Abstract

   Cancer-associated fibroblasts (CAFs) are a heterogeneous cell
   population that plays a crucial role in remodeling the tumor
   microenvironment (TME). Here, through the integrated analysis of
   spatial and single-cell transcriptomics data across six common cancer
   types, we identified four distinct functional subgroups of CAFs and
   described their spatial distribution characteristics. Additionally, the
   analysis of single-cell RNA sequencing (scRNA-seq) data from three
   additional common cancer types and two newly generated scRNA-seq
   datasets of rare cancer types, namely epithelial-myoepithelial
   carcinoma (EMC) and mucoepidermoid carcinoma (MEC), expanded our
   understanding of CAF heterogeneity. Cell–cell interaction analysis
   conducted within the spatial context highlighted the pivotal roles of
   matrix CAFs (mCAFs) in tumor angiogenesis and inflammatory CAFs (iCAFs)
   in shaping the immunosuppressive microenvironment. In patients with
   breast cancer (BRCA) undergoing anti-PD-1 immunotherapy, iCAFs
   demonstrated heightened capacity in facilitating cancer cell
   proliferation, promoting epithelial-mesenchymal transition (EMT), and
   contributing to the establishment of an immunosuppressive
   microenvironment. Furthermore, a scoring system based on iCAFs showed a
   significant correlation with immune therapy response in melanoma
   patients. Lastly, we provided a web interface
   ([41]https://chenxisd.shinyapps.io/pancaf/) for the research community
   to investigate CAFs in the context of pan-cancer.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12943-023-01876-x.

   Keywords: Spatial transcriptomics, Single-cell RNA sequencing,
   Pan-cancer analysis, Cancer-associated fibroblasts, Tumor
   microenvironment, Tumor immunotherapy

Introduction

   Tumors display extensive heterogeneity, with cancer cells engaging in
   reciprocal interactions with their microenvironment, forming a complex
   ecosystem [[42]1]. Cancer-associated fibroblasts (CAFs), as one of the
   most prominent and abundant cell populations in the tumor
   microenvironment (TME) [[43]2], have garnered significant attention in
   recent years. CAF's intricate interactions with stromal components and
   immune cells play a crucial role in orchestrating TME reorganization,
   encompassing processes such as angiogenesis, extracellular matrix (ECM)
   remodeling, and immune evasion [[44]3–[45]5]. At present, the crucial
   role of CAFs has been largely overlooked by most therapies, including
   immunotherapy and chemotherapy [[46]3]. Our current understanding of
   the interplay between CAFs and components of TME is insufficient to
   support the development of reliable treatment strategies. Further
   research is needed to deepen our understanding of these interactions
   and pave the way for effective therapeutic interventions.

   In recent years, the application of single-cell transcriptomics has
   unraveled the heterogeneity of CAFs within many cancer types, such as
   bladder carcinoma (BC) [[47]6], head and neck squamous cell carcinoma
   (HNSCC) [[48]7], papillary thyroid carcinoma (PTC) [[49]8], and lung
   cancer (LC) [[50]9]. Furthermore, two recent unbiased studies based on
   single-cell RNA sequencing (scRNA-seq) explored the heterogeneity and
   plasticity of CAFs from a pan-cancer perspective and revealed the
   conservation of CAF phenotypes across cancer types [[51]10, [52]11].
   Although scRNA-seq provides an unprecedented opportunity to
   systematically dissect the heterogeneity of CAFs, the loss of spatial
   information during tissue dissociation hinders the study of the
   crosstalk between CAFs and TME. Recently developed spatial
   transcriptomics (ST) can obtain whole-transcriptome data within tissue
   sections, thereby preserving the spatial position information of cells
   [[53]12]. Therefore, orthogonal integration of scRNA-seq data and ST
   data will help determine the spatial distribution characteristics of
   CAFs and further dissect the cellular communication between CAFs and
   TME.

   In this study, we have delineated the landscape of CAFs in six common
   cancer types and described the unique functional features of these
   subtypes. We also analyzed scRNA-seq data of three additional common
   tumors and two newly sequenced rare tumors to expand our understanding
   of CAF heterogeneity. A spatial single-cell transcriptomic atlas
   spanning six tumors, including 744,289 cells, generated by integrating
   scRNA-seq data and ST data was used to describe the spatial
   distribution characteristics of CAFs and to characterize the complex
   interactions between CAFs and TME. Notably, a score generated based on
   inflammatory CAFs (iCAFs) showed a significant correlation with the
   response of melanoma patients to immunotherapy. In summary, our
   integrated data resources provide novel insights and guidance for the
   development of therapeutic strategies targeting CAFs in TME.

Results

Construction of a pan-cancer spatial single-cell transcriptome atlas

   To establish a spatial single-cell landscape in pan-cancer, we acquired
   scRNA-seq data from 69 samples of 56 patients diagnosed with one of the
   six prevalent cancer types, along with ST data from 22 tissue slices of
   22 patients (Fig. [54]1a and b; Table S[55]1 and S[56]2). Among them,
   the ST data of 10 tissue slices had corresponding scRNA-seq data from
   the same patient (Fig. [57]1c; Table S[58]1 and S[59]2). The data we
   collected included six types of cancer: BRCA, colorectal cancer (CRC),
   liver hepatocellular carcinoma (LIHC), ovarian cancer (OVCA), prostate
   adenocarcinoma (PRAD), and uterine corpus endometrial carcinoma (UCEC)
   (Fig. [60]1a; Table S[61]1 and S[62]2). After strict quality control
   and filtration, a total of 163,919 cells in the scRNA-seq data and
   59,529 spots in ST data were retained for downstream analysis
   (Fig. [63]1d and S[64]1a). In the scRNA-seq dataset, the median number
   of unique molecular identifiers (UMIs) per cell was 3955, and the
   median number of genes per cell was 1425 (Figure S[65]1b and c). For ST
   analysis, the median number of UMIs per spot was 11,139 and the median
   number of genes per spot was 3,863 (Figure S[66]1d and e). To minimize
   the batch effect between different scRNA-seq datasets, we independently
   analyzed each dataset. Taking CRC as an example, we used graph-based
   clustering and identified seven major clusters based on typical markers
   of different cell types (Table S[67]3), including epithelial cells,
   fibroblasts, endothelial cells, T&NK, B cells, myeloid cells and mast
   cells (Fig. [68]1e and f). CopyKAT was used to estimate the single-cell
   copy number variation (CNV) landscape of tumors, in order to
   distinguish malignant epithelium from non-malignant epithelium
   (Fig. [69]1e). The myeloid cells were further divided into monocytes,
   macrophages and dendritic cells (Fig. [70]1e). The CD8 + T cells,
   CD4 + T cells, regulatory T cells (Treg cells) and natural killer cells
   (NK cells) were identified from the T&NK cluster (Fig. [71]1e).
   Similarly, cells from the other 5 types of cancer were clustered into
   roughly the same subgroups (Figure S[72]2, S[73]3, S[74]4 and S[75]5).
   Of note, we detected neutrophils in the scRNA-seq data from the inDrop
   platform, which were not detected in the scRNA-seq data from the
   10 × Genomics platform (Figure S[76]2). Neutrophils are very fragile
   and have low RNA content [[77]13], which may be the main reason for the
   capture failure.

Fig. 1.

   [78]Fig. 1
   [79]Open in a new tab

   A pan-cancer spatial single-cell transcriptome atlas. a Schematic
   depicting the study design. The cancer types included in this
   pan-cancer study were displayed in the first image on the left, created
   by Figdraw. b The number of samples in the pan-cancer analysis of
   scRNA-seq and ST. c Pie chart showing the proportion of ST sections
   that have corresponding scRNA-seq data from the same patient compared
   to those without such corresponding scRNA-seq data. d The number of
   cells in the pan-cancer analysis of scRNA-seq and ST. e Uniform
   Manifold Approximation and Projection (UMAP) plots showing the major
   cell types in CRC. f Bubble heatmap showing the expression of marker
   genes for the major cell types in CRC. g Spatial cell charting of CRC
   using CellTrek

   CellTrek is a computational toolkit that enables direct mapping of
   individual cells back to their spatial coordinates in tissue sections
   based on scRNA-seq and ST data [[80]14]. Unlike ST deconvolution
   methods, this approach transferred ST coordinates to single cells,
   thereby achieving single-cell resolution [[81]14]. We applied it to
   quality-controlled scRNA-seq and ST data in pan-cancer to reconstruct
   spatial single-cell atlases. Even without corresponding scRNA-seq data
   from the same patient, ST datasets were still largely covered by
   scRNA-seq datasets based on the co-embedding results (Figure S[82]6).
   Due to some cells being repeatedly mapped, we ultimately obtained a
   pan-cancer spatial single-cell transcriptomic atlas containing 744,289
   cells (Fig. [83]1d and g).

CAF heterogeneity in pan-cancer

   To compare the similarity of the main cell lineages of different cancer
   types, we constructed a phylogenetic tree (Figure S[84]7a). Compared
   with the biased distribution of epithelial cells, fibroblasts from
   different cancer types clustered together (Figure S[85]7a), indicating
   that fibroblasts had similar transcriptional features in different
   cancer types. Interestingly, NK cells and B cells originating from UCEC
   demonstrated unique features (Figure S[86]7a), implying that the TMEs
   across diverse cancer types could have potentially exerted distinct
   effects on immune cell phenotypes. We subsequently investigated the
   heterogeneity of fibroblasts in scRNA-seq datasets of the 6 cancer
   types (Fig. [87]2a). The reclustering of the fibroblast cluster
   identified four CAF subtypes, as well as pericytes and smooth muscle
   cells (SMCs) (Fig. [88]2a). After applying Harmony for batch
   correction, all cells with local inverse Simpson's Index (LISI) greater
   than 1 indicate that no obvious batch effects were observed (Figure
   S[89]8). CFD + fibroblasts showed high expression of chemokines (CCL11,
   CXCL12, and CXCL14) (Fig. [90]2b; Table S[91]4), similar to the
   previously reported iCAFs in various types of tumors such as BC [[92]6]
   and PTC [[93]8]. GO enrichment analysis of its marker genes showed
   their association with the response to mechanical stimulation, reactive
   oxygen species, epithelial cell proliferation, immune system, and cell
   migration (Fig. [94]2c). POSTN + fibroblasts showed high expression
   levels of several ECM remodeling genes (MMP11, CTHRC1, COL1A1, COL1A2,
   COL3A1, COL10A1, and COL11A1) and enriched signatures of ECM
   (Fig. [95]2b and c; Table S[96]4), which were consistent with the
   previously reported matrix CAFs (mCAFs) in cervical squamous cell
   carcinoma (CESC) [[97]15]. Interestingly, a cluster of cells was
   related to the response to hypoxia and canonical glycolysis
   (Fig. [98]2c), resembling the reported metabolic CAFs (meCAFs) in
   pancreatic ductal adenocarcinoma (PDAC) [[99]16]. Notably, we also
   found a cluster of cells that exhibited higher expression of a set of
   cell cycle-related genes (CENPF, NUSAP1, PTTG1, STMN1, TOP2A, and
   TUBA1B) (Fig. [100]2b; Table S[101]4), which was consistent with
   proliferative CAFs (pCAFs) in a previous pan-cancer study of CAFs
   [[102]10]. Immunofluorescence on tissue microarrays from BRCA patients
   further substantiated the existence of the four CAF subtypes (Figure
   S[103]9). Next, we further investigated the heterogeneity of CAFs using
   the AUCell algorithm, based on the functional features of CAFs
   summarized by Lavie et al. [[104]17] (Fig. [105]2d; Table S[106]5).
   iCAFs exhibited the highest activity in immune-related functions,
   including complement activation, chemokine production, and inflammatory
   response (Fig. [107]2d). Additionally, the biological processes of
   angiogenesis, wound healing, regulation of ECM organization and
   collagen biosynthetic process were all enriched in mCAFs
   (Fig. [108]2d). As expected, meCAFs exhibited a high level of
   glycolytic activity (Fig. [109]2d). Interestingly, in addition to the
   cell cycle, pCAFs were also involved in IFN − I production and muscle
   contraction (Fig. [110]2d).

Fig. 2.

   [111]Fig. 2
   [112]Open in a new tab

   CAF heterogeneity in pan-cancer. a UMAP plots showing the integration
   of fibroblasts across six different cancer types by Harmony. b
   Differential expression analysis showing the upregulated genes for each
   fibroblast subtype. An adjusted p value < 0.05 is indicated in red,
   while an adjusted p value ≥ 0.05 is indicated in blue. c GO enrichment
   analysis of upregulated genes in each CAF subtype. d Heatmap showing
   pathway activities scored by AUCell in each CAF subtype. e Proportion
   of CAF subtypes across multiple cancer types. f Heatmap showing the ORs
   of CAF subtypes in each cancer type. g Scatter plot showing the RSSs in
   each CAF subtype. The top 5 regulons are highlighted. h SCAP analysis
   of metabolic pathways in meCAFs. i Slingshot trajectory analysis of
   CAFs. j GeneSwitches analysis of pathway activity changes in the
   transition pathway from pericytes to iCAFs

   Although iCAFs and mCAFs were the major CAFs celltypes across 6 cancer
   types, different subtypes of CAFs still exhibited significant cancer
   preferences (Fig. [113]2e and f). iCAFs were enriched in BRCA and CRC,