Aspirin blocks formation of metastatic intravascular niches by inhibiting platelet-derived COX-1/thromboxane A2

Because metastasis is associated with the majority of cancer-related deaths, its prevention is a clinical aspiration. Prostanoids are a large family of bioactive lipids derived from the activity of cyclooxygenase-1 (COX-1) and COX-2. Aspirin impairs the biosynthesis of all prostanoids through the irreversible inhibition of both COX isoforms. Long-term administration of aspirin leads to reduced distant metastases in murine models and clinical trials, but the COX isoform, downstream prostanoid, and cell compartment responsible for this effect are yet to be determined. Here, we have shown that aspirin dramatically reduced lung metastasis through inhibition of COX-1 while the cancer cells remained intravascular and that inhibition of platelet COX-1 alone was sufficient to impair metastasis. Thromboxane A2 (TXA2) was the prostanoid product of COX-1 responsible for this antimetastatic effect. Inhibition of the COX-1/TXA2 pathway in platelets decreased aggregation of platelets on tumor cells, endothelial activation, tumor cell adhesion to the endothelium, and recruitment of metastasis-promoting monocytes/macrophages, and diminished the formation of a premetastatic niche. Thus, platelet-derived TXA2 orchestrates the generation of a favorable intravascular metastatic niche that promotes tumor cell seeding and identifies COX-1/TXA2 signaling as a target for the prevention of metastasis.


Introduction
Prostanoids are a family of bioactive lipids comprising prostaglandins (e.g., PGD 2 , PGE 2 , PGF 2a ), thromboxane A 2 (TXA 2 ), and prostacyclin (PGI 2 ). The rate-limiting step of prostanoid biosynthesis is catalyzed by cyclooxygenase (COX), an enzyme with 2 isoforms, COX-1 and COX-2. Both COX-1 and COX-2 have virtually identical enzymatic activity, mediating the conversion of arachidonic acid into PGG 2 and then into PGH 2 , the common precursor of all prostanoids (1). However, the spectrum of prostanoids synthesized by each isoform differs in vivo as a result of distinct expression patterns and functional coupling to prostanoid synthases in different cell types (2). For example, COX-2 is induced in endothelial cells and macrophages during inflammation and wound healing and couples with PGE 2 synthase in those cells to produce proinflammatory PGE 2 (2,3). In contrast, COX-1 is constitutively expressed. In platelets COX-1 couples with TXA 2 synthase (TXAS) to generate prothrombotic TXA 2 upon procoagulant stimuli (e.g., collagen, thrombin, and adenosine diphosphate [ADP]) (4)(5)(6). Because of the differential expression of prostanoid synthases and COX-1 and COX-2, the activity of the 2 isoforms is rarely redundant.
The importance of COX and prostanoid pathways in metastasis is apparent from reports showing that their inhibition greatly curtails metastasis. NSAIDs, including aspirin, that inhibit both COX-1 and COX-2 generally reduce metastasis in clinical studies and murine models (7)(8)(9). In some reports specific COX-2 inhibition blocks metastasis (10,11), but not in others (12). Looking at the downstream prostanoids, inhibition of TXA 2 or of PGE 2 synthesis also reduces metastasis in animal models, while PGI 2 has been reported to inhibit metastasis (11)(12)(13)(14)(15)(16)(17)(18), with some exceptions (12,19). These reports raise the question of whether some prostanoids might be suitable targets for metastasis prevention or therapy.
The possibility of using COX or prostanoid synthesis inhibition as a preventive strategy for metastasis has been highlighted by both clinical and experimental studies. Aspirin is given clinically in a variety of doses to reduce cardiovascular events or inflammation. Because of its unique combination of irreversible inhibition of COX enzymes and short circulating half-life, low-dose aspirin preferentially inhibits COX-1 in platelets, reducing the production of prothrombotic TXA 2 and other prostanoids (20). Thus, low-dose aspirin is given for prophylaxis of myocardial infarction and stroke. Higher doses of aspirin inhibit both COX isoforms in other tissues (21). In particular, the reduction of COX-2-derived PGE 2 exerts antiinflammatory effects. Case-control studies and meta-analysis of randomized controlled trials have shown that aspirin given for these unrelated purposes reduces metastatic cancer (22,23). This effect was significant over a range of primary tumor types, with a Because metastasis is associated with the majority of cancer-related deaths, its prevention is a clinical aspiration. Prostanoids are a large family of bioactive lipids derived from the activity of cyclooxygenase-1 (COX-1) and COX-2. Aspirin impairs the biosynthesis of all prostanoids through the irreversible inhibition of both COX isoforms. Long-term administration of aspirin leads to reduced distant metastases in murine models and clinical trials, but the COX isoform, downstream prostanoid, and cell compartment responsible for this effect are yet to be determined. Here, we have shown that aspirin dramatically reduced lung metastasis through inhibition of COX-1 while the cancer cells remained intravascular and that inhibition of platelet COX-1 alone was sufficient to impair metastasis. Thromboxane A 2 (TXA 2 ) was the prostanoid product of COX-1 responsible for this antimetastatic effect. Inhibition of the COX-1/TXA 2 pathway in platelets decreased aggregation of platelets on tumor cells, endothelial activation, tumor cell adhesion to the endothelium, and recruitment of metastasis-promoting monocytes/ macrophages, and diminished the formation of a premetastatic niche. Thus, platelet-derived TXA 2 orchestrates the generation of a favorable intravascular metastatic niche that promotes tumor cell seeding and identifies COX-1/TXA 2 signaling as a target for the prevention of metastasis.
Aspirin blocks formation of metastatic intravascular niches by inhibiting platelet-derived COX-1/ thromboxane A 2 they are crucial and which prostanoids might be responsible for their effects.
Here we demonstrate using a variety of different models that specific inhibition of COX-1 in platelets is sufficient to inhibit metastasis to the same extent as aspirin whereas inhibition of COX-2 does not reduce metastatic colonization. We further show that COX-1 blockade leading to inhibition of TXA 2 synthesis in platelets is sufficient to inhibit metastasis. Lastly we provide evidence that the antimetastatic effect of COX-1 inhibition is generally limited to the early stages of metastasis and that inhibition of COX-1 or of TXA 2 synthesis prevents the formation of an intravascular metastatic and premetastatic niche.

Results
Reduction of metastasis by aspirin correlates with the inhibition of thrombosis. We treated mice with different doses of aspirin (ASA; low, medium, and high), which were based on the low, medium, and high doses used in humans according to a body surface area dose conversion method and on previous literature (8,(36)(37)(38). Inhibition of COX-1 was evaluated using serum levels of TXB 2 , a stable metabolite of TXA 2 generated by platelet COX-1 activity during clotting (ex vivo) ( Figure 1A and ref. 39). Greater than 95% reduction in TXB 2 ex vivo is thought to indicate physiological inhibition of COX-1 (40). The medium and high doses, but not the low dose, of aspirin reduced TXB 2 by more than 95% ( Figure 1B) and, accordingly, reduced COX-1-dependent (arachidonic acid and U46619, a stable analog of TXA 2 ) agonist-induced platelet aggregation (Figure 1, C and D). COX-1-independent (ADP) platelet aggregation was not affected (Figure 1, C and D). Importantly, low-dose aspirin did not reduce serum TXB 2 more than 95% over 6 days after the treatment began, suggesting that the drug does not accumulate over time (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/JCI121985DS1). more prominent effect on adenocarcinomas (23). While aspirin also reduces metastasis in murine models (8,11,24,25), it is not generally established whether the reduction of metastasis can be attributed to inhibition of COX-1 or COX-2.
During hematogenous metastasis, tumor cells have adapted to co-opt diverse cell types, including platelets, endothelial cells, and immune cells, to survive the hostile environment of the blood circulation (26). In particular, cancer cells engage platelets to form small aggregates or clots on their surface through expression of tissue factor (TF) (27) or P-selectin ligands (28,29). Platelet aggregation and activation itself supports tumor cell survival in the circulation, either directly or through recruitment of myeloid cells, and can initiate cancer cell spreading on the surface of the blood vessel and transendothelial migration (27,(29)(30)(31)(32)(33). Different prostanoids are synthetized by a variety of cells in the vascular niche and are essential to maintain cardiovascular homeostasis, but the levels and source of prostanoids often change dramatically between physiological and pathological conditions (3,34,35). Thus, in addition to uncertainly as to which COX isoform is responsible for supporting metastasis, it is unclear in which cell compartment Diagram of COX-1 and COX-2 products. Serum TXB 2 represents COX-1 activity in platelets. Plasma PGE 2 represents COX-2 activity in monocytes. (B) TXB 2 in serum from C57BL/6 mice treated with vehicle or aspirin (ASA; low, medium, or high) for 2 days before blood sampling (n = 3). (C and D) Agonist-induced aggregation of CD61-stained platelets from mice treated with vehicle or aspirin for 2 days. Arachidonic acid, U46619, and ADP were the agonists (n = 7 for vehicle group, 4 for all other groups). (E and F) Experimental design (E) and ex vivo PGE 2 levels (F) in plasma from mice in B (n = 4). Data are represented as mean + SD (B and F), mean ± range (C). One-way ANOVA with Tukey's multiple-comparisons test. *0.01 < P ≤ 0.05; **0.001 < P ≤ 0.01; ***P ≤ 0.001. Spontaneous metastasis was also inhibited by aspirin. BALB/c mice with 4T1-GFPderived subcutaneous tumors received vehicle or aspirin treatment ( Figure 2E). Tumor growth was similar in both treatment groups, although aspirin treatment was associated with enhanced tumor regression (Supplemental Figure 4A). Aspirin decreased numbers of lung and liver metastases, of disseminated tumor cells in the lungs (Figure 2, F-I), and of circulating tumor cells (CTCs) (Supplemental Figure 4B) and the invasive ability of those CTCs (Supplemental Figure 4, C-E).
These data confirmed the inhibitory effect of aspirin on metastasis at doses that inhibit COX-1 activity and thrombosis, suggesting that aspirin affects metastasis establishment through an antithrombotic effect.
COX-1 inhibition is sufficient to reduce metastasis. Since aspirin inhibits both COX-1 and COX-2 at metastasis-suppressive doses, we determined the effect on metastasis of selective inhibitors of COX-1 (SC-560) or COX-2 (NS-398). Isoform specificity was confirmed by reduction of serum TXB 2 for COX-1 ( Figure 3A) and plasma PGE 2 for COX-2 ( Figure 3B and Supplemental Figure 2). COX-1 inhibition by SC-560 significantly reduced the number of metastatic lung nodules from B16F10 cells (Figure 3, C and D) compared to the medium and high doses of aspirin ( Figure 2B). COX-2 inhibition by NS-398 did not reduce the numbers (Figure 3, C and D), making it unlikely that metastatic seeding requires PGE 2 . However, NS-398-treated mice had smaller metastatic colonies (Figure 3, E and F), compatible with the reported involvement of COX-2 in tumor cell proliferation (15). Using other models, SC-560 also reduced experimental lung metastasis from MC-38-GFP, 4T1, and MDA-MB-231-CFP cells (Supplemental Figure 3) and spontaneous lung and liver metastasis, pulmonary dissemination ( Figure 3, G-J), and CTCs and their invasiveness from 4T1-GFP tumor-bearing mice (Supplemental Figure 4, B-E).
Fewer experimental metastases were generated by B16F10 cells in COX-1 -/-(Ptgs1 -/-, indicated here as COX-1 -/-) mice than in wild-type (COX-1 +/+ ) (Figure 4, A and B). As expected, COX-1 -/-Since COX-2 is not significantly expressed in blood cells in the absence of inflammation, we assayed COX-2 inhibition using plasma PGE 2 after COX-2 induction by LPS (Figure 1, A and E, and ref. 41). All doses of aspirin reduced plasma PGE 2 levels, demonstrating inhibition of COX-2 ( Figure 1F). Systemic PGE 2 metabolites (PGE 2 M) were also reduced (Supplemental Figure 2). The antiinflammatory effect of low-dose aspirin has been previously suggested (37,42). Thus aspirin inhibited COX-2 at all doses but only inhibited COX-1 with physiological significance at the medium and high doses. Hence, the medium dose is the minimum dose to achieve antithrombotic effects in our model, similar to low-dose aspirin in humans.
The effects of aspirin on experimental metastasis were assessed in mice treated with aspirin starting 2 days before the i.v. injection of syngeneic B16F10 melanoma tumor cells (Figure 2A). Aspirin at the medium and high doses reduced the number of metastatic lung nodules by more than 50% (Figure 2, B and C). The number of colonies inversely correlated with aspirin intake ( Figure  2D). Aspirin (medium dose) similarly reduced the number of metastatic lung nodules from MC-38-GFP, 4T1, and MDA-MB-231-CFP cells (Supplemental Figure 3), indicating a widespread inhibitory effect of aspirin on metastasis. was given from day -2 to day +1 (-2→+1) for the intravascular phase, from day +1 to day +4 (+1→+4) for the extravasation phase, or from day +4 to day +21 (+4→+21) for the extravascular phase ( Figure  5A). Aspirin and SC-560 given during the intravascular phase of metastasis resulted in reduced numbers of metastatic lung nodules similar to treatment throughout. In contrast, administration during the later extravasation and extravascular phases did not affect metastasis ( Figure 5, B and C). NS-398 did not change metastasis regardless of the schedule ( Figure 5D). Aspirin and SC-560, but not NS-398, reduced the number of tumor cells in the lung 24 hours after injection ( Figure 5, E and F). A similar reduction was obtained in COX-1 -/mice ( Figure 5, E and F). These findings suggest that COX-1 in host cells, rather than in tumor cells, is required for the pulmonary retention of tumor cells and the onset of metastasis. TXA 2 signaling driven by COX-1 is essential for metastasis. Circulating TXA 2 is the most abundant product of COX-1 in the circulatory system, mainly produced by activated platelets (4). To ask whether TXA 2 was a critical intermediary in the COX-1dependent development of metastasis, we administered picotamide (PICO), a dual inhibitor of TXAS and antagonist of TXA 2 receptor (TP), to mice ( Figure 6A) (44). This treatment reduced mice had decreased serum TXB 2 levels ( Figure 4C) and reduced platelet aggregation (Figure 4, D and E) compared with COX-1 +/+ mice. Taken together, these data indicate that the inhibition of COX-1 is sufficient to impair metastasis development. They further suggest that inhibition of COX-1 mimics the antimetastatic effect of aspirin. These results led us to ask (a) what phase of metastasis is affected by inhibition of COX-1; (b) which product of COX-1 enables metastasis; and (c) which COX-1-expressing cells are responsible for these effects.
Activity of COX-1 is required during the intravascular phase of metastasis. Disseminating tumor cells remain in the bloodstream for 1-4 days prior to extravasation, a time that varies depending on the model used (43). After tail vein injection, B16F10 cells were mainly intravascular after 24 hours and underwent extravasation between days 1 and 3 with the majority extravasated by day 4 (Supplemental Figure 5, A and B). The number of B16F10 cells rapidly decreased after injection, with only approximately 25% of the total adhered cells surviving in the lung vasculature 1 day after injection (Supplemental Figure 5C). Multicellular colonies were first noted on day 3, and micrometastases were seen on day 4 (Supplemental   Figure 7A), suggesting that aggregation was triggered by the tumor cells. Additionally, platelets neither aggregated nor associated with the lung vasculature of naive mice (Supplemental Figure 7B), excluding the possibility that platelet aggregation resulted from euthanasia and its accompanying decreased blood flow. Treatment with aspirin, SC-560, and picotamide, but not NS-398, decreased the number and the size of clots per tumor cell ( Figure 8, A-C). Similar results were obtained after coincubation of the 2 cell populations in vitro (Supplemental Figure 8, A-C). Pretreatment of platelets with aspirin and SC-560 diminished platelet aggregation on tumor cells, while pretreatment of tumor cells had no effect (Supplemental Figure 8, D-F). Additionally, COX-1 +/+ platelets, but not COX-1 -/platelets, B16F10 cells, nor primary lung microvascular endothelial cells (LMVECs) cells, generated TXB 2 either alone or in coculture ( Figure 8D). Thus, COX-1 in platelets associated with B16F10 cells is a major source of TXA 2 , and its inhibition affects platelet aggregation and thrombus expansion on tumor cells. We then asked whether COX-1 inhibition in reintroduced platelets would reduce metastasis. After platelet depletion by R300 antibody, platelets isolated from vehicle-, ASA-, SC-560-, NS-398-, or PICO-treated mice or COX-1 -/mice were infused after the antibody had dissipated but before platelets regained control levels (Supplemental Figure 9, A and B). Platelet depletion inhibited metastasis formation by B16F10 cells (Supplemental Figure 9C) (26,45). The reinfusion of COX-1 +/+ platelets in both COX-1 +/+ ( Figure  4, B-E). Similarly, the TP antagonist vapiprost reduced platelet aggregation but not plasma TXA 2 (Supplemental Figure 6, A-C), compatible with TP antagonism. Vapiprost impaired the early persistence of tumor cells in the lungs (Supplemental Figure 6, D and E), further supporting the notion that TXA 2 signaling is required for early phases of metastatic seeding.
We then supplemented TXA 2 using U46619, restoring basal levels of plasma TXA 2 /TXB 2 ( Figure 7A) and platelet aggregation (Figure 7, B and C) in aspirin-treated mice. This restored the numbers of persistent tumor cells in the lung to control values in aspirin-treated mice 1 day after injection (Figure 7, D and E) and of experimental metastases, even when U46619 was discontinued 1 day after tumor cell injection (Figure 7, F-H). These results suggest a central role for COX-1-derived TXA 2 in the inhibition of metastasis by aspirin.
TXA 2 synthesis by COX-1 in platelets is required for metastasis. TXA 2 is synthesized by activated platelets and is a potent agonist of platelet aggregation and a secondary mediator of thrombus expansion (4). Platelets aggregate on the surface of B16F10 cells through a TF-dependent mechanism (27). We asked whether inhibition of COX-1 leading to reduced TXA 2 levels results in decreased aggregation of platelets on tumor cells and reduced metastasis (8,  A and B) B16F10 metastatic lung nodules in COX-1 +/+ (C57BL/6) or COX-1 -/mice (n = 7), as in Figure 2A. (C) Concentration of serum TXB 2 from COX-1 +/+ or COX-1 -/mice (n = 4). (D and E) Agonist-induced aggregation of platelets from COX-1 +/+ or COX-1 -/mice (n = 8 and 4). Data are represented as mean + SD (A and C), median ± range (D). Unpaired t test, 2-tailed (A and C); 1-way ANOVA with Tukey's multiple-comparisons test (D). *0.01 < P ≤ 0.05; **0.001 < P ≤ 0.01; ***P ≤ 0.001. The COX-1/TXA 2 axis in platelets contributes to an intravascular metastatic niche. Microemboli are formed with tumor cells, platelets, and myeloid cells at sites of activated endothelium. The myeloid cells promote the survival of disseminating cells and their development into metastasis (27,29,32,48). Using Cx3cr1 gfp/+ mice to visualize monocytes and macrophages (27,49), we found that aspirin (medium and high doses) and SC-560, but not NS-398 and picotamide, reduced clustering of monocytes/macrophages around the intravascular tumor cells (Figure 9, A and B). The magnitude of monocyte recruitment correlated with the extent of the platelet clots ( Figure 9C). Treatment with aspirin, SC-560, and picotamide also reduced the extent of endothelial activation as indicated by E-selectin and VCAM-1 expression in vessels adjacent to platelet-tumor cell aggregates (Figure 9, D-F). Neither monocyte/macrophage recruitment nor endothelial activation was observed in naive mice (Supplemental Figure 7). Additionally, inhibition of COX-1/TXA 2 was associated with a larger diameter of lung vessels (Supplemental Figure 11, A-C), suggesting a decrease of vasoconstriction that might further prevent the accumulation of aggregates.
Analogous effects resulted from coinfusion of COX-1 -/platelets and B16F10 cells in COX-1 +/+ mice, with a decrease in platelet aggregation on tumor cells (Figure 10, A-C), association of tumor The COX-1/TXA 2 pathway contributes to a pulmonary premetastatic niche. The ability of disseminated tumor cells to colonize distant sites is enhanced by the systemic effects of a primary tumor, generating a premetastatic niche (50). To test the effect of inhibition of the COX-1/TXA 2 pathway on the establishment of a lung premetastatic niche, mice bearing B16F10 subcutaneous tumors were treated with aspirin and injected i.v. with tumor cells to induce lung metastasis before the occurrence of spontaneous metastasis ( Figure 11A). Aspirin treatment was started after the initiation of tumor growth and interrupted 2 days before tumor cell injection to avoid a direct effect of platelet inhibition on metastatic seeding. The increased numbers of meta-static lung nodules, indicative of the establishment of a premetastatic niche, were completely abrogated by treatment with aspirin ( Figure 11B). Aspirin did not affect the number of nodules in mice without subcutaneous tumors ( Figure 11B), further supporting the prometastatic role of intact COX-1/TXA 2 axis in platelets at the moment of tumor cell injection.
Lung preconditioning has been linked to the recruitment of myeloid cells with the support of the coagulation system (27,51,52). The numbers of Cx 3 CR1-GFP + monocytes/macrophages in the lungs of mice bearing tumors were greater than those in lungs of naive mice. Aspirin abolished this increase in monocytes/macrophages in the premetastatic lungs ( Figure 11, C and D) but did not affect the numbers in naive mice ( Figure 11C). Taking into account the role of platelets in the recruitment of myeloid cells (27,29,33) and the effect of the TXA 2 inhibitor picotamide on the establishment of spontaneous metastasis (Figure 6, J and K), together these specifically required, we tested clopidogrel, an antagonist of the P2Y12 ADP purinergic receptor, and eptifibatide, an inhibitor of α IIb β 3 integrin (also known as GPIIb/IIIa), both used clinically to reduce platelet aggregation (53)(54)(55)(56). Clopidogrel and eptifibatide significantly reduced ADP-induced platelet aggregation (Figure 12, A and B) without affecting plasmatic TXB 2 levels in vivo data suggest that the establishment of a lung premetastatic niche depends on the COX-1/TXA 2 pathway in platelets. TXA 2 signaling, not other platelet activation pathways, is required for the establishment of the intravascular metastatic niche. To understand whether platelet aggregation generally is critical for creating a metastatic niche or whether TXA 2 signaling is more , aspirin, or aspirin + U46619 (n = 6, 6, and 5) 1 day after injection (B16F10-CMFDA, white). Scale bars: 1 mm (black bar), 100 μm (white bar). (F) Experimental design of aspirin with or without U46619 treatment. Two days before B16F10 cell injection, mice were treated with vehicle, aspirin, or aspirin + U46619 for 3 weeks (ASA + U46619) or supplemented until 1 day after injection, followed by treatment with aspirin alone [ASA + (U46619 -2→+1)]. (G and H) B16F10 metastatic lung nodules in mice treated with vehicle, aspirin, aspirin + U46619, or aspirin + (U46619 -2→+1) (n = 9, 10, 9, and 8). Data are represented as mean + SD (A, D, and G), median ± range (B). One-way ANOVA with Tukey's multiple-comparisons test. *0.01 < P ≤ 0.05; **0.001 < P ≤ 0.01; ***P ≤ 0.001.

Discussion
In this paper we have provided evidence that aspirin reduces metastasis through the inhibition of platelet COX-1 and its product TXA 2 . Inhibition of COX-1 activity or TXA 2 signaling alone by pharmacological or genetic means was sufficient to reduce metastasis in a range of models. This novel finding directly implicates the activity of COX-1/TXA 2 in platelets before and during the intravascular transit of tumor cells, while it is not necessary for the persistent growth of the metastatic lung nodules. The inhibition of ( Figure 12C), compatible with a functional COX-1/TXA 2 pathway in platelets. Unlike aspirin, clopidogrel and eptifibatide did not affect the early persistence of B16F10 melanoma cells in the lungs (Figure 12, D and E), suggesting that TXA 2 signaling in the context of platelet aggregation is essential for the establishment of the early metastatic niche. All together our data describe a signaling network centered on platelet-derived TXA 2 that can be inhibited by aspirin treatment, leading to a reduced seeding efficiency and metastasis ( Figure 13). Aspirin has distinctive pharmacological properties at different doses, mainly derived from the differential inhibition of COXs in different body compartments. The antimetastatic effect of aspirin was seen at doses that inhibited COX-1/TXA 2 , whereas the inhibition of COX-2/PGE 2 alone was not sufficient, COX-1/TXA 2 in platelets impairs multiple consecutive steps of the hematogenous transit of tumor cells, leading to the reduction of tumor cells in the lung vasculature. Thus, COX-1 activity and TXA 2 production in platelets contribute to the generation of a permissive early metastatic niche ( Figure 13). monocytes/macrophages in proximity to the tumor cells (32,61). Monocyte chemoattractant protein-1 (CCL2/MCP-1) and CCL5 release by endothelial cells following TXA 2 signaling might amplify recruitment (29,64). Altogether, local release of TXA 2 leads to the formation of hematogenous microemboli with prometastatic properties. The recruitment of monocytes/macrophages was also reduced by aspirin at the level of the premetastatic niche, leading to reduced lung seeding. These data support the notion that cancerinduced thrombosis via the COX-1/TXA 2 pathway plays a central role in the conditioning of metastatic sites both before and after the arrival of CTCs (27). The inhibition of COX-2 decreases metastasis in some models (10, 11) but not others (12). In our experiments, NS-398 did not reduce seeding of B16F10-and 4T1-derived lung metastasis, but it decreased the size of metastatic lung nodules from B16F10 cells, consistent with COX-2 enhancing proliferation and immune evasion in experimental models (65,66). We noted that inhibition of COX-2 significantly inhibited metastasis by one colorectal cancer cell line, MC-38-GFP. Some colorectal cancers depend on COX-2 for progression (66), and we confirmed that MC-38 cells express much higher levels of COX-2 than B16F10 cells (S. Lucotti, unpublished observations). Thus, the sensitivity to COX-2 inhibition might be indicated by COX-2 expression in cancer cells (67).
In contrast, our data point to COX-1 inhibitors reducing metastasis through a microenvironment-centered mechanism. COX-1 inhibition was effective on cell lines regardless of their COX-1 expression (S. Lucotti, unpublished observations), and COX-1 -/mice had reduced metastasis, suggesting that aspirin has an antimetastatic effect independent of tumor cell expression of COX-1. Using Oncomine gene expression data we found that COX-1 expression in the primary tumor did not correlate with risk of metastatic cancer nor with the antimetastatic effect of aspirin suggesting a prominence of COX-1 rather than COX-2 in the metastatic process. We exclude COX-independent targets (57) since analogous results were obtained with inhibitors of other steps in the COX pathway and in COX-1 -/mice. The antimetastatic effect of aspirin was seen in low-and medium-dose trials (75-300 mg/d), and increased doses did not show additional benefit (23), consistent with platelet COX-1 as the main target for the antimetastatic effect of aspirin. To the best of our knowledge, COX-1 has been previously implicated only marginally in the development of metastasis (58). COX-1 can be expressed by a variety of cell types (21). Reinfusion of platelets in plateletdepleted mice only restored metastasis if the platelets contained active COX-1, showing that it is the platelet supply of COX-1 that is essential to metastasis.
Further, although COX-1 can generate a variety of active prostaglandins, the reduction of TXA 2 is responsible for the antimetastatic effect of aspirin. Infusion of a synthetic analog of TXA 2 restored the metastatic phenotype during aspirin treatment. Platelets aggregate on the surface of tumor cells and function as circulating reservoirs of TXA 2 . Autocrine TXA 2 signaling in platelets further enhances their aggregation on tumor cells, which supports metastasis (8,26,45). Additionally, paracrine TXA 2 signaling generates a favorable environment for tumor cell seeding through vascular constriction and induction of E-selectin and VCAM-1 through the TP receptor on endothelial cells (59,60). Cytokines released from intracellular granules of activated platelets also induce endothelial cell activation (61). Endothelial activation correlates with tumor cell survival within the lung vasculature (32), and E-selectin and VCAM-1 might facilitate tumor cell adherence to the endothelium directly (62) or via bound platelets (63). We demonstrated enhanced adhesion of tumor cells to an LMVEC monolayer in the presence of platelets, analogous to the results in vivo. Concomitantly, endothelial activation facilitates the homing and retention of metastasis-promoting However, aspirin significantly increases the risk of severe gastrointestinal symptoms and complications, especially over longterm use. Our data, together with previous clinical trials ( [71][72][73], suggest that selective TXA 2 inhibitors such as picotamide might present an alternative to target platelet TXA 2 while sparing other gastroprotective COX-1 products (i.e., PGI 2 ), and thus might be a safer therapeutic option for the prevention of metastatic disease.

Methods
Animals. C57BL/6 (C57BL/6J), BALB/c (BALB/cAnNCrl), and SCID (CB17/Icr-Prkdc scid /IcrIcoCrl) mice were purchased from Charles River Laboratories and Cx3cr1 gfp/+ mice (B6.129P-Cx3cr1 tm1Litt /J) from The Jackson Laboratory (49). COX-1 -/mice (74) were provided by TDW and JAM. Seven-to ten-week-old female mice were used for experiments involving drug treatment and/or tumor cell injection, while older naive mice with a C57BL/6 background were used for blood withdrawal and platelet isolation. Drugs were administered through drinking water, given ad libitum and changed every second day. Cell lines and staining. B16F10 murine melanoma cells (a gift from John L. Francis, Center for Thrombosis Research, Florida Hospital, Orlando, Florida, USA; ref. 75) were cultured in RPMI 1640 medium (Sigma-Aldrich), 4T1/4T1-GFP murine breast cancer cells, MC-38-GFP murine colorectal cancer cells, and MDA-MB-231-CFP human breast cancer cells (ATCC) were cultured in DMEM (Sigma-Aldrich) in a 5% CO 2 humidified atmosphere at 37°C. Media were supplemented with 10% heat-inactivated FBS (Gibco), 2 mM l-glutamine, 25 mM HEPES, 50 U/ml penicillin, and 5 μg/ml streptomycin (Ther-(Supplemental Figure 12). These data are consistent with the hypothesis that tumor cells with higher procoagulant activity are more sensitive to the antimetastatic effect of COX-1 inhibition (11,24). We showed that aspirin only at doses inhibiting COX-1 and specific COX-1 inhibitors blocked metastasis in 4 different metastasis models as well as spontaneous metastasis by 4T1. A limitation of our study, however, is that we only detailed the mechanisms for B16F10 melanoma cells. We could hypothesize that a similar inhibition of platelet COX-1 would occur in a range of models.
It is well established that platelet function greatly supports metastatic spread (68). Pharmacological or genetic inhibition of P2Y12 receptor and α IIb β 3 integrin has been previously found to be associated with reduced numbers of experimental metastasis (53,55,56). We failed to find any effect of the P2Y12 receptor antagonist clopidogrel and the α IIb β 3 integrin inhibitor eptifibatide on the early metastatic seeding of B16F10 cells. These findings suggest that alternative pathways of platelet aggregation are not required for the establishment of the intravascular metastatic niche, but they do not exclude the possibility that these pathways support later stages of metastasis such as epithelial-mesenchymal transition and extravasation (55,69).
Overall, this work identifies COX-1 and its product TXA 2 as potential pharmacological targets to inhibit the intravascular phase of metastasis, consistent with the use of low-to mediumdose aspirin as adjuvant therapy for cancer patients. The recently started phase III ADD-ASPIRIN Trial (70) will examine the effects of aspirin in the prevention of tumor relapse and metastasis. sterile water and delivered in saline at 0.5 mg/kg/d through i.p. injection (78). Isolation and staining of platelets. After sacrifice with an overdose of pentobarbital (665 mg/kg, i.p., or 332.5 mg/kg, i.v.), blood was collected from mice by cardiac puncture in syringes containing 3.2% (wt/ vol) sodium citrate (Thermo Fisher Scientific) or ACD buffer (83 mM Na 3 C 6 H 5 O 7 , 111 mM dextrose, 71 mM citric acid) (Sigma-Aldrich and Fisher Thermo Scientific), at 1:10 vol/vol ratio to blood.
Drug formulation for animal studies. Aspirin (ASA), purchased as dl-lysine acetylsalicylate (Aspégic injectable, Sanofi Aventis), was dissolved in sterile deionized water and resuspended in drinking water at 30 mg/l (low) (37), 180 mg/l (medium) (38), or 625 mg/l (high) (8). SC-560 (Cayman Chemical) dissolved in DMSO (Sigma-Aldrich) was resuspended at 24 mg/l in drinking water supplemented with 0.2% (vol/ vol) polyethylene glycol 200 (PEG200) and 0.01% (vol/vol) Tween-20 (both from Sigma-Aldrich) (77). NS-398 (Cayman Chemical) dissolved in DMSO was resuspended at 12 mg/l in drinking water supplemented with 0.9% wt/vol sodium chloride (Sigma-Aldrich) (77). Picotamide (PICO; Abcam) dissolved in 100% ethanol (Sigma-Aldrich) was resuspended in drinking water at 30 mg/l. U46619 (Cayman Chemical) was diluted in DMSO and delivered at 50 μg/kg through a 180-mg/l aspirin solution. Vapiprost (Vapiprost hydrochloride, Santa Cruz Biotechnology) dissolved in sterile water was resuspended in drinking water at 20 mg/l. All drinking water contained 2% wt/vol sucrose (Sigma-Aldrich). Clopidogrel [(±)-Clopidogrel (hydrochloride), Cayman Chemical] dissolved in 100% ethanol was delivered in saline at 10 mg/kg/d through i.p. injection (56). Eptifibatide (Sigma-Aldrich) was resuspended in Figure 13. Platelet-derived TXA 2 promotes metastasis by generating a permissive intravascular metastatic niche. Model of metastasis promotion by TXA 2 derived from COX-1 activity in platelets. The aggregation of platelets on tumor cells stimulates the aspirin-sensitive de novo synthesis of TXA 2 , which enhances the expansion of clots on tumor cells and leads to further TXA 2 synthesis. Concomitantly, TXA 2 -TP interaction induces the contraction of vascular smooth muscle cells, the activation of endothelial cells, and the recruitment of monocytes/macrophages to tumor cells, providing a permissive niche for metastasis seeding. Spontaneous metastasis assay. 5 × 10 5 4T1-GFP cells were injected s.c. in the right flank of female BALB/c mice. Tumors were measured using a digital caliper, and the volume was calculated as height × length × width × π/6. When the tumor reached 20-30 mm 3 , mice were randomly allocated to 4 treatment groups (vehicle, aspirin, SC-560, or picotamide). When tumor reached 800 mm 3 , lungs were perfused/ isolated for ex vivo imaging of the left lung. Lung and liver metastatic nodules were counted.
Isolation of CTCs and Transwell invasion assay. Blood from BAL-B/c mice was drawn in syringes containing ACD buffer (1:5 vol/ vol). Whole blood was diluted in an equal volume of PBS, layered on Ficoll-Paque PLUS media solution (GE Healthcare), and centrifuged at 400 g for 30 minutes at 19°C, according to the manufacturer's instructions. The mononuclear cell layer was isolated, and GFP + cells were counted as CTCs.
For Transwell migration and invasion assays, 5000 CTCs were resuspended in serum-free DMEM and seeded on a Transwell insert (8 μm pore size; BD Biosciences) coated with growth factor reduced Matrigel matrix (2 μg/μl; BD Biosciences). The bottom chamber contained DMEM supplemented with 2% FBS as chemoattractant. After 20 hours, cells adherent to the bottom well were fixed with 2% PFA, and GFP + cells were counted with a Celigo S Imaging Cytometer (Nexcelom Bioscience LLC).
Measurement of prostanoids. After 2 days of treatment with vehicle or drugs, blood was collected through the vena cava of terminally anesthetized C57BL/6 mice. For serum, blood was left to clot for 30 minutes at room temperature and centrifuged at 850 g for 15 minutes at 4°C. For plasma, blood collected with ACD buffer (1:10 vol/vol) was centrifuged at 1000 g for 15 minutes at 4°C. B16F10 cells (3 × 10 4 ) and/ or platelets (50 × 10 6 ) were coincubated for 24 hours, followed by centrifugation of conditioned medium at 1300 g for 10 minutes to remove detached cells and platelets. TXB 2 concentration was measured through a Thromboxane B 2 EIA kit (Cayman Chemical). 6-Keto-PGF 1α concentration was measured in anticoagulated plasma through a 6-keto-PGF 1α ELISA kit (Enzo Life Sciences).
For ex vivo PGE 2 , whole anticoagulated blood was incubated with 10 μg/ml LPS (Sigma-Aldrich) or saline for 24 hours at 37°C. Plasma was assayed with a PGE 2 ELISA kit (Abcam). For in vivo PGE 2 , C57BL/6 mice were injected with 5 mg/kg LPS or saline, and anticoagulated blood was collected through vena cava 4 hours after injection. Plasma (version 7.6.5). Depletion of the single-platelet population is representative of platelet aggregation, and can be visualized through the appearance of a comet tail of platelet aggregates (79).
Ex vivo platelet aggregation on tumor cells. CMFDA-stained B16F10 cells were seeded at 10 4 cells per well in collagen I-biocoated multichambers (BD Biosciences). The following day, 30 × 10 6 PKH26-stained platelets were added together with vehicle or drugs. After 2 hours, cells were fixed with 2% paraformaldehyde (PFA) in PBS and mounted with Vectashield mounting medium containing DAPI (Vector Laboratories). Tumor cells and platelets were pretreated with vehicle or drugs for 2 hours at 37°C or 30 minutes at 30°C, respectively, washed twice with PBS or washing buffer, and coincubated.
MRI scan of lungs. MRI on formalin-fixed lungs embedded in 4% agarose was performed at 4.7T or 7.0T (VNMRS, Agilent) using a 25-mm-inner-diameter quadrature birdcage coil (Rapid). T2-weighted fast spin echo 3D scan was acquired (echo spacing 9.35 ms, echo train length 8, effective echo time [TE] 37.41 ms, repetition time [TR] 200 ms) with a field of view (FOV) of 32 × 32 × 32 mm 3 to ensure complete coverage of the coil (and sample). Scan time was approximately 27 minutes per sample for an isotropic resolution of 125 μm. Ten samples were queued for unsupervised MRI measurement using an in-house-developed carriage system utilizing a stepper motor driven by an Arduino controller (http://www.arduino.cc). Tumor burden was quantified by manual segmentation via ImageJ (version 1.46r, NIH) and itk-SNAP software (version 3.6.0) (80).
Ex vivo whole lung imaging assay. CMAC-stained B16F10 cells (5 × 10 5 ) and PKH26-stained platelets (9 × 10 8 ) were injected into opposite tail veins of Cx3cr1 gfp/+ mice. After 8 or 24 hours, isolated lungs were placed in a specially designed chamber with a coverslip glass (0.16-0.19 mm thick) at its bottom. To visualize lung endothelium, anti-CD31-PE antibody (50 mg/kg; 102408, BioLegend) was injected in the vena cava 5 minutes before sacrifice, followed by vena cava ligation. Lungs were inflated with 0.5 ml of air and remained inflated during the imaging (27,31,81). Tumor cell extravasation was evaluated visually from microscopic FOV or through reconstruction of tumor cells and vessel surface with Imaris software (versions 8.2 and 9, Bitplane). The size of platelet microclots and monocyte/macrophage clusters was calculated through MAT-LAB (R2017a) code written in-house. Alexa Fluor 633) or a gallium arsenide phosphide (GaAsP) array (PE, Cx 3 CR1-GFP). Channels were acquired sequentially to minimize bleed-through of emitted light. Stacks of 15-40 slices at 1-to 2-μm intervals from random FOV or tile scans of whole left lung (×10 or ×20) were acquired. Statistics. Statistical analysis was performed with GraphPad Prism (version 5.02). D'Agostino and Pearson omnibus normality test was applied to assess data distribution. For normally distributed data, unpaired t test (2-tailed) or 1-way ANOVA with Tukey's test or Pearson's test was used. For non-normally distributed data, Mann-Whitney test, Kruskal-Wallis with Dunn's multiple-comparisons post hoc test, or Spearman's test was used. Outliers were identified through Grubbs's test (α = 0.05, GraphPad QuickCalc outlier calculator) and excluded. Differences were considered significant with a P value lower than 0.05.
Study approval. Animal procedures were in accordance with UK Animal law (Scientific Procedures Act 1986), including local ethics approval at the University of Oxford under project license 30/3413.