcomes underscore a central principle: platelet extravasation alone is insufficient to promote tumor growth unless coupled to appropriate intratissue effector programs. Systemic CXCR4 inhibition affects tumor, stromal, immune, endothelial, and platelet compartments and therefore does not establish a platelet-specific mechanism. However, these experiments demonstrate that CXCR4 signaling contributes to platelet extravasation in vivo. Although cell-type–specific contributions of CXCR4 blockade on tumor growth will require further investigation, the availability of plerixafor supports the potential translational relevance of targeting CXCR4 signaling in cancer. Consistent with the established role of CXCL12-CXCR4 signaling in maintaining endothelial integrity and limiting permeability,70,71 a permeability-driven mechanism would predict increased platelet extravasation after CXCR4 blockade. However, plerixafor treatment of tumor-bearing mice reduced platelet extravasation into tumors. Moreover, direct measurement of tumor albumin content in tumors resected from plerixafor- and placebo-treated mice revealed no significant difference between groups, arguing against altered endothelial permeability as the underlying mechanism. Although this analysis primarily reflects CXCR4-dependent effects and does not exclude potential contributions from CXCR7, CXCR4 remains the principal receptor regulating CXCL12-dependent endothelial barrier function. A further layer of functional specialization emerges at the level of platelet degranulation. Dense granule release, governed by Munc13-4,26 was dispensable for platelet extravasation but essential for tumor growth, whereas α-granule release, regulated by Munc18-2,55 preserved endothelial barrier function and limited platelet passage. Loss of Munc18-2 increased platelet extravasation without altering tumor growth, reflecting the mixed progrowth and antigrowth cargo of α-granules, including angiogenic, immunomodulatory, and inhibitory mediators.7,72, 73, 74, 75, 76, 77 An important caveat in these results is that Munc13-4 deficiency, while primarily impairing dense granule secretion, may also secondarily attenuate α-granule release through reduced ADP-mediated amplification. However, several lines of reasoning suggest that this indirect effect does not account for our observations. First, platelet extravasation was unaffected in Munc13-4–deficient mice, whereas it was markedly altered in Munc18-2–deficient animals, indicating that α-granule–dependent regulation of endothelial barrier function remains functionally distinguishable from dense granule pathways in vivo. Second, the TME provides multiple platelet agonists, including tumor-derived ADP, thrombin, and exposed collagen, which can independently drive α-granule secretion and may compensate for reduced platelet-derived ADP. Thus, although ADP-dependent amplification contributes to platelet activation, our data support a model in which Munc13-4–dependent dense granule release plays a nonredundant role in promoting tumor growth, separable from α-granule–mediated regulation of vascular permeability. Alterations in vascular integrity could potentially contribute to changes in platelet extravasation in our studies. Several of the proteins examined, including FAK, PECAM-1, Munc13-4, and Munc18-2, are also expressed in endothelial cells and are known regulators of endothelial barrier function. To minimize confounding effects from endothelial-intrinsic signaling, we used platelet-specific genetic models, thereby focusing the analysis on platelet-dependent mechanisms. Importantly, except for platelet-specific CLEC-2 deficiency, none of the tumor models exhibited increased hemorrhage, arguing against a generalized disruption of vascular integrity as the primary driver of platelet extravasation. Because CLEC-2 deficiency produced a hemorrhagic phenotype, these data were not used to assess tumor growth or platelet extravasation. Together, these observations support the interpretation that the observed effects reflect regulated platelet trafficking rather than nonspecific leakage. However, tumor vascular leakiness represents an important alternative mechanism that could contribute to platelet accumulation outside the vessel lumen. Although we did not detect differences in endothelial permeability in our models, we cannot fully exclude the possibility that genetic or pharmacologic perturbations alter tumor vascular integrity and thereby influence platelet extravasation. From a therapeutic perspective, our findings point to 2 complementary strategies: blocking the entry gate via inhibition of PECAM-1, CXCR4, or FAK to prevent platelet infiltration into tumors and silencing the intratumoral payload by targeting Munc13-dependent secretion to disrupt platelet-mediated angiogenesis, stromal activation, and immune suppression while preserving systemic hemostatic function. Conflict-of-interest disclosure: A.K.S. reports consulting for Onxeo, Foundation Medicine, and Kaida; serving on data and safety monitoring boards for Advenchen and Mural Oncology; and research funding from Pfizer. The remaining authors declare no competing financial interests. Acknowledgments This work was supported, in part, by National Cancer Institute, National Institutes of Health grants R01 CA231141 (V.A.-K.), and R01 CA177909 and R01 CA275762 (V.A.-K. and A.K.S.), the John S. Dunn Research Foundation (S.T.C.W.), the American Cancer Society, the Frank McGraw Memorial Chair in Cancer Research (A.K.S.), and the McCollough Professor in Cancer Research (V.A.-K.). Authorship Contribution: H.L. designed and performed experiments, analyzed data, and drafted the manuscript; W.C.A. and R.G.D. performed experiments and analyzed data; M.V. and S.T.C.W. designed and performed imaging experiments; M.S.C. performed experiments and analyzed data; J.Y.A. and S.T.C.W. analyzed and segmented image data using deep learning; A.K.S. designed experiments, interpreted data, and edited the manuscript; and V.A.-K. designed and supervised experiments, interpreted data, and wrote the manuscript. Footnotes The RNA sequencing data analyzed in this study were obtained from publicly available data sets deposited in the Broad Institute Single Cell Portal under accession number SCP2640. The full-text version of this article contains a data supplement. Contributor Information Anil K. Sood, Email: aksood@mdanderson.org. Vahid Afshar-Kharghan, Email: vakharghan@mdanderson.org. Supplementary Material Supplemental Table and Figures References Associated Data Supplementary Materials Supplemental Table and Figures