Differential Expression of Homing Receptor Ligands on Tumor-Associated Vasculature that Control CD8 Effector T-cell Entry

The importance of CD8⁺ T cells in immunologic control of solid tumors is well established. However, many tumors show poor CD8⁺ T-cell infiltration, which limits the effectiveness of new generation immunotherapeutics (1–4). Understanding the mechanisms underlying CD8⁺ T-cell infiltration holds the promise of improving clinical outcomes. CD8⁺ T-cell entry into peripheral tissues involves a series of sequential interactions between homing receptors (HR) on activated CD8⁺ T cells and their corresponding ligands on vascular endothelial cells (EC; ref. 5). Expression of the HRs ESL (this is an abbreviation for E-selectin ligand, which, confusingly, is the receptor for E-selectin, a homing receptor ligand expressed on blood vasculature), α₄β₁ and α₄β₇ integrins, and the chemokine receptor CCR9 on activated CD8⁺ T cells is determined by the secondary lymphoid organ in which priming occurs (6–9). Conversely, expression of LFA-1 and several chemokine receptors, including CXCR3, on activated CD8⁺ T cells does not depend on the priming site (9).

The tropism of T cells expressing distinct HR depends upon expression of the corresponding HR ligands by tissue vasculature. E-selectin, a HR ligand that enables entry of ESL⁺ effectors, is selectively expressed on skin vasculature (10). MadCAM-1 and CCL25, which enable entry of α₄β₇⁺CCR9⁺ T cells, are HR ligands selectively expressed on gut vasculature (11, 12). Conversely, HR ligands VCAM-1, ICAM-1, and chemokines CXCL9, 10, and 11, which enable entry of α₄β₁⁺, LFA-1⁺, and CXCR3⁺ effectors, respectively, are expressed on inflamed vasculature of many tissues (13, 14). Thus, tissue-specific and inflammation-induced expression of HR ligands determines which tissues are infiltrated by CD8⁺ effectors primed in different secondary lymphoid organs.

Relatively little is known about the HR ligands expressed on tumor vasculature or the HR/ligand interactions that lead to CD8⁺ T-cell entry. LFA-1/ICAM-1 interactions are common mediators of leukocyte engagement with other cells and, not surprisingly, play an important role in entry into murine melanoma and human glioblastoma (15–17). CXCR3 interactions with CXCL9/10 mediate entry into subcutaneous (SC) and intracranial murine melanomas (18–20) and correlated with the representation of CD8⁺ T cells in melanomas growing in lungs (21). α₄β₁/VCAM-1 interactions are required for CD8⁺ T-cell accumulation in intracranial melanoma (22) and in SC melanoma following kinase inhibitor-induced VCAM-1 upregulation (19). VCAM-1 expression correlated with increased T-cell representation in pancreatic islet cell carcinoma, glioblastoma, and melanoma (16, 23, 24), but a cause-and-effect relationship was not established. E-selectin/ESL interactions are important for entry into B16-F10 melanomas following inflammatory stimuli, and in human squamous cell carcinoma, TLR-mediated upregulation of E-selectin is correlated with increased representation of CLA⁺ CD8⁺ T cells (15, 25). However, the importance of E-selectin/ESL interactions in mediating entry in the absence of inflammatory stimuli has not been addressed. In human colorectal cancers, high MadCAM-1 gene expression is associated with increased CD8⁺ T-cell representation (26), but to date, the role of α₄β₇ in entry into tumors has not been evaluated. Thus, although studies have identified individual HR that can contribute to CD8⁺ T-cell infiltration into some tumors, a comprehensive analysis of the different molecules that mediate entry into any particular tumor, and how this varies with anatomical location, has not yet been conducted.

Here, we systematically identified the molecules required for CD8⁺ T-cell entry into B16 melanomas grown in SC and intraperitoneal (IP) compartments. We compared HR ligand expression on B16 melanoma vasculature with that of adjacent tissue vasculature and determined what cells and cytokines drive expression of HR ligands on tumor vasculature. Finally, we determined which HR/ligand interactions are required for CD8⁺ T-cell entry into tumors growing in SC or IP locations. Applying this knowledge clinically has the potential to enhance the efficacy of multiple cancer immunotherapies.

C57Bl/6 (B6; Charles River/NCI); Rag1^−/−, IFNγ^−/−, and E-selectin^−/− (all The Jackson Laboratory); and OT-1 Rag1^−/−, TNFα^−/−, and Thy1.1 (all Taconic Biosciences) mice were purchased. CXCR3^−/− (27), AAD (28), and FH (29) mice were bred in our facility. Animals were housed in pathogen-free facilities. The University of Virginia ACUC approved all procedures in accordance with the NIH Guide for the Care and Use of Laboratory Animals.

B16-F1, B16-OVA (30), B16-AAD (31), MC38 (Steven Rosenberg, NIH) transfected to express cytoplasmic ovalbumin, or LLC-OVA (E. Podack, University of Miami) were injected IP, IV, intracranial, or SC. All lines were Mycoplasma negative and used within 10 passages. Expression of OVA after cell passages was confirmed by flow cytometry, but cells were not otherwise authenticated. Experiments were performed 11 to 14 days after injection.

Frozen 0.7-μm sections were fixed in acetone:ethanol, blocked sequentially with anti-Fc (2.4G2; Bio X Cell), Avidin/Biotin Blocking Kit (Vector), H₂O₂ and NaN₃, then stained with either CD31-FITC (eBioscience) or CD31-AF647 (BioLegend), and MadCAM-1-Biotin, VCAM-1-Biotin, or ICAM-1-Biotin (all eBioscience). Streptavidin DyLight550 (Thermo Fisher) was used as secondary. PerkinElmer TSA Biotin Kit was used for amplification. Images were collected on an AxioImager with Apotome (Zeiss). ImageJ software (NIH) was used to quantify CD31⁺ pixels that were also VCAM-1⁺, MadCAM-1⁺, HA⁺, or ICAM-1⁺ (32).

Bulk H-2K^b+, ovalbumin-specific OT-I Thy1.1⁺ T cells were adoptively transferred into B6 mice, which were then immunized by SC, IV, or IP routes with SIINFEKL peptide-pulsed, activated bone marrow–derived dendritic cells (BMDC; ref. 9). Alternatively, CD8⁺ T cells from FH or FH x CXCR3^−/− transgenic mice were adoptively transferred into AAD mice, and immunized with YMDGTMSQV peptide–pulsed activated BMDC. FTY720 (Novartis) was administered daily to retain effector CD8⁺ T cells in draining lymph nodes (LN). On day 5, draining LNs and/or spleens were homogenized and treated with RBC lysis buffer (Sigma). CD8⁺ T cells were enriched using anti-CD8⁺ magnetic beads (Miltenyi) and 300,000 to 500,000 were injected IV into tumor-bearing animals.

Effectors were blocked with either 100 μg anti-rat IgG (Jackson ImmunoResearch), anti-α₄ (PS/2; ATCC), anti-CD44 (IM7), anti-α₄β₇ (DATK32), or anti-CD11a (M17/4; all Bio X Cell) for 30 minutes before injection into tumor-bearing animals. HR ligands were blocked by IP injection of 100 μg anti-VCAM-1 (M/K-2.7) or anti-MadCAM-1 (MECA-367; Bio X Cell) 6 hours prior to effector transfer.

Tissue was incubated in medium containing 0.42 U/mL Liberase TM (Roche) for 15 minutes at 37°C, homogenized and CD31⁺ cells purified using anti-CD31 magnetic beads (Miltenyi Biotec) and the Possel AutoMACS protocol.

LNs and spleens from B6, TNFα^−/−, or IFNγ^−/− mice were homogenized and treated with RBC lysis buffer (Sigma). CD8⁺ T cells were enriched using anti-CD8⁺ magnetic beads, and 5,000,000 were transferred into Rag1^−/− mice. Three days after transfer, 400,000 B16-OVA cells were injected SC.

Cells were Fc blocked (Bio X Cell) and stained with fluorescent antibodies to CD31, CD45, CD8, Thy1.1 (all eBioscience); CXCL9 (BioLegend), E-Selectin, P-Selectin (both BD); E-selectin fusion protein and P-selectin fusion protein (both R&D). BD Cytofix/Cytoperm Kit was used for fixation/permeabilization. CD31-enriched cells were resuspended in Dapi and run live. Lymphocytes were fixed in 2% PFA. Cells were run on FACS Canto II (BD) or Cytoflex (Beckman Coulter) flow cytometers. FlowJo software was used for analysis.

All analyses were performed using unpaired Student t tests.

Skin-associated vascular ECs express E-selectin, whereas gut-associated vascular ECs express MadCAM-1. To determine whether tumor-associated vasculature expressed these ligands, and whether expression depended on tissue origin of the tumor or site of growth, we evaluated CD31⁺ vascular ECs isolated from SC or IP B16-F1 melanoma tumors expressing ovalbumin (B16-OVA) by flow cytometry or by immunofluorescent microscopy. ECs from SC tumors expressed E-selectin, but few expressed MadCAM-1 even following signal amplification (Fig. 1A and B). In contrast, EC from IP tumors expressed MadCAM-1, whereas E-selectin was very low. We observed similar patterns of MadCAM-1 on ECs from IP and SC tumors of ovalbumin-expressing Lewis lung carcinoma (LLC-OVA) and of E-selectin on ECs from SC and IP tumors of ovalbumin-expressing MC38 (Supplementary Fig. S1). These results suggest that anatomic microenvironmental influences that drive expression of E-selectin and MadCAM-1 on normal tissue vasculature also pattern tumor vasculature.

We next evaluated expression of HR ligands that are upregulated on endothelium of many inflamed tissues. Although P-selectin is sometimes considered a skin-associated molecule, it is more broadly expressed (33), and P-selectin was expressed similarly on SC and IP tumor vasculature (Fig. 1C). ICAM-1 and hyaluronic acid (HA), LFA-1 and CD44 ligands, respectively, were also expressed comparably on SC and IP tumor vasculature (Fig. 1D and E). CXCL9, a CXCR3 ligand, was expressed in CD31⁺ ECs from SC and IP tumors, although the percent positive cells was significantly higher in the latter (Fig. 1F). Conversely, VCAM-1, the ligand for α₄β₁ integrin, was expressed on a higher percentage of CD31⁺ pixels from SC tumors, and was detectable without tyramide amplification, whereas visualizing VCAM-1 on IP tumor vessels required amplification (Fig. 1G). Thus, SC and IP tumors have distinct microenvironmental features that cause differential expression of VCAM-1 and CXCL9.

We also assessed HR ligand expression on the vasculature of tumors growing in lung and brain, common sites of melanoma metastasis. VCAM-1 was expressed on tumor ECs from both sites, but at a lower percentage compared with SC and IP tumors (Supplementary Fig. S2). Interestingly, E-selectin was expressed on brain but not lung tumor EC, whereas MadCAM-1 was not expressed on either. These findings validate VCAM-1 as being broadly expressed, whereas E-selectin and MadCAM-1 are controlled by the anatomical site.

Because HR ligand expression on vascular endothelium is upregulated by inflammatory cytokines, we compared it on vasculature from tumors and normal tissue. The percentage of CD31⁺ ECs from SC tumors that were E-selectin⁺, VCAM-1⁺, or CXCL9⁺ was significantly higher than that of ECs from skin (Fig. 1H). Similarly, the percentage of CD31⁺ pixels that were VCAM-1⁺ was significantly higher in IP tumors than the gut (Fig. 1I). Surprisingly, the percentage of CD31⁺ pixels that were MadCAM-1⁺ was significantly lower in tumor than gut. The elevated expression of HR ligands on SC and IP tumor vasculature is consistent with the idea that inflammatory stimuli in the tumor microenvironment are responsible.

To determine whether HR ligand expression depended on adaptive immunity, we compared B16-OVA tumors grown in B6 and Rag1^−/− mice. The fraction of CD31⁺ cells expressing VCAM-1 was reduced by 70% to 90% in tumors from Rag1^−/− mice (Fig. 2A and B). Similarly, expression of CXCL9 by CD31⁺ cells was negligible (Fig. 2B). There was no statistically significant difference in E-selectin expression (Fig. 2). We also evaluated HR ligand expression on the vasculature of parental B16-F1 tumors, which lack a strong antigen and are less well-infiltrated by CD8⁺ T cells (refs. 30, 32; Supplementary Fig. S3). VCAM-1 and CXCL9 expression on B16-F1–associated vasculature was significantly lower than that of B16-OVA tumors, whereas E-selectin expression was comparable (Fig. 2A and B). These results identify a role for adaptive immunity in upregulating expression of VCAM-1 and CXCL9, but not E-selectin, on SC tumor vasculature.

We also analyzed patterns of coexpression of these HR ligands by individual CD31⁺ cells. Using the total percentage of cells expressing each marker, we calculated the expected levels of coexpression based on random distribution, and compared these to what was observed (Fig. 2C). The percentage of cells expressing only E-selectin was consistent with independent control of expression. However, the percentage of cells expressing either VCAM-1 or CXCL9 alone was significantly lower than predicted, while populations expressing these ligands together were significantly higher. This suggests a coordinated control of VCAM-1 and CXCL9, consistent with effector cells releasing inflammatory mediators in proximity to individual ECs.

CXCL9 expression on inflamed vasculature is upregulated by IFNγ (34), whereas VCAM-1 expression can reportedly be upregulated by either IFNγ or TNFα (35, 36). To determine the role of these cytokines in driving CXCL9 and VCAM-1 expression on tumor vasculature, SC B16-OVA tumors were grown in B6, TNFα^−/−, and IFNγ^−/− mice. As expected, CXCL9 expression in EC from tumors grown in IFNγ^−/− mice was negligible, pointing to a nonredundant role for this cytokine (Fig. 3A). VCAM-1 expression was also negligible on ECs from these tumors, but was comparable on vasculature of tumors grown in B6 or TNFα^−/− mice (Fig. 3A). Because the immune infiltrate of B16-OVA tumors was dominated by CD8⁺ T cells, we asked whether they were the source of IFNγ controlling VCAM-1 and CXCL9 expression. SC B16-OVA tumors were grown in Rag1^−/− mice previously repleted with bulk CD8⁺ T cells from WT, IFNγ^−/−, or TNFα^−/− animals. CXCL9 and VCAM-1 expression in tumors from WT CD8⁺ T-cell repleted mice was comparable with that of tumors from B6 mice and from mice repleted with TNFα^−/− CD8⁺ T cells, confirming a lack of control by this cytokine (Fig. 3B). However, expression was substantially lower in tumors from mice repleted with IFNγ^−/− CD8⁺ T cells (Fig. 3B). Despite differences in ligand expression, infiltration of WT, IFNγ^−/−, or TNFα^−/− CD8⁺ T cells was similar. Thus, despite reports of TNFα-mediated control of VCAM-1 expression, these results establish a nonredundant role for IFNγ in upregulating CXCL9 and VCAM-1 on tumor-associated vasculature, and the sufficiency of CD8⁺ T cells to be its source.

To determine which vascular HR ligands and HR mediate effector CD8⁺ T-cell entry into SC and IP tumors, we generated effectors with different HR expression profiles by IV, SC, or IP immunization (9). These effectors were adoptively transferred into Thy1.1 congenic mice bearing late-stage SC or IP tumors, and those that had infiltrated tumors were quantitated 18 hours later. IV-primed effectors uniformly express α₄β₁, CD44 (Supplementary Fig. S4), CXCR3, and LFA-1 (9). After transfer of 300,000 to 500,000 effectors, 0.1% to 2% had infiltrated both SC and IP B16-OVA tumors comparably 18 hours later (Fig. 4). This is consistent with the low infiltration observed in other murine and human studies of adoptively transferred T cells (37).

To evaluate the importance of individual HR and ligands, we pretreated these CD8⁺ effectors with HR blocking antibodies prior to adoptive transfer, pretreated recipient mice with HR ligand blocking antibodies, or used mice that carried genetic HR or HR ligand deletions. Antibody blockade of α4β1 integrin on CD8⁺ T effectors almost completely eliminated their infiltration into SC and IP tumors (Fig. 4A). Their representation in the spleen was unaffected, demonstrating that lack of accumulation in the tumor was not due to antibody-mediated killing. Blockade of VCAM-1 on tumor vasculature also substantially inhibited entry of these effectors (Fig. 4B), demonstrating that α₄β₁/VCAM-1 interactions are essential for CD8⁺ T-cell entry into both SC and IP tumors despite differences in the level of VCAM-1 expression. Entry into SC tumors was also completely inhibited using antibodies to LFA-1 (Fig. 4C) and CD44 (Fig. 4D). It is likely that LFA-1 and CD44 act by binding to ICAM-1 and HA, respectively, as these well-established ligands are displayed on tumor vasculature. The greater than 90% inhibition of infiltration observed with individual blockade of α4β1, CD44, and LFA-1 indicates that these HRs act nonredundantly to mediate entry of IV-primed CD8⁺ effectors into SC and IP tumors.

To assess the importance of CXCR3, we used CXCR3^−/− mice bred to mice expressing a transgenic T-cell receptor specific for tyrosinase⁺ HLA-A2 (29, 38), and assessed infiltration of IV-primed effectors into B16 tumors expressing AAD, a recombinant form of HLA-A2 (31). Infiltration of both SC and IP B16-AAD tumors by CXCR3^−/− effectors was reduced more than 80% (Fig. 4E). Also, WT effectors transferred into mice bearing 3-day-old tumors controlled outgrowth, whereas CXCR3^−/− effectors had no effect (Fig. 4F). Thus, CXCR3 plays an essential role in the infiltration of CD8⁺ effectors into tumors in multiple anatomical locations and is essential for tumor control.

CD8⁺ T-cell effectors primed by IP immunization are over 70% α₄β₇⁺ (ref. 9; Supplementary Fig. S4). An additional 10% express α₄ in the absence of α₄β₇, and are thus α₄β₁⁺. The level of α₄β₁ on α₄β₇⁺ cells cannot be determined. Because VCAM-1 expression is low on IP tumor vasculature, we asked whether MadCAM-1 could provide an alternate ligand for α₄β₇⁺ CD8⁺ T-cell entry. Surprisingly, MadCAM-1 blockade had no effect on IP-primed effector entry into IP tumors (Fig. 5A). However, VCAM-1 blockade inhibited infiltration by about 80%, and no additional effect was seen with simultaneous VCAM-1 and MadCAM1 blockade (Fig. 5B). Treatment of these effectors with either anti-α₄, which would block both α₄β₇ and α₄β₁, or anti-α₄β₇, also inhibited entry into IP tumors (Fig. 5B). α₄β₇ has been reported to bind VCAM-1 as well as MadCAM-1 (11), suggesting that IP-primed effectors enter IP tumors through interactions of α₄β₇, and possibly α₄β₁, with VCAM-1, while MadCAM-1 is uninvolved.

Because more MadCAM-1 was expressed on IP tumor vasculature than VCAM-1 (Fig. 1), its lack of involvement led us to question whether it was expressed on the luminal surface. Thus, we injected fluorescent anti-CD31 and anti-MadCAM-1 into IP tumor-bearing mice 30 minutes before harvesting tissue. As expected, the CD31 antibody decorated the vasculature of Peyer's patches, axillary/brachial LNs, and tumors, and the MadCAM-1 antibody decorated vasculature of Peyer's patches but not of axillary/brachial LNs. However, it failed to stain IP tumor vasculature (Fig. 5C). Using the same approach, anti-VCAM-1 did stain the tumor vasculature (Fig. 5D). Thus, although expressed on tumor vascular ECs, MadCAM-1 is not available on the luminal surface for interaction with α₄β₇.

CD8⁺ T-cell activation in skin-draining LN generates effectors, over 50% of which express ESL in the absence of α₄β₁, whereas another 25% to 30% coexpress both (Supplementary Fig. S4). Because no commercially available E-selectin antibodies abrogate interaction with ESL⁺ cells (39), we used two approaches to determine if ESL/E-selectin interactions mediated CD8⁺ T-cell entry into SC tumors. First, we transferred a 50/50 mix of SC and IP-primed effectors into WT mice bearing SC or IP tumors. Infiltration of IP tumors by both effector populations was comparable, and also comparable with the infiltration of SC tumors by IP-primed effectors. However, SC effectors infiltrated SC tumors to a substantially greater extent than IP effectors (Fig. 6A). This is consistent with the idea that ESL/E-selectin interactions selectively augment SC effector entry into SC tumors. To establish this definitively, we evaluated entry of SC-primed effectors into SC tumors growing in WT and E-selectin^−/− mice. Infiltration into tumors grown in E-selectin^−/− mice was about half relative to tumors grown in WT mice (Fig. 6B). These results indicate that E-selectin⁺ vasculature selectively facilitates ESL⁺ CD8⁺ T-cell entry into SC tumors.

This study characterized vascular HR ligand expression in murine tumors growing in distinct anatomic locations, the factors controlling ligand expression, and the molecular requirements for CD8⁺ T-cell entry (Fig. 7). Our results identify both tumor anatomical location and immune microenvironment as important elements controlling expression of these molecules. Whereas E-selectin augmented CD8⁺ T-cell entry selectively into SC tumors, MadCAM-1 was irrelevant because it was not expressed on the vascular luminal surface. VCAM-1, through interaction with α₄β₁ and likely α₄β₇, together with one or more ligands for CXCR3 supported broad-based infiltration of CD8⁺ T cells into both anatomic locations. Our results identify mechanisms that may underlie the lack of infiltration that is observed in many human tumors, and point toward strategies to enhance infiltration for therapeutic purposes.

Previous studies evaluated expression of individual HR ligands on tumor vasculature and concluded that the expression of E-selectin (15, 25), ICAM-1 (15), CXCL9 (20, 21), and VCAM-1 (19) was often low. Expression of VCAM-1 correlates with T-cell representation in several human cancers (16, 23, 24). Expression of VCAM-1, ICAM-1, and MadCAM-1 in human colorectal cancers correlates with T-cell representation and patient survival (26). Here, we showed that patterns of E-selectin and MadCAM-1 expression on the vasculature of different murine SC and IP tumors mimicked those on skin and gut vasculature, respectively (10, 11). VCAM-1, CXCL9, and ICAM-1 were expressed on tumor vasculature in both locations, in keeping with their broad expression on inflamed tissues. Our observation that E-selectin and MadCAM-1 expression depends on anatomic location is consistent with the sprouting model of tumor angiogenesis, in which vasculature branches from healthy tissue vasculature toward areas of hypoxia (40). Signals that lead to differences in HR ligand expression on skin and gut vasculature are not understood. In addition, the observation that MadCAM-1 expression is confined to the abluminal surface has not been previously reported and indicates that currently unidentified factors are controlling its polarity. Thus, although tumor-associated vasculature is considered to be aberrant, our work reveals that it retains tissue-associated characteristics.

Previous studies demonstrated that proinflammatory stimuli augment low level HR ligand expression on poorly immunogenic tumors (15, 19, 25), and this is associated with enhanced entry of CD8⁺ T cells. Here, we showed that expression of VCAM-1 and CXCL9 on the vasculature of B16-F1 tumors grown in WT mice and B16-OVA tumors grown in Rag^−/− mice was very low, and that the immune response to tumor-associated OVA in WT mice was sufficient to drive increased expression. VCAM-1 and CXCL9 expression on tumor endothelial cells was coordinately controlled by IFNγ released from CD8⁺ T cells, and was significantly higher than that of normal tissue vasculature. E-selectin expression, although higher than that of normal tissue, was not significantly influenced by adaptive immunity. Increased HR ligand expression is associated with increased T-cell representation in the tumor microenvironment and improved tumor control (41). Collectively, these studies suggest that HR ligand expression on most tumors limits CD8⁺ effector T-cell entry, but can be enhanced by augmenting endogenous immunity, directly activating tumor vascular EC, or manipulating the immunosuppressive tumor microenvironment. A somewhat surprising observation was that infiltration of WT and IFNγ^−/− CD8⁺ T cells into tumors grown in Rag1^−/− animals was similar, despite the lack of HR ligand expression on tumors infiltrated by IFNγ^−/− CD8⁺ T cells. This suggests that initial infiltration of these cells, presumably involving these same HR ligands, is driven by another mechanism, and in Rag1^−/− mice this is most likely an innate immune cell. These results suggest HR ligand expression on tumor vasculature must be sustained through continual restimulation by IFNγ, creating a positive feedback loop in which newly infiltrating T cells that release IFNγ maintain or increase expression of HR ligands, leading to recruitment of additional CD8⁺ effectors that perpetuate the process. This positive feedback loop may be down-modulated by the immunosuppressive microenvironment of tumors, and upregulated by therapeutic interventions that augment the immune response or directly increase HR ligand expression on tumor vasculature.

We established that α₄β₁, CD44, and LFA-1 played nonredundant roles in enabling effector CD8⁺ T-cell entry into both SC and IP tumors. LFA-1/ICAM-1 interactions commonly mediate firm adhesion and can mediate CD8⁺ T-cell entry into therapeutically treated tumors (15), so their demonstrated importance in our study was not unexpected. α₄β₁ acts at both the initial step of the adhesion cascade (42) and the terminal step, in lieu of LFA-1 (43). The nonredundancy of α₄β₁ and LFA-1, along with the well-established role of LFA-1 in the terminal step (44), suggests that α₄β₁ acts in the initial slow rolling step to enable effector entry into tumors. Although elevated expression of CD44 is a marker of T-cell activation and memory, its role in T-cell entry has been evaluated infrequently. CD44 acts cooperatively with α₄β₁ to mediate T-cell entry into the peritoneum (43). Based on this, we propose that α₄β₁ and CD44 coordinately mediate the initial slow rolling step.

CXCR3 is a positive prognostic indicator in melanoma patients, and its ligands have been identified as part of a gene signature associated with enhanced patient survival and response to immunotherapy (45, 46). CXCL9 and CXCR3 were shown to be critical in CD8⁺ T-cell entry into SC or intracranial B16 tumors (18, 20). Our results confirm and extend this earlier work by showing that CXCR3 is the only chemokine receptor needed for CD8⁺ T-cell entry into B16 tumors growing in IP and SC locations, that it was necessary for T-cell–mediated tumor control, and that CXCL9 expression was dependent on adaptive immunity. In keeping with other work, we propose that CXCR3 activates the high-affinity form of LFA-1 for firm adhesion. The importance of VCAM-1 is emphasized by the fact that it supports entry of α₄β₁⁺ T cells even when expressed at relatively low levels on IP tumor vasculature, and that it may also enable infiltration α₄β₇⁺ T cells. However, E-selectin and MadCAM-1 have more limited value. Taken together, our work identifies a single set of HRs and ligands that supports CD8⁺ T-cell entry into a variety of tumors growing in different anatomic locations.

In mice, intradermal immunization leads to T cells with predominant expression of ESLs and very few expressing α₄β₁ (9). In this regard, the commonly used intradermal route of vaccine administration may limit the effectiveness of cancer vaccines for lesions in non–skin-associated sites. Intravenous vaccination generates effectors that uniformly express α₄β₁ and should be evaluated as a cancer vaccine administration route. Similarly, the HRs expressed on tumor-infiltrating lymphocytes or peripheral blood lymphocytes that have been activated for use in CAR-T–, bispecific antibody–, or recombinant TCR–based adoptive cell therapy approaches are rarely examined. Given the massive numbers of such cells that are administered for therapeutic effect, and the higher success rates in treating liquid tumors in which infiltration is not constrained by vasculature, we believe there are significant opportunities to optimize HR expression to improve the success of this approach.

HR ligand expression on tumor vasculature must also be high enough to support entry. Although tumor vasculature supports infiltration of some lymphocytes, it could be increased if HR ligand expression were enhanced and in turn augment the clinical response rates. Adoptive cell therapies have shown efficacy in melanoma patients with preexisting CD8⁺ T-cell infiltrates (37). It may be that cytokines secreted by preexisting CD8⁺ T cells drive increased vascular HR ligand expression, enabling more efficient infiltration of newly transferred cells. Also, direct enhancement of VCAM-1 or CXCL9 on tumor vasculature could initiate a feedback loop such that even patients without preexisting CD8⁺ T-cell infiltrates may respond to therapy. In addition to adoptive cell therapy, radiotherapy and checkpoint blockade inhibitors are also most effective in patients with preexisting CD8⁺ T-cell infiltrates (47, 48). Understanding the requirements for CD8⁺ T-cell entry into tumors in multiple anatomic locations, combined with targeted upregulation of HR ligands, could enhance the efficacy of these therapies still further.

A.R. Ferguson is Director, Clinical Science, for Gritstone Oncology. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.N. Woods, A.R. Ferguson, V.H. Engelhard

Development of methodology: A.N. Woods, J.D. Peske, R.K. Gregg, V.H. Engelhard

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.N. Woods, J. Zeng, A.B. Dutta, J.D. Peske, E.F. Tewalt

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.N. Woods, A.L. Wilson, N. Srivinisan, J. Zeng, A.B. Dutta, J.D. Peske, E.F. Tewalt, A.R. Ferguson

Writing, review, and/or revision of the manuscript: A.N. Woods, A.L. Wilson, V.H. Engelhard

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.N. Woods

Study supervision: V.H. Engelhard

This study was supported by USPHS grants CA185955, AI068836, and CA78400 (V.H. Engelhard), USPHS fellowships CA121916 (R.K. Gregg) and AI072818 (A.R. Ferguson), USPHS training grants AI007496 (A.N. Woods, E.F. Tewalt, R.K. Gregg, and A.R. Ferguson), GM007267 (A.B. Dutta), GM007267 and CA009109 (J.D. Peske), DOD Horizon Award CA150903 (A.L. Wilson), and fellowship PF-10-156-01-LIB from the American Cancer Society (E.F. Tewalt).

The authors thank Kara Cummings, Holly Davis, and the UVA Research Histology Core for excellent technical support.

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