Pheochromocytomas (PCCs) are rare neuroendocrine tumors that originate from chromaffin cells in the adrenal gland (Fishbein et al, 2017; Toledo et al, 2017). PCCs have been shown to display a remarkable diversity of driver alterations, including germline and somatic mutations as well as somatic fusion genes (Favier et al, 2015; Fishbein et al., 2017; Pillai et al, 2017). This diversity is reflected in the current molecular taxonomy of PCCs The Cancer Genome Atlas (TCGA), including for example the pseudo-hypoxic type (germline mutations in SDHx, FH, and VHL, etc.), the kinase-signaling type (germline or somatic mutations in RET, NF1, TMEM127, MAX, and HRAS, etc.), and the Wnt-signaling type (somatic mutations in CSDE1 and somatic gene fusions affecting MAML3) (Crona et al, 2017). However, 50∼60% of PCCs fail to be classified using this system (Lenders et al, 2014).

As understanding of PCC clinicopathological and immunophenotypic characteristics has deepened, the former concept of ‘benign or malignant PCC’ has been abandoned; since 2017, all PCCs are regarded as malignant tumors with metastatic potential according to the WHO pathological classification (Lam, 2017). There is no single histo-morphological feature indicating the malignant potential of PCCs; a number of multifactorial systems have been proposed (Kulkarni et al, 2016). A cytomorphometric study reported that extra-adrenal location, coarse nodularity, confluent necrosis, and the absence of hyaline globules are associated with malignant behavior (Lewis, 1971). Several scoring systems that for example consider vascular invasion, tumor size, diffuse growth, and mitotic activity have been used for risk assessment of malignant behavior of PCCs (Sherwin, 1959; Sisson et al, 1984; Symington & Goodall, 1953). A scoring system—the Pheochromocytoma of the Adrenal gland Scaled Score (PASS)—is weighted for histologic features and has been used to separate tumors with a potential for a biologically aggressive behavior (PASS > or = 4) from tumors with benign lesion (PASS < 4) (Thompson, 2002). However, there have been conflicting reports regarding the utility of the PASS scoring system in correctly predicting PCC malignancy (Agarwal et al, 2010; Strong et al, 2008; Wu et al, 2009).

Immune checkpoint inhibitors (ICIs) have achieved remarkable results in a variety of solid tumors in recent years (Brahmer et al, 2012; Herbst et al, 2014; Ribas & Wolchok, 2018). Currently, there are only 2 clinical trials investigating the efficacy of ICIs for the treatment of metastatic PCCs: one is still in the recruitment stage, and the other showed 43% non-progression rate (NPR) and 0% overall response rate (ORR) in 7 PCC patients (Dai et al, 2020; Naing et al, 2020). The composition of immune cells in the tumor immune microenvironment is an essential indicator for predicting likely responses to immunotherapy, and serves as the basis of immunotherapy (Joyce & Fearon, 2015; Schumacher & Schreiber, 2015). However, there are no studies reporting the composition of immune cells in PCCs. The few published studies investigating the immune microenvironment of PCCs have been limited to the expression of PD-L1 at the histological level and to assessment of the tumor mutation burden (TMB) at the genomic level, and these results only seem to suggest that PCCs are immune-cold (Bratslavsky et al, 2019; Guo et al, 2019; Pinato et al, 2017). Therefore, deepening understanding of the immune cell composition in the tumor immune microenvironment should greatly aid in the effective application of immunotherapies to treat PCCs (Uher et al, 2021).

In the present study, we performed single-cell RNA sequencing (scRNA-seq) analysis of 11 tumor tissues and 5 adjacent tissues from 4 sporadic PCC patients with unclassified mutations and 1 hereditary PCC patient with Von Hippel-Lindau (VHL) syndrome caused by germline mutation in VHL. We found less intra-tumoral heterogeneity than inter-individual heterogeneity of PCCs, which is inconsistent with the PASS score prediction, thus highlighting limitations of the PASS scoring system. For the unclassified PCCs, we distinguished metabolism-type and kinase-type PCCs based on copy number variants (CNVs) and gene expression profile data. These two PCC types also displayed distinct characteristics of tumor evolution and cell-cell communication. Although multispectral immunofluorescence staining showed a lack of CD8+ T cell infiltration in both metabolism-type and kinase-type PCCs, only the kinase-type PCC exhibited downregulation of HLA-Ⅰ molecules that possibly regulated by RET, suggesting the potential of combined therapy with kinase inhibitors and immunotherapy for kinase-type PCCs, whereas the application of immunotherapy to metabolism-type PCCs (with antigen presentation ability) is likely unsuitable. Our study presents a single-cell transcriptomics-based molecular classification and microenvironment characterization of PCCs and provides clues for potential therapeutic strategies to treat PCCs.


A Landscape View of Cell Composition in PCCs

To characterize the molecular correlates of unclassified PCCs based on known prominent driver alterations, we performed single-cell RNA sequencing (scRNA-seq) on 16 collected specimens from 5 PCC patients (P1-P5). Four of these patients (P1-P4) suffered from sporadic PCCs: whole exome sequencing (WES) of these tumor tissues detected somatic mutations as compared with their adjacent tissues, including in JAK2, ARHGEF39, KMT2D, and MST1 (Supplemental Table S1); note that none of these fit into the three previously molecularly defined groups in the TCGA molecular taxonomy (Crona et al., 2017). The cohort also included 1 hereditary PCC patient (P5) with Type 2 Von Hippel-Lindau (VHL) syndrome, with a germline missense mutation in VHL (Supplemental Table S1).

We collected 1 resected tumor specimen from P1 (P1_T1), 3 specimens from distinct intra-tumoral sites from P2 (P2_T1, P2_T2, and P2_T3) and P5 (P5_T1, P5_T2, and P5_T3), as well as 2 specimens from distinct intra-tumoral sites from P3 (P3_T1 and P3_T2), and P4 (P4_T1 and P4_T2). To enable comparisons, one adrenal tissue specimen adjacent to the tumor was collected from each patient (P1_A, P2_A, P3_A, P4_A, and P5_A) (Fig. 1A). Tumor cell morphology was observed with microscopy of hematoxylin-eosin stained tumor tissues (Fig. S1), and immunocytochemistry staining results showed strong expression of Chromogranin A (CGA) in all tumor tissues, validating that the collected samples were PCC tumor specimens (Fig. S1). The histological features of samples were evaluated with the PASS scoring system for predicting the malignant potential of PCC patients. The PASS scores ranged from 2 to 9, showing obvious heterogeneity among the examined intra-tumoral sites and inter-individual comparisons for the 5 PCC patients (Supplemental Table S2).

Integration Analysis across 5 PCC Patients Revealing the Cell Composition of the PCC Microenvironment.

(A) Schematic of the experimental pipeline. 11 tumor specimens and 5 adjacent specimens were isolated from 5 PCC patients, dissociated into single-cell suspensions, and analyzed using 10x Genomics Chromium droplet scRNA-seq. (B) UMAP plots displaying 133,894 cells isolated from 16 specimens collected from 5 PCC patients. (B) UMAP plots of cells colored by specimens, patients, and tissue types. (C) UMAP plot showing 13 main cell types from all specimens. (D) Dot plot of representative marker genes for each cell type. The color scale represents the average marker gene expression level; dot size represents the percentage of cells expressing a given marker gene.

The specimens were digested into single cell suspensions, and 3’-scRNA-seq (Chromium Single Cell 3′ v3 Libraries) analysis was performed on each sample. After quality control filtering to remove cells with low gene detection and high mitochondrial gene coverage, we retained 133,894 individual cells (Fig. S2A-S2C). SCTansformed normalization and principal component analysis (PCA) of the Seurat suite(Butler et al, 2018; Satija et al, 2015; Stuart et al, 2019) were then employed for unsupervised dimensionality reduction prior to clustering. Uniform Manifold Approximation and Projection (UMAP) was used for visualizing the distinct specimens, patients, and tissue types (Fig. 1B and 1C).

We classified the 67 detected clusters into 13 cell types according to their expression profiles for recognized marker genes and Spearman correlation analysis between clusters: the classified cell types included adrenal cells (marked by DLK1 and RBP1), endothelial cells (marked by PECAM1 and VWF), fibroblasts (marked by THY1 and PLAC9), smooth muscle cells (marked by ACTA2 and TAGLN), monocytes/macrophages (marked by CD14 and CD163), neutrophils (marked by S100A8 and S100A9), T cells (marked by IL7R and CD3D), natural killer cells (NKs, marked by KLRD1 and GNLY), proliferating cells (marked by TOP2A and MKI67), plasma cells (marked by IGLC2 and IGHG1), B cells (marked by MS4A1 and CD79A), mast cells (marked by KIT and TPSB2), and sustentacular cells (marked by S100B and CRYAB) (Fig. 1C and 1D, Fig. S2D and S2E). Thus, our clustering and cell type annotation analysis identified diverse adrenal cells, stromal cells, and immune cells within the PCC microenvironment.

The PASS Score does not Informatively Reflect Heterogeneity as Inferred from the scRNA-seq Data

In light of the remarkable diversity of drivers including germline and somatic mutations reported among PCCs (Crona et al., 2017), we investigated PCC heterogeneity at the inter-individual and intra-tumoral levels based on the single-cell transcriptional profiles. A Spearman correlation analysis based on average gene expression in each specimen showed that the intra-tumoral heterogeneity was less than inter-individual heterogeneity (Fig. 2A), inconsistent with the PASS score prediction. We further analyzed the cell type composition in each specimen and found similar fractions of cell types between the collected intra-tumoral sites despite different PASS scores (Fig. 2B), indicating relatively small intra-tumoral differences in both gene expression and cell type composition.

Correlation Analysis Reveals Less intra-tumoral Heterogeneity than inter-individual Heterogeneity.

(A) Heatmap plotting the correlation coefficient among 16 specimens. The color keys from blue to red indicate the correlation coefficient from low to high. (B) Bar plot showing the percentage of cell types in 16 specimens. (C) Heatmap of the correlation coefficient among 5 PCC patients. The color keys from blue to red indicate the correlation coefficient from low to high. (D) Bar plot depicting the frequency distribution of cell types in 5 PCC patients.

The inter-individual correlation analysis indicated that P4 showed low correlation with the other three sporadic PCCs patients (P1-P3) (Fig. 2C), and further analysis of cell type fractions showed a significantly elevated proportion of adrenal cells (77%) in P4; the adrenal cells comprised between 16–36% of the total cells in the P1-P3 specimens (Fig. 2D). P4 also had an obviously distinct distribution of stromal cells vs immune cells (9% stromal cells and 14% immune cells) as compared to P1-P3 (28-62% stromal cells and 22-43% immune cells) (Fig. 2D). Consistent with the reported vascular endothelial hyperplasia phenotype in VHL syndrome (Vortmeyer et al, 2013), it was unsurprising that we observed poor correlation between the sporadic patients (P1-P4) and the VHL patient (P5), reflecting a higher proportion of endothelial cells in P5 (35%) as compared to P1-P4 (3-10%) (Fig. 2C and 2D).

Based on these observations, we proposed that there was a certain degree of limitation in the PASS scoring system that is based on the histologic observation of tumor tissues, which is consistent with the report that some patients with a low degree of malignancy based on the PASS scoring system exhibit invasion or metastasis (Agarwal et al., 2010; de Wailly et al, 2012; Wu et al., 2009), and the incorporation of PASS scoring system with gene expression profile might facilitate further understanding of PCC pathobiology and/or aid in the clinical evaluation of PCC.

Identification of Tumor Clusters through CNV analysis

To further characterize malignant chromaffin cells, we collected adrenal cells, which theoretically include adrenocortical cells and chromaffin cells (Yates et al, 2013). The copy number karyotyping of aneuploid tumors (CopyKAT) algorithm (Gao et al, 2021) was applied, which estimates genomic copy number variants (CNVs) from scRNA-seq data by employing an integrative Bayesian segmentation approach to distinguish malignant chromaffin cells from normal cells. Our analysis identified two clusters among adrenal cells (normal cells and tumor cells), with tumor cells showing extensive chromosomal amplification and/or deletion (Fig. 3A). Consistently, the inferred aneuploid tumor cells highly expressed Chromogranin B (CHGB), which is a marker of malignant chromaffin cells (Wiedenmann et al, 1988; Winkler & Fischer-Colbrie, 1992) (Fig. 3B and 3C). The inferred diploid normal cells expressing high levels of CYP17A1 and HSD3B2 were defined as adrenocortical cells (Kubota-Nakayama et al, 2016), while the inferred diploid normal cells showing low CHGB expression were defined as normal chromaffin cells (Fig. 3B and 3C). We also found a cluster of inferred diploid normal cells with high expression of VWF (a marker for vascular endothelial cells) in the VHL patient (P5) (Fig. 3B and 3C).

Single-cell Copy Number Profiles in PCC Clusters Inferred by CopyKAT.

(A) Heatmap indicating the CNV patterns of inferred normal cells, tumor cells, tumor cluster A, and tumor cluster B. Blue, white, and red respectively indicate deletion from a chromosome, normal chromosome, and amplification on a chromosome. (B) UMAP plot of the inferred normal cells (blue) and tumor cells (red) identified by CopyKAT. (C) Feature plots showing the marker gene expression levels in inferred normal cells and tumor cells. (D) UMAP plot depicting the inferred normal cells (blue), tumor cluster A (purple), and tumor cluster B (orange) identified by CopyKAT. (E) Feature plots displaying the marker gene expression levels in inferred tumor cluster A and B.

We noticed two CNV patterns among the inferred aneuploid tumor cells, with the distinction evident in chromosomes 3, 11, and 17 (Fig. 3A). These CNV profiles were reflected in distinct tumor clusters (tumor cluster A and B) through CopyKAT prediction (Fig. 3D). We observed that cells in inferred tumor cluster A exhibited high expression of NDUFA4L2 and COX4I2, while cells in inferred tumor cluster B expressed high levels of PNMT and RET (Fig. 3D and 3E), suggesting the different gene expression between tumor cluster A and B. PNMT is known as an enzyme that catalyzes the conversion of norepinephrine to epinephrine and that RET is a member of the receptor tyrosine kinase family, while NDUFA4L2 and COX4I2 respectively function in the oxidative phosphorylation and glycolysis pathways (Drilon et al, 2018; Mahmoodi et al, 2020; Sinkler et al, 2017). Taken together, we distinguished malignant chromaffin cells from adrenal cells through CNV analysis and found two clusters of malignant chromaffin cells with different single-cell copy number profiles.

Definition of Metabolism-type and Kinase-type tumors among unclassified PCCs

To explore the gene expression profile between tumor cluster A and B, we then used the FindAllMarkers function of the Seurat suite (Butler et al., 2018; Satija et al., 2015; Stuart et al., 2019) to identify differentially expressed genes (DEGs) of these tumor clusters (Fig. 4A and 4B). A functional enrichment analysis based on gene ontology (GO) annotation was performed to further characterize the biological function of tumor clusters: the tumor cluster A DEGs showed functional enrichment for terms including generation of metabolites and energy, response to oxygen levels, response to hypoxia, and ATP metabolic process and some notably upregulated genes of tumor cluster A included known energy metabolism-related genes such as NDUFA4L2, COX4I2, RGS4, and AQP1 (Fig. 4C). For tumor cluster B, the GO analysis showed enrichment for terms including cytoplasmic translation, neuron projection development, axon development, and cell growth, and there was obvious upregulation of PNMT, RET, and genes encoding other kinases known to function in proliferation, as well as genes that participate in the regulation of neuroendocrine functions (e.g., CALM1, PENK, and NPY) (Fig. 4D).

PCC Patients were Classified into Metabolism-type and Kinase-type.

(A) Heatmap plotting the scaled expression patterns of major marker genes in each tumor cluster. The color keys from pink to yellow indicate relative expression levels from low to high. (B) Dot plot depicting up-regulated genes of tumor cluster A (top) and tumor cluster B (bottom). The x-axis specifies the rank of DEGs and the y-axis specifies the natural logarithm of the FC. Dotted vertical and horizontal lines reflect the filtering criteria. Dot size represents the natural logarithm of the FC of genes. The color keys from blue to red indicate the P-value from high to low. (C, D) GO enrichment analysis of the up-regulated genes in tumor cluster A (C) and tumor cluster B (D) indicating the top altered 10 terms in the biological process of gene ontology. The x-axis specifies the number of genes enriched in the pathways. The color keys from shallow to deep indicate the P-value from high to low. (E) Immunohistochemistry staining of CGA, NDUFA4L2, COX4I2, PNMT, and RET markers in formalin-fixed paraffin-embedded PCC tissue sections matched to scRNA-seq specimens. Scale bar, 100 μm.

Notably, immunocytochemistry staining of serial tumor sections showed strong accumulation of PNMT and RET in P4, while NDUFA4L2 and COX4I2 accumulation was characteristic for the P1-P3 and P5 sections, providing protein-level evidence to validate the differential trends detected from the scRNA-seq results (Fig. 4E). As a consequence, we propose a metabolism-type (P1-P3) and kinase-type (P4) classification for unclassified sporadic PCCs. Consistently, kinase-signaling PCCs by TCGA classification include germline and somatic mutations in RET, and are uniquely able to secrete epinephrine, owing to their high expression of PNMT; P5 could also be classified into the metabolism-type we defined, which is known to display reprogramming of cellular energy metabolism caused by VHL mutation and subsequent HIF dysregulation (Chappell et al, 2019). These results support that the previously unclassified PCCs can be classified as metabolism-type or kinase-type based on transcriptional programs, with these types partially corresponding to the pseudo-hypoxic type and kinase-signaling type of PCCs from the TCGA classification.

Trajectory Analysis Reveals Subclonal Dynamics of PCC Types

We next examined tumor subclonal evolution patterns for the metabolism-type and kinase-type PCCs, by applying the Dynverse algorithm to order cells along a pseudotime trajectory (Saelens et al, 2019). We first clustered metabolism-type PCC cells into 2 subclusters based on DEGs and found similar proportions of tumor subcluster 1 and 2 at both the early stage and late stage of tumor evolution (Fig. 5A and 5B). Late up-regulated genes include S100A4 and TAGLN that promote epithelial-mesenchymal transition and cell invasion, which highly expressed in subcluster 2 (Fig. 5C), suggesting the higher malignant potential of subcluster 2 as compared with subcluster 1. Metabolism-related genes such as NDUFA4L2, ATP5MG, NDUFB4, and COX17 were highly expressed at early stage, while SOD3 and COX7A1 were up-regulated at the late stage (Fig. 5C), revealing the consistently active metabolism phenotype of metabolism-type PCCs cells. We also found the high expression of tumor suppressor gene SPINT2 at early stage and subsequent up-regulation of the oncogene JUNB (Fig. 5C), revealing an increase of malignant potential over the tumor evolution of metabolism-type PCCs. In short, these results indicate that cells from metabolism-type PCCs display phenotype of consistently active metabolism and increasingly malignant potential.

Pseudotime Analysis of PCC Tumor Evolution by Dynverse.

(A, B) Pseudotime trajectory of metabolism-type PCC cells colored by pseudotime (A) and tumor subclusters (B). (C) Pseudotime heatmap plotting the expression levels of genes across the transition from beginning (left) to end (right). The color keys from blue to red indicate the gene expression levels from low to high. (D, E) Pseudotime trajectory of kinase-type PCC cells colored by pseudotime (D) and tumor subclusters (E). (F) Pseudotime heatmap showing the expression levels of genes across the transition from beginning (left) to end (right). The color keys from blue to red indicate the gene expression levels from low to high.

We then clustered kinase-type PCC cells into 2 subclusters and found that the major clone has been changed from subcluster 1 to 2 along the trajectory (Fig. 5D and 5E). Further analysis of DEGs in the pseudotime trajectory showed early high expression of CALM1 and CALM2 in the subcluster 1, and late up-regulation of RET in the subcluster 2 (Fig. 5F), suggesting the persistent dysregulation of kinase signals over tumor evolution. We also noticed the early expression of PNMT, while neuron-specific genes such as MAP2, MAP1B, L1CAM, and CNTN1 were up-regulated later in the pseudotime trajectory (Fig. 5F), suggesting the decrease of epinephrine synthesis and the appearance of neuron-like phenotypes during tumor evolution, consistent with previously reported neuron-like phenotypes in PCC cells with high expression of RET (Califano et al, 1995; Powers et al, 2003; Powers et al, 2009). Taken together, these analyses have revealed the transcriptional clonal dynamics of metabolic-type and kinase-type PCCs.

Cellular Communication Analysis to Further Characterize the Tumor Microenvironment

Having characterized differences in the cell type composition, gene expression, and clonal evolution between metabolism-type and kinase-type PCCs, we subsequently expanded our investigation to cell-cell communication occurring within the distinct tumor microenvironments of these two types of PCCs. We adopted CellChat tool to quantitively analyze intercellular communication networks (Jin et al, 2021). We observed the activation of the annexin signaling pathway in metabolism-type PCCs (Fig. 6A). Further analysis of participated ligand-receptors showed major participation of ANXA1-FPR1 pair in the annexin signaling pathway (Fig. 6B), and high co-expression of ANXA1 and FPR1 in neutrophils and monocytes/macrophages (Fig. 6C). It has been reported that the annexin signaling pathway is known to be activated under a strong inflammatory reaction (Gastardelo et al, 2014; Gavins et al, 2003; Gerke & Moss, 2002; Hayhoe et al, 2006). Combined with our findings of a higher proportion of neutrophils and monocytes/macrophages in metabolism-type as compared with kinase-type (Fig. 2D), we speculate that metabolism-type PCCs may display elevated inflammatory responses. We then performed Gene Set Enrichment Analysis (GSEA) on metabolism-type and kinase-type PCCs by using hallmark gene sets in the MSigDB databases (Liberzon et al, 2011). As expected, the inflammatory response signaling pathway showed enrichment in metabolism-type PCCs as compared with that in kinase-type PCC (Fig. 6D). Taken together, metabolism-type PCCs exhibited a strong inflammatory reaction and the activation of annexin signaling pathway.

CellChat Analysis Reveals Cell-cell Communication Patterns in PCC Types.

(A) Circle plot showing an inferred annexin signaling pathway network in the metabolism-type PCC microenvironment. The edges connecting the circles represent the communication probability between any two kinds of cell types. The color of the edge denotes directionality (i.e., senders vs receivers). (B) Bar graph plotting the quantification of the relative contributions of individual ligand-receptor pairs to the overall annexin signaling pathway. (C) Violin plot of the expression distribution of ANXA1, FPR1, and FPR2 in main cell types of the metabolism-type PCC microenvironment. (D) GSEA enrichment plots of the inflammatory response signaling of metabolism-type patients (left) and the kinase-type patient (right). (E) Circle plot depicting an inferred VEGF signaling pathway. (F) Bar graph displaying the quantification of the relative contributions of individual ligand-receptor pairs to the overall VEGF communication network. (G) Violin plot plotting the expression distribution of VEGFA, VEGFB, PLGF, VEGFR1, and VEGFR2 in main cell types of P5 PCC microenvironment. (H) Circle plot of an inferred FGF signaling pathway in the kinase-type PCC microenvironment. (I) Bar graph showing the quantification of the relative contributions of individual ligand-receptor pairs to the overall FGF signaling pathway. (J) Violin plot plotting the expression distribution of FGF1, FGF18, and FGFR1 in main cell types of the kinase-type PCC microenvironment.

Considering the heterogeneity in cell type composition between sporadic and VHL patients in metabolism-type, we further analyzed cellular interactions in the VHL patient. We noticed the activation of VEGF signaling pathway in P5 (Fig. 6E), and major contribution of PLGF-VEGFR1, VEGFA-VEGFR1 and VEGFA-VEGFR2 pairs to VEGF signaling pathway (Fig. 6F). Specifically, VEGFA and VEGFR1 were highly expressed in tumor cells, VEGFR1 and VEGFR2 in endothelial cells, and PLGF in fibroblasts and smooth muscle cells, respectively (Fig. 6G), suggesting the crosstalk between tumor cells and fibroblasts or endothelial cells in tumor microenvironment through these ligand-receptor pairs, which mediated in the tumor angiogenesis and metastasis. The observed activation of VEGF pathway is consistent with the vascular endothelial hyperplasia caused by VHL mutation in VHL syndrome (Vortmeyer et al., 2013) and the elevated proportion of endothelial cells in this patient (Fig. 2D).

We subsequently analyzed the kinase-type PCC patient, and identified the activation of FGF signaling network (Fig. 6H) and major involvement of FGF1-FGFR1 in the FGF pathway (Fig. 6I). The smooth muscle cells, fibroblasts, and B cells exhibited high expression of FGF1, while tumor cells, smooth muscle cells, T cells, and B cells highly expressed FGFR1 (Fig. 6J). The analyses suggested the cell-cell communication among tumor cells, stroma cells, and immune cells through the FGF1-FGFR1 pair, which regulated the tumor cell proliferation (Katoh, 2016). These results indicate that kinase-type PCCs may benefit from FGFR inhibitors, which have been shown to exert inhibitory effects on tumor proliferation (Liang et al, 2012). Together, we described the cellular communication networks in microenvironment of metabolism-type and kinase-type PCCs.

Immune Infiltration in the PCC Microenvironment Implies Patient Responses to Immunotherapy

We further analyzed the immune microenvironment of metabolism-type and kinase-type PCCs. Visualization of cellular communication showed apparent crosstalk between tumor cells and T cells in metabolism-type patients, while we were surprised to find a lack of apparent communication between tumor cells and T cells in the kinase-type patient (Fig. 7A). These results suggested that kinase-type PCC cells could evade immune surveillance of T cells and further indicated the immune escape potential of kinase-type PCCs. Considering leukocyte antigen-class Ⅰ (HLA-Ⅰ) surface level is known to represent the antigen-presenting abilities of tumor cells (Brea et al, 2016; Jhunjhunwala et al, 2021; Oh et al, 2019), we assessed HLA-Ⅰ expression in PCC cells: the expression levels of HLA-A, HLA-B, and HLA-C were significantly lower in kinase-type PCC cells than in metabolism-type PCC cells (Fig. 7B). Consistently, immunohistochemistry staining of serial sections of tumor tissues showed lower expression of HLA-A in P4 as compared to that in other patients (Fig. 7C), providing a plausible explanation for the lack of interactions detected between tumor cells and T cells in the kinase-type PCC. In short, we discovered that the kinase-type PCC displayed an impaired HLA-Ⅰ expression profile consistent with immune escape, while metabolism-type PCCs showed antigen-presenting abilities.

Prediction of the Immune Escape Potential and Immunotherapy Response of PCC Patients.

(A) Circle plots depicting inferred cellular interactions in metabolism-type (left) and kinase-type (right) PCC microenvironment. (B) Box plots showing the expression levels of HLA-A, HLA-B, and HLA-C in metabolism-type and kinase-type PCC patients. (C) Immunohistochemistry staining of CGA and HLA-A in formalin-fixed paraffin-embedded PCC tissue sections matched to scRNA-seq specimens. Scale bar, 100 μm. (D) UMAP plot showing 8 immune cell types detected from all PCC specimens. (E) Dot plot of representative marker genes for each immune cell type. The color scale represents the average marker gene expression level; dot size represents the percentage of cells expressing a given marker gene. (F) Comparison of the proportion of immune cell types in tumor vs adjacent tissues. (G) Multispectral immunofluorescent staining for the juxtaposition of PCC cells (marked by CGA), CD4+ T cells (marked by CD4), CD8+ T cells (marked by CD8), M1 macrophages (marked by CD68), and M2 macrophages (marked by CD163) in formalin-fixed paraffin-embedded PCC tissue sections matched to scRNA-seq specimens. The white, yellow, cyan, green, pink, and blue spots indicated cells with high expression of CGA, CD4, CD8, CD68, CD163, and DAPI proteins in PCC tissue sections, respectively. S, stroma; T, tumor. Scale bar, 100 μm.

We additionally assessed the immune microenvironment of PCCs by subclustering immunotherapy-related immune cells and identified 8 clusters including central memory CD4+ T cells (CD4+ TCM, marked by CD3D, CD3E, CD3G, CD4, CCR7, and IL7R), effector memory CD4+ T cells (CD4+ TEM, marked by CD3D, CD3E, CD3G, CD4, and IL7R), CD8+ T cells (marked by CD3D, CD3E, CD3G, CD8A and CD8B), natural killer T cells (NKTs, marked by CD3E, KLRD1 and GNLY), natural killer cells (NKs, marked by KLRD1 and GNLY), B cells (marked by MS4A1 and CD79A), M1 macrophages (marked by CD68), and M2 macrophages (marked by CD68 and CD163) (Fig. 7D and 7E). Further analysis of immune cell type composition in metabolism-type and kinase-type PCCs showed an increase in the proportion of CD4+ T cells, CD8+ T cells, and M1 macrophages in tumor tissues as compared with that in adjacent tissues, while the proportion of M2 macrophages in adjacent tissues decreased as compared with that in tumor tissues (Fig. 7F).

The proportion and location of infiltrating immune cells in tumor tissues have been shown to be informative regarding responsivity to immunotherapy (Sade-Feldman et al, 2018; Stanton & Disis, 2016). In particular, the CD8+ T cells play an essential role in anti-tumor immunity; these cells can recognize tumor antigens displayed on the surface of tumor cells by HLA-Ⅰ molecules (Jhunjhunwala et al., 2021). As the scRNA-seq data do not support discernment of whether immune cells have infiltrated tumors or are merely present in the sampled materials, we performed multispectral immunofluorescence (mIF) staining on CD4+ T cells, CD8+ T cells, M1 macrophages, and M2 macrophages in formalin-fixed paraffin-embedded (FFPE) tumor tissues of P1-P5. We observed that CD8+ T cells, as well as CD4+ T cells, M1 macrophages, and M2 macrophages were rare (and were only present in tumor stroma of P1-P5), indicating apparent immune escape in both metabolism-type and kinase-type PCCs (Fig. 7G). Combined with previously reported negative regulatory effects of kinases (such as RET, ALK, and MEK) on HLA-Ⅰ expression on tumor cells (Brea et al., 2016; Oh et al., 2019), we speculate that the possible reason for inability in recruiting CD8+ T cells of kinase-type PCCs is the down-regulation of HLA-Ⅰ in tumor cells regulated by RET, while the mechanism of immune escape in metabolism-type PCCs (with antigen presentation ability) needs to be further explored. Our results also indicate that the application of immunotherapy to metabolism-type PCCs is likely unsuitable, while kinase-type PCCs may have the potential of combined therapy with kinase inhibitors and immunotherapy.


The genomic research of PCCs has made continuous progress in recent years (Bausch et al, 2017; Crona et al., 2017; Fishbein et al., 2017; Neumann et al, 2019), but there have been obvious breakthroughs for more efficacious therapies. Isotope therapies can only relieve the clinical symptoms in 50% of patients (without reducing tumor size) (Fitzgerald et al, 2006; Roman-Gonzalez & Jimenez, 2017), while most chemotherapies and targeted therapies exhibit low remission rates and severe side effects in clinical studies (Druce et al, 2009; Hamidi, 2019; O’Kane et al, 2019; Oh et al, 2012), limiting the development of PCC treatments. Although immunotherapy has achieved success in the therapy of solid tumors, it is difficult to evaluate the curative effect in PCCs. To address this problem, two key features associated with the tumor immunotherapy need to be elucidated—tumor heterogeneity and the tumor microenvironment (Hegde & Chen, 2020; Li et al, 2018; Zhang & Zhang, 2020). Regarding PCC heterogeneity, genomic studies have confirmed the inter-individual heterogeneity based on driver mutations (Dahia, 2014; Toledo et al., 2017). However, there is lack of relevant research on intra-tumoral heterogeneity, which has an impact on disease progression and sensitivity to immunotherapy (Vitale et al, 2021). For the PCC microenvironment, the cell composition and cellular crosstalk have not been explored. By investigating the heterogeneity and microenvironment characteristics of PCCs, we proposed a feasible method to judge the heterogeneity of PCCs, and also provided a clue for the application of immunotherapy in PCC patients.

The clinical and histopathological diagnosis of malignancy and heterogeneity of PCCs is particularly difficult, and is still limited by the lack of reliable prognostic markers (de Wailly et al., 2012). Although the PASS scoring system was proposed as a tool for discriminating potentially malignant PCCs from benign ones, it still cannot be reliably applied to PCCs for predicting malignancy or clinical prognostication, owing to observer variation and nonrepeatability (Agarwal et al., 2010; Kimura et al, 2014; Thompson, 2002; Wu et al., 2009). In order to investigate the association between PASS scoring system and heterogeneity of PCCs, we studied by scRNA-seq and observed less heterogeneity among different intra-tumoral sites than that among PCC patients, which is inconsistent with the prediction of the pathological PASS scoring system. Our findings suggest that the biopsy used to guide clinical treatment in the future may not need multi-point sampling, whereas and single-point sampling (while avoiding obvious necrotic areas) plus scRNA-seq analysis can informatively represent the characteristics of PCCs. Our findings also demonstrate that the PASS score is not suitable to reflect the intra-tumoral heterogeneity and inter-individual heterogeneity of PCCs, and helps explain the inconsistency between PASS score prediction and clinical outcomes in previous studies (Agarwal et al., 2010; Wu et al., 2009). Since the development of a reproducible modified scoring system for PCCs remains a challenging task (Wu et al., 2009), the combination of molecular diagnostic methods like single-cell sequencing with pathological tools might be easier to implement in the clinical practice of PCC diagnosis and care (Crona et al., 2017; Papathomas et al, 2021).

According to the results of genomics research in recent years, 40∼50% of PCCs have been classified into a certain molecular pathway (pseudohypoxia, kinase, or Wnt), but the remaining 50-60% are still unclassified (Crona et al., 2017; Lenders et al., 2014). PCCs that classified into these molecular pathways may have the opportunity to carry out drug clinical trials against corresponding targets, such as HIF2α, VEGF, and RET, etc. (Favier et al., 2015; Nölting et al, 2019; Toledo & Jimenez, 2018), while unclassified PCCs still have no clues or basis for drug treatment. According to the WES results of the 5 patients in our study, except for one patient with VHL germline mutation, the others could not be classified based on detected germline or somatic mutations. Therefore, we developed a new classification of PCCs based on scRNA-seq, which are able to accurately capture the transcriptional features of PCC cells. We found that PCC cells from 4 patients highly expressed metabolism-related genes, while that from another patient exhibited high expression of kinase-related genes. Although previous studies applying bulk RNA-seq analysis of PCCs have also reported highly expressed genes (Batchu et al, 2022; Flynn et al, 2015), the interpretation of bulk RNA-seq data can be complicated by interference from the large number of non-tumor cells (Potter, 2018; Suvà & Tirosh, 2019), so it is difficult to determine whether the highly expressed genes are reliable molecular features of PCCs. Although PCC cells from more patients need to be examined, we have defined at least two types of PCCs: metabolism-type and kinase-type.

One definition of the tumor immune microenvironment is the combination of the antigen presentation ability of tumor cells, the infiltration of immune cells, and the interaction between them (Joyce & Fearon, 2015; Schumacher & Schreiber, 2015). Although our immunofluorescence staining data showed a lack of CD8+ T cell infiltration in both metabolism-type and kinase-type PCCs, only kinase-type PCCs exhibited downregulation of HLA-Ⅰ molecules. The expression of HLA-Ⅰ is a marker for tumor antigen presentation and CD8+ T cell infiltration (Chowell et al, 2018; Jhunjhunwala et al., 2021; Perea et al, 2018). Previous studies have demonstrated that kinases such as RET, MAP2K1, ALK, and FGF can promote tumor growth via helping tumor cells to evade the immune system by downregulating HLA-Ⅰ expression (Brea et al., 2016; Oh et al., 2019). The kinase-type PCCs that we defined exhibited activation of RET and FGF signals, which led us to consider the possible inhibitory effect of kinases on HLA-Ⅰ expression in PCCs. Recently, RET inhibitors have been approved for the treatment of non-small cell lung cancer with RET mutation (Drilon et al, 2020). For kinase-type PCCs, whether RET inhibitors can enhance the antigen presentation ability of tumor cells, and further provide chance for combination of kinase-targeted therapy and immunotherapy, needs to be verified by follow-up research. The metabolism-type PCCs, however, with antigen presentation ability and also the lack of CD8+ T cell infiltration, the application of immunotherapy is likely unsuitable, and the mechanism(s) of immune escape in these tumors needs to be further explored.

There are potential limitations of our study. Despite the patient cohort in our study including a hereditary PCC patient, only one patient with a VHL germline mutation was included; other types of hereditary PCCs with germline mutations (for example in SDHx and FH) were not included. The number of sporadic cases was also limited, mainly caused by the low incidence of PCCs. In addition, the specimens were all from primary tumors, while no metastatic tumors were included. Together, these considerations underscore the necessary for additional studies of PCCs in the future.

In summary, our study presents the molecular classification and tumor microenvironment characterization of PCCs through scRNA-seq. Our identification of less intra-tumoral heterogeneity than inter-individual heterogeneity of PCCs is inconsistent with the PASS score prediction and suggests the limitation of the PASS scoring system. Among unclassified PCCs, we defined metabolism-type and kinase-type at the single-cell transcriptome level. We observed a lack of CD8+ T cell infiltration in both metabolism-type and kinase-type PCCs. The kinase-type PCC showed downregulation of HLA-Ⅰ that possibly regulated by RET, suggesting the potential of combined therapy with kinase inhibitors and immunotherapy in kinase-type PCCs. For the metabolism-type PCCs, which have antigen presentation ability and also exhibit immune escape, the application of immunotherapy is likely unsuitable. The proposal of this molecular classification and our characterization of these PCC types can contribute to strategy development and clinical trials of PCC treatments in the future.


Ethical Regulations

The research presented here complies with all relevant local, national, and international regulations. For all PCC patient specimens, informed written consent was obtained prior to donation. The Peking University First Hospital Review Board (Protocol 300-001) approved the study.

Patient Cohort

Five PCC patients were included, and all patients had signed the consent forms at the Department of Urology in Peking University First Hospital. Fresh tumor specimens were collected during surgical resection. The sporadic patients (P1-P4) were performed surgical resections of the tumors at right adrenals, and the VHL patient (P5) underwent left adrenalectomy. We collected 1 resected tumor specimen from P1 (P1_T1), 3 specimens from distinct intra-tumoral sites from P2 (P2_T1, P2_T2, and P2_T3) or P5 (P5_T1, P5_T2, and P5_T3), and 2 specimens from distinct intra-tumoral sites from P3 (P3_T1 and P3_T2), or P4 (P4_T1 and P4_T2). To enable comparisons, one adrenal tissue specimen adjacent to the tumor was collected from each patient (P1_A, P2_A, P3_A, P4_A, and P5_A). A total of sixteen specimens were carefully dissected under the microscope and confirmed by a qualified pathologist.

Tissue Processing

The fresh tumor specimens were stored in the tissue preservation solution (JSENB) and washed with Hanks Balanced Salt Solution (HBSS, HyClone) for 3 times and minced into 1-2 mm pieces. Then the tissue pieces were digested with 2 ml tissue dissociation solution at 37℃ for 15 min in 15 ml centrifuge tube with sustained agitation. After digestion, using 40 μm sterile strainers to filter the samples and centrifuging the samples at 1,000 rpm for 5 min. Then the supernatant was discarded, and the sediment was resuspended in 1 ml phosphate buffered saline (PBS, HyClone). To remove the red blood cells, 2 ml red blood cell lysis buffer (BD) was added at 25°C for 10 min. The solution was then centrifuged at 500 g for 5 min and suspended in PBS. The cells were stained with trypan blue (Sigma) and microscopically evaluated.

Single-cell Library Construction and Sequencing

Utilizing the 10x Genomics Chromium Single Cell 3′ v3 Library Kit and Chromium instrument, approximately 17,500 cells were partitioned into nanoliter droplets to achieve single-cell resolution for a maximum of 10,000 individual cells per sample. The resulting cDNA was tagged with a common 16 nt cell barcode and 10 nt Unique Molecular Identifier during the RT reaction. Full-length cDNA from poly-A mRNA transcripts was enzymatically fragmented and size selected to optimize the cDNA amplicon size (approximately 400 bp) for library construction (10x Genomics). The concentration of the 10x single-cell library was accurately determined through qPCR (Kapa Biosystems) to produce cluster counts appropriate for the HiSeq4000 or NovaSeq6000 platform (Illumina). In all, 26 × 98 bp (3′ v3 libraries) sequence data were generated targeting between 25 and 50K read pairs/cell, which provided digital gene expression profiles for each individual cell.

General scRNA-seq Data Analysis

For single-cell RNA-seq analysis, we used CellRanger (10x Genomics, v.6.1.2) to pre-process the single-cell RNA-seq data after obtaining the paired-end raw reads. Cell barcodes and unique molecular identifiers (UMIs) of the library were extracted from read 1. Then, the reads were split according to their cell (barcode) IDs, and the UMI sequences from read 2 were simultaneously recorded for each cell. Quality control on these raw readings was subsequently performed to eliminate adapter contamination, duplicates, and low-quality bases. After filtering barcodes and low-quality readings that were not related to cells, we mapped the cleaned readings to the human genome (GRCh38) and retained the uniquely mapped readings for UMIs counts. Next, we estimated the accurate molecular counts and generated a UMI count matrix for each cell by counting UMIs for each sample. Finally, we generated a gene-barcode matrix that showed the barcoded cells and gene expression counts.

The R package Seurat (v.4.0.2) was used for all subsequent analysis (Butler et al., 2018; Satija et al., 2015; Stuart et al., 2019). For quality control of single-cell RNA-seq, a series of quality filters was applied to the data to remove those barcodes which fell into any one of these categories: too few genes expressed (possible debris), too many UMIs associated (possible more than one cell), and too high mitochondrial gene expression (possible dead cell). The cutoffs for these filters were as recommended by the Seurat package. Low-quality cells and outliers were discarded, and the single cells that passed the QC criteria were used for downstream analyses.

Cell Clustering and Cell Type Annotation

The Seurat software package (v.4.0.2) was used to perform cell clustering analysis to identify major cell types (Butler et al., 2018; Satija et al., 2015; Stuart et al., 2019). All Seurat objects constructed from the filtered UMI-based gene expression matrixes of given samples were merged. We first applied the SCTransform (v.0.3.2) function to implement normalization, variance stabilization, and feature selection through a regularized negative binomial model. Then, the principal component analysis (PCA) was applied for linear dimensionality reduction with the top 3000 variable genes. According to standard steps implemented in Seurat, highly variable numbers of principal components (PCs) 1-50 were selected and used for clustering using the Uniform Manifold Approximation and Projection (UMAP) method. We identified cell types of these clusters based on the expression of canonic cell type markers or inferred by CellMarker database (Zhang et al, 2019). The cluster markers were also certified using the FindAllMarkers function of the Seurat suite, and cell types were manually annotated based on certified markers finally.

Correlation Analysis of scRNA-seq

After integration, for each cluster, each sample, and each patient, we compared the gene expression to others to identify the significant highly expressed genes (adjusted P-value < 0.05 and log fold change > 0). Then the average gene expressions in each cluster, each sample, and each patient were calculated. The pairwise correlations were then estimated.

DEGs Identification and Enrichment Analysis

The cluster-specific genes were identified by running Seurat containing the function of FindAllMarkers on a log-transformed expression matrix (min. pct = 0. 25, only. Pos = TRUE, and logfc.threshold = 0.25). We also identified the differentially over-expressed genes between two clusters with the Wilcoxon Rank-Sum Test with the FindMarkers function in Seurat (adjusted P-value < 0.05, only.pos = T, and logfc.threshold = 0.1), and the cluster-specific overrepresented GO biological process was calculated with the compareCluster function in the clusterProfiler package (v.4.2.2) of R (Yu et al, 2012). We also used the GSEA with the curated gene sets to identify the pathways that were induced or repressed in between the cell clusters. In brief, the mean gene expression level was calculated and the log twofold change (FC) between the specific cell cluster and the other cells was applied as the test statistic. The 50 hallmark gene sets in the MSigDB databases ( were used for the GSEA analysis (Liberzon et al., 2011).

Single-cell RNA CNV Detection and Clustering

All cells were classified as either normal or tumor based on the genome-wide copy number profiles computed from the gene expression UMI matrix using the Bayesian segmentation approach, CopyKat (v.1.0.8) (Gao et al., 2021). Aneuploid single cells with genome-wide copy number aberrations were predicted to be tumor cells. Diploid cells were predicted to be normal cells. The CopyKat-based predictions were further confirmed by single-cell gene expression profiles.

Trajectory Analysis of scRNA-Seq Data

In order to perform a detailed comparison among different trajectory modeling tools, the Dynverse (v.0.1.2) tool was used (Saelens et al., 2019). Based on the scoring system provided by Dynverse and following a careful inspection of the generated trajectories, we applied ‘SCORPIUS’ and ‘EMBEDDR’ to infer the development trajectories. Finally, we selected the genes that were differentially expressed on different stages through the trajectory and plotted the pseudotime heatmap.

Cellular Communication Analysis

Cell-cell interactions based on the expression of known ligand-receptor pairs in different cell types were inferred using CellChatDB (v.1.1.3) (Jin et al., 2021). In brief, we followed the official workflow and loaded the normalized counts into CellChat and applied the preprocessing functions ‘identifyOverExpressedGenes’, ‘identifyOverExpressedInteractions’, and ‘projectData’ with standard parameters set. As database we selected the ‘Secreted Signaling’ pathways and used the pre-compiled human ‘Protein-Protein-Interactions’ as a priori network information. For the main analyses the core functions ‘computeCommunProb’, ‘computeCommunProbPathway’, and ‘aggregateNet’ were applied using standard parameters and fixed randomization seeds. Finally, to determine the senders and receivers in the network, the function ‘netAnalysis_signalingRole’ was applied on the ‘netP’ data slot.

Whole Exome Sequencing and Analysis

Genomic DNA extracted from tumor tissues were sent for whole exome sequencing. The exomes were captured using the Agilent SureSelect Human All Exon V6 Kit and the enriched exome libraries were constructed and sequenced on the Illumina NovaSeq 6000 platform to generate WES data (150 bp paired-end reads, > 100×) according to standard manufacturer protocols. The cleaned reads were aligned to the human reference genome sequence UCSC Build 19 (hg19) using Burrows-Wheeler Aligner (BWA) (v.0.7.12) (Li & Durbin, 2009). All aligned BAM were then performed through the same bioinformatics pipeline according to GATK Best Practices (v.3.8) (McKenna et al, 2010). We obtained germline variants based on variant calling from GATK-HaplotypeCaller. We then used GATK-MuT ect2 to call somatic variants in tumors and obtained a high-confidence mutation set after rigorous filtering by GATK-FilterMutectCalls. All variants were annotated using ANNOVAR (v.2018Apr16) (Wang et al, 2010).

Immunocytochemistry and Multispectral Immunofluorescent Staining

Immunocytochemistry and multispectral immunofluorescent staining experiments were conducted according to standard protocols using antibodies against formalin-fixed paraffin-embedded (FFPE) tissue specimens. The antibodies used are listed as follows: CGA (ABclonal, A9576), NDUFA4L2 (Proteintech, 66050-1-lg), COX4I2 (Santa, sc-100522), PNMT (Abcam, ab154282), RET (Abcam, ab134100), CD4 (Abcam, ab133616), CD8 (Invitrogen, MA1-80231), CD68 (Invitrogen, MA5-12407), CD163 (Abcam, ab182422), and HLA-A (ABclonal, A11406).

Data availability

Fastq raw data and processed data can be requested by contacting the corresponding authors. Human transcriptome reference used for our analysis is available at 10x Genomics website ( or in zenodo (

Code availability

The single-cell RNA sequencing data was processed using cellranger v.6.1.2 ( and analyzed with the R package Seurat v.4.0.2 ( Custom code that was created for scRNA-seq analysis can be requested by contacting the corresponding authors.


This work was supported by grant from Ministry of Science and Technology of China (No. 2021YFA1300603 to L.S.), grant (Z200020 to L.S.) from the Natural Science Foundation of Beijing, grant (81874158 to L.S.) from the National Natural Science Foundation of China, and grant (2022CX08 to Z.Z.) from National High Level Hospital Clinical Research Funding. We thank Dr. Jiaqian Wang (YuCe Biotech Co., Ltd) for provided technical support.

Author contributions

Z.Z., L.S., and K.G. conceived the idea and designed the study. Y.X. prepared samples for scRNA-seq. S.Q. performed scRNA-seq data analysis. Q.S., S.F., W.A., Y.X., and W.H. performed immunocytochemistry and multispectral immunofluorescent staining. S.Y., K.Z., and X.Z. provided technical assistance. Z.Z., L.S., and Y.W. provided guidance in data analysis and interpretation of the results. S.Q. drafted the manuscript, with significant input from Z.Z., L.S., and K.G. All authors reviewed the manuscript prior to submission.


The authors declare no conflicts of interest that pertain to this work.