Genomic Characteristics and Single-Cell Profiles After Immunotherapy in Fumarate Hydratase-Deficient Renal Cell Carcinoma

Fumarate hydratase (FH)-deficient renal cell carcinoma (FHRCC) is a rare type of renal cell carcinoma (RCC; ref. 1). Historically, only hereditary leiomyomatosis and renal cell cancer syndrome (HLRCC) have been recognized as pathologic subtypes owing to their typical presentation and distinguishable histological features (2). Advances in genetic testing have revealed that FHRCC can arise from germline mutations with a typical HLRCC presentation or biallelic somatic loss of FH gene (3). FH deficiency in IHC is the key diagnostic criterion for FHRCC due to the lack of specificity in morphologic features (4). FHRCC has poor prognosis, with a median survival of 18 to 24 months in the advanced stages (1, 3, 5). Current therapeutic exploration focuses primarily on the combination of vascular endothelial growth factor (VEGF) or mTOR/VEGFR therapy, and one phase II trial reported remarkable results with the combination of bevacizumab and erlotinib (1, 5, 6). Meanwhile, ICI, which changes the treatment landscape of clear cell RCC, is scarcely investigated in FHRCC.

A recent study suggested that FHRCC is an immunogenic subtype characterized by high levels of tumor-infiltrating lymphocytes, PD-L1 expression, and activated chemotaxis signaling. Clinical benefit from ICI-based treatment was noticed in this cohort (3). Given these intriguing results and that no more than 10 patients received ICI-based treatment in prior studies, this study aimed to comprehensively analyze the genomic characteristics and immune microenvironment of a larger FHRCC cohort receiving ICI-based treatment and elucidate the reason for the heterogeneous response at the single-cell level.

This was a retrospective review of the institutional database of patients diagnosed with RCC with a non–clear cell morphologic subtype between 2012 and 2021. A total of 30 patients with FHRCC were identified through pathology reports; of them, 76.7% (23/30), 13.3% (4/30), and 10.0% (3/30) had papillary, tubular, and tubular cystic, nested with adenoid, respectively. All cases were reviewed by experienced uropathologists and confirmed using genetic testing. All patients in the cohort underwent whole-exome sequencing (WES), and four patients underwent single-cell RNA sequencing after ICI-based treatment (Fig. 1A). Fresh tissue samples for single-cell RNA sequencing were obtained from the biopsies. Three were located in the lymph nodes, and one was located on the abdominal wall. A total of 16 formalin-fixed paraffin-embedded (FFPE) tissue blocks sliced into 5-μm-thick sections were collected for IHC and analysis.

This study was approved by the Ethics Review Board of Sun Yat-sen University Cancer Center (Project no.: B2022-034-01) and was performed in accordance with the ethical standards of the Helsinki Declaration (7). Written informed consent was obtained from all patients.

Tumor DNA from FFPE sections was extracted using the QIAamp DNA FFPE Tissue Kit (Qiagen), and genomic DNA was extracted from white blood cells (WBC) using the Blood Genomic DNA Mini Kit (CWBiotech). Sequencing libraries were prepared using the KAPA Hyper Prep Kit (KAPA Biosystems) according to the manufacturer's instructions. Pools of four to six libraries were used to hybridize to the capture panel with 16 hours at 47/65°C (Nimblegen/IDT). Libraries were quantified using a Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific). The final libraries were sequenced on an Illumina Nextseq500 (PE 75). For exome sequencing, we used the Integrated DNA Technologies xGen Exome Research Panel v1.0 (8). Germline variants, called the Genome Analysis Tool Kit, from WBC samples were first filtered with a threshold of minimum coverage of 50× and allele frequency of over 30%. Variants, noncoding regions, and synonymous mutations were filtered out. Similar filtering rules were applied for somatic mutations from FFPE samples, and mutation frequency was required to be eight times higher than those in the WBC sample control. All SNVs/indels were annotated with ANNOVAR, and each SNV/indel was manually checked using Integrative Genomics Viewer.

CellRanger v6.0.2 software was used to align raw fastq files to GRCh38 human genome reference seq and obtain single-cell raw gene expression matrices. Raw gene expression matrices were imported and processed using the Seurat R package (version 3.1.5; ref. 9). Low-quality cells, defined as follows, were removed: (i) cells with >2,500 or <200 expressed genes, (ii) mitochondrial count ratio >0.2, and (iii) cells with obvious co-expression of two types of marker genes, T cells, and B cells. Each patient's seurat object was merged into one, and batch effects were minimized using the Harmony R package (version 1.0; ref. 10). Gene expression matrix of the remaining high-quality cells were normalized to the total cellular UMI counts with “log” method, and the normalized expression was scaled (scale.factor = 10,000). Highly variable genes (top 2,000) were extracted using the Seurat FindVariableGenes function by “mvp” method, and the mean cutoff value ranged from 0.1 to infinity, and the dispersion cutoff value ranged from 0.5 to infinity. Principal component analysis was performed on the highly variable genes. Clusters were found using the FindNeighbors and FindClusters functions (dims.use = 1:15, resolution = 0.5). Uniform manifold approximation and projection (UMAP) was used for dimension reduction and visualization of every cell (11).

The cell lineage trajectory of CD8⁺ T cells was inferred using the Monocle2 R package (12). The “differentialGeneTest” function was used to derive differentially expressed genes from each cluster, and genes with a P value <0.01 were used to order the cells in pseudotime analysis. After constructing the cell trajectories, genes that varied according to pseudotime were identified using the “differentialGeneTest” function in Monocle2 and were used to perform Gene Ontology (GO) enrichment analysis. The average expression [measured by log2 (TPM + 1)] of pro-apoptosis–related genes were used to define the apoptosis score for CD8⁺ T cells. The pro-apoptotic genes are listed in Supplementary Table S1. Glycolysis-related genes were previously described by Sun and colleagues (13).

Genes with frequency greater than 10% in WES were chosen for IHC staining. Because antibodies to ANKRD11, KMT2B, BCORL1, and NCOR2 were not available at that time, IHC was performed on KMT2C, ARID1A, PTCH1, ATM, ZFHX3, and NOTCH3. Two uropathologists independently evaluated the samples. The staining intensity was assessed as follows: 0 (negative staining), 1 (weak staining), 2 (moderate staining), and 3 (strong staining). Percentage scores were assigned as follows: 1 (1%–25%), 2 (26%–50%), 3 (51%–75%), and 4 (76%–100%). The intensity and percentage scores assigned to each sample were multiplied to give a final score of 0 to 12. Positive PD-L1 expression was scored using the tumor proportion score (TPS). Seven samples after ICI-based treatment were stained with CD8A⁺ followed by PD-1 antibodies to verify exhausted CD8⁺ T cells. Antibodies used were as follows: anti-KMT2C (Bioss, 1:600 dilution), anti-ARID1A (Abcam, 1:200 dilution), anti-PTCH1 (Abcam, 1:500 dilution), anti-ATM (Abcam, 1:2,000 dilution), anti-ZFHX3 (Thermo Fisher Scientific, 1:1,000 dilution), anti-NOTCH3 (Abcam, 1:100 dilution), anti-PD-L1 (Cell Signaling Technology, 1:100 dilution), anti-FOXP3 (Abcam, 1:100 dilution), anti-CD8 (Ascend Biology, 1:100 dilution), and anti-PD-1 (Ascend Biology, 1:100 dilution).

mIF was performed using a PANO 7-plex IHC kit (catalog no. 0004100100; Panovue) according to the manufacturer's protocol. The average density (cells/mm²) of CD8⁺ T cells, PD-L1⁺ T cells, PD-1⁺ T cells, and Foxp3⁺ Treg cells was visualized with PanoVIEW VS200 and counted using InForm image analysis software (Version 2.4; PerkinElmer).

The patients were followed up every 3 months. Treatment responses were evaluated according to the RECIST v1.1 (14). Objective response rate (ORR) was defined as complete response (CR) or partial response (PR), and the disease control rate (DCR) was defined as achieving CR, PR, or stable disease. Progression-free survival (PFS) was calculated from treatment initiation to disease progression, whereas overall survival (OS) was defined as the time from initial metastasis to the last follow-up or death.

Comparisons were performed using the Wilcoxon test. Correlation was performed using the Spearman rank correlation analysis. Survival curves were constructed using the Kaplan–Meier method and were compared using the log-rank test. All data are presented as mean ± SD. All statistical analyses were performed using R (http://www.rproject.org) and SPSS version 25.0 (SPSS Inc.). Statistical significance was set at P values less than 0.05.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

In total, 21 of the 30 patients (70%) had germline mutations and 9 had somatic mutations. The baseline patient characteristics are summarized in Supplementary Table S2. The median age at initial diagnosis was 36 years (range, 16–63 years), and the male-to-female ratio was 2.75:1. Male patients tended to be older at the time of initial diagnosis (P = 0.009). In total, 29 of the 30 patients (96.7%) presented with metastases, including 15 (50.0%) and 14 (46.7%) patients with synchronous and metachronous metastases, respectively. The most common metastases were in the lymph nodes (15/29, 51.7%) and bones (9/29, 31.0%).

Four patients received single-cell RNA sequencing after ICI-based treatment. Patient 1 (P04) was a 23-year-old male who presented with stage IV disease involving bilateral kidneys, regional, and neck lymph nodes (LN). After progression on sunitinib, he switched to ICI with axitinib with 23 months of PR. His neck LNs were sampled after 13 months of ICI-based treatment. Patient 2 (P12) was a 29-year-old male with synchronous metastases of bone, liver, and LNs. After 3 months of pazopanib, he insisted on the addition of ICI. A biopsy of left supraclavicular lymph node metastases was performed after 5 months of progressive disease (PD) of ICI-based treatment. Patient 3 (P05) was a 46-year-old male who underwent left-sided radical nephrectomy and developed liver metastases 10 months after surgery. After 8 months of sunitinib, he experienced progression in retroperitoneal LNs and liver. He developed abdominal wall metastasis after 15 months of ICI plus axitinib, which was resected for single-cell RNA sequencing. Patient 4 (P01) was a 49-year-old male patient who developed vertebral oligometastasis 2 months after radical nephrectomy. After multiple lines of tyrosine kinase inhibitor (TKI) failure, he received ICI with envatinib for bone, liver, and LN polymetastases. He progressed only after 2 months of ICI-based treatment, so neck LN was sampled after 8 months of ICI-based treatment.

The median TMB was 0.14 mutations per megabase (interquartile range: 0.02–1.18 mutations per megabase), suggesting that TMB-low may be common in FHRCC, and the survival benefit could not be compared between the TMB-high and TMB-low groups. The common mutant genes apart from FH were KMT2C (8/30, 26.7%), ARID1A (6/30, 20%), PTCH1 (6/30, 20%), ANKRD11 (5/30, 16.7%), BCORL1 (4/30, 13.3%), KMT2B (4/30, 13.3%), NCOR2 (4/30, 13.3%), ATM (4/30, 13.3%), ZFHX3 (4/30, 13.3%), and NOTCH3 (4/30, 13.3%; Fig. 1B). A total of 19 FH gene mutations were identified (Fig. 1C), including 10 missense mutations, 5 frameshift mutations, and 4 nonsense mutations. The most common mutation locations of the FH gene were c.698G>A and c.563A>T. In addition, two patients (P25 and P26) were detected intron splicing mutations of FH gene and one patient (P23) was detected copy-number variant. In the gene set enrichment analysis, the most frequent mutations occurred in pathways of transcription factors (11/30, 36.7%), genome integrity (10/30, 30%), and chromatin SWI/SNF complex (8/30, 26.7%; Fig. 1D). The association between treatment response and genetic mutation were analyzed in the top three mutated genes. Compared with targeted therapy, nonfrontline ICI-based treatment in KMT2C (P = 0.028), ARID1A (P = 0.014), and PTCH1 (P = 0.083) mutant cohorts achieved better DCR (Fig. 2A). No significant difference was observed in KMT2C and ARID1A wild-type cohorts (Supplementary Fig. S1A). Meanwhile, there was no significant difference in PFS between the different gene types, possibly owing to the limited sample size (Supplementary Figs. S1B–S1G).

Using IHC, we explored tumor-infiltrating levels of CD8⁺ T cells and PD-L1, as well as the expression levels of KMT2C, ARID1A, PTCH1, ATM, ZFHX3, and NOTCH3 based on the results of WES and available antibodies. KMT2C was negative and ARID1A was strongly positive in all samples. The median H scores of PTCH1 and ATM were 6 and 10, respectively. The median TPS was 15% for PD-L1, and the median density of tumor-infiltrating CD8⁺ T cells reached 113/mm² [extensive CD8⁺ TIL (≥100 mm²): 9/16, 56.3%]. On mIF, both PD-1 and FOXP3 were negative. However, extensive CD8⁺ T cells (median: 159/mm²) and a moderate TPS of PD-L1 (median: 13.3%) were found in the tumor areas (Fig. 2B). The percentage of tumor-infiltrating CD8⁺ T cells was negatively correlated with ATM expression level (R = −0.624, P = 0.01; Fig. 2C and D). The ATM-high group showed better tumor response to non-frontline ICI-based treatment than did the ATM-low group (DCR: 100% vs. 50%; P = 0.035), whereas the percentage of CD8⁺ T cells did not correlate with the tumor response to ICI-based treatment (Fig. 2E and F). In addition, no correlation was observed between the PD-L1 expression and the response to ICI-based treatment (Fig. 2G).

After quality control and removal of the batch effect between the samples, 32,479 single cells were clustered into 19 major clusters using the uniform manifold approximation and projection method. Cluster-specific genes were used to annotate cell types using classic markers described in previous studies (9, 11, 12, 15–22). We identified the following: (i) one cluster of epithelial cells (CRYAB+ and SPP1+); (ii) six subtypes of T cells (namely, exhausted T cells (CD8A+ and PDCD1+), cytotoxicity T cells (CD8A⁺, GZMK⁺, and PDCD1⁻), T CD4 naïve memory (CD3D⁺ and CCR7⁺), Treg (CD4⁺, FOXP3⁺), T-RSAD2 (CD3D⁺ and RSAD2⁺), and T-TRM (CD3D⁺ and RSAD2⁺); (iii) NK cells (KLRC1⁺ and GNLY⁺); (iv) monocytes (CD68⁺ and S100A8⁺); (v) three subclusters of B cells (CD79A⁺ and MS4A1⁺; namely, B naïve memory (CD68⁺ and S100A8⁺), B-RSAD2 and B plasma, and mast cells (TPSAB1⁺ and TPSB2⁺); (vi) fibroblasts (PLA2G2A⁺ and CXCL14⁺); and (vii) vascular endothelial (VWF⁺ and PLVAP⁺) and profiling cells (TOP2A⁺ and MKI67+; Fig. 3A–C; Supplementary Table S3). T cells were the most abundant in the TME. Notably, exhausted T cells were present only in PD patients (Fig. 3D). We further validated the finding in seven cases of FHRCC using IHC. The percentage of exhausted CD8⁺ T cells remained significantly higher in PD patients than that in PR patients (Fig. 3E), similar results also observed in a recent clear cell renal cell carcinoma study (23). Interestingly, we found a cluster of fibroblasts in PR patients with high expression of PLA2G2A and CXCL14 (Fig. 3C), which was reported to be a novel fibroblast subtype associated with poor prognosis but better immunotherapy response in pancreatic ductal adenocarcinoma (24). This subtype of fibroblast deserves further investigation.

We explored the dynamic immune states and gene expression of FHRCC CD8⁺ T-cell infiltration by inferring state trajectories using Monocle (12, 25). The analysis showed that the CD8 RSAD2 cells were at the beginning of the trajectory path, whereas the CD8 TRM cells and most exhausted CD8⁺ T cells were at the terminal state (Fig. 4A and B). We then performed clustering of genes with a pseudotemporal expression pattern, whereby the ordering of genes was clustered into five clusters. Genes in cluster 3, including FASLG, CASP9, PKM, and PDCD1, were highly expressed at the end stage, and GO analysis suggested that their function was mainly enriched in the pathways of negative regulation of immune system process, intrinsic apoptotic signaling pathway, T-cell apoptotic process, and glycolysis (Fig. 4C). Genes in cluster 2, including CCR6, IL7R, and ANXA1, tended to be downregulated during pseudotime. GO enrichment analysis showed that their functions were enriched in a series of immune activation functions, such as T-cell activation, T-cell differentiation, T-cell differentiation involved in immune response, regulation of immune effector processes, positive regulation of lymphocyte proliferation, and positive regulation of adaptive immune response (Fig. 4C).

Analysis of energy metabolism pathways showed that glycolysis-related genes were gradually upregulated during pseudotime (Fig. 4D). Apoptosis-related genes, including CASP9 and FASLG, were upregulated during pseudotime, indicating an increased apoptotic trend of CD8⁺ T cells in FHRCC (Fig. 4E). Importantly, although the patients were exposed to ICI-based treatment before biopsy, the exhausted marker genes (e.g., PDCD1, LAG3, TIGIT, CTLA4, and HAVER2) and related exhausted score were gradually upregulated. Further, the T-cell exhaustion–related transcription factors, including TOX, TOX2, HOPX ARNT, ETV1, and EOMES, were upregulated during the trajectory, further confirming the exhausted state of these cells (Fig. 4F and G). In summary, CD8⁺ T cells were progressively dysfunctional with gradual upregulation of exhausted markers and an increase in apoptotic trend, indicating an immune tolerance status during pseudotime trajectories in FHRCC. Similar findings were reported in ICI therapy exposed breast cancer and lung cancer using pre- and posttreatment matched samples (26, 27). The expression of exhaustion-related genes of CD8⁺ T cells gradually increased during pseudotime trajectories.

Complex cellular responses are initiated by ligand-receptor binding and the subsequent activation of specific signaling pathways. We conducted an analysis of cell–cell communication using CellChat (28), which could quantitatively characterize and compare the inferred cell–cell communication based on the average expression of the ligands and receptors in cell populations. As the exhausted CD8⁺ T cells were enriched in patients with PD only, we analyzed cell–cell communication in PR and PD groups. Besides, we performed cell–cell communication analysis of each patient to avoid inter-individual differences. To eliminate the influence of different cell numbers on the communication intensity between PD and PR patients, stratified and random sampling was utilized to select the same number of PR patients from each sample and cell types of PD patients for analysis.

Exhausted CD8⁺ T cells and cytotoxic CD8⁺ T cells in PD and PR patients played a dominant role in cell–cell communication, showing the greatest incoming and outgoing interaction strength (Fig. 5A; Supplementary Fig. S2A). Cells from PD patients had a greater number of possible inferred interactions. Average interaction strength, which assessed the probability of observing an interaction given the expression levels of the receptor and its ligand, was smaller in PR patients (Fig. 5B; Supplementary Fig. S2B). PD and PR patients showed a different cell–cell communication pattern. The pathways enriched in PD patients included: VEGFR and TIGIT for VEGF, FASLG for PVR and FAS, and CD45 for LGALS9. Meanwhile, PD patients showed decreased signaling pathways in ANGPTL and COMPLEMENT (Fig. 5C; Supplementary Fig. S3A), which are known antitumor processes. We also noted higher interaction strength of the FAS/FASLG pathway among subtypes of T cells in PD patients than in PR patients (Fig. 5D and E; Supplementary Fig. S3B). This indicated increased apoptosis of T cells, especially exhausted T cells, in PD patients. Moreover, the interaction strength of the GALETIN pathway was stronger in PD patients than in PR patients (Fig. 5F and G). GALETIN pathway with LGALS9 (Gal9) is a well-known immunosuppressive signal in cancer (12, 25). Upon analysis at individualized level, there was little heterogeneity among the 3 PD patients (P01, P05, and P12). The trends of cell–cell communication among PD patients were similar, and their differences to PR patients remained alike (Supplementary Figs. S2 and S3).

In summary, cell–cell communication patterns differed according to clinical outcomes following ICI-based treatment. These communication patterns suggest increased immunosuppressive signaling in patients with PD.

The median follow-up period was 22.52 months (range, 2.77–86.80 months). Six patients were dead at the end of the last follow-up. Twenty-nine patients with metastatic disease underwent systemic therapy. The drugs, duration of treatment, and efficacy are summarized in Supplementary Table S4. First-line therapy was exclusively targeted therapy. Second-line therapy included TKI- (27.6%, 8/29) or ICI-based treatment (anti-PD-1 inhibitors with or without TKIs, 58.6%, 17/29): 70.6% (12/17) axitinib with anti-PD-1 inhibitor, 5.9% (1/17) pazopanib with anti-PD-1 inhibitor, 5.9% (1/17) cabozantinib with anti-PD-1 inhibitor, 5.9% (1/17) bevacizumab with anti-PD-1 inhibitor, and 11.8% (2/17) anti-PD-1 inhibitor monotherapy. Among patients who received nonfrontline therapy, those who received ICI-based treatment showed higher ORR (17.6% vs. 0%, P = 0.046) and DCR (64.7% vs. 12.5%, P = 0.004) than those who received TKIs alone (Fig. 6A). The median PFS was higher in patients with ICI-based treatment than in patients with TKI monotherapy (11.57 months vs. 5.50 months, P = 0.154; Fig. 6B). Among patients with germline mutation, the ORR (16.7% vs. 0%, P = 0.086) and the DCR (66.7% vs. 14.3%, P = 0.011) were higher after ICI-based treatment. The 1-year PFS was also better for ICI-based treatment than for TKI monotherapy in this population (41.3% vs. 17.1%, P = 0.174; Fig. 6C). Meanwhile, both the ORR (16.7% vs. 20.0%, P = 0.712) and the DCR (66.7% vs. 60.0%, P = 0.755) to ICI-based treatment were similar between the germline and somatic mutation subsets. The 1-year and 18-month OS was 100.0% and 88.2% in the entire cohort, 100.0% and 92.3% in the germline subset, and 100.0% and 75.0% in the somatic subset (P = 0.036, Fig. 6D). The treatment durations from first-line therapy of the 29 patients and 2 typical patients are illustrated in Fig. 6E–G.

FHRCC is highly malignant, but the urgent need for treatment remains unmet due to the inconsistent results from limited studies, both at the genetic and therapeutic levels. Our study reinforces the low mutation burden of FHRCC and demonstrates an immune-active microenvironment in this cancer type. On the basis of a larger FHRCC cohort receiving ICI-based treatment, our single-cell sequencing provides details on the functional status of CD8⁺ T cells after immunotherapy and finds enhanced immunosuppressive signaling as the underlying reason for primary ICI-based treatment resistance.

Our cohort of patients with FHRCC presented low TMB and high frequencies of KTM2C and ARID1A mutations. Low TMB is a consistent feature of FHRCC across studies; however, somatic mutations reflect the heterogeneity of this disease (1, 3, 5). Previous reports collectively showed that common somatic mutations in FHRCC are involved in the regulation of cell proliferation (NF2, PIK3CA), chromatin remodeling (ARID1A), and DNA damage repair (KMT2C, ATM; refs. 1, 3, 5, 29, 30). With regards to WES findings, in particular, Sun and colleagues observed the most common mutations were TTN, NF2, and ARID1B genes, which is different from our cohort. Nevertheless, frequent alterations in genes involving SWI/SNF complex were observed in both Sun's and our cohorts (3). SWI/SNF is a chromatin remodeler implicated in transcription and DNA damage repair (DDR), mutation of which was associated with better sensitivity to anti-PD-1 therapy in clear cell RCC (31). This common alteration might contribute to the relatively good response to ICI treatment of Sun's and our cohort, in contrast with Gleeson's population (5).

Our patients with KMT2C, ARID1A, and PTCH1 mutations achieved better DCR following ICI-based treatment. KMT2C involves in DDR, and ARID1A is a SWI/SNF complex gene that interacts with mismatch repair protein to promote DDR (32, 33). Mutations affecting DDR may lead to genomic instability, neoantigen production, PD-L1 upregulation, and enhancement of the antitumor immune response to ICI-based treatment (34). In RCC, loss-of-function mutations of KMT2C and ARID1A are associated with the longer PFS in ICI-based treatment than in TKI monotherapy (35). Our results suggest that targeting the DNA damage repair pathway could be considered in FHRCC. A trial exploring the combination of anti-PD-1 inhibitor and anti-PARP inhibitor, which targets the DNA damage repair pathway, in metastatic RCC including patients with FHRCC is ongoing (NCT04068831). PTCH1 belongs to the sonic hedgehog pathway that plays a role in tumorigenesis. In colorectal cancer, PTCH1 mutation created a favorable immune contexture with higher immunocyte infiltration, which improves tumor response to immunotherapy (36). The potential immunogenicity of PTCH1 mutation might explain the improved response of our patients.

Our study demonstrated that extensive CD8⁺ T-cell infiltration is common (56%) in FHRCC. We also found an inverse correlation between ATM expression and CD8⁺ T-cell infiltration, which has also been observed in breast cancer (37). ATM is a kinase belonging to the DDR pathway, the alteration of which may activate the innate immune response and increase neoantigens. ATM inhibition may result in an increase in type I IFN that enhances the production of T-cell–recruiting chemokines (38). Thus, lower level of ATM expression may be associated with increasing immunogenicity that drives T-cell infiltration. Sun and colleagues also observed extensive CD8⁺ T-cell infiltration in 46% of patients with FHRCC and considered it as the reason for improved disease control in 6 ICI-based–treated patients (3). However, we found no association between CD8⁺ T-cell infiltration and response to ICI-based treatment, similar to previous reports of clear cell RCC (39, 40). Therefore, the treatment response during ICI-based treatment involves a complicated interplay between genetic alterations and immune infiltration. IHC cannot identify the functional status of T cells, highlighting the need for in-depth exploration at the single-cell level.

In this study, we performed single-cell RNA sequencing in 4 patients receiving ICI-based treatment. Immunotherapy could lead to reprogramming of the TME, and thus focusing on the TME exposed to immunotherapy could identify potential causes and mechanisms of immunotherapy resistance. We found that the TME of FHRCC was primarily dominated by T cells, consistent with previous findings on clear cell RCC (8, 23, 39, 41–44). Notably, despite the high level of CD8⁺ T cells infiltrating the FHRCC tissues, most of the CD8⁺ T cells seemed to be exhausted in PD patients, which was validated in 7 ICI-exposed patients. The ratio of exhausted CD8⁺ T cells was significantly higher in PD patients. Similar results were also observed in clear cell RCC. Exhausted CD8⁺ T cells was higher in ICI treatment–resistant patients (23). However, in lung cancer, the exhausted signature of CD8⁺ T cells tended to be higher in ICI therapy–responded tissues (26). These inconsistent results may be related to the heterogeneity among cancer species. Our finding confirms that the degree of T-cell infiltration on IHC is insufficient to predict the response to ICI-based treatment when the T-cell functional status is unknown.

In pseudotime trajectory analysis of CD8⁺ T cells, glycolysis-related genes were gradually upregulated. This finding indicated that CD8⁺ T cells were transformed from a memory phenotype to an exhaustion phenotype (45). The gradually increasing trajectory of glycolysis pathway could also reflect the gradual increase in hypoxia in the TME as tumor progresses. Although it is absorbing to reverse T-cell exhaustion through PD-1 blockade, in current study, exhaustion-related genes were gradually upregulated during the pseudotime with continuous exposure to ICI-based treatment. We presume that ICI-based treatments are not sufficient to retrain terminally exhausted T cells, where multiple immune inhibitory molecules accumulate. It was reported that exhausted precursor CD8⁺ T cell was the key effector cell for ICI-based treatment, instead of exhausted CD8⁺ T cell (26). Considering that we do not have matched pre- and posttreatment samples, whether these conclusions are fully applicable to FHRCC deserves further investigation.

Cell–cell communication analysis showed that the number and strength of inferred interactions of cellular communication were remarkably higher in PD patients than in PR patients. Moreover, interactome analysis showed an increased number of novel receptor-ligand pairs such as TIGIT and GALECTIN, which are T-cell–suppressive factors, in PD patients. TIGIT, which belongs to the poliovirus receptor nectin family, is an inhibitory receptor expressed mainly on T cells (46). Co-blocking of PD-1 and TIGIT can revive terminally exhausted T cells in bladder cancer, and anti-TIGIT therapy has achieved promising outcomes in early-phase clinical trials (19, 47). Galectins are glycan-binding proteins involved in tumor growth, apoptosis, and immune resistance. PD-1 can interact with galectin-9 and TIM-3 to attenuate PD-1⁺TIM-3⁺ T-cell death, and galectin-9 inhibition can expand tumor-infiltrating T cells (48). Our results provide potential immune therapeutic targets and possible combination strategies for patients resistant to ICI-based treatment. In individualized analysis, although the genetic mutation profiles were not identical among PD patients, cell–cell communication patterns were similar. These suggest that there might be some common patterns of ICI resistance among FHRCC patients. Considering the defects of single-cell sequencing technology, the analysis methods, and the significance of posttranscriptional regulation, a more accurate map of cellular interactions is still worth further investigation at protein level in FHRCC.

Analysis of FHRCCs with different causes showed superior survival in those caused by germline mutations. Gleeson and colleagues also reported that despite similar molecular backgrounds, patients with germline mutations had longer OS than did those with somatic mutations (28.1 months vs. 13.6 months; ref. 5). Notably, patients with germline mutations in this study responded better to ICI-based treatment, which may have contributed to the improved survival in this group. The use of ICI-based treatment in FHRCC is controversial because response rates vary greatly owing to the different strategies for ICI-based treatment. In general, the ORR and DCR of ICI alone, either as monotherapy or dual therapy, are only 0% to 18% and 36% to 38%, respectively, and these are lower than those of TKI monotherapy (1). However, ICI-based treatment could achieve an ORR and DCR of 58% and 100%, respectively, and a PFS of 13.3 months (3). This could explain our relatively good response rates (ORR, 17.6%; DCR, 64.7%) and better PFS (11.6 months) of ICI-based treatment because of the dominance of ICI plus TKI. Other combination regimens, including mTOR/VEGF and bevacizumab/erlotinib, also show remarkable results in FHRCC (1, 5, 6). Although no consensus has been reached, current data suggest that combination therapies involving antiangiogenic drugs are promising for FHRCC.

Our study had some limitations. First, the study had a limited sample size and follow-up time due to the rarity of FHRCC. Furthermore, single-cell sequencing was based on a single tumor sample collected at a single time point from each patient; this likely did not capture the spatial and temporal characteristics of FHRCC. Finally, different TKIs were used in combination with ICIs. Further trials are needed to establish accurate treatment guidelines.

Immune infiltration is frequent in FHRCC, and response to ICI-based treatment depends on the functional status of tumor-infiltrating lymphocytes. Our study elucidates diverse genetic alterations, provides data on the efficacy of ICI-based treatment, and deepens the understanding of ICI-based treatment failure in FHRCC. Further, the findings provide evidence to additional factors targeting T-cell function in current immunotherapies for FHRCC.

No author disclosures were reported.

P. Dong: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing. X. Zhang: Conceptualization, data curation, software, formal analysis, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. Y. Peng: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. Y. Zhang: Conceptualization, resources, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. R. Liu: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. Y. Li: Resources, data curation. Q. Pan: Resources, data curation. W. Wei: Resources, data curation. S. Guo: Resources, methodology, writing–review and editing. Z. Zhang: Resources, methodology, writing–review and editing. H. Han: Resources, methodology, writing–review and editing. F. Zhou: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. Y. Liu: Conceptualization, resources, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. L. He: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

The study was supported by two grants from the National Natural Science Foundation of China (Grants Nos. 81772483 for Liru He and 82102988 for Ruiqi Liu). The authors would like to thank for all the valuable contributions from the patients and their families in this study. And we would like to thank Editage (www.editage.cn) for English language editing.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.