Inflammation induced by tumor-associated nerves promotes resistance to anti-PD-1 therapy in cancer patients and is targetable by interleukin-6 blockade

Summmary While the nervous system has reciprocal interactions with both cancer and the immune system, little is known about the potential role of tumor associated nerves (TANs) in modulating anti-tumoral immunity. Moreover, while peri-neural invasion is a well establish poor prognostic factor across cancer types, the mechanisms driving this clinical effect remain unknown. Here, we provide clinical and mechniastic association between TANs damage and resistance to anti-PD-1 therapy. Using electron microscopy, electrical conduction studies, and tumor samples of cutaneous squamous cell carcinoma (cSCC) patients, we showed that cancer cells can destroy myelin sheath and induce TANs degeneration. Multi-omics and spatial analyses of tumor samples from cSCC patients who underwent neoadjuvant anti-PD-1 therapy demonstrated that anti-PD-1 non-responders had higher rates of peri-neural invasion, TANs damage and degeneration compared to responders, both at baseline and following neoadjuvant treatment. Tumors from non-responders were also characterized by a sustained signaling of interferon type I (IFN-I) – known to both propagate nerve degeneration and to dampen anti-tumoral immunity. Peri-neural niches of non-responders were characterized by higher immune activity compared to responders, including immune-suppressive activity of M2 macrophages, and T regulatory cells. This tumor promoting inflammation expanded to the rest of the tumor microenvironment in non-responders. Anti-PD-1 efficacy was dampened by inducing nerve damage prior to treatment administration in a murine model. In contrast, anti-PD-1 efficacy was enhanced by denervation and by interleukin-6 blockade. These findings suggested a potential novel anti-PD-1 resistance drived by TANs damage and inflammation. This resistance mechanism is targetable and may have therapeutic implications in other neurotropic cancers with poor response to anti-PD-1 therapy such as pancreatic, prostate, and breast cancers.


Introduction
The development of cancer immunotherapy, specifically that of programmed cell death 1 (PD-1) blocking antibodies, has ushered in a new era in oncological care. Anti-PD-1 therapy has induced profound and durable tumor regression in specific patient subsets across multiple cancer types 1 . Yet, the majority of patients still do not respond to anti-PD-1 treatment [2][3][4][5] .
Colossal efforts have been invested in identifying potential resistance mechanisms to anti-PD-1 therapy. CD8 + T cell activity within the tumor microenvironment (TME), a key effector of anti-tumor immune activity 6,7 , has been extensively studied, leading to discoveries of new immune checkpoints and immunotherapies 8,9 . However, elimination of cancer cells by CD8 + T cells is only the final chord in an intricate symphony. To migrate into a tumor, become activated, proliferate, and resist exhaustion, CD8 + T cells must interact with not only cancer cells but also multiple other immune cells in the TME that regulate T cell activity 10 . Moreover, the associations of fibroblasts 11 and intra-tumoral bacteria 12 with clinical response to anti-PD-1 therapy suggest that other, non-immune residents of the TME may also regulate the antitumoral immune response.
Tumor infiltration into TANs, known as perineural invasion (PNI), is a well-established adverse prognostic factor in cancer 17,18 , especially in cutaneous squamous cell carcinoma (cSCC) 19 . However, little is known about the role of TANs in regulating anti-tumoral immune activity 20 . This limited knowledge contrasts with the established evidence of bidirectional communication between the peripheral nervous system (PNS) and immune system. The PNS supports hematopoiesis, regulates immune responses against infections, and participates in the creation of immune memory [21][22][23] . Injured peripheral nerves attract immune cells such as M2 macrophages [24][25][26] key players in tumor progression 27,28 and resistance to anti-PD-1 therapy 29 to promote nerve healing and regeneration 24,30 . Yet, the immune-nerve-cancer reciprocal relationship remains largely uncharted. Here, we delineated the intra-tumoral immune and neural phenotypes among cSCC patients who underwent anti-PD-1 therapy and demonstrate the role of TANs in resistance to anti-PD-1 therapy.

Results
Cancer-induced nerve damage is associated with poor clinical response to anti-PD-1 therapy.
To evaluate the potential role of TANs in clinical response to anti-PD-1 therapy we used tumor samples from 55 patients with stage II-IVA cSCC who were enrolled in two clinical trials (NCT03565783 and NCT04154943). All patients underwent neoadjuvant anti-PD-1 therapy with Cemiplimab (Regeneron Pharmaceuticals) followed by surgery (Figure 1a, see baseline and neoadjuvant-treated sample distribution in Supplementary Figure 1). None of the patients underwent radiation treatments prior to the anti-PD-1 therapy. All patients received at least two cycles of Cemiplimab. In one trial (NCT04154943), the patients were allowed to receive up to 4 cycles of neoadjuvant Cemiplimab if they did not progress radiologically or clinically and tolerated the treatment 31 (Figure 1b). Responders (n=31) were defined as patients with less than 10% viable tumor cells at surgery; non-responders (n=16) were defined as patients with more than 50% viable tumor cell in the neoadjuvant-treated surgical specimens, as previously described 31,32 . Patients who had 10%-50% viable tumor cells in the surgical specimens (n=8) were excluded from our cohort a-priori 31,32 since this patient population has been inconsistently assigned to both the responders and non-responders groups in previous neoadjuvant clinical trials [33][34][35][36] . Some patients received adjuvant standard of care treatments after surgery, based on the judgement of the treating physician 31 .
Our first step in examining the potential role of TANs in clinical response to anti-PD-1 therapy was to assess tumors for the presence of PNI, as PNI is the most established and clinically relevant form of cancer-nerve interaction 37 . At baseline, non-responders had a significantly higher incidence of PNI compared to responders (71% versus 20%, respectively, p=0.041, Figure 1c). The definition of PNI is not based on functional evidence of nerve damage -PNI is a histo-morphological phenomenon, defined as the presence of tumor cells abutting or in close proximity to a nerve with encirclement of at least a third of the nerve circumference by tumor; or the presence of cancer cells within the epineurial, perineurial, and/or endoneurial compartments of a nerve 18 Table 1). These transcriptional alterations were over-expressed in neoadjuvant treated tumors of non-responders compared to responders (FDR 0.014, Figure 1f).
To test whether TAN damage may promote resistance to anti-PD-1 therapy, we used two neuromodulated cSCC mouse models 41 . First, we eliminated nerves from the TME by excising and plucking the nerves innervating the skin of immunocompetent SKH1-Elite (SKH1-Hr hr , Charles River 41 ) mice. This procedure, called denervation, was done while preserving skin vasculature, and absence of nerves from the skin was confirmed by histology.
Sham surgery was performed in the control group (Figure 1g). Skin denervation was confirmed one week post-denervation using behavioral testing 41 . SCC cells (B6, ultraviolet induced, SKH1-Hr hr derived 42 ) were orthotopically injected to the denervated skin. Seven days after cancer inoculation, mice were treated with either anti-PD-1 or IgG2 control. Denervated mice demonstrated improved tumor response to anti-PD-1 therapy with significantly lower tumor volumes compared to the control groups (P = 0.03, Figure1h, Supplementary figure 2a). Next, we sought to validate the potential impact of TAN damage on response to anti-PD-1. Nerve damage was induced using surgical axotomy (Figure 1i). In this mouse model, severed nerves are left in place 41 , resulting in Wallerian degeneration 43,44 (anterograde disintegration of axons and their transected myelin sheaths). One-week post-axotomy, cutaneous B6 SCC cells were orthotopically injected into the numb dermatome, followed by treatment with anti-PD-1.

Cancer cells damage nerves by inducing nerve demyelination and degeneration
To decipher the mechanism of TAN damage, we examined the interaction between SCC cells and neurons in vitro. Freshly harvested murine dorsal root ganglia (DRG) neurons were kept intact to maintain the integrity of the explant and prevent compromise of the cellcell contact between neurons, Schwan cells, and endoneurial macrophages 46,47 . DRG neurons were co-cultured with murine SCC cells (Moc1 and B6). As seen in Video 1, the SCC cells were neurotropic, and within 72 hours made direct contact with the axon. The ultra-structural changes associated with the direct cancer-neuron contact were assessed using electron microscopy (EM). Scanning EM images were obtained on day 5 of the co-culture and To confirm that the mechanism driving cancer induce nerve damage is de-myelination, multiplex immunofluorescence stains were conducted on tumor samples from an independent validation cohort of 86 treatment naïve cSCC patients. This cohort included patients with localized (T1-3 50 ) disease who underwent Mohs surgery at the University of Texas MD Anderson Cancer Center. This external cohort was used to test our hypothesis that cancer induced nerve damage occurs early in the disease course and represent an intrinsic cancer cell trait, rather than a marker of advanced disease. Tissue sections were stained for general nerve markers (beta-3-tubulin, B3T), markers of nerve damage (cJUN and ATF3), and markers of de-myelination (degraded myelin base protein, dMBP, and galactosylceramidase, GALC, Figure 2f). We found a significant correlation between nerve insult (ATF3 + cJUN + ) and demyelination (dMBP+, pearson's correlation co-efficient=0.87 p<0.0001, Figure 2g). Due to the proximity of blood vessels to nerve in the TME (neurovascular bundles), we sought to rule out a vascular injury that might contribute to the nerve damage. Immunohistochemistry stain against ERG, a marker for endothelial cells, revealed that nerve damage was not associated with a vascular injury (Supplementary Figure 3). These findings further confirmed that TAN damage is associated with peripheral demyelination.
Demyelination is a hallmark of central neurodegenerative diseases, such as Parkinson's disease, Alzheimer's disease, and amyotrophic lateral sclerosis [51][52][53] . Hence, we sought whether transcriptomic pathways associated with these central neurodegenerative diseases might be present in peripheral nerves exposed to cancer. Freshly harvested human DRG neurons were co-cultured with human cSCC cells (IC8 45 ) for 5 days. Cells were sorted, and NeuO + cells (live neurons) underwent RNA sequencing. Compared to a neuron-only controls, neurons that were co-cultured with cancer cells significantly downregulated genes involved in homeostasis, neuronal repair, and neuronal survival pathways, including the CREB pathway, FAK signaling, synaptogenesis, phagosome formation, calcium signaling, and SNARE complex, FDR < 0.01, Figure 2h). To assess for potential direct effect of the anti-PD-1, human DRG neurons were co-cultured with cSCC cells with and without anti-PD-1 antibodies (Cemiplimab, Regeneron). Next, we assessed for evidence of CAPND in our human cSCC clinical trials cohort.
The degeneration-regeneration homeostatic status of TANs was assessed via NanoString GeoMx Digital Spatial Profiler (DSP). The protein neuron profiling panels included markers of neural degeneration (e.g., α-synuclein, LRRK2 and Park5/7) and neuro-inflammation (e.g., Following a peripheral nerve injury, neurons and Schwan cells attract immune cells to the peri-neural niche to initiate an inflammatory response aimed at nerve healing and regeneration 44 . Hence, we hypothesized that CAPND was associated with the presence of proinflammatory, tumor promoting immune activity. To test this hypothesis, we assessed potential differences in the peri-neural niche immune activity between responders and non-responders. This architectural analysis was done using the DSP protein expression data. Peri-neural niches of neoadjuvant-treated non-responders showed correlation between markers of neuronal response to injury and various immune markers, including immune makers associated with tumor progression such as CD163 (tumor associated macrophages), FOXP3 (T regulatory cells, Tregs), and the immune checkpoints VISTA and IDO-1 (Figure 3b). In contrast, peri-neural niches of responders showed mainly an inverse correlation between markers of neuronal response to injury and immune markers. These findings were validated using multiplex immunofluorescence stains of the peri-neural niches (Figure 3c). Analysis of the peri-neural niches (defined as an area within 150 m from the epicenter of TANs 59 ) in neoadjuvant-treated samples showed that CD68 + CD163 + cells, as well as CD8 + PD1 + and CD8 + LAG3 + cells (exhausted CD8 + T cells) were more abundant in non-responders compared to responders (p=0.055, p=0.078, and p=0.095, respectively, Figure 3d). Collectively, these findings suggested co-localization of a CAPND and an inflammatory, tumor promoting immune activity.
Next, we validated these spatial findings in the cSCC clinical trial cohort. Among the neoadjuvant-treated tumors, region with CAPND phenotype co-localized the tumor-promoting inflammation phenotype higher compared to regions without CAPND (n = 688 of 6571 and 596 of 3019, p < 0.001 Fig. 3h). To further validate these finding, a similar spatial transcriptomic analysis was conducted on tumors derived from our nerve injury mouse hSCC model (see above) treated with the Cemiplimab. The CAPND phenotype was enriched among axotomized mice compared with sham operated mice (Supplementary Figure 7a). These enriched regions were spatially associated with increased tumor-promoting inflamatorry activity in axotomized mice compared to sham operated mice, but not with the anti-tumoral immunity phenotype (Supplementary Figure 7b). Taken together, these results suggested a functional role for CAPND in facilitating an inflammatory, tumor-promoting immune activity that affect the general TME immune tone and hence dampen the clinical efficacy of anti-PD-1 therapy.

Blockade of TAN-induced inflammatory signals enhanced anti-PD-1 efficacy
To further validate the expansion of pro-nerve healing, tumor-promoting inflammation from the per-neural niche to the rest of the TME, we profiled intra-tumoral immune difference between responders and non-responders from our clinical trials cohort. Immunohistochemical staining of tumor samples demonstrated no differences in CD8 + T-cell abundance between responders and non-responders either before or after treatment (Figure 4a). Since CD8 + T-cells could properly infiltrate tumors of non-responding patients, we hypothesized that these T-Cells encountered a hostile TME, leading to their functional impairment. To test this hypothesis, we first stained for PDL-1, since PDL-1 acts as a negative feedback loop suppressing CD8+ T-cell activation 60 inflammation. This resistance mechanism may be relavent to other, non-cSCC neurotropic cancer with an overall poor response to anti-PD-1 therapy such as pancreatic 85 , prostate 86 , and breast 87 cancers.
A key finding of our study is that the TAN-derived anti-PD-1 resistance may be clinically targetable and reversible. Our murine model results demonstrated that combined blockade of PD-1 and the pro-inflammatory cytokine IL-6 improved anti-PD-1 efficacy ( Figure   4). While our results are preliminary, blocking inflammatory signaling to enhance anti-PD-1 clinical efficacy is an exciting and rapidly evolving field, which is already being tested in metastatic melanoma and non-small cell lung cancer patients (NCT04940299, NCT03999749).
As another potential therapeutic approach, the inflammatory signaling might be blocked by addressing its root causenerve degeneration. Neuroprotective agents may, theoretically, dampen CAPND. Moreover, markers of nerve degeneration may serve as future bio-markers to identify patients with lower chances of responding to anti-PD-1. While the current study did not provide evidence in humans for the efficacy of such treatments or biomarkers, it is among the first to introduce the concept of TAN-derived modulation of anti-tumoral immunity, hence supporting future research in this field.
A major limitation of this study is the fact that different patient-based analyses had different sample sizes (Supplementary Figure 1).