Effector lymphocyte-induced lymph node-like vasculature enables naïve T-cell entry into tumors and enhanced anti-tumor immunity

The presence of lymph node (LN)-like vasculature in tumors, characterized by expression of peripheral node addressin and chemokine CCL21, is correlated with T-cell infiltration and positive prognosis in breast cancer and melanoma patients. However, mechanisms controlling the development of LN-like vasculature and how it might contribute to a beneficial outcome for cancer patients are unknown. Here we demonstrate that LN-like vasculature is present in murine models of melanoma and lung carcinoma. It enables infiltration by naïve T-cells that significantly delay tumor outgrowth after intratumoral activation. Development of this vasculature is controlled by a mechanism involving effector CD8 T-cells and NK cells that secrete LTα3 and IFNγ. LN-like vasculature is also associated with organized aggregates of B-lymphocytes and gp38+ fibroblasts that resemble tertiary lymphoid organs that develop in models of chronic inflammation. These results establish LN-like vasculature as both a consequence of and key contributor to anti-tumor immunity.

L ymph nodes (LNs) contain specialized blood vessels called high endothelial venules (HEVs). HEVs display peripheral node addressin (PNAd) and CCL21, and mediate entry of naive and memory T cells expressing the cognate ligands L-selectin and CCR7 (ref. 1). HEVs are not normally found outside lymphoid tissue but are induced at sites of chronic inflammation 2 . They have been recently detected in human tumours and are associated with a positive prognosis [3][4][5][6] . This suggests that PNAd and CCL21 on tumour vasculature are important elements of immunological tumour control, but the mechanisms inducing their expression and their function in supporting anti-tumour immunity are unknown.
In peripheral LN, HEV morphology and adhesion molecule expression are maintained by dendritic cells (DCs) that express lymphotoxin (LT) a 1 b 2 , which acts via the LTb receptor (LTbR) on blood endothelial cells 7,8 . In inflamed non-lymphoid tissues, PNAd and CCL21 expression is often associated with the development of organized structures resembling LN termed tertiary lymphoid organs (TLOs). Control of PNAd in TLOs is thought to be similar to control in LN. Inhibiting LTbR signalling blocks PNAd expression in many TLO models [9][10][11][12] and DCs regulate the presence of PNAd þ vasculature and associated TLO in inflamed lungs 13,14 . PNAd þ vasculature can be induced by transgenic expression of LTa and LTb in the pancreas and kidney 15,16 , or by transgenic expression of CCL21 in the pancreas and thyroid via an LTbR-dependent pathway 17,18 . Similarly, transgenic expression of LTa or CCL21 in tumours leads to induction of PNAd þ vasculature [19][20][21] . However, these transgenic models do not allow one to determine the mechanisms regulating spontaneously arising PNAd þ vasculature. In non-transgenic tumour models, the density of intratumoral DCs 22 and regulatory T-cell (Treg) depletion 23 have been associated with the presence of LN-like vasculature, but the mechanisms controlling its development remain unknown.
Although it is generally assumed that tumour-infiltrating CD8 T cells are effector cells that differentiated in tumour-draining LN, we previously showed that naive T cells also infiltrate tumours 24 . Tumour-infiltrating naive T cells differentiate into functional effector cells in the tumour 24 and promote its destruction 25,26 . However, this work did not establish the mechanisms that supported naive T-cell entry. Here we investigated this using murine tumour models established in the absence of transgenic expression of chemokines or cytokines. We show that tumours spontaneously develop LN-like vasculature and identify novel molecular mechanisms, dependent on endogenous effector lymphocytes that drive its formation. We also demonstrate that LN-like vasculature is the major portal through which naive T cells enter tumours, and that infiltrating naive T cells are able to delay tumour outgrowth. These findings place intratumoral LN-like vasculature in a positive feedback loop that is both a consequence of and contributor to anti-tumour immunity.

Results
Tumours develop LN-like vasculature. Recent studies have identified LN-like vasculature in human tumours as a prognostic marker of enhanced patient survival [3][4][5][6] . Thus, we evaluated whether similar vessels developed in murine tumours. By immunofluorescence, we detected PNAd on CD31 þ endothelium in subcutaneous (s.c.) and intraperitoneal (i.p.) B16-OVA tumours in C57BL/6 mice (Fig. 1a-c; low-power images in Supplementary Fig. 1a,b). No staining was observed with isotype control antibody (Fig. 1c). PNAd was also expressed on vasculature of OVA-expressing Lewis lung carcinoma tumours (LLC-OVA) and B16 expressing a tyrosinase epitope as a model antigen (B16-AAD), in both s.c. and i.p. locations ( Fig. 1d-g). The fraction of PNAd þ vessels in tumours (B5-10%) was much smaller than in LN (Fig. 1h). PNAd detection on tumour vasculature also required tyramide amplification, while detection on LN HEV did not, indicating a significantly lower level of expression. In i.p. tumours, a fraction of PNAd þ endothelial cells exhibited the cuboidal morphology typical of LN HEV, with PNAd apparent at both the luminal and abluminal surfaces (Fig. 1i,j). Otherwise, PNAd was expressed on endothelial cells with a flat morphology, typical of the overall tumour vasculature (Fig. 1a,b). To verify that PNAd was expressed on the luminal surface, we injected MECA-79 antibody intravenously (i.v.) before tumour harvest. This labelled the majority of LN HEVs and tumour vessels, which in serial sections were PNAd þ , based on our standard staining protocol (Fig. 1k,l). No luminal staining was detected after injecting an isotype control antibody (unpublished). In both tumour sites, PNAd þ vessels coexpressed MAdCAM-1 and VCAM-1 ( Supplementary Fig. 1c-g). However, VCAM-1 was expressed more highly on PNAd-negative vessels. In i.p. but not in s.c. tumours, there were also many vessels that expressed MAdCAM-1 without PNAd (Supplementary Fig. 1e-g). PNAd þ vasculature was also present in genetically induced melanomas from BRAF V600E PTEN À / À transgenic mice 27 ( Supplementary  Fig. 2a).
We also detected messenger RNA for chemokines Ccl21 and Ccl19 in s.c. and i.p. tumour lysates, although at lower levels than in LN (Fig. 1m). B16-OVA cells cultured in vitro did not express detectable levels of either chemokine. CCL21 protein was displayed on the vasculature of all tumours evaluated ( Fig. 1n-p), but CCL19 protein was undetectable, probably reflecting the B10-fold lower expression levels (Fig. 1m). Staining serial sections established that CCL21 was coexpressed on PNAd þ vessels (Fig. 1q). Thus, multiple murine tumour models spontaneously develop LN-like vasculature expressing PNAd and CCL21.
LN-like vasculature enables naive T-cell infiltration. We previously showed that naive T cells directly infiltrated tumours 24 ; thus, we tested whether LN-like vasculature served as their entry point. We transferred naive Thy1.1 congenic OT-I cells into C57BL/6 mice with established s.c. or i.p. B16-OVA tumours and evaluated the absolute number of infiltrating OT-I cells after 1 or 18 h. This number was consistent across experiments and largely independent of tumour size (unpublished observations). We used 18 h as a standard time point for most of our experiments, in keeping with other studies [28][29][30][31] . However, infiltration was largely complete by 1 h (Fig. 2a). Co-transferred naive (CD44 lo ) polyclonal T cells infiltrated tumours equivalently to naive OT-I cells, demonstrating that infiltration was not dependent on antigen specificity (Fig. 2b). The efficiency of naive T-cell infiltration into tumours was lower than that into LN; however, it was consistent with the lower PNAd expression on tumour vasculature (Fig. 2c).
Disruption of chemokine signalling in transferred naive OT-I cells by pertussis toxin pretreatment prevented their infiltration into tumours (Fig. 2d), indicating entry is chemokine dependent. To directly determine whether naive T-cell infiltration required interactions with LN-like vasculature, OT-I cells were pretreated with either anti-L-selectin or anti-CCR7 blocking antibodies before adoptive transfer. T-cell entry into LN and both s.c. and i.p. tumours was significantly inhibited when either molecule was blocked ( Fig. 2e-g). CCR7 signalling activates integrin LFA-1 to cause arrest and extravasation via interactions with ICAM-1 (ref. 32). ICAM-1 was widely expressed on tumour vessels (unpublished observations) and pretreatment of naive OT-I cells with anti-CD11a to block LFA-1 also prevented infiltration into both LN and tumours ( Fig. 2e-g). To determine whether infiltration was mediated by L-selectin binding to MAdCAM-1 or PNAd, we transferred OT-I cells into i.p. tumour-bearing mice pretreated to block luminally expressed molecules with injection of either anti-MAdCAM-1, anti-PNAd or isotype control. MAdCAM-1 blockade reduced infiltration into Peyer's patches but had no effect on tumour infiltration (Fig. 2h) 25 . This did not affect overall LN vascularity (Fig. 3a,b), but significantly decreased PNAd expression on LN vessels and resulted in loss of HEV morphology (Fig. 3a,c). As expected, residual PNAd on flat endothelium in LTbR-blocked mice was abluminal, whereas PNAd on control HEV was displayed on both luminal and abluminal surfaces (Fig. 3a). In a parallel group of LTbR-Ig-treated mice that received naive OT-I T-cells, loss of luminal PNAd led to significantly reduced LN infiltration (Fig. 3d). However, LTbR-Ig blockade had no effect on PNAd expression on i.p. tumour vasculature ( Fig. 3e-g). Concordantly, naive OT-I cells transferred to a parallel cohort of LTbR-blocked animals entered i.p. tumours in greater numbers than controls, demonstrating that PNAd was still expressed on the luminal surface (Fig. 3h). LTbR-Ig treatment did not alter expression of Ccl21 in either LN or tumours (Fig. 3i,j). Thus, the mechanisms that control luminal PNAd on LN HEV and tumour vasculature are distinct. significantly decreased (Fig. 4a,b). However, naive T-cells infiltrated i.p. tumours grown in B-cell-deficient mMT À / À mice normally (Fig. 4c). Furthermore, infiltration into i.p. tumours in b2M À / À mice, which retain B cells and CD4 T cells but lack endogenous CD8 T cells and natural killer (NK) T cells, was poor ( Fig. 4b). However, infiltration into i.p. tumours in CD1d À / À mice was normal, ruling out a requirement for NKT cells and pinpointing CD8 T cells as necessary (Fig. 4d). A role for CD8 T cells in induction of LN-like vasculature was also suggested in genetically induced melanomas from BRAF V600E PTEN À / À transgenic mice. These tumours are well infiltrated by endogenous CD8 T cells 33 and the majority contained PNAd þ vasculature ( Supplementary Fig. 2b). In contrast, tumours from BRAF V600E PTEN À / À mice with an activating mutation in b-catenin 34 are devoid of infiltrating CD8 T cells 35 and none of these tumours contained PNAd þ vasculature ( Supplementary Fig. 2b).
To directly establish the importance of endogenous CD8 T cells, we reconstituted Rag2 À / À mice with total CD8 T cells purified from C57BL/6 mice before implantation of i.p. tumours. We then assessed development of LN-like vasculature directly by measuring expression of PNAd and CCL21, and indirectly by monitoring infiltration of transferred naive OT-I cells, in separate cohorts of mice (Fig. 4e). PNAd was completely absent from i.p. tumour vasculature in Rag2 À / À mice and expression of Ccl21 was significantly decreased (Fig. 4f-h). In i.p. tumours from CD8repleted Rag2 À / À mice, however, expression of PNAd and Ccl21, and infiltration of transferred naive OT-I cells were restored to levels seen in wild-type (WT) mice ( Fig. 4f-i). It seemed likely to be that this was directly mediated by CD8 effector T cells from the repleting pool that had infiltrated the tumour, because equivalent numbers of endogenous CD8 T cells were found in tumours in repleted Rag2 À / À mice and C57BL/6 mice (Fig. 4j).
Cohort (1)  (e) Experimental protocol used for the repletion of Rag2 À / À mice with CD8 T cells before tumour implantation. On day 14 after tumour implantation, tumours from one cohort of mice were harvested for immunofluorescence and reverse transcriptase-PCR. A second cohort of mice received naive OT-I cells and tumours were processed for flow cytometry to assess infiltration into tumours. Representative images (f) and summary data (g, n ¼ 3-4) for PNAd expression in tumours in the indicated mice. (h) Ccl21 expression (n ¼ 2) in i.p. tumour lysates was quantified as in In keeping with this, B16-F1 tumours, which lack a strong antigen, show reduced infiltration by endogenous CD8 T cells in C57BL/6 mice, compared with B16-OVA tumours (Fig. 4k,l). In addition, PNAd expression on B16-F1 tumour vasculature was significantly lower than on B16-OVA vasculature, but still greater than in B16-OVA tumours from Rag2 À / À mice (Fig. 4m). These data establish that PNAd and CCL21 expression on i.p. tumour vasculature is determined by the extent of infiltration of endogenous effector CD8 T cells, which is in turn related to the strength of the antigenic stimulus that activates them.
T and NK cells induce LN-like vasculature in s.c. tumours. In contrast to i.p. tumours, transferred naive OT-I cells infiltrated s.c. tumours in control and Rag2 À / À mice equivalently, whereas s.c. tumours in b2M À / À hosts continued to be poorly infiltrated (Fig. 5a). This suggested that NK cells, which are hyporesponsive in b2M À / À but not Rag2 À / À animals 36,37 , acted redundantly with endogenous CD8 T cells to induce LN-like vasculature in s.c. tumours. Indeed, although depletion of NK cells from WT mice had no effect, naive T-cell infiltration into s.c. tumours was significantly diminished when NK cells were depleted from Rag2 À / À mice and restored in NK-depleted Rag2 À / À mice reconstituted with CD8 T cells (Fig. 5b,c). In addition, PNAd and Ccl21 expression was significantly decreased in s.c. tumours from NK-depleted Rag2 À / À mice (Fig. 5d-f). Therefore, NK cells act redundantly with endogenous CD8 T cells to induce LN-like vasculature in s.c. tumours.
Cytokine production differs between tumour sites. Equivalent numbers of NK cells and CD8 T cells were present in both i.p. and s.c. tumours, demonstrating that NK induction of LN-like vasculature in s.c. but not i.p. tumours was not due to differential representation (Fig. 5g). To assess the effector activities of these cells, we treated tumour-bearing mice with Brefeldin A for 4 h and analysed the in vivo cytokine profiles immediately after harvest. About 20% of CD8 T cells in both tumour locations produced interferon-g (IFNg) (Fig. 5h). However, although B30% of NK cells in s.c. tumours produced IFNg, o5% of NK cells in i.p. tumours did so (Fig. 5h). In addition, significantly larger fractions of NK cells and CD8 T cells produced tumour  ARTICLE necrosis factor-a (TNFa) in s.c. tumours than in i.p. tumours (Fig. 5i). Thus, NK and effector CD8 T-cell functional activities depend on the location of tumour growth and the involvement of NK cells in inducing LN-like vasculature correlates with their secretion of IFNg.
Tumour-infiltrating naive T cells delay tumour outgrowth. We next determined whether naive T cells that infiltrated tumours via LN-like vasculature exerted a positive or negative effect after differentiating in the tumour mass. We transferred naive OT-I cells into C57BL/6 or TNFR1/2 À / À mice bearing early (day 8) s.c. B16-OVA tumours and monitored outgrowth and survival. At the time of naive T-cell transfer, we also initiated daily treatment with the sphingosine 1-phosphate receptor modulator FTY720,     (Fig. 9a). However, CD44expressing OT-I cells in the antigen-free mesenteric LN, which redistribute from the tumour-draining LN 42 , were reduced by over 80% by FTY720 treatment (Fig. 9a). Despite sequestration of LN-activated effector T cells, naive OT-I cells that entered tumours in WT mice via LN-like vasculature significantly delayed tumour outgrowth and prolonged survival (Fig. 9b,c). In contrast, naive OT-I cells had no effect on tumours in TNFR1/2 À / À hosts, which lack LN-like vasculature (Fig. 9b,d). Thus, naive CD8 T cells that infiltrate tumours via LN-like vasculature enhance anti-tumour immunity. This is consistent with our previous demonstration that they become fully differentiated intratumoral effector cells 24 .
Organized lymphoid tissue develops in i.p. tumours. As PNAd and CCL21 expression in inflamed tissues has been associated with formation of TLO, we determined whether this also occurred in tumours. In i.p. tumours, we found large follicular aggregates of B lymphocytes immediately surrounding some but not all sites of PNAd expression (Fig. 10a). These aggregates were co-extensive with a reticular network of gp38 þ cells that did not stain for the LEC marker LYVE-1 (Fig. 10a), suggesting that they represented the gp38 þ CD31 neg CCL21 þ FRC-like cells identified by flow cytometry (Fig. 7a). T cells and DCs were also found in these structures, but they were not organized into a discreet zone (Fig. 10b,c), as has been seen in some TLO. Nonetheless, in conjunction with our earlier work 24 , these results suggest that organized lymphoid tissue associated with LN-like vasculature may serve as sites for the activation and regulation of recently entering naive T cells.

Discussion
Our observation of LN-like vasculature in multiple tumour models adds to a growing body of literature demonstrating its presence in mouse 21,23 and human 4,6,43 tumours. The presence of PNAd-and CCL21-positive LN-like vasculature in human tumours has been associated with a positive prognosis. We also previously showed that naive T cells enter tumours and differentiate into effectors 24 . Here we link these two observations, demonstrating that PNAd and CCL21 control naive T-cell entry into tumours. In contrast to LN HEV and many TLO models, development of intratumoral LN-like vasculature did not depend on LTbR signalling. Instead, it depended on other cytokines secreted by tumour-infiltrating CD8 or NK effector lymphocytes. PNAd on tumour endothelium was induced by LTa 3 signalling through TNFRs. CCL21 expression by endothelial cells and associated gp38 þ fibroblasts was induced by IFNg downstream of TNFR signalling in i.p. tumours. However, its control in s.c tumours is more complex. It probably involves IFNg, which is made by both CD8 T cells and NK cells in these tumours and redundant factor(s) that remain to be identified. Interestingly, effector functions of intratumoral NK cells also varied with location of tumour growth, altering their ability to induce PNAd and CCL21 expression. Our results support a model in which an initial influx of LN-primed effector lymphocytes into the tumour induces the development of Control of PNAd expression on LN HEV and in TLO models has been primarily attributed to DCs that express LTa 1 b 2 , which engages LTbR 7,8,13,14 . PNAd expression in human breast cancer has also been correlated with the presence of LTb-producing DCs 22 . Other reports have also demonstrated a role for B cells in inflamed LN 44,45 and an indirect role for CD4 T cells by initiating DC-mediated induction of PNAd in a thyroiditis TLO model 18 . However, our results with mice deficient in effector lymphocyte populations clearly demonstrate that the simple presence of DCs is insufficient for inducing PNAd expression in B16 tumours. Similarly, neither B cells nor CD4 T cells are necessary or sufficient for inducing PNAd in this tumour model system. Instead, PNAd expression required endogenous effector CD8 T cells in i.p. tumours and either CD8 T cells or NK cells in s.c. tumours. Effector lymphocytes acted directly to induce PNAd by secreting LTa 3 , which engaged TNFRs on endothelial cells. The expression of TNFR1 rather than TNFR2 on endothelial cells in B16 tumours suggests that signalling through this receptor is responsible.
Another recent report showed PNAd expression in methylcholanthrene-induced tumours required Treg depletion, although the molecular mechanism controlling PNAd expression was not defined 23 . However, in B16 tumours growing in WT mice and in genetically induced melanomas, PNAd is expressed despite the presence of Treg cells. Taken together, we suggest that the role of Tregs in regulating PNAd is indirect, by limiting the accumulation and/or effector activities, including LTa 3 secretion, of activated lymphocytes. The fact that Treg depletion was not required for PNAd expression in these melanoma models probably reflects in part the differences in antigenic strength between ovalbumin and neoantigens formed in genetically induced melanomas or methylcholanthrene-induced fibrosarcomas, tipping the balance towards more robust effector activity. Consistent with this, weakly antigenic B16-F1 tumours had reduced accumulation of effector lymphocytes and lower levels of PNAd expression than B16-OVA tumours.
At the molecular level, LTbR signalling is critical for regulating PNAd expression on LN HEV 8 and in several TLO models [9][10][11][12]46 . Previous work by Ruddle and colleagues 15,16,47,48 also dissected the roles of LTa and LTb, using a model of transgenic overexpression in the pancreas and kidney. In those studies, a primary role for LTbR signalling was also demonstrated, as PNAd þ vasculature was not seen when the LTa transgene was expressed in the absence of LTb 15 . Abluminal PNAd was present when only the LTa transgene was expressed in mice with endogenous LTb, whereas luminal PNAd was dependent on transgenic expression of both LTa and LTb 16 . In other circumstances, however, PNAd expression can be LTbR independent. For example, when LTbR is deleted from endothelial cells, PNAd remains present on cells that have otherwise lost the typical HEV morphology 49 . PNAd is also expressed on the abluminal surface of LN HEV in LTb knockout animals 16 or mice treated with LTbR-Ig 8,45 . A role for LTa 3 -TNFR signalling in controlling this residual abluminal PNAd has been suggested 16 but not directly demonstrated. Here we have shown that PNAd expression on tumour-associated LN-like vasculature is LTbR independent and LTa 3 -TNFR dependent.
Other characteristics of LN-like vasculature in tumours differ from LN HEV or other TLO models. First, PNAd expression levels on tumour vasculature are low compared with LN HEV, and PNAd in tumours was largely expressed on flat endothelium. Importantly, LTa 3 -induced PNAd on tumour vasculature was clearly luminal, enabling L-selectin-mediated naive T-cell infiltration that could be blocked by i.v. administration of PNAd antibody. LN-like vasculature co-expressed MAdCAM-1, which is found on HEV in mesenteric LN and is also induced by LTa 3 (refs 15,47). MAdCAM-1 did not contribute to naive T-cell infiltration, however, perhaps because it is not properly posttranslationally modified to serve as an L-selectin ligand. A critical step in the generation of PNAd is the sulfation of core glycoproteins by the sulfotransferases GlcNAc6ST-1 and GlcNAc6ST-2 (ref. 50). GlcNAc6ST-2 is expressed specifically in HEV and is primarily responsible for generating luminal PNAd 51 . GlcNAc6ST-1 is expressed more widely and contributes to both luminal and abluminal PNAd [52][53][54] . Our finding of lowlevel expression of PNAd on both luminal and abluminal surfaces suggest the induction of GlcNAc6ST-1 rather than GlcNAc6ST-2, by LTa 3 -TNFR signalling. In any case, molecular control of PNAd expression on tumour vasculature appears distinct from what has been reported in other systems, perhaps as a consequence of disregulated angiogenesis.
A second component of LN HEV expressed on tumour vasculature was the chemokine CCL21. CCL21 expression has also been reported in human tumours 3,4,6 , but similar to PNAd its regulation and role in the tumour microenvironment is poorly understood. CCL21 expression depends on TNFa and LT in the spleen 55,56 and transgenic TLO models 16,57 , but its expression in adult LN is independent of these cytokines 8,45 , and no alternative regulatory mechanisms have been defined. In tumours CCL21 expression was controlled by effector lymphocytes, and in i.p. tumours both TNFR signalling and CD8 effector secretion of IFNg were required. We are not aware of other reports that CCL21 is induced by IFNg. In fact, one report demonstrated that IFNg was required for a transient drop in CCL21 levels in the inflamed spleen 58 . Nevertheless, our data suggest that IFNg directly induces CCL21 expression in both endothelial cells and gp38 þ cells associated with LN-like vasculature. As i.p. tumours in TNFR1/2 À / À mice lack both PNAd and CCL21, while i.p. tumours in IFNg À / À mice are deficient in CCL21 only, initial LTa 3 -TNFR signalling may enable subsequent responsiveness of CCL21 expression to IFNg signalling. CCL21 expression by gp38 þ cells is intriguing, because these cells share phenotypic characteristics with LN FRCs 38 , which are a major source of CCL21 in LN. gp38 þ cells were also the major source of CCL21 in tumours, but were in proximity to PNAd þ endothelial cells that also expressed CCL21 on their own. Thus, CCL21 displayed on LN-like vasculature that enables naive T-cell entry probably reflects both direct production by the endothelium and CCL21 transcytosed 59 from adjacent gp38 þ cells. We also found that these gp38 þ cells are organized around portions of LN-like vasculature into reticular networks in conjunction with large aggregates of B lymphocytes in i.p. tumours. The functionality of gp38 þ cells as components or organizers of these TLO-like structures, the functionality of the structures themselves and the reasons they develop at only some sites of LNlike vasculature remain to be defined. Nonetheless, these observations point towards a level of immune organization within tumours that may offer additional targets for intervention.
The cellular and molecular differences in induction of LN-like vasculature in i.p. and s.c. tumours highlight a microenvironmental heterogeneity that depends on the location of growth. This is typified by the greater secretion of IFNg and TNFa by NK cells in s.c. tumours, enabling them to induce LN-like vasculature in this location. The enhanced functionality of NK cells in s.c. tumours could reflect the recruitment of distinct NK subpopulations that cannot reach i.p. tumours based on differential expression of chemokine receptors 60 . Alternatively, each location may display different levels of activating or inhibitory NK receptor ligands 61 , shifting the balance in s.c. tumours towards NK-cell activation. The presence of immunosuppressive cytokines in i.p. tumours such as TGFb could also inhibit NK-cell effector activities. These differences may be driven by properties of the tumour microenvironment that are determined by external cues, as has been described in the regional control of DC programming of tissue selective T-cell trafficking 62 . Locationbased microenvironmental heterogeneity may affect efforts to enhance immune responses in various metastatic sites.
Evidence for both positive and negative effects of LN-like vasculature on anti-tumour immunity have been reported in murine models 20,21 . In several human studies, PNAd and CCL21 expression have been associated with a positive prognosis 3-6 . As our results demonstrate that the level of LN-like vasculature is correlated with the extent of accumulation by endogenous effector lymphocytes that induce it, this prognostic association could simply reflect that the presence of LN-like vasculature is a proxy for effector lymphocyte infiltration. However, here we have shown that the development of LN-like vasculature directly supports an ongoing anti-tumour response by enabling the infiltration of naive CD8 T cells, which can differentiate into effector cells in the tumour microenvironment. It is important to note that our demonstration of this effect occurred in an experimental setup that necessarily eliminated the normal constant traffic of LN-activated effectors as well as any ongoing traffic of naive T cells into the tumour. The overall impact of a more sustained influx of naive tumour-infiltrating CD8 T cells may therefore be even greater. Thus, induction of LN-like vasculature in tumours has the potential to be a key contributor to anti-tumour immunity by generating a self-sustaining infiltration of T cells into the tumour mass.
Tumour cells and injections. B16-F1 melanoma cells, variants expressing cytoplasmic ovalbumin (B16-OVA) or a chimeric major histocompatibility complex molecule that presents tyrosinase (B16-AAD), and LLC-OVA (a gift from E. Podack, University of Miami) (4 Â 10 5 cells) were injected i.p. or s.c. Tumors were allowed to establish for B14 days prior to naïve T-cell transfer or other analyses. Paraffin embedded melanomas from BRAF V600E PTEN À / À mice and BRAF V600E PTEN À / À with an activating mutation in b-catenin 27,34 were a kind gift from Dr Marcus Bosenberg (Yale).
In-vivo effector function. Tumour-bearing mice were injected i.p. with 250 mg of Brefeldin-A (Sigma-Aldrich). Four hours later, tumour-infiltrating lymphocytes were isolated as described above and stained for surface molecule expression, then stained for intracellular IFN-g or TNF-a by using CytoFix/CytoPerm (BD Biosciences).
Immunofluorescence microscopy. Tissues were flash frozen in liquid nitrogen, embedded in optimum cutting temperature compound and cut into 6-mm sections. Sections were fixed with acetone/ethanol and blocked with PBS/5% BSA, anti-CD16/32, 3% H 2 O 2 , 0.1% NaN 3 and Avidin/Biotin Blocking Kit (Vector Laboratories). Paraffin-embedded melanomas from BRAF V600E PTEN À / À mice were deparaffinized, rehydrated and subjected to heat-induced antigen retrieval for 20 min at 95°C in citrate buffer, pH 6.0, before blocking steps. Sections were stained with biotinylated or fluorescence-conjugated antibodies (5-20 mg ml À 1 ) to CD31 (390), ICAM-1 (YN1/1.7.4), CD11c (N418), CD3e (145-2C11), CD8a (53-6.7), B220 (RA3-6B2), LYVE-1 (ALY7), gp38 (8. Image analysis. Quantification of CD31 þ and PNAd þ areas was performed using ImageJ software (NIH) on original fluorescence images taken at identical exposures across samples. Consistent thresholds were applied to each image to identify CD31 þ and PNAd þ pixels. The percentage of CD31 þ pixels of total imaged area and the percentage of PNAd þ pixels within the region of interest of CD31 þ area was calculated. Quantification of CD8 T-cell infiltration was performed by counting the number of cells present per 20 Â field. Multiple sections per tumour and random fields per section were used for analysis. For image presentation, brightness and contrast were linearly adjusted and colour-merged images were generated using Photoshop CS6 software (Adobe).
Bone marrow chimeras. Mice were irradiated (650 rad Â 2) and reconstituted with a minimum of 2 Â 10 6 bone marrow cells depleted of CD4 þ and CD8 þ T cells by magnetic beads (Miltenyi Biotec). Chimeras were maintained on Sulfatrim (sulfamethoxazole/trimethoprim) water for 3 weeks and were allowed to reconstitute for at least 8 weeks before use.
Tumour control. B16-OVA tumours were implanted s.c. into WT or TNFR1/2 À / À mice. On day 8 after implantation, mice were injected i.v. with 4 Â 10 6 OT-I/Thy1.1 cells depleted of CD44-expressing cells by magnetic beads and treated daily with 1 mg kg À 1 FTY720 (a gift from V. Brinkmann, Novartis Pharma AG, Basel, Switzerland) in sterile saline i.p., or treated with FTY720 without receiving naive T cells. Tumour size was monitored daily by caliper measurements and mice were killed when tumours reached 16 mm in any one dimension.
Statistical analysis. P-values were calculated for comparisons between two groups by unpaired t-tests and for comparisons between three or more groups by one-way analysis of variance with post tests, to correct for multiple comparisons as indicated in legends. Po0.05 was considered statistically significant. All graphs and statistics were calculated using Graph Pad Prism version 6.0.