Adipose-derived mesenchymal stem cells differentiate into heterogeneous cancer-associated fibroblasts in a stroma-rich xenograft model

Cancer-associated fibroblasts (CAFs) are the key components of the densely proliferated stroma in pancreatic ductal adenocarcinoma (PDAC) and contribute to tumor progression and drug resistance. CAFs comprise heterogeneous subpopulations playing unique and vital roles. However, the commonly used mouse models have not been able to fully reproduce the histological and functional characteristics of clinical human CAF. Here, we generated a human cell-derived stroma-rich CDX (Sr-CDX) model, to reproduce the clinical tumor microenvironment. By co-transplanting human adipose-derived mesenchymal stem cells (AD-MSCs) and a human PDAC cell line (Capan-1) into mice, the Sr-CDX model recapitulated the characteristics of clinical pancreatic cancer, such as accelerated tumor growth, abundant stromal proliferation, chemoresistance, and dense stroma formed from the heterogeneous CAFs. Global RNA sequencing, single-cell based RNA sequencing, and histological analysis of CAFs in the Sr-CDX model revealed that the CAFs of the Sr-CDX mice were derived from the transplanted AD-MSCs and composed of heterogeneous subpopulations of CAF, including known and unknown subtypes. These lines of evidences suggest that our new tumor-bearing mouse model has the potential to address an open question in CAF research, that is the mechanism of CAF differentiation.


Co-transplantation of adipose-derived MSCs and human PDAC cell line induced high tumor growth and chemotherapy resistance.
To obtain an in vivo model that mimics clinical PDAC characteristics (i.e., stromal-rich and capable of cancer gland formation) 13,28 , a human PDAC cell line (Capan-1) was co-transplanted with several commercially available human MSCs (Fig. 1a). Among adipose-, bone marrow-, or umbilical cord tissue-derived human MSCs, AD-MSCs exhibited the most prominent tumor-promoting effect and histology similar to the clinical PDAC (Fig. S1) 27 . AD-MSC was therefore selected as a source of human stromal cells to develop a new mouse model. Tumors in the Sr-CDX mice were significantly larger than those in the conventional CDX (Capan-1 only) mice throughout the observation period (Fig. 1b). Notably, the average tumor weight in the Sr-CDX group was approximately four times greater than that in the conventional CDX group at day 23 (Fig. 1c). AD-MSCs alone did not induce tumor formation. Similar results were obtained using another cell line, MIAPaCa-2 ( Fig. S2a,b). The chemotherapy resistance of Sr-CDX mice, a burden of clinical PDAC, which is largely attributed to the role of the stromal desmoplasia, was then assessed. The Sr-CDX and conventional CDX (Capan-1 only) groups were treated thrice with gemcitabine through intraperitoneal administration. The effect of gemcitabine was clearly limited in the Sr-CDX group as opposed to the CDX group, recapitulating the chemotherapy resistance in Sr-CDX mice (Fig. 1d).

Co-transplantation of human adipose-derived MSCs and Capan-1 cells recapitulated the dense stroma and cancer gland formation, histological characteristics of human PDAC. In contrast
to the conventional CDX mice, in which the stromal amount is quite poor, Sr-CDX mice showed morphological similarity with human clinical PDAC tumors, which involves an abundant stroma surrounding the ductal glands consisting of cancer cells (Fig. 2a). When the stromal fibrosis was highlighted using blue color with Masson's trichrome (MT) staining (Fig. 2a) and quantified, the stromal collagen area in Sr-CDX mice was significantly larger than that in the CDX mice (Fig. 2b). Co-transplantation of MIAPaCa-2 with AD-MSCs showed an increase in the stromal area, but the formation of glandular tubular structures was not seen (Fig. S2c).
It is interesting to note whether the origin of stromal cells in Sr-CDX is the transplanted AD-MSCs or other cells of the mouse host. To confirm the transplanted AD-MSCs, the cells were labeled with green fluorescent protein (GFP) in lentiviral transduction 27 . The evolution of stromal cells over time could be traced by immunohistochemistry (IHC) against GFP. It was confirmed that almost all stromal cells in the Sr-CDX mice were GFPpositive, indicating that the dense stromal cells originated from the transplanted AD-MSCs and its derivatives (Fig. 2c). The GFP-positive stromal area under the total visual field gradually increased until day 14 (Fig. 2d). It was also explored whether the AD-MSCs differentiated into CAFs in Sr-CDX mice. IHC of the CAF marker αSMA revealed that αSMA-positive cells were located in close proximity to the E-cadherin-positive cancer cells, indicating that these are possibly myCAFs (Fig. 2e), and that these αSMA-positive myCAFs were derived from the GFP-positive AD-MSCs (Fig. 2f). CAFs in human clinical PDAC are known to be heterogeneous, including the αSMA-positive myCAFs located adjacent to the ductal glands and IL-6-positive iCAFs located apart from the cancer cells (Fig. 2g). It was observed that stromal cells in the Sr-CDX mice exhibited a immunohistochemical profile similar to that of the human PDACs, such as αSMA-positive, IL-6 positive, or double-positive profiles Gene-expression patterns of Sr-CDX CAFs showed known CAF-related gene profiles. RNA sequencing (RNA-seq) analysis was performed to investigate the global gene-expression patterns of the original mono-cultured AD-MSCs, Sr-CDX CAFs, and clinical CAFs. The latter two were obtained using the outgrowth method (Fig. 3a). When transcriptomes of the Sr-CDX CAFs and original AD-MSCs were compared, 2608 genes were differentially expressed, of which 1176 were upregulated and 1432 downregulated in Sr-CDX (Fig. 3b). These upregulated genes included the iCAF and myCAF markers such as IL-6, Leukemia Inhibitory Factor (LIF), actin alpha 2 (ACTA2), and Tropomyosin (TPM1) (Fig. 3b, Fig. S3, Table S1).
To narrow down the list of CAF-related gene signatures in the Sr-CDX CAFs from 1176 genes, clustering analysis was performed by adding two clinical CAFs (Fig. 3c). After selecting the clusters involving the genes upregulated in Sr-CDX CAFs, compared to those of the original AD-MSC (cluster B, E, and F in Fig. 3c), the expression pattern was clearly observed in both clinical CAFs. The B cluster contained 252 genes (black box in Fig. 3c, Table S2). These 252 genes included the myCAF markers such as ACTA2 and COL15A1 (Fig. 3d in red characters) and iCAF markers such as IL-6 and hyaluronan synthase 2 (HAS2, Fig. 3d in green characters, Table S2) 10 . To investigate the functional changes, Gene Ontology (GO) enrichment analysis was performed with the 252 genes in cluster B. The representative GO terms identified in the biological process category were "cell adhesion" (p = 2.23 E-13), "extracellular matrix organization" (p = 4.27 E-12), "blood vessel development" (p = 2.20 E-8), and "blood vessel morphogenesis" (p = 1.09 E-7, Fig. 3e and Table S3). Briefly, soon after harvesting and digesting the tumors comprising GFP-labeled AD-MSCs and red fluorescent protein (RFP)-labeled Capan-1 cells, fluorescence-activated cell sorting (FACS) was used to isolate viable GFPpositive Sr-CDX CAF cells (Fig. 4a). The gene signatures of original mono-cultured AD-MSCs, including 10,994 single cells, revealed that AD-MSCs formed three large clusters with unique gene signatures, according to the t-distributed stochastic neighbor embedding (t-SNE) analysis (Fig. 4b). The myCAF marker genes (ACTA2 and TPM1) and common CAF marker genes (COL1A1 and fibroblast activation protein alpha, FAP) were highly expressed in clusters 1 and 2, whereas expression of iCAF-or apCAF-related marker genes was not observed in the original AD-MSCs (Fig. 4b,c, and Fig. S4a,b). Therefore, clusters 1 and 2 were defined as "myCAF-like clusters" (Fig. 4b).
The gene signatures of Sr-CDX CAFs containing 699 single GFP-positive cells formed seven distinct clusters according to t-SNE analysis, showing increased subpopulations (Fig. 4d). The common CAF marker genes (COL1A1 and FAP) and myCAF marker genes (ACTA2 and TPM1) were highly expressed in clusters 5-7, whereas iCAF marker genes (IL-6 and LIF) and apCAF marker genes (CD74 and HLA-DRA) were highly expressed in clusters 1-4 and 7 ( Fig. 4e and Fig. S4c). The gene-expression levels of several reported CAF markers in seven clusters were investigated [9][10][11] ; the heat map showed varying levels of CAF marker expression among the seven clusters (Fig. 4f). In clusters 1-4, iCAF and apCAF markers were expressed at different levels, whereas myCAF and common CAF markers were not expressed. In cluster 5, myCAF markers were highly expressed, whereas other CAF marker expression levels were relatively low. In cluster 6, both myCAF and common CAF marker expression levels were high. In cluster 7, myCAF marker was highly expressed; iCAF (LIF and LMNA), apCAF (HLA-DRA), and common CAF markers were also expressed. Clusters 1-4 were defined as iCAF/apCAF-like clusters, clusters 5 and 6 as myCAF-like clusters, and cluster 7 as an ambiguous myCAF/iCAF/apCAF -like cluster. Cluster-specific genes were extracted from seven clusters, and the GO enrichment analysis revealed functional differences between each CAF cluster ( Fig. S4d-e). In addition, IHC was performed using the extracted cluster markers, and their presence in both clinical PDAC and Sr-CDX was confirmed ( Fig. 4g-h).

Discussion
We evaluated the potential of AD-MSCs in providing heterogeneous CAF populations, as well as its contribution in generating a stroma-rich TME, with the CAF differentiation into multiple subtypes. Hitherto, PSCs and BM-MSCs have been considered as CAF precursors, and have been of interest. It has been proved that PSCs interact with cancer cells as well as with other cellular elements in the stroma, including immune cells, endothelial cells and neuronal cells, to set up a growth permissive microenvironment for PDACs 29 . In addition, recent studies have shown that PSCs possess the ability to differentiate into CAF subtypes, myCAF and iCAF 11 . BM-MSCs have shown the capability to adhere to the tumor site and promote tumor growth 18,20 , and be involved in the acquisition of chemotherapy resistance 21 . On the other hand, AD-MSCs have also been reported to differentiate into CAFs and be involved in tumor growth and desmoplastic reactions; however, the number of reports is relatively small 15,16 . Although GEMMs, such as the KPC (LSL-KrasG12D/+; LSL-Trp53R172H/+; Pdx-1-Cre) mouse model 30 , are excellent; verifying the relationships between the origin and heterogeneity of CAFs in those mice is difficult as well as clinical pancreatic cancer. In this study, immortalized AD-MSCs were efficacious in inducing CAF differentiation and succeeded in recapitulating the clinical PDAC characteristics, such as promoting tumor growth, drug resistance, and stromal desmoplasia, more accurately than the conventional in vivo models such as CDX, PDX, and GEMM.
RNA-seq analysis demonstrated that various CAF markers, including myCAF and iCAF markers, were differentially upregulated in Sr-CDX CAF (Fig. 3b). GO enrichment analysis, however, highlighted only the GO  www.nature.com/scientificreports/ terms related to the myCAF functions (Fig. 3e). This discrepancy may be explained by the higher expression levels of myCAF markers in Sr-CDX CAFs than iCAF or apCAF markers. Therefore, CAF heterogeneity was examined further in our Sr-CDX models at the single-cell level. www.nature.com/scientificreports/ Pioneering studies have already suggested the existence of multiple subpopulations of CAFs with distinct transcriptional profiles 10,31-33 . The presence of three representative subpopulations of CAFs was also reconfirmed in the CAFs of Sr-CDX models, as demonstrated by the presence of roughly two groups related to myCAF (clusters 5-7) and iCAF (clusters 1-4); additionally, the apCAF population overlapped mainly with the iCAF population ( Fig. 4d-e). The presence of a fourth subtype of CAF, a purported mixture of myCAFs and iCAFs, has been proposed by Neuzillet et al. and named subtype C 9 ; this subtype was visible both in the scRNA-seq analysis (cluster 7) and in IHC analysis (Fig. 2g,h, and 4d-h). In addition to the roles of myCAFs and iCAFs, the GO analysis indicated that the cluster 7 CAFs may be involved in "immune responses" that are associated with antigen presentation. Among the iCAF clusters containing four subclasses, CAFs belonging to the cluster 1 are enriched with apoptosis-related GO terms and may play a role in preventing cancer cell apoptosis and assisting cancer cell growth. It may have been related to tumor growth and chemotherapy resistance in the Sr-CDX mice (Fig. 1a-d and 5). CAFs in clusters 4 and 5 with enriched mitochondria-related GO terms may demonstrate activated metabolism and resistance to hypoxia and chemotherapy. CAFs in clusters 5-7 enriched with ECM-related GO terms may promote the production of abundant ECM, resulting in abundant stroma of the tumor in Sr-CDX and desmoplasia in clinical PDAC (Fig. 2a and 5). Notably, our Sr-CDX model contains a heterogeneous subpopulation of CAFs despite using only AD-MSCs as the source of CAFs, providing an excellent system for investigating CAFs. Considering that AD-MSCs alone did not form tumors in vivo, AD-MSCs could have differentiated into CAFs by interacting with Capan-1 cells. As we have shown in our in vitro co-culture experiments 27 , CAFs could be differentiated by contact or non-contact interaction with Capan-1. In that case, TGFβ and other factors secreted by cancer cells would be expected to be one of the factors that cause AD-MSCs to differentiate into CAFs 11,34 . In addition, cancer cells are expected to be more diverse in vivo TME compared to in vitro 35 . The heterogeneous and diverse CAFs generated from AD-MSCs in this study are considered to be differentiated by various signals from diverse cancer cells or by signals with concentration gradients in the tumor microenvironment.
These results could be attributed to the use of immortalized AD-MSCs. In fact, human telomerase reverse transcriptase (hTERT) immortalized fibroblasts have been used to retain important phenotypic traits in breast cancer research 36 , and also are reported to be useful to investigate the heterogeneous functions and phenotypes of CAFs 37 . The use of immortalized cells may allow human CAFs to maintain its heterogeneity and important functions in the tumor.
In conclusion, our Sr-CDX model, using the AD-MSCs, accurately reproduced the pathological features, chemotherapy resistance, and CAF heterogeneity observed in the clinical PDAC, and confirmed that the AD-MSCs can differentiate into multiple subpopulations of CAFs (Fig. 5). This novel Sr-CDX model has the potential to elucidate the mechanism of CAF differentiation and explain the contribution of CAFs in the progression of pancreatic cancer.

Methods
Cells and culture conditions. The immortalized human AD-MSC cell line ASC52telo (ATCC SCRC-4000) and the human pancreatic cancer cell lines Capan-1 (ATCC HTB-79) and MIAPaCa-2 (CRL-1420) were utilized as per the previous reports 22,27 . The isolation and culture conditions for cell lines and CAFs from the primary tumors were described previously 27 . Briefly, AD-MSCs were labeled with GFP, whereas Capan-1 cells were labeled with RFP via lentiviral transduction. HEK293T cells were seeded to a 6 well plate and co-transfected with pLKO.1-puro eGFP (Sigma-Aldrich, St. Louis, MO, USA), psPAX2 and pMD2.G at a ratio of 2:1.5:1.2 μg/ well. Twenty-four hours later the viral supernatant was collected and filtrated with 0.45 μm membranes. The AD-MSCs were infected with the viral supernatant in the presence of Polybrene (10 μg/ml) for 24 h. After 72 h, puromycin was added to the medium to purify GFP-positive cells. To establish the clinical CAFs isolated from PDAC patients, PDAC tumor sections were minced and two or three 2-mm tumor pieces were plated onto a gelatin-treated 3.5-cm dish in HFDM-1 medium (Cell Science & Technology Institute, Sendai, Japan) supplemented with 1% streptomycin-penicillin and 5% FBS. The dishes were then incubated under 5% CO 2 , 20% O 2, at 37 °C. After a couple days of incubation, fibroblasts grew around the tumor fragments, and then the tumor fragments were removed. When an enough number of cells were observed, they were passaged or stored. Under these culture conditions, CAFs selectively grown, while the remaining PDAC cells were depleted following a few passages.  -0317-B). A total of six resected pancreatic cancer specimens were used in this study. For the application of these clinical samples to research purposes, written informed consent was obtained from all patients.

Mouse model and in vivo experiments.
Eight-week-old female nude mice (Balb/c nu/nu) were purchased from Japan CLEA Inc. (CLEA Japan, Tokyo, Japan) for use in this study. Mice were bred and housed under specific pathogen-free conditions at the Animal Center of AIST and University of Tsukuba and were randomized into two groups: Sr-CDX group and conventional CDX group (control group, Capan-1 only). In the CDX model, 1 × 10 6 AD-MSCs and 1 × 10 6 Capan-1 cells were subcutaneously transplanted into the mice, and the sizes of the generated tumors were monitored. After testing various ratios, the mixing ratio of AD-MSC and Capan-1 was determined to be 1:1. The tumor volume and weight of Sr-CDX group (n = 3) were compared   Inc., Cincinnati, OH, USA), according to the manufacturer's instructions. The library preparation and sequencing were conducted at Macrogen Japan using a Truseq library prep kit and NovaSeq 6000 (Illumina, San Diego, CA, USA) to produce 150-bp paired-end reads. The acquired data from two independent devices for each condition were mapped and quantified using STAR (2.7.1a) and RSEM (1.3.1). hg38 was used as the reference genome, and Ensemble GRCh38 was used for the gene annotation 39,40 . Subsequently, differentially expressed genes were analyzed by iDEP.91 41 . To investigate the function of the differentially expressed genes, GO enrichment analysis was performed using the database for annotation, visualization, and integrated discovery (DAVID) 42,43 . The raw sequences in FASTQ format were deposited at DNA Data Bank of Japan (DDBJ, accession number DRA010287).
Sample preparation, staining, and sorting for single-cell RNA-seq. Tumor  Single-cell data processing and analysis. The single-cell RNA-seq library was constructed using Chromium Controller and Chromium Single Cell 3′ Reagent Kits v3.1 (#CG000204, 10× Genomics, Pleasanton, CA, USA) according to the manufacturer's protocol. The library was sequenced using NovaSeq6000 (Illumina) in accordance with the manufacturer's instructions to obtain 100-bp paired-end reads. After sequencing, FASTQ files were generated using Cell Ranger ver. 2.1.0 mkfastq (10× Genomics). Dimensionality reduction with the t-distributed stochastic neighbor embedding (t-SNE) algorithm for visualization and k-means clustering was performed using Cell Ranger. The raw sequences in FASTQ format are available on DDBJ (accession number DRA010288).

Statistical analysis.
Data are represented as the mean ± standard deviation (SD). The differences between each group were compared using unpaired two-tailed Student's t-tests or one-way analysis of variance (ANOVA) with Tukey's multiple comparison test. All data were evaluated using ANOVA in the statistical analysis software package SPSS version 25.0 (IBM SPSS Statistics). A value of P < 0.05 was considered statistically significant. The error bars in the figures represent the SD.

Data availability
RNA-seq data are available in the DDBJ Sequence Read Archive under accession number DRA010287 and DRA010288. The raw data are available from the corresponding author upon reasonable request.