Metastatic progression of pheochromocytoma and paraganglioma occurs via parallel evolution

Abstract

Pheochromocytoma (PCC) and paraganglioma (PGL) are neuroendocrine tumors derived from chromaffin cells of the adrenal medulla and ganglia of the autonomic nervous system. Approximately one-third are causatively associated with pathogenic germline variants. Metastatic disease develops in up to 25% of patients with PCC/PGL, for whom therapeutic options are limited, and no targeted treatments exist. Tumor evolution in metastatic PCC/PGL has not been well delineated. We performed whole-exome sequencing of paired specimens from 27 patients with metastatic PCC/PGL to better understand cancer progression. Tumors demonstrate high rates of loss-of-function variants in chromatin remodeling and DNA damage repair genes, suggesting potential therapeutic targets. Low rates of shared somatic variants were observed between primary tumors and metastases, with evidence of independent monoclonal pathogenic variants in metastatic tumors. These findings suggest that PCC/PGL metastases develop via monoclonal seeding and parallel progression.

Introduction

Pheochromocytoma (PCC) and paraganglioma (PGL) are neuroendocrine tumors which originate from chromaffin-derived tissues of the adrenal medulla and ganglia of the autonomic nervous system¹. PCC/PGL frequently secrete catecholamines, resulting in cardiovascular and metabolic effects. PCC/PGL have the highest rates of heritability among all solid tumors and are the first human tumors to be causatively associated with oncogenic pathogenic variants^2–4. Between 27 and 40% of PCC/PGL are due to germline pathogenic variants, with more than 20 identified driver genes^5–7. Germline pathogenic variants are a known risk factor for aggressive and metastatic disease, which develops in 10–25% of PCC/PGL^8,9. There is a critical need for new treatments for metastatic PCC/PGL, as therapeutic options are extremely limited for unresectable and metastatic disease^9–11.

An improved understanding of tumor genomics has led to the identification of novel approaches to targeted therapy. Three primary molecular subtypes of PCC/PGL have been characterized. These include Cluster 1 (pseudohypoxia), Cluster 2 (kinase signaling), and Cluster 3 (Wnt-altered). Alterations in Cluster 1 genes with consequent protein loss, including SDHx, VHL, FH, EPAS1, and EglN/PHD2, cause constitutive activation of hypoxia-inducible transcription factors (HIF1α and HIF2α)⁵. This common downstream pathway is a promising therapeutic target for a genomically heterogenous disease. HIF2α has been proposed as a biomarker for poor outcomes in PGL, and the HIF2α inhibitor belzutifan (MK-6482) has recently been approved as monotherapy for metastatic PCC/PGL, after Phase 2 trials demonstrated a 26% objective response rate^12,13. Studies in Cluster 1 mutated tumors have demonstrated that elevated levels of succinate act via lysine demethylases to suppress the homologous recombination DNA repair (HRD) pathway required for the repair of DNA double-strand breaks, inducing sensitivity to poly-ADP ribose (PARP) inhibitors in xenografted animal models and in vitro studies, but have not been rigorously evaluated in humans^14–16. In contrast, Cluster 2 genes (RET, NF1, TMEM127, MAX) affect more diverse molecular pathways, although a strong interaction exists between MAX/MYC and HIF and RAS signaling, suggesting a potential biological link between pseudohypoxia and kinase-signaling subtypes^17,18. Recent investigations show that sunitinib, a tyrosine kinase inhibitor, demonstrates anti-tumor efficacy in metastatic PCC/PGL¹⁹.

Although germline variants have been well characterized in both primary and metastatic PCC/PGL, somatic alterations have been predominantly studied in primary tumors until very recently^20–28. Recent investigations of SDHB-altered PCC/PGL demonstrated late somatic events in metastatic disease, including TERT and ATRX alterations, suggesting that these are common secondary genomic drivers²⁹. To date, no studies have focused on the role of somatic variants in tumor evolution and the progression of metastatic disease across the spectrum of germline variants. This approach has been fruitful in other malignancies, where comparative analysis of primary and metastatic tumors suggests that driver alterations may arise in tumor subclones and can inform targeted therapy^30,31. Therefore, the goal of this study was to characterize somatic alterations in paired primary and metastatic PCC/PGL tumors to improve understanding of metastatic tumor evolution and to inform potential therapy.

Results

Study cohort

We studied paired germline-primary-metastasis samples from 11 patients and paired germline-metastasis samples from 16 patients diagnosed with metastatic PCC or PGL. In total, we sequenced 48 formalin-fixed, paraffin-embedded tumors from 27 unrelated patients, including 20 PCC (3 primary tumors; 17 metastases) and 28 PGL (8 primary tumors; 20 metastases). Seventeen patients had pathogenic germline variants in known susceptibility genes (SDHA, 2; SDHB, 13; SDHC, 1; RET, 1; 29 total tumors). Twelve patients were male (44.4%), 22 were Non-Hispanic White (81.5%) and five were Non-Hispanic Black (18.5%). The median age at diagnosis was 36 years (IQR: 24.5–49); and seven patients presented with distant metastases at initial diagnosis. In patients without metastatic disease at diagnosis, the median time to recurrence was six years. The median follow-up duration was 12 years (IQR: 7–18). Kaplan–Meier curves for recurrence-free and overall survival are shown in Fig. 1. Complete patient characteristics including prior treatments are presented in Supplementary Table 1.

Fig. 1. Kaplan–Meier survival curves.

A Survival curve displaying recurrence-free survival. The median time to recurrence was 5 years amongst all patients and 6 years amongst patients who were not diagnosed with metastatic disease at the time of initial diagnosis. B Survival curve displaying overall survival. The median time-to-death from all-causes was 14 years (IQR: 7–19).

Somatic variant analysis reveals frequent pathogenic variants in extracellular matrix, chromatin remodeling, and DNA damage response genes

We used whole-exome sequencing (WES) to evaluate the architecture of somatic variants across the study cohort. We examined individual variants to ascertain whether loss or gain-of-function pathogenic variants were enriched in metastatic disease compared to primary disease. We identified 658 variants in cancer-associated genes that met loss-of-function criteria and 163 variants in cancer-associated genes that met gain-of-function criteria (see Methods and Supplementary Figs. 1–3 for information on variant calling and filtering; see Supplementary Data 1 for identified variants).

Among the loss-of-function variants, we identified 98 missense variants with REVEL score >0.7, 37 splice acceptor and splice donor variants, 346 start lost or stop gained variants, and 177 frameshift variants. The median tumor mutational burden (TMB) in the cohort was 1.76 mutations/Mb (IQR: 0.71–13.8; range 0.0–143.9) which is consistent with previous estimates^5,20. TMB varied significantly by tumor type (1.19 mutations/Mb in PCC versus 2.79 mutations/Mb in PGL; Wilcoxon’s test p = 0.041) but did not vary by disease stage (1.71 mutations/Mb per primary versus 2.72 mutations/MB per metastasis; Wilcoxon’s test p = 0.524), germline alterations (1.72 mutations/Mb per non-SDHx-mutated tumor versus 1.95 mutations/Mb per SDHx-mutated tumor; Wilcoxon’s test p = 0.246), or among synchronous versus metachronous metastases (1.71 mutations/Mb per synchronous metastases versus 1.57 mutations/Mb per metachronous metastases; Wilcoxon’s test p = 0.536) (see Fig. 2 and Supplementary Fig. 4A). The number of cancer gene associated loss- and gain-of-function variants did not vary based on tumor type, disease stage, germline alterations, or among synchronous versus metachronous metastases (see Supplementary Fig. 4B).

Fig. 2. Landscape of putative pathogenic variants in primary and metastatic pheochromocytomas and paragangliomas.

Oncoplot displaying the genes in loss-of-function and gain-of-function pathogenic variants in COSMIC Cancer Gene Census Tier 1/2 genes most commonly found. A. The top panel shows the tumor mutational burden for each sample. The second panel categorizes tumors by type (PCC versus PGL), tumor category (primary versus metastatic), and the presence of germline mutations. B. Loss-of-function and gain-of-function mutations. Mutations are shown for each sample, with gray boxes indicating no detected mutations and colored boxes representing identified mutations. Colors correspond to the predicted functional impact of the mutations. The percentages to the left indicate the proportion of samples with mutations in each gene, while the right-hand bar chart summarizes the frequency and types of mutations for each gene. Only genes mutated in at least six samples are included.

We found frequent loss- or gain-of-function variants in extracellular matrix, chromatin remodeling, and DNA damage repair related genes (Fig. 2; see Supplementary Fig. 5 for sensitivity analysis; see Supplementary Table 2 for rates of variants in previously reported genes). Among extracellular matrix genes, 31% of tumors had variants in MUC16 (n = 15); 12% had variants in FAT1 (n = 6); and 12% had variants in MUC4 (n = 6). Among chromatin remodeling genes, 21% of tumors had variants in KMT2D (n = 10); 15% had variants in ATRX (n = 7); 12% had variants in CREBBP (n = 6); 10% had variants in KMT2C (n = 7); and 10% had variants in KMT2A (n = 7). Among DNA damage response genes, 10% had variants in BRCA2 (n = 5); 8% had variants in ATM (n = 4); 8% had variants in ATR (n = 4); 6% had variants in TP53 (n = 3); and 6% had variants in BRCA1. When accounting for transcript size, ID3, KMT2D, and SOX21 had the greatest mutations/Mb (533, 441, and 399 respectively; see Supplementary Data 2). In total, 39.6% of tumors (n = 19) had variants in either chromatin remodeling or DNA damage repair genes. The rates of these variants were not significantly different in patients with or without known germline pathogenic variants in SDHx genes (50% versus 36.4%; Fisher’s exact test, p = 0.696). Finally, we utilized dNdScv³² to identify genes under positive selection pressure when considering a global false discovery rate of 0.1: MUC3A (Q < 0.001) and SCML2 (Q = 0.079) were identified as being under positive selection.

Copy number variant analysis reveals frequent arm-level changes of chromosome 1p and focal changes in 1q21.2, 8q24.3, 9q13, 19p13.3, and 19q13.2

We analyzed the copy number variant profiles by applying Sequenza and annotSV to paired germline-tumor samples. Large variations were observed in the estimated purity, ploidy, and number of copy events per sample (see Supplementary Fig. 6). The purity of the samples varied significantly by tumor type (0.75 in PCC versus 0.55 in PGL, p < 0.001, Wilcoxon’s test) and germline alteration (0.74 in non-SDHx versus 0.59 in SDHx, p = 0.002, Wilcoxon’s test) but not by disease stage or between synchronous and metachronous metastases. The median ploidy across all samples was 1.9, and ploidy did not vary according to tumor type, disease stage, germline variants, or between synchronous and metachronous metastases. We observed whole genome duplication in ~15% of our samples, which is consistent with previous reports. The median number of copy number events per tumor was 111 (IQR: 77–154, range: 57–2884), with one sample demonstrating evidence of chromothripsis. The number of copy number events also did not vary according to tumor type, disease stage, germline variants, or between synchronous and metachronous metastases.

Global copy number changes are shown in Fig. 3A. Arm-level losses were observed in chromosomes 1p, 3p, 3q, 11p, and 17p, neutral loss-of-heterozygosity events in chromosomes 11q and Xq, and arm-level gains in chromosomes 7q, 8q, 12p, 12q, 13q, 19p, and 20q in more than 20% of the samples (see Supplementary Fig. 7). We used GISTIC to identify chromosomal bands that were significantly enriched for deletions or amplifications. Primary tumors contained two chromosomal bands — 1q21.2 and 9q13, which were enriched for deletions at an FDR < 0.05, and contained COSMIC Tier 1 cancer genes (see Fig. 3A). Metastatic tumors contained two chromosomal bands, 8q24.3 and 19q13.2, that were enriched for deletions and one chromosomal band, 19p13.3, that was enriched for amplifications (see Fig. 3A, B) at an FDR < 0.05. These peaks contained the COSMIC Tier 1 cancer genes, RECQL4, AKT2, and FSTL3.

Fig. 3. Landscape of copy number variants in primary and metastatic pheochromocytomas and paragangliomas.

A Cohort-wide summary of copy number events. The upper panel shows the average copy number changes at each locus across the cohort, while the lower panel illustrates copy number changes for individual samples. Samples are ordered identically to those in Fig. 1A. B GISTIC analysis of deletion events in primary and metastatic tumors. Peaks with a false discovery rate (FDR) < 0.05 are highlighted in blue. COSMIC Cancer Gene Census Tier 1 genes located within these peaks are annotated. C GISTIC analysis of amplification events in primary and metastatic tumors. Peaks with an FDR < 0.05 are highlighted in red. COSMIC Cancer Gene Census Tier 1 genes located within these peaks are annotated.

OncoDriveFML reveals frequent driver events in metastatic tumors with germline SDHB pathogenic variants

We utilized the Cancer Genome Interpreter to identify potential drivers of metastatic disease. OncoDriveFML identified 814 somatic driver mutations across the entire cohort, of which 536 occurred in SDHB-altered tumors and 713 occurred in metastases (see Supplementary Fig. 8 and Supplementary Data 3). The most mutated genes included SPEN (n = 20), MUC16 (n = 15), KMT2D (n = 11), BRIP1 (n = 10, FAT1 (n = 10), ATRX (n = 8), CREBBP (n = 8), KMT2C (n = 8), LRP1B (n = 8), and NF1 (n = 8). Overall, 65.2% of SDHB-altered tumors exhibited driver mutations compared with 12% of non-SDHB-altered tumors (n = 18 and n = 3, respectively; P < 0.001 Fisher’s exact test). CREBBP and EP300 are histone acetyltransferases and alterations in both genes have been implicated in both hematologic and solid malignancies^33,34. Additionally, OncoDriveFML identified 206 copy number driver events of which 186 occurred in metastatic tumors and 82 occurred in SDHB mutated tumors. The most common copy number events included amplifications in AURKA (n = 7) and amplifications in AREG, CDK4, CDK6, EGFR, EREG, FRS2, GRM3, MDM2, MET, or MYC (n = 6; see Supplementary Fig. 9 and Supplementary Data 4). Driver copy number events were clustered within a subset of tumors.

Paired primary-metastatic tumors demonstrate low rates of shared somatic variants and evidence of monoclonal seeding followed by parallel evolution

When evaluating the differences in somatic mutation profiles between primary-metastasis tumor pairs, we observed a low fraction of somatic variants that were present in both tumors (see Fig. 4A). When looking specifically among potential pathogenic variants (n = 821), we identified one patient with a TP53 variant that was present in both their primary tumor and metastases; however, no other primary-metastasis tumor pairs demonstrated this pattern. When examining all variants that met our criteria for read depth amongst patients with primary-metastasis tumors pairs (n = 27,309), we found that the median percentage of pathogenic variants that were shared between paired samples was 9.7% (range: 0.33–36.1%). To confirm that this finding was not artifactual, we performed an identity-by-descent analysis that confirmed the relatedness of our paired samples (see Fig. 4A) and manually reviewed pathogenic variants in cancer-associated genes for paired primary and metastatic tumors using IGV to confirm that there was no evidence of pathogenic variants being missed by the bioinformatics pipeline.

Fig. 4. Timing and clonality of somatic mutations and copy number events in metastatic pheochromocytomas and paragangliomas, with chronology of primary tumor development and metastatic seeding patterns.

A Relatedness of paired primary-metastatic and metastatic-metastatic tumor samples, shown as the shared fraction of somatic variants (light gray) and germline variants identity by descent (gray). B Proportion of primary-private, metastasis-private, and shared clonal and subclonal pathogenic variants across all tumors, and in pheochromocytomas and paragangliomas separately. C Timing estimates for pathogenic variants in the genes shown in Fig. 1A, derived using MutationTimeR. Most variants were subclonal, with few being classified as early, clonal mutations. D Oncoplot of timing estimates for copy number gains, neutral loss-of-heterozygosity events, and whole-genome duplications. The left panel categorizes tumors by type (PCC versus PGL), tumor category (primary versus metastatic), and germline variants. The center panel shows aggregated timing estimates for copy number changes across chromosomal arms, while the right panel summarizes timing estimates for all mutations in a sample. E Jaccard similarity index estimates for primary-metastatic tumor pairs. Metastases were classified as monoclonal when the Jaccard index was <0.30. F Estimates of primary tumor expansion age for PCC and PGL, based on tumor doubling times and sizes at diagnosis reported in the literature. G Estimates of metastatic seeding time, inferred from plausible primary tumor expansion ages and the ratio of metastasis-private clonal mutations to primary-private clonal mutations.

Based on this observation of a low fraction of shared somatic variants between primary-metastasis tumor pairs, we sought to further elucidate the evolutionary characteristics of metastatic PCC and PGL through molecular clock analysis. We applied MutationTimeR³⁵ to estimate the relative timing and clonality of the somatic and copy number variants in our samples. When examining primary-metastatic tumor pairs, we found that metastases demonstrated a higher proportion of clonal pathogenic variants than primary tumors (70% versus 30%, p < 0.001, two-proportions z-test; see Fig. 4B). Across all pathogenic variants, 38.0% were classified as subclonal (n = 312); only 7.0% were classified as early, clonal variants (n = 57) (see Fig. 4C). Copy number gains, neutral loss-of-heterozygosity, and whole genome duplications were estimated to occur early in molecular time in almost all samples (see Fig. 4D).

Metastatic founder populations and the time of metastatic seeding occur years before the diagnosis of primary tumors in metastatic PCC/PGL

Using methodology developed by Hu et al.³⁶, we inferred the seeding pattern of individual metastases by calculating the Jaccard similarity index for primary-metastatic tumor pairs using the number of metastasis-private clonal variants, primary tumor-private clonal variants, and shared subclonal variants as inferred by MutationTimeR. In 93.3% of primary-metastasis tumors pairs (n = 14), the Jaccard similarity index was <0.30, suggesting a monoclonal seeding pattern of metastases (see Fig. 4D). We next estimated the time from the emergence of the metastatic founder population to the diagnosis of the primary tumor (hereafter, primary tumor expansion age) using a Gompertz growth model informed by literature-derived priors on the doubling time and size at diagnosis of PCC/PGL (see Supplementary Table 3 and Supplementary Table 4)^37–44. For each patient with primary–metastasis pairs, we bootstrapped posterior distributions of expansion age, estimating that the origin of metastatic lineages preceded primary diagnosis by a median of 17.7 years (IQR 10.8–39.9; see Fig. 4F and Supplementary Table 5).

Lastly, we estimated metastatic seeding times ($t_s$) for all primary-metastasis tumors pair in our cohort, accounting for uncertainty in the number of metastasis-private clonal variants and primary tumor-private clonal variants, the rate at which private and clonal mutations accrue in the primary tumor sample, and the primary tumor expansion-age. Among the 15 pairs, five tumors generated unreliable metastatic seeding times estimates (defined as $t_s>0$) due to large ratios of metastasis-private somatic variants to primary-private somatic variants which can occur when a metastasis was seeded after primary tumor diagnosis and accumulated private clonal mutations after surgical resection of the primary tumor or when metastatic seeding occurred before diagnosis of the primary accompanied by a large number of private clonal mutations during metastatic growth. Given the inability to distinguish between these two phenomena in bulk sequencing, $t_s$ for these tumors are considered unreliable.

In the remaining 10 primary-metastasis tumors pairs, we estimated that metastases were seeded ~12.5 years before the diagnosis of the primary tumor (IQR: 9.1–29.7) (see Fig. 4G) and Supplementary Table 6). Synchronous metastases were estimated to be seeded 32 years prior to diagnosis while metachronous metastases were estimated to be seeded 9.9 years prior to diagnosis (Wilcoxon’s test p = 0.067). Taken together, these data suggest that early copy number changes contribute to tumorigenesis and metastasis in PCC/PGL and allow for monoclonal seeding of metastases, with subsequent development of independent variants in primary and metastatic tumors.

Discussion

Molecular characterization is essential for modern cancer treatment. Genomic subtyping can explain tumorigenesis, predict prognosis, guide surveillance regimens, and determine the candidacy for targeted therapy. We performed whole-exome sequencing of paired primary and metastatic PCC/PGL across the spectrum of germline variants to better understand tumor progression and identify potential therapeutic targets. We found that loss-of-function variants in chromatin remodeling and DNA damage repair are common in PCC/PGL, including ATRX which is associated with aggressive PCC/PGL, and has implications for therapy^5,20,26,29. We further assessed tumor evolution in metastatic disease. Importantly, we found low rates of shared somatic variants between primary tumors and metastases, with evidence of independent monoclonal variants in metastatic tumors. The metastatic lineages originated at a median of 17.7 years prior to diagnosis of the primary tumor, with metastatic seeding occurring at a median of 12.5 years prior to diagnosis. These findings suggest that metastases represent early events in PCC/PGL tumor development, with the subsequent parallel evolution of primary and metastatic tumors.

Chromatin remodeling and DNA damage repair

Chromatin remodeling and DNA damage have been implicated in tumor progression in PCC/PGL through several pathways. We found that PCC/PGL frequently demonstrated significant loss-of-function variants in chromatin remodeling and DNA damage repair genes, including ATRX. ATRX is a member of the SWI/SNF family of chromatin remodelers, with essential roles in telomere maintenance and cell cycle integrity⁴⁵. Somatic variants in ATRX are present in a minority of PCC/PGL cases, both alone and in conjunction with germline alterations in Cluster 1 genes, including SDHB, VHL and FH^26,29,46. ATRX is associated with clinically aggressive behavior and is an independent risk factor for metastatic disease⁴⁷. In our cohort, ATRX and other driver mutations were significantly more frequent in SDHB-altered tumors, consistent with a recent multi-omic study of SDHB-altered paired metastatic PCC/PGL²⁹. Prior work suggests that ATRX and TERT represent mutually exclusive alterations, suggesting a redundant role in metastatic progression; however, our investigation did not evaluate TERT promoter alterations²⁹. Importantly, ATRX variants have implications for treatment; in other cancer types, ATRX-altered cells appear susceptible to therapies targeting the DNA damage response, including PARP inhibition^48,49. Notably, SDHx-altered tumors demonstrated sensitivity to PARP inhibitors in vitro and in xenografted models, suggesting that DNA damage response therapies may represent a viable treatment strategy via multiple mechanisms^14,16,50.

Chromatin remodeling is also integral to the mechanism of tumor progression in SDHx-altered PCC/PGL. Alterations in SDHx and other Krebs-cycle genes, including FH lead to the accumulation of the oncometabolite succinate, suppressing homology-dependent DNA repair¹⁶. Oncometabolite accumulation directly disrupts chromatin signaling via inhibition of the lysine demethylase KDM4B, which is facilitated by the aberrant hypermethylation of H3K9⁵¹. Global hypermethylation is well-documented in SDHx PCC/PGL, in which accumulation of succinate inhibits 2-OG-dependent histone and DNA demethylases, leading to epigenetic silencing. Interestingly, the hypermethylator phenotype is most specific to tumors deficient in Krebs-cycle genes, with metastatic disease demonstrating more heterogenous methylation profiles^52–54. Interestingly, our data suggest other potential mechanisms of epigenetic modification. In our cohort, OncoDriveFML identified frequent alterations in CREBBP and EP300, which are two histone acetyltransferases that are key epigenetic regulators. EP300 is one of the most commonly altered genes in cancer and can occur in conjunction with CREBBP mutations³³. CREBBP/EP300 alterations are associated with a hypermethylated subtype of oropharyngeal carcinoma⁵⁵. Taken together, these findings suggest that multiple molecular pathways may contribute to hypermethylation and chromatin disruption in PCC/PGL.

Tumor evolution

In the traditional linear progression model of metastatic disease, evolution of the primary tumor leads to the accumulation of somatic variants with late dissemination of metastatic clones with high degrees of similarity to the primary tumor⁵⁶. In contrast, in our cohort, paired primary and metastatic PCC/PGL showed low rates of shared somatic variants. This is consistent with prior investigations that show significant genomic heterogeneity between primary and metastatic tumors, with acquisition of secondary driver alterations in metastases²⁹. Molecular clock analysis revealed that metastatic founder lineages arose early in primary tumor development, at a median of 17.7 years prior to diagnosis. Metastatic seeding events occurred at a median of 12.5 years prior to diagnosis, allowing for ongoing evolution of both primary and metastatic lesions over a long time period. These findings are consistent with a parallel progression model of metastasis, in which early disseminating cells from primary tumors evolve separately. This behavior has been observed in breast, sarcoma, bladder, and other cancers and leads to significant genetic heterogeneity between primary tumors and metastases^57,58. PCC/PGL typically have an indolent progression and may have a long latency period between initial diagnosis and development of metastatic disease, consistent with the parallel progression model. Our data do not permit the granular timing of driver alterations with respect to tumor evolution; however, prior work suggests that alterations in TERT and ATRX may represent late somatic events²⁹.

One of the expected results of parallel progression is that primary tumors and metastases demonstrate greater genomic heterogeneity than that predicted by traditional linear progression. This heterogeneity can result from two processes: polyclonal seeding, where metastases originate from divergent lineages in the primary tumor, and clonal expansion where metastatic tumor cells continue to evolve under pressure from the tumor microenvironment. Data from multi-cancer analyses of breast, colorectal, and lung cancer tumors suggest that polyclonal seeding is particularly common in untreated nodal and distant organ metastases³⁶. In contrast, our data showed a low Jacquard similarity index between tumors, suggesting that PCC/PGL metastases represent monoclonal expansion events. This finding has important treatment implications because of the potential for genomic heterogeneity and therefore differing therapeutic responsiveness between primary tumors and metastases. Tumor phylogenetic analysis and inter-metastasis comparisons are needed to further characterize metastatic seeding patterns in PCC/PGL. It will be of utility to determine whether oligometastatic deposits share a common clonal origin, potentially rendering multiple metastases susceptible to treatment with systemic targeted therapies. Somatic profiling of metastatic tumors is not currently in clinical use for PCC/PGL but might also facilitate the identification of clinically actionable variants in metastatic PCC/PGL, which are not present in the primary tumors. Importantly, systemic treatment may cause a “bottle neck” phenomenon, selecting for resistant cells and leading to the expansion of monoclonal metastatic tumors. The dearth of effective systemic therapeutic options for PCC/PGL inherently limits treatment pressure. However recent studies show that in paired tumors taken before and after treatment with DNA alkylating chemotherapy with cyclophosphamide, vincristine and dacarbazine (CVD), post-treatment tumors demonstrate doubling in single nucleotide variants, and overexpression of MGMT, which is an established temozolomide resistance mechanism in glioblastoma²⁹. Temozolomide is a newer alkalyting agent which has demonstrated modest tumor and biochemical efficacy in PCC/PGL in small, retrospective series^59,60. SDHx alterations are associated with response to temozolomide in patients with metastatic PCC/PGL⁶¹. Temozolomide is currently under evaluation both as a single agent and in combination with Olaparib (Alliance A021804), which will offer important insight into dual treatment with targeted therapies⁶².

Limitations

Limitations of this study should be considered. First, our cohort is small, which limits statistical power and may reduce the generalizability of observed patterns. Second, the sample set is also enriched for SDHx-altered tumors, raising the possibility of selection bias and potentially under-representing other molecular subtypes. Third, we lacked confirmatory experimental or functional data, so mechanistic interpretations of putative pathogenic and/or driver variants remain provisional. Fourth, analyses were performed on FFPE material using whole-exome sequencing, which can be affected by variable DNA quality, fixation-related artifacts, and limited ability to detect non-exonic events or complex structural variation. Finally, our study relies on a single genomic modality and does not incorporate complementary data types like transcriptomic, epigenetic, or proteomic data.

Conclusion

Analysis of metastatic-primary tumor PCC/PGL pairs demonstrated high rates of loss-of-function variants in chromatin remodeling and DNA damage repair genes, suggesting potential molecular therapeutic targets. Our data also demonstrate low rates of shared somatic variants between primary tumors and metastases and evidence of independent monoclonal variants in metastatic tumors. Taken together, these findings suggest that PCC/PGL metastases develop via monoclonal seeding and parallel progression. Further investigations evaluating inter-metastatic genomic variation will aid in our understanding of disease progression and potential treatment options for PCC/PGL.

Methods

Study cohort

Study subjects were identified from an Institutional Review Board approved, prospectively maintained tissue repository and database (Protocol #812495) of PCC/PGL at the University of Pennsylvania. Enrolled subjects provided informed consent for the study of clinical and genomic data, and the study adhered to the Declaration of Helsinki. The clinical variables assessed included demographic data, clinical data, treatment history, genomic testing, pathology, and survival outcomes. Socioeconomic data were not collected, and therefore, were not reported in this study. Sex- and race-based analyses were not performed because of the small sample size, which may limit the generalizability of this study.

Whole-exome sequencing

DNA was extracted using standard protocols. One ug of DNA was used for library preparation. Library preparation was performed using the NEBNext FFPE DNA Repair Kit and NEBNext Ultra II Library Prep Kit (New England BioLabs, Ipswich, MA, USA; RRID:SCR_013517), as previously described^63,64. After library preparation, whole-exome sequencing was performed using the SureSelectXT Human All Exon V7 (Agilent Technologies, Santa Clara, CA; RRID:SCR_013575), capturing four exomes per bait, using dual indexes, and sequencing on an Illumina NovaSeq 6000 (RRID:SCR_016387) at the Penn Next Generation Sequencing Core with 100 base paired end sequencing. Whole-exome sequencing achieved a mean depth of 68× across the cohort (median depth 54×).

Somatic variant analysis

Fastq files from whole-exome sequencing were aligned to the hg38 assembly of the human genome using BWA-MEM (v0.7.17; RRID:SCR_010910)⁶⁵. Variant discovery was performed using Mutect2 (v4.2.0.0; RRID:SCR_026692)⁶⁶, Strelka2 (v2.8.4; RRID:SCR_005109)⁶⁷, Vardict (v1.4; RRID:SCR_023658)⁶⁸, Lancet (v1.1.0)^69,70, and Varscan2 (v2.4.4; RRID:SCR_006849)⁷¹. Somatic variants were called using Varlociraptor (v8.4.6) with a statistical model to account for FFPE artifacts and a false discovery rate of 5%⁷². Subsequent analyses were limited to variants with a read depth of >10. To identify pathogenic loss-of-function (LOF) variants, we excluded somatic variants with a gnomAD frequency ≥0.01 and protein truncating variants located in the last exon. We then filtered for (1) frameshift variants, (2) nonsense variants, (3) missense variants with a REVEL score greater than 0.7, and (4) splice acceptor (+1, +2) or splice donor variants (−1, −2). To identify pathogenic gain-of-function (GOF) variants, we excluded all variants meeting the LOF criteria and then filtered for variants that had (1) either pathogenic/likely pathogenic ClinVar annotations or Catalog of Somatic Mutations in Cancer (COSMIC) pathogenic annotations, and (2) mutation types matching those documented as pathogenic for their respective genes in COSMIC. We filtered the LOF and GOF variants that met their respective criteria in genes documented in the COSMIC v98 Tier 1 or Tier 2 Cancer Gene Census (RRID:SCR_002260)⁷³. Germline pathogenic variants were abstracted from clinical genetic testing and confirmed using germline sequencing data. We ensured tumor-normal samples and primary-metastasis tumor samples were correctly paired using identity-by-descent analysis with plink v1.9 using the --genome command⁷⁴.

Copy number variant analysis

We called copy number variants in whole-exome sequencing data with Sequenza (v3.0; RRID:SCR_016662)⁷⁵. Using annotSV (v3.4.2)⁷⁶, we annotated CNVs as deletions (no copies), losses (one copy), neutral losses of heterozygosity (two copies), gains (3–4 copies), or amplifications (≥5 copies). We estimated arm-level losses within each tumor by calculating the percentage of a chromosome affected by a deletion or loss. We calculated arm-level gains by calculating the percentage of a chromosome affected by a gain or amplification above the ploidy of the sample. We considered tumors as having arm-level losses or gains if more than 50% of the chromosome arm was affected by a CNV of that class. Focal copy number amplifications and deletions were identified using GISTIC (v2.0.23; RRID:SCR_000151)⁷⁷ on the GenePattern server (RRID:SCR_003201).

Molecular clock analysis

We used MutationTimeR to estimate relative timing of somatic variants with respect to local copy-number gains³⁵. Input variants were restricted to sites with tumor read depth >10. For each mutation, we provided: (A) tumor-specific reference and alternate allele counts from our somatic variant calling pipeline; (B) allele-specific copy-number states, tumor purity, average ploidy, and whole-genome duplication status inferred using Sequenza; and (C) patient sex abstracted from the clinical record. MutationTimeR integrates these inputs to infer mutation copy number and assign timing categories (clonal early, clonal late, clonal NA, subclonal, or NA). Classification was made with respect to established methodology: early/late if the mutation occurred preceding or after the copy number gains, and clonal/subclonal based on if the mutation was present in all tumor cells or only in a fraction of them. These classifications were used for downstream analyses including seeding patterns, primary tumor expansion age, and metastatic seeding time as previously described.

Seeding pattern inference

For each primary–metastasis tumor pair, we counted metastasis-private clonal mutations (Lm), primary-private clonal mutations (Lp), and primary–metastasis shared subclonal mutations (Ws) according to MutationTimeR timing and clonality labels. We computed the Jaccard similarity index (JSI) as shown in Eq. (1):

$$\mathrm{JSI}=\frac{W_s}{L_m+L_p+W_s}$$ 1

Based on prior simulations, pairs were classified as monoclonal seeding if JSI ≤ 0.30 and polyclonal seeding if JSI > 0.30.

Primary tumor expansion age analysis

Primary tumor expansion age (T) was estimated under a Gompertz growth model with a fixed carrying capacity $k={10}^{11}$ cells. To incorporate uncertainty in tumor size at diagnosis and tumor doubling time, we derived empirical priors from a literature review for tumor volume at time of diagnosis in cm³ ($N_p$) and doubling time (DT) which were weighted by the number of tumors contributing to each estimate. We fit bounded kernel density estimates to these weighted priors and drew 5000 bootstrap samples per patient. Tumor volumes were converted to cell counts assuming ${10}^8$ cells per cm³. For each bootstrap draw, we computed the Gompertz growth rate parameter $\beta$ from the sampled doubling time (DT) and sampled diagnostic size ($N_p$) using the formula as shown in Eq. (2):

$$\beta =\frac{\mathit{ln}\left[\frac{\mathit{ln}\left(K\right)-\mathit{ln}\left(N_p\right)}{\mathit{ln}\left(K\right)-\mathit{ln}\left(2N_p\right)}\right]}{\mathit{DT}}$$ 2

We then derived $\alpha =\log \left(k\right)\beta$ and calculated expansion age as shown in Eq. (3):

$$T=-\frac{1}{\beta }\mathit{ln}(\mathit{1}-\frac{\mathit{ln}\left({\mathit{N}}_{\mathit{p}}\right)\mathit{\beta }}{\mathit{\alpha }})$$ 3

Bootstrap draws were retained only if $\beta >0$, $T>0$, and $T$ was less than the patient’s age at diagnosis. We summarized the posterior distribution of T per patient using the median and 95% interval (2.5th–97.5th percentiles). To visualize growth trajectories, we computed expected tumor size over time for each patient using Eq. (4):

$$S\left(t\right)=e^{\left(\frac{\alpha }{\beta }\left[1-e^{-\beta t}\right]\right)}$$ 4

evaluated on a fine time grid (0.01 years) for the median parameter sets. Time axes were shifted so that 0 corresponded to diagnosis (negative values denote years prior to diagnosis).

Metastatic seeding time analysis

Metastatic seeding time ($T_s$) for each primary–metastasis pair was estimated by propagating uncertainty in both the primary expansion age and mutational counts. For each pair, we combined the patient-specific posterior draws of $\mathit{T}$ with (i) an $\alpha$ prior drawn from a normal distribution N(0.13, 0.02) as in Hu et al.³⁶ where $\alpha$ represents the rate at which the primary tumor acquired private clonal mutations and (ii) Poisson resampling of observed mutation counts $L_m$ and $L_p$ to reflect counting uncertainty. Pairs with fewer than three primary-private clonal mutations were excluded. For each of 5000 aligned draws we computed $T_s$ as shown in Eq. (5):

$$T_s=\left(1-\alpha \frac{L_m}{L_p}\right)T$$ 5

retaining only draws where $0\le T_s\le$ age at diagnosis. For each pair, we reported the median $T_s$ and 95% interval (2.5–97.5th percentiles).

Quantification and statistical analysis

Categorical variables were reported as frequencies with percentages. Continuous variables were reported as mean with standard deviation (SD) or median with interquartile range (IQR) for normally distributed and non-normally distributed variables, respectively. Continuous parametric variables were compared as a difference of means using a two-tailed Student’s t test; continuous non-parametric variables were compared using a Wilcoxon–Mann–Whitney test. Categorical variables were compared using the Chi-Square test or Fisher’s exact test, as appropriate for group size. α was set at 0.05, and p < 0.05 was considered significant. Kaplan–Meier survival analysis was performed. The primary outcomes included recurrence free survival (RFS) which was defined as the time from diagnosis to radiologic or biochemical evidence of recurrent PCC/PGL, and overall survival (OS) which was defined as the time from diagnosis to death due to PCC/PGL. Statistical inference was performed using R (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria).

Author contributions

Conceptualization: A.M.P., H.W., K.L.N. Software: A.M.P., B.W., J.P. Formal analysis: A.M.P., J.P., B.W., H.W. Investigation: K.D.A., W.A., J.S. Resources: D.L.C., B.W., B.B., M.B. Data curation: A.M.P., B.W., J.P. Writing – original draft: A.M.P., H.W. Writing – review and editing: All authors. Visualization: A.M.P., H.W. Supervision: K.L.N., H.W. Funding acquisition: H.W.

Data availability

Whole-exome sequencing data are available via dbGaP (registry #61171). The analytic code is available at: https://github.com/ampregnall/Metastatic-PPGL-Genomics.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41698-026-01291-7.