Podoplanin-expressing macrophages promote lymphangiogenesis and lymphoinvasion via Galectin 8-mediated adhesion to tumor lymphatics

Podoplanin-expressing macrophages (PoEMs) represent a perilymphatic TAM subset

Among all the cells of the immune system, we found that in orthotopic 4T1 breast tumors PDPN was almost exclusively expressed in TAMs but not in other tumor-infiltrating leukocytes, as assessed by FACS (Figure 1A). We also measured the transcript and protein levels of PDPN in different cell types, using freshly sorted tumor lymphatic endothelial cells (tLECs) as a reference. Strikingly, PDPN-expressing macrophages (PoEMs) had as much PDPN as tLECs (Figure 1B-1C). F4/80⁺ skin macrophages or circulating monocytes from these tumor-bearing mice were, respectively, scarcely positive or negative for PDPN (Figure S1A-S1B). In line, PoEMs were nearly absent in resected lymph nodes from both healthy and tumor-bearing mice (Figure S1C-S1D). These data suggest that PDPN expression becomes evident when monocytes differentiate into macrophages and experience some typical features of the tumor microenvironment such as hypoxia and/or TGFβ, which are known to promote Pdpn transcription (Suzuki et al., 2008; Tejchman et al., 2017).

Following, we characterized TAM heterogeneity and distribution within the tumoral space. PDPN⁺ TAMs (PoEMs) constituted about 30% of the entire TAM population infiltrating 4T1 breast tumors as shown by FACS (Figure 1D). Under histological assessment, we found that PoEMs were more proximal to the lymphatics than PDPN-negative TAMs (non-PoEMs) (Figure 1E-1G). Confocal imaging on thick sections from 4T1 breast tumors further confirmed that PoEMs localized in the proximity or adhered to the lymphatic vessel walls (Figure 1H). Next, we utilized the Rosa26.mTmG reporter line crossed with an inducible, macrophage specific Csf1r.iCre deleter, which allowed us to distinguish TAMs (EGFP⁺) from all the other stromal components (Tomato⁺). In the syngeneic E0771 orthotopic breast cancer model, we observed that round-shaped, PDPN⁺, EGFP⁺ TAMs were found adjacent to the LV walls (Figure 1I).

When characterizing the PoEM phenotype, we observed that MHC II^lo/CD206^hi or CD11c^-/CD204^hi F4/80⁺ cells (pro-tumoral, M2-like) were highly positive for PDPN, while MHC II^hi/CD206^lo or CD11c⁺/CD204^lo TAMs (anti-tumoral, M1-like) were PDPN-low or negative (Figure 1J-1K and Figure S1E-S1G) (Pucci et al., 2009). Besides, RNAseq analysis of PoEMs vs. PDPN-negative TAMs revealed more transcriptomic differences (accession number at the GEO database: GSE126722). Gene ontology (GO) enrichment analysis pointed at a highly significant upregulation (q value<0,005) of processes related to extracellular matrix (ECM) remodeling and reorganization in PoEMs. In contrast, GO terms related to growth factor and cytokine signalling did not show any significant changes (q value>0,005) (Figure S1H and not shown). Many matrix metalloproteinases (MMPs) and collagens, the main regulators of matrix organization and invasion (Barcus et al., 2017; Clark and Vignjevic, 2015) were found strongly upregulated in PoEMs vs. non-PoEMs (Figure 1L), while the majority of (lymph)angiogenic growth factors and cytokines did not show differential expression (Figure 1M). Overall, Vegfc and Vegfd (annotated as Figf), the two ligands of lymphangiogenic VEGFR3, were poorly expressed compared to Vegfa (Figure S1I). These data suggest that PoEMs represent a subset of TAMs potentially involved in matrix turnover and invasion.

Pdpn deletion in TAMs inhibits tumor lymphatic growth and metastasis and impairs perilymphatic localization of TAMs

To study the biological function of PDPN in TAMs, WT and Pdpn KO bone marrow (BM) cells were used to reconstitute the immune system of lethally irradiated WT Balb/c recipient mice, thus generating WT→WT and Pdpn KO→WT chimerias (Figure S2A). Six weeks after BM reconstitution, syngeneic 4T1 breast cancer cells were injected orthotopically in both WT→WT and Pdpn KO→WT mice. Despite comparable tumor growth, pulmonary metastases were 60% lower in KO→WT vs. WT→WT mice (Figure 2A-2E and Figure S2B-S2C). This was not caused by differential cancer cell extravasation and lodging to the lungs, as the number of pulmonary nodules following intravenous injection of 4T1 cancer cells was comparable in WT→WT and KO→WT mice (Figure S2D). Similar to human triple-negative breast cancer (TNBC), 4T1 cancer cells disseminate preferentially via the lymphatic route, and to a lower extent via blood vessels (Mohammed et al., 2011; Pulaski and Ostrand-Rosenberg, 2001; Ran et al., 2010). We therefore focused our attention on tumor LVs and lymph node metastasis. Density, total area and lumen size of lymphatics were strongly (up to 75%) reduced in KO→WT vs. WT→WT mice, as assessed by the usage of various lymphatic endothelium markers i.e., LYVE1, VEGFR3 and PROX1 (Figure 2F and Figure S2E-S2K). This impaired lymphangiogenesis was associated with decreased Evans Blue drainage from tumor to the inguinal lymph nodes (Figure S2L). Consistently, the number of CK14⁺ breast cancer cells in the draining lymph nodes was 75% lower in KO→WT vs. WT→WT mice (Figure 2G-2H). On the other hand, tumor blood vessel density and area, as well as pericyte coverage, permeability and tumor hypoxia (parameters linked to cancer cell intravasation into the blood stream) (Mazzone et al., 2009) were not affected by Pdpn deletion (Figure S2M-S2S).

In order to further compare the rates of hematogenous and lymphatic dissemination, we assessed lung metastasis in WT→WT and KO→WT mice wherein the draining LNs were resected (or not) prior to orthotopic 4T1 tumor implantation. Lymph node resection in WT→WT mice reduced lung metastasis by 55% but it did not result in any additional effect in the KO→WT mice, suggesting that tumor blood vessel intravasation and hematogenous metastasis stayed unaltered (Figure 2I-2J). Consistent with our observation that PoEMs are absent in LNs, the blood or lymph vessel density in the draining lymph nodes of WT→WT and KO→WT tumor-bearing mice were the same (Figure S2T-S2V). Moreover, the injection of 4T1 cells directly into the draining LNs resulted in comparable lung metastasis (Figure S2W).

Based on the finding that PDPN in TAMs is relevant for LV growth and lymphatic metastasis, we hypothesized that standard anti-lymphangiogenic therapy would be ineffective in Pdpn KO→WT tumor-bearing mice. In WT→WT mice, a VEGFR3-blocking antibody (clone mF431C1) (Laakkonen et al., 2007) reduced 4T1 tumor lymph vessel formation and metastasis by 50% and 40%, respectively (Figure 2K-2N). However, anti-VEGFR3 had no effect on tumor lymphangiogenesis and dissemination in KO→WT chimeras. Tumor growth was comparable in all the groups (Figure 2O). Altogether, these data indicate that PoEMs promote tumor lymphangiogenesis and lymphatic metastasis, likely favoring cancer cell entry into the lymphatic circulation.

As TAMs were the main population of leukocytes expressing PDPN (Figure 1A), we analyzed CD11b⁺, F4/80⁺ macrophages from 4T1 tumors (Figure S3A) and confirmed Pdpn deletion in KO→WT mice by both FACS and qRT-PCR (Figure S3B-S3C). Pdpn knockout did not influence overall TAM density (Figure S3D-S3E). However, the interaction of TAMs with LVs was strongly impaired in absence of PDPN (Figure 2P-2Q and Figure S3F). In contrast, TAM contact with tumor blood vessels did not change significantly in Pdpn WT vs. KO→WT mice (Figure S3G-S3H). Although PDPN was relevant for TAM attachment to the lymphatics, its deletion did not affect the expression of main M1 or M2 macrophage polarization markers, defining pro-inflammatory or pro-(lymph)angiogenic macrophage subsets, respectively (Figure S3I-S3L). Aligned, whole tumors and TAMs from Pdpn WT and KO chimeras had comparable expression of Vegfa, Vegfc, and Vegfd (Figure S3M-S3O). In line with the RNAseq data, while TAMs have high levels of Vegfa, they represent a smaller source of Vegfc and Vegfd as compared to the rest of the tumor.

In orthotopic EMT6.5 breast tumors, Pdpn KO TAMs also had impaired perilymphatic distribution, diminished LV area, and reduced metastatic burden, despite unaltered tumor growth (Figure S4A-S4H). Moreover, in mice lacking Pdpn specifically in TAMs upon Tamoxifen treatment (Figure 2R), orthotopic E0771 breast tumors displayed defective perilymphatic TAM localization (Figure 2S-2T and Figure S4I), as well as diminished lymphangiogenesis and metastasis, while tumor growth was comparable in all the groups (Figure 2U-2W and Figure S4J-S4O). Blood vessel density, pericyte coverage, tumor hypoxia and TAM infiltration were not altered by macrophage-specific Pdpn deletion (Figure S4P-S4U).

Finally, existing LVs in adult skin (Figure S5A-S6B), as well as pathological (lymph)angiogenesis in cauterized corneas (Figure S5C-S5F) were both comparable in WT→WT and KO→WT chimeras, consistent with the absence of PoEMs in these conditions (not shown). Overall, we found that PDPN expression is: i) the highest in pro-tumoral M2-like TAMs, ii) crucial for PoEM localization around the lymphatics, but iii) dispensable for determining their differentiation state into canonical M1 vs. M2 macrophages. By impairing the recruitment of a PDPN⁺ TAM subset to the perilymphatic space, tumor lymphangiogenesis in KO→WT mice is reduced, with consequent drop of lymphatic metastasis.

Perilymphatic TAM localization is mediated by GAL8 expression in LECs

Given the perilymphatic localization defect of macrophages, we focused our attention on a known PDPN interactor − galectin 8 (GAL8), a secreted glycan-binding protein abundantly expressed by LVs, modulating cell adhesion and migration, codified by the gene LGALS8 (Troncoso et al., 2014). First, we analyzed GAL8 expression in 4T1 tumor sections and found that this protein is expressed by tumor lymphatics and by some CD45⁺ infiltrating immune cells, but not by tumor blood vessels (Figure 3A-3C). In vitro, LECs defined as CD31⁺, PDPN⁺ cells (Figure S6A) expressed LGALS8 but did not express the most-studied PDPN ligand − CLEC2 (Figure S6B). In line with the in vivo distribution, LGALS8 expression in cultured LECs was 3,2-times higher than in BECs (Figure S6C), and not detected in 4T1 or E0771 breast cancer cells (Figure S6D).

Next, we validated GAL8 binding to Pdpn WT bone marrow-derived macrophages (BMDMs), also expressing PDPN at the RNA and protein levels (Figure S6E-S6F). GAL8 binding was impaired by 70% in Pdpn KO BMDMs (Figure 3D-3E). When measuring the migratory capacity of macrophages in a Transwell assay, WT BMDMs migrated efficiently towards recombinant GAL8, whereas Pdpn KO BMDMs showed almost complete lack of migratory response (Figure 3F-3G). Interestingly, in the WT condition, the migrated BMDMs were PDPN-positive, whereas the non-migrated ones were negative for PDPN (Figure 3H-3I). This finding was consistent with the observation that macrophages differentiated in vitro from BM cells constitute a mixed population, in which about 20% of cells is PDPN-positive and the remaining 80% is PDPN-negative (Figure S6E). Furthermore, WT but not KO BMDMs were able to migrate towards a monolayer of LECs (Figures 3J). Silencing of LGALS8 in LECs with three different siRNAs (or combinations of those) (Figure 3J-3L) or pharmacologic inhibition of GAL8 with thiodigalactoside (TDG) (Figure 3M) impaired the migration of WT BMDMs to the same extent as the knockout of Pdpn, but it did not further affect the migratory behavior of Pdpn KO macrophages. As a positive control, we show that the migratory capacity towards CCL2 was comparable for both WT and Pdpn KO BMDMs (Figure 3J). LGALS8 reconstitution in silenced LECs restored macrophage chemoattraction, excluding possible off-target effects (Figure 3N and S6G-S6H). In line, Pdpn deficiency strongly impaired the ex vivo migration of sorted TAMs towards GAL8 (Figure 3O-3P). Therefore, our data demonstrate that GAL8 expression by the LVs is important for macrophage migration to LECs.

In order to validate the significance of lymphatic vessel-derived GAL8 for TAM recruitment into the perilymphatic space, we generated a pseudo-specific knockout of GAL8 in tumor lymphatics. To this end, Lgals8 WT and KO recipient mice were reconstituted with BM cells from Csf1r.iCre x Pdpn^lox/lox mice (or from Csf1r.iCre x Pdpn^wt/wt controls) (Figure 4A). Since in tumors GAL8 was expressed by leukocytes and lymphatics only (Figure 3A-3C), this procedure restricted Lgals8 deficiency to LVs, whereas CD45⁺ immune cells still expressed GAL8 (Figure 4B). Genetic deletion of Lgals8 in LECs strongly decreased tumor lymphatic vessel area and the attachment of Pdpn WT TAMs to tumor lymphatics (Figure 4C-4E). This was accompanied by a reduction in lung metastasis (Figure 4F-4G). Conversely, Lgals8 deletion in the context of Pdpn KO TAMs did not further affect tumor LVs, perilymphatic TAM distribution or the metastatic burden observed in Pdpn KO chimeras. All experimental groups showed comparable tumor growth (Figure 4H).

Next, we utilized TDG in order to provide a proof-of-concept of pharmacologic GAL8 targeting, even though TDG is not a specific GAL8 inhibitor as its mode of action also restrains Galectin 1, Galectin 3, and Galectin 9 (Tribulatti et al., 2012). In this approach, we chose another breast cancer model - 4T1 (instead of E0771 tumors as in the abovementioned genetic targeting). Intratumoral injection of TDG (Ito et al., 2011) in WT→WT mice led to similar effects as those observed in the Lgals8 KO recipients, namely to decreased lymphangiogenesis, less TAM-LV interactions and reduced lung metastasis (Figure 4I-4N). In Pdpn KO→WT mice, TDG treatment did not additionally influence the lymphatic phenotype. Although galectins can affect blood vessel formation and CD8⁺ T cell activation (Ito et al., 2011; Markowska et al., 2010), tumor growth, tumor blood vessel density, or CD8⁺ T cell infiltration were not influenced by TDG treatment in our model (Figures 4O-4S). Altogether, these data argue that GAL8 in breast tumor lymphatics and PDPN in TAMs participate in the same molecular process and emerge as key factors of a previously unexplored metastatic cascade in breast cancer.

PoEM migration and adhesion depend on GAL8-mediated integrin β1 activation

GAL8 binding to PDPN was reported to favor clustering and activation of integrin β1 (CD29) in LECs (Chen et al., 2016). Given the adhesion defects observed in Pdpn KO BMDMs, we wondered whether similar effects occur on PoEM (macrophage) surfacewe. Indeed, we found that GAL8 stimulation induced integrin β1 activation in Pdpn WT but not in KO BMDMs (Figure 5A-5B) and that GAL8 binding to Itgb1-silenced BMDMs was decreased to the same level as in Pdpn KO BMDMs (Figure S6I-S6K). No additional impairment of GAL8 binding was observed upon silencing of integrin β1 in Pdpn KO BMDMs. Next, we found that PDPN and β1 integrin interaction was promoted by GAL8 stimulation in Pdpn WT but not in Pdpn KO BMDMs, as assessed by a proximity ligation assay (PLA) (Figure 5C-5D) and co-immunoprecipitation for PDPN and integrin β1 subunit (Figure S6L). In line with the fact that integrin β1 signaling is crucial for macrophage migration (Meng and Lowell, 1998), the deactivation of β1 on WT BMDMs (with a blocking antibody or with siRNAs) impaired their in vitro adhesion to LECs (Figure 5E-5F) or their migration towards GAL8 (Figure 5G). This effect was specifically dependent on integrin β1, but not on integrin β2 (Figure S6M-S6O). Altogether, these data support the idea that the binding of GAL8 to PDPN on macrophage surface promotes the formation of a multicomplex with β1 integrin and ultimately its activation − a prerequisite for macrophage migration and adhesion to LECs.

PDPN-mediated adhesion of PoEMs to LECs promotes lymphatic growth and cancer cell lymphoinvasion

To assess the functional relevance of our findings, we conducted a capillary network formation assay where Pdpn WT or KO BMDMs were seeded together with LECs. In these conditions, LECs cultured with Pdpn KO BMDMs formed simpler, less dense lymphatic networks in comparison to those observed in the presence of WT BMDMs (Figure S6P-S6Q). Then, using a 3D sprouting assay in a type I collagen gel, we demonstrated that BMDMs increase the number and length of sprouts observed in spheroids of LECs alone. Moreover, this lymphangiogenic effect was much stronger when the same assay was performed by adding PoEMs to the spheroid (instead of a total BMDM pool) (Figure 5H-5J). Neither Pdpn KO nor PDPN-negative BMDMs elicited any sprouting effect. In contrast, PoEMs, non-PoEMs, total Pdpn WT and KO BMDMs triggered a similar proliferative boost when co-cultured with LECs (Figure S6R); LEC apoptosis was scarce and not increased in any of the conditions tested (Figure S6S). Altogether, these data indicate that PDPN in PoEMs is important for sustaining lymphatic vessel sprouting but not for LEC proliferation and survival.

Macrophage proximity to blood vessels has been shown to modulate endothelial junctions and favor blood vessel intravasation of cancer cells through their direct interaction, resulting in the formation of TAM-cancer cell-BEC triads (Harney et al., 2015). We therefore set up a system where we evaluated how different sorted populations of F4/80⁺ BMDMs were remodeling VE-cadherin junctions of a confluent LEC monolayer. We found that the opening of the endothelial layer was much stronger in the presence of PDPN⁺ macrophages but not with PDPN^- or Pdpn KO macrophages (Figure 5K-5L). Aligned with these results, transendothelial migration of Calcein-labelled 4T1 cancer cells was promoted by PDPN⁺ but not by PDPN^- or Pdpn KO macrophages (Figure 5M). Pre-incubation of macrophages with the integrin β1 blocker prevented PoEM-induced formation of interendothelial gaps and 4T1 cancer cell transmigration (Figure 5K-5M). Thus, PoEM adhesion to the lymphatic endothelium is sufficient to induce gaps in a LEC monolayer and facilitate cancer cell transmigration.

In light of the above findings, we assessed the presence of triads at the lymphatic vessel site by implanting GFP-expressing 4T1 tumors into BM chimeras. In WT→WT mice, clusters of LECs, macrophages and a GFP+ cancer cells were frequently found, while their presence was reduced by 35 % in the KO→WT mice (even when normalized for strongly reduced LV area) (Figure 5N-5O and Figure S6T-S6U). Together, these data suggest that PoEMs at the LVs are instrumental for the promotion of lymphatic growth and cancer cell lymphoinvasion.

PDPN in PoEMs serves as an upstream regulator of matrix remodeling

Based on the initial RNAseq-based observations (Figure 1L and Figure S1H), we hypothesized that PDPN could be involved upstream in the transcriptional regulation of certain genes. We arbitrarily focused on the top 100 upregulated (and not downregulated) transcripts (FDR<0,05), taking into account the mediators of matrix composition and digestion i.e., collagens and MMPs, located at the right side of the volcano plot (Figure 1L). We identified a list of 4 MMPs (Mmp2, Mmp9, Mmp12, Mmp13) and 8 collagen subunit genes (Col1a1, Col1a2, Col3a1, Col5a2, Col5a3, Col5a1, Col6a2, Col6a3) which were highly upregulated in PoEMs vs. non-PoEMs. Using independent cohorts of mice, we confirmed that PoEMs had the highest expression levels of all these genes, while non-PoEMs, Pdpn WT and KO TAMs showed lower transcript values (Figure 6A-6M). Vegfa, Vegfc, and Vegfd were equally expressed in both PoEMs and non-PoEMs as well as in their total pool, and their expression did not change in Pdpn KO TAMs (Figure 6N-6P).

Lymphatic capillaries attach directly to interstitial matrix, mainly composed of fibrillar type I collagen (Detry et al., 2011). Interestingly, PoEMs, constituting a TAM subset proximal to tumor lymphatics, abundantly express Col1a1 and Col1a2 subunits. Consistently, bundles of type I collagen were richly deposited by ex vivo cultured PoEMs, and less by PDPN-negative TAMs and Pdpn KO TAMs, while total Pdpn WT TAMs displayed an intermediate collagen I deposition state, as assessed by stainings on the decellularized matrix (Figure 6Q-6R). Histologically, 4T1 tumor sections had thinner sleeves of type I collagen around LVs in KO→WT vs. WT→WT mice (Figure 6S-6T).

CTX-I are C-terminal telopeptides that are released during type I collagen digestion (Garnero et al., 2003). Consistent with the highest levels of Mmp2, Mmp9, Mmp12, Mmp13 (and other MMPs that were further down in the list of upregulated genes), PoEMs displayed the strongest collagen degradation activity (Figure 6U) and the highest gelatinase activity, the latter ascribed to MMP2, MMP9 and MMP12 (Chelluboina et al., 2015; Jacob et al., 2013; Li et al., 2012) (Figure 6V). Consistently, tumor LVs in the KO→WT mice had 60% less perilymphatic MMP12 accumulation than in WT→WT mice (Figure 6W-6X). In tumors, matrix remodeling and MMPs offer new paths for migration and invasion, but also allow the release of growth factors that otherwise would remain bound to matrix components, to the cell surface, or sequestered by growth-factor binding proteins (Bergers et al., 2000; Mott and Werb, 2004). In line, our next observations revealed that soluble vs. matrix-bound VEGFC and VEGFD ratios (main lymphangiogenic growth factors) were much lower in ex vivo cultured Pdpn KO TAMs than in WT TAMs (despite comparable protein levels in WT and KO TAM extracts, or in whole tumors from WT→WT and KO→WT mice) (Figure 6Y-6Z and Figure S7A-S7D). This suggests that the chemoattractant capacity of these growth factors (and possibly of many others) is reduced in KO→WT mice, as impaired matrix remodeling makes them poorly available in solution. Of note, TAMs constitute a very poor source of VEGFC as compared to other cellular components in 4T1 tumors (Figure S7E).

To assess if the transcriptional regulation of MMPs and type I collagens depended on GAL8, we analyzed PoEMs, PDPN-negative TAMs and total TAMs isolated from Lgals8 total KO mice, thus preventing any interaction between PDPN and GAL8 in vivo. Under these conditions we could still appreciate the upregulation of Mmp9, Mmp12, Mmp13, Col1a1 and Col1a2 in PoEMs vs. non-PoEMs, whereas the absence of GAL8 did not influence this expression pattern (Figure S7F-S7K). Lymphangiogenic genes i.e., Vegfa, Vegfc and Vegfd, were instead comparable in all sorted populations, suggesting that neither PDPN nor GAL8 was impacting on their transcription levels (Figure S7L-S7N). Overall, PDPN in TAMs appear to be not only instrumental for favoring the proximity of PoEMs to tumor LVs, but it also regulates the extracellular matrix (ECM) turnover by enhancing collagen formation and degradation. Hence, PoEMs promote lymphangiogenesis and lymphoinvasion both directly by remodeling the matrix in the proximity of growing tumor LVs, and indirectly by favoring growth factor liberation and accessibility.

Potential clinical relevance of PoEMs in breast cancer patients

Finally, we translated these findings to human breast cancer patients. Like in murine tumors, 30% of the TAMs expressed PDPN (as assessed in a small cohort of patients) (Figure 8A). Furthermore, freshly sorted human PoEMs, but not PDPN-negative TAMs, carried the ability to migrate towards soluble GAL8 (Figure 8B). Likewise,, human monocyte-derived PDPN⁺ macrophages had enhanced ability to migrate towards LECs, as compared to non-PoEMs (Figures 8C). This migratory capability was lost when LECs were silenced for LGALS8. Finally, we analyzed tumor samples from a unique cohort of patients with bilateral synchronous breast tumors, in which one tumor was lymph node positive (N+), while the contralateral tumor was lymph node negative (N0). Adhesion of PDPN⁺ TAMs to the lymphatic walls correlated with the presence of metastasized regional tumor LNs (p value range = 0,00037– 0,049). In patients with bilateral tumors that were both N0, we could not observe differences in the number of PDPN⁺ TAMs interacting with the lymphatic vessel walls (Figure 8D-8F). When analyzing the N+ subgroup, patients with both LN and organ metastasis (N+,M+) had significantly higher numbers of PoEMs per tumor lymphatic vessel than patients with lymph node metastasis but no organ metastasis (N+, M0) (Figure 8G). All together, we demonstrate that PoEMs are present in human breast cancer and that, due to their response to LEC-derived GAL8, they localize in direct proximity to tumor lymphatics. Our data represent a proof-of-concept of PoEMs as a potential prognostic marker in breast cancer patients.