Stromal nicotinamide N-methyltransferase orchestrates the crosstalk between fibroblasts and tumour cells in oral squamous cell carcinoma: evidence from patient-derived assembled organoids

Nicotinamide N-methyltransferase (NNMT) has been reported to be linked to methylation reprogramming in cancer cells. However, the role of NNMT in the tumour microenvironment (TME) remains elusive. Here, we found that the expression of NNMT was elevated in the stroma of oral squamous cell carcinoma (OSCC). Using a fibroblast-attached organoids (FAOs) model, we confirmed that stromal NNMT expression contributed to the generation of assembled tumour organoids. In a tumour regeneration assay with co-implanted OSCC cells and cancer-associated fibroblasts (CAFs), the tumour-initiating activity was reduced when NNMT was silenced in CAFs. In contrast, overexpression of NNMT in paracancerous fibroblasts (PFs) accelerated tumour growth in co-inoculation experiments. Notably, fibroblast-specific NNMT can regulate type I collagen deposition in both FAOs and xenografts. Further investigations confirmed that the stromal NNMT-aggravated oncogenic activities were attenuated by treatment with inhibitors of either collagen synthesis (e.g. losartan, tranilast, and halofuginone) in fibroblasts, or the focal adhesion kinase (FAK) signal (i.e. defactinib) in cancer cells. Mechanistically, overexpression of NNMT reduced the enrichment of H3K27me3 at the promoter of the gene encoding lysyl oxidase (LOX), a key enzyme that regulates the cross-linking of collagen I. Overall, we propose that the NNMT-LOX-FAK cascade contributes to the crosstalk between cancer cells and fibroblasts during OSCC development, and that NNMT-centric extracellular matrix remodelling is a novel therapeutic target for patients with OSCC.


INTRODUCTION
Stromal cell reprogramming in the tumour microenvironment (TME) is an emerging hallmark of cancer [1][2][3][4]. Of the novel TME targets, nicotinamide N-methyltransferase (NNMT) is a metabolic enzyme that catalyses the transfer of methyl groups from S-adenosylmethionine (SAM) to nicotinamide [5]. Since NNMT functions as a consumer of methyl donors (i.e. SAM), the overexpression of NNMT can also lead to global histone hypomethylation [5]. For a long period of time, NNMT was only considered an epigenetic regulator in cancer cells [6][7][8][9]. More recently, proteomic analysis of tumour specimens, including not only cancer cells but also stromal cells, has clarified that upregulation of NNMT is highly restricted to cancer-associated fibroblasts (CAFs) [10]. Given this evidence, the detailed mechanisms by which stromal NNMT affects tumour biology deserve further investigation.
Oral squamous cell carcinoma (OSCC) is commonly infiltrated by CAFs [11], which not only secrete oncogenic cytokines and growth factors to facilitate tumour growth but also mediate extracellular matrix (ECM) remodelling by building up matrisome proteins, such as type I collagen [12][13][14]. Of note, the clinical relevance of type I collagen in cancer remains context-dependent [15][16][17][18][19][20]. In several studies based on human cancer cell lines, the density of collagen I is associated with enhanced tumour cell survival, proliferation and migration and may inhibit anticancer drug delivery [16][17][18]. On the other hand, in a genetically engineered mouse model, deletion of the gene encoding collagen I (i.e. Col1a1) in myofibroblasts unexpectedly accelerates cancer progression in animals [19,20]. Due to these controversial findings, further studies are needed to understand collagen biogenesis in the TME.
Organoid technology provides powerful strategies to acquire new insight into the human TME [21][22][23]. For example, the 3D coculture system of pancreatic cancer organoids with pancreatic stellate cells, either directly or indirectly, has pushed forwards the discovery of intratumor CAF heterogeneity in vivo, with myofibroblast-like CAFs located adjacent to the tumour nest and inflammatory CAFs located more distant from cancer cells [24]. In recent studies, the assembled organoid system was further exploited to model the spatial integration of stromal cells into tumour units, which allowed for the systematic investigation of cancer-stroma crosstalk [25]. As a case in point, we have recently developed a fibroblast-attached organoids (FAOs) model, in which the spatial localisation of CAFs surrounding tumour nest was better recapitulated compared to traditional 3D coculture system [26,27]. Considering the potential role of NNMT in ECM remodelling [10], here, we employed FAOs to clarify whether stromal NNMT can regulate collagen deposition and tumour development in OSCC.

RESULTS
Expression of NNMT is elevated in the stromal compartment of the OSCC TME To determine the prognostic value of NNMT in patients with OSCC, we examined the expression of NNMT in the TCGA database using the Gene Expression Profiling Interactive Analysis (GEPIA) website. The results of the survival analysis indicated that in a cohort of patients with head and neck cancer (n = 518), the overexpression of NNMT was correlated with worse overall survival (Fig. 1A). To verify the expression pattern of NNMT in OSCC specimens, a panel of tumour (n = 55) and paracancerous samples (n = 8) were collected to perform immunohistochemical (IHC) staining for NNMT. The results revealed that there was no difference in the expression of NNMT in epithelial cells between OSCC and oral mucosa tissues (Fig. 1B, Supplementary Fig. 1A, B). Nonetheless, stromal NNMT expression in OSCC was significantly elevated compared with that in mucosa samples (Fig. 1B, C, Supplementary  Fig. 1A). More importantly, the stromal NNMT expression in tumours with more advanced clinical stage (i.e. [3][4] was relatively higher than those with earlier stage (i.e. 1-2) (Fig. 1C), indicating a positive correlation between stromal NNMT expression and OSCC development. Since cancer-associated fibroblasts (CAFs) represent a central component in the stroma of OSCC [11], to validate the expression of NNMT in CAFs and tumour cells, patient-derived organoids (PDOs) and corresponding fibroblasts were generated using OSCC and tumour-adjacent tissues. The results of the immunofluorescence (IF) staining confirmed that the tumour organoids were stained negative for NNMT and α-SMA (Fig. 1D), a typical marker for activated fibroblasts [28], and this staining pattern was similar to that of OSCC cells in vivo. Whereas, the colocalization of NNMT and α-SMA in stromal cells, as shown in the IF staining image of OSCC tissue, was highly recapitulated in CAFs (Fig. 1D). In line with these findings, the upregulation of NNMT in CAFs compared to either cancer cells or paracancerous fibroblasts (PFs) was further demonstrated by immunoblotting (IB) analysis (Fig. 1E). To clarify whether the interaction with OSCC cells can promote the expression of NNMT in fibroblasts, we obtained healthy gingival fibroblasts (HGFs) from patients undergoing tooth extraction (n = 3). Then, HGFs and PDOs were used to perform transwell coculture assays according to a reported protocol [24]. The results of the IB analysis clearly showed that coculture with PDOs promoted the expression of NNMT in HGFs (Fig. 1F). Overall, these data indicate that enhanced NNMT expression is an exclusive feature of CAFs in the OSCC TME. To better study the role of CAFspecified NNMT in OSCC development, OSCC cells and parallel CAFs were aggregated to construct fibroblast-attached organoids (FAO) (Supplementary Fig. 1C). Results of the bright field imageing, Hematoxylin/Eosin (HE) and IF staining clearly illustrated that the spatial integration of CAFs, located on both periphery and centre of tumour cells units, were better modelled in FAOs than in traditional tumour organoids ( Supplementary Fig. 1D-F). More importantly, the colocalization pattern of NNMT and α-SMA in vivo were highly recapitulated in FAOs (Fig. 1D), indicating that the FAO may be a competent model to study the role of stromal NNMT in OSCC development.
Stromal NNMT expression is essential for the regeneration of FAOs It is known that the activation of CAFs is closely correlated with tumour development [12]. To validate the role of NNMT in CAF activation, several lines of CAFs from patients with OSCC were treated with an inhibitor of NNMT (i.e. NNMTi). The results of the in-gel colony formation assay clearly showed that treatment with NNMTi reduced length of extensions of activated CAFs colonies ( Fig. 2A, B). In addition, IB results confirmed that the expression of activated fibroblast markers, such as COL1A1 and α-SMA, was decreased upon treatment with NNMTi ( Fig. 2C), demonstrating that the expression of NNMT contributes to the maintenance of the CAF phenotype. To clarify the role of stromal NNMT in OSCC development, CAFs were transfected with a short-hairpin RNA targeting the expression of NNMT (sh-NNMT). Decreased expression of NNMT in CAFs transfected with sh-NNMT was confirmed by immunoblotting assay (Supplementary Fig. 2A, B). Then, the CAFssh-NNMT were employed to construct FAOs. We found that knockdown of NNMT not only blocked the morphological activation of CAFs incorporated into tumour organoids, but also reduced the colony diameter and serial passage efficiency of the entire FAO ( Fig. 2D-F). In contrast, overexpression of NNMT (oe-NNMT) in corresponding PFs ( Supplementary Fig. 1C), a potential precursor cell of CAFs [13], can induce the morphological activation of fibroblasts incorporated into tumour organoids and promote the colony diameter, as well as the serial passage efficiency of the entire FAO ( Fig. 2G-I). More interestingly, in CRISPR-mediated knockout experiments, the colocalization pattern of NNMT and α-SMA in FAOs was degenerated when NNMT was knocked out in CAFs (Fig. 2J, K). Overall, these results indicate that stromal NNMT expression is essential for the symbiotic environment of OSCC cells with CAFs.
Stromal NNMT expression can promote the tumour-initiating activity of OSCC cells To exemplify whether stromal NNMT contributes to tumour growth in vivo, sphere-forming OSCC cells at different dosages, as well as CAFs transfected with either sh-NNMT or control vector (sh-Cont), were co-implanted into BALB/c nude mice for a limited dilution assay (LDA). The results demonstrated that in the coimplanted xenograft model, the tumour-initiating frequency (TIF) of sphere cells was decreased when NNMT expression in CAFs was knocked down (Fig. 3A). Consistent with this finding, knockdown of NNMT in CAFs reduced the growth parameters (e.g. volume and weight) of the co-implanted xenografts (Fig. 3B, C), demonstrating that stromal NNMT expression contributes to OSCC tumour growth. To further investigate the role of stromal NNMT expression in OSCC development, IF and IHC staining were performed to analyse the cellular composition of the co-implanted xenografts. The results showed that knockdown of NNMT reduced the frequency of CAFs (i.e. GFP + cells) in xenografts (Fig. 3D, E), indicating that the expression of NNMT is essential for the biogenesis of CAFs in the TME. More importantly, knockdown of NNMT in CAFs decreased the expression of CD44, a typical cancer stem cells (CSCs) marker [29], and Ki67, a proliferative cell marker Fig. 1 Expression of NNMT is elevated in stromal compartment of OSCC-TME. A Kaplan-Meier curves from TCGA database showed that high NNMT expression in head and neck cancer (HNSC) patients was positively correlated with poor survival. B Representative images showed the immunohistochemical (IHC) staining of NNMT in tumour samples, as well as the adjacent mucosal specimens from patients with OSCC. Scale bars, 100 μm. C Quantification analysis of IHC staining showed the stromal NNMT expression scores in OSCC tumour with clinical stage at 1-2 (n = 25) and 3-4 (n = 30), as well as that in oral mucosal specimens (n = 8). **p < 0.01; ****p < 0.0001. D Representative immunofluorescence (IF) staining images showed the expression pattern of NNMT and α-SMA in OSCC specimens and corresponding tumour organoids, paracancerous frbroblasts (PFs), cancer-associated fibroblasts (CAFs), and fibroblast-attached organoid (FAO). Scale bars, 50 μm. E Immunoblotting assay validated that the expression of NNMT in CAFs was elevated than that in corresponding tumour organoid and PFs. Data were obtained using specimens from patient-2, 6, 9, (P-2, 6,9), and the clinical information of relevant patients were described in our previous study (J Dent Res. 2021. 100(2): 201-208). F Immunoblotting assay showed the expression of NNMT in health gingival fibroblasts (HGFs) that were co-cultured with several lines of OSCC-organoids.
[30], in co-implanted xenografts (Fig. 3D, F, G), implying that stromal NNMT overexpression may promote the cancer stemness and malignant progression of OSCC in vivo. To investigate whether NNMT can trigger the properties of CAFs in their precursor cells, sphere-forming OSCC cells at different dosages, and PFs transfected with either oe-NNMT or control vector (oe-Cont), were co-implanted into BALB/c nude mice for a limited dilution assay (LDA). The results showed that the overexpression of NNMT in PFs increased the TIF of sphere cells (Fig. 3H) and promoted the parameters (e.g. volume and weight) of the coimplanted xenografts (Fig. 3I, J). Simultaneously, the frequency of fibroblasts (i.e. GFP + cells) in xenografts, as well as the cells stained positive for CD44 and Ki67, were drastically increased upon PF-restricted NNMT overexpression ( Fig. 3K-N). Taken together, these results clearly demonstrate that stromal NNMT contributes to tumour-initiating activity of OSCC cells.
Stromal NNMT expression can modulate collagen deposition in the OSCC TME Studies have reported that overexpression of NNMT is associated with extracellular matrix (ECM) remodelling in the TME [10]. Since type I collagen is the most abundant component of the ECM [14], we questioned whether NNMT can regulate collagen I deposition in OSCC TME. To address this issue, bioinformatics analysis was carried out to gain some preliminary evidence. The results of Gene Expression Profiling Interactive Analysis (GEPIA) showed that in patients with head and neck cancer, the expression of NNMT was closely correlated with that of the genes related to type I collagen biogenesis (Fig. 4A, B). Consistent with this, the IF staining of serial sectioning images confirmed that the expression of NNMT and COL1A1 in OSCC were both restricted to stromal compartment ( Fig.  4C), where the cells were stained negative for Pan-CK (Fig. 4C), a marker of cancer cells [31]. To clarify the role of stromal NNMT in collagen I deposition, a series of experiments was performed upon the FAOs and the co-implanted xenograft model. The results of IF staining showed that knockout of NNMT in CAFs drastically decreased the expression of COL1A1 in FAOs (Fig. 4D, E). More importantly, in co-implanted xenograft with CAFs and sphere cells, the results of Masson trichrome and IHC staining confirmed that the abundance of type I collagen, as well as the frequency of both COL1A1 + and α-SMA + cells, were drastically reduced when NNMT was silenced in CAFs ( Fig. 4F-H). In contrast, in co-implanted xenografts with PFs and sphere cells, overexpression of NNMT in PFs elevated the amount of type I collagen and increased the expression of COL1A1 and α-SMA ( Fig. 4I-K). In summary, these data illustrate that stromal NNMT expression can regulate type I collagen deposition in the OSCC TME.

Collagen deposition is involved in stromal NNMT-aggravated oncogenic behaviours
To explore whether type I collagen in the TME can promote OSCC development, sphere-forming OSCC cells and CAFs transfected with a short-hairpin RNA targeting the expression of Col1a1 (sh-Col1a1) were employed for FAO construction. The results disclosed that in two individual patient-derived FAOs, knockdown of Taken together, these results indicate that the CAF-derived type I collagen may contribute to OSCC development. To authenticate whether the collagen I deposition is involved in the stromal NNMT-mediated oncogenic activities, sphere-forming OSCC cells and PFs transfected with oe-NNMT were integrated to construct FAOs. Then, the FAOs (with PFs-oe-NNMT) were treated with several potential inhibitors of type I collagen synthesis (i.e. losartan, tranilast and halofuginone) at the indicated dosage, which can decrease the expression of COL1A1 in fibroblasts (Supplementary Fig. 5F, G), but show rare toxicity upon single tumour cell-derived organoids (Fig. 5E, F). As a result, the colony diameter of FAOs was significantly reduced upon treatment with inhibitors of collagen ( Fig. 5E, F), indicating that NNMT-mediated collagen I deposition may contribute to CAF-cancer cross-talk. To confirm the role of NNMT-mediated collagen deposition in tumour growth in vivo, sphere-forming OSCC cells and PFs transfected with oe-NNMT were co-implanted into BALB/c nude mice for the drug intervention assay. The results demonstrated that intraperitoneal injection (ip) with losartan slowed the growth dynamics of co-implanted xenografts with PFs-oe-NNMT ( Fig. 5G-I). In addition, the expression of CD44 in xenografts was decreased upon losartan treatment, which was accompanied by a reduction in type I collagen deposition in vivo ( Fig. 5J-M). Overall, these results suggest that stromal NNMT expression can modulate type I collagen deposition to promote OSCC development.
Stromal NNMT expression promotes cancer stemness by sustaining FAK signalling Studies have reported that sustaining cancer stemness, which plays a central role in tumour development [4], is susceptible to the interaction of cancer stem cells (CSCs) with the surrounding ECM [33]. In line with this, our RNA-sequencing analysis of sphereforming OSCC cells (SFCs), which enriched the CSC population [34], confirmed that the expression of genes related to collagenmediated signals, such as focal adhesion kinase (FAK) and ECMreceptor interaction [35,36], was elevated in SFCs (Supplementary Fig. 6A-C). Since FAK is a central effector of ECM-related signal transduction [14,34], to determine whether FAK in cancer cells contribute to the role of ECM in regulating cancer stemness, sphere-derived cells were transfected with a short-hairpin RNA targeting the expression of PTK2 (Fig. 6A), the gene encoding FAK [37]. Then, the transfected OSCC cells were used for several experiments, including a non-anchoring sphere-forming assay, Fig. 2 Stromal NNMT expression is essential for the regeneration of FAOs. A Time tracking images showed that treatment with NNMT inhibitor (NNMT-i) impaired the morphological activation of CAFs colonies, which were growing in matrigel at 3D situation. Scale bars, 100 μm. B Quantification analysis showed that the extensions length of CAFs colonies (quantified colonies, n = 5 in each group) was significantly reduced upon treatment with NNMT-i. ****p < 0.0001. C Immunoblotting assay confirmed that treatment with NNMT-i decreased the expression of fibroblasts activation markers, such as COL1A1 and α-SMA in CAFs. D Representative images showed that, in CAF-OSCC cells cluster-derived FAOs, knockdown of NNMT (sh-NNMT) in CAF attenuated the morphological activation of fibroblasts and reduced the colony size of FAOs. Scale bars, 100 μm. E Quantification analysis verified that, in CAF-OSCC cells cluster-derived FAOs, knockdown of NNMT (sh-NNMT) in CAF reduced the colony diameter of FAOs (quantified colonies, n = 15 in each group). ***p < 0.001. Experimental repeat, n = 3. F Serial passage assay demonstrated that, in CAF-OSCC cells cluster-derived FAOs, knockdown of NNMT (sh-NNMT) in CAF blocked the reconstitution of FAOs. *p < 0.05; ***p < 0.001. G Representative images showed that, in PF-OSCC cells cluster-derived FAOs, overexpression of NNMT (oe-NNMT) in PFs triggered the morphological activation of fibroblasts in FAOs. Scale bars, 100 μm. H Quantification analysis verified that, in PF-OSCC cells cluster-derived FAOs, overexpression of NNMT (oe-NNMT) in PFs increased the colony diameter of FAOs (quantified colonies, n = 15 in each group). ***p < 0.001. Experimental repeat, n = 3. I Serial passage assay demonstrated that, in PF-OSCC cells clusterderived FAOs, overexpression of NNMT (oe-NNMT) in PFs facilitated the reconstitution of FAOs. *p < 0.05; **p < 0.01. J Immunoblotting assay confirmed decreased expression of NNMT and several fibroblasts activation markers, such as COL1A1, α-SMA and FAP in CAFs that were transfected with a single guide RNA toward human NNMT (sgNNMT), which was based on the CRISPR/cas9 knockout system. K Representative IF staining images showed that, CAF-OSCC cells cluster-derived FAOs, knockout of NNMT in CAFs destroyed the spatial distribution pattern of α-SMA in FAOs. Scale bars, 100 μm.   4 Stromal NNMT can modulate the collagen deposition in OSCC-TME. A, B Correlation analysis from GEPIA database showed the transcriptional relevant of NNMT and collagen deposition-related genes, such as COL1A1 and COL3A1, in cohort of patients with head and neck cancer. C IF co-staining assay showed the expression pattern of Pan-CK with NNMT and COL1A1 in serial sectioning images of OSCC specimens. Scale bars, 100 μm. D Representative IF staining images showed that, in CAF-OSCC cells cluster-derived FAOs, knockout of NNMT in CAFs destroyed the spatial distribution pattern of COL1A1 in FAOs. Scale bars, 100 μm. E Quantification analysis showed that, in CAF-OSCC cells cluster-derived FAOs (quantified colonies, n = 5 in each group), knockout of NNMT in CAFs decreased the frequency of cell stained positive for COL1A1. ***p < 0.001. Experimental repeat, n = 3. F-H Representative Masson/IHC staining images and quantification analysis validated that, in CAF-SFC co-implanted xenografts (n = 10 in control group and 6 in knockdown group), knockdown of NNMT in CAF lead to less frequency of type I collagen (G) and reduced expression of COL1A1 (H) in tumour. ***p < 0.001. Scale bars, 100 μm. Experimental repeat, n = 3. I-K Representative Masson/IHC staining images and quantification analysis validated that, in PF-SFC co-implanted xenografts (n = 8 in control group and 10 in overexpression group), overexpression of NNMT in PF increased the abundance of type I collagen (J) and promoted the expression of COL1A1 (K) in xenograft tumour. ***p < 0.001. Scale bars, 100 μm. Experimental repeat, n = 3. and a collagen titrating organoid formation assay. Interestingly, the results showed that knockdown of PTK2 in OSCC cells did not affect their sphere-forming efficiency under non-anchoring culture conditions (Fig. 6B, C). Otherwise, PTK2 knockdown impaired the ability of OSCC cells to form organoids in Matrigel supplemented with collagen I (Fig. 6B, D, Supplementary Fig. 6D-G), indicating that FAK in cancer cells support the role of collagen I in promoting cancer stenmess. Since the CAFs are major provider cells of collagen I [14], to clarify whether FAK in cancer cells affect the CAF-cancer cross-talk, the FAOs regeneration assay were employed for PTK2 knockdown OSCC cells. As a result, the colonies size of FAOs was decreased when PTK2 was silenced in OSCC cells (Fig. 6B, E), demonstrating that FAK in cancer cells, similar to NNMT in fibroblasts, is essential for the CAF-cancer interplay. To further verify the role of FAK signalling in FAO propagation, SFCs were pretreated with defactinib, an inhibitor of FAK activity [38]. The IB results confirmed that the phosphorylation of FAK was significantly reduced upon defactinib treatment at the indicated dosage (Fig. 6F), which showed rare toxicity to single tumour cell-derived organoids (data not shown). Then, pretreated SFCs, as well as either CAFs or PFs-oe-NNMT were clustered to perform FAOs regeneration assay. The results confirmed that  ) showed that, in CAF-OSCC cells cluster-derived FAOs, knockdown of COL1A1 in CAF (CAF-sh-Col1a1) reduced the serial passage efficiency of FAOs colonies. **p < 0.01. Experimental repeat, n = 3. E, F Representative images and quantification analysis showed that, in PF-OSCC cells cluster-derived FAOs, in which the NNMT in PFs was overexpressed, treatment with inhibitors of type I collagen synthesis, such as losartan, tranilast, and halofuginone, can reduced the colony size of FAOs. Of note, similar treatment rarely affects the colony size of simple tumour cell-derived organoid. Scale bars, 100 μm. ***p < 0.001. G-I Representative images and quantification analysis showed that, in PF-SFC co-implanted xenograft, in which the NNMT in PFs was overexpressed, the administration of losartan decreased the size (G), volume (H) and weight (I) of xenograft tumour. **p < 0.01; ***p < 0.001. J-M Representative IF/Masson/IHC staining images and quantification analysis validated that, in PF-SFC co-implanted xenograft, in which the NNMT in PFs was overexpressed, the administration of losartan reduced the CD44 expression (K), type I collagen abundance (L), and COL1A1 expression (M) in xenograft tumour. Scale bars, 100 μm. ***p < 0.001. pretreatment with defactinib decreased the colony diameter of FAOs (Fig. 6G-J), and more importantly, reduced the expression of CD44 in FAOs (Fig. 6K, L). Finally, to validate the role of FAK signalling in tumour growth when stromal NNMT was upregulated, SFCs, as well as PFs transfected with oe-NNMT, were coimplanted into BALB/c nude mice and then administered defactinib (ip). The results showed that treatment with defactinib can relieve the growth dynamics of co-implanted xenografts with PFs-oe-NNMT (Fig. 6M-O). Collectively, these data suggest that stromal NNMT-mediated collagen deposition may promote OSCC development via FAK signalling.

NNMT promotes LOX transcription by reducing histone methylation
To understand the mechanisms by which NNMT regulates collagen deposition in the TME, PCR analysis was performed to assess the expression of collagen biosynthesis-related genes, as described in the literature [39][40][41], in both CAFs and the corresponding PFs. Specifically, the expression of genes encoding not only COL1A1 but also lysyl oxidase (LOX), an enzyme that modulates fibril collagen cross-linking [41], was decreased in CAFs upon NNMT knockdown (Fig. 7A). In line with this finding, overexpression of NNMT in PFs promoted the expression of COL1A1 and LOX (Fig. 7B), implying that the transcription of COL1A1 and LOX is faithfully involved in NNMT-mediated signal transduction. Studies have reported that NNMT can alter gene expression by triggering histone hypomethylation [10]. To verify whether NNMT can regulate histone methylation in fibroblasts, an immunoblotting assay was performed to detect methylation at typical residues, such as H3K4, H3K9 and H3K27 [42]. We found that the expression of H3K27me3 was increased in CAFs upon NNMT knockdown and likewise decreased in PFs in response to NNMT overexpression (Fig. 7C, D). Notably, in PFs overexpressing NNMT, the expression of H3K27me3 was further reduced by treatment with redundant nicotinamide or methionine (Fig. 7E, F), which are both metabolic substrates of NNMT [10]. To verify whether NNMT can promote the transcription of COL1A1 and LOX by suppressing H3K27 methylation, PFs overexpressing NNMT were treated with GSK-J1, an inhibitor of H3K27me3 demethylases, such as JMJD3/KDM6B and UTX/KDM6A [43], for further experiments. The immunoblotting results confirmed that treatment with GSK-J1 increased the methylation level of histones at H3K27 and simultaneously decreased the expression of LOX in PFs overexpressing NNMT (Fig. 7G). More importantly, the results of the ChIP assay confirmed that overexpression of NNMT in PFs reduced the enrichment of H3K27me3 at the promoters of COL1A1 and LOX (Fig. 7H, I), indicating that the transcription of COL1A1 and LOX is epigenetically regulated by NNMT. Consistent with these findings, in FAOs with PFs transfected with oe-NNMT, treatment with GSK-J1 at indicated dosage, which show rare toxicity upon single tumour cell-derived organoids (Supplementary Fig 7A-C), reduced the frequency of fibroblasts in FAOs, and simultaneously impaired their growth dynamic (Fig. 7J-L, Supplementary Fig 7D). In conclusion, these results clearly demonstrate that stromal NNMT can regulate collagen deposition by epigenetically regulating the transcription of LOX, and thereby promoting OSCC development (Fig. 8).

DISCUSSION
In this study, the assembled tumour organoid and co-implanted xenograft models were used to show that stromal NNMTmediated collagen I deposition can promote OSCC stemness by sustaining FAK signalling. Furthermore, we clarified that NNMT suppressed histone methylation at the promoter of LOX, the gene encoding key enzyme that regulate collagen cross-linking. Together, we propose that the NNMT-LOX-FAK axis contributes to the crosstalk between cancer cells and fibroblasts, and that NNMT-centric ECM remodelling is a potential target for patients with OSCC.
The 2-way interaction of cancer cells with the TME, in which fibroblasts account for a large proportion, plays a vital role in solid tumour development [12]. On the one hand, a variety of secretionand/or contact-dependent mechanisms are engaged by cancer cells to trigger phenotypic transition in precursor cells of CAFs [12,13], such as tissue resident fibroblasts adjacent to tumours (i.e. PFs) [13]. On the other hand, CAFs have an enhanced ability to produce oncogenic cytokines, generate more matrisomes, and contract collagenous matrices [12]. However, the symbiotic environment of CAFs and tumour cells is not perfectly simulated in current cancer models [13]. For instance, even the use of subcutaneous co-injection has been widely reported [44][45][46], the microenvironment of subcutaneous tissues is largely different from that of the primary TME. To resolve this issue, orthotopic transplantation more convincingly restores the primary TME [31], although its application may be challenging due to technical difficulties. More recently, studies have implicated the potential of organoid technology in studying CAF-cancer crosstalk [21]. In support of this, Öhlund et al. showed that distinct CAF phenotypes, such as myofibroblast-like and inflammatory CAFs, may be induced by direct or indirect coculture of precursor cells with tumour organoids [24], providing the pioneer evidence to reveal the intratumoral CAF heterogeneity [24]. In line with this, others and our group further exploited assembled organoid systems, such as fibroblast-attached organoids (FAOs), in which the spatial integration of CAFs into the tumour nest was stably modelled in vitro [27,47]. Here, we showed that the expression pattern of CAF-specific markers in vivo, including NNMT, α-SMA, COL1A1, and LOX, was steadily mimicked in FAOs. More interestingly, the expression of CAF-specific markers was essential for the regeneration of assembled organoids, indicating that tumour development relies on not only cancer cell proliferation but also CAF activation. Overall, we provide evidence that the assembled organoids are competent model for studying the cancer-TME interplay.
NNMT is a versatile metabolic regulator that contributes to the homoeostasis of the cellular methylation landscape [5]. Physiologically, NNMT is highly expressed in liver tissue and functions as the catalyst of S-adenosylmethionine (SAM), the major provider of methyl groups, and nicotinamide, a derivative of vitamin B3, production [5]. Several studies have indicated that the expression of NNMT is casually elevated in cancer cells and that targeting NNMT activity may inhibit tumour growth [6][7][8][9][48][49][50]. Recently, a proteomics study of patients with metastasis further clarified that Fig. 6 Stromal NNMT promotes the cancer stenmess via sustaining FAK signal. A Immunoblotting confirmed reduced expression of FAK in sphere-derived OSCC cells that were transfected with a short-hairpin RNA targeting expression of PTK2 (sh-PTK2). B The sphere, organoid, and FAO reconstitution assay were carried out for the OSCC cells that were transfected with sh-PTK2 or control vector (sh-Cont). Scale bars, 100 μm. C Quantification analysis showed that knockdown of PTK2 didn't affect the colonies diameter of tumour spheres (quantified colonies, n = 15 in each group). ns, no significant. Experimental repeat, n = 3. D Quantification analysis showed that knockdown of PTK2 reduced the colonies diameter of tumour organoids (quantified colonies, n = 15 in each group). Experimental repeat, n = 3. ***p < 0.001. E Quantification analysis showed that knockdown of PTK2 in cancer cells reduced the colonies diameter of FAOs (quantified colonies, n = 15 in each group). Experimental repeat, n = 3. ***p < 0.001. F Immunoblotting assay showed the phosphorylation level of FAK at different site, including Tyr397 and Tyr925, in sphere-derived cells upon treatment with Defactinib. G, H Representative images and quantification analysis showed that, in CAF-OSCC cells cluster-derived FAOs, pretreatment of cancer cells with Defactinib reduced the colony diameter of FAOs. Scale bars, 100 μm. ***p < 0.001. I, J Representative images and quantification analysis showed that, in PF-SFC cluster-derived FAOs, in which the NNMT in PFs was overexpressed, pretreatment of cancer cells with Defactinib reduced the colony diameter of FAOs (n = 15 in each group). Experimental repeat, n = 3. Scale bars, 100 μm. ***p < 0.001. K-L Representative images and quantification analysis showed that, in CAF-OSCC cells cluster-derived FAOs, pretreatment of cancer cells with Defactinib decreased the staining intensity of CD44 in FAOs (quantified cells, n = 40 cells in each group). Experimental repeat, n = 3. Scale bars, 100 μm. M-O Representative images and quantification analysis showed that, in PF-SFC coimplanted xenograft, in which the NNMT in PFs was overexpressed, the administration of Defactinib decreased the size (M), volume (N) and weight (O) of xenograft tumour. **p < 0.01; ***p < 0.001.
the upregulation of NNMT is restricted to the stromal rather than the epithelial compartment of metastatic tumours [10]. Given this evidence, the biological significance of NNMT in the TME is context-dependent and deserves further investigation. In the present study, we obtained data from an OSCC tissue array, assembled tumour organoids, and co-implanted xenografts. Unexpectedly, we found that NNMT expression was conservatively restricted to the tumour stroma. In fibroblast-colony formation, FAO regeneration and tumour initiation assays, the maintenance of CAF phenotypes and the oncogenic properties of CAFs were both impaired when NNMT was pharmacologically inhibited or genetically silenced. More importantly, the overexpression of NNMT in PFs, a precursor cell of CAFs, can promote not only the activation of fibroblasts but also the proliferation of OSCC cells, both in vitro and in vivo. Overall, we confirmed that stromal NNMT overexpression contributes to TME evolution and OSCC development.
In many types of solid tumours, the ECM becomes highly dynamic, partly due to deregulated matrisome deposition [14]. Among the matrisome proteins, type I collagen is the most abundant component of cancer-related ECM [14]. Notably, collagen synthesis activity relies on not only the translation of procollagen molecules but also on multistep posttranslational modifications, which are mediated by two molecular chaperones, three collagen hydroxylases, two collagen glycosyltransferases, two specific proteinases, and one specific oxidase [39][40][41]. In particular, the dysregulation of collagen oxidase (i.e. LOX) has been widely reported to be involved in TME alterations and tumour development [51]. For instance, Cox et al. showed that LOX-mediated collagen cross-linking is responsible for fibrosisenhanced metastasis; [17] Haj-Shomaly et al. showed that chemotherapy induces prometastatic pulmonary ECM remodelling by upregulating LOX in T cells, which can be targeted with LOX inhibitors to suppress metastasis; [52] Grasset et al. showed that the inhibition of LOX can relieve the mechanosensitization of EGF-dependent cancer cell collective invasion [18]. In the present study, the correlation between NNMT expression and type I collagen deposition was confirmed in FAOs and co-implanted xenograft models. Moreover, the functional reconstitution of FAOs was impaired when the gene encoding COL1A1 was silenced. In PFs overexpressing NNMT, the pro-tumour effects of fibroblasts were reduced upon treatment with an inhibitor of collagen synthesis. In mechanistic investigations, the transcription of LOX is faithfully responding to NNMT knockdown or overexpression in CAFs or PFs. Most importantly, H3K27 trimethylation at the LOX promoter, which is a typical suppressive histone marker, was reduced by NNMT overexpression. In summary, we clearly demonstrated that LOX-dependent type I collagen deposition is involved in NNMT-aggravated oncogenic behaviours.
CSCs are the key driving force of patients with OSCC [11,26,34]. During the past decade, it has been increasingly recognised that the maintenance of cancer stemness is affected by the interplay of CSCs with the surrounding TME niche [1,4], which is underlaid by not only soluble factor-mediated cell-cell communication [46,47,53], but also ECM receptor-mediated cell-matrix interactions [45,54]. For instance, Condello et al. showed that in ovarian cancer, the scaffold protein tissue transglutaminase (TG2) modulated spheroid proliferation and the tumour-initiating capacity of CSCs by sustaining integrin-mediated Wnt signalling; [54] Ohta et al. found that upon chemotherapy disrupts COL17A1 and breaks the dormancy in colon cancer stem cells through FAK-YAP signal axis [33]. In the present study, our RNA-seq results confirmed that the expression of genes related to the FAK signal, which is a key signal transducer involved in the ECM-receptor interaction [33,35,36], was elevated in sphere-forming OSCC cells. Furthermore, the ability of sphere cells to reconstitute FAOs was impaired when the FAK signal was pharmacologically inhibited or genetically silenced. In both FAOs and xenografts models, the NNMT-aggravated oncogenic properties of fibroblasts were also reduced when the FAK signal was constrained. In conclusion, we confirmed that stromal NNMT can promote cancer stemness by sustaining the collagen-FAK signalling cascade.
Notably, there are several open questions and/or limitations with the present dataset. First, molecular insight into NNMT in distinct subsets of CAFs remains to be further studied. Although the loss-of-function of NNMT in CAF activation was validated in this study, the conclusion was only supported by the analysis of myofibroblast phenotypes, such as α-SMA expression and collagen depositing ability. Since the functional heterogeneity of CAFs has been widely reported [13], it would be valuable to question whether NNMT can regulate other properties of CAFs, such as their ability to trigger inflammatory reactions, or their role in local-regional metastasis [31]. Second, the transcription network in response to NNMT overexpression is still complex and deserved more concentration. For instance, it remained unknown why the expression of several other collagen synthesis-related genes was differentially regulated by NNMT knockdown. Future study based on transcriptomics and epigenomics analyses would likely help resolve this dilemma. Thirdly, the in-depth mechanisms of collagen biology were not thoroughly interrogated at present. Since LOX, the key enzyme that mediate collagen cross-linking, was presumed to be the target of NNMT, it would be interesting to analysis how the alteration of collagen spatial structure, which should be evaluated by multiphoto microscopy rather than simple IHC staining, may affect TME evolution. Last but not the least, the use of subcutaneous co-injection models in this context was not a perfect choice. To more authentically understand the role of NNMT in OSCC TME, the orthotopic transplantation in oral cavity is essential for the ensuing investigations.
However, our findings strongly indicate that collagen deposition in TME is drastically retarded when NNMT in CAFs is inhibited, and the NNMT-LOX-FAK axis may be a signal cascade that orchestrates the crosstalk between fibroblasts and tumour cells. Overall, we believe that future studies based on stromal NNMT would benefit the clinical treatment of patients with OSCC.

MATERIALS AND METHODS Clinical samples
Collection of specimens from patients with OSCC was obtained with written informed consent processed according to IRB-approved guidelines at the Ethics Committee of School and Hospital of Stomatology at Wuhan University (IRB-ID:2021A18). The clinical information of samples involved in tissue microarray (OSCC samples, n = 60; oral mucosa samples, n = 8) and organoid biobank (n = 12) has been deposited in our previous studies [26,55]. Fig. 7 NNMT promotes LOX transcription via receding the histone methylation. A Collagen deposition-related gene expression was determined by quantitative PCR in CAFs, in which the expression of NNMT was knocked down. *p < 0.05; **p < 0.01. B Collagen depositionrelated gene expression was determined by quantitative PCR in PFs, in which the expression of NNMT was overexpressed. *p < 0.05; **p < 0.01. C Immunoblotting assay showed the expression level of H3K27me1, H3K27me2, and H3K27me3 in CAFs that were transfected with sh-NNMT. D Immunoblotting assay showed the expression level of H3K27me1, H3K27me2, and H3K27me3 in PFs with overexpressed NNMT. E, F Immunoblotting assay confirmed that, supplement of metabolic substrate of NNMT, including Nicotinamide and Methionine, can reduce the expression level of H3K27me3 in PFs with overexpressed NNMT. G Immunoblotting assay showed that, treatment with GSK-J1, a potent inhibitor of H3K27me3/me2-demethylases (i.e. JMJD3/KDM6B and UTX/KDM6A), promoted the expression of H3K27me3, and decreased the expression of LOX and COL1A1 in PFs with overexpressed NNMT. H, I Chip assay showed that, the enrichment of H3K27me3 at the promoter of LOX and Col1a1 was reduced upon overexpression of NNMT in PFs. *p < 0.05; **p < 0.01. J-L Time tracking images showed that, in PF-OSCC cells cluster-derived FAOs, in which the NNMT in PFs was overexpressed, treatment with GSK-J1 reduced the frequency of fibroblasts (i.e. GFP + cells) in FAOs (J, K), and decreased the FAOs colonies diameter (J, L). Quantified colonies, n = 12 in each groups. The actual GFP + cells number was calculated by live observing under fluorescence microscope. Experimental repeat, n = 3. Scale bars, 100 μm. ns, no significant. **p < 0.01.

Sphere-forming assay and RNA-seq library construction
Sphere-forming assay was performed as previously described [34]. For RNAseq library construction, total RNA of OSCC cells from patient-derived organoid-2 [26] were isolated by the guanidine thiocyanate method using standard protocols [29]. The raw data has been deposited to SRA under accession number PRJNA800217. To enrich the CSCs population, the mature spheres (with diameter >50 μm) were trypsinized into single cells for serial passage, at least 3 cycles, which were then engaged for drugs pretreatment assays [34]. For defactinib pretreatment assays, the sphere-derived cells were seeded in 6 well plates (50,000 cells/well) and incubated with stem cells medium that supplemented with defactinib at 20 μM for 24 h.

Fibroblast-colony formation assay
Briefly, 10 5 CAFs were suspended in ultra low attachment plates for 24-48 h to generate fibroblasts clusters, which were embedded in matrigel to allow colony formation. Details were as previously described [27].

Construction and regeneration of FAOs
The organoid lines 2, 6, and 9, as well as the corresponding CAFs/PFs [26], were employed as the source of cancer cells and fibroblasts to construct FAOs. Briefly, the CAF-OSCC cells clusters were generated according to previously described protocol [27]. Since day 2 after CAF-OSCC cells clusters were embedded in matrigel (Corning#356237), the morphology and colony size of clusters were recorded under an inverted microscope (Leica Microsystems). Specifically, the cells with elongated spindle-like branch were defined as activated CAFs [30], the colonised buds with length > 50 μm were defined as invasive frontier structure. To determine colonies efficiency, the clusters with maximum diameter > 200 μm, or more than 5 invasive frontier structures attached by activated CAFs were counted under an inverted microscope (Leica Microsystems), and the forming efficiency (%) = scored FAOs number/total plating CAF-OSCC cell clusters. In all the drugs treatment assays upon FAOs model, which contained more than one cellular components, the selection of treatment dosage was based on the following consideration: (i) Treatment with drugs can reduce the expression of hypothesised proteins (e.g. COL1A1) in fibroblasts; (ii) Treatment with drugs showed rare toxic effects upon tumour cell-derived colonies formation.

Animal experiments
For animal experiments, female, 4-to 6-week-old BALB/C nude mice were purchased from Beijing HFK Bioscience Co., Ltd. (Beijing, China), and were maintained according to protocols approved by the Ethical Committee on Animal Experiments of the Animal Care Committee of Wuhan University (S07921070J). The mice were randomly divied into different groups, and no animals were excluded in this study. For co-implanted assay (n = 20 in each group), the sphere-forming cells and fibroblasts were resuspended in a PBS/Matrigel (BD Biosciences) mixture (1:1 volume), which were then injected in the into the subcutaneous tissue of mouse flanks using 27gauge needles. For drug administration assay (n = 5 in each group), the Losartan and Defactinib, all purchased from MedChemExpress (MCE, USA), were injected intraperitoneally into the BALB/C nude mice bearing with coimplanted xenograft. Tumour sizes were blindly measured every 2 days and calculated using the formula (width2 × length)/2.

Transfection materials
The human NNMT-CRISPR plasmids were purchased from Wuhan bioeagle Co.,Ltd. (China). The lentivirus containing overexpressed NNMT, or the short-hairpin RNA targeting the expression of NNMT or COL1A1, were purchased from Shanghai Genechem Co.,Ltd. (China). Transfection of cells were performed according to the manufacturer's instruction.

RNA isolation and qPCR
Total RNA from CAFs and PFs was isolated using RNAiso Plus (Takara, Tokyo, Japan) according to the manufacturer's instruction. Details were as previously described [56]. The primer sequence of related genes was clarified in Supplementary Table 1.

Chromatin immunoprecipitation (ChIP) assay
For PFs with overexpressed NNMT expression, the ChIP assay was conducted according to the manufacturer's instruction of the ChIP kit (Thermo Fisher Scientific, Rockford, USA). Details were as previously described [56]. Fig. 8 Schematic of the proposed CAF-cancer cross-talk mechanism. Tumour-stroma interplay hijacks NNMT as a master regulator to trigger LOX-mediated collagen I deposition, which then enhance the FAK signal in cancer cells, and thereby promoting OSCC development.