Inhibition of Cyclin-dependent Kinase Counteracts Neuron-promoted Bladder and Endometrial Tumorigenesis: A Systematic Analysis Authors: MIN-HONG WANG, MEI-REN PAN, YU-HSUAN HUNG Published: 2026 ABSTRACT Background/AimThe tumor microenvironment is an indispensable component of tumor development. Although fibroblasts and immune cells have been widely studied in tumorigenesis, the role of neurons in cancer has received limited investigation. This study aimed to evaluate the impact of neural signaling on tumorigenesis across cancer types and identify potential therapeutics for neuron-associated tumor development.Materials and MethodsA systematic analysis was performed using the Neuro_Cancer database (https://bio-bigdata.hrbmu.edu.cn/Neuro_Cancer/) and its integrated tools, including neural signal activity, clinical survival, neuro-immune interaction, drug association, and spatial transcriptomic pattern analyses. These analyses were applied across cancer types to identify those strongly influenced by neural activity. Experimental validation was conducted using human bladder and endometrial cancer cell lines co-cultured with neurons, followed by treatment with the cyclin-dependent kinase inhibitor RO-3306.ResultsPrognostic analysis identified bladder, thyroid, and endometrial cancers as tumor types strongly influenced by neural activity. Heatmap analysis of neural activity demonstrated obvious differences in neural intensity, classified as high, moderate, or low, among these cancer types. Neuro-immune interaction analysis showed that lymphocytes and granulocytes were involved in tumorigenesis associated with high neural activity. Drug association analysis identified RO-3306 as a potential therapeutic candidate for counteracting neuron-supported tumor development. Experimental validation showed that co-culture with neurons increased the proliferation of bladder and endometrial cancer cells, whereas RO-3306 treatment abolished this effect. Spatial transcriptomic analysis revealed that neural activity was highest in the core of bladder and endometrial tumors, followed by the tumor boundary, tumor budding area, and normal tissue.ConclusionThese bioinformatic and experimental results highlight the importance of neural signaling in the tumorigenesis of bladder, thyroid, and endometrial cancer and indicate that cyclin-dependent kinase inhibitors may counteract neuron-supported tumor development. Introduction The tumor microenvironment is an essential component of cancer progression and consists of multiple cellular and non-cellular elements, including fibroblasts, extracellular matrix, immune cells, vascular components, and neurons (1-14). Interactions between these components regulate tumor proliferation, invasion, metastasis, and therapeutic responses. Although immune regulation in cancer has been extensively investigated, the contribution of neural signaling to tumor progression has received comparatively less attention. Regarding prognosis, while Zhu et al. found an impact of 5-hydroxytryptamine (5-HT; serotonin) on colon cancer stem cells (CSCs), they also found that 5-HT predicted poor prognosis (15). The authors found that 5-HT increased colon cancer sphere formation in mouse tumors and patient specimens. Inhibition of 5-HT-related enzymes (tryptophan hydroxylase-2) or transporters (serotonin transporter) eliminated this colon CSC population, as well as colon tumorigenesis. The inhibition of 5-HT-related receptors, such as 5-hydroxytryptamine receptor members 1B, 1D, and 1F, had a similar effect on colon CSC suppression. Suppression of these two enzymes reduced the expression of β-catenin. Staining of patient specimens revealed significant upregulation of 5-HT, leucine-rich repeat-containing G protein-coupled receptor 5 (colon CSC marker), β-catenin, and MYC proto-oncogene, bHLH transcription factor (β-catenin target) in advanced tumors. Schmidt et al. applied a spatial analysis of biopsies from 142 patients with head and neck cancer and found that a greater number and larger size of perineural invasion-positive nerves, as well as their closer proximity to the tumor, were linked to worse clinical outcomes, including both overall and disease-specific survival (16). By categorizing perineural invasion-positive nerves as either tumor-proximal or tumor-distal, the authors observed significantly lower levels of myelin basic protein, a myelin marker, in the tumor-proximal group. This finding is noteworthy because its reduced expression is associated with cancer-induced nerve injury, as reported by Baruch et al. (17). Further linking these findings to the increasingly recognized role of cancer immunity, Wang et al. found that nociceptor neurons promoted pancreatic ductal adenocarcinoma via suppression of natural killer (NK) cells (18). Using resiniferatoxin (RTX) to ablate nociceptor neurons, they found that RTX altered mouse behavior and reduced tumor growth after cancer cell inoculation. Single-cell RNA sequencing revealed that NK cells were the main population upregulated in the radiotherapy-treated group, and the activation of these NK cells by RTX was confirmed by flow cytometry based on the expression of tumor necrosis factor α and granzyme B. Depletion of NK cells by antibodies in tumor-bearing mice blunted the antitumor effects of RTX. However, blockade of calcitonin-related polypeptide α and receptor activity-modifying protein 1 (an accessory protein of calcitonin-related polypeptide α) signaling by rimegepant mimicked the effects of RTX. Single-cell RNA sequencing further revealed the contribution of interleukin 15 (IL-15) in nociceptor neuron-affected pancreatic ductal adenocarcinoma, as depletion of IL-15 with an antibody caused the same effect as the NK cell-depleting antibody. IL-15 is expressed in cancer-associated fibroblasts and plays an important role in NK cell proliferation. Staining of patient specimens by multiplex immunohistochemistry confirmed the negative correlation between nociceptor neurons and NK cells, as well as the prognostic importance of the former. Wang and colleagues focused on a single neural protein family, the leucine-rich repeat neuronal (LRRN) family, in breast cancer and found its expression predicted better prognosis and was positively associated with dendritic cells (DCs) and M1 macrophages (19). Despite the increasing recognition of neural involvement in cancer, several important questions remain unresolved. Firstly, the cancer types most strongly associated with neural activity have not been systematically prioritized across cancers. Secondly, the relationships among neural signaling, the tumor immune microenvironment, spatial distribution, and therapeutic vulnerability remain insufficiently characterized. Thirdly, experimental validation of neuron-associated tumor progression identified through large-scale analyses remains limited. Recently, Gao et al. developed the Neuro_Cancer database, an integrated pan-cancer platform designed to investigate neuron-associated regulation in cancer through transcriptomic and spatial analyses (20). This database integrates bulk transcriptomic data from The Cancer Genome Atlas, spatial transcriptomic datasets, single-cell transcriptomic datasets, and pharmacogenomic resources, including the Genomics of Drug Sensitivity in Cancer (GDSC) database. Using neuron-associated gene signatures, the platform enables analysis of neural activity, prognosis, immune interaction, drug association, and spatial transcriptomic distribution across cancer types. Neural activity within the platform was inferred from neuron-associated transcriptomic signatures, and tumors were categorized into high, moderate, and low neural activity groups according to the neural signal intensity generated by the Neuro_Cancer analytical pipeline (20). Building upon this resource, the present study aimed to systematically identify cancer types strongly associated with neural activity and evaluate the relationships among neural signaling, immune perturbation, spatial distribution, and therapeutic response. In addition, experimental co-culture models were used to validate whether neuron-associated cancer cell proliferation might be counteracted by cyclin-dependent kinase inhibition (CDKi). Materials and Methods Bioinformatic analysis. Bioinformatic analyses were performed using the Neuro_Cancer database developed by Gao et al. (20), an integrated pan-cancer platform containing transcriptomic, clinical, spatial transcriptomic, and pharmacogenomic datasets across multiple cancer types. The database incorporates bulk RNA sequencing data from The Cancer Genome Atlas, single-cell transcriptomic datasets, spatial transcriptomic datasets, and drug-sensitivity datasets from public resources, including GDSC (21, 22). Dataset collection, preprocessing, neural signature construction, neural activity scoring, and validation procedures were performed in the original Neuro_Cancer study by Gao et al. (20). Neural activity scores implemented within the platform were derived from neuron-associated gene-expression signatures, as previously described (20). Tumors were categorized into high, moderate, and low neural activity groups according to the neural signal intensity generated by the Neuro_Cancer hierarchical clustering pipeline. To identify cancer types strongly associated with neural signaling, prognostic analyses were performed using overall survival comparisons between patients with tumors with high and low neural activity assessed by the log-rank test. Heatmaps of neural activity were subsequently analyzed according to the neural activity intensity groups provided by the platform. Neuro-immune interaction analyses were performed within the platform to evaluate associations between neural activity and inferred immune cell infiltration profiles integrated within the Neuro_Cancer database. Drug association analysis using the Neuro_Cancer database was performed to evaluate candidate therapeutic compounds predicted to counteract tumors exhibiting high neural activity based on drug-sensitivity profiles and their estimated half-maximal inhibitory concentration values derived from integrated pharmacogenomic datasets, including GDSC (21, 22). Candidate compounds were ranked according to the statistical significance of their predicted association of counteracting high neural activity status. Spatial transcriptomic analyses were performed using publicly available datasets integrated within Neuro_Cancer (from 239 tissue sections and 630,158 spots across 22 cancer types collected from published studies). These comprise representative bladder and endometrial cancer datasets containing annotated tumor regions which were selected for visualization of neural activity distribution. Specifically, the publicly available bladder cancer spatial transcriptomic dataset included in the Neuro_Cancer study that contained a larger annotated tumor region was selected for analysis. Cell lines and reagents. The human bladder cancer cell line 5637 (HTB-9TM) and the endometrial cancer cell line RL95-2 were obtained from the Bioresource Collection and Research Centre (Hsinchu, Taiwan, ROC). The human neuroblastoma cell line SH-SY5Y was obtained from the American Type Culture Collection (Manassas, VA, USA). HTB-9TM cells were cultured in RPMI medium (Procell, Houston, TX, USA). RL95-2 cells were cultured in Dulbecco’s modified Eagle’s medium/F12 medium (Procell). SH-SY5Y cells were cultured in Eagle’s Minimum Essential Medium/F12 medium (American Type Culture Collection). All media were supplemented with 10% fetal bovine serum (GeneTex, Hsinchu, Taiwan, ROC) and 1% penicillin-streptomycin (Procell). SH-SY5Y cells were differentiated into neuron-like cells using 10 μM retinoic acid (MedChemExpress, Monmouth Junction, NJ, USA) for 5 days, as previously described (23). Cell co-culture and drug treatment. To evaluate the effect of neuronal co-culture on cancer cell proliferation and the response to CDKi, human bladder cancer HTB-9TM cells and endometrial cancer RL95-2 cells were directly co-cultured with differentiated SH-SY5Y neuron-like cells. For co-culture experiments, 25,000 HTB-9TM cells or 50,000 RL95-2 cells were seeded together with 50,000 differentiated SH-SY5Y cells. Cancer cells were prelabeled with DiO fluorescent membrane dye (Biotium, Fremont, CA, USA) before co-culture to facilitate selective visualization and counting. Co-culture was performed for 24 h in the presence or absence of 25 μM RO-3306 (MedChemExpress). Additional control groups included cancer cells cultured alone with or without RO-3306 treatment. Vehicle-treated controls received equivalent concentrations of dimethyl sulfoxide. After treatment, fluorescently labeled cancer cells within representative microscopic fields were manually counted under a fluorescence microscope (IX50; Olympus, Tokyo, Japan). Data are presented as the mean ± standard error of the mean of three independent biological replicates. Statistical analysis. Statistical analysis for the bioinformatic analysis was provided by the Neuro_Cancer database using R, and statistical analysis for the cell line experiments was performed using Student’s t-test in GraphPad (GraphPad Software, Boston, MA, USA). Associations and differences with values of p<0.05 were considered statistically significant. All software and algorithms used in this study are freely or commercially available and are listed in the materials and methods section of (20). Any additional information required to reanalyze the data reported here is available from the corresponding author upon request. Results High neural activity predicted poor prognosis in multiple cancer types. To identify the cancer types heavily affected by neural activity, we analyzed prognosis according to neural activity (high versus low) across the following types: sarcoma, thymoma, skin cutaneous melanoma, mesothelioma, lung squamous cell carcinoma, lung adenocarcinoma, liver hepatocellular carcinoma, stomach adenocarcinoma, pancreatic adenocarcinoma, esophageal carcinoma, rectum adenocarcinoma, colon adenocarcinoma, cholangiocarcinoma, ovarian serous cystadenocarcinoma, uterine carcinosarcoma, uterine corpus endometrial carcinoma, cervical squamous cell carcinoma and endocervical adenocarcinoma, breast invasive carcinoma, bladder urothelial carcinoma, testicular germ cell tumor, prostate adenocarcinoma, kidney renal papillary cell carcinoma, kidney renal clear-cell carcinoma, kidney chromophobe, head and neck squamous cell carcinoma, and thyroid carcinoma. We found that the cancer types most heavily affected by neural activity were bladder (log-rank p<0.0001), thyroid (log-rank p<0.0001), and endometrial (log-rank p=0.03963), in which higher neural activity predicted worse overall survival (Figure 1). We therefore focused on these cancer types for further analysis. High neural activity predicted poor prognosis in multiple cancer types. The Neuro_Cancer database clinical survival tool was applied to identify cancer types heavily affected by neural activity, and identified bladder (A), thyroid (B), and endometrial (C) cancer. The presence of lymphocytes and granulocytes was modulated by neural signaling in bladder, thyroid, and endometrial cancer. To determine whether differential neural activity resulted in distinct gene-expression patterns and contributed to prognostic differences in bladder, thyroid, and endometrial cancer, we applied the Neuro_Cancer database heatmap tool and found that high, moderate, and low levels of neural activity produced distinct heatmaps (Figure 2). Heatmap analysis revealed differential neural activity in bladder, thyroid, and endometrial cancer. The Neuro_Cancer database neural signaling activity tool was applied to visualize high (red), moderate (orange), and low (blue) levels of neural signaling in bladder (A), thyroid (B), and endometrial (C) cancer. Because the tumor immune microenvironment is one of the main components determining disease outcome (24), the associations between immune cells and neural activity in bladder, thyroid, and endometrial cancer were analyzed. Compared with moderate/low neural activity, high neural activity in bladder cancer was associated with lower levels of DCs and CD8+ lymphocytes (Figure 3A). In thyroid cancer, high neural activity was associated with a lower level of DCs but higher levels of regulatory T-cells (Tregs) and M2 macrophages (Figure 3B). In endometrial cancer, high neural activity was associated with lower levels of follicular T-helper (Tfh) cells and γδ T-cells, but a higher level of naïve CD8+ T-cells (Figure 3C). Neuro-immune analysis revealed that dendritic cells (DCs), lymphocytes, and granulocytes were associated with neural activity in bladder, thyroid, and endometrial cancer. The Neuro_Cancer database neuro-immune interaction tool was applied and the relative abundance of various cells was determined using quanTIseq, CIBERSORT, CIBERSORT_abs and xCell algorithms (20). The main immune cell types heavily affected by neural activity in bladder (A), thyroid (B), and endometrial (C) cancer are shown ranked by p-value. According to (20), comparisons between groups were performed using a two-tailed Wilcoxon rank-sum test. The upper edge of the box represents the third quartile, the middle line represents the median, and the lower edge represents the first quartile. The upper and lower whiskers indicate the range of non-outlier values. The underlying data were from 239 tissue sections and 630,158 spots across 22 cancer types collected from published studies [see (20) for details]. Tfh: Follicular T-helper cells; Tgd: γδ T-cells; Tregs: regulatory T-cells. CDKi counteracted neuron-promoted tumorigenesis in bladder, thyroid, and endometrial cancer. To counteract neuron-supported tumorigenesis in bladder, thyroid, and endometrial cancer, we applied the Neuro_Cancer database drug-association tool and found that high neural activity was associated with sensitivity to the CDKi RO-3306 (bladder cancer, p=1.536e-39; thyroid cancer, p=2.331e-14; and endometrial cancer, p=1.418e-06) (Figure 4). Drug analysis revealed that cyclin-dependent kinase inhibitor efficacy was significantly associated with high neural activity in bladder, thyroid, and endometrial cancer. Drug analysis using the Neuro_Cancer database drug-association tool showed an association between high neural activity and sensitivity to the cyclin-dependent kinase inhibitor RO-3306 in bladder (A), thyroid (B), and endometrial (C) cancer. According to (20), comparisons between groups were performed using a two-tailed Wilcoxon rank-sum test. The upper edge of the box represents the third quartile, the middle line represents the median, and the lower edge represents the first quartile. The upper and lower whiskers indicate the range of non-outlier values. Experimental validation was carried out by in-vitro cell co-culture with SH-SY5Y cells with neural differentiation induced by retinoic acid and treatment using RO-3306 at its half-maximal inhibitory concentration [25 μM, from the Neuro_Cancer (20) and GDSC (21, 22) databases]. This experiment showed that although the proliferation of human bladder cancer cells (HTB-9TM) and endometrial cancer cells (RL95-2) increased when co-cultured with these differentiated SH-SY5Y cells, this effect was blunted by RO-3306 (Figure 5). A human thyroid cancer cell line without subtype specification for comprehensive analysis was not available for study. However, this experiment confirmed that CDKi counteracted neuron-supported tumorigenesis in bladder and endometrial cancer cells. Experimental validation showed that the cyclin-dependent kinase inhibitor RO-3306 counteracted neuron-assisted cancer cell proliferation in bladder and endometrial cancer cell lines. The human neuroblastoma cell line SH-SY5Y was differentiated into neuron-like cells using retinoic acid and co-cultured with the human bladder cancer cell line HTB-9TM (A) or the endometrial cancer cell line RL95-2 (B); a human thyroid cancer cell line without subtype specification for comprehensive study was not available. DMSO: Dimethyl sulfoxide control. Data are the mean ± the standard error of the mean. Statistical analysis was performed with Student’s t-test. Spatial transcriptomic analysis revealed high neural activity in the core of bladder and endometrial tumors. Finally, we analyzed spatial transcriptomics to determine whether different tumor areas exhibited different levels of neural activity. Spatial transcriptomic analysis in bladder cancer revealed the highest neural activity in the tumor core, followed by the tumor boundary, tumor budding area, immune cells, and normal tissue (Figure 6A). In endometrial cancer, although no immune cell infiltration was observed, the spatial pattern from tumor core to normal tissue was preserved (Figure 6B). Although no spatial transcriptomic data were available for thyroid cancer analysis, these results further strengthen the importance of neural activity in these cancer types. Spatial transcriptomic analysis revealed differential neural activity across the tumor core, tumor boundary, tumor budding area, in immune cells, and in normal tissue in bladder and endometrial carcinoma. The Neuro_Cancer database spatial pattern tool was applied, and neural activity in different compartments in bladder (A) and endometrial (B) cancer is shown. No spatial transcriptomic data were available for thyroid cancer analysis. The upper edge of the box represents the third quartile, the middle line represents the median, and the lower edge represents the first quartile. The upper and lower whiskers indicate the range of non-outlier values. Discussion The present findings suggest that neural activity is associated with substantial alterations in the tumor immune microenvironment. In bladder cancer, high neural activity is associated with reduced DCs and CD8+ T-cells, which is consistent with previous observations that reduced DC infiltration and impaired CD8+ T-cell responses are associated with unfavorable bladder cancer outcomes (25, 26). Similarly, thyroid cancer with high neural activity was found to exhibit increased levels of regulatory T-cells (Tregs) and M2 macrophages, both of which have been associated with immune suppression and recurrence of thyroid cancer (27-29). In endometrial cancer, alterations involving Tfh cells, γδ T-cells, and CD8+ T-cells further support the potential role of neural signaling in shaping immune regulation within the tumor microenvironment (30-32). Although direct investigations regarding the crosstalk between neurons and DCs/CD8+ T-cells in bladder cancer, DCs/Tregs/M2 macrophages in thyroid cancer, and Tfh cells/γδ T-cells/CD8+ T-cells in endometrial cancer remain limited, previous studies have suggested the existence of complex neuroimmune interactions within the tumor microenvironment. In 2023, Patwardhan et al. reviewed the interplay between neurons, DCs, and lymphocytes in bladder cancer (33). In this context, neuron-derived factors, including brain-derived neurotrophic factor and catecholamines, may support bladder cancer progression, and tumor cells subsequently express programmed cell death 1 (PD1, CD274) and IL-10 to suppress lymphocyte and DC activity, thereby contributing to immune escape. Later, in 2025, other researchers revisited the therapeutic importance of neurons across multiple cancer types. Using cancer models of the head, neck, pancreas, and skin, Baruch et al. found that cancer-induced nerve injury (CINI) promoted resistance to PD1-based immunotherapy (17). Using an initial cohort of 47 patients with advanced head and neck cancer, the authors collected patient specimens before and after anti-PD1 treatment. Spatial analyses of RNA and proteins were performed to understand the relationship between CINI and resistance to anti-PD1. In accordance with their previous study (34), they found that perineural invasion was associated with immunotherapy resistance and validated this observation using datasets from other cancer types. Mouse experiments with immunocompetent C57BL/6 mice confirmed that axotomy-induced nerve injury led to resistance to anti-PD1 therapy, with real-time evidence demonstrated by scanning electron microscopy. This type of nerve injury, characterized by expression of activating transcription factor 3 and degradation of myelin basic protein, is associated with exhausted CD8+ T-cells (positive for lymphocyte-activating gene 3) and overall immunosuppression. On the other hand, Zhao et al. used brain and pancreatic cancer models to show that radiotherapy-induced nerve senescence helped cancer cells survive (35). As senescence is intertwined with metabolism, especially lipid metabolism, the authors performed targeted metabolic analysis and found that prostaglandin E2 (PGE2) was significantly upregulated when neurons became senescent following exposure to conditioned medium from cancer cells treated with chemotherapeutic agents. Downstream signaling analysis revealed the regulation of PGE2 by asparagine endopeptidase (AEP), eukaryotic translation initiation factor 4A1 (eIF4A1), and prostaglandin E synthase 3 (PTGES3); radiotherapy and chemotherapy triggered AEP to cleave eIF4A1, which in turn stabilized PTGES3 mRNA, subsequently increasing PGE2. Blockade of either AEP or PGE2 re-sensitized tumors to radiotherapy. Our study also showed that a CDKi may counteract neuron-promoted tumorigenesis, especially in the bladder and endometrium. The observation that immune cell-enriched regions also exhibit neural activity is consistent with increasing evidence supporting bidirectional neuroimmune interactions within the tumor microenvironment (36). One potential approach to further dissect the complex interactions among neurons, immune cells, and cancer cells is the application of microfluidic chip systems, which are increasingly being used in cancer research (37). Future studies combining neuron-cancer co-culture systems with patient-derived immune cells obtained under appropriate institutional review board approval may help further clarify the mechanisms through which neural signaling regulates the tumor immune microenvironment and tumorigenesis. Conclusion Systematic analysis using the Neuro_Cancer database revealed that the cancer types most heavily affected by neural activity were bladder, thyroid, and endometrial. Our findings suggest that increased neural activity is associated with poorer survival outcomes, marked alterations of the immune microenvironment, sensitivity to certain therapeutic agents such as CDKi, and prominent spatial enrichment of neural signals within tumor cores, collectively highlighting neural activity as a clinically relevant feature of tumor biology and a potential therapeutic vulnerability. Conflicts of Interest None. Authors’ Contributions Conceptualization: M.H.W., Y.H.H. Methodology: Y.H.H. Validation: Y.H.H. Formal analysis: Y.H.H. Resources: M.R.P., Y.H.H. Data curation: Y.H.H. Writing: Y.H.H. Supervision: Y.H.H. Project administration: Y.H.H. Funding acquisition: Y.H.H. Funding This study was supported by grant KMUH113-3R70 from KMUH to Y.H.H. This study is supported partially by Kaohsiung Medical University Research Center Grant (KMU-TC114A204). Artificial Intelligence (AI) Disclosure During the preparation of this manuscript, a large language model (ChatGPT, OpenAI) was used solely for language editing and stylistic improvements in select paragraphs. No sections involving the generation, analysis, or interpretation of research data were produced by generative AI. All scientific content was created and verified by the authors. Furthermore, no figures or visual data were generated or modified using generative AI or machine learning-based image enhancement tools.