WHAT IS ALREADY KNOWN ON THIS TOPICWHAT THIS STUDY ADDS

This study reports the first assessment of heterogeneous presence of lymphoid structures in primary and metastatic melanomas involving skin/subcutaneous/soft tissues. These lymphoid structures exhibit diversity in their morphologic appearance on H&E assessment, and varied cellular composition based on multiplex immunofluorescence analyses, but consistent expression of the 12-chemokine gene expression score. Notably, these features are negatively correlated with pigmentation/neural network gene signatures. Furthermore, melanomas tend to present with predominantly immature lymphoid structures in mixed locations, which appear related to improved overall survival and the absence of lymph node involvement in patients with primary melanomas.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

The provided histological assessment of lymphoid aggregates (LA) and TLS can be implemented in clinical practice through the assessment of H&E-stained samples. Description of heterogeneous LA structures and differential presence in primary versus metastatic skin/soft tissue tumors provides a basis for studying the impact of future immunotherapy options on LA/TLS neogenesis and/or the evolution of existing LA/TLS in the melanoma tumor microenvironment. The findings for negative correlations between pigmentation/neural networks warrant further investigation to discern underlying mechanisms of action in inhibiting the formation and maturation of lymph node structures in the melanoma microenvironment.

Introduction

Tertiary lymphoid structures (TLS) represent lymph-node-like structures that conditionally form in peripheral organs impacted by chronic disease, with their presence being associated with improved patient prognosis in the solid tumor setting.1–8 Historically, TLS have been characterized based on the presence of germinal centers (GC), B and T-cell infiltrates which form organized structures along with specialized follicular dendritic cells (FDCs) and high endothelial venules (HEVs).9 However, it has become increasingly apparent that tumors often exhibit heterogeneous degrees and patterns of immune cell infiltration, leading to the belief that clusters of immune infiltrates/aggregates may represent precursor states of TLS in situ.10 The degree to which clinicopathologic features are correlated with immune infiltrates is largely unexplored, but there is evidence that they do exert some influence. For example, the presence, location and composition of TLS have been shown to vary by melanoma subtype, with desmoplastic melanoma exhibiting a higher frequency of TLS with differences observed for location(s) within the tumor microenvironment (TME), and proliferation status when compared with other histological subtypes of melanoma.11

While TLS have received the lion’s share of attention, they appear to develop relatively rarely in primary melanomas, which has limited their ability to predict prognosis and treatment response in a clinically meaningful manner. Recognizing that most melanomas exhibit some degree of lymphoid infiltrate, a systematic approach for scoring tumors for lesser degrees of inflammation is ripe for exploration. Immune infiltrates characterized by either T or B-cell clusters are called lymphoid aggregates (LA). Immature/early TLS contain B and T cells along with dendritic cell (DC)-LAMP+ mature DC. Mature primary follicle-like TLS are characterized by the presence of CXCR5+PD1+ T follicular helper (TFh) cells and CD21+ FDC. Secondary follicle-like TLS are characterized by the presence of BCL6+ B and TFh cells, CD21+CD23+ FDCs, and high-affinity antibody-producing plasma B cells within GC. While the use of mature TLS as biomarkers for patient with cancer outcomes has been extensively studied, the prognostic impact of LA and immature TLS have only been partially elucidated with early TLS even being proposed as mediators of local immunosuppression in support of tumor progression.2 9

Several gene expression scores have been promoted as being prognostic factors for the tissue presence of immune infiltrates.12–14 One of these is a 12-chemokine (12-CK) gene expression score including CCL2-5, CCL8, CCL19, CCL21, CXCL9-11, and CXCL13 that has been previously reported to serve as a predictor for TLS presence in colorectal and melanoma specimens in association with improved patient overall survival (OS).6 14

In the current study, we assessed pathologic, transcriptomic and prognostic characteristics of LA in clinical specimens of primary and metastatic melanoma. We performed a histological assessment to determine the degree, pattern, and location of lymphoid infiltrates in tumor tissue. We then correlated these findings with clinical and pathological characteristics of melanomas, with a further resolution of cellular composition and transcriptomic signatures used to discriminate the prognostic metrics of lymphoid structures in patients with melanoma. We observed a differential presence of LA and TLS in primary versus metastatic melanomas consistent with their favorable roles in patients with melanoma in support of extended OS. In primary melanomas, significant associations were noted between the presence of LA and clinicopathological characteristics of tumors and tumor-free regional lymph node status.

Material and methodsData source and study population

Patient specimens were obtained from the Biospecimen Repository of the Melanoma Center of the University of Pittsburgh Medical Center Hillman Cancer Center under the institutional review board-approved University of Pittsburgh Cancer Institute Melanoma Center Human Biological Sample and Nevus Image Banking and Analysis Protocol (UPCI 96-99). Patients with a histological diagnosis of melanoma, known male/female status, age ≥18 years old were included in this study. Informed consent was obtained prior to biological specimen collection and banking. Demographic data (age, sex), clinical data (stage, regional lymph node status), pathological data (histological subtype, Breslow thickness, presence/absence of regression) were collected from electronic medical records and pathological reports. The Cancer Genomic Atlas skin cutaneous melanoma cohort (TCGA-SKCM) was used for the identification of primary and metastatic melanoma samples, collection of clinical and pathological data, and tissue transcriptomic analyses.

Histological assessment of LA and TLS

H&E slides were reviewed by two board-certified dermatopathologists (AL/DVDS and JM), who classified the lesions according to the following established histologic criteria: LAs were defined as (1) clusters, where groups of ≥50 lymphocytes were present, but were irregularly shaped; (2) nodules, if presenting as circular/spheroid aggregates of >50 lymphocytes with concentric placement, and (3) TLS, if present as nodular LA containing GCs, (4) absent, if none of the above. LA and TLS localization were categorized as peritumoral if present at the invasive margin of tumor, stromal if separated from the tumor by normal stromal tissue, and intratumoral if present within the tumor parenchyma (online supplemental figure 1A–F).

Multispectral immunofluorescence and image analysis

Fresh tissue samples were formalin-fixed, embedded in a paraffin block, sectioned (4 microns), and mounted onto slides. Automated staining of tissues was performed on the Leica Bond RX. For staining, Akoya Bioscience’s Opal 6-Plex Detection Kit (cat# NEL871001KT) was used according to the manufacturer’s instructions. Cells were stained with the following primary antibody/conjugated opal pairs: CD8/Opal480, CD21/Opal520, CD4/Opal570, SOX10/Opal620, PNAd/opal690, CD20/Opal780 (n=24) (online supplemental figure 2). The Akoya Bioscience’s PhenoImager HT platform and inForm analysis software were used for whole slide scanning, spectral unmixing, and phenotyping. Multi-layer tag image file format (TIFF) images were exported from inForm (Akoya) and loaded into HALO (Indica Labs, New Mexico, USA) for segmentation and quantitative analysis. Each tissue image was segmented into individual cells using the 4, 6-Diamidino-2-Phenylindole (DAPI) marker that stains cell nuclei. For each fluorescent marker, a binary status (positive or negative) was determined by comparing the fluorescent intensities against an established threshold for that specific antibody. The comparisons of cellular composition between nodule and cluster for marker genes CD8, CD20, CD21 and PNAd were performed using the Wilcoxon rank test. Comparisons of cellular composition among intratumoral, peritumoral, and stroma for marker genes CD8, CD20, and PNAd were performed using the Wilcoxon rank test. The correlation score and p value among CD8, CD20, and CD21 cells were calculated using the Spearman method (n=24). We further assessed cell local entropy among cell types using the entropy gradient function method in the R package SPIAT. This program computes entropy as a function of radius and determines the gradient for small radii. We assessed the cell colocalization between two cell types using area under the curve of the cross-K function (AUC) method in the R package SPIAT. Ripley’s K function is used to determine if the spatial distribution of point patterns across a study area is clustered, dispersed or randomly distributed. The cross-K function is a generalization of Ripley’s K function for multitype point patterns that summarizes the spatial relationship between two or more types of points. Comparing the observed cross-K function with the expected cross-K function gives the spatial relationship between the two-point processes. To quantify the patterns, we calculate the area between the curve of the cross-K function and the expected Poisson Process E by subtracting the AUC, resulting in a metric, DAUC. DAUC was then normalized by the total area of the cross-K plot, enabling comparisons of this metric between images of different sizes. If DAUC is positive, the observed cross K curve is above the Poisson process curve and the two-point processes are aggregated and indicative of high colocalization. If DAUC is negative, the observed cross K curve is below the Poisson process, indicating separation of the patterns and low colocalization of signals.

Gene expression profiling in TCGA-SKCM melanoma specimens and functional enrichment analyses

Melanoma patient clinicopathological and messenger RNA expression data were retrieved from the TCGA-SKCM pan-cancer atlas version using the cBioPortal database (https://www.cbioportal.org/). Gene expression profiles were compared between samples with and without LA/TLS. Significant differentially expressed genes (DEGs) were identified based on an adjusted p value (padj)<0.05 and a fold change >2 (n=1,165). The top 100 highly expressed DEGs were selected for the generation of an expression heatmap. Expressed genes were then ranked by padj value and fold change using the formula −log10(padj+0.01)×log2FC. Gene Set Enrichment Analysis was conducted by comparing the pre-ranked gene list against the Hallmark, C5, and C7 pathways from the MSigDB database. Statistical significance for term enrichment was set at a Benjamini-Hochberg padj<0.05. The previously validated 12-chemokine gene expressions score was calculated using the first principal component of the log2-transform gene expression and compared between groups using the Wilcoxon test, as described previously.13 14

Statistical analysis

Categorical data were summarized by proportions and percentages. Continuous data were summarized by their mean, SD, median and IQR. A Fisher’s exact test was used to determine if an association existed between lymphoid aggregate and regional lymph node status, LA and disease type (primary/metastatic), LA and histology type, LA and regression (yes/no), and TLS and melanoma type (primary or metastatic). To determine the association between LA and Breslow thickness, the distribution of Breslow values was examined with a boxplot and Shapiro-Wilk test and determined to be not normal. After performing a log transformation on Breslow values, the data were normally distributed. An analysis of variance table was then used to compare the means of log (Breslow) across lymphoid aggregate types. For the association between OS and LA, TLS with GC, or LA and TLS location, a univariate Cox proportional hazards model was used for each variable. Multivariable Cox proportional hazard models were used to examine associations between survival endpoints and LA presence, LA type or TLS while controlling for covariates such as age, sex, stage, and sample type.

ResultsPatient with melanoma

We screened a total of 670 samples for eligibility and excluded 392 samples due to their being obtained from metastatic melanoma involving lymph node, or harvested after neoadjuvant therapy, or those obtained through core biopsy or from specimens of frozen tissue samples only, or those of inferior quality for interpreting digital images (TCGA SKCM). In total, we analyzed 278 samples from the TCGA SKCM (n=149) and the University of Pittsburgh Medical Center (UPMC) melanoma cohort (n=129). These included primary melanomas (n=195) and melanomas metastatic to skin/subcutaneous/soft tissues (n=83) (online supplemental figure 3). Key baseline characteristics are presented in online supplemental table 1. Based on histological assessment, LAs were present in 72% of cases, while TLS with GC were detected in only 11% of cases. The presence of LA and TLS was significantly associated with sample type. Samples with the presence of LA were predominantly identified in primary compared with metastatic melanomas, and TLS were more common in metastatic compared with primary melanomas (p=0.02) (figure 1A,B). LA presented mainly as clusters only (n=92, 46.2%), with fewer presenting as nodules only (n=34, 17.1%), and the remaining (n=72, 36.7%) samples containing both types of LA (figure 1C). LA were most commonly present in multiple locations in the TME (n=103, 51.7%), followed by being located only in peritumoral areas (n=70, 35.2%) (figure 1D). In most cases, TLS were located peritumorally (n=14, 45.2%) followed by their detection in stromal locations (n=11, 35.5%) (figure 1E). In metastatic skin/subcutaneous tumor samples with surrounding adipose tissue, TLS with GC were present in 10 out of 15 (67%) cases and primarily located within the tumor surrounding adipose tissue (figure 1F).

Figure 1

Overview of lymphoid aggregate (LA) and tertiary lymphoid structure (TLS) characteristics based on H&E assessments. Stacked bar plots comparing the percentages of (A) LA and (B) TLS among primary and metastatic samples. (C) Bar plot displaying LA subtypes. Bar plots showing locations of (D) LA and (E) TLS. (F) H&E slide showing subcutaneous nodule of melanoma, which is surrounded by adipose tissue, red arrow indicates TLS forming within adipose tissue.

Prognostic significance of lymphoid structures and their relation to clinicopathological characteristics among primary melanomas

The presence of any type of LA was associated with improved OS (log-rank p value=0.04) (figure 2A). In multivariable analysis, after adjusting for age, sex, stage, and sample type, LAs continued to be associated with improved OS (HR=0.52, 95% CI (0.31 to 0.87), p=0.01) (figure 2B). The prognostic significance did not vary based on the subtype of LA (cluster vs nodule vs both, p=0.6) (figure 2C) or location of LA (intratumoral vs mixed vs stromal vs peritumoral, p=0.1) (figure 2D). The presence of TLS with LA was not associated with a significant improvement in OS in comparison to LA only after adjusting for age, sex, stage, and sample type (p=0.3) (figure 2E). Of note, the presence of LA was associated with negative regional lymph node (RLN) status (70.3% vs 29.7%, p=0.02) (figure 3A) and with histological subtype of melanoma (p=0.01). LAs were more prevalent in nodular melanomas and less prevalent in the lentigo maligna subtype. LAs trended to be related to tumor regression (p=0.054) but not Breslow thickness (p=0.36) (figure 3B–D).

Figure 2

Prognostic significance of lymphoid structures in relation to overall survival. (A) Kaplan-Meier plot comparing overall survival probability between patients with LA and those without LA. Cox proportional hazard regression models for overall survival adjusting for (B) LA presence, age, sex, stage, and sample type, (C) LA subtype, (D) LA location. (E) Hazard regression plot comparing overall survival between patients with LA with TLS and LA only with adjustment for age, sex, stage, sample type. IT, intratumoral; LA, lymphoid aggregates; PT, peritumoral; TLS, tertiary lymphoid structures.

Figure 3

Assessment of lymphoid aggregates in primary melanomas in relation to regional lymph node (RLN) status and histopathological features. (A) RLN positivity, p=0.02, (B) histological subtype, p=0.01(C) regression, p=0.05 and (D) log-transformed Breslow thickness, p=0.36. RLN, regional lymph node, SS, superficial spreading.

Cellular composition of lymphoid structures based on subtype and location within the TME

Multiplex immunofluorescence analysis of lymphoid structures revealed marked heterogeneity in the presence of lymphoid structures within the TME, and their cellular composition, ranging from predominant CD8+ tumor infiltrating lymphocyte to TMEs with varied cellular composition in lymphoid structures (figure 4A,B). Accounting for both type of structure and location, CD8+ T cells trended to be predominantly present in intratumoral clusters (p=0.07), whereas CD20+ B and CD21+ DCs were more often found in peritumoral nodules (p=0.003 and p=0.004, respectively), with no statistically significant difference observed for the presence of PNAd in the various lymphoid structures (figure 4C–F). Accounting for the location of lymphoid structures only, we noted significant differences in cell distributions. CD8+ T cells were primarily located peritumorally versus in stromal locations (p=0.008), CD20+ B cells exhibited primarily stromal (p=0.005 and p=0.045 vs peritumoral and intratumoral, respectively) positioning, with no significant location differences noted for PNAd expression (figure 4G–I). CD20+ B cells were significantly correlated with CD8+ T cells in the TME (p<0.0001) (figure 4J). Further assessment of cell colocalization over whole tissue revealed a high degree of colocalization for CD20+ B cells and CD4+ cells, and for CD20+ B cells with CD8+ T cells across all samples, presenting with positive DAUC scores, with the observed cross K curve falling above the Poisson process curve. B cells were also colocalized with PNAd+cells in 20 out of 21 cases where both cell types were detectable (individual case presentation, figure 5A–D). These results highlight the finding that within the TME, B cells are often coordinately colocalized with T cells and PNAd+ endothelial cells as opposed to being observed as isolated, pure cell populations. Regarding the relationship between CD20+ B cells and SOX10+ melanoma cells, their colocalization was observed in 18 out of 24 (75%) cases, which likely reflects structures formed in the stroma of a single case where there was no colocalization, the DAUC score was negative, and the observed cross K curve falls below the Poisson process (individual case presentation, figure 5E). The localized entropy plots, with the color (from green, yellow, orange to pink) indicating the entropy value are illustrated for an individual case in figure 5F–H, I, demonstrating the colocalization of mixed clusters of CD20+ B cells and CD4+ T cells, CD20+ B cells and PNAd+ cells, and CD20+ B cells and CD21+ DCs.

Figure 4

Cellular composition of lymphoid structures and heterogeneity based on subtypes, location evaluated by multiplex immunofluorescence assessment of lymphoid infiltrates. (A) Multiplex immunofluorescence analysis of lymphoid structures. Stack bar plots represent the proportion of each cell. Layer 1 represents the whole slide, and each region of interest was annotated and presented in x axis. (B) Heatmap represents the proportion of cell types for each slide across cluster and nodule from intratumorally (n=17), peritumorally (n=72), and stromal (n=42) annotated regions of interest. (C) Box plots comparing the composition of CD8+T cells, (D) CD20+ cells, (E) CD21+ cells, (F) PNAd across subtypes and location of lymphoid aggregates. (G) Box plot shows the distribution of CD8 cells at IT, (n=17), PT (n=72), and ST (n=42). (H) CD20 cell distribution (I). PNAd distribution. The p value was calculated between the two groups by using the Wilcoxon rank test. (J) Correlation between CD20 and CD8 counts over the whole slide, Spearman correlation coefficient and p value are depicted over figure showing strong positive correlation. IT, intratumoral; PT, peritumoral; ST, stromal.

Figure 5

Cell colocalization quantified using the cross-K function method for an individual case using CD20 as a reference and measurement of entropy. The black line represents the input image, the red line represents a randomly distributed point pattern. With the increase of radius (x axis), the black line diverges further from the red line, meaning that there is at least one mixed cluster of two types of points, which is also suggested by the positive value of the AUC score for (A) CD20-CD4, (B) CD20-CD8, (C) CD20-PNAD, and (D) CD20-CD21. (E) For CD20-SOX10, the AUC score was close to 0, suggesting the two types of cells do not exhibit a positional relationship. Entropy measurement based on CD20 as reference (F) with CD4 (G), CD21 (H) or PNAD (I). Higher entropies (blue) are obtained when the cell types considered are balanced and suggest high cell colocalization, and lower entropies arise when a particular cell type(s) is rare and indicate low cell colocalization. AUC, area under the curve.

Transcriptomic profiling and functional Gene Set Enrichment Analyses of melanoma samples with and without LA

A heat map of the top 100 DEGs is depicted in figure 6A. Consistent with the known role of antibody production and antigen presentation capacity of B cells within the TME, we noted that Ig genes and major histocompatibility complex (MHC) class II genes were among the highest-ranking DEGs. MSigDB hallmark gene set analyses for the upregulated genes linked to LA revealed enrichment for hallmark pathways such as interferon gamma response, IL6_JAK_STAT3 signaling, TNFA_signaling via NFKB and NOTCH_signaling, WNT_BETA_CATENIN_signaling, epithelial_mesenchymal_transition, hypoxia, hedgehog signaling pathways represented the major downregulated gene pathways Gene Onctology (GO) analyses revealed several terms in downregulated genes including the BP terms “synapse organization”, “pigmentation”, “pigment cell differentiation” “developmental pigmentation”, and “sensory_system_development”, the CC terms “glutamatergic_synapse” and “presynaptic_active_zone” and the MF term “voltage_gated_calcium_channel_activity”. REACTOME pathway analyses identified upregulated DEG pathways (MHC class II antigen presentation, cell surface interactions at the vascular wall, CD22_mediated BCR-regulation pathways) and coordinate downregulated genes pathways including protein_protein_interactions at synapses, signaling by NOTCH1, GABA_receptor activation, signaling by WNT in cancer, glutamate_neurotransmitter release cycle, among others (figure 6B). A full list of pathways enriched in up/downregulated genes is presented in online supplemental table 2. Notably, we observed that the 12-CK gene expressions score was significantly higher in samples with evidence of LA with or without TLS versus none (p<0.001) (figure 6C).

Figure 6

Transcriptomic profiling and functional Gene Set Enrichment Analyses of melanoma samples with and without LA. (A) Heatmap illustrating differentially expressed genes in samples with/without LA. (B) Bar plots showing upregulated and downregulated MSigDB hallmark gene sets, Gene Ontology pathways, and REACTOME pathways in LA samples. (C) Box plot showing 12-chemokine score in samples with LA±TLS which are combined in LA group and without LA (No-LA). Table further shows 12-CK score in LA and TLS groups, separately. P value was calculated using the Wilcoxon rank-sum test. LA, lymphoid aggregates; TLS, tertiary lymphoid structures.

Discussion

TLS have attracted significant attention over the last decade as important components of the tumor immune microenvironment. They appear predictive of favorable patient prognosis or improved response to immune checkpoint blockade therapy. Notably, many previous studies focused solely on surveys of mature TLS with GC (aka secondary follicles) to link with patient outcomes. In a neoadjuvant cohort of patients with resectable melanoma treated with anti-programmed-death-1 (PD-1)-based immunotherapy, B cells and TLS were most strongly associated with superior treatment response.2 Another study reviewing the prognostic role of TLS in patients with metastatic melanoma concluded that the presence of TLS in the melanoma TME is associated with improved patient OS.3 These studies assessed the role of TLS predominantly in the setting of metastatic lymph node samples. In patients with solid tumors treated with anti-PD-1 therapy, the presence of mature TLS appeared to predict the antitumor efficacy of the intervention. In that study, a range of solid tumor types were investigated for the presence of TLS in both tumor core biopsies and surgical excision samples.15

Our study is the first to systemically characterize, quantitate, and analyze the prognostic value of LA in melanoma, highlighting their favorable correlation with patient OS. Histological and morphological assessments of immune infiltrates in melanoma specimens revealed distinct shapes, patterns, and locations of lymphoid structures within the TME. Histologically assessed TLS with GC were identified predominantly in metastatic sites of skin/subcutaneous/soft tissue with an incidence of 11% among all samples evaluated. LA were predominantly found in primary melanoma samples, where they were associated with histological subtype of melanoma and tumor-negative regional lymph node status. The latter finding is of particular interest since it supports the hypothesis that lymphoid structure formation in primary tumors may alter the trafficking of tumor to regional lymph nodes. One potential mechanism that requires further investigation is that local production of LN chemokines in LA may distract tumor cells from exhibiting tropism to LN via lymphatic vessels. Another plausible mechanism is that TLS are playing an active role in the superior immune-mediated control of tumor metastasis from the regional TME to distal sites.

Intriguingly, we observed that mature TLS with GC in advanced melanoma often form within adipose tissues surrounding the tumor nest, a finding that warrants further investigation of the cytokine/chemokine milieu of adipose tissues, which may predispose the local TME for formation and maturation of lymphoid structures. This finding is of particular interest when considered in the context that obesity is associated with improved progression-free survival and OS in patients with metastatic melanoma, with the underlying mechanism(s) for the obesity paradox representing an area of intense clinical interest and active investigation.16 To further illuminate the modulatory impact of adipose tissue on the formation and maturation of TLS, the downstream effects of adiponectin which is a cytokine produced by adipose tissue, was reviewed using the Immune Dictionary Immune Response Enrichment Analysis software (https://www.immune-dictionary.org/app/home). Top upregulated and downregulated genes in B cells following treatment with adiponectin are shown in online supplemental figure 4. Notably, top genes upregulated by adipokines were related to immunoglobulin production (ie, Ighg2b, Jchain, Ighg2c, Iglc1).

In this study, it was of keen interest to study heterogeneity in the location(s) in which lymphoid structures formed within the TME. Our results show that the cellular composition of lymphoid structures varies based on the location of LA within the TME, with B cells predominantly identified in the stromal versus intratumoral or peritumoral space. However, due to the high percentage of cases containing mixed locations of LA, it is difficult to assess the unique prognostic impact of LA found in specific locations within the melanoma TME. In fact, our data did not reveal any prognostic significance for LA based on their specific sublocalizations within the TME. With that being said, the question remains as to how these structures transcriptionally/functionally vary based on their disparate locations in the TME. CD20+ B cells were positively correlated with CD8+ T cells in the TME. Further assessment of the colocalization of cells confirmed the finding that CD20+ B cells in the TME tend to colocalize with both CD4+ T cells and CD8+ T cells along with PNAd+ vascular endothelial cells in forming lymphoid structures as opposed to being present as isolated CD20+ B-cell populations randomly dispersed throughout the TME.

DEG analyses performed comparing LA and non-LA containing samples appear to highlight the negative involvement of neural networks and neurotransmitter release pathways in lymphoid structure development/maintenance in the melanoma TME. This may provide some insight into the potential inhibition of lymphoid structure formation driven by perineural invasion, a concept that clearly requires further prospective study for validation. Notably, in translational melanoma models it was recently shown that sensory neurons inhibit TLS formation in cutaneous sites of disease.17 Another interesting finding in our study was the selective reduction of pigmentation pathway-associated DEGs in the presence of lymphoid structures within the TME. Whether this finding is due to increased local expression of lymphoid structure-associated interferon genes known to inhibit expression of pigmentation-associated genes in melanocytes/melanoma or to the possibility that pigment/melanin may have a negative impact via soluble mediators on LA/TLS formation and maturation also warrants future investigation.

Among differentially expressed genes between LA versus non-LA samples there was upregulation of T helper 1-oriented genes (IFNG, CXCL9-11) along with IL21 and IL17F which are key cytokines produced by T helper type 17 cells that enhance B-cell differentiation, isotype switching, and germinal center formation. Among LA samples there was an increase in CCL19 and CXCL13 expression which enhance B-cell recruitment and T–B-cell compartmentalization within TLS, as well as TNFRSF13B and TNFRSF17 which mediate B cell and plasma cell survival (online supplemental figure 5).18–22

In our study, we also compared the 12-CK gene expressions score in melanoma samples with and without LA. This 12-CK gene expression score was developed to gage the systemic presence of TLS in patients and focuses on chemokines known to recruit various B, T and myeloid cell populations into (inflamed) tissue sites. Our analysis revealed higher mean 12-CK gene expression scores in melanoma samples containing LA (with or without TLS with GC) in comparison to specimens devoid of LA. This finding confirms the 12-CK gene expression score as an informative gage for detecting the presence of heterogeneous forms of lymphoid structures (beyond TLS) in the melanoma TME.

Our study has several strengths. First, our report substantiates the prognostic significance of LA in a well-defined and homogeneous cohort of melanoma samples with primary lesions in the skin and metastatic lesions in skin/subcutaneous and soft tissues, in the context of both the tumor and surrounding stroma. We show that the presence of LA/TLS in metastatic tumor organ/tissue sites is heterogeneous, making it imperative to study the role of these lymphoid structures in properly sampled and evaluated specific tissue sites to more accurately interpret their association with patient prognosis. Furthermore, the description of LA/TLS in metastatic lymph node samples requires further refinement and should be segregated from other metastatic lesions (such as visceral sites) when attempting to interpret the potential role(s) played by LA/TLS in the TME and in patient with cancer outcomes. Second, our samples were of rigorous high quality, as we excluded all samples that were obtained through core biopsy. While we believe that LA/TLS may be identified in core biopsy material, we have also observed that results from core biopsies often underestimate the true representation of these lymphoid structures compared with whole excision samples. In our opinion, tissues obtained through different procedures such as core versus excisional biopsy should not be mixed when comparatively analyzing LA/TLS. Third, in our study, we refrained from reporting the number of TLS and ratio to tumor area, as we noted that the number of TLS varies significantly among sections of individual tumors when assessing large excision samples. Therefore, we focused more on the qualitative characteristics/maturity of LA/TLS as opposed to simply enumerating lymphoid structures. Lastly, we include analyses of a large cohort of primary melanoma cases where we report the true prevalence and clinical significance of lymphoid structures in these primary skin lesions. Our future studies will focus on specific metastatic sites of disease since we believe that findings cannot be generalized across disparate metastatic sites, the performance of expanded multiplex imaging analyses on samples that were reviewed from the TCGA SCKM cohort, and the performance of mechanistic studies to delineate pathways underlying our postulates that tumor-associated adipose tissues facilitate TLS formation in metastatic sites while pigmentation and neural networks limit or prevent TLS formation.

This study has some potential limitations. Heterogeneous presence of LA/TLS in various tumor sections may limit the interpretation of the prognostic role of lymphoid structures especially when assessing TCGA-SKCM cases. Future work will focus on software development to more objectively qualify and quantify exact lymphoid structure type and number of lymphoid structures. Functional heterogeneity of lymphocytes among LA and TLS is also of importance especially while assessing primary and metastatic tissues and will need to be further investigated along with the evaluation of the role of LA/TLS in immune checkpoint blockade response among primary tumor samples. The impact of immune checkpoint blockade therapy on LA/TLS formation and assessment of pre-therapy and post-therapy TME changes are of clinical significance that require further evaluation through gene transcriptomic analyses that consider sampling strategy differences which may limit the interpretation of morphological-defined LA/TLS cases as these relate to clinical relevance.

In summary, our study provides a comprehensive assessment of lymphoid structures in the melanoma TME and reveals that the presence of LA (in addition to TLS) is associated with improved OS in patients with melanoma. LA were observed to form predominantly in primary melanomas, with a greater abundance of TLS-GC detected in metastatic melanoma lesions in skin/subcutaneous sites. LA tended to form in divergent locations within TME. The presence of LA was associated with tumor-negative regional lymph node status. The 12-CK gene expression score was elevated in tumors that contained LA. Our findings for an increased presence of TLS with GC within metastatic skin/subcutaneous lesions surrounded by adipose tissue, and for negative correlations between pigmentation/neural networks and the presence of LA/TLS will require further investigation to delineate potential mechanisms that promote/restrict LA/TLS formation/fate/function in the melanoma TME, with this information then used to inform future design of interventional approaches designed to promote therapeutic LA/TLS for improved patient outcomes.

Data availability statement

Data are available upon reasonable request.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study used samples collected retrospectively from patients as part of the University of Pittsburgh Cancer Institute Melanoma Center Human Biological Sample and Nevus Image Banking and Analysis Protocol (HCC 96-099) CR19080226-007 where participants gave informed consent to participate in the study beofre taking part.This study did not enroll human subjects in itself therefore this was deemed as an exempt by University of Pittsburgh IRB MOD20020208-008. Additional IRB Advarra IRB (Local) Pro00071214 approval was obtained at Moffitt Cancer Center to conduct work focused on mainly image analysis and TCGA-SKCM data review and analyses.

Acknowledgments

This work has been supported in part by the Biostatistics and Bioinformatics Shared Resource at the H. Lee Moffitt Cancer Center & Research Institute, an NCI designated Comprehensive Cancer Center (P30-CA076292), by the Moffitt Tissue and Molecular Genomics Cores; both comprehensive cancer center facilities designated by the National Cancer Institute (P30-CA076292), was funded by the NCI-NIH (P30 CA076292, and P50 CA168536), CJG Fund, Chris Sullivan Fund, V Foundation, and the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation (J.J.M.), and the Donald A. Adam Melanoma and Skin Cancers Center of Excellence, ASCO Conquer Cancer Foundation and Melanoma Research Foundation (L.K.), NIH P01 CA234212 (W.J.S), and P50CA254865 Melanoma and Skin Cancer Program SPORE (J.M.K).