Sensory Nerves Impede the Formation of Tertiary Lymphoid Structures and Development of Protective Antimelanoma Immune Responses

The tumor microenvironment (TME) exerts significant influence over the development and progression of cancer. Over the last decade, the peripheral nervous system (PNS) has been firmly established as a critical component of the TME, where it influences tumor initiation and growth, metastasis, and cancer-related pain (1). Because the majority of past studies have focused on the autonomic nervous system, the role of sensory nerves (SN) in cancer progression remains comparatively less understood. It is now generally accepted that adrenergic innervation promotes cancer progression. Adrenergic nerve density increases during prostate cancer development and correlates with more aggressive disease in breast, colorectal, lung, and prostate cancer models (2–6). Sensory axonogenesis occurs in head and neck carcinoma and cervical cancer; however, it remains unclear whether all solid tumors exhibit sensory innervation (7–10). Several studies utilizing sensory denervation have demonstrated the protumor impact of SNs (11, 12); however, inhibitory effects on tumor growth and metastasis have also been reported (13, 14).

An investigation of SN impact on the TME requires precise targeting of the somatic afferent fibers supplying the tumor. This represents a significant technical challenge for application to many types of solid tumor models, with the murine anatomy relevant to cutaneous sensory innervation being ideally suited for such studies. Specifically, surgical access to the peripheral neurons composed primarily of sensory fibers, as well as their dorsal root ganglia (DRG), provides a unique opportunity to study the impact of SNs on skin-associated malignancies. In this regard, cutaneous sensory innervation has been shown to contribute to tumorigenesis and cancer progression in the skin. For example, basal cell carcinoma arises preferentially within the mechanosensory touch dome epithelia (15). We have previously shown that sensory DRG coinoculated with orthotopic melanoma results in accelerated tumor growth in vivo (16). However, selective cutaneous sensory denervation in the context of skin cancer progression has been heretofore understudied.

Melanoma is by far the most aggressive form of skin cancer, with a steadily increasing incidence worldwide (17). The host immune system's response to melanoma is essential to restricting cancer progression, with the magnitude and functional phenotypes of tumor-infiltrating lymphocytes (TIL) serving as predictors of patient clinical outcome and response to immunotherapy (18). Hence, the development of therapies promoting a proinflammatory TME is of significant translational interest. We have previously reported that SNs and peripheral neuroglia, specifically Schwann cells (SC), influence immune cell composition within the TME of melanomas (10, 16, 19–21). However, a direct link between SN-modulated immune responses and melanoma progression has never been confirmed.

Utilizing surgical and chemical sensory denervation methods, we investigated the role of SNs in melanoma progression in mouse models. Both denervation methods significantly slowed the growth of subcutaneous B16F10 melanomas in vivo, without significantly affecting tumor-associated blood or lymphatic vascular densities. Depletion of CD8⁺ lymphocytes prevented denervation-restricted tumor growth. We showed that sensory denervation led to the development of augmented melanoma-specific immune responses. When compared with controls, denervated tumors contained higher densities of mature high endothelial venules (HEV) associated with the formation of classic tertiary lymphoid structures (TLS), which associate with a more favorable melanoma prognosis (22, 23). Furthermore, tumors arising within denervated skin demonstrated improved recruitment and activation of both myeloid and lymphoid cells, reduced presence of regulatory immune cells, increased type-1 polarization of TILs, an increased T-cell clonality and expanded B-cell repertoire when compared with control tumors. Transcriptional profiling of human melanomas confirmed our findings in murine melanoma models. Specifically, the content of neuron-associated mRNA in human melanoma inversely correlated with antitumor immune response and TLS formation pathways and predicted superior overall survival. Our results implicate sensory innervation as an immunoregulatory mechanism within the melanoma TME that deserves attention as a target for the development of interventional therapeutic strategies to improve patient outcomes.

B16F10 murine melanoma cells (ATCC, CRL-6475) were cultured in RPMI-1640 (Cytiva, SH30255.01), supplemented with 10% FBS (R&D Systems, S11550H), 0.1% 2-mercaptoethanol (Gibco, 21985-023), 1% Antibiotic Antimitotic Solution (Sigma, A5995), 1% HEPES (Gibco, 15630-080), 1% NEAA Mixture (Lonza, 13-114E), and 1% L-glutamine (Gibco, 25030-081). Cells were used at passages (P) 3 to 6 for tumor initiation.

All studies were conducted in accordance with the NIH guidelines for the Care and Use of Laboratory Animals and approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh (Pittsburgh, PA). C57BL/6 mice (8-week-old) were purchased from the Jackson Laboratory and were housed in a pathogen-free facility under controlled temperature, humidity, and a 12-hour light/dark cycle with a commercial rodent diet and water available ad libitum. The Plp-CreERT/tdTomato^f/f mice were generated by crossing B6.Cg-Tg(Plp1-cre/ERT; Jackson Lab, Stock No: 005975) with B6.Cg-Gt(ROSA)26Sor^{tm14(CAG−tdTomato)} (Jackson Lab, Stock No: 007914). To induce Cre recombination, tamoxifen (Sigma, T5648) in corn oil (20 mg/mL) was injected intraperitoneally (75 mg/kg) every 24 hours for five consecutive days.

Surgical sensory cutaneous denervation was adapted from a previously described technique (24). Briefly, C57BL/6 mice were anesthetized with isoflurane (Covetrus, 11695067772), using a single animal vaporizer unit (E-Z Systems, 108SA) at a flow rate of 2 L/minute, and an approximately 1.5-cm skin incision was made along the midline back skin. Six dorsal cutaneous SNs on the left side were tied off with 9–0 prolene sutures (Ethicon, 1754G) at their bifurcation from the main nerve trunk. Nerves were transected approximately 2 mm distally to the sutures, carefully dissected, and 5–7 mm of distal nerves were removed. Skin was sutured with 7–0 prolene sutures (Covidien, SP-1696G). Sham surgery included a skin incision without nerve transection. Two months after surgery and prior to melanoma injection, complete local sensory denervation on left back and flank was confirmed with a pin prick test. Chemical sensory denervation was performed as previously described (25). Resiniferatoxin (Cayman Chemical) was injected intraperitoneally at escalating doses of 30 μg/kg, 70 μg/kg, and 100 μg/kg for three consecutive days. Denervation was confirmed by tail-flick response to noxious heat stimulation after 21 days. Denervated and age-matched control mice were injected intradermally with B16F10 melanoma cells (10⁵ cells in 100 mL sterile PBS) into the left dorsolateral skin. Tumor area (mm²) was measured with a caliper every second day starting at day 7. To deplete CD8⁺ T cells, a previously described method was utilized (26). Briefly, mice were injected i.p. with 150 μg anti-mouse CD8a (clone 2.43; Bio X Cell) on days 0, 3, 6, and 9 after B16F10 injection. CD8 depletion was confirmed by flow cytometry analysis of splenocyte single-cell suspensions utilizing nonoverlapping anti-mouse CD8a (clone CT-CD8a, Thermo Fisher). Rat IgG2b kappa antibody (clone LTF-2, Bio X Cell) was used as isotype control.

On days 16 to 17 after B16F10 cell injections, tumors were harvested and minced, incubated in DMEM (Corning, 10-013CV) containing collagenase D (1 mg/mL, Sigma, COLLD-RO) and DNase (1 mg/ mL, Sigma, DN25-1G) for 45 minutes at 37°C, and passed through 70-μm cell strainers (Falcon, 352350). Spleens and inguinal lymph nodes were passed through 70-μm cell strainer, washed in MACS buffer (Miltenyi Biotec, 130-091-221), and incubated for 5 minutes at 4°C in 2 mL of RBC lysis buffer (BioLegend, 420302). Single-cell suspensions were stained in MACS buffer with surface antibodies and Fixable Viability Dye eFluor 780 (FVD, Invitrogen, 1:1,000) for 1 hour at 4°C. For intracellular staining, cells were first fixed and permeabilized with the FoxP3/Transcription Factor Staining kit (eBioscience) for 30 minutes at room temperature, stained with surface antibodies and FVD, followed by incubation with intracellular and intranuclear antibodies for 60 minutes at room temperature. Flow cytometry was performed on BD LSRFortessa (BD Biosciences), and data were analyzed using FlowJo software (BD Biosciences, flowjo.com). Antibodies used are listed in Supplementary Table S1. Gating strategy and parameters are provided in Supplementary Material.

Spleens and inguinal lymph nodes of tumor-bearing mice 16 days after B16F10 cell injection were harvested, dissociated as indicated above, passed through a 70-μm cell strainer, washed once with PBS, and incubated for 5 minutes at 4°C in 2 mL of RBC lysis buffer. Cell mixtures were cocultured with irradiated B16F10 cells (10⁴ Rads, Gammacell 1000 Elite, Nordion) at a 20:1 ratio (leukocytes/melanoma cells), 37°C in RPMI containing IL15 (PeproTech, 10 ng/mL) for 5 days. Cells were washed in PBS, split, and coincubated with B16F10 cells (10:1 ratio) in RPMI media for 24 hours. IFNγ in 10 μL supernatants (1:10 dilution) were determined using the Mouse IFNγ ELISA Set (BD Biosciences), based on standard curves obtained utilizing the kit IFNγ, according to the manufacturer's protocol. Measurements were performed with an iD5 plate reader (Molecular Devices) and analyzed using SoftMax Pro 7.1 (Molecular Devices).

Tumors were harvested on days 16 to 17 after B16F10 cell injections. Tissues were fixed with 2% paraformaldehyde (PFA) for 2 hours, incubated in 30% sucrose in PBS for 24 hours, frozen in 2-methylbutane (Sigma) in liquid nitrogen, embedded in Tissue Plus OCT (Thermo Fisher), and processed into 50-μm thick sections. Sections were permeabilized with 0.1% Triton X-100 (Sigma, T9284) in PBS for 10 minutes, blocked with 5% goat serum (Gibco, 16210–072) with 2% BSA in PBS for 45 minutes, and washed with 0.5% BSA in PBS. Immunostaining was performed for 16 hours at 4°C with primary antibodies (Supplementary Table S2), followed by secondary antibodies (Supplementary Table S2) for 1 hour at room temperature and then 4′,6-diamidino-2-phenylindole (DAPI, Sigma, 1 μg/mL). Sections mounted in gelvatol medium (Sigma) were imaged using a Nikon A1+ confocal microscope. Image analyses were performed with NIS-Elements AR 4.40 software (Nikon).

For RNA extraction, mouse B16F10 tumors stored in RNAlater (Invitrogen) were homogenized in TRIzol reagent using a bead homogenizer Bullet Blender (Next Advance) at speed 8 for 5 minutes using Navy Rhino tubes (Next Advance, NAVYR1). RNA was isolated using TRIzol reagent (Thermo Fisher) as per the manufacturer's protocol and quantified using DeNovix DS-11 spectrophotometer. For cDNA synthesis, 2 μg RNA was converted to cDNA using a QuantiTect Reverse Transcription Kit (Qiagen). Quantitative (q)RT-PCR was performed using TaqMan probes labeled with FAM-ZEN/IBFQ (Supplementary Table S3), an endogenous control (Actin) labeled with VIC-MGB-PL, and TaqMan Fast Advance Master Mix (Applied Biosystems). For select gene targets, qRT-PCR was performed using Fast SYBR Green Master Mix (Applied Biosystems) with appropriate primers (Supplementary Table S3), and Gapdh as endogenous control. Reactions were run and analyzed on StepOnePlus thermocycler (Applied Biosystems). Relative fold changes were calculated and normalized based on the 2^–ΔΔCt method.

RNA sequencing (RNA-seq) was performed as previously described, with several modifications (10). Briefly, TRIzol reagent (Thermo Fisher, 15596018) was used to extract total RNA from homogenized tumors 16 days after B16F10 cell injections as described for qRT-PCR. Quality and concentration of RNA were determined with Qubit (Thermo Fisher) and Agilent Bioanalyzer, and samples with RNA integrity numbers (RIN) 7.5 to 8.7 were used. Sequencing libraries were constructed by using Illumina TruSeq Stranded mRNA Prep with a read length of 2 × 75 bp. Approximately 1.5 × 10⁸ total reads, paired-end, were obtained per tumor (Illumina NextSeq 500/550 v2.5 Mid Output, 150 cycles). The reverse-stranded paired-end RNA-seq reads were checked for the presence of adapters and high-quality bases using FastQC (v 0.11.9). The raw fastq files were lane separated, which required concatenation to generate complete, paired-end fastq files. The universal adapters in the high-quality reads were already trimmed, and, therefore, did not require the use of an adapter trimming tool. The reads were mapped against the Ensembl mouse reference genome (GRCm38 mm10) using the HISAT2 (v 2.1.0) mapping tool. The output file from HISAT2 was converted from SAM format to BAM format using SAMtools (v 1.10). Counts for expressed genes were generated using HT-Seq (v 0.11.2) and were output in text format. These count text files were imported into the Bioconductor R-package, edgeR (v 3.32.1), which was utilized to identify the differentially expressed genes (DEG) based on the criteria of the genes having an expression count of absolute value log base 2 greater than 1 between two experimental conditions and a false discovery rate (FDR) < 0.1. After DEGs were identified, the list of genes with their differential expression values was uploaded to Ingenuity Pathway Analysis (IPA, Qiagen). Gene Ontology (geneontology.org) and Enrichr (maayanlab.cloud/Enrichr) were also utilized for pathway analyses. Bar graphs were generated using R-package ggplot2 (v 3.3.5). Heatmaps were generated using R-package gplots (v 3.1.1).

The Cancer Genome Atlas (TCGA) program was used to obtain RSEM normalized RNA-seq data for 473 patients with skin cutaneous melanoma (SKCM; ref. 27). Eighty-two patients were selected whose RNA-seq data represented the primary melanoma tumor, not a metastatic lesion, and was procured prior to any interventional treatment. Patients were divided into two groups based on the sum of transcript per million (TPM) values of gene sets discussed in Results. The median value of summed TPM score was used to partition patients into high- and low-neuronal content groups (Supplementary Table S4), and DEGs between these two groups were determined using DESeq2 (28). Genes upregulated in the low-neuronal content group (vs. high-neuronal content group) with P < 0.05 and fold change greater than 1.5 were utilized in pathway analyses with IPA and Enrichr. Bar graphs were generated using R-package ggplot2 (v 3.3.5). Kaplan–Meier analyses and log-rank tests were performed using R-package survival (v 3.2-13).

Primary cutaneous melanoma tissues were collected after obtaining signed informed consent from three subjects. Samples were processed and analyzed by IF as described above for mouse melanomas, utilizing primary and secondary antibodies listed in Supplementary Table S2. The study was performed after the approval by the institutional review board of the University of Pittsburgh, in accordance with an assurance filed with and approved by the U.S. Department of Health and Human Services, and was conducted in accordance with U.S. Common Rule ethical guidelines.

Genomic DNA from whole tumors was extracted on day 16 after B16F10 cell injections, purified using DNAeasy Blood and Tissue kit (Qiagen), and shipped frozen to Adaptive Biotechnologies (Seattle, WA). Amplification was carried out utilizing TCRβ and BCR heavy chain (IgH) V/J gene-targeting multiplexed primers, and sequencing of CDR3 regions was performed as previously described (29, 30). Potential amplification bias was measured and corrected using synthetic templates mimicking natural V(D)J rearrangements (29). Productive rearrangements excluded out of frame sequences and sequences with stop codons within the CDR3 region. Data were analyzed using the ImmunoSEQ analysis platform (Adaptive Biotechnologies).

Results are expressed as mean ± SD. Two-group analyses were performed using an unpaired Student t test. Three or more groups with one independent variable were analyzed using one-way ANOVA with the Tukey multiple comparisons test. Three or more groups with two independent variables were analyzed using two-way ANOVA with Tukey multiple comparisons test. Analyses were performed using GraphPad Prism software. All tests were two-tailed, and a P value < 0.05 was considered to indicate statistical significance.

RNA-seq raw data are available in the GEO database under accession number GSE193065. Immunosequencing raw data are available at https://doi.org/10.21417/OK2022MELANOMA. All data published in this report are available on reasonable request.

Surgical denervation of dorsal thoracic sensory cutaneous fibers (Fig. 1A) was achieved with a modification of a previously reported method (24). Briefly, proximal trunks of the transected and stripped nerves were intraoperatively tied with sutures (Supplementary Fig. S1A and S1B), thereby preventing axon sprouting and reinnervation, and resulting in complete, reproducible, and durable sensory cutaneous denervation associated with the loss of nociception of the back skin (see Materials and Methods). IF analyses of axonal and glial markers demonstrated a significant reduction in numbers of cutaneous neurons after surgical sensory denervation (SSD), including thick dermal and thin subepidermal nerve bundles, as well as free epidermal nerve endings (Supplementary Fig. S1C). To verify denervation, we also utilized transgenic mice, Plp-CreERT/tdTomato^f/f (B6 background), which undergo tamoxifen-mediated expression of red fluorescent protein in the SCs of the peripheral neurons (Supplementary Fig. S1D and S1E; refs. S31–S33). Using this model, we confirmed the loss of the cutaneous neurons after SSD (Fig. 1B). Next, we examined whether SSD affected the vascular and lymphatic composition of the skin. Neither the density of CD31⁺ vascular nor Lyve-1⁺ lymphatic endothelial networks in denervated skin were affected by SSD (Supplementary Fig. S1F–S1H).

We next examined the rate of B16F10 melanoma growth in the surgically denervated skin of syngeneic C57BL/6 mice. Compared with control mice, SSD resulted in significantly slowed tumor growth (Fig. 1C and D). In contrast, sham surgery (skin incision and undermining without nerve transection) did not affect tumor growth. To confirm these results using a nonsurgical complementary method, we administered resiniferatoxin (RTX), a potent agonist of transient receptor potential cation channel, subfamily V, member 1 (TRPV1), which selectively ablates TRPV1⁺ neurons (25). Intraperitoneal injection of RTX led to the loss of nociceptive cutaneous TRPV1⁺ nerve fibers and their corresponding SC (Fig. 1E), which was confirmed in mice one month later based on their lack of response to noxious heat stimulation (see Materials and Methods). Similar to SSD, chemical denervation of the afferent cutaneous neurons resulted in a significantly slower growth rate for B16F10 melanomas in vivo (Fig. 1F). As we had observed in surgically denervated non–tumor-bearing skin, densities of intratumoral CD31⁺ vascular and Lyve-1⁺ lymphatic vessels were not affected by the loss of SNs (Fig. 1G and H). The expression of tissue hypoxia marker, Hif1a, also remained unchanged in tumors after SSD (Fig. 1I). We next examined whether intervention-induced antimelanoma immune responses contributed to slowed tumor growth after SSD. Antibody-mediated depletion of CD8a⁺ T lymphocytes (26) at the time of tumor development abrogated the effect of SSD on melanoma growth (Fig. 1J), suggesting that sensory innervation of the skin supports melanoma progression by regulating local immune responses to tumor cells.

Further investigation revealed that compared with controls, tumors arising within surgically denervated skin contained more CD45⁺ leukocytes (Fig. 2A; Supplementary Fig. S2). The percentages of infiltrating CD4⁺ and CD8⁺ T cells, CD19⁺ B cells, and CD11b⁺ cells were all elevated in the denervated TME, whereas the frequency of F4/80⁺ macrophages remained constant (Fig. 2B). Significantly fewer regulatory T cells (Treg) were found in the SSD tumors vs. control tumors (Fig. 2C), and intratumoral percentages of F4/80⁺CD206⁺ M2 tumor-associated macrophages (TAM) and Ly6G⁺ granulocytic myeloid-derived suppressor cells (MDSC), but not Ly6G^–Ly6C^hi monocytic MDSCs, were also diminished after SSD (Fig. 2D). Significantly fewer CD8⁺PD-1⁺ and CD4⁺PD-1⁺ T cells were identified in denervated tumors, suggesting that these effector cells were less exhausted and presumably functionally competent (Fig. 2E; ref. 34). Finally, CD4⁺ T cells expressing CD73, an ecto-5′-nucleotidase that catalyzes the conversion of extracellular AMP to adenosine, a potent immunosuppressive factor in TME (35), were found less frequently in denervated tumors (Fig. 2F). Flow-cytometric analyses of TILs from RTX-denervated mice confirmed the findings obtained in our surgically denervated animal models (Supplementary Fig. S3).

In contrast to the TME, the spleens of mice after SSD or RTX-mediated sensory ablation did not exhibit altered immune profiles compared with spleens isolated from control animals, with the exception of a decrease in splenic CD8⁺PD-1⁺ T-cell content after SSD (Supplementary Figs. S4–S6). The composition of immune cells in the TME after CD8a depletion was not reflective of the status of cutaneous sensory innervation (Supplementary Fig. S7). CD8a depletion led to diminished populations of T cells (CD8⁺, CD4⁺, and Treg), but not CD11b⁺ cells, in control and denervated tumors; the only significant difference between the two groups was fewer intratumoral CD11b⁺F4/80⁺ cells after SSD (Supplementary Fig. S7). Spleens of both the innervated and denervated groups displayed a 15-fold decrease in CD8⁺ T cells after CD8a depletion relative to nondepleted controls (Supplementary Fig. S8). The only two significant differences in the composition of splenic immune cells between the CD8a-depleted innervated and CD8a-depleted denervated groups were increased frequencies of CD4⁺PD-1⁺ and CD8⁺PD-1⁺ populations in the latter cohort (Supplementary Fig. S8). The changes in the TME immune composition fomented by skin denervation were observed only in tumor-bearing mice. Specifically, when we compared non–tumor bearing control mice to non–tumor bearing denervated mice, we did not observe any differences in either the composition or the phenotype of lymphoid and myeloid immune cells in their skin, inguinal lymph nodes, or spleens (Supplementary Fig. S9). These results suggest that the absence of sensory innervation in the skin leads to a more immunocompetent and protective cutaneous TME.

The density of CD8⁺ TILs in melanoma has long been considered a positive prognostic marker of patient overall survival and response to (immuno)therapy (18). IF confirmed our flow cytometry results, with CD8⁺ TIL densities in skin tumors devoid of SNs being significantly higher than those found in innervated tumors (Fig. 3A and B). To elucidate their functional status, we determined the percentages of interferon (IFN)γ^hi and granzyme (Gzm)-B^hi CD8⁺ TILs by flow cytometry. In surgically denervated melanoma-bearing mice, more CD8⁺ TILs were IFNγ^hi, whereas the percentage of GzmB^hi cells among CD8⁺ TILs remained unchanged compared with controls (Fig. 3C). IFNγ^hi and GzmB^hi CD8⁺ TILs in the TME of RTX-treated mice were significantly increased relative to controls (Fig. 3D). TME expression of Ifng from surgically and chemically denervated skin was also increased relative to tumors isolated from the skin with intact nerves (Fig. 3E). To further determine whether melanomas elicited stronger effector T-cell responses in denervated mice, single-cell suspensions of lymph nodes and spleens from control and denervated tumor-bearing mice were isolated and restimulated with B16F10 melanoma cells in vitro. Lymphocytes obtained from the inguinal lymph nodes and spleens of surgically denervated tumor-bearing mice produced significantly more IFNγ in response to melanoma restimulation in vitro versus those isolated from control mice (Fig. 3F). Sensory denervation with RTX yielded similar results (Fig. 3G). These findings suggest that antitumor T-cell responses in denervated mice are superior to those in mice with intact sensory cutaneous innervation.

To better understand the impact of denervation on the immune responses to melanoma, we performed RNA-seq analyses of B16F10 tumors isolated from the skin of control versus SN-depleted mice. Of the 409 DEGs with a 0.1 FDR cutoff, 364 were upregulated and 44 were downregulated (Supplementary Table S5). Several of the downregulated genes, including P2rx3, Kcnh2, S100β, Slc6a9, Abat, Kcnn1, and Avil, were highly expressed by neurons or by select non-PNS cells in response to neuronal signals. Gene Ontology (GO) analysis of the 44 downregulated genes revealed enrichment of a single pathway: positive regulation of nervous system process (GO:0031646, P = 2.09e−6, FDR = 3.3e−2, 48-fold enrichment), suggesting an expected reduction in neuronal signaling in the TME after SSD. Corroborating our flow cytometry findings, expression of Ptprc (CD45), Cd4, Itgam (CD11b), and Itgax (CD11c) was significantly increased in denervated tumors relative to controls (Supplementary Table S5). Utilizing IPA and considering all 409 DEGs, we found that the majority of the most significantly enriched pathways (lowest P values) broadly reflected the activation of anticancer immune responses (Fig. 4A; Supplementary Table S6). Pathway analysis with Enrichr confirmed denervation-dependent activation of innate and adaptive immune response elements, which could be subdivided into the following categories: activation of the immune response, leukocyte adhesion and diapedesis, Th1 and Th2 responses, B-cell activation, antigen processing and cross-presentation, and interferon signaling (Supplementary Fig. S10; Supplementary Table S7). We also examined enriched physiologic pathways using IPA, which could be parsed into the following groups: recruitment, homing, adhesion, transmigration, chemotaxis, and activation of leukocytes (T and B lymphocytes, antigen-presenting cells), and formation/quantity of lymphoid tissue (Fig. 4B; Supplementary Table S8). Taken together, these data corroborate superior activation of multiple branches of the protective host immune response within the cutaneous TME lacking sensory innervation.

The appearance of the “lymphoid organ formation” pathway in our RNA-seq results led us to further investigate the relationship between SSD and TLSs in the melanoma TME. TLSs are ectopic lymph node–like structures containing variably organized B- and T-cell aggregates surrounding lymphatic and high endothelial blood vessels (36) and have been identified in several types of cancers, including melanoma (37, 38). The presence and the extent of TLS formation in the melanoma TME are predictive of favorable patient response to immunotherapy and extended overall survival (22, 23). TLSs in tumors may amplify cytokine-mediated signaling and improve antigen presentation and tumor-specific antibody production (39). In pancreatic carcinomas, small nerve fibers have been reported to associate with lymphoid aggregates, but the relationship between TLSs and innervation in tumors remains poorly understood (40). Based on the results of our transcriptional profiling of the TME, we next examined the association between SNs and TLS formation in our cutaneous melanoma model. From the 409 DEGs, we identified 54 genes previously associated with TLS formation (22, 39, 41–43). Expression of all the gene products in this ad hoc panel was upregulated in denervated tumors (Fig. 4C). We also assessed the expression of consensus pro-TLS factors, including chemokines associated with inflammatory infiltrates and TLS formation (38, 41, 44). We observed that expression of Ccl3, Ccl5, Ccl19, Ccl21, Cxcl9, Cxcl13, and Tnfa were all upregulated in melanomas lacking sensory innervation (Fig. 4D). Compared with control tumors, denervated tumors also expressed elevated lymphotoxin β receptor (LTβR) agonists Lta, Ltb, and Tnfsf14/LIGHT, which are implicated in secondary lymphoid organ (SLO) and TLS development (Fig. 4E; refs. 39, 41, 42). IPA pathway analyses similarly revealed an enrichment trend (P = 0.07) for lymphotoxin β receptor signaling in denervated tumors (Supplementary Table S6). Cumulatively, these results suggest that SSD promotes TLS formation in the cutaneous melanoma TME.

To expand on these findings, we morphometrically analyzed melanoma tumors for the physical presence of TLSs. As in the case of SLO, tumor-associated TLSs contain HEVs, specialized postcapillary venules expressing high sulfated sialomucins that mediate homing and transmigration of lymphocytes from the blood (36, 37, 39). Sulfated sialomucins, ligands for L-Selectin/CD62L, are concentrated on high endothelial cells (HEC) and serve as classic biomarkers for HEVs. Our genomic analyses revealed a trend for upregulated expression of several gene products associated with sialomucins in denervated versus innervated tumors, including Cd34 (FC = 1.59, P = 0.0199), Podxl (FC = 1.5, P = 0.0538), and Emcn (FC = 1.58, P = 0.0445). qRT-PCR analyses confirmed upregulation of Cd34, Podxl, and GlyCAM1 in tumors after SSD (Fig. 4F). Expression of other HEV leukocyte-homing receptors (45), Icam1 and Selp, were also increased after denervation (Fig. 4C and F). Expression of chemokines present on the luminal surface of HEVs, which mediate lymphocyte trafficking, including Ccl19 and Ccl21 (and their receptor Ccr7), Cxcl9, Cxcl12, Cxcl13, were also upregulated in denervated tumors (Fig. 4C and D). We also found that tumors from denervated skin expressed elevated HEC/HEV adhesion molecules, including Vcan (FC 2.5, P = 0.0058), Mcam (FC 1.47, P = 0.011), Fxyd5 (FC 1.63, P = 0.012), Icam1 (FC 3.71, P = 4.17e−5), Vcam1 (FC 4.44, P = 1.2e−7; Fig. 4C and F; Supplementary Table S5; refs. 39, 45, 46). Ligands of these leukocyte-homing domains, such as Itgb2, Itgb7, Itgal, and Selplg, also had higher expression in denervated versus innervated tumors (Fig. 4C; Supplementary Table S5).

To further quantify and characterize HEVs in B16F10 melanomas, we used IF and MECA-79 antibody, which binds sulfated peripheral node addressin (PNAd) of HECs (47). The fluorescence intensity of HEVs stained with anti–MECA-79 was significantly elevated in denervated tumors versus controls (Fig. 5A and B). HECs in denervated tumors exhibited a more pronounced cuboidal morphology and contained more CD34 and ICAM1 versus HECs in control tumors (Fig. 5A and B). Using IF, we also examined the extent of lymphocyte clusters surrounding PNAd⁺ HEVs as an indicator of TLS formation. In SN-ablated tumors, PNAd⁺ HEVs had denser adjacent aggregates of CD3⁺, CD4⁺, and B220⁺ lymphocytes (Fig. 5C and F). CD8a depletion prior to melanoma initiation significantly diminished peri-HEV CD3⁺ and B220⁺ cells in denervated tumors (Supplementary Fig. S11). To better determine whether PNAd⁺ HEVs and TLS in melanomas form near nerves, we examined B16F10 tumors isolated from the skin of Plp-CreERT/tdTomato^f/f mice. We observed that nerves were located in close proximity to PNAd⁺ HEVs in innervated but not SSD tumors, as expected (Fig. 5G). The absence of nerves near PNAd⁺ HEVs, therefore, correlated with significantly larger clusters of CD3⁺ lymphocytes around the tumor-associated blood vessels (Fig. 5G). When taken together, these findings suggest that SN ablation in the cutaneous TME promotes maturation of HEVs, infiltration by immune cells, and their organization into TLSs.

As in the case of SLO, tumor TLSs are posited to serve as local sites for T- and B-lymphocyte recruitment, antigen presentation, and the generation of tumor-specific effector T-cell and antibody responses. In melanoma and other cancers, naïve CD62L⁺ T cells are recruited into TLSs within the TME, where they are (cross)primed, expanded, and differentiated into effector cells that mediate antitumor immunity (48, 49). Increased densities and the maturation of TLSs in tumors have been linked to expansions in the (antitumor) T- and B-cell repertoires (49–52). We, therefore, determined the effect of sensory innervation on the intratumoral T-cell (TIL-T) and B-cell (TIL-B) repertoires. We performed TCR-seq of V(D)J complementarity-determining region 3 (CDR3) of TCRβ rearrangements on control versus denervated melanomas (29, 30). We found that the number of unique TCRβ clonotypes was not significantly different between the two groups, indicating that the TIL-T population richness is not affected by the absence of intratumoral SNs (Fig. 6A). Simpson clonality, however, was significantly increased in denervated tumors, indicating a reduced evenness (i.e., increased oligoclonality) of TCRβ rearrangements in the TIL-T population (Fig. 6B). The frequency of the most represented TCRβ clonotypes among all productive rearrangements was elevated in SSD tumors relative to controls (Fig. 6C), as was the sum of the frequencies of the top 5 most abundant TCRβ clonotypes (Fig. 6D; Supplementary Fig. S12). Differential analysis of the clonotypes with abundance > 5 shared between control and denervated tumors, but were represented at a significantly higher frequency in either group, revealed a trend suggestive of local, intratumoral expansion (increase in frequency) of these shared TCRβ clonotypes in the SSD group (Fig. 6E and F), and the number of unique TCRβ clonotypes with abundance >50 was significantly increased in denervated tumors relative to controls (Fig. 6G and H). Analysis of the overlapping data sets suggested a trend for an increase in the number of shared CDR3 sequences (amino acids) among the denervated tumors, as well as between denervated and control tumors, relative to the shared sequences among the control tumors (Fig. 6I and J). Overall, these results suggest that sensory denervation leads to the clonal expansion of shared and unique TIL-T clonotypes in association with enhanced TLS formation in the TME.

We also analyzed the TIL-B repertoire by immune-sequencing the BCR H chain (IgH) CDR3 region on whole tumor explants (29). The number of unique BCR IgH clonotypes, a reflection of clonal diversity (richness), was significantly increased in SSD tumors relative to controls (Fig. 6K). Simpson clonality, a measure of the population clonal polarization, showed a decreasing trend in tumors with ablated nerves (Fig. 6L). The fraction of productive BCR rearrangements, in-frame rearrangements without a stop codon that can generate a functional protein receptor, was increased in TIL-B populations isolated from denervated tumors (Fig. 6M), and the sum frequency of the top 5 intratumoral BCR IgH clonotypes was not affected by SSD (Fig. 6N). When taken together, these results confirmed our RNA-seq, IF, qRT-PCR, and flow cytometry analyses in supporting that sensory nerve ablation leads to increased recruitment and expansion of B cells into/within the melanoma TME. An increase in intratumoral BCR clonal diversity at the expense of clonal polarization in denervated mice suggests that TIL-B clones may have been preferentially recruited from SLO rather than generated in TLSs. Indeed, whereas B cells were significantly more numerous around PNAd⁺ HEVs of SSD versus innervated tumors on days 14 to 21 after melanoma implantation (Fig. 5), we did not observe formal germinal center formation in the TME at these time points (with germinal centers postulated to serve as local sites for B-cell clonal expansion and somatic recombination in TLS in situ) (23).

We next determined whether the negative correlation between skin sensory innervation and the enhanced development of antitumor immune responses observed in our mouse models was reflective of human disease. To that end, we performed analyses of SKCM RNA-seq data publicly available from TCGA. From gene-expression profiling data (n = 473 melanoma samples), we selected 82 primary cutaneous lesions that were procured prior to any interventional treatment. Patients were divided into two groups based on the summed tumor-associated expression score (TPM) composed of seven genetic markers associated with sensory neurons: Scn10a, Calca, Tac1, Gal, Nefh, Kcnn1, and Kcnh2 (Supplementary Table S4; ref. 53). These genes were chosen based on their expression specificity in sensory neurons in both mice and humans (54, 55). Two of these genes, Kcnn1 and Kcnh2, were downregulated in our surgical denervation mouse model (Supplementary Table S5). Because melanocytes and the PNS share a common neural crest cell origin, we excluded neuronal and glial genes known to be expressed by melanoma cells (56). Human primary melanomas with summed TPM expression scores of these seven genes larger than the overall group median were considered as having higher neuron-associated mRNA content (NAmRNA) and vice versa (Fig. 7A). High and low NAmRNA content groups did not differ based on a range of other parameters, including age, gender, ethnicity, pathologic stage, or site of disease (Supplementary Table S9).

Analysis of DEGs between the high and low NAmRNA groups was then carried out. Multiple proinflammatory cytokines, chemokines, and other immune mediators were significantly upregulated in the low versus high NAmRNA group, including Ccl2, Ccl7, Ccl8, Ccl13, Cxcl5, Cxcl13, Tnfsf13, Il6, Il15, and others (Supplementary Table S10). Pathway analyses with IPA (Supplementary Table S11) and Enrichr (Supplementary Table S12) revealed an enrichment of identical immune response pathways in the low NAmRNA group to those we observed in our SSD mouse model. Specifically, pathways involving the activation of antigen processing and presentation, TCR and BCR signaling, and interferon gamma signaling were upregulated in the human melanomas with low levels of neuron-associated transcripts (Fig. 7B; Supplementary Fig. S13A). Physiologic IPA pathways enriched in the low NAmRNA group—recruitment, homing, transmigration and chemotaxis of lymphoid/myeloid cells, lymphopoiesis, and formation of lymphoid tissue—were identical to the mouse denervation model (Fig. 7C; Supplementary Table S13). We performed IF to further evaluate the role of the innervation on TLS formation in human primary cutaneous melanoma. We detected PNAd⁺ HEVs, which were either in close proximity to the PGP9.5⁺ neurons or not associated with neurons and noted a significantly reduced density of CD3⁺ lymphocytes around PNAd⁺ HEVs, which were located next to nerves versus HEVs with no near-neighbor neurons (Fig. 7D and E). Finally, because TLS formation has been shown to improve melanoma survival, we compared the available survival data in our high and low NAmRNA cohorts. Overall survival was significantly greater among patients with low NAmRNA melanomas (Fig. 7F). Because the unequivocal detection of local translation in primary afferents within tissues requires spatial transcriptomics (57), we grouped the same patient cohort based on tumor expression of 15 genes associated with SCs (Mag, Gdnf, Gap43, Ncam1, Gldn, Plp1, Ntn1, Prx, Ntf3, Ntf4, Ntrk3, Cntn2, Cntn4, GabbR1, Areg), avoiding genes highly expressed by melanoma cells (58). Analysis of DEGs in low versus high SC mRNA content groups was carried out (Supplementary Table S14), which revealed an upregulation of proinflammatory and HEV maturation pathways, including leukocyte diapedesis and lymphotoxin β receptor signaling, in the low SC mRNA group (Supplementary Table S15). As in the case of NAmRNA analysis, overall survival was significantly greater among patients with low SC mRNA content melanomas (Supplementary Fig. S13B). Taken together, these data suggest that in mouse (and human) melanomas, the presence of sensory neurons (and neuron-associated mRNA content) in the TME may restrict local inflammation and TLS formation, thereby limiting protective antitumor immune responses in support of disease progression and reduced overall survival.

The main novel finding in our study reflects the regulatory role mediated by melanoma-associated sensory neurons on the development and function of immune responses within the TME. We observed that both complete SSD and chemical ablation of TRPV1⁺ sensory neurons of cutaneous melanomas resulted in the activation of antitumor immune programming in the TME, as evidenced by a reduction in intratumoral Treg, M2 TAM, and MDSC immunosuppressive populations, and a coordinate increase in effector immune cell presence. We did not observe any denervation-associated changes in the densities of vascular and lymphatic endothelial cells or expression of hypoxia markers in the denervated TME, suggesting that modulation of the immune system is a primary mechanism by which sensory neurons support melanoma growth in the skin. In this regard, we observed that antibody-mediated depletion of CD8⁺ T cells mitigated the effect of sensory nerve ablation on melanoma progression, supporting the key role of antitumor CD8⁺ T cells in disease outcome. In the absence of sensory neurons, HEVs in the TME were more mature, and leukocyte recruitment into the TME was enhanced, with these immune cells orchestrated into nonclassic TLSs lacking apparent germinal centers but containing dense clusters of T cells and B cells. The augmented influx of the myeloid and lymphoid cells into denervated melanomas also resulted in an expanded intratumoral BCR IgH repertoire and the increased (cross)priming of T cells, in association with an expanded T-cell repertoire and improved TIL effector functionality. These results suggest that tumor-associated sensory neurons (TRPV1⁺ fibers, specifically) actively restrict the evolution of protective antitumor immune responses in the cutaneous TME in vivo.

TLS formation has been increasingly recognized as an important biological process associated with a more favorable cancer patient prognosis and an indicator of superior patient response to immunotherapy (22, 23). The evidence for a neuroimmune circuit regulating SLO and TLSs is only now emerging in the setting of select inflammatory conditions (53, 59, 60). Very recently, nerve fibers were reported to colocalize with TLSs in pancreatic cancer (40), which has a dismal clinical prognosis and is associated with severe immune dysfunction (61). We present here, for the first time, experimental evidence supporting neuroimmune regulation affecting HEV maturation and TLS formation within the TME. TLSs serve as gateways for CD62L⁺ naïve/central memory lymphocyte infiltration from blood into tumors, a function which is largely mediated by mature PNAd⁺ HEVs (39). Various inflammatory immune cells have been implicated in the neogenesis and maturation of HEVs in the TME, including both lymphoid (CD8⁺ T cells, B cells, NK cells; refs. 48, 62) and myeloid (CD11c⁺ dendritic cells, CD68⁺ macrophages; refs. 62, 63) cells types. We observed that denervation was accompanied by a decrease in intratumoral populations of two major types of regulatory immune cells, Treg and Gr1⁺ neutrophils, which have previously been shown to inhibit TLS formation (64, 65). We did not observe CD11c⁺ clusters in TME-associated TLS-like structures in our models, which rather resembled canonical immature TLSs composed of CD4⁺ T-cell and B-cell aggregates, but without the formation of mature (B cell-rich) germinal centers. Although antibody-mediated CD8a⁺ T-cell depletion abrogated the antitumor benefit provided by skin denervation, it remains mechanistically unclear exactly how SNs impede TLS formation in tumors. We would hypothesize that afferent neurons may either directly hinder HEC maturation, block the recruitment or the activity of lymphoid tissue-inducing cells, or promote the infiltration of regulatory cells that are contraindicated for TLS formation (Fig. 7G). Clearly, further translational/clinical investigation of the dynamic interactions within the neuroimmune axis of the TME is warranted.

The functions of peripheral sensory neurons are tightly coupled to SC activity (66). Because neurons are comprised of nerve fibers and SC, denervation methods aimed at removing one of these components inadvertently affects the other. Thus, complete surgical, chemical, or genetic ablation of the nerve/SC unit leads to challenges in determining the relative contribution of either element to the observed phenotype (1). We have previously shown that DRG neurons and SC may accelerate melanoma progression via neuroregenerative and immunomodulatory mechanisms (10, 16). Here, we utilized complimentary surgical and chemical denervation methods that removed the majority of sensory nerve fibers and SC from the skin prior to melanoma initiation. Therefore, it is impossible to conclude with certainty which of these components of the sensory neuron is responsible for hindering antitumor immunity in the TME. Additional cell-specific inducible nonablative genetic models will be required to conclusively determine how afferent nociceptive fibers (TRPV1⁺ and TRPV1^– fibers) and/or their associated SC regulate local immune responses in the TME.

Our analyses of melanoma patient cohort should be interpreted with caution. Tumoral presence of transcripts commonly associated with DRG, which are located some distance from cutaneous tumors, does not directly indicate innervation, as these mRNAs may be from nonneuronal cells. Although local translation in primary afferent axons has been reported for a number of DRG-associated factors (including Gap43, Calca, and Scn10A; ref. 57), spatial transcriptomics and multiplexed quantitative tissue microscopy of patient melanomas and outcome measures must be performed to unequivocally determine the impact of tumor innervation on immune responses, TLS formation, and overall survival. Our data indicate that grouping patients based on the content of SC-associated transcripts in tumors also suggests an inverse relationship between tumor innervation and inflammation/overall survival. Other interpretations are possible, such as the relationship of neural-like differentiation of melanoma cells (neurotropic, desmoplastic) to an immunosuppressive TME and poor survival. However, clinical data do not currently support such an interpretation (67). Further studies utilizing human melanomas are required to better understand the impact of tumor afferent innervation on protective immune responses and disease progression.

Selective targeting of sensory neurons for cancer therapy has thus far not been realized. Ablation and deactivation of sensory innervation using varying methods in different mouse tumor models, including pancreatic, gastric, breast, and skin cancers, have yielded conflicting results (11–16). Such discrepancies may arise from organ-specific differences in tumor innervation or antitumor immunity, or other TME factors, as well as distinct SN-targeting techniques used in these studies. Further reflecting this disparity, modulations of nociceptor signaling axes for cancer therapy have proven inconclusive. For example, substance P (SP), a tachykinin neuropeptide released by sensory neurons, was reported to promote cancer progression over a decade ago (68). However, aprepitant, an antagonist of the neurokinin-1 receptor (NK1R which binds SP) used to treat chemotherapy-related nausea and vomiting, has thus far failed to demonstrate antitumor efficacy (69). Targeting TRPV1, a capsaicin receptor expressed on nociceptor neurons, has been successful in the treatment of neuropathic pain, but its utility in cancer therapy remains equivocal (70). The use of botulinum neurotoxins, which prevent neurotransmitter and neuropeptide release from the termini of afferent and efferent neurons, has similarly been limited to treating cancer-related pain, rather than being applied as a cancer therapeutic agent (71). Further studies will be required to discern the impact of PNS-targeting drugs on immune status within the TME and how they may be combined with existing immunotherapies for possible synergistic treatment benefit in the cancer setting. In this regard, we detected upregulated expression of several anti-inflammatory factors in the denervated melanoma TME, including Arg1, Ptgs2/Cox2, and PD-L1, suggesting near-term assessment of combination protocols implementing local denervating agents along with agents that antagonize these compensatory (immuno)regulatory pathways, such as checkpoint inhibitors (72). We further anticipate that prospective studies of SN-specific regulatory mechanisms in the cancer setting will likely identify additional therapeutic candidates for use in monotherapies or combination immunotherapies.

M.R. Shurin reports grants from ASC during the conduct of the study. No disclosures were reported by the other authors.

K. Vats: Data curation, investigation, methodology, writing–original draft. O. Kruglov: Data curation, investigation, methodology, writing–original draft. B. Sahoo: Formal analysis, methodology. V. Soman: Formal analysis, writing–original draft. J. Zhang: Investigation, methodology. G.V. Shurin: Data curation, formal analysis, methodology. U.R. Chandran: Formal analysis. P. Skums: Formal analysis. M.R. Shurin: Investigation, writing–review and editing. A. Zelikovsky: Formal analysis, writing–review and editing. W.J. Storkus: Methodology, writing–review and editing. Y.L. Bunimovich: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.

This project was supported by the Research Scholar Grant, RSG-19-088-01-CSM, from the American Cancer Society (Y.L. Bunimovich), and by the Hillman Fellows for Innovative Cancer Research Program funded by the Henry L. Hillman Foundation (Y.L. Bunimovich). This project utilized the University of Pittsburgh Imaging Core that is supported in part by award P01HL114453, and the Hillman Cancer Genomics Facility and Bioinformatics Services, supported in part by award P30CA047904. The results published here are in part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. The authors wish to thank Maureen A. Lyons for assisting with RNA-seq, and Drs. Brian M. Davis and Daniel H. Kaplan for their careful reviews and helpful comments provided during the preparation of this manuscript.

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.