Tertiary lymphoid structures (TLS) are organized lymphoid aggregates that have been identified at sites of chronic antigen exposure, including organ transplants, localized autoimmune reactions and tumors. Characterized by distinct T and B cell areas and lymph-node-like vasculature expressing peripheral node addressin (PNAd),1 2 TLSs have been identified in primary and metastatic melanoma lesions,3–5 and are associated with improved patient survival and response to checkpoint blockade.3 4 6–13 However, these associations have been based on univariate analyses or multivariate models which did not account for some relevant clinical and immunological variables.4 6 13 Another important limitation of prior studies has been a disproportionate reliance on TLS found in metastases to lymph nodes (LNs)4 5 13 which; as secondary lymphoid organs, may not reflect TLS arising de novo at other sites. Thus, uncertainty persists regarding whether TLS are independently associated with clinical outcomes in melanoma, particularly when associated with non-LN metastases.

TLSs vary in composition and expression of functional markers, and such differences may alter their associations with clinical outcomes. For example, TLSs in melanoma have been associated with improved outcomes, while in breast cancer and hepatocellular carcinoma, TLS have been associated with increased tumor grade and decreased overall survival (OS).14–16 Furthermore, in transplanted lungs, TLS are associated with enhanced tolerance to the transplanted organ.17 These contrasting effects of TLS on immune control may also depend on TLS maturation, which progress from immature early TLS (eTLS), to primary follicle-like TLS (pTLS) and subsequently to secondary follicle-like TLS (sTLS). Maturation stage has correlated with clinical outcomes in colorectal and lung cancers, however a similar assessment has not been conducted in melanoma.4 5 7–9 18

Additional questions remain regarding how lymphocyte activation within the TLS (which are usually peritumoral) relates to lymphocyte activation within the tumor parenchyma and to overall tumor control. Among intratumoral immune cells within melanoma metastases, proliferating T-cells and those differentiated towards Th1/Tc1 phenotypes having been associated with improved survival, while exhausted T-cells and regulatory T-cells (Tregs) have been associated with worse clinical outcomes.19–24 Likewise, our group and others have shown that intratumoral B cell density has been associated with prolonged survival of melanoma patients before the checkpoint blockade era25–27 and with higher response rates to checkpoint blockade.4 5 However, it is unknown whether intra-TLS lymphocyte composition or activation correlates with clinical outcomes; likewise, it is unknown whether intra-TLS lymphocyte function mirrors intratumoral lymphocyte function.

To advance understanding of the prognostic significance of TLS, we evaluated TLS and intratumoral regions for immune infiltrates using multiplex immunofluorescence histology (mIFH) in a homogeneous population of cutaneous melanoma metastases (CMM). We hypothesized that mature TLS and would be associated with improved OS in a multivariate model controlling for clinical variables and intratumoral CD8⁺ T-cell density. We also hypothesized that among patients with TLS, survival would be shorter in those with high proportions of immature TLS, and in those with high proportions of intra-TLS lymphocytes expressing markers of T cell exhaustion (EOMES⁺), Tregs (FoxP3⁺), or CD21⁺CD20⁺ B-cells. Similarly, we hypothesized that among patients with a TLS, high densities of intra-TLS lymphocytes expressing markers associated with immune activation, such as proliferation (Ki67⁺), somatic hypermutation by B-cells (AID⁺), or Th1/Tc1 differentiation by T-cells (Tbet⁺) would be associated with improved outcomes. Finally, we predicted that lymphocyte activity within the TLS would correlate with lymphocyte activity within the tumor.

In this sample of CMM from 64 checkpoint-blockade naïve patients with stage IIIB–IV melanoma, 47% had tumor-associated TLS. This prevalence rate is concordant with a prior report in which TLS were identified in 14 out of 29 (48%) CMM.18 Prior studies of TLS in melanoma have predominantly studied LN metastases4 5; however, the present study excluded LN metastases in the interest of a more homogeneous sample population that avoids potential biases in studying TLS within pre-existing secondary lymphoid structures. Although a prior report did not identify mature TLS in CMM,5 we found that 21% of all TLS were mature (sTLS). TLSs were associated with higher densities of intratumoral CD8⁺ T-cells, B-cells and plasma B-cells, and while there were strong correlations between T and B cell proliferation within TLS, intra-TLS B cell proliferation did not correlate with intratumoral B cells or CD8⁺ T cells proliferation. Taken together, these data suggest that overlap may exist between the factors that promote TLS formation and those that drive intratumoral lymphocyte invasion, as TLS^pos lesions were more densely infiltrated by lymphocytes at the tumor center. Nonetheless, lymphocyte activity within the TLS was not correlated with lymphocyte function within tumor parenchyma, implying that intratumoral lymphocyte activity is regulated by factors distinct from those which regulate intra-TLS lymphocyte activity. With respect to clinical outcomes, TLS⁺ CMM were independently associated with improved OS in a multivariate model incorporating key clinical variables and intratumoral CD8⁺ T-cell infiltration (HR=0.51, p=0.04). In a subgroup analysis of only TLS⁺ lesions, higher fractions of intra-TLS B-cells undergoing somatic hypermutation and decreased fractions of CD21⁺CD20⁺ B-cells were associated with improved OS in separate multivariable hazard models, strongly suggesting that TLS may exert either immunogenic or immunoregulatory effects on tumor control, which are driven by intra-TLS B-cell phenotype.

TLSs are defined as lymphoid aggregates that are segregated into B- and T-cell compartments and are associated with vascular endothelial cells expressing PNAd, a ligand that is also expressed on HEV in secondary lymphoid organs.1 TLS neogenesis has been linked to secretion of pro-inflammatory cytokines, notably CXCL13. CXCL13 is a proinflammatory chemokine which recruits B cells to peripheral tissues and is upregulated in TLS-associated colorectal cancers and expressed by perivascular stromal cells in TLS-associated lung squamous cell carcinoma.2 8 33 In separate studies, our group is investigating the mechanistic role of fibroblasts in TLS generation both in murine models and in pilot studies of CXCL13 expression by fibroblasts in human TLS. These pilot data suggest that CXCL13 positive cells are more abundant in TLS than the tumor parenchyma.34

As described previously, TLS may be further classified as eTLS, consisting of B-cell and T-cell aggregates, pTLS, which are characterized by networks of FDC, or sTLS, which demonstration germinal center formation, and mimic the architecture of a germinal center within a true secondary lymphoid follicle.4 7 8 10 Previously, TLS maturation was correlated with reduced tumor recurrence in colorectal cancer and with improved survival in chemotherapy-naïve lung cancer patients.7 8 In melanoma, TLS with germinal centers (sTLS) were previously identified almost exclusively in TLS associated with LN metastases, which is likely reflective of the fact that earlier works were performed using samples from heterogeneous tissue types and included only a very small number of CMM, limiting TLS characterization in these lesions.4 5 Thus, these data represent the largest characterization of melanoma-associated TLS maturation outside the LN. Most TLS were eTLS; however, a significant fraction (21%) of TLS in these CMM were sTLS.

Patients with TLS⁺ metastases had longer OS in a multivariate analysis controlling for clinical variables and intratumoral CD8⁺ T-cell density. These patients were all treated prior to the checkpoint blockade era; thus, patient outcomes reflect spontaneous antitumor immune responses. This extends an earlier report wherein TLS⁺ lesions were associated with improved OS in a multivariate analysis only when tumors were also densely infiltrated by intratumoral CD8⁺ T-cells.5 As intratumoral CD8⁺ T-cell density is a well-recognized marker of improved prognosis in melanoma and other cancers,28 35–38 it would be valid to question whether differences in survival associated with TLS⁺ tumors that were also densely infiltrated by CD8⁺ T-cells can be attributed to the TLS themselves or are simply a reflection of intratumoral CD8⁺ T-cell density.

Importantly, our multivariate models controlled for CD8⁺ T-cell tumor infiltration and for surgical intent, which was not incorporated into the model used in earlier reports. In patients with metastatic disease, surgical intent can be either curative, in which all known metastatic lesions are removed at time of surgery, or palliative, in which known metastatic disease is left unresected during metastasectomy. It has been well-documented that complete resection, reflective of curative surgical intent, correlates with patient outcomes after metastasectomy39 40; thus, our finding that TLS⁺ lesions tended to be more commonly resected with curative intent (70% vs 50%, p=0.17) suggests that multivariate models assessing associations between TLS and patients outcomes should control for surgical intent in patients with metastatic disease. After controlling for surgical intent, patient age, patient sex and disease stage and intratumoral CD8⁺ T-cell infiltration, we confirmed that the presence of TLS was independently associated with improved OS.

It has been suggested that TLS may improve outcomes by enhancing the activity of intratumoral CD8⁺ T-cells.5 In the present study, TLS⁺ CMM were more densely infiltrated by intratumoral CD8⁺ T-cells, B-cells and plasma B-cells, which have been previously associated with improved outcomes in melanoma.17 This may indicate that TLS support antigen-driven proliferation of the T- and B-cells that infiltrate the tumors associated with them. Intriguingly, the density of intratumoral CD8⁺ T-cells correlated with increasing fraction of sTLS but not with fraction of eTLS or pTLS, suggesting that TLS maturation may be correlated with cytotoxic T-cell infiltration into the tumor center. However, it is notable that proliferation rates of intra-TLS lymphocytes were not correlated with proliferation rates of intratumoral lymphocytes, suggesting that factors driving proliferation of T-cells and B-cells within TLS differ from those infiltrating among tumor cells.

We hypothesized that among TLS⁺ metastases, markers of immune activation or regulation within the TLS would be associated with better and worse patient outcomes, respectively. Specifically, we hypothesized that increased intra-TLS lymphocyte proliferation, T-cell differentiation to a CD4⁺ Th1 or CD8⁺ Tc1 phenotype and increased somatic hypermutation by B-cells would be associated with improved OS. Contrary to these hypotheses, the fractions of B-cells and T-cells proliferating within the TLS were not associated with changes in OS, which is consistent with prior findings.5 Also, intra-TLS Th1-dominance of CD4⁺ T-cells populations and Tc1-dominance of CD8⁺ T-cell populations were not associated with associated with improved OS. The fraction of B-cells undergoing somatic hypermutation, however, was significantly associated with improved OS.

Others have demonstrated that B-cells in CMM-associated TLS express AID and undergo both antibody class switching and clonal amplification, which are the hallmarks of antibody affinity maturation.18 Affinity maturation occurs in B-cells undergoing somatic hypermutation (AID⁺) and is dependent on duration of antigen presentation by local antigen presenting cells in the presence of helper T-cell stimulation.41 B-cells within the tumor-adjacent TLS are colocalized with FDC and CD4⁺ T-cells and are likely exposed to higher concentration of tumor antigen than B-cells in tumor draining LNs. Consequently, there is reason to speculate that antibody affinity maturation within CMM associated TLS could produce more effective antitumor antibodies than antibodies produced by B-cells in tumor draining LNs. Interestingly, even in a small study, wide variation in the percent of intra-TLS B-cells expressing AID (0%–60%) has been observed.18 In the present larger study of TLS-associated CMM, we confirm that intra-TLS B-cells express markers of somatic hypermutation and that expression is highly variable across TLS. Furthermore, we expand on this prior work, demonstrating that TLS⁺ patients with a high proportion of intra-TLS B cells undergoing somatic hypermutation had improved OS relative to other TLS⁺ patients, and that this association persisted in a multivariate model controlling for patient sex and surgical intent.

We also investigated the possibility that some TLS may have immune suppressive or regulatory features, as others have also theorized that some TLS may serve as immunotolerant niches harboring malignant cells, particularly if the TLS is enriched with cell types that promote immune tolerance or exhaustion.9 12 42 We found broad variance among patients in the intra-TLS proportions of several putative immunoregulatory cell types, including Tregs, CD21⁺CD20⁺ B cells, and CD8⁺EOMES⁺ T cells, which can be associated with T-cell exhaustion.32 43 We hypothesized that high proportions of these cell types would be associated with worse patient outcomes. In contrast to our hypothesis, density of intra-TLS Treg cells was not associated with patient outcomes. High proportions of EOMES⁺CD8⁺ cells were correlated with worse OS in a univariate analysis, but this was not significant in multivariate analysis. However, CD21⁺CD20⁺ B-cell fraction was strongly correlated with worse OS and was significant in a multivariate proportional hazards model controlling for patient sex and curative-intent resection.

The significance of CD21 as a B-cell marker in cancer is poorly understood. In melanoma patients, others have noted that intratumoral B-cells exhibited lower CD21 expression than peripheral blood B-cells; however, no specific assessment was made of the function or prognostic significance of these cells.4 Intriguingly, it has recently been demonstrated that circulating CD21^low B-cells are upregulated in melanoma patients after the initiation of checkpoint blockade, and that these B-cells were a more clonal population than circulating CD21⁺ B-cells.44 In autoimmune disease, low CD21 expression by B-cells is associated with decreased fractions of T-reg cells in patients with common variable immune deficiency (CVID).45 Decreasing fractions of B-cells expressing CD21 correlated with severity of autoimmune reactions in patients with rheumatoid arthritis, systemic lupus erythematosus, CVID and Sjogren’s syndrome.30 46 47 These B-cells have been found to home preferentially to peripheral tissues that are targeted in autoimmune conditions, such as the synovial fluid of patients with rheumatoid arthritis.3 Given the established role of CD21⁺ B-cells in regulating humoral autoimmune disease, and the increase in circulating CD21^low B-cells after checkpoint blockade, it is tempting to speculate that CD21⁺ B-cells may also play a role in attenuating the antitumor immune response in TLS, and our findings would be consistent with this hypothesis.

Many questions remained unanswered regarding how TLS enhance or inhibit tumor control in CMM. TLS have been found to predict response to checkpoint blockade and were associated with improved survival in patients who received checkpoint blockade therapy.4 5 As samples from our study predated the checkpoint blockade era, it would be worthwhile to investigate whether the associations between intra-TLS B-cell phenotypes and patient survival identified herein also correlate with patient response to checkpoint blockade. Likewise, the possibility that TLS maturation may influence response to checkpoint blockade, or the activation states of these intratumoral lymphocytes, have yet to be explored.

We observed a strong association between patient outcomes and fraction of B-cells expressing CD21⁺ within TLS, identifying a potential subset of TLS which may contribute to immune regulation and inhibition of tumor control. Similarly, the favorable prognostic association of AID expression by B cells in TLS suggests that antibodies produced by B cells in TLS may enhance immunologic tumor control. For all of these, further work is needed to elucidate the mechanisms underlying these associations. It is tempting to speculate that novel therapies targeting these B-cell populations, in conjunction with checkpoint blockade, may improve outcomes in patients with TLS-associated advanced melanoma.

In summary, this study provides the first detailed assessment of a population of patients with non-LN metastases evaluated for TLS and their clinical associations. Practically speaking, CMMs are common in melanoma and are readily accessible for biopsy, and therefore could provide useful windows to the host-tumor relationship. TLS were identified in 47% of lesions, and were peritumoral in location, but were associated with increased intratumoral lymphocyte infiltration. However intra-TLS lymphocyte function was distinct from intratumoral lymphocyte activity. A significant fraction of the identified TLS had matured into pTLS or sTLS. The presence of TLS was associated with improved OS in a multivariate hazard models controlling for surgical intent and intratumoral CD8⁺ T-cell activity. Intra-TLS B-cell activity was also strongly correlated with OS in multivariate hazard models, with high fractions of B-cells undergoing somatic hypermutation or expressing CD21 associated with better and worse patient outcomes, respectively. These data suggest that each TLS may exist in a unique immunological niche which is distinct from the intratumoral niche, and influence of this niche on B-cell activity within the TLS is a strong predictor of systemic tumor control.