Single-cell RNA sequencing uncovers a neuron-like macrophage subset associated with cancer pain

Abstract

Tumor innervation is a common phenomenon with unknown mechanism. Here, we discovered a direct mechanism of tumor-associated macrophage (TAM) for promoting de novo neurogenesis via a subset showing neuronal phenotypes and pain receptor expression associated with cancer-driven nocifensive behaviors. This subset is rich in lung adenocarcinoma associated with poorer prognosis. By elucidating the transcriptome dynamics of TAM with single-cell resolution, we discovered a phenomenon "macrophage to neuron-like cell transition" (MNT) for directly promoting tumoral neurogenesis, evidenced by macrophage depletion and fate-mapping study in lung carcinoma models. Encouragingly, we detected neuronal phenotypes and activities of the bone marrow-derived MNT cells (MNTs) in vitro. Adoptive transfer of MNTs into NOD/SCID mice markedly enhanced their cancer-associated nocifensive behaviors. We identified macrophage-specific Smad3 as a pivotal regulator for promoting MNT at the genomic level; its disruption effectively blocked the tumor innervation and cancer-dependent nocifensive behaviors in vivo. Thus, MNT may represent a precision therapeutic target for cancer pain.

Fig. 1.. Macrophage-specific scRNA-seq reveals a previously unknown neuron-like TAM subset.

(A) t-Distributed stochastic neighbor embedding (t-SNE) of 10×-based scRNA-seq of LysM-tdTomato⁺ macrophage lineage cells from the LLC tumors on LysM-Cre/ROSA-tdTomato mice. (B) GO analysis of up-regulated DEGs in the macrophage lineage–derived Tubb3⁺ cells (red cluster) reveals their association with neurogenesis as shown by GO terms “Development_Neurogenesis_Synaptogenesis” (P = 0.01980) and “Development_Neurogenesis_in_general” (P = 0.02536). (C) Expression level of neuronal genes in the t-SNE plot of mouse Tubb3⁺ TAMs. (D) Two-photon confocal live imaging detects calcium efflux in the LysM-tdTomato⁺ macrophage-derived neuronal-like cells (red) in the LLC tumor ex vivo, as shown by their concentrated OGB-1 (green fluorescent calcium indicator dye) intensity. (E) Clodronate liposomes (Lip-Clodronate)–mediated macrophage depletion significantly reduces Tubb3⁺ TAMs in LLC tumor compared to their control group [liposome phosphate-buffered saline (Lip-PBS)], as shown by immunofluorescence (n = 4; ***P < 0.001 versus Lip-PBS, t test), and (F) is associated with a significant reduction of spontaneous nocifensive behaviors (total counts of flinching and licking observed in 30 min) of the LLC-bearing mice on day 20 [n = 4 (duplicate); ***P < 0.001 versus Lip-PBS, t test]. Scale bars, 50 μm.

Fig. 2.. Presence of the neuron-like TAM subset in NSCLC.

(A) Existence of the TUBB3⁺CD68 neuron-like TAM subset (yellow) and their expression levels of neuronal genes (red) in a surgical resection of human NSCLC biopsy detected by 10× scRNA-seq. (B) Detection of the NeuN⁺ TAM in patient biopsies of renal cell carcinoma and hepatocellular carcinoma by immunofluorescence (quantification in fig. S4B). (C) Confocal microscopy with z-stack scanning confirms a TUBB3⁺ TAM subset (TUBB3⁺ CD68, yellow) in NSCLC biopsy. (D) The levels of TUBB3⁺ TAM in tumor and normal lung tissues were quantified with flow cytometry analysis (n = 5; t test, ***P < 0.001 versus Normal). Cohort analysis of adenocarcinoma NSCLC shows significance in (E) correlation (% area of total TUBB3⁺ cells versus ratio of TUBB3⁺ TAMs/total CD68; ***P < 0.0001, Pearson correlation; n = 102) and (F) higher mortality of patients with high TUBB3⁺ TAM level, quantified by multiplex immunohistochemical staining (n = 94; P < 0.033, log-rank test). (G) The up-regulated DEGs of human TUBB3⁺ CD68⁺ TAMs are highly associated with neuronal diseases in GO analysis (Huntington’s disease, Parkinson’s disease, and Alzheimer’s disease, red). Scale bars, 50 μm.

Fig. 3.. MNT in TME.

(A) RNA velocity analysis shows a macrophage lineage–derived Tubb3⁺ population (cluster 4, purple) derived from the F4/80⁺ TAM (cluster 0, blue) in the LysM-tdTomato cell scRNA-seq dataset from Fig. 1A. (B) Their top expression genes show an obvious macrophage and neuronal signatures in clusters 0 and 4. (C) UMAP of the LysM-tdTomato cell scRNA-seq dataset from Fig. 1A, where the Tubb3-rich cluster (cluster 4, purple) is linked but is discrete from Cd68- and Adgre1 (F4/80)–rich cluster (cluster 0, blue) among seven clusters of the macrophage lineage cells in the TME. (D) Pseudotime analysis consistently demonstrates a complete transition from Tubb3⁻ Cd68⁺ TAM into Tubb3⁺ Cd68⁻ neuron-like cells in the TME at the transcriptional level. (E) Fate-mapping study evidences the existence of macrophage lineage–derived Tubb3⁺LysM-tdTomato⁺ population in the LLC tumor on day 15 (quantification in fig. S8, A and B). (F) Adoptive transfer of GFP⁺ BMDMs results in the production of GFP⁺ Tubb3⁺ cells in the LLC tumor on day 25 detected at the protein level by immunofluorescence (n = 5; quantification in fig. S8C). Scale bars, 50 μm.

Fig. 4.. Neuronal phenotypes and activity of MNTs.

(A) Diphtheria toxin–mediated macrophage depletion reduces Tubb3⁺F4/80⁺ MNTs in the LLC tumor of LysM/iDTR mice, which is successfully rescued by the adoptive transfer of BMDM as shown by immunofluorescence and quantification [**P < 0.01 versus Control, ^###P < 0.01 versus DT, one-way analysis of variance (ANOVA); n = 5]. (B) Morphological changes of BMDMs under stimulation with TGF-β1 (5 ng/ml for 5 days), cancer cell conditioned medium LLC-CM [10% (v/v) for 7 days], and their combination (TGF-β1 for 5 days and then LLC-CM for 7 days) in vitro (quantification in fig. S8D). (C) The de novo expression of neuronal differentiation markers in BMDM under TGF-β1 (TGF) and LLC-CM (CM) is detected by Western blot analysis (quantification in fig. S8E; ***P < 0.001 versus Control; ^###P < 0.001 versus TGF-β1, one-way ANOVA; n = 3). (D) Unbiased bioinformatic analysis reconstructs a neurotransmitter-centric gene network by using the up-regulated DEGs of MNTs (Tubb3⁺ TAM in Fig. 1B). (E) Strong expression of Tubb3 (green) and neuron-like dendrites is detected in the in vitro generated MNTs but not in BMDMs (quantification in fig. S11B) as shown by immunofluorescence, where (F) capsaicin (activator of nociceptive vanilloid receptor) (41) markedly induces calcium efflux of the MNTs but not BMDMs as detected by the change of OGB-1 fluorescent intensity (ΔF/F). (G) Local adoptive transfer of in vitro generated MNTs significantly increased spontaneous nocifensive behaviors of the LLC-bearing NOD/SCID mice compared to their LLC only control in vivo [***P < 0.001 versus Control, t test; n = 5 (duplicate)], (H) where MNTs in LLC tumor highly express pain mediators Ano1 and Trpc1 detected by macrophage lineage scRNA-seq dataset from Fig. 1A. Scale bars, 50 μm (A) and 25 μm (B and E).

Fig. 5.. Smad3 is essential for MNT formation in the TME.

(A) Representative image of the NSCLC cohort shows hyperactivation of SMAD3 (p-SMAD3, blue) in human TUBB3⁺CD68⁺ MNTs as shown by confocal imaging. (B) NSCLC cohort reveals a positive correlation between TAM-specific SMAD3 activation (p-SMAD3⁺CD68⁺) and MNT abundancy in NSCLC, detected by quantification of % stained area in multiplex immunohistochemistry (*P = 0.0233, Pearson correlation = 0.1965; n = 102). Reduction of mouse Tubb3⁺F4/80⁺ MNTs in the LLC tumor of mice with Smad3-KO TME (Smad3-KO mice), detected by (C) immunofluorescence and (D) flow cytometry analysis [n = 5 (duplicate); ***P < 0.001 versus Smad3-WT, t test]. (E) LLC cell inoculation markedly induces cancer-associated nocifensive behaviors (total counts of flinching and licking in 30 min) in the WT mice, which is significantly prevented in the Smad3-KO mice [***P < 0.001 versus tumor-free control (CTL); ^###P < 0.001 versus tumor-bearing WT mice, one-way ANOVA; n = 5 (duplicate)]. (F) LLC-CM Smad3 dependently induces de novo mRNA expression of neuronal differentiation markers Pou4f1, NeuN, and Tubb3 in BMDMs in vitro (***P < 0.001 versus Control, ^###P < 0.001 versus WT + LLC-CM, one-way ANOVA; n = 6). Local adoptive transfer of WT BMDMs (WT-BMDM) increases (G) neuronal marker expressing MNTs (F4/80⁺ Tubb3⁺, F4/80⁺ NeuN⁺) in the LLC tumors of NOD/SCID mice, (H) contributing to the significant increase of their cancer-associated nocifensive behaviors (total counts of flinching and licking observed in 30 min), which are effectively canceled in mice that received Smad3-KO BMDMs (KO-BMDM) (***P < 0.001 versus Control, ^###P < 0.001 versus WT-BMDM, one-way ANOVA; n = 5). Scale bars, 50 μm.

Fig. 6.. Targeting Smad3 effectively blocks MNT-driven neurogenesis in vivo.

(A) ECR browser detects a direct binding site of Smad3 of the conserved region of mouse and human Tubb3 genomic sequence at 5′UTR (purple). (B) TGF-β1 (5 ng/ml) stimulation enriches the physical interaction of Smad3 protein on the 5′UTR of Tubb3 genomic sequence in BMDM in vitro as confirmed by ChIP-seq. The Tubb3⁺LysM-tdTomato⁺ MNTs are FACS-collected from the LLC-bearing LysM-Cre/ROSA-tdTomato mice, and their Smad3-bound chromatins are pulled down and submitted for DNA sequencing. The identified Smad3 binding sites in the genome of the in vivo MNTs are shown as (C) density distribution reads per kilobase million (RPKM) plot, (D) heatmap, (E) genome region chart, and (F) motif analysis. (G) There are 7 neuron differentiation and 12 nervous system development-related genes (red) detected as Smad3 direct target genes in the tumoral MNTs as revealed by GO analysis. Treatment with the pharmaceutical Smad3 inhibitor SIS3 dose-dependently blocks the formation of (H) Tubb3⁺ and NeuN⁺ MNTs and (I) cancer-associated spontaneous nocifensive behaviors (total counts of flinching and licking observed in 30 min, duplicate) of the LLC-bearing mice in vivo [***P < 0.001 versus solvent control, ^##P < 0.01 versus SIS3 (5 μg/g), one-way ANOVA; n = 5]. Scale bars, 50 μm.