1 INTRODUCTION

Phaeochromocytomas and paragangliomas (PPGL) are rare neuroendocrine tumours that arise from chromaffin tissue from the adrenal medulla (phaeochromocytoma) or neural crest progenitors of extra-adrenal sympathetic or parasympathetic paraganglia (paraganglioma).1, 2 The clinical signs and symptoms vary according to the localisation of the tumours and to their hormonal activity. Treatment options include surgery, peptide receptor radionuclide therapy, targeted therapies, chemotherapy and radiotherapy. Morbidity and mortality are high in patients with metastatic disease, which account for 10%–20% of PPGL.3, 4

PPGL are considered to be the most heritable tumours and over the past two decades, the identification of more than a dozen PPGL susceptibility genes5 has transformed the management of PPGL patients. More recently, the significant proportion of germline negative PPGL has motivated interest in the role of somatic sequencing for PPGL in both research and clinical settings.6-8 Furthermore, tumour sequencing has become more amenable in the era of next-generation sequencing (NGS), which offers a faster, cheaper and higher throughput option to the conventional method of Sanger sequencing. Custom NGS panels for tumour have followed in the successful path of germline targeted assays and testing can be applied to paraffin-embedded tissues as well as fresh frozen samples.7

In 2017, The Cancer Genome Atlas provided a comprehensive genomic characterisation by analysing a cohort of 173 patients with PPGL8 of which 27% of patients had a germline and 39% a somatic genome alteration. At the somatic level, five PPGL driver genes (HRAS, NF1, EPAS1, RET and CSDE1) and eight hotspots and cancer-relevant genes (BRAF, IDH1, FDFR1, VHL, ATRX, TP53, SETD2 and ARNT) were identified and in some tumours an overexpression of MAML3 fusion genes was also noted.8 Two subsequent studies confirmed the presence of a somatic driver mutation in 32% and 37% of PPGL patients with NF1, HRAS, RET and VHL being the most frequently affected genes.7, 9 Notable findings in previous studies were the association of somatic ATRX variants with aggressive tumour behaviour and the detection of mosaic mutations in SDHB and VHL.9, 10 Mosaicism may be underestimated in patients with PPGL if germline DNA alone is tested.

On the basis of their underlying driver mutation at a germline or somatic level, PPGL can be divided into three main clusters.8 The first cluster includes tumours with mutations in citric acid cycle genes such as SDHx, FH, MDH2, as well as VHL gene mutations. The transcriptional signature of ‘cluster 1’ tumours is defined by abnormal stabilisation of HIF alpha transcription factors leading to pseudohypoxia.6 ‘Cluster 2’ tumours are characterised by an upregulation of kinase signalling pathways involving the mitogen-activated protein kinase pathway and the mechanistic target of rapamycin (mTOR) pathway and include mutations in genes such as RET, NF1, TMEM127 and MAX. Finally, the third cluster is defined by activation of the Wnt/beta-catenin pathway. Perturbations in this pathway have been exclusively described in sporadic PPGL with somatic variants in CSDE1 and MAML3 fusion genes.8

Despite significant advances in our understanding of PPGL tumourigenesis, a number of barriers to optimal clinical practice still exist. First, risk stratification and prediction of malignant potential have remained a major challenge.10, 11 With the exception of germline SDHB mutations, no robust molecular marker for the aggressive disease is currently known.12 This poses a challenge for clinical surveillance practices as potential metastatic cases may have no germline genetic diagnosis.13, 14 A lack of effective treatment options for PPGL is another significant unmet need in clinical practice.15, 16 Precision therapeutics based on molecular tumour characteristics are a desirable and crucial next step to improving outcomes and quality of life for patients with these rare tumours.17

The primary aim of this study was to explore the prevalence and role of somatic driver variants in a large UK cohort of patients with PPGL using an NGS strategy to analyse ‘mutation hotspots’ in 68 human cancer genes.

2 MATERIALS AND METHODS 2.1 Study design and participants

Two separate cohorts were recruited between 2018 and 2021. For the development cohort, patients from Cambridge University Hospitals, Guy's and St. Thomas' NHS Foundation Trust, London and from St. Bartholomew's Hospital in London were included. For the validation cohort, tumour samples were recruited from different PPGL referral centres across Great Britain. For both cohorts, the diagnosis of PPGL was based on procedures provided by international clinical practice guidelines18, 19 and was confirmed by histology in every case.

All patients provided written informed consent for sample and data collection as well as genetic testing (South Birmingham REC and East of England—Cambridge South REC, reference number: 5175 and East London and Cambridge East MREC 06/Q0104/133).

2.2 Development cohort

Tumour and matched germline DNA samples were prospectively collected from patients with a new diagnosis of PPGL who underwent surgery or patients under ongoing clinical care for whom tumour tissue was available. Both sporadic and familial cases were included. Detailed clinical information (i.e., sex, age of onset, tumour localisation and extension, metastatic disease, secretion pattern and family history) was collected. In June 2021, follow-up information including recurrent disease (multiple tumours or metastatic disease) and survival was assessed for all patients.

2.3 Validation cohort

Tumour DNA samples were retrospectively collected from patients with sporadic and hereditary PPGL tumours. Matched germline DNA was not available for these patients. Clinical information including sex, age of onset and primary localisation of the tumour was accessible, but other clinical characteristics and follow-up data were not available. Results of germline genetic testing were collected when possible.

2.4 Targeted gene panel and sequencing technique

Tumour and matched germline DNA were sequenced and analysed using a custom-designed NGS panel based on the Ion AmpliSeq™ Cancer Panel covering ‘mutation hotspots’ in 68 human cancer genes and additional bespoke content to cover all exons and flanking sequences of 12 PPGL-related genes plus EPAS1- and VHL-targeted exons (Table S1).

2.5 Bioinformatics analysis

All samples were aligned to the hg38 version of the reference human genome using bwa 0.7.17 in alt contig aware mode as described by the authors.20 The generated SAM file was compressed into a BAM file and sorted by genomic position using samtools 1.9.21 The sorted BAM files were subject to Base Quality Score Recalibration and Indel Realignment as specified in the Genome-Analysis Toolkit (GATK) 22 best practices.23, 24 For somatic variant calling the following GATK's MuTect26 was used. A panel of normals (PON) was generated using the germline samples with GATK's (version 4.0.3.0) Mutect2 and CreateSomaticPanelOfNormals algorithms. Variants were called in all tumours using the PON and the matched germline sample with the GATK's MuTect2 algorithm to generate a VCF file.25 Finally, the VCF files were filtered with GATK's FilterMutectCalls algorithm. The resulting VCF file was annotated and prioritised using annovar.26

2.6 Variant filtering

Synonymous variants and noncoding variants were removed. Variants were removed if the variant allele frequency was <10% or the minor allele frequency (MAF) greater than 0.1% in EVS6500 and/or 1000 Genomes. All variants with a read depth less than two standard deviations below the mean coverage (<500 reads) were filtered out. Variants in the intronic and intergenic regions, synonymous variants, variants which failed the ‘artefact-in-normal’ and ‘base quality’ (minimum base quality below 20) filters, were also discarded. Finally, variants that were classified as ‘benign’ or ‘likely benign’ on the Catalogue of Somatic Mutations in Cancer (COSMIC) (https://cancer.sanger.ac.uk/cosmic) or ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/) were removed. For those tumour samples without a matched germline, further variant filtering was performed if a common germline variant or single nucleotide polymorphisms was identified (Figure 1).

Flowchart for variant filtering and classification. *Minimum base quality below 20. **On the basis of the data from the Catalogue of Somatic Mutations in Cancer (COSMIC) (https://cancer.sanger.ac.uk/cosmic) or ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/). ***<500 reads was less than two standard deviations below the mean coverage). °Including multiple variants in the same tumour. ^Validation of suspected driver variants was performed using; (i) Sanger sequencing for 10 cases (D73,D77,D78,D79,D84,D87,D88,D95,D97,D98), (ii) Nuclear magnetic resonance spectroscopy to detect 2-hydroxyglutarate for case D84, (iii) hybrid capture-based sequencing for case D87, and (iv) SDHB immunohistochemistry for case D77, D86)

2.7 Variant classification

For the purpose of this study, a somatic variant was defined as a potential driver variant if the variant allele frequency was >10%. Sanger sequencing validation was performed on 10 samples with suspected somatic driver variants. Other validation methods including SDHB immunohistochemistry, ex-vivo tumour metabolomics using NMR spectroscopy and hybrid capture-based sequencing were performed on single tumour samples to validate specific somatic driver variants. Identified driver variants were classified as; pathogenic, likely pathogenic or a variant of uncertain significance (VUS) based on evidence available from the Catalogue of Somatic Mutations in Cancer (COSMIC) (https://cancer.sanger.ac.uk/cosmic) or ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/), dbSNP, Single Nucleotide Polymorphism database (http://www.ncbi.nlm.nih.gov/snp); EVS, exome variant server (http://evs.gs.washington.edu/EVS); ExAC, Exome Aggregation Consortium (http://exac.broadinstitute.org); LOVD, Leiden Open (source) Variation Database (http://www.lovd.nl).

2.8 Statistical analysis

Statistical tests were performed using the statistical software package R (R Foundation for Statistical Computing). Summary statistics include median and interquartile ranges (IQR) for continuous variables and frequency and percentage for categorical variables.

For detailed information about study methodology please see the Supporting Information Appendix.

3 RESULTS

One hundred patients were analysed in the development cohort and 41 patients in the validation cohort.

3.1 Baseline characteristics of the pooled cohort

In the pooled patient data set (development and validation cohort), 76 (54%) patients were male and the median (IQR) age at diagnosis was 47 (36, 62). Germline genetic testing results were available for all but one patient in the development cohort, and for 31 (76%) cases in the validation cohort. A third of tested patients (45/130, 35%) harboured a germline mutation, most frequently in the SDHx genes. The most frequent tumour location was adrenal in 89 (63%) patients, followed by an extra-adrenal abdominal location in 34 (24%) patients. Multiple tumours were noted in 10 (7%) patients and median [IQR] tumour size was 44.5 mm (31.5, 62.5).

In the development cohort, 52 patients (52%) had a noradrenaline-only secreting tumour, 16 patients (16%) had metastatic disease and 3 patients died from metastatic PPGL during the study period. Baseline characteristics of the pooled patient data set as well as the individual development and validation cohorts are shown in Table 1.

Table 1. Baseline characteristics of study patients Pooled cohorts Development cohort Validation cohort Number of patients 141 100 41 Sex (male), n (%) 76 (54) 55 (55) 21 (51) Age at diagnosis (years), median [IQR] 47 [36, 62] 48 [37, 66] 42 [35, 49] Genotype, n (%) No mutation 85 (61) 67 (67) 18 (44) Mutation 45 (32) 32 (32) 13 (32) No information 11 (8) 1 (1) 10 (24) Genotype affected gene, n (%) SDHB 19 (14) 12 (12) 7 (17) SDHD 6 (4) 2 (2) 4 (10) SDHA 6 (4) 5 (5) 1 (2) SDHC 2 (1.4) 2 (2) 0 (0) VHL 4 (3) 3 (3) 1 (2) TMEM127 3 (2) 3 (3) 0 (0) RET 2 (1.4) 2 (2) 0 (0) NF1 1 (0.7) 1 (1) 0 (0) MAX 1 (0.7) 1 (1) 0 (0) FH 1 (0.7) 1 (1) 0 (0) Tumour localisation, n (%) Adrenal 89 (63) 67 (67) 22 (54) Extra-adrenal abdomen 34 (24) 24 (24) 10 (24) Extra-adrenal mediastinum 2 (1.4) 2 (2) 0 (0) Head and neck 14 (10) 5 (5) 9 (22) Bladder 2 (1.4) 2 (2) 0 (0) Multiple tumours, n (%) 10 (7) 8 (8) 2 (5) Maximum tumour size (mm), median [IQR] - 44.5 [31.5, 62.5] - Metastatic disease, n (%) - 16 (16) - Death, n (%) - 3 (3) - Secretory pattern, n (%) - - Nonfunctional - 9 (9) - Adrenaline - 4 (4) - Noradrenaline - 52 (52) - Mixed - 32 (32) - Family history, n (%) - 11 (11) - 3.2 Somatic and matched germline sequencing

Tumour DNA (all primary tumours) was extracted from paraffin-embedded tumour samples and fresh frozen tissue in 136 (96.5%) and five (3.5%) samples, respectively. Matched germline DNA was extracted from blood in 97 patients and from adjacent normal tissue in two patients of the development cohort.

3.3 Quality assessment of sequencing assay

The mean coverage calculated across all sequencing runs was 2171.64 reads, median coverage was 2402.86 and the standard error was 91.62482 (SD 1044.68). The coverage ranged from 30.66 to 7071.21 reads (see Figure S1). A higher frequency of C>T variants consistent with DNA damage from formalin fixation was noted in the FFPE samples, however, this mutational signature was not significant at a higher allele frequency (>5%).

3.4 Detection of somatic variants of the pooled cohort

Somatic sequencing revealed the presence of one or more potential somatic driver variants in 37 (26%) patients of the pooled cohort including 26 pathogenic variants and 19 variants of uncertain significance (see Figure 1). Excluding patients with VUS, 25 (18%) of patients were found to have one (except V36 had two) pathogenic or likely pathogenic variant.

The most frequent affected genes (affected by both pathogenic variants and VUS) were NF1 (n = 7), VHL (n = 5), HRAS (n = 4), EPAS1 (n = 4) and RET (n = 3). All but three somatic variants were detected in patients without a germline mutation (exceptions: D86 with germline and somatic SDHA variant and a VUS in KRAS, V11 with germline SDHD variant and somatic VUS in FH, V38 with germline SDHB variant and two somatic VUS in SDHA) (see Tables 2 and S3).

Table 2. Molecular classification of detected driver somatic variants in the development cohort ID Gene Variant rs ID Variant type Variant classification Variant allele frequency (%) Validated by Sanger sequencinga D2 HRAS c.182A>T, p.Gln61Leu rs121913233 Nonsynonymous Pathogenic 38 Yes D5 RET c.2753T>C, p.Met918Thr rs74799832 Nonsynonymous Pathogenic 39 No D10 RET c.2753T>C, p.Met918Thr rs74799832 Nonsynonymous Pathogenic 39 No D15 SDHD c.14G>A, p.Trp5Ter rs104894310 Stop gain Pathogenic 31 No D22 VHL c.371C>T, p.Thr124Ile rs193922610 Nonsynonymous Likely pathogenic 36 No D23 VHL c.250G>A, p.Val84Met rs5030827 Nonsynonymous Pathogenic 36 No D30 RET c.1898T>G, p.L633R - Nonsynonymous Uncertain 30 No D33 SDHA c.1679C>T, p.T560M rs775350508 Nonsynonymous Uncertain 14 No D40 EPAS1 c.1589C>T, p.A530V - Nonsynonymous Uncertain 51 No D51 BRAF c.1801A>G, p.K601E rs121913364 Nonsynonymous Uncertain 40 No D53 NF1 c.3338delT, p.L1113fs - Frameshift Likely pathogenic 45 No D54 NF1 c.2014G>T, p.G672X - Frameshift Likely pathogenic 60 No D56 NF1 c.3513delG p.K1171fs - Frameshift Likely pathogenic 10 No D61 KIF1B c.1204C>T, p.L402F rs764084679 Nonsynonymous Uncertain 30 No D62 EPAS1 c.1681C>T, p.Q561X - Stop gain Uncertain 22 No D65 FGFR3 c.1125T>A, p.Y375X - Stop gain Likely pathogenic 45 No D67 TP53 c.527G>A, p.C176Y rs786202962 Nonsynonymous Likely pathogenic 13 No D68 VHL c.386T>C, p.Leu129Pro rs1559428119 Nonsynonymous Uncertain 22 No D73 HRAS c.182A>C, p.Gln61Pro rs121913233 Nonsynonymous Likely pathogenic 29 No D77 SDHB c.423+1G>A rs398122805 Splice site Likely pathogenic 15 Yes D78 VHL c.482G>A, p.Arg161Gln rs730882035 Nonsynonymous Likely pathogenic 33 Yes D79 NF1

c.2927_2933delCTGAAGG,

p.Thr976fs

- Frameshift Likely pathogenic 36 Yes D84 IDH1 c.394C>T, p.Arg132Cys rs121913499 Nonsynonymous Likely pathogenic 40 Yes D86 KRAS c.88G>A, p.Arg115Leu - Nonsynonymous Uncertain 13 No D86 SDHA c.1270G>T, p.Glu424X - Stop gain Likely pathogenic 27 Yes D87 FBXW7 p.Cys384fs - Frameshift Uncertain 70 Yes D88 VHL c.245G>T, p.Arg82Leu rs794726890 Nonsynonymous Likely pathogenic 25 Yes D95 HRAS c.182A>C, p.Gln61Pro rs121913233 Nonsynonymous Likely pathogenic 41 No D97 NF1 c.7925delC, p.Ser2642fs - Frameshift Likely pathogenic 15 Yes D98 NF1 c.2098delA, p.Thr700fs - Frameshift Likely pathogenic 40 Yes Note: Clinical and genetic characteristics of the validation cohort are shown in Table S3. aYES means that Sanger sequencing was performed and the variant confirmed. NO means that Sanger sequencing was not performed.

Pathogenic variants in ‘cluster 1’ genes (e.g., SDHx, FH and VHL genes) were more frequent at the germline level, whereas ‘cluster 2’ genes (such as RET, HRAS and NF1) were most frequently mutated at the somatic level (see Figure 2).

Distribution of somatic and germline variants according to molecular cluster