Identification of a novel RASD1 somatic mutation in a USP8-mutated corticotroph adenoma

Cushing's disease (CD) is caused by pituitary corticotroph adenomas that secrete excess adrenocorticotropic hormone (ACTH). In these tumors, somatic mutations in the gene USP8 have been identified as recurrent and pathogenic and are the sole known molecular driver for CD. Although other somatic mutations were reported in these studies, their contribution to the pathogenesis of CD remains unexplored. No molecular drivers have been established for a large proportion of CD cases and tumor heterogeneity has not yet been investigated using genomics methods. Also, even in USP8-mutant tumors, a possibility may exist of additional contributing mutations, following a paradigm from other neoplasm types where multiple somatic alterations contribute to neoplastic transformation. The current study utilizes whole-exome discovery sequencing on the Illumina platform, followed by targeted amplicon-validation sequencing on the Pacific Biosciences platform, to interrogate the somatic mutation landscape in a corticotroph adenoma resected from a CD patient. In this USP8-mutated tumor, we identified an interesting somatic mutation in the gene RASD1, which is a component of the corticotropin-releasing hormone receptor signaling system. This finding may provide insight into a novel mechanism involving loss of feedback control to the corticotropin-releasing hormone receptor and subsequent deregulation of ACTH production in corticotroph tumors.


INTRODUCTION
Cushing's syndrome is caused by chronic exposure to elevated glucocorticoids via exogenous and endogenous sources. The typical clinical features of Cushing's syndrome are mutation in the GTP-binding site of the gene RASD1 that we hypothesize contributes to the pathogenesis of CD in this patient because of the involvement of RASD1 in regulation of ACTH production by glucocorticoid feedback. Additionally, based on the allelic fractions of the mutations, these tumor cells exhibit either subclones or polyclonal origin. These findings challenge the current model that corticotroph adenomas are genetically homogeneous.

Clinical Presentation and Family History
A 32-year-old woman with no known family history of pituitary diseases, who was one of identical triplet sisters, presented with a history of recently diagnosed type 2 diabetes mellitus and weight gain, easy bruising, and subjective plethora. On review of systems she endorsed occasional acne, moodiness before her menstrual period, and chronic insomnia. She denied neuromuscular weakness, significant mood changes, difficulty concentrating, or hirsutism. Evaluation confirmed CD: 24-h urinary free cortisol (UFC) levels were 154 and 90.4 µg (nl < 50 µg); midnight salivary cortisol (MSC) levels were 0.118, 0.142, and 0.917 µg/dl (nl < 0.112 µg/dl); serum cortisol was 4.1 µg/dl after 8 mg dexamethasone (nl < 1.8 µg/dl); random morning plasma ACTH level was 50 pg/ml; and pituitary magnetic resonance imaging (MRI) showed a clearly defined right-sided 4-mm lesion (Fig. 1). She was also found to have dyslipidemia and fatty liver on MRI of the abdomen. She was diagnosed with CD and underwent transnasal transsphenoidal adenomectomy without complication. Immunohistochemistry confirmed a corticotroph adenoma (Fig. 2) with a low MIB-1 index. Pathologic examination of tissue slices estimated 40%-50% tumor cellularity.

Genomic Analyses
WES of the blood-derived normal DNA and formalin-fixed paraffin-embedded (FFPE)derived tumor DNA from the patient was carried out on the Illumina HiSeq 2500 platform, yielding a mean sequencing depth of 145× and 315×, respectively, that was usable for variant calling (Table 1). Germline (constitutional) variants and somatic mutations were called; supporting binary alignment (BAM) read alignments for each somatic mutation  (Robinson et al. 2011;Thorvaldsdóttir et al. 2013), yielding 36 passing calls whose class, type, and trinucleotide context are depicted in Figure 3 (the variant genomic coordinates are provided as Supplemental File 1). C>T transitions were the most common single-nucleotide variant (SNV) type, as is common in many cancer mutation signatures (Alexandrov et al. 2013); although importantly Alexandrov et al. (2013) did not include mutations outside protein-coding exons, which are included in Fig. 3. Notably, the mutations were dominated by deletions, only two of which were in protein-coding regions. No insertions were observed (Table 2).  Picard terminology is as defined on https://broadinstitute.github.io/picard/picard-metric-definitions.html, with the Picard output field name given in parentheses. "Duplication" includes both optical/sequencing and polymerase chain reaction (PCR) duplicates according to Picard and is computed on the entire genome (i.e., not just in exome target regions). "Usable" read bases are bases aligned to exome-target regions and not in duplicate reads.
Novel RASD1 mutation in a corticotroph adenoma

Molecular Case Studies
After variant annotation, 13 of 36 somatic mutation calls were predicted to alter the amino acid sequence of a protein isoform (Table 2). This count of 13 protein-altering mutations is consistent with the low counts previously reported in other WES studies of corticotroph adenomas: median 7, range 3-23 (Reincke et al. 2015) and median 5, range 1-9 (Ma et al. 2015). The difference with Ma et al. (2015) can be explained by the higher sequencing depth in our study and differences in variant-calling procedure described in Ma et al. (2015); calls with allelic fraction of <20% were discarded, which would have eliminated most of our calls (Table 2). Validation was carried out on 12 of the 13 mutations via targeted amplicon sequencing on a second next-generation sequencing (NGS) platform (Pacific Biosciences RSII, PacBio), confirming 100% of the attempted mutations as somatic (present in tumor, absent in normal) and also obtaining similar allelic fractions to the original Illumina-based calls. Following the heuristic from Alexandrov et al. (2013) that WES Figure 3. (A) Distribution of the 36 somatic mutation calls passing manual review by type and class. All passing calls were included regardless of predicted impact (i.e., synonymous, intronic, and intergenic calls are included). For single-nucleotide variants (SNVs), type is given by the pyrimidine of the mutated base pair as per previous conventions (Alexandrov et al. 2013). (B) SNVs from A were further categorized into 96 trinucleotide classes defined by the immediately flanking bases (as previously described in Alexandrov et al. 2013).
Novel RASD1 mutation in a corticotroph adenoma     interrogates ∼30 megabases (Mb) of protein-coding exons in the human genome, we obtained a somatic mutation rate of 0.43 mutations/Mb for protein-altering mutations and 0.53 mutations/Mb for mutations in coding exons (13 protein-altering; three synonymous). When the latter is compared with somatic mutation rates across many cancer types in Alexandrov et al. (2013), the somatic mutation rate in our patient was low, a finding consistent with previous WES studies of corticotroph adenomas (Ma et al. 2015;Reincke et al. 2015).
The allelic fractions of somatic mutations ranged from 3% to 26%. Under the hypothesis that one of these mutations initiates or precedes neoplastic transformation of a single cell that then undergoes clonal expansion, the tumor purity of the sequenced specimen may therefore be <52%, consistent with the estimate of 40%-50% from pathologic examination. Although the presence of copy-number alterations (CNAs) can confound such an estimate, we did not observe aneuploidy or CNA events that confound the allelic fractions in Table 2 (using the SAAS-CNV tool [Zhang and Hao 2015] as before ]; data not shown).
We cross-referenced our somatic mutation calls with the Catalogue of Somatic Mutations in Cancer (COSMIC) database (Forbes et al. 2015) to determine whether any had previously been observed in a tumor; no exact matches were found and one approximate match is noted in Table 2. We also cross-referenced these calls with ClinVar (Landrum et al. 2016), as some variants are known to occur as both somatic driver mutations and germline variants in inherited/familial neoplasm syndromes. We then reviewed the gene annotations in the list of 13 protein-altering mutations in the context of potential involvement in the molecular pathways implicated in CD or in corticotroph tissues in general. Notably, our patient's adenoma had the USP8 p.P720R mutation that has previously been described as involved in the pathogenesis of CD (Ma et al. 2015;Reincke et al. 2015), occurring in 35%-62% of CD-causing corticotroph adenomas (Perez-Rivas and Reincke 2016); this was also the sole mutation identified in the ClinVar cross-reference. The high allelic fraction of this mutation (20%-22%, depending on sequencing platform; third highest on the list) is consistent with the hypothesis that it is a driver mutation present early in the clonal expansion of the tumor.
A novel mutation, p.K34M, in the GTP-binding region of RASD1 was identified in this tumor, at an allelic fraction (3%) indicative of a subclone with respect to cells containing USP8 p.P720R. RASD1 was originally discovered as an inducible gene in dexamethasone-stimulated AtT-20 mouse corticotroph cells (Kemppainen 1998). Based on this connection to the cell type under study, further computational analysis of the functional significance of this mutation was conducted. Alignment of several related small GTPases, including the well-studied oncogenes KRAS, NRAS, and HRAS, revealed that p.K34M is found within the G1 motif and is likely involved in binding to GTP (Fig. 4). However, the precise contacts between the enzyme and substrate are not clear given that the G1 motif diverges in the RASD family from the other small GTPases for which substrate-bound crystal structures are available. Other genes in Table 2 were reviewed for possible connections to CD or corticotroph biology but were not judged to have a direct connection like USP8 or RASD1; hence, these two genes are the focus of this report.

Treatment Outcomes
Consistent with successful removal of the patient's ACTH-secreting tumor, her postoperative day 1 serum cortisol level was 3.4 µg/dl. She was discharged on physiologic oral hydrocortisone replacement and subsequently tapered off after 4 mo. Following discontinuation of oral hydrocortisone, UFC, 8 mg dexamethasone suppressed serum cortisol, and MSC concentrations were all within the normal range, consistent with CD remission. CD symptoms Novel RASD1 mutation in a corticotroph adenoma

G4 motif:
[NT]KXD  CaaX amino-terminal motif that undergoes posttranslational modification (a denotes any aliphatic amino acid). RASD2, encoding the protein Rhes, is shown because it is the closest human homolog to RASD1 (63% protein sequence identity); the two form the RASD subfamily that is distinct from other Ras family proteins (<45% protein sequence identity). DIRAS1 and DIRAS2 are shown because they are the next closest homologs to RASD family proteins. NRAS, HRAS, and KRAS are shown because they are well-characterized oncogenes. Functional impact of the RASD1 mutations (in red) has been demonstrated experimentally: p.G31V (Cismowski et al. 1999(Cismowski et al. , 2000Vaidyanathan et al. 2004), p.G36V (Cismowski et al. 1999), p.A178V (Graham et al. 2001), and p.C278S (Graham et al. 2001;Vaidyanathan et al. 2004). Mutations in NRAS, HRAS, and KRAS (in red) are widely known oncogenic mutations and are also recurrent somatic mutations across multiple neoplasm types in COSMIC (Forbes et al. 2015(Forbes et al. ) (accessed 2016, except for HRAS amino acid A146 (no mutations of any type in COSMIC, although p.A146V results in constitutive activation [Feig and Cooper 1988] and may be germline pathogenic in Costello syndrome, ClinVar accession RCV000013445.18), and NRAS amino acid K117 (no mutations of any type in COSMIC and no published evidence on any K117 mutation). Amino acid ranges are given in parentheses next to the gene symbols. Amino acids are color-coded according to biochemical class (yellow, nonpolar; green, polar; blue, basic; pink, acidic). For positions where one of several amino acids is possible, the possibilities are given in brackets. X, any amino acid; LOF, loss of function; GOF, gain of function.
Novel RASD1 mutation in a corticotroph adenoma

DISCUSSION Tumor Genetic Heterogeneity Models
The current case identifies a novel RASD1 mutation in a USP8-positive corticotroph adenoma. The different allelic fractions between the USP8 and RASD1 somatic mutations in the studied tumor cells suggest that this ACTH-secreting tumor is genetically heterogeneous. Two models for heterogeneity are proposed as follows. In model A (Fig. 5A), the USP8 and RASD1 mutations may be synergistic, with the USP8 mutation occurring early in tumorigenesis, leading to abnormal proliferation of ACTH-secreting cells. At a later time point in the pathogenesis of the tumor, one of these USP8-mutant cells acquires a RASD1 mutation, resulting in a subclone of RASD1-mutant/USP8-mutant cells. In model B (Fig. 5B), the USP8 and RASD1 mutations are mutually exclusive, giving rise to subclones of cells with different mutation combinations (i.e., USP8-mutant/RASD1-wild type vs. USP8-wild type/RASD1-mutant). In both models, the tumor is genetically heterogeneous. Our hypothesis is that under both these models, RASD1 is a contributor to cell proliferation and ACTH secretion, but occurs in a small subpopulation of the tumor cells. Although these findings do not clearly distinguish between monoclonal versus polyclonal origin of the tumor, they nevertheless indicate that the tumor is genetically heterogeneous and suggest further studies into the interplay between multiple possible drivers.

Rationale for RASD1 as a Contributor to Pathogenesis
This study identified a mutation in RASD1 that may alter binding to GTP on the basis of RASD1 structural homology with well-studied, oncogenic small GTPases KRAS, NRAS, and HRAS. It is appealing to speculate that RASD1 p.K34M will have a reduced affinity to GTP and will therefore be less active (decreased capacity to interact with downstream proteins) versus wild type. However, because of the low allelic fraction (presumed subclonal nature) of this mutation, we cannot conclusively determine whether the mutation is homozygous or heterozygous. Several studies have defined a role for RASD1 in inhibition of Gα s signaling (Graham et al. 2001(Graham et al. , 2004. This may occur through an interaction with Gα i (Cismowski et al. 2000) . Normal feedback regulation within the hypothalamic-pituitary-adrenal axis

Molecular Case Studies
involves glucocorticoid induction of genes associated with suppression of corticotropinreleasing hormone receptor (CRHR) signaling. RASD1 was identified by its virtue of being strongly induced by dexamethasone in mouse corticotroph cell lines and pituitaries (Kemppainen 1998;Tu and Wu 1999;Brogan et al. 2001). Taken together, RASD1 may be a transcriptionally inducible negative regulator of CRHR-Gα s signaling in corticotrophs. In cells with nonfunctional RASD1, this loop is no longer intact and CRHR signaling may become insensitive to negative feedback from glucocorticoids, thereby allowing for continued ACTH secretion. This alteration within corticotroph signaling and regulation may occur within the population of USP8 mutant cells or in a separate subclone. Separate molecular signaling schemes for these situations are depicted in Figure 6, illustrating the effect on the promoter of the gene POMC (whose protein product is processed to become ACTH, which is then secreted). It is intriguing to consider the possibility that loss of RASD1 function and mutation of USP8 may be additive or synergistic in relation to the pathophysiology of CD (Fig. 6D). However, since we did not carry out experimental validation of RASD1 p.K34M function for this study, we must caution that our claims regarding its involvement in disease biology are only hypothetical and based mainly on what is known about RASD1 biology from prior studies. Further studies should investigate the functional significance of this novel RASD1 mutation in the pathogenesis of ACTH tumors.

METHODS WES and Targeted Validation
Paired-end (2×100 bp) WES on Illumina HiSeq 2500 (Illumina) and targeted amplicon validation on PacBio RSII (Pacific Biosciences) was carried as previously described , with the following modifications. For WES, the SureSelect Human All Exon V5 hybridization capture system (Agilent) was used. Libraries from three tumor and nine normal samples were multiplexed in a 2:1 tumor:normal ratio and sequenced on all eight lanes of a High Output flow cell; only two of these samples are presented in this work (others to be published in a future work).

Variant Calling
WES FASTQ files from the normal and tumor sample were combined into a patient-specific "cohort" and run through an in-house pipeline (Linderman et al. 2014)

Molecular Case Studies
SNV calls from Mutect (N = 158) and indel calls from Varscan2 (N = 97) were loaded into a custom MySQL (Percona MySQL Server Community Edition 5.6.14-rel62.0.483.rhel6) database schema using in-house scripts and annotated using RVS (Hakenberg et al. 2016) and SnpEff 4.0b (Cingolani et al. 2012) using the Ensembl (Aken et al. 2016) version 75/ GRCh37 resource bundle. Somatic calls whose population allele frequency in Exome Aggregation Consortium (ExAC) (Lek et al. 2016) exceeded 1% were discarded on the presumption that they are any combination of contamination, a variant present but missed in the normal sample, a low-level artifact could not be pathogenic because it was too common in general population. All remaining SNV (N = 152) and indel (N = 64) calls were manually reviewed in IGV (Robinson et al. 2011;Thorvaldsdóttir et al. 2013) and the UCSC Genome Browser (Rosenbloom et al. 2015) to inspect supporting alignment quality in the BAM files and mappability of the genomic region in the hg19 human genome assembly, paying attention to whether a variant call was located in a short tandem repeat or a low-complexity sequence region (Benson 1999), a region with self-homology/duplication in the reference genome, or a region of low alignability according to the GEM track from ENCODE/CRG (Derrien et al. 2012). Uncertain calls, many of which were due to a low-level C>A substitution artifact also present in the normal or due to artifacts in padding regions, were manually rejected at this step, resulting in a final list of 25 SNV and 11 indel calls, which are shown in Figure 3, the protein-altering subset of which is shown in Table 2. The multiple sequence alignment for Figure 4 was made using the European Molecular Biology Laboratory European Bioinformatics Institute (EMBL-EBI) Clustal Omega web tool (Goujon et al. 2010;Sievers et al. 2011) (http://www.ebi.ac.uk/Tools/msa/clustalo/, used 2016-09-24, default settings) and edited using Unipro UGENE v1.24.2 (Okonechnikov et al. 2012). Protein sequence identity of the RASD subfamily was determined by blastp of RASD1 and RASD2 against all human proteins in the RefSeq protein database (Altschul et al. 1997(Altschul et al. , 2005 (http://blast.ncbi.nlm.nih.gov, used 2016-09-25).

ADDITIONAL INFORMATION Data Deposition and Access
All somatic mutation calls passing manual review (including those predicted to not alter protein) are provided as Supplemental File 1 and have been submitted (COSP42647) to Novel RASD1 mutation in a corticotroph adenoma C O L D S P R I N G H A R B O R Molecular Case Studies the COSMIC database (Forbes et al. 2015). Consent could not be obtained for public release of raw sequencing data.

Ethics Statement
The study was approved by the Institutional Review Board at the Mount Sinai Medical Center. The patient gave written informed consent before participation, including permission to publish the results.