Platelet-Specific PDGFB Ablation Impairs Tumor Vessel Integrity and Promotes Metastasis

The vasculature constitutes an essential barrier between the tissue parenchyma and the circulating blood and provides key signals to quickly recruit circulating blood cells to sites of injury, inflammation, and infection (1). In addition to the endothelial cells facing the lumen, blood vessels are composed of supporting (mural) cells attached to the abluminal side of the endothelium. Mural cells in capillaries, pericytes, help to maintain integrity of the vasculature, and diseases such as diabetic retinopathy and vascular malformations are characterized by pericyte loss (2–4). During development, proper pericyte recruitment to the vasculature is dependent on the interaction between platelet-derived growth factor B (PDGFB) expressed by the endothelium and PDGF receptor β (PDGFRβ) expressed by the pericytes (5–9). In addition, the endothelial-derived PDGFB must be anchored to the extracellular matrix (ECM) in close proximity to the vasculature, thereby creating a gradient that attracts PDGFRβ-expressing pericytes to form close attachments to the endothelium. This anchoring of PDGFB to the ECM is achieved by the “retention motif,” which is a heparan sulfate–binding region in the C-terminal domain of the growth factor (10–12). In a tumor, pericytes are commonly less numerous and also less tightly attached to the endothelium (13, 14). There could be several explanations, such as loss of the PDGFB gradient due to expression of PDGFB by tumor cells also (11) and/or increased proteolytic activity in the tumor microenvironment, which could release PDGFB bound to the ECM, thereby disrupting the gradient. An imbalance between angiopoietin 1 (Ang1) and Ang2 signaling through the Tie2 receptor on endothelial cells can also affect pericyte coverage in tumor vasculature (15).

We previously made the unexpected finding that platelet depletion (acute thrombocytopenia) in RIP1-Tag2 (RT2) mice with late-stage pancreatic neuroendocrine carcinoma induced a dramatic loss of NG2-positive (NG2⁺) pericytes from the tumor vessels (16). Platelet depletion also resulted in significantly impaired perfusion of tumor blood vessels and a concomitant reduction in tumor cell proliferation. Indeed, platelets have been described as “guardians of the tumor vasculature” (17), because acute thrombocytopenia induces hemorrhage in tumors (18). Similar observations have also been made in other situations involving active remodeling of the vasculature (19). The mechanism underlying this role of platelets in the tumor vasculature is not fully understood, but was reported to be independent of the capacity of platelets to aggregate. Instead, it was concluded that a secreted factor was responsible for the protective effect exerted by platelets on the tumor vessels (18). Whether loss of pericytes is a contributing factor to the tumor vascular damage induced by acute thrombocytopenia is unknown.

By which mechanism could platelets promote pericyte recruitment to tumor vessels? Platelets represent a major source of PDGFB, which is stored in their α-granules and released upon activation. In the tumor microenvironment, platelets are constantly activated because of aberrant expression of tissue factor (TF), leading to generation of thrombin, a potent platelet activator (20). In addition, the discontinuous tumor endothelium exposing subendothelial compartments, containing for example collagen, is another strong activation signal for platelets (21). As described above, PDGFB expressed by the endothelium is critical for proper pericyte recruitment to blood vessels (8, 9). However, in a tumor, there is an ongoing extensive vascular remodeling and additional sources of PDGFB could potentially contribute to maintain a sufficient level of pericyte coverage. Platelets activated in the tumor vasculature could provide such an additional source of PDGFB. To test this hypothesis, we generated genetically modified mice that lack PDGFB specifically in the megakaryocyte/platelet lineage using the Cre/loxP system. These mice were further crossed to the transgenic RT2 mouse, which develops orthotopic pancreatic neuroendocrine tumors. Using this model, as well as two additional transplanted models for tumor growth, we address the consequences of lacking PDGFB in platelets for the vascular phenotype and malignant progression.

Animal work was approved by the local ethics committee (5.8.18-09777/2018) and performed according to the United Kingdom Coordinating Committee on Cancer Research guidelines for the welfare of animals in experimental neoplasia (22). Mice of different genotypes were generated and compared as littermates. Sample size (n) is given in each figure legend. Analyses were performed in a blinded fashion where relevant, including quantification of immunohistologic stainings, the angiogenic switch, and measurement of tumor volumes. Randomization was not applicable, as no treatments were performed. All mice used in the study were on C57BL/6 background. To deplete PDGFB specifically in platelets, Pf4-Cre mice (C57BL/6-Tg(Pf4-cre)Q3Rsko, The Jackson Laboratory; ref. 23) were crossed with Pdgfb^fl/wt mice (B6.129P2-Pdgfb^tm2Cbet, The Jackson Laboratory; ref. 8) to generate Pf4-Cre;Pdgfb^fl/wt mice, which were then bred with Pdgfb^fl/wt mice again to acquire the Pf4-Cre;Pdgfb^fl/fl mice. To obtain RT2-positive (01XD5, NCI-MMHCC) mice that lack PDGFB in platelets, Pdgfb^fl/fl mice were crossed with RT2 mice to generate RT2;Pdgfb^fl/wt mice, which were bred with Pf4-Cre;Pdgfb^fl/wt mice to acquire RT2;Pf4-Cre;Pdgfb^fl/fl mice. DNA extracted from tail biopsies was used for genotyping by PCR. Primers used for PCR are listed in the online Supplementary Materials and Methods. From 10 weeks of age, RT2-positive mice received drinking water supplied with 10% sucrose, to relieve hypoglycemia.

A complete list of antibodies used in this study, including concentrations, as well as primer sequences used for genotyping or qPCR, are included in the online Supplementary Materials and Methods.

Mice were depleted from platelets by an intraperitoneal injection with an antibody against Gp1bα (R300, Emfret Analytics). Platelet depletion was performed from 11 weeks of age with repeated injections every third day with a total six injections per mouse. Antibody dosage was 4 μg antibody/g body weight at first injection and 2 μg antibody/g body weight at subsequent injections.

Cryosections (5 μm) fixed in ice-cold methanol were used for all immunostainings if not specified. Fixation with ice-cold acetone was used for CD41 stainings of embryonic livers and for PDGFRβ stainings. For megakaryocyte staining (anti-CD41) of embryonic livers, IHC was performed using a TSA-Biotin Kit, according to the manufacturer's instructions (NEL700A001KT, Perkin Elmer). Slides were developed using 3-amino-9-ethylcarbazole solution, counterstained with hemalum solution, and mounted with Aquatex (VWR International). For megakaryocyte stainings (anti-CD41) of bone marrow, tibias were dissected, fixed in 3% paraformaldehyde (PFA) for 24 hours, and put in 10% EDTA/PBS solution (pH 7.2) for 14 days to decalcify the bones. The tissue was then embedded in OCT and sectioned.

Pericyte coverage was evaluated by quantification of the NG2/PDGFRβ/desmin-positive area and calculation of the ratio to the CD31-positive area (NG2 or PDGFRβ or desmin/CD31). The amount of extravasated fibrinogen (vascular leakage) was evaluated by quantification of the fibrinogen-positive area and subtraction of the CD31-positive area. Perivascular platelets were evaluated by quantification of the CD41-positive area within approximately 7 μm distance from the CD31-positive area.

Blood was sampled from the tail vein, left to coagulate overnight at 4°C, and centrifuged at 1,500 × g for 5 minutes to collect serum. ELISA was performed according to the instruction of the Quantikine ELISA Kit for mouse/rat PDGF-BB (MBB00, R&D Systems) and mouse/rat TGFβ1 (MB100B, R&D Systems). For PDGFB ELISA, mouse serum was diluted 1:10 in Calibrator Diluent RD6-3 supplied with the kit. For TGFβ1 ELISA, mouse serum was diluted 1:150 in Calibrator Diluent RD5-53 supplied with the kit.

Mice were anesthetized by intraperitoneal injection of 2% avertin. Lungs were dissected and digested with 5 mg/mL Collagenase II and 50 μg/mL DNase I (both from Sigma-Aldrich) in DMEM for 40 minutes at 37°C in a shaking incubator. Lung lysates were passed through a 70-μm cell strainer and erythrocyte lysis was performed on ice. The samples were blocked with Fc block (101302, BioLegend) and stained with anti-CD31-PE (553373, BD Biosciences) and anti-CD45-APC (103112, BioLegend) for 40 minutes. After washing with FACS sorting buffer (2% BSA/1 mmol/L EDTA/PBS), samples were stained with DAPI for live/dead discrimination. Endothelial cells (CD45⁻CD31⁺) were sorted using a FACSaria III (BD Biosciences), collected in PBS, centrifuged, and RNA was extracted using the NucleoSpin RNA Plus Kit (740984, Macherey-Nagel). cDNA was generated using the iScript cDNA Synthesis Kit (1708891, Bio-Rad) and KAPA SYBR FAST qPCR Kit (KK4608, Kapa Biosystems) was used for the PCR reaction. Primers are listed in online Supplementary Materials and Methods.

Lungs from C57BL/6 mice were cut into small pieces using a scalpel and added into a C-Tube (130-096-334, Miltenyi Biotec) with DMEM supplemented with collagenase (7 mg/mL, LS004176, Worthington) and DNase (1 μg/mL, D4513, MilliporeSigma). Lungs were then mechanically dissociated on a MACS Octo Dissociator (Miltenyi Biotec) using the m_LDK_1 program. Erythrocytes were lysed using Red Blood Cell Removal Solution (130-094-183, Miltenyi Biotec). Endothelial cell isolation was performed following the “isolation of endothelial cells from mouse lung” protocol from Miltenyi Biotec, using Mouse CD45 Microbeads (130-110-618, Miltenyi Biotec) for depletion of immune cells and Mouse CD31 Microbeads (130-097-418, Miltenyi Biotec) for CD31 enrichment, following the instructions provided by the manufacturer. RNA was extracted using the RNeasy Plus Micro Kit (74004, Qiagen). Eighty-four genes related to endothelial cell biology were analyzed with the RT² Profiler PCR Array for Mouse Endothelial Cell Biology (PAMM-015Z, Qiagen), using a Bio-Rad CFX96 instrument, according to instructions from the manufacturer. Normalization was performed against β2-microglobulin. Analyzed genes are listed in the online Supplementary Data (EC qPCR array).

Mice were anesthetized by intraperitoneal injection of 2% avertin. Blood was drawn by cardiac puncture and ACD buffer (38 mmol/L citric acid; 75 mmol/L trisodium citrate; and 100 mmol/L dextrose) was added to 40% of the total blood volume for anticoagulation. Platelet counts were analyzed using a Sysmex XP-300 Hematology Analyzer (Sysmex).

Mice were anesthetized by intraperitoneal injection of 2% avertin. Blood was drawn by cardiac puncture with ACD buffer as described under Platelet counts. ADP (10 μmol/L, A2754, MilliporeSigma) or 0.01 U/mL thrombin (HCT-0020, Haematologic Technologies) was used to activate platelets. Degree of platelet activation was measured as surface expression of GpIIb/IIIa (M023-2, Emfret Analytics) and P-selectin (M130-2, Emfret Analytics), detected by flow cytometry (FC500, Beckman Coulter) as described previously (24).

Mice were anesthetized by intraperitoneal injection of 2% avertin, and FITC-lectin (Lycopersicon Esculentum, FL-1171, Vector Laboratories) was administered by retro-orbital injection of 150 μL solution (0.5 mg/mL) and allowed to circulate for 2 minutes, before perfusion with 10 mL of 1× PBS (pH 7.4) and 10 mL of 2% PFA. Tumors and organs were stored in 30% sucrose at 4°C overnight, frozen in OCT Cryomount (45830, Histolab), and stored at −70°C. Cryosections (5 μm) of tumors and organs were stained with anti-CD31 to visualize the total vessel area. FITC-lectin perfusion was evaluated by quantifying the ratio of FITC-lectin/CD31-positive area.

Mice were anesthetized by intraperitoneal injection of 2% avertin and perfused with 10 mL of 1× PBS (pH 7.4) and 10 mL of 2% PFA. The pancreas was dissected from 7-week-old mice and angiogenic islets were counted in a stereo dissection microscope at ×16 magnification. Tumors were isolated from the dissected pancreas of 14-week-old mice and tumor volumes were calculated [(π/6) × width² × length].

The MC38 mouse colon carcinoma cells were obtained through Prof. Kristofer Rubin (Uppsala University, Uppsala, Sweden) and cultured in DMEM Glutamax (31966021, Invitrogen) containing 10% FBS (Euroclone) and 1% penicillin/streptomycin (SVA, Uppsala, Sweden) at 37°C and 5% CO₂. The cells went through maximum five passages after thawing and were screened for Mycoplasma using PCR (last date for testing: March 3, 2020; ref. 25). The cell line was not authenticated in our laboratory. MC38 cells (0.5 × 10⁶) in 100 μL PBS were injected subcutaneously in RT2-negative wild-type (WT) and pl-PDGFB–knockout (KO) mice and allowed to engraft for 18 days. Before sacrifice, mice were FITC-lectin perfused as described above. Tumors were dissected and frozen in OCT Cryomount (45830, Histolab).

HCmel12 melanoma cells were obtained from Prof. Thomas Tueting (University Hospital Magdeburg, Magdeburg, Germany; ref. 26). Cells were cultured in RPMI1640 medium (R0883, Invitrogen) containing 10% FBS (Euroclone), 1% penicillin/streptomycin (SVA), and 1% l-glutamine at 37°C and 5% CO₂, and regularly screened for Mycoplasma by LONZA MycoAlert Mycoplasma Detection Kit (LT07-701; last date for testing: May 15, 2019). The cell line was not authenticated in our laboratory. HCmel12 cells (0.5 × 10⁶ cells in 100 μL PBS) were injected subcutaneously into RT2-negative WT and pl-PDGFB–KO mice, and allowed to engraft for 24 days. Mice were sacrificed by cervical dislocation, and lungs were dissected and frozen in OCT Cryomount (45830, Histolab).

Spontaneous liver metastasis was analyzed in RT2 mice and spontaneous lung metastasis was analyzed in mice with subcutaneously injected HCmel12 melanoma cells. RT2 mice (14 weeks of age) were anesthetized with 2% avertin and perfused with 10 mL of 1× PBS and 10 mL 2% PFA. Mice with HCmel12 tumors were sacrificed by cervical dislocation 24 days after tumor cell injection. Tissues were frozen in OCT Cryomount (45830, Histolab) and metastases were quantified in 5 μm cryosections from livers (RT2 mice) and lungs (HCmel12 mice) stained with hematoxylin and eosin (H&E; 01820 and 01650, Histolab). Tissue sections from four different levels of the organ were analyzed with 200 μm between each level. The average number of metastases per level was calculated for each individual mouse.

Circulating tumor cells (CTC) were quantified in blood from RT2 mice, by qPCR for T-antigen and insulin (primer sequences in online Supplementary Materials and Methods). Blood was drawn by heart puncture (anticoagulated with ACD buffer as described under platelet counts) and mixed with RNA Later Solution (AM7020, Invitrogen). RNA was isolated from the samples using the RiboPure Blood Kit (Life Technologies/Ambion). The iScript cDNA Synthesis Kit (1708891, Bio-Rad) and KAPA SYBR FAST qPCR Kit (KK4608, Kapa Biosystems) were used for cDNA synthesis and PCR reaction.

B16 melanoma cells (ATCC) were cultured in DMEM Glutamax (31966021, Invitrogen) containing 10% FBS (Euroclone) and 1% penicillin/streptomycin (SVA) at 37°C and 5% CO₂. The cells were not authenticated after purchase, but retained melanin production and were regularly tested negative for Mycoplasma using PCR (last date for testing: March 3, 2020; ref. 25). B16 cells (0.5 × 10⁶ cells in 100 μL PBS) were injected in WT and pl-PDGFB–KO mice via the tail vein. Mice were sacrificed after 14 days and lungs were dissected. Metastatic analysis was performed by counting visible metastatic foci on the surface of the lung lobes.

Proximity ligation assay (PLA) was performed using the Duolink PLA Products (MilliporeSigma) according to a standard protocol at the PLA Proteomic Facility SciLifeLab (Uppsala University, Uppsala, Sweden). Briefly, 3.7% PFA-fixed tumor cryosections were permeabilized with 0.5% Triton-X 100 in PBS and blocked at 37°C during 1 hour in Duolink blocking solution with 200 μg/mL donkey anti-mouse IgG (715-005-150, Jackson ImmunoResearch). Cryosections were incubated at 4°C overnight with goat anti-mouse PDGFRβ and anti-mouse pY-100 primary antibodies, washed for 5 minutes for three times with TBST, followed by 1 hour incubation with Duolink PLA Secondary Probes (MilliporeSigma), anti-mouse PLUS (DUO92001), and anti-goat MINUS (DUO92006) at 37°C. Cryosections were washed in TBST and incubated for 30 minutes at 37°C in ligation solution. The PLA signal was amplified by rolling-circle amplification by using phi29 polymerase (27). Nuclei were detected by Hoechst. Cryosection slides were mounted for microscopy analysis by using SlowFade Gold Antifade Reagent (S36936, Invitrogen). Single primary antibody with PLA probes and PLA probes without primary antibodies were routinely included as negative controls. The phospho-PDGFRβ signaling was evaluated by quantifying the PLA signal–positive area. Costaining for NG2 was performed by adding the rabbit anti-mouse NG2 antibody together with the PLA primary antibodies and the anti-rabbit Alexa488 secondary together with the Duolink secondary probes.

Imaging of tissue sections stained by immunofluorescence or H&E was done using a Nikon Eclipse 90i Microscope and the NIS Elements 3.2 software. Images were analyzed using the ImageJ 1.45s Software (NIH, Bethesda, MD). Chromogenic slides were imaged with Leica DM LB Bright-Field Microscope and Olympus Digital Camera. Studio Lite software was used for image acquisition. Images of embryos were obtained with a dissecting microscope (model S6; Leica) connected to a Sony camera with Image-Pro version 3.0.1 software.

All statistical analyses in this study were performed using the nonparametric two-tailed Mann–Whitney test, because sample sizes are small (n < 50) and, therefore, difficult to test for normality. In this study, n always represents individual samples. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P < 0.0001. Exact P values are given in the respective figure legend.

We have previously reported that platelet depletion in RT2 mice with pancreatic neuroendocrine carcinomas resulted in an almost complete loss of NG2⁺ pericytes (16), and we first confirmed this finding (Fig. 1A). Because pericytes may show plasticity in marker expression, we addressed whether other pericyte markers were also lost. Immunostaining for the pericyte markers PDGFRβ (Fig. 1B) and desmin (Fig. 1C) in tumor tissue from control or platelet-depleted RT2 mice revealed that cells expressing these pericyte markers were also significantly reduced in tumor vessels after induction of thrombocytopenia. These results suggest that the abundance of differentiated pericytes in pancreatic tumors is indeed reduced following platelet depletion. This finding might reflect a failure of corecruitment of pericyte progenitors to newly formed tumor vessels, or that newly recruited pericytes die or dedifferentiate. Irrespective of these alternative mechanisms, RT2 tumor vessels fail to acquire a coat of differentiated pericytes after platelet depletion. In agreement with previous findings (18), we also detected increased leakage from tumor vessels after platelet depletion (Fig. 1D).

Activated platelets secrete PDGFB, a factor also expressed by endothelial cells and known to be essential for pericyte recruitment during developmental angiogenesis (6, 8). To investigate whether platelet-derived PDGFB mediates recruitment of pericytes to tumor vessels, we generated a conditional knockout of PDGFB in megakaryocytes and, consequently, platelets. Mice expressing the Cre-recombinase under the platelet factor 4 (PF4) promoter (Pf4-Cre), specific for the megakaryocyte lineage (23), were cross-bred to Pdgfb^fl/fl mice, containing loxP sites flanking exon 4 in the Pdgfb gene (8). Theoretically, this should result in functional inactivation of floxed Pdgfb alleles in all megakaryocytes and hence platelets (Fig. 2A). Different genotypes were detected with specific primers as described in Materials and Methods and generated PCR products of distinct sizes as illustrated in Fig. 2B. Mice lacking PDGFB in the megakaryocyte lineage (Pf4-Cre;Pdgfb^fl/fl) were viable and fertile. It was previously demonstrated that the floxed Pdgfb allele is functional (8), therefore, mice with genotypes Pdgfb^fl/fl, Pdgfb^fl/wt, Pdgfb^wt/wt or Pf4-Cre served as PDGFB WT controls.

Macroscopic examination of Pf4-Cre;Pdgfb^fl/wt and Pf4-Cre;Pdgfb^fl/fl embryos at different gestational stages (E14.5, E15.5, and E17.5) revealed no abnormal phenotype (Supplementary Fig. S1A–S1C). Also the number of viable Pf4-Cre;Pdgfb^fl/fl embryos were the same as the expected (Supplementary Table S1). Pdgfb mutants displayed the same number and distribution pattern of CD41-positive megakaryocytes (platelet precursors) in embryonic livers, an important site of hematopoiesis during development (Supplementary Fig. S1D–S1F). Equal number of bone marrow megakaryocytes were also detected by immunostaining for CD41 in tibias from adult mice (Supplementary Fig. S1G–S1I). These data suggest that platelet-specific knockout of PDGFB (pl-PDGFB KO) does not influence the number of megakaryocytes, either during development or in the adult mouse. Importantly, platelet counts in pl-PDGFB–KO and WT mice were similar (Supplementary Fig. S2A), as was the proportion of platelets among other cell types in peripheral blood (Supplementary Fig. S2B).

To address whether the Cre-loxP recombination in megakaryocytes was sufficiently efficient to remove PDGFB from platelet releasates, we analyzed PDGFB protein levels in serum samples from mice with different genotypes using ELISA. Pf4-Cre;Pdgfb^fl/fl mice had undetectable levels of PDGFB in serum in contrast to Pf4-Cre;Pdgfb^wt/wt and Pf4-Cre;Pdgfb^fl/wt, where the fl/wt mice displayed approximately half of the serum PDGFB concentration (3.2 ng/mL) compared with the wt/wt mice (5.3 ng/mL; Fig. 2C). These data demonstrate a functional Cre-loxP recombination, and hence removal of PDGFB from platelet granules. Interestingly, this result also shows that all detectable PDGFB in mouse serum is derived from platelets. To confirm that the conditional knockout did not affect endothelial PDGFB expression, CD31⁺ cells extracted by FACS from lungs derived from WT or pl-PDGFB–KO mice were analyzed for expression of PDGFB by qPCR. No difference in endothelial PDGFB expression was detected between WT and pl-PDGFB–KO mice (Fig. 2D).

To address whether platelet-specific deletion of PDGFB could have secondary effects on the vasculature, we analyzed the expression of 84 distinct genes related to endothelial cell biology by qPCR arrays in lung endothelial cells sorted from pl-PDGFB–KO and WT mice. For seven of the 84 genes, there was no detectable signal in one or more of the samples. For the remaining 77 genes, all samples generated a detectable signal and the analysis showed that there were no statistically significant differences in gene expression between pl-PDGFB–KO and WT mice (see online Supplementary Data: EC qPCR array).

To make sure that platelet activation was not compromised by the lack of PDGFB, we monitored P-selectin and fibrinogen receptor expression on the platelet surface after stimulation of whole blood with thrombin and ADP, respectively, using FACS. Platelets from Pf4-Cre;Pdgfb^fl/fl mice responded equally well to the activation stimulus as platelets from control mice with intact PDGFB levels in their platelets (Fig. 2E and F), demonstrating that lack of PDGFB does not affect the capacity of platelets to become activated.

To address whether genetic ablation of PDGFB in platelets has functional consequences for healthy vasculature, we analyzed perfusion, leakage, and pericyte coverage in kidney tissue from mice with or without PDGFB in their platelets (Fig. 3). Specifically, we selected kidney glomeruli for analysis because previous work has shown this structure to be particularly sensitive to the genetic loss of Pdgfb or its receptor-encoding gene Pdgfrb (5, 9, 10, 28). We could not find any differences between the genotypes with respect to vascular perfusion (Fig. 3A), fibrinogen extravasation (Fig. 3B), or the amount of NG2⁺ vessels (Fig. 3C). We, therefore, conclude that platelet-derived PDGFB is dispensable for pericyte recruitment and does not affect vascular function in a healthy tissue such as the kidney.

Mice with Pf4-Cre and floxed Pdgfb alleles were cross-bred with the RT2 mouse, which develops pancreatic neuroendocrine carcinoma through a series of well-defined steps (Supplementary Fig. S3A; ref. 29). From this breeding, a number of different genotypes were generated, as illustrated in Supplementary Fig S3B. In addition to the genotypes described in Fig. 2, transmission of the oncogenic RT2 allele was assayed by PCR, generating a 449-bp fragment. Because of the similarities in size between the RT2 and Cre PCR products, the presence of Cre was analyzed separately (Supplementary Fig. S3C).

A proportion of the hyperplastic and dysplastic lesions that develop in the insulin-producing islets of Langerhans in the pancreas of RT2 mice undergo an angiogenic switch with onset at around 7–8 weeks of age (29, 30). This means that the islets initiate active angiogenesis, which is largely governed by increasing hypoxia and elevated levels of VEGFA (31). The angiogenic islets can be detected under a microscope by their red color in the dissected pancreas. Quantification of the angiogenic switch revealed no significant difference between the two groups of mice (Supplementary Fig. S4A). The angiogenic islets subsequently develop into adenomas and ultimately carcinomas as the mice age. Analyzing the total tumor burden at week 14 of age, we could not find any difference in tumor volumes between WT and pl-PDGFB–KO mice (Supplementary Fig. S4B). However, in 8 of total 18 analyzed pl-PDGFB KO mice, the tumors were connected to blood-filled cysts of varying size, which was never observed in the WT mice (Supplementary Fig. S4C).

Immunostaining for platelets (CD41) revealed a clear increase in the number of adhering, and thus activated, platelets in tumor tissue compared with healthy kidney tissue, as expected (Supplementary Fig. S4D). No difference was detected between WT and pl-PDGFB–KO mice with respect to the amount of tumor adherent platelets, quantified either as total amount (Supplementary Fig. S4E) or platelets in close proximity to vessels (Supplementary Fig. S4D–S4F), further underscoring that there is no activation defect in platelets deficient for PDGFB.

To address whether genetic ablation of PDGFB in platelets affects the tumor vasculature, we analyzed perfusion, leakage, and pericyte coverage in tumor tissue from 14-week-old RT2 WT and pl-PDGFB–KO mice (Fig. 4). We found a significant reduction in FITC-lectin perfused tumor vessels in mice with PDGFB-deficient platelets compared with control mice (Fig. 4A). Moreover, there was a strong and significant increase in the amount of extravasated fibrinogen detected by immunostaining of tumor tissue from pl-PDGFB–KO mice compared with WT (Fig. 4B), indicating increased leakage and reduced functionality of the tumor vasculature. Immunostaining for NG2 to detect pericytes revealed a significant reduction of NG2 expression in the tumor vessels of mice that lacked PDGFB in their platelets (Fig. 4C). Because of putative plasticity of pericytes with respect to marker expression, we also performed immunostaining for desmin and PDGFRβ, additional pericyte markers. In agreement with the NG2 data, there was a significant reduction of vascular-associated cells with desmin (Fig. 4D) or PDGFRβ (Fig. 4E) immunoreactivity, indicating that pericyte recruitment to tumor vessels is compromised in the absence of platelet-derived PDGFB.

To determine whether the requirement of platelet-derived PDGFB for pericyte recruitment and vessel functionality is specific for the RT2 model, or whether this finding applies to other tumor types also, we performed subcutaneous injections of MC38 mouse colon carcinoma cells in RT2-negative WT and pl-PDGFB–KO mice and analyzed the vascular phenotype in established tumors (Supplementary Fig. S5). In agreement with the findings in the RT2 mice, FITC-lectin perfusion was reduced (Supplementary Fig. S5A) and fibrinogen leakage was enhanced (Supplementary Fig. S5B) in mice lacking PDGFB in their platelets. Moreover, pericyte coverage in the MC38 tumor vasculature was reduced in mice with PDGFB-deficient platelets, as judged by immunostaining for NG2 (Supplementary Fig. S5C) and desmin (Supplementary Fig. S5D). This finding indicates that platelet-derived PDGFB may be of general importance for pericyte recruitment to tumor vessels.

To address whether lack of platelet-derived PDGFB affects signaling through PDGFRβ in the tumor microenvironment, PDGFRβ phosphorylation in RT2 tumors from 14-week-old WT and pl-PDGFB–KO mice was analyzed by in situ PLA. Positive PLA signals, indicating phosphorylated PDGFRβ, were detected in the WT individuals, but strongly reduced in tumor tissue from pl-PDGFB–KO mice (Fig. 5A). Coimmunostaining for NG2 to detect pericytes showed that the PLA signal was mainly located to pericytes (Fig. 5B). Negative controls (lacking primary or secondary antibody, or using noncomplementary PLA primers) did not generate any signal. These data support the conclusion that PDGFB derived from platelets contributes to modulation of the microenvironment in tumor tissue.

To address whether the compromised integrity of tumor vessels in Pf4-Cre;Pdgfb^fl/fl mice affects dissemination of tumor cells, we analyzed WT (n = 23) and pl-PDGFB–KO (n = 17) RT2 mice for metastases in the liver using H&E staining. Not all RT2 mice develop metastasis during the time we are allowed to keep them because of ethical regulations. During this time, incidence of metastasis was not statistically different between WT (n = 11/23) and pl-PDGFB–KO (n = 11/18) mice. However, quantification of the number of metastases/liver section revealed a significant increase in the mice with PDGFB-deficient platelets compared with mice with intact PDGFB (Fig. 6A). To confirm the finding of increased spontaneous metastasis in pl-PDGFB–KO mice we used the HCmel12 melanoma model, which grows invasively in the skin when injected subcutaneously and forms spontaneous metastasis in the lung (26). Also, in this model there was a decrease in NG2⁺ pericytes in the primary tumor vasculature of pl-PDGFB–KO mice compared with WT (Supplementary Fig. S6A). Of 22 HCmel12-injected mice/group, 9 WT and 15 pl-PDGFB–KO mice developed lung metastasis after 24 days. Quantification of the number of metastases/lung section revealed a significant increase in pl-PDGFB–KO mice (Supplementary Fig. S6B).

Analysis of CTCs in WT and pl-PDGFB–KO RT2 mice using qPCR for both T-antigen and insulin, expressed by the tumor cells, showed a significantly higher level of CTCs in pl-PDGFB–KO mice (Fig. 6B). To further address the mechanism behind the increased metastasis in pl-PDGFB–KO mice, we performed intravenous injections of B16 melanoma cells in WT and pl-PDGFB–KO mice. When the tumor cells entered directly into the circulation, without having to intravasate, there was no significant difference between the two genotypes with respect to the number of experimental metastases that formed in the lungs (Fig. 6C). Taken together, these data indicate that lack of platelet-derived PDGFB promotes metastasis by enhanced entry of tumor cells into the circulation.

Enhanced intravasation of tumor cells could be regulated by different forces. One possibility is that the reduced pericyte investment occurs already at an early stage during tumor progression, which destabilizes the vascular integrity and enables easier passage for tumor cells into the circulation. To address this possibility, we analyzed pericyte coverage in pancreatic islets from 7-week-old WT and pl-PDGFB–KO RT2 mice. At this age, the pancreatic islets have not yet developed into carcinomas, but the angiogenic switch has been initiated (29, 30). Immunostainings for NG2 (Supplementary Fig. S7A) and desmin (Supplementary Fig. S7B) revealed a reduced pericyte coverage in the vasculature already at this early stage of tumor development. In agreement, increased vascular leakage was also detected at this stage (Supplementary Fig. S7C). Reduced vascular function can cause hypoxia, a known driver of epithelial-to-mesenchymal transition (EMT) and invasiveness (32, 33). To address the level of hypoxia in WT and pl-PDGFB–KO tumors from RT2 mice, we analyzed expression of the hypoxia-inducible transcriptional target, HIF1α, by qPCR. As illustrated in Fig. 7A, we found a significant increase in HIF1α expression in tumor tissue from pl-PDGFB–KO compared with WT mice. Immunostaining for another hypoxia-inducible molecule, BNIP3, whose expression is primarily driven by HIF1α (34), revealed elevated levels in RT2 tumor tissue from mice with PDGFB-deficient platelets (Fig. 7B). Moreover, we found increased expression of carbonic anhydrase 9 (CA9), induced by hypoxia, in HCmel12 tumors from pl-PDGFB–KO mice (Supplementary Fig. S6C).

To address potential differences in EMT between WT and pl-PDGFB KO during early-stage tumorigenesis, expression of the epithelial markers E-cadherin and ZO-1 was analyzed by immunostaining. We found a significant reduction in the amount of E-cadherin (Fig. 7C) and ZO-1 (Fig. 7D) in the pancreatic islets from 7-week-old RT2 mice with PDGFB-deficient platelets. Moreover, expression of the mesenchymal marker vimentin was increased in tumors from pl-PDGFB–KO mice (Fig. 7E). Collectively, these data indicate an accelerated EMT in tumors from pl-PDGFB–KO mice.

Platelets are a major source of TGFβ and platelet-derived TGFβ was previously identified as an important driver of tumor cell EMT and metastasis (35). To confirm that the level of TGFβ in platelets was not affected in the pl-PDGFB–KO mouse, we performed an ELISA for TGFβ in serum collected from WT and pl-PDGFB–KO mice. As expected, there was no difference between the mice with respect to TGFβ levels (Supplementary Fig. S7D).

On the basis of these data, we conclude that the enhanced metastasis seen in mice with PDGFB-deficient platelets is caused by reduced tumor vessel integrity and perfusion, elevated tumor hypoxia and EMT, eventually resulting in increased intravasation of tumor cells.

The growth factor, PDGFB, was originally isolated from platelets because of its abundance in this cell type (36), but its functional role in platelets has never been addressed using genetic models. In this study we have generated mice genetically deficient for PDGFB in megakaryocytes and hence platelets. These mice are viable and fertile with no obvious abnormalities. Pf4-Cre–mediated excision of PDGFB in platelets was confirmed by ELISA, showing undetectable levels of PDGFB in serum from Pf4-Cre;Pdgfb^fl/fl mice. The absence of PDGFB in serum from Pf4-Cre;Pdgfb^fl/fl mice also demonstrates that this growth factor is not freely circulating in the blood, but instead stored and released from platelets upon activation. In this way it is possible to create a high local increase in the concentration of PDGFB. The conditional deletion of PDGFB did not affect the number of bone marrow megakaryocytes, platelet counts, or the proportion of platelets in peripheral blood. This is in-line with a previous study using bone marrow chimeras, which demonstrated that lack of PDGFB expression in the hematopoietic compartment does not interfere with normal hematopoiesis (37). Importantly, platelet activation in response to ADP and thrombin, measured as expression of the fibrinogen receptor and P-selectin, was not affected by the lack of PDGFB.

Physiologically, platelets are activated upon injury because of exposure of subendothelial compartments containing TF, leading to thrombin generation, and collagen. This induces rapid arrest, conformational change, and degranulation of platelet intracellular stores. Platelet granules contain a large collection of growth factors and are a major source of VEGF and TGFβ in our bodies (38, 39). Tumors have acquired the capacity to activate platelets due to aberrant expression of TF and due to their discontinuous endothelium, exposing subendothelial spaces. As a consequence, tumors are continuously exposed to the plethora of growth factors released from platelet granules, illustrating the expression that “tumors are like wounds that do not heal” (40).

We found that tumor vessels in RT2 mice lacking platelet PDGFB displayed reduced vascular function and decreased coverage by NG2⁺, PDGFRβ⁺, and desmin⁺ mural cells, a finding that was confirmed in two additional models. These data indicate that platelet-derived PDGFB has a functional role in pericyte recruitment to the tumor vasculature. Release of PDGFB from activated platelets occurs in close proximity to the endothelium where the activating stimuli are present. Platelet-derived PDGFB should thus be present in a similar location as the endothelial-derived PDGFB, with the highest concentration retained around the endothelium (Fig. 7F). In the healthy noninjured tissue, arrest and activation of circulating platelets does not occur and consequently their granule content will not be released to contribute to processes in their microenvironment. This was also confirmed in our own immunostainings for platelets in healthy kidney compared with tumor tissue. In agreement, we found no difference between mice with intact or PDGFB-deficient platelets with respect to pericyte coverage or vascular function in kidney tissue from healthy mice.

Despite the reduced pericyte coverage in tumors from mice with PDGFB-deficient platelets, we did not observe any difference in tumor volumes between the genotypes in any of the three tumor models analyzed (RT2, MC38, and HCmel12). Interestingly, it has been demonstrated that blood vessels in healthy tissue can lose up to 50% of their pericytes and still retain a normal morphology (8). Even more severe pericyte deficiency is compatible with life, but this generates an abnormal vasculature both with respect to anatomy and function (8, 10). We did, indeed, detect a significant reduction in tumor vascular function in pl-PDGFB–KO mice. In addition, almost 50% of the analyzed tumors from RT2 mice with PDGFB-deficient platelets presented with blood-filled vacuoles, indicating the existence of an abnormal vascular phenotype in these mice. Previous studies of the effect of reduced pericyte coverage in tumor vasculature have been performed using the PDGFB retention motif–mutant (Pdgfb^ret/ret) mice, which display a systemic pericyte deficiency (10). In agreement with our findings, increased hemorrhage was detected in T241 fibrosarcoma subcutaneously implanted in Pdgfb^ret/ret mice, while no difference in tumor growth was detected compared with WT mice (11). Another study has shown that tumor growth in pericyte-deficient Pdgfb^ret/ret mice may be increased or unchanged, depending on the specific tumor type (41). Enhanced tumor growth in Pdgfb^ret/ret mice was connected to an accumulation of immunosuppressive myeloid-derived suppressor cells in the tumor tissue, caused by hypoxia-induced upregulation of IL6 (42).

Spontaneous metastasis was enhanced in RT2 and HCmel12 mice with PDGFB-deficient platelets, measured as nodules/tissue section, and we found an elevated level of CTCs in pl-PDGFB–KO RT2 mice. However, when tumor cells were injected directly in the circulation, no difference in experimental lung metastasis was found between WT and pl-PDGFB–KO mice. These data indicate that the increase in metastasis in pl-PDGFB–KO mice is because of increased intravasation of tumor cells. This finding could potentially be explained by a disrupted endothelial barrier in pericyte-deficient tumor vasculature in pl-PDGFB–KO mice, facilitating tumor cell passage. Pericyte-deficient vessels can also create a hypoxic microenvironment (42), which can drive EMT and invasive behavior of malignant cells (32, 33). A role for pericytes in limiting metastases in RT2 mice was previously reported by Xian and colleagues (43), showing that RT2 mice on the pericyte-deficient Pdgfb^ret/ret background display enhanced metastasis and leaky vasculature. In two studies using pericyte ablation, it was shown that pericytes can limit hypoxia-induced EMT and metastasis (44, 45). In human colorectal carcinoma, increased metastatic events were found to correlate to low degree of pericyte coverage in the primary tumor (46). We could indeed detect elevated expression of hypoxia-inducible targets such as HIF1α, BNIP3, and CA9 in tumor tissue from pl-PDGFB–KO mice compared with WT, as well as a more mesenchymal tumor phenotype, based on the expression levels of E-cadherin, ZO-1, and vimentin. These changes in the primary tumor likely explains the enhanced intravasation of tumor cells in pl-PDGFB–KO mice.

Collectively, our data suggest that specific platelet granule proteins like PDGF help to preserve vascular integrity in tumors, and most likely during wound healing, thus limiting metastasis. This is in contrast to the well-described overall prometastatic properties of platelets (47).

The Pf4-Cre mice contain an extra copy of the Cxcl3, Cxcl5, Cxcl7, and Cxcl15 genes (23). To rule out that increased expression of these chemokines plays a role in the effects reported in this study, we have analyzed the expression in tumor tissue from Pf4-Cre⁺ and Pf4-Cre⁻ RT2 mice. We could not detect any difference in expression levels of Cxcl3, Cxcl5, Cxcl7, and Cxcl15 between the two genotypes (Supplementary Fig. S8A–S8D).

The receptor for PDGFB on pericytes, PDGFRβ, has an additional ligand in the PDGF family of growth factors, namely PDGFD (48, 49). Similar to PDGFB, PDGFD is expressed by the endothelium (50, 51). In contrast to PDGFB, total knockout of PDGFD generates viable and fertile offspring, but with a mild vascular phenotype, including disorganized NG2⁺ pericytes in cardiac vessels (51). When crossed to the RT2 model, tumor growth was inhibited both in Pdgfd^+/− and Pdgfd^−/− mice, a finding that was attributed to lack of stimulation of a small subpopulation of PDGFRβ-positive tumor cells (52). Pericyte recruitment was, however, not impaired in RT2 tumors in PDGFD-deficient mice, which indicates a specific role of PDGFB in this context. It was speculated that the lack of a retention motif in PDGFD could enable this growth factor to travel further from the vasculature and therefore target other cell types (52). Whether PDGFD is present in platelets remains to be demonstrated, but on the basis of the above data, it is unlikely that PDGFD could compensate for the lack of PDGFB in pericyte recruitment.

Complete knockout of PDGFB in mice is embryonically lethal due to microvascular bleedings caused by lack of pericytes and vascular smooth muscle cells (6, 28). Conditional knockout of PDGFB under the control of the Tie 1 promotor generates viable and fertile offspring with a varying degree of pericyte coverage in the vasculature (8, 9). Tie 1 is expressed in endothelial cells, but its expression has also been reported in hematopoietic cells (53–55). If, or to what degree, PDGFB expression in megakaryocytes and platelets is affected in Tie 1–directed conditional knockout of PDGFB is currently unknown.

Despite the large collection of growth factors and other types of molecules stored in platelet granules, their individual roles in physiologic or pathologic situations are largely unexplored. One exception is TGFβ that was conditionally knocked out from platelets using the PF4 promoter (35). A study by Labelle and colleagues demonstrated that platelet-derived TGFβ plays a crucial role in promoting EMT and metastasis in cancer. It has been assumed that PDGFB secreted by activated platelets upon injury contributes to wound healing and tissue regeneration, but this remains to be demonstrated.

PDGFB secreted from activated platelets likely adds to the deposits of this growth factor in the ECM surrounding the vasculature and, thus, contributes to the vascular PDGFB gradient, required for proper recruitment of pericytes. This finding represents a novel mechanism describing how activated platelets can help to preserve the endothelial barrier in actively remodeling vasculature.

C. Betsholtz reports grants from Swedish Cancer Foundation, Swedish Science Council, and Knut and Alice Wallenberg Foundation during the conduct of the study. A.-K. Olsson reports grants from The Swedish Cancer Society, The Swedish Research Council, and Ruth and Nils-Erik Stenbäck Foundation during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

Y. Zhang: Conceptualization, formal analysis, supervision, investigation, visualization, methodology, writing-original draft, writing-review and editing. J. Cedervall: Conceptualization, formal analysis, supervision, investigation, visualization, writing-original draft, writing-review and editing. A. Hamidi: Formal analysis, investigation, visualization, writing-original draft. M. Herre: Formal analysis, investigation, visualization, writing-original draft. K. Viitaniemi: Formal analysis, investigation, visualization. G. D'Amico: Formal analysis, supervision, investigation, visualization. Z. Miao: Formal analysis, investigation. R. Valsala Madhavan Unnithan: Formal analysis, investigation. A. Vaccaro: Formal analysis, investigation. L. van Hooren: Formal analysis, investigation. M. Georganaki: Formal analysis, investigation. Å. Thulin: Formal analysis, investigation. Q. Qiao: Formal analysis, investigation. J. Andrae: Supervision, writing-review and editing. A. Siegbahn: Resources. C.-H. Heldin: Resources. K. Alitalo: Resources, supervision, methodology. C. Betsholtz: Supervision, methodology, writing-review and editing. A. Dimberg: Resources, supervision, writing-review and editing. A.-K. Olsson: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

The in situ PLA experiments were performed at the PLA Proteomics facility, which is supported by Swedish Science for Life Laboratory. This work was supported by grants from The Swedish Cancer Society (CAN 2017/522), The Swedish Research Council (2016-03036), and Ruth and Nils-Erik Stenbäck Foundation (all to A.-K. Olsson).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.