Signaling networks in cancer stromal senescent cells establish malignant microenvironment

Significance

Cancer stromal cells play a pivotal role in the development of cancer. However, these cells exhibit significant heterogeneity in their regulatory functions of tumorigenesis, and the precise mechanisms underlying these processes remain to be elucidated. Our study identifies a pivotal function of signaling networks among p16^high cancer stromal senescent cells in establishing a tumor-promoting microenvironment in pancreatic ductal adenocarcinoma (PDAC) allograft models. These findings suggest that targeting p16^high cancer stromal cells could be a strategy for PDAC suppression.

Abstract

The tumor microenvironment (TME) encompasses various cell types, blood and lymphatic vessels, and noncellular constituents like extracellular matrix (ECM) and cytokines. These intricate interactions between cellular and noncellular components contribute to the development of a malignant TME, such as immunosuppressive, desmoplastic, angiogenic conditions, and the formation of a niche for cancer stem cells, but there is limited understanding of the specific subtypes of stromal cells involved in this process. Here, we utilized p16-Cre^ERT2-tdTomato mouse models to investigate the signaling networks established by senescent cancer stromal cells, contributing to the development of a malignant TME. In pancreatic ductal adenocarcinoma (PDAC) allograft models, these senescent cells were found to promote cancer fibrosis, enhance angiogenesis, and suppress cancer immune surveillance. Notably, the selective elimination of senescent cancer stromal cells improves the malignant TME, subsequently reducing tumor progression in PDAC. This highlights the antitumor efficacy of senolytic treatment alone and its synergistic effect when combined with conventional chemotherapy. Taken together, our findings suggest that the signaling crosstalk among senescent cancer stromal cells plays a key role in the progression of PDAC and may be a promising therapeutic target.

Cellular senescence is a phenomenon characterized by irreversible proliferative arrest in response to excessive cellular stress. In addition to cell cycle arrest, various other hallmarks of cellular senescence have been extensively reviewed (1), including resistance to apoptosis, epigenetic alterations, and senescence-associated secretory phenotypes (SASPs).

Cellular senescence is a double-edged sword, exerting opposing effects in tumorigenesis. This phenomenon has generally been regarded as a tumor-suppressive process by preventing the proliferation of cells carrying transforming mutations (2). However, the accumulation of senescent cells during natural aging leads to chronic inflammation, emerging as a risk factor for overall tumor incidence (3). Furthermore, chemotherapy, radiotherapy, or other cell cycle inhibitors have been shown to induce cellular senescence in cancer cells (4). These intratumoral senescent cells may recruit immune cells through the secretion of proinflammatory factors, thereby enhancing blood vessel permeability and immune surveillance against cancer (5). On the other hand, the cytokines secreted by senescent cells promote angiogenesis, metastasis, and extracellular matrix (ECM) remodeling (6). While stromal cells lacking transforming mutations are prone to senescence induction, the characteristics and identification of senescent stromal cells are not as well-understood as those of senescent cancer cells.

Pancreatic ductal adenocarcinoma (PDAC) is one of the worst prognostic solid tumors due to its nonspecific symptoms until advanced stages. Both conventional cytotoxic chemotherapy and immunotherapeutic strategies such as immune checkpoint inhibition have limited responses (7, 8). In the context of PDAC, the stromal barrier within the tumor microenvironment (TME) emerges as a pivotal factor conferring treatment resistance. The composition of the TME varies across tumor types, commonly featuring immune cells, stromal cells, and blood vessels. The most notable histological feature in the TME of PDAC is the desmoplastic reaction, which indicates that fibroblasts are activated by cancer cells and produce the extensive fibrotic region surrounding the tumor (9). The resultant desmoplasia creates a mechanical barrier rich in ECM components such as collagen, elastin, and hyaluronic acid. This barrier contributes to hypoperfusion and hypertension, restricting vascularization, drug delivery efficiency, and immune cell infiltration (10).

Low immune activity has also been implicated in PDAC. Antitumor immunity primarily relies on dendritic cells, inflammatory macrophages, and helper T cells. However, the activation of myeloid-derived suppressor cells (MDSCs), anti-inflammatory tumor-associated macrophages (TAM), and regulatory T cells (Treg) create an immunosuppressive TME (11). Very recently, the tumor-supporting function of senescent macrophages in lung cancer has been described (12, 13). These findings highlight the importance of the TME managed by stromal senescent cells in cancer progression. However, the role of the other senescent cell types, such as fibroblasts and endothelial cells, and the crosstalk between senescent stromal cells remain largely unknown. Furthermore, simultaneous targeting of both the TME and cancer cells has recently been described as having beneficial therapeutic potential over the conventional approaches (14).

To gain a comprehensive understanding of the interaction between cancer and stromal senescent cells, this study utilized p16-Tom (p16-Cre^ERT2/Rosa26-LSL-tdTomato) mice (15) and conducted allograft transplantation of primary PDAC cells (16). This approach allowed for a thorough investigation into the biological characteristics of senescent stromal cells and their crosstalk within the TME during tumorigenesis and tumor progression.

Results

Identification and Characterization of p16^high Stromal Cells in an Allograft PDAC Model.

Cellular senescence plays contradictory roles in the field of oncology. There is still a wide divergence of opinions as to whether cellular senescence promotes or inhibits malignancy. These two perspectives mainly focus on how senescent somatic cells influence tumorigenesis and how therapy-induced senescent cancer cells remodel drug resistance and metastasis (17, 18). TME is composed of cancer cells and noncancer cells, and these cells, along with the cytokines they secrete, impact the cancer progression and the therapeutic effectiveness, such as immune checkpoint blockade treatments (19). However, the role of senescent stromal cells and their interconnectivity in the TME is still largely unknown. To address this, we used the p16-Tom mouse model as a recipient to quantify and visualize the senescent noncancer cells in the PDAC allograft model (Fig. 1A). The p16^high noncancer cells were labeled with tdTomato expression by tamoxifen administration, whereas the donor cancer cells were not labeled. Interestingly, we found that the spatial distribution of tdTomato-positive (Tom⁺) cells was predominantly located in the central region of the tumors (Fig. 1B), and this pattern was positively correlated with the overall abundance of Tom⁺ cells. Hypoxia in the tumor core has been shown to contribute to selection pressure for cancer stemness and aggressiveness (20). To understand whether these p16^high noncancer cells in the tumor have a positive or negative impact on the cancer progression, we performed scRNA-seq to analyze the transcriptome to infer the roles and cell types of Tom⁺ noncancer cells in the tumor after excluding circulating cells by perfusion (SI Appendix, Fig. S1). We first performed unsupervised clustering to divide these stromal cells into 8 clusters (SI Appendix, Fig. S2A). Using SingleR in combination with marker genes, we were able to classify these clusters into three cell types: fibroblasts, macrophages, and endothelial cells (Fig. 1 C and D and SI Appendix, Fig. S2 B and C). Since some cells identified as dendritic cells and monocytes within the cluster were classified as macrophages, we used FACS analysis to further validate that among CD45⁺/CD11b⁺ cells, Tom⁺ cells are predominantly macrophages (SI Appendix, Fig. S2 D and E). A higher level of tdTomato was confirmed in Tom⁺ cells compared to Tom⁻ cells (Fig. 1E). The composition of Tom⁺ cells was 43.4% fibroblasts, 17.3% endothelial cells, and 38.1% macrophages (SI Appendix, Fig. S3A). We further validated the existence of these three cell types in the tumor through section analysis, using immunofluorescence staining followed by confocal microscopy (Fig. 1F). According to the FACS analysis, 2.3, 0.9, and 2.2% of the cells in fibroblasts, endothelial cells, and macrophages, respectively, were Tom⁺ (SI Appendix, Fig. S3B). In addition to subcutaneous transplantation, we also orthotopically transplanted cancer cells into the pancreas (21) and labeled p16^high stromal cells according to the same schedule. We also detected the presence of these three types of p16^high cells within the tumor by immunostaining (SI Appendix, Fig. S3 C and D).

Fig. 1.

p16^high cancer stromal cells consist of TECs, CAFs, and TAMs. (A) The experimental schedule of PDAC allograft model and tamoxifen administration in p16-Tom mice. (B) The representative fluorescence image of tumors from the mouse treated as in (A). The quantification and distribution of Tom⁺ cells were shown on the Right panel. Each dot represents one tumor sample, and the dot lines indicate the 95% CI. The Z-score explanation scheme is shown in the Upper Right panel. One-sample t test and simple linear regression were performed for Z-score and correlation, respectively (Scale bar, 1 mm.) (n = 11). (C) UMAP visualization of single-cell transcriptomes of tumor stromal cells from the mouse model as in (A). The dots were categorized by cell types and split by libraries derived from Tom⁺ or Tom⁻ cells. (D) The violin plots showing the expression levels of cell type marker genes. Expression levels represent normalized UMI counts. (E) The violin plots showing the expression levels of tdTomato in the indicated cells. (F) Confocal immunofluorescence images of tumors as in (A) using the indicated antibodies. (Scale bar, 100 µm.)

scRNA-seq Reveals Tumor-Promoting Features in p16^high Cells.

Tumor endothelial cells (TECs), cancer-associated fibroblasts (CAFs), and TAMs play a critical role in the TME by supporting the tumor progression of PDAC (14, 19, 22). They regulate various aspects of the tumor, including immune suppression, angiogenesis, ECM remodeling, and growth factor production. We performed the transcriptome analysis to explore the connection between cellular senescence in stromal cells, their associated properties, and the functional impact of p16^high stromal cells on the TME.

In the p16^high TEC population, we identified 104 up-regulated and 80 down-regulated differentially expressed genes (DEGs), including Stat3, Egr1, Nrp2, Plxna4, Selp, Spp1, Sparc, Ubb, and Uba52 (Fig. 2A and SI Appendix, Fig. S4 A and B). These genes have been reported previously, suggesting that p16^high endothelial cells might be involved in angiogenesis, metastasis, reduction of vascular permeability, and impaired protein quality control (23–31). In addition, we collected Tom⁺ and Tom⁻ TECs for qPCR analysis and found that in addition to a significant increase in p16, markers of senescent cells such as Glb1 (32), PAI-1 (33), and the SASP marker IL6 were also significantly elevated (SI Appendix, Fig. S4C). This indicates a significant enrichment of senescent TECs in the Tom⁺ TEC population. Gene ontology (GO) and gene set enrichment analysis (GSEA) revealed up-regulation of Th17 activation, SHH, and TNF-α signaling pathways (Fig. 2B and SI Appendix, Fig. S5A). On the other hand, antigen presentation, immune cell chemotaxis, E2F, mTORC, and OXPHOS were enriched in p16^low TECs (Fig. 2B and SI Appendix, Fig. S5B). These transcriptomic features show that angiogenesis, immunosuppression, and cellular senescence may be mainly contributed by p16^high TECs (34–40). Consistently, the analysis of regulatory network inference showed that angiogenesis-related transcription factors, including Egr1, Foxo1, and Stat3 (41), were potentially activated in p16^high but not in p16^low TECs (SI Appendix, Fig. S6A). Activated NF-κB signaling and downregulation of cell-junction proteins indicated the hyperpermeability of the vascular structure, which was correlated with metastasis and intravasation processes (42). Furthermore, the binding motifs of Runx1 and Runx2 were significantly enriched in the promoter regions of up-regulated DEGs in p16^high endothelial cells (SI Appendix, Fig. S6B), suggesting that TNF-α stimulation or AP-1 signaling may play an important role in promoting angiogenesis (43).

Fig. 2.

The scRNA-seq analysis reveals the tumor-supportive function of p16^high stromal cells. (A) The violin plots showing the expression levels of indicated DEGs in Tom⁺ and Tom⁻ TECs. (B) The GSEA plots showing the enriched terms between Tom⁺ and Tom⁻ TECs. (C) The violin plots showing the expression levels of indicated DEGs in Tom⁺ and Tom⁻ CAFs. (D) The GSEA plots showing the enriched terms between Tom⁺ and Tom⁻ CAFs. (E) The violin plots showing the expression levels of indicated DEGs in Tom⁺ and Tom⁻ TAMs. (F) The GSEA plots showing the enriched terms between Tom⁺ and Tom⁻ TAMs. DEGs were identified by FDR < 0.05 and Log2FC > 0.1. All GSEA terms were identified by adjusted P-value < 0.05 using the B–H method. Red and blue colors represent terms which were enriched in up-regulated DEGs and down-regulated DEGs, respectively.

In PDAC, hyperactivation of CAFs is the dominant origin and cause of the desmoplastic reaction (44). In the CAF population, 279 up-regulated and 443 down-regulated DEGs were identified (SI Appendix, Fig. S7A). Among them, we observed higher expression levels of several collagens in p16^high CAFs (Fig. 2C and SI Appendix, Fig. S7B), suggesting that p16^high CAFs may contribute more significantly to tumor fibrosis (44). Interestingly, we also noted that several cytokines (Thbs4 and Smoc2) involved in angiogenesis were upregulated in p16^high CAFs (Fig. 2C and SI Appendix, Fig. S7B) (45, 46), which have been described to be downstream of the TGF-β pathway (47, 48). Furthermore, the higher p16 protein levels in Tom⁺ CAFs than in Tom⁻ CAFs were also confirmed (SI Appendix, Fig. S7C). GO and GSEA revealed upregulation of ECM remodeling, VEGF signaling, TGF-β response, and oxidative stress in p16^high CAFs (Fig. 2D and SI Appendix, Fig. S8A). Surprisingly, p16^low CAFs showed relatively high expression of proinflammatory cytokines (Cxcl1, Cxcl2, and Il6) (SI Appendix, Fig. S7B), which might be caused by the colocalization between the nonfibrogenic region and the inflammatory zone (49). Regulatory network analysis revealed that Foxo3 is more activated in p16^high CAFs, whereas AP-1 is inhibited (SI Appendix, Fig. S9A). This was consistent with the results in GSEA, where E2F targets were significantly enriched in p16^low CAFs. Taken together, these results suggest that p16^high CAFs were in a state of cell cycle arrest (50). Importantly, the binding motif enrichment analysis revealed that numerous collagen-related genes were positively regulated by Creb3l1 (SI Appendix, Fig. S9B).

The polarization of macrophages toward either the M1 or M2 phenotype is crucial in determining the activity of the immune system in the TME (51). In our scRNA-seq dataset, transcriptome comparison revealed 203 up-regulated and 381 down-regulated DEGs in p16^high TAMs (SI Appendix, Fig. S10A). The expression levels of M2 polarization-related genes, including Arg1, Arg2, CD36, Il4ra, and Il10rb, were higher in p16^high TAMs (Fig. 2E and SI Appendix, Fig. S10B) (52–54). In contrast, genes associated with inflammation, such as Il1a and Nfkb1, as well as the antigen presentation gene H2-Ab1, which is mainly expressed by M1 macrophages, showed higher expression in p16^low cells (Fig. 2E and SI Appendix, Fig. S10B) (55). These results suggested that p16 expression was partially correlated with the anti-inflammatory properties of TAMs. Similarly, GO and GSEA also suggested that p16^high TAMs play a role in wound healing and promoting angiogenesis, which were previously described as functions of M2 macrophages (Fig. 2F and SI Appendix, Fig. S11A) (56). Notably, the hypoxia response and OXPHOS were enriched in p16^high and p16^low TAMs, respectively, clearly indicating that spatial distribution and metabolic homeostasis are distinct in these two subtypes of TAMs (Fig. 2F and SI Appendix, Fig. S11B). Consistently, the regulatory network analysis showed that the M2 transcription factors, Cebp and Maf, were activated, whereas the M1 transcription factor, Nfkb1, was inactivated in p16^high TAMs (SI Appendix, Fig. S12 A and B) (57, 58).

We conducted the cell–cell communication inference to investigate the signal transduction network between each stromal cell type in TME (SI Appendix, Fig. S13). In the TECs, p16^high cells expressed more Agrn and Hbegf compared to p16^low cells, which was consistent with the angiogenesis and fibrogenic features of p16^high stromal cells (59–61). On the other hand, Apln expressed by p16^low TECs was reported to inhibit tumor growth through recruiting immune cells (62). Furthermore, Il1b and Il6 derived from p16^low CAFs were involved in the induction of inflammation (63, 64). Interestingly, although Tgfb1 secreted from TAMs was not altered by p16 expression, the Tgfb3 expression was higher in p16^high CAFs than in p16^low CAFs. This result implied that p16^high CAFs were more sensitive to the stimulation and participated in the tumor fibrosis (65, 66).

Elimination of p16^high Cancer Stromal Cells Suppresses Tumor Progression.

To evaluate the potential therapeutic value of targeting senescent cells within tumors, we first tested the efficacy of the senolytic agent ABT263 in combination with the chemotherapeutic agents Gemcitabine plus Abraxane (GA) (SI Appendix, Fig. S14A) (16). Both GA and ABT263 treatments significantly inhibited tumor growth, and their combination showed a greater improvement (SI Appendix, Fig. S14B). In this design, ABT263 targeted both senescent noncancer stromal cells and senescent cancer cells potentially induced by chemotherapy (5, 67). To further validate the conclusion from our scRNA-seq analysis that p16^high stromal cells support tumor growth, we used p16-Cre^ERT2-DTR (p16-Cre^ERT2/Rosa26-LSL-DTR) mice to specifically examine the effect of removing p16^high stromal cells (Fig. 3C). As expected, the elimination of p16^high stromal cells by diphtheria toxin (DT) significantly reduced p16 expression throughout the tumor and suppressed PDAC tumor growth compared to littermate control (p16-Cre^ERT2) (Fig. 3B and SI Appendix, Fig. S14C).

Fig. 3.

Elimination of p16^high stromal cells reduces tumor growth. (A) The experimental schedule of the PDAC allograft model treated with tamoxifen and diphtheria toxin (DT) in p16-Cre^ERT2 and p16-Cre^ERT2-DTR mice. (B) The growth curves of the tumors described in (A) (n = 16). (C) The representative images (Left panels) and quantification results (Right panels) of Ki67 IHC staining in the tumors from the indicated groups described in (A) (Scale bar, 100 µm.) (n = 8). (D) The representative images (Upper panels) and quantification results (Lower panels) of CD31 staining in the center or periphery regions of the tumors from the indicated groups described in (A) (Scale bar, 500 µm.) (n = 8 for p16-Cre^ERT2 and n = 7 for p16-Cre^ERT2-DTR). (E) The representative image of Sirius red staining. The dotted line indicates the central fibrosis region of the tumor derived from p16-Cre^ERT2 mice. (Scale bar, 1 mm.) (F) The representative images (Left panels) and quantification results (Right panel) of Sirius red staining in the central or peripheral regions of the tumors from the indicated groups described in (A) (Scale bar, 100 µm.) (n = 8). Data are presented as mean ± SEM. Two-way ANOVA with Sidak’s test (A, D, and F) and unpaired t test (C) were performed.

To elucidate the molecular basis of tumor suppression resulting from the removal of p16^high stromal cells, we initially examined the population of cancer stem cells (CSCs) because TME largely contributes to establishing the CSC niche. Elimination of p16^high stromal cells did not affect the population of CSCs, suggesting that p16^high stromal cells did not affect stem cell niche in PDAC (SI Appendix, Fig. S14D). In addition, Ki67 and cleaved-CASP3 staining results suggested that proliferation was slightly but significantly suppressed in the p16-Cre^ERT2-DTR group, while apoptosis was unaffected (Fig. 3C and SI Appendix, Fig. S14E). We then investigated vascularization and observed a significant reduction in the total area of CD31⁺ cells, both in the periphery and at the center, following the removal of p16^high stromal cells (Fig. 3D). We also assessed tumor fibrosis, which could serve as a barrier to host immune cells. Intriguingly, we found that tumor fibrosis was detected in the center but not in the periphery, aligning with the predominant localization of p16^high stromal cells in the central region (Fig. 3E). After the elimination of p16^high stromal cells, the fibrosis in the tumor center was dramatically improved (Fig. 3F). Since subcutaneous fibroblasts and pancreatic stellate cells may have different characteristics, we repeated this experimental design using orthotopic transplantation (SI Appendix, Fig. S15A). The results showed that both tumor weight and the degree of fibrosis were consistently improved after the removal of p16^high stromal cells (SI Appendix, Fig. S15 B and C). Taken together, these results imply that p16^high stromal cells contribute to the creation of a cancer-prone microenvironment, influencing aspects such as angiogenesis and fibrosis.

p16^high Stromal Cells Repress Anticancer Immunity.

Transcriptomic analysis of p16^high TAMs revealed the anti-inflammatory phenotypes, suggesting that their elimination could reactivate immune cell infiltration. However, single-cell analysis could not clearly distinguish between macrophages and monocytic myeloid-derived suppressor cells (M-MDSCs). Thus, we quantified Tom⁺ cells in each subpopulations of CD11b⁺ cells including macrophages, M-MDSCs, and inflammatory monocytes by FACS analysis using Ly6C and F4/80 (SI Appendix, Fig. S16A). The results showed that Tom⁺ cells were significantly more abundant in macrophages as compared to M-MDSCs and inflammatory monocytes (Fig. 4A). The population of infiltrating CD8⁺ T cells among CD45⁺ cells was significantly increased in the tumor from p16-Cre^ERT2-DTR mice (Fig. 4B and S16B), which was consistent with the more abundant Gzmb⁺ cells observed in the p16-Cre^ERT2-DTR group (Fig. 4C). Indeed, we found that the majority of Tom⁺ TAMs were ARG1⁺ representing M2-like phenotypes, whereas only a small proportion of Tom⁻ TAMs were ARG1⁺ (Fig. 4D). To further validate whether Tom⁻ and Tom⁺ macrophages differ in their ability to repress T cell surveillance, we cotransplanted cancer cells and Tom⁺ or Tom⁻ TAMs sorted from the tumors of p16-Tom mice into p16-Cre^ERT2-DTR mice (Fig. 4E). After DT treatment, we found that the number of infiltrating T cells was significantly suppressed by cotransplanted Tom⁺ macrophages (Fig. 4F). Furthermore, we cocultured these two types of TAMs with spleen CD8⁺ T cells and showed that Tom⁻ macrophages activated CD8⁺ T cells (as indicated by CD69⁺/CD8⁺ and CD44⁺CD69⁺/CD8⁺), whereas Tom⁺ macrophages did not exhibit this ability (Fig. 4G and SI Appendix, Fig. S16C). Although DEGs between Tom⁺ and Tom⁻ TAMs did not include TGF-β, a representative cytokine secreted by M2 macrophages, immunoblotting of tumor lysates revealed the lower TGF-β signaling activity in p16-Cre^ERT2-DTR than in p16-Cre^ERT2 mice, as assessed by the phosphorylation of SMAD2/3 (Fig. 4H and SI Appendix, Fig. S14C). Finally, to fully comprehend the alteration in cytokine expression, we performed cytokine array analysis using tumor lysates from p16-Cre^ERT2-DTR and p16-Cre^ERT2 mice. The expressions of CCL2 and ICAM-1 were diminished following the removal of p16^high stromal cells (Fig. 4I). CCL2 is reported to activate MDSC and suppress anticancer immunity (68). It is noteworthy that CCL2 levels are correlated with poor prognosis in pancreatic cancer patients (69–71).

Fig. 4.

p16^high TAMs are involved in immune suppression in the TME. (A) The percentages of Tom⁺ cells among macrophages, M-MDSC, and inflammatory monocytes quantified by FACS (n = 3). (B) The FACS analysis of the CD8⁺ population in CD45⁺ tumor-infiltrating immune cells from the tumor described in Fig. 3A (n = 10). (C) The representative images (Left panels) and quantification results (Right panels) of Gzmb IF staining in the tumors from the indicated groups described in Fig. 3A (Scale bar, 100 µm.) (n = 8). (D) The representative images (Left panels) and quantification results (Right panels) of ARG1 staining on the TAMs sorted from p16-Tom mice (Scale bar, 100 µm.) (n = 3). (E) The experimental schedule of cotransplanting Tom⁺ and Tom⁻ TAMs with cancer cells in p16-Cre^ERT2-DTR mice. (F) The representative images (Left panels) and quantification results (Right panels) of CD3 IHC staining in the tumors from the indicated groups described in (E) (Scale bar, 100 µm.) (n = 5). (G) The FACS analysis of the activation of T cells cocultured with Tom⁺ and Tom⁻ TAMs. The splenic CD8⁺ T cells were cocultured with sorted Tom⁺ or Tom⁻ TAMs for 18 h and then the staining for CD44 and CD69 was performed to estimate the T cell activation (n = 3, 6, 4 in each group). (H) Quantification results of intensity ratio between p-SMAD2/3 and SMAD2/3 in immunoblotting images shown in SI Appendix, Fig. S14C (n = 8). (I) Quantification of the cytokine array on total lysates from the tumor described in Fig. 3A. Data are presented as mean ± SEM. Unpaired t test (B, C, D, and H), paired t test (F), one-way ANOVA followed by Tukey HSD test (A and G), and two-way ANOVA followed by FDR adjustment (I) were performed.

Tumor Fibrosis Is Regulated by Macrophage-Derived TGF-β- p16^high Fibroblast CREB3L1 Axis.

Since the scRNA-seq analysis of p16^high stromal fibroblasts revealed their implications in tumor fibrosis and increased response in the TGF-β pathway, we speculate a potential association between these two processes. To address this, we first investigated the effect of TGF-β treatment on the expression of genes involved in angiogenesis and fibrosis. The treatment of TGF-β1 in primary mouse pulmonary fibroblasts resulted in the upregulation of Smoc2, Col1a2, Col3a1, and Col5a1 as well as p16 (Fig. 5A). Transcription factor binding motif analysis of upregulated DEGs in p16^high CAFs identified CREB3L1 as a key transcription factor for collagen expression (SI Appendix, Fig. S9B). We therefore investigated the activation status of CREB3L1 in senescent fibroblasts and its relationship with the expression of collagen genes. The level of cleaved CREB3L1 (an active form) was increased in senescent cells compared to uninduced cells, and this activation was further stimulated by TGF-β1 treatment (Fig. 5B) (72). The cleavage of CREB3L1 could be inhibited by the treatment of S1P/S2P inhibitor, PF429242. Simultaneously, the expression of Col1a1 was promoted by TGF-β1 treatment, and this effect was stronger in senescent cells (Fig. 5C). Importantly, the TGF-β1-induced upregulation of Col1a1 in senescent cells was rescued by PF429242, whereas it had no effect in uninduced cells. On the other hand, phosphorylated SMAD2/3 [known downstream of TGF-β and capable of activating collagen gene expression (73)] increased with TGF-β1 treatment but remained unaffected by PF429242 (Fig. 5B). These data suggest that in senescent cells, CREB3L1, compared to SMAD2/3, plays a more pivotal role in transmitting the TGF-β1 signal to collagen gene expression. Overexpression of an active form of CREB3L1 (amino acid 1–375) induced the expression of Col1a1, Col1a2, Col3a1, Col5a1, and Col12a1, all of which are known to be involved in tissue fibrosis (Fig. 5 D–F) (74). Taken together, these results suggest that TGF-β signaling activates CREB3L1 to induce fibrotic collagen expression specifically in p16^high stromal fibroblasts.

Fig. 5.

CREB3L1 mediates the TGF-β stimulation to promote collagen expression in senescent fibroblasts. (A) The relative mRNA levels determined by qPCR of the indicated genes in uninduced pulmonary fibroblasts treated with mock or 5 ng/mL TGF-β1 for 48 h (n = 3). (B) Immunoblotting images using the indicated antibodies on cell lysates with the indicated treatments. The TGF-β1 and PF429242 were treated in cells with 5 ng/mL and 2 µM, respectively, for 48 h. (C) The relative mRNA levels of Col1a1 determined by qPCR in the pulmonary fibroblasts with indicated treatments (n = 3). (D) Immunoblotting images using the indicated antibodies on cell lysates with or without overexpression of 3xFlag-CREB3L1 (amino acids 1–375). (E) Schematic representation of the domains and cleavage sites of CREB3L1. (F) The relative mRNA levels determined by qPCR of the indicated genes in uninduced pulmonary fibroblasts with or without overexpression of 3xFlag-CREB3L1 (amino acids 1–375) (n = 3). (G) The schematic image showing the signaling network within p16^high cancer stromal cells. The brown dashed arrows in the figure represent pathways that were not demonstrated in this study but have been mentioned in previous research. Data are presented as mean ± SEM. Unpaired t test (A and F) and two-way ANOVA followed by Sidak’s test (C) were performed.

Discussion

Cancer is believed to be driven by a complex ecosystem consisting of a variety of noncancer cells, including immune cells, fibroblasts, and endothelial cells, and their myriad interactions with cancer cells and among themselves. Initially, these noncancer cells and their environment were considered bystanders to malignant transformation. Still, later, these cells and their interactions were thought to provide cues to understanding the molecular basis of cancer progression. However, very little is known about the full picture of cancer stromal cells such as their crosstalk and the subtype specificities in malignant transformation. The p16^INK4a is an inhibitor of CDK4/6 and plays a critical role in maintaining permanent cell cycle arrest during cellular senescence. In the early stages of pancreatic neoplasia, the senescence mechanism contributed to the suppression of tumorigenesis. However, since p16 mutation is frequently observed in PDAC, it is possible that cancer cells use this mechanism to escape from cellular senescence and progress to PDAC development. Recently, cellular senescence in stromal cells has also been identified in cancer tissues, with senescent TAMs functioning as suppressors of cancer immunity (12, 13). In the present study, we successfully uncovered the complete picture of signaling networks between p16^high senescent cancer stromal cells in a PDAC allograft model.

ScRNA-seq analysis of p16^high TECs, CAFs, and TAMs revealed the following networks. 1) TGF-β activates CREB3L1 to increase collagen expression in p16^high CAFs and subsequently induces cancer fibrosis. 2) TGF-β induces Smoc2 expression in p16^high CAFs and potentially enhances VEGF response in p16^high TECs (45). 3) p16^high stromal cells contribute to the promotion of angiogenesis. This observation may be caused by a direct elimination of p16^high TECs, which exhibit angiogenesis transcriptome signatures, or by an indirect effect of reducing p16^high CAFs, which express Smoc2 and promote angiogenesis. 4) p16^high TAMs show stronger M2-like phenotypes than p16^low TAMs to suppress cancer immunity (Fig. 5G). These signaling networks collaboratively establish a malignant microenvironment that facilitates the progression and metastasis of cancer cells. Importantly, p16^high TECs, CAFs, and TAMs displayed a substantial decrease in the expression of E2F target genes compared to p16^low cells, suggesting that they are in a senescent state. Intriguingly, we detected a much higher population of p16^high cancer stromal cells in the tumor center than in the periphery. Therefore, hypoxic conditions in cancer tissues may induce or regulate p16^high cancer stromal cells. Indeed, transcriptomic analyses of p16^high TAMs, TECs, and CAFs revealed the upregulation of hypoxia-related genes and downregulation of oxidative phosphorylation-related genes.

Elimination of p16^high cancer stromal cells showed drastic changes in the cancer microenvironment such as 1) improvement of cancer fibrosis in the center, but not in the periphery of the cancer. 2) Increased in the population of CD8⁺ cells in cancer. 3) Suppression of angiogenesis in cancer tissues. The use of ABT263 to target total senescent cells effectively inhibits tumor growth; however, it remains unclear whether this effect is due to the elimination of senescent cancer cells, senescent noncancer cells, or a direct cytotoxic effect on cancer cells (75). In this study, we demonstrated that selective elimination of p16^high stromal cells also significantly inhibited tumor growth, as confirmed by the fact that DT treatment could eliminate only stromal senescent cells but not cancer senescent cells in the p16-Cre^ERT2-DTR with PDAC allograft model. Targeting overall senescent cells within tumor may be a promising therapeutic strategy. The tumor suppressive effect of senolysis may be due to the increase in the activated CD8⁺ cytotoxic T cells and the amelioration of the desmoplastic reaction, which likely acts as a barrier to immune cell infiltration. Interestingly, the combined application of senolytic agents with chemotherapy revealed an additive tumor-suppressive effect, indicating that targeting senescent stromal cells might hold promise for cancer therapy.

TGF-β-mediated collagen expression has been reported to be regulated by SMAD2/3 (73). CREB3L1 is known to be a downstream mediator of TGF-β signaling that promotes collagen expression (76). In senescent fibroblasts, we found that CREB3L1 is a major transcription factor responsible for collagen expression, demonstrated by the significant rescue of TGF-β-mediated collagen expression upon PF429242 treatment. Surprisingly, CREB3L1-dependent collagen induction was specific to senescent fibroblasts, as PF429242 had almost no effect on collagen expression in uninduced cells. Therefore, a CREB3L1 inhibitor may be an alternative strategy to improve desmoplastic reactions in PDAC.

In conclusion, our findings suggest that cancer stromal senescent networks contribute to the establishment of a procancer malignant microenvironment, characterized by immune suppression, angiogenesis, fibrosis, and enhanced invasion and metastasis of PDAC cells. Consequently, the combination of targeting stromal senescent cells, conventional chemotherapies, and immune checkpoint blockade may provide a novel therapeutic strategy for PDAC.

Materials and Methods

Detailed descriptions of Materials and Methods are in SI Appendix.

Mouse Model.

The p16-Tom mice were heterozygous and generated by crossing p16Ink4a-Cre^ERT2 mice with Rosa26-CAG-LSL-tdTomato-WPRE mice (15). p16-Cre^ERT2-DTR mice were generated by crossing p16Ink4a-Cre^ERT2 mice with Rosa26-SA-LSL-DTR-IRES-tdTomato mice. Details are in SI Appendix.

Cell Culture.

All cells were cultured in DMEM (Nacalai tesque) supplemented with 10% fetal bovine serum (FBS) (Sigma) and 1x penicillin/streptomycin/amphotericin B (Nacalai tesque). Cells were kept at 37 °C under normoxia conditions except for pulmonary fibroblasts under hypoxia conditions with 5% oxygen. See SI Appendix for details.

Allograft Model of Pancreatic Adenocarcinoma.

The mouse pancreatic tumor cell clone was kindly provided by Dr. Ben Z. Stanger (16) which was isolated from late-stage primary tumors from C57BL/6 background KPCY (Kras^LSL-G12D/+; Trp53^LSL-R172H/+; Pdx1-Cre; Rosa26^YFP/YFP) mice (77). See SI Appendix for details.

Chemotherapeutic and Senolytic Treatments.

In the chemotherapy regimen, gemcitabine (G) (Sandoz) at a concentration of 12 mg/mL in PBS was administrated at a dosage of 120 mg/kgBW, and Abraxane (nab-paclitaxel, A) (Taiho Pharmaceutical) at a concentration of 12 mg/mL in PBS was administrated at a dosage of 120 mg/kgBW (the dose of paclitaxel would be 12 mg/kgBW). See SI Appendix for details.

In Vitro T Cell Inhibition Assay.

In vitro T cell inhibition assay was performed using the spleen obtained from C57BL/6 mice. All antibodies used are listed in SI Appendix, Table S1. See SI Appendix for details.

Sphere Forming Assay.

The 100 PDAC cells (YFP⁺/DAPI^-) FACSorted from p16-Cre^ERT2 and p16-Cre^ERT2-DTR mice were subjected to sphere formin assay. See SI Appendix for details.

Plasmid Construction, Lentivirus Production, Tissue Dissociation, Flow Cytometry Analysis, scRNAseq Analysis, Histological Immunostaining, Protein Extraction, Immunoblotting, RNA Isolation, and Real-Time PCR.

Standard methods were adopted. All antibodies and primers used in this study are listed in SI Appendix, Tables S1 and S2 respectively. Details are in SI Appendix.

Cytokine Array Analysis.

The tumors collected from p16-Cre^ERT2 or p16-Cre^ERT2-DTR mice were subjected to cytokine array analysis. See SI Appendix for details.

Statistical Analysis.

For animal experiments, all the male mice were randomly assigned to each group and independently followed the same age-dependent schedule in each experimental design. The sample sizes were not predetermined by pilot studies. Blind designs were not employed in this study because of the automatic analyses obtained using the image analyzer with the same criteria. Data are presented as means ± SEM unless otherwise noted. Comparisons between the two groups were made by an unpaired two-tailed Student’s t test. Multicomparisons of one-variable data were carried out by one-way ANOVA followed by a post hoc Tukey’s test or Dunnett’s test. Multicomparisons of multiple-variable data were performed by two-way ANOVA followed by Sidak’s multiple comparisons test. For all representative findings, triplicate or multiple independent experiments were performed, and similar results were obtained.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

scRNA-seq datasets data have been deposited in Gene Expression Omnibus (GEO) (GSE252684) (78). All materials and mouse models are available from corresponding authors on request.