RORγt⁺ Innate Lymphoid Cells Promote Lymph Node Metastasis of Breast Cancers

Breast cancer is the most common malignant neoplasm with significant morbidity and mortality. The ability of cancer cells to invade lymphatics stratifies breast cancers into distinct prognostic groups (1). The molecular mechanisms mediating this tumor cell entry remain unclear, but studies have established important roles for the lymphoid chemokines CXCL13, CCL19, and CCL21 (2).

An important early step in the construction of lymphoid organs is the recruitment of lymphoid tissue inducer cells (LTi) by CXCL13 and CCL21, which are recognized via the receptors CXCR5 and CCR7, respectively (3–5). LTis are members of the innate lymphoid cells (ILC) family. Recent moves to propose a uniform nomenclature divide these cells into three groups (6), and LTis represent the prototypic cell type of the “group 3” RORγt⁺ family of ILCs. We will refer to these cells henceforth as ILC3. ILC3 play a major role in lymphoid tissue development both in the embryo (7) and in adult life (8, 9). Within the secondary lymphoid structures, ILC3 produce lymphotoxin (LT)α₁β₂, which binds LTβR on mesenchymal stromal cells (MSC), stimulating the production of CXCL13, CCL19, and CCL21, as well as the TNF family member, RANKL, promoting lymphocyte recruitment and compartmentalization (10).

The presence or role of these cells has not yet been explored in breast cancers. Here, we demonstrate that CCL21-dependent recruitment of ILC3s into mammary tumors results in a CXCL13-dependent positive feedback loop between ILC3 and MSCs. Antibody-blocking experiments in BALB/c and Rag1^−/− mice demonstrated that CCL21, CXCL13, ILC3, and RANKL all promote metastasis to the lymph node. We report the novel identification of RORγt⁺ILC3s within the human tumor microenvironment, their association with more agressive breast cancer subtypes, and lymphatic metastasis.

Tissue samples and data from patients were obtained from The King's Health Partners (KHP) Cancer Biobank at Guy's Hospital (London, United Kingdom; REC nr.: 07/40874/131).

Experiments were performed in accordance with the UK Home Office Animals Scientific Procedures Act, 1986 and the UKCCCR guidelines. Tumors were established by the injection of 4T1.2 cells into the mammary fat pad of 6- to 8-week-old BALB/c mice (Charles River Laboratories) and Rag1^−/− mice (BALB/c background, The Jackson Laboratory). CXCL13 or CCL21 were neutralized by intravenous injection of 0.5 μg goat antibodies (R&D Systems) starting on the first day after tumor establishment and repeated every 3 days until the end of the experiment. ILCs were depleted by intraperitoneal injection of 0.25 mg anti-CD90.2 (clone 30H12, Bio X Cell) starting on day 3 after tumor establishment and repeated every 3 days until the end of the experiment.

The KHP Cancer Biobank of the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) dataset was profiled using the Illumina HT12 platform. Frozen tissue sections were subjected to histopathologic review to assess the presence of invasive tumor, and only samples with >70% tumoral DNA were included. Samples were quantile normalized, and a ComBat BeadChip correction applied [N = 234; 176 estrogen receptor–positive (ER⁺) samples, 58 ER⁻ samples]. PAM50 subtype was assigned as in refs. 11–13.

Sixty fresh-frozen tumor sections were randomly selected from the METABRIC patient cohort for ILC staining as described previously (14). Confocal tile scan images were obtained using an LSM510 Metamicroscope (Carl Zeiss). Image analysis for RORγt⁺ILC quantification was carried out using MacBiophotonics ImageJ software. Detailed IHC protocols are described in Supplementary Information.

The mouse breast cancer cell line 4T1.2 (derived from a mammary carcinoma in a BALB/c mouse; ref. 15) and human bone marrow–derived MSCs (HS-5) were cultured in DMEM (Invitrogen) complete media. Extracellular matrix (ECM) invasion assays, based on the Boyden chamber principle, were carried out using 96-well Cell Invasion Assay Kit (ECM555, Chemicon International) as per the manufacturers' instructions. To confirm identity, short tandem repeat profiling was performed on all cell lines.

For NKp46⁻ILC3 sorting experiments, splenocyte suspensions were prepared from BALB/c mice and cells stained with CD3, CD11c, B220R, CD127, CD90.2, and NKp46 and sorted by using a FACSAria. The NKp46⁻ILC3 were identified as CD3⁻CD11c⁻B220⁻CD127⁺CD90.2⁺NKp46⁻ cells. Purity was confirmed at >90%. To extract intratumoral ILC3, tumors were minced and incubated with collagenase/hyaluronidase at 37°C for 60 minutes and passed through a filter to form a single-cell suspension. Cells were stained as per sorting experiments. Flow cytometry reference beads (PeakFlow blue; Invitrogen) were added to the samples before analysis for quantification of cells in each tumor. The absolute number of cells/mg of tumor was calculated using the formula: density of x cells = (number of beads added to each sample multiplied by count of x cells/count of beads)/tumor weight. For multi-photon experiments, 5–6 × 10⁴ sorted NKp46⁻ILC3 were injected (intravenously) into tumor-bearing mice on the same day. Immune cell populations from tumors and draining lymph nodes (DLN) from mice after treatment with either neutralizing anti-CXCL13, anti-CCL21, or IgG control (R&D Systems) were isolated as described above. Antibodies used are included in Supplementary Table S1.

Cells were cultured in DMEM complete medium supplemented with 25 mmol/L HEPES. For NKp46⁻ILC3-MSC coculture experiments, MSCs were grown in 9.4 × 10.7 mm ibidi 8-well slide chambers. Image acquisition was performed using an Olympus IX71 inverted microscope housed within an environment chamber maintained at 37°C. Sequential phase contrast images were captured every 10 minutes for a total of 10 hours. NKp46⁻ILC3-MSC clustering was measured as described in Supplementary Information.

Tumors were snap frozen and lysed by homogenization in 100 mmol/L Tris pH 7.5, 150 mmol/L NaCl, 1 mmol/L EGTA, 1 mmol/L EDTA, 1% (v/v) Triton X-100, and 0.5% (w/v) sodium deoxycholate. ELISAs were performed using DuoSet Kits (R&D Systems).

MSCs were cultured overnight in 6-well plates to 30% confluency. Cells were transfected with RNAiMAX in serum-free OptiMEM and siRNAs at 20 nmol/L. Details of the siRNA used are in Supplementary Table S2.

Mammary imaging window (MIW) surgery was performed 10 days after injection of 1 × 10⁶ 4T1.2 cells into the mammary fat pad as described previously (16). For multi-photon experiments, 1 × 10⁶ MSCs (control or knockdown) followed 24 hours later by 5 × 10⁴ sorted NKp46⁻ILC3 cells were intravenously injected into mice. Twenty-four hours later, mice were placed in a microscope-attached imaging box kept at 32°C and imaged for a maximum period of 3 hours/day for 3 consecutive days. Image processing and image reconstructions were done using MacBiophotonics ImageJ software.

Permutation tests for small samples with multiple ties were performed using the “coin” package in R-2.13.0 (17). Predictive value of ILC score for high lymph node burden was determined using Cox multivariate proportional hazards model. GraphPad was used for other data analysis. P values <0.05 were considered significant.

To investigate whether ILC3s are recruited into a TME, we used a mouse model of triple-negative breast cancer (TNBC) with 4T1.2 cells in BALB/c mice that develop metastatic disease via lymphatics (Fig. 1A; ref. 18). Upon tumor induction, the number of NKp46⁻ILC3 (19) were determined at different times in tumors, DLNs, and nondraining lymph nodes (NDLN; Fig. 1A and B). FACS and immunofluorescence staining for CCR6, RORγt, and CD4 further confirmed the gated cells to be ILC3 (Supplementary Fig. S1). The number of NKp46⁻ILC3 cells in tumors peaked at day 14 (day 10 vs. day 14, P = 0.0019, unpaired t test), whereas the number in DLN peaked later, at day 18 (day 10 vs. day 18, P = 0.0041 unpaired t test; Fig. 1B). NKp46⁻ILC3 cell density within the NDLN did not change significantly, acting as an internal control. Confocal imaging of primary tumors and DLNs taken at day 14 and day 21, respectively, for markers discriminatory for RORγt⁺ILC3 (defined as RORγt⁺CD127⁺CD3⁻) as previously published (20) confirmed the presence within our mouse model (Fig. 1C and D). In contrast to the temporal pattern of NKp46⁻ILC3 infiltration (Fig. 1B), absolute numbers of CD3⁺ T cells (Fig. 1E) and CD19⁺ B cells (Fig. 1F) decreased in tumors over time, whereas the number of T and B cells in the DLN continued to increase until day 24.

CXCL13 has an essential role in ILC3 function (5), and lymphoid structures resembling the LN paracortex develop in tumors expressing high levels of CCL21 (21). Therefore, we investigated whether either of these chemokines could play a role in the recruitment of NKp46⁻ILC3 cells to tumors in our model. We confirmed that tumor NKp46⁻ILC3 express both CCR7 and CXCR5 and are thus capable of responding to CCL21 and CXCL13, respectively (Fig. 1G). We then analyzed the levels of CCL21 and CXCL13 present in primary tumors and serum at various times after tumor establishment. CCL21 levels peaked in both the tumor and serum at day 12, before declining rapidly (Fig. 1H). CXCL13 in the tumor also peaked early (day 10–12), but levels in serum lagged behind, peaking at day 14. In contrast to CCL21, CXCL13 oscillated, with tumor CXCL13 beginning to rise again at day 20 and serum CXCL13 concentration increasing at day 24 (Fig. 1I).

To examine the effect of CCL21 or CXCL13 blockade on NKp46⁻ILC3 recruitment to tumors in vivo, tumor-bearing mice were treated with control or neutralizing anti-CXCL13 and anti-CCL21 antibodies, starting one day after tumor cell implantation and repeated every 3 days. Tumors were analyzed for NKp46⁻ILC3 at day 14, a time point at which maximum number of these cells had previously been shown to be present in the tumors (Fig. 1B). When compared with isotype controls, anti-CCL21, but not anti-CXCL13 neutralizing antibodies, significantly reduced NKp46⁻ILC3 recruitment to the primary tumor (Fig. 1J).

During embryogenesis, clustering of NKp46⁻ILC3s and production of CXCL13 and CCL21 by activated lymphoid tissue organizer cells (LTo, closely linked to stromal cells of mesenchymal origin; ref. 22) are responsible for initiating a feedback loop with further NKp46⁻ILC3 recruitment and subsequent amplification of LT receptor signaling (23). Given the lineage relationship between MSCs, which exhibit a marked tropism for tumors (24), and LTo cells that are known to interact with ILC3s, we hypothesized that ILC3 interaction with CXCL13-producing stromal cells may modulate the chemokine profile of the TME. Within our in vitro model, MSCs secrete high concentrations of CCL21 and CXCL13 chemokines (Supplementary Fig. S2).

Time-lapse microscopy demonstrated NKp46⁻ILC3-MSC clustering (Fig. 2A, top; Supplementary Video S1), with cells remaining closely associated for as long as 7 hours (Fig. 2A, red arrow in bottom; Supplementary Video S2). There was no effect on proliferation of ILC3 on contact or coculture with MSCs (Supplementary Fig. S3A). We quantified cell clustering of NKp46⁻ILC3-MSC (Fig. 2B) and investigated how knockdown of CXCL13 and CCL21 in MSC (Supplementary Fig. S3B and S3C) affected this clustering rate. Transient siRNA knockdown of CXCL13, but not of CCL21, resulted in a decrease of NKp46⁻ILC3-MSC clustering around (P < 0.0001, unpaired t test; Fig. 2C). CXCL13-mediated clustering may be synergistic with the initial CCL21-mediated recruitment of NKp46⁻ILC3 into the primary tumor, as the CCL21-recruited NKp46⁻ILC3 are required to promote significant CXCL13 production by interaction with MSC.

Next, we used an intravital MIW with multi-photon imaging to assess NKp46⁻ILC3-MSC interaction in vivo (16). These visualization experiments were conducted to demonstrate how the fluorescent MSCs (which are allogenic and therefore could have a finite half-life once injected in vivo) may interact with ILC in the relatively short term and whether this interaction is CXCL13 dependent. 4T1.2 cells were injected into the mammary fat pad and an MIW placed over the tumor 10 days after inoculation (Fig. 2D, i). Tumor-bearing mice were treated with either neutralizing anti-CXCL13 or isotype antibody (as described for Fig. 1H). Fluorescently labeled MSCs and NKp46⁻ILC3 were intravenously injected 48 or 24 hours prior to imaging, respectively (Fig. 2D, ii). In control antibody-treated mice, NKp46⁻ILC3 were clustered and in close proximity to MSCs. However, in mice injected with neutralizing anti-CXCL13 antibody, NKp46⁻ILC3 and MSCs were not close with each other (P < 0.0001, unpaired t test, Fig. 2E and F). These in vivo imaging results support the in vitro observation that NKp46⁻ILC3-MSC clustering is CXCL13 dependent.

To test the hypothesis that CCL21 and CXCL13 might play a role in promoting metastasis of tumor cells to lymph node, we treated 4T1.2 tumor-bearing mice with neutralizing antibodies against CCL21 or CXCL13, or with an antibody to deplete NKp46⁻ILC3 and examined the DLN for evidence of metastasis.

In vivo, neither anti-CXCL13 nor anti-CCL21 treatments affected tumor growth (Fig. 3A). The weight of the DLNs were significantly reduced in both cohorts (anti-CXCL13 P = 0.0156; and anti-CCL21 P = 0.0017 one-way ANOVA) compared with the control cohort (Fig. 3B). Immunohistochemical analysis of DLN for tumor load with anti-pan-cytokeratin revealed fewer tumor foci within the DLN of mice treated with anti-CXCL13 or anti-CCL21 compared with control antibody-treated mice (Fig. 3C). Measurements of the total surface area of tumor foci (μm²) demonstrated a significant decrease in the tumor load in the DLN of mice treated with anti-CXCL13 or anti-CCL21 (P < 0.05 one-way ANOVA; Fig. 3C and D).

Given the involvement of CXCL13 and CCL21 in B- and T-cell homeostasis (2), we assessed the effect of anti-CXCL13 and anti-CCL21 blockade on lymph node metastasis in Rag1^−/− mice, which lack B and T cells. Rag1^−/− mice have much smaller lymph nodes; these lymph node samples were therefore formalin fixed to help preserve the morphology better. We report a decrease in the number of pan-cytokeratin positive tumor cells in DLN of tumor-bearing Rag1^−/− mice treated with blocking anti-CCL21 or anti-CXCL13 (anti-CXCL13 P = 0.004; anti-CCL21 P = 0.005; Fig. 3E and F). These results suggest that T and B cells are not involved in the CXCL13- and CCL21-dependent tumor cell migration into lymph nodes.

To strengthen the link between ILC and chemokines in lymph node metastasis, we depleted ILCs with anti-CD90.2, as described previously (Fig. 3G; ref. 25). It is noteworthy that anti-CD90.2 does not specifically deplete ILC3 and is also able to deplete T cells. Therefore, these experiments were also carried out in Rag1^−/− mice. We found a significant decrease in the tumor burden in the DLN of mice treated with anti-CD90.2 (P = 0.04 Mann–Whitney test; Fig. 3H). Therefore, CCL21, CXCL13, and ILCs themselves, and no B or T cells, all promote metastasis of breast cancer cells to the DLN.

As our in vivo results suggested an inhibitory effect of CXCL13 or CCL21 blockade on 4T1.2 cell invasion into the DLN, we used ECM invasion assay to directly investigate the effects of increasing concentrations of CXCL13 and CCL21 on tumor cell invasion. EGF stimulation of 4T1.2 and NIH3T3 served as positive and negative controls, respectively. Recombinant CXCL13 or CCL21 did not significantly increase the invasion of 4T1.2 cells at concentrations between 10 and 100 ng/mL (Fig. 4A).

During lymph node development, the interaction of NKp46⁻ILC3-MSC stimulates RANKL production by MSC, and RANKL signals back to the NKp46⁻ILC3, establishing a positive feedback (26). CXCL13 has recently been shown to promote RANKL expression in stromal cells in oral squamous cell carcinoma (27), and RANK signaling in several breast cancer cell lines induces epithelial–mesenchymal transition (EMT), promoting cell migration and invasion (28). To test the relationship between CXCL13 and RANK signaling in vitro, we first confirmed, as shown previously (27), that although 4T.12 cells expressed RANK receptor in vitro (Fig. 4B), they were not themselves the source of RANKL (Fig. 4C). Levels of more than 200 pg/mL of RANKL were observed in MSC-conditioned media, supporting the hypothesis that the source of RANKL within the tumor is likely to be stromal (Fig. 4C). Stimulation with CXCL13, but not CCL21, increased the expression of RANKL in MSCs (paired t test 50 ng/mL vs. control: P < 0.01; Fig. 4D and E). We also confirmed that MSCs expressed CXCR5 as suggested by the above experiment (Supplementary Fig. S4). We next investigated whether increasing concentrations of RANKL would increase 4T1.2 cell invasion. Addition of RANKL to 4T1.2 (between 10 and 100 ng/mL) was observed to significantly increase the ability of the tumor cells to invade through the matrix (Fig. 4F).

To investigate the relationship between CXCL13 and RANKL expression in vivo, we analyzed the changes in the levels of RANKL in the sera of 4T1.2 tumor-bearing mice at a number of time points after tumor establishment. RANKL levels peaked at day 18 (P < 0.0001, unpaired t test; Fig. 4G), approximately 4 days after the first serum peak in CXCL13 (Fig. 1H). In mice treated with anti-CXCL13, levels of RANKL at day 14 were significantly reduced (P < 0.001 unpaired t test; Fig. 4H), in support of the idea that CXCL13 drives RANKL production in vivo. A significant reduction in RANKL was also observed in anti-CCL21–treated mice (P < 0.001 unpaired t test; Fig. 4H).

These findings led us to hypothesize that RANKL, like CCL21 and CXCL13, might promote metastasis of tumor cells to DLN. Treatment with anti-RANKL neutralizing antibody did not affect the growth of the primary tumor (Fig. 4I). Immunohistochemical analysis of DLN for tumor load revealed no metastasis in the majority of antibody-treated mice (n = 5/7), and the mean area of tumor metastasis was lower in the antibody-treated mice than the controls (P < 0.01, unpaired t test; Fig. 4J). RANKL blockade using a neutralizing antibody did not significantly affect the recruitment of NKp46⁻ILC3 into the primary tumors, but the numbers in DLN were significantly lower, compared with the controls (P < 0.05, one-way ANOVA; Fig. 4K).

We further analyzed the gene expression of RORγt⁺ILC3-associated/lymphoid chemokines CXCL13, CCL19, and CCL21 and their receptors, CXCR5 and CCR7, in a subset of 234 samples of breast cancer from the METABRIC (see Supplementary Table S3 for patient characteristics; ref. 29). Unsupervised hierarchical cluster analysis of the transcriptional profile in these samples revealed that this cohort could be categorized on the basis of their expression of RORγt⁺ILC3-associated/lymphoid chemokines and their receptors (Fig. 5A). “Basal-like” breast cancers (PAM50-intrinsic subtype assignments; ref. 30) presented high expression of these genes (31/53 basal-like tumors lie in the top-branch cluster, n = 89; P = 0.0007, two-tailed Fisher exact test; Fig. 5A). Further cross-validation of these results was seen in four independent breast cancer datasets (Supplementary Fig. S5). RORγt⁺ILC3-associated/lymphoid chemokine and their receptor genes were highly specific (no association with other lymphoid chemokine genes, such as the ligand–receptor pair CCL20–CCR6, which attracts immature DC, effector/memory T cells and B cells) and showed significant internal pairwise correlation (P < 10⁴, Fig. 5B).

We next stained frozen primary tumor sections for markers for RORγt⁺ILC3 (defined as RORγt⁺CD127⁺CD3⁻), as we previously published (20). RORγt⁺ILC3 were present in approximately half of the sections examined (Fig. 5C and D). These cells were in proximity to CD3⁺ T cells (Fig. 5C) and found within tertiary lymphoid structures (TLS), as previously defined (Supplementary Fig. S6; ref. 31). We hypothesized that tumors with higher levels of RORγt⁺ILC3-associated chemokines would have a higher number of RORγt⁺ILC3. To test this, we performed a blinded study in primary breast cancer sections (total patients n = 59). The number of RORγt⁺CD127⁺CD3⁻ cells/mm² (of total area/section) varied considerably from case to case (range, 0–56/mm²; Fig. 5D), but patients with high tumor ILC3 counts were also likely to have a high gene expression score for the ILC3-associated chemokines (Fig. 5D; P < 0.001, Spearman correlation permutation test).

We next assessed the correlation of CXCL13 and CCL21 protein and gene expression levels with ILC3 scores. Fifty-nine cases with known ILC3 scores were stained for CXCL13 or CCL21 expression. For CXCL13, only stromal cells stained for this chemokine (Fig. 5E, i). In contrast, CCL21 staining was positive for both tumoral and stromal cells. We quantified the relationship between ILC3 and stromal CXCL13/CCL21 staining. ILC3 presence correlated positively with CXCL13 staining but not with stromal CCL21 (Fig. 5E, ii). These additional data strengthen our preclinical data (Fig. 2), with CXCL13 upregulation in stromal cells as a secondary event to the recruitment of ILC3 to the primary tumor.

We next stained tumor sections for the lymphatic endothelial cell marker, podoplanin, and evaluated sections for evidence of tumor cell invasion into lymphatics (Fig. 6A, i). We considered lymphatic invasion to have occurred if at least one tumor cell cluster was clearly visible in the lymphatic vascular space (red arrow in Fig. 6A, i). RORγt⁺ILC3 were present in 82% (14/17) of tumor samples with lymphatic tumor cell invasion but only in 27% (8/30) of samples without lymphatic tumor cell invasion. Similarly, 73% (22/30) of samples without lymphatic tumor cell invasion had no RORγt⁺ILC3 present in the tumor, whereas only 17% (3/17) of samples without RORγt⁺ILC3 cells displayed lymphatic invasion (Fig. 6, ii; P < 0.003, Fisher exact test). We did not find an association between lymphatic invasion and CD3⁺ T cells or with CD3⁺CD127⁺RORγt⁺ (most likely representing TH17 cells), strengthening the specificity of the correlation between RORγt⁺ILC3 cells and lymphatic invasion (Supplementary Table S2). We next investigated whether the association between RORγt⁺ILC3 counts and lymphatic invasion translated into a high lymph node tumor burden (i.e., four or more metastatic lymph nodes at surgical resection) within our dataset. In basal-like breast cancer, raised RORγt⁺ILC3 counts were found to also correlate with a higher burden of lymph node metastases (P = 0.02, permutation-based Mann–Whitney; Fig. 6B). Given our in vitro and in vivo findings, we investigated whether lymph node burden was related to gene expression of CCL21 and CXCL13 in the primary tumor. We found that in basal-like breast cancers, a high lymph node tumor burden was associated with significantly increased levels of CCL21 (P = 0.0043; lymph node positive, 4+ vs. 0; two-tailed Mann–Whitney; Fig. 6C). Although CXCL13 levels were also increased in patients suffering from a high lymph node burden, this association was not significant (P = 0.15; Fig. 6D). These correlations were not statistically significant in other breast cancer subtypes (HER2⁺ or luminal A/B), suggesting that the proposed mechanisms may only operate in specific breast cancer subtypes. In a multivariate (Cox) proportional hazards model and taking basal-like and HER2-enriched tumors together, the RORγt⁺ILC3 score achieved 84% prediction accuracy for high lymph node burden, higher than traditional clinicopathologic parameters (e.g., grade, tumor size, receptor status; Fig. 6E).

Recent years have seen a growing appreciation for the pleiotropic nature of the TME (32). The importance of ILC3s in normal lymphoid organogenesis has been accepted for a long time, but their role in the TME has only recently begun to be investigated. Work by Shields and colleagues described, in a murine model of melanoma, a mechanism by which CCL21-expressing tumors recruit ILC3 cells, which transform the TME contributing to a tolerant milieu that promotes immune evasion (21). Study by Eisenring and colleagues showed that RORγt⁺ILCs are required for IL12 to exert its antitumor activity (33). Similarly, a protective function of NKp46⁺ILC3 (distinct from the NKp46⁻ILC3) has been reported in lung cancers (34). These findings are not necessarily at odds, as whether ILC3s promote or prevent cancer progression is likely to depend on the type of cancer and whether they recruit immune cells into a tolerogenic (21) or inflammatory (33) microenvironment. We report the presence of RORγt⁺ILC3 in human breast cancers and that they have a previously unrecognized function in facilitating tumor invasion into the lymphatic system through modulation of the local lymphoid chemokine milieu. We show that ILC3 recruitment into a TNBC tumor model is CCL21 dependent, while CXCL13 regulates their clustering with stromal cells (see Fig. 7).

The CCL21/CCR7 axis plays a role in the progression of different malignancies (35, 36). These studies focus on the direct effects of CCL21 on CCR7-expressing tumor cells, rather than on how CCL21 may modify the TME. We show that CCL21 is expressed both in primary human breast cancers and in a mouse model of TNBC. In the mouse model, the peak of CCL21 expression in tumors was closely followed by ILC3 recruitment, an association we show to be causal through its prevention by CCL21 blockade, consistent with the melanoma xenograft study correlating tumor expression of CCL21 and ILC3 cell recruitment (21).

CXCL13 was not required for ILC3 recruitment to tumors but was important for the induction of ILC3–MSC clustering and RANKL upregulation by MSC. MSCs are recruited to tumors early in development via mechanisms reminiscent of those that operate in chronic wound healing (37, 38). Once activated, they secrete CXCL13, CCL21, and CCL19 and secrete lymphangiogenic factors, such as VEGF-C (39). This interaction, in the TME, may promote neo-lymphangiogenesis, increasing the number of lymphatic vessels into which tumor cells are able to migrate and thus increasing opportunities for lymphatic metastases.

In addition to its role in promoting clustering, CXCL13 stimulates increased RANKL production by MSC. This is likely to facilitate DLN metastasis by promoting EMT in breast cancer cells, enhancing their ability to migrate and metastasize (28, 40). This explains the reduction in serum levels of RANKL observed in anti-CXCL13–treated mice. ILC3 interaction with CXCL13-producing stromal cells may constitute a positive feedback, with CXCL13 reinforcing the ILC3–MSC interaction as shown in vitro and by intravital imaging. These are likely to explain our observations that patients with tumor cell invasion into the lymphatics were more likely to have a higher RORγt⁺ILC3 score compared with patients without lymphatic vessel invasion, as well as the significant association between RORγt⁺ILC3 counts and greater risk of an increased number of lymph node metastasis in basal-like breast cancer patients.

CXCL13 is highly expressed in clinical samples from some breast cancer patients (41), but there is conflicting evidence on how it affects disease progression. Although high CXCL13–CXCR5 expression positively correlates with classical determinants of poor prognosis (41, 42), it serves as a good prognostic marker within this high-risk subgroup of breast cancer patients (42–44). Within the TME, the role of immune cells and/or chemokines is particularly complex (45, 46). Here, we report that downstream of CCL21-mediated recruitment of intratumoral ILC3, CXCL13 promotes lymphatic invasion of tumor cells via the RANK–RANKL signaling pathway. However, CXCL13 is also a powerful chemoattractant for lymphocytes (47). This is in line with our finding that basal-like breast cancers, which frequently bear a prominent lymphocytic infiltrate, presented a high score for the lymphoid chemokine/chemokine receptor gene signature, and our data demonstrating a decreased number of tumor-infiltrating lymphocytes (TIL) in anti-CCL21- or anti-CXCL13–treated mice. In addition, TILs and TLS are described as key prognostic and predictive markers for specific breast cancer subtypes (47–50). Therefore, the well-established role of CXCL13 as a chemoattractant could explain why, in a subset of cases, it seems to play a protective role. In our cohort of patients, CCL21 expression and RORγt⁺ILC3 presence in the primary tumor were associated with increased DLN metastasis in basal-like breast cancer, but not in HER2⁺ or luminal A/B subtypes. It is noteworthy that these data may not translate into worse prognosis for patients, and additional studies are required to fully understand the clinical significance of these findings.

One important consideration for any future development of chemokine-based therapeutic interventions is the interplay between the CCL19–CCL21/CCR7 and CXCL13/CXCR5 axes within tumors. Both have been implicated as important drivers of leukocyte trafficking and lymphoid organogenesis in physiologic situations (51). However, it is important to make a distinction between the two chemokines in the pathologic context of the TME as CXCL13, but not CCL21, is required for ILC3–MSC clustering, which we proposed here to be an important regulatory mechanism in tumor cell migration through RANKL production by MSC.

In summary, we propose that, in our tumor model, ILC3s are recruited to the tumor by CCL21, have a pivotal role in facilitating lymphatic vessel invasion by tumor cells, and they do this via two CXCL13-mediated positive feedback loops. Further investigation into how ILC3, MSC, CCL21, CXCL13, and RANKL are coordinated to establish a network of interactions between the tumor cells and their microenvironment is required.

No potential conflicts of interest were disclosed.

Conception and design: S. Irshad, F. Flores-Borja, S.A. Quezada, A.N.J. Tutt, T. Ng

Development of methodology: S. Irshad, F. Flores-Borja, V. Male, G.O. Fruhwirth, S. Ameer-Beg, L.M. Carlin, D.R. Withers, B. Vojnovic

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Irshad, F. Flores-Borja, J. Monypenny, R. Evans, V. Male, P. Gordon, A. Cheung, P. Gazinska, F. Noor, F. Wong, P.R. Barber, N. Woodman, D. Patel, J. Owen, S.G. Martin, S.E. Pinder, C.E. Gillett, S.P. Poland, F. McCaughan, L.M. Carlin, B. Vojnovic

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Irshad, F. Flores-Borja, K. Lawler, R. Evans, V. Male, A. Grigoriadis, G.O. Fruhwirth, M. Rodriguez-Justo, D.R. Withers, P. Lane, B. Vojnovic, S.A. Quezada, P. Ellis

Writing, review, and/or revision of the manuscript: S. Irshad, F. Flores-Borja, V. Male, G.O. Fruhwirth, S.G. Martin, S.E. Pinder, U. Hasan, S.A. Quezada, P. Ellis, A.N.J. Tutt, T. Ng

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Irshad, N. Woodman, D. Patel, M. Rodriguez-Justo, J. Owen, S. Ameer-Beg

Study supervision: S. Ameer-Beg, A.N.J. Tutt, T. Ng

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.