Recurrent activating mutation in PRKACA in cortisol-producing adrenal tumors.

Adrenal tumors autonomously producing cortisol cause Cushing's syndrome. We performed exome sequencing of 25 tumor-normal pairs and identified 2 subgroups. Eight tumors (including three carcinomas) had many somatic copy number variants (CNVs) with frequent deletion of CDC42 and CDKN2A, amplification of 5q31.2 and protein-altering mutations in TP53 and RB1. Seventeen tumors (all adenomas) had no somatic CNVs or TP53 or RB1 mutations. Six of these had known gain-of-function mutations in CTNNB1 (β-catenin) or GNAS (Gαs). Six others had somatic mutations in PRKACA (protein kinase A (PKA) catalytic subunit) resulting in a p.Leu206Arg substitution. Further sequencing identified this mutation in 13 of 63 tumors (35% of adenomas with overt Cushing's syndrome). PRKACA, GNAS and CTNNB1 mutations were mutually exclusive. Leu206 directly interacts with the regulatory subunit of PKA, PRKAR1A. Leu206Arg PRKACA loses PRKAR1A binding, increasing the phosphorylation of downstream targets. PKA activity induces cortisol production and cell proliferation, providing a mechanism for tumor development. These findings define distinct mechanisms underlying adrenal cortisol-producing tumors.

Cortisol production is normally tightly regulated. Stress signals lead to hypothalamic release of corticotropin releasing hormone (CRH) and vasopressin, inducing pituitary secretion of adrenocorticotropic hormone (ACTH), which binds its G-protein coupled receptor (melanocortin receptor 2) in adrenal fasciculata, activating adenylate cyclase[16-19]. The resulting cAMP binds the regulatory subunit of PKA, inducing release of the PKA catalytic subunit, allowing phosphorylation of downstream targets such as CREB[20,21]. This induces cortisol biosynthesis and cell proliferation[21-23].

Cushing syndrome (CS) results from excessive cortisol production, causing metabolic abnormalities including central obesity, myopathy, diabetes, hypertension and osteoporosis[2,24]. This arises from tumors that constitutively produce CRH or ACTH, or from adrenal tumors that autonomously produce cortisol[1]. Somatic activating mutations in CTNNB1 (beta-catenin) or GNAS (encoding Gαs, which activates adenylate cyclase) have been found in a subset of these adrenal tumors[5-8]. Additionally, germline or mosaic mutations in GNAS, PRKAR1A (encoding the regulatory subunit of PKA), or cAMP phosphodiesterases (PDE11A or PDE8B) cause increased cAMP signaling and syndromes that include CS typically with adrenal hyperplasia.[11-13,15,25-27]. Thus, increased cAMP signaling in adrenal fasciculata is sufficient to increase cortisol production and adrenal cell mass.

Nonetheless, the pathogenesis of the vast majority of autonomous cortisol-producing adrenal tumors is unknown. We initially studied 25 patients diagnosed with autonomous cortisol-producing tumors. The diagnosis was based on elevated 24-hour urinary cortisol with suppressed ACTH level and/or pathological (i.e. non-suppressed) dexamethasone suppression test, and an adrenal tumor[28]. Eleven subjects had clinical stigmata of CS (Methods, Supplementary Table 1). Surgical pathology revealed tumors of mean size of 5.2 cm; 22 tumors were classified as adrenocortical adenomas (ACA) and 3 as adrenocortical carcinomas (ACC).

Exome sequencing on tumor and matched normal DNA was performed. 90% of the targeted bases in tumors had 20 or more independent reads (Supplementary Table 2). Mean tumor purity was estimated at 38%, and somatic single nucleotide variants (SNVs) and somatic copy number variants (CNVs) were called (see Methods). Mutations of interest were validated by direct Sanger sequencing.

The results identified two distinct groups of tumors with qualitative differences CNV number. Eight tumors (CNV+, including all three carcinomas) had 13.4 ± 3.9 CNVs (range 7–18, Supplementary Fig. 1) while 17 (CNV−) had zero. CNV+ tumors had a higher somatic SNV mutation rate (8.0 × 10−7 vs. 2.8 × 10−7; P = 1.3 × 10−3, Mann-Whitney U test) and more protein-altering somatic mutations per tumor (16.1 vs. 5.6; P = 8 × 10−4, Mann-Whitney U test; Fig. 1a). These differences remained significant when carcinomas were removed from analysis (mutation rate 5.4 × 10−7, P = 0.02; protein-altering mutations 10.4, P = 0 .01).

Figure 1

Somatic mutations in cortisol-producing adrenal tumors. (a) The number of somatic mutations in each tumor is shown. Tumors are either adrenocortical carcinomas (ACC, in orange), CNV+ (in off-white) or CNV− (in white). (b) Recurrent copy number variants and GISTIC peaks identified in tumors. Likely driver genes are listed in parenthesis. Red, amplification; blue, deletion. (c) Somatic protein-altering mutations in genes frequently mutated in tumors. Each track describes presence (colored) or absence (off-white, white) of mutation in tumor for gene specified.

CNV+ tumors had significant focal losses of 3 chromosome segments (GISTIC[29] q-value < 0.25; Supplementary Fig. 2). Deletions of 1p36.12 and 9p21.3 included cancer-related genes CDC42 and CDKN2A, respectively; 10p12.33 deletion contained miR-511, which shows reduced expression in ACCs[30]. A focal gain at 5q31.2 (q-value = 0.16) included WNT8a. Wnt signaling is a proliferative signal in adrenal cortex[5], and gain of function mutations in beta-catenin are common in adrenal tumors[31]. Two CNV+ tumors had known activating mutations in CTNNB1 (Supplementary Table 3).

CNV+ tumors also had somatic mutations in cancer-related genes TP53 and RB1, in all cases in segments with loss of heterozygosity (LOH; Fig. 1c). Mutations in these genes were not found in CNV− tumors (P = 5.5 × 10−3, Fisher’s exact test). LOH at TP53 in all cases resulted from loss of one copy of the entire chromosome 17 (Supplementary Fig. 1); TP53-associated cancers such as colon, lung and breast typically have deletion confined to 17p[32-35]. Interestingly, 17q includes PRKAR1A raising the question of whether heterozygous deletion of PRKAR1A might contribute to cortisol excess in these tumors. No other genes were significantly mutated (Supplementary Table 3). Tumors clinically classified as adenomas in the CNV+ group all had Weiss scores of 0 or 1, and, where studied, low Ki67 staining (Supplementary Table 1), providing no histologic evidence of carcinoma.

The remaining 17 tumors had a mean of only 1.6 silent and 5.6 protein-altering mutations (Fig. 1a). Known gain of function somatic mutations[31,36]. were found in CTNNB1 (four tumors; p.Ser45Pro, p.Asp32Gly) or GNAS (two tumors; p.Arg201His, p.Gln227Glu) (Supplementary Table 3).

Additionally, the identical somatic A>C mutation in PRKACA (encoding the alpha isoform of the catalytic subunit of cAMP-dependent protein kinase A), resulting in a p.Leu206Arg (NM_002730.3:c.617A>C) substitution (Fig. 2a, Supplementary Table 4), was found in 6 CNV-tumors. This mutation has not been previously reported (absent from >10,000 exomes in public and Yale databases, and the Catalogue of somatic mutations in cancer, COSMIC). The read counts in all cases were consistent with heterozygous somatic mutations. These mutations were mutually exclusive with CTNNB1 and GNAS mutations (Fig. 1c, P = 0.043, Fisher’s exact test). Both wildtype (WT) and mutant PRKACA transcripts were detected in tumor cDNA (Supplementary Fig. 3). No other genes were mutated in more than one tumor.

Figure 2

PRKACA Leu206 interacts with PRKAR1A. (a) Representative chromatograms of Sanger DNA sequence of normal and tumor DNA for PRKACA codons 205–207; the tumor shows a heterozygous somatic p.Leu206Arg mutation. (b) Multiple sequence alignment of the activation loop of PRKACA. This segment is highly conserved from yeast to humans, with leucine at the position orthologous to Leu206 through invertebrates, and isoleucine in yeast. (c) Crystal structure of PRKACA and inhibitor (PDB: 1ATP)[9]. The inhibitory peptide (Arg94-Arg-Asn-Ala-Ile98) is shown in cyan while the P+1 loop of PRKACA is shown in yellow. Leu206 and Leu199 of PRKACA make van der Waals contacts with Ile98 of the inhibitory peptide, while arginines of the inhibitory peptide form salt bridges with glutamates of the catalytic cleft[10,38,39]. (d) Schematic illustration of the key interactions between the regulatory and catalytic subunit (adapted from ref. 10). (e) Close-up view of the interaction between Ile98 and Leu206, and (f) likely rotamer with Arg206 substitution.

Segments of PRKACA, GNAS and CTNNB1 harboring gain of function mutations were amplified by PCR of DNA from 31 additional cortisol-producing adrenal tumors and matched germline samples (available for 20 subjects) followed by Sanger sequencing (Supplementary Table 1). Seven additional tumors without matched normal DNA had exome sequencing. These comprised 33 ACAs and 5 ACCs. The PRKACA p.Leu206Arg mutation was identified in 7 additional tumors, collectively in 13 of 63 tumors (21%; 24% of all ACAs; 35% of ACAs associated with overt CS). The probability of finding the identical somatic mutation at any base in the exome 13 or more times by chance in 63 tumors is < 10−62 (Binomial test). Activating CTNNB1 and GNAS mutations were found in a total of 10 (15.9%), and 3 tumors (4.8%), respectively. PRKACA, CTNNB1 and GNAS mutations remained mutually exclusive (P = 0.02, Monte Carlo simulation).

PRKACA is the most highly expressed catalytic isoform in human adrenal (Supplementary Table 5); the most highly expressed regulatory subunit is PRKAR1A. PRKACA phosphorylates the R-R-X-S/T-Φ motif (Φ is hydrophobic)[37]. The regulatory subunit binds to PRKACA catalytic cleft via a pseudosubstrate sequence R-R-G-A-I[9]; the isoleucine of PRKAR1A fits in a hydrophobic cleft in the catalytic domain formed by Leu206 and Leu199[38,39] (Fig. 2c, 2d[10]). Leu206 is conserved in orthologs from invertebrates to humans, while yeast has the isomer Ile (Fig. 2b). The p.Leu206Arg substitution is likely to disrupt this interaction (Fig. 2e, 2f), resulting in constitutive PKA activity.

We expressed PRKAR1A and either PRKACAWT or PRKACAL206R (both with C-terminal FLAG epitope tag), in equimolar ratios in HEK293T cells. PRKAR1A shows low endogenous levels that are increased by transfection (Fig. 3a). Cell lysates were immunoprecipitated (IP) with anti-FLAG to pull down PRKACA, followed by Western blotting using anti-PRKAR1A antibodies. In 3 independent transfections, immunoprecipitation with PRKACAWT robustly pulled down PRKAR1A, while no PRKAR1A was detected after IP with PRKACAL206R (Fig. 3a).

Figure 3

PRKACAL206R does not bind regulatory subunit and shows increased phosphorylation of CREB and ATF1. (a) FLAG-tagged constructs of wild type (Wt) or mutant (M) catalytic subunits (PRKACAWT and PRKACAL206R) and the regulatory subunit (PRKAR1A) were expressed in HEK293 cells in indicated combinations. Cell lysates were immunoprecipitated with anti-FLAG and the products of IP were analyzed by Western blotting with antibodies specific for FLAG and PRKAR1A. While IP of PRKACAWT pulls down the regulatory subunit, IP of PRKACAL206R does not. Duplicates shown are from independent transfections. (b) Western blots with antibodies specific for CREB phosphorylated at Ser133 (pCREB) and ATF1 phosphorylated at Ser63 (pATF), CREB, ATF1, and FLAG in cells expressing FLAG-tagged PRKAR1A and either wildtype or mutant PRKACA(both tagged with FLAG). Representative results of triplicate experiments are shown. Bar graph shows results of quantitation (mean ± SEM; n = 3). Cells expressing PRKACAL206R have 3 to 4-fold higher levels of pCREB and pATF1 than cells expressing PRKACAWT (**P<0.05). neg, no transfection of PRKACA or PRKAR1A.

Unbound PRKACAL206R should result in increased phosphorylation at targets such as Ser133 of CREB (CREBSer133-P). Western blotting of lysates from cells expressing wildtype or mutant PRKACA and PRKAR1A with an antibody specific for phospho-Ser133 demonstrated that PRKACAL206R produced ~4-fold increased CREBSer133-P versus PRKACAWT (P = 0.037, T-test; Fig. 3b). Similarly increased phosphorylation was seen at Ser63 in ATF1, another PKA target (P= 0.002, T-test; Fig. 3b). Immunohistochemistry for CREBSer133-P showed increased staining in eight PRKACA-mutant tumors compared to 5 adenomas without PRKACA, GNAS or CTNNB1 mutations (median 55% versus 10%, Supplementary Fig. 4, Supplementary Table 6). Additional histologic features are described in Supplementary Fig. 5.

Adenomas with PRKACA or GNAS mutations were significantly smaller than adenomas without (28.7 ± 7.3 mm versus 39.2 ± 15.9 mm, s.d.; P = 0.035, Mann-Whitney U test) and patients presented at significantly younger ages (45.3 ± 13.5 versus 52.5 ± 11.9 years, s.d.; P = 0.045, Mann-Whitney U test). They were also significantly associated with overt CS (13/16 with PRKACA or GNAS mutation with overt CS, versus 16/39 without, P = 0.008, Fisher’s exact test).

These results implicate a recurrent gain-of-function mutation in PRKACA in cortisol-producing adrenal adenomas. This was the most frequently mutated gene in our cohort. Mutant PRKACA has drastically diminished binding to its regulatory subunit, leading to increased PKA enzymatic activity.

Because of strong prior evidence that increased cAMP signaling is sufficient for cell proliferation and cortisol production[7,40-42], these PRKACA mutations are likely sufficient for tumor formation (Supplementary Fig. 6). The absence of other mutations likely to contribute to neoplasia supports this notion. The same applies to activating GNAS mutations. These findings are analogous to observations in adrenal aldosterone-producing adenomas, in which the single mutations are sufficient for cell proliferation and hormone production[43-45]. In both diseases, the coupling of signaling pathways to both hormone production and cell proliferation provides a mechanism for increasing the mass of hormone-producing cells when demand for hormone is high. These observations suggest that PKA inhibitors could be of benefit in controlling tumor growth and hormone production in tumors bearing this mutation.

Mutations in PRKACA, GNAS and CTNNB1 were found in 12 of 17 CNV− tumors. The pathogenesis of the remaining five tumors without known drivers remains unknown, leaving open the question of whether they might result from mutations outside the coding region.

The mutual exclusivity of PRKACA, GNAS and CTNNB1 mutations suggests they phenocopy one another. Previous studies have found that increased cAMP signaling increased Wnt/beta-catenin signaling[46,47]; similarly, PKA can phosphorylate beta-catenin, inhibiting ubiquitination and increasing signaling[48,49]. Wnt signaling promotes proliferation of adrenal fasciculata cells, and there is evidence for a role in activation of the steroid biosynthetic pathway[50]. Nonetheless, activating CTNNB1 mutations are also found in non-hormone producing adrenocortical tumors and other cancers[31,51,52]. Further work will be required to establish the factors that determine which tumors with CTNNB1 mutations autonomously produce hormones.

While this paper was in review, this same somatic PRKACA mutation was reported in 37% of adrenal adenomas associated with overt Cushing syndrome and 22% of all cortisol-producing adenomas from an independent cohort. Use of a cell-based FRET assay for PKA activity provided evidence of constitutive activity of the mutant protein[53].

The striking differences in mutation spectrum between CNV+ and CNV− tumors suggest two distinct mechanisms for these tumors. Interestingly, the somatic mutation spectrum of CNV+ tumors diagnosed by rigorous histologic criteria as adenomas was more similar to that of CNV+ carcinomas and other cancers than CNV− adenomas. Larger cortisol-producing adrenal tumors are believed to make cortisol inefficiently owing to lack of activation of the cortisol biosynthetic pathway, while smaller tumors are more efficient due to induction of cortisol biosynthetic enzymes[54]. The current findings support this observation, because smaller tumors more frequently have PRKACA and GNAS mutations, activating the cAMP signaling pathway.

The mutation spectrum in CNV+ adenomas suggests they may have greater malignant potential. These tumors may come to early attention due to clinical signs and symptoms of cortisol excess. Interestingly, some adrenal cortisol-producing tumors initially classified as benign have recurred with invasive or metastatic disease[55]. Similarly, other studies in mice and humans have also suggested a multistep model of progression from adrenal hyperplasia to adenoma and carcinoma[56-58]. These observations suggest that CNV+ adenomas might warrant closer clinical follow-up.

While all tumors in our study were diagnosed as having autonomous cortisol production by experienced endocrinologists and surgeons in tertiary referral hospitals, supporting data for 9 patients referred from outside hospitals had incomplete documentation in available medical records (Supplementary Table 1). Exclusion of these tumors (14% of the total) did not significantly alter the conclusions. These 9 tumors comprised 3 carcinomas and 6 adenomas with no mutation in PRKACA or GNAS, and 1 with CTNNB1 mutation.

Lastly, activating GNAS mutations are found in growth-hormone secreting pituitary tumors and thyroid adenomas[59]. The finding that PRKACA mutation can phenocopy activating GNAS mutations in adrenal tumors raises the possibility that mutations in PRKACA or other genes that increase cAMP signaling might be found in these other tumors as well.

Online Methods

Patients and specimens

Matched fresh-frozen cortisol-producing adrenocortical tumors and either venous blood samples or adjacent normal tissue were obtained from patients undergoing adrenalectomy at Yale New Haven Hospital, Uppsala University Hospital. Additional tumors and matched normal samples were obtained from formalin-fixed paraffin-embedded samples in Yale Department of Pathology, University of Bonn and University of Düsseldorf. The research protocols of all studies were approved by local IRBs.

All tumors included in this study were determined by experienced endocrinologists and endocrine surgeons at tertiary referral hospitals to have ACTH-independent cortisol production, resulting in either subclinical or overt Cushing syndrome at the time of surgery, with surgical pathology consistent with that diagnosis. Patient demographics, tumor characteristics, laboratory test results and clinical features are summarized in Supplementary Table 1. Some clinical records were incomplete owing to evaluation having been performed at outside hospitals, with detailed records missing from the tertiary hospital medical record.

Tumors were classified by an endocrine pathologist blinded to genotypes; tumor histology was categorized by gross assessment of tumor, percentage of oncocytic cells, and nuclear atypia graded from unremarkable to severe atypia. Weiss scores[60,61] were determined for each tumor, and proliferative index using Ki-67 (clone MIB-1, Dakopatts, dil 1:300, Epitope retrieval: high pH for 20′ Leica Bond III) was performed in tumors on the judgment of the pathologist. Nuclear staining in tumor cells was assessed as positive, and a percentage was calculated after assessing 4000 to 5000 tumor cell nuclei (10 high power fields).

DNA and RNA were isolated from non-FFPE samples using the Allprep kit (Qiagen) following manufacturer’s instructions. For FFPE samples, DNA was prepared with the BiOstic FFPE Tissue DNA Isolation kit (MO BIO Laboratories) using a modified protocol.

Whole exome sequencing and analysis

Targeted capture was performed using the NimbleGen 2.1 Exome reagent followed by sequencing on the Illumina platform as previously described[44]. Sequences were aligned to NCBI Build 36 of the human genome using the ELAND program (Illumina). Somatic mutations were called based on the significance of differences in reference and non-reference read distributions between tumor and matched normal samples. Calls were further evaluated by manual inspection of read alignments.

The somatic mutation rate and estimations of tumor purity were calculated as previously described[44]. MutSig[62] was used to determine if genes were mutated more often than expected by chance. CNVs were identified by comparing coverage depth ratios of tumor and matched normal samples after normalizing for mean coverage depth of each exome, and changes in minor allele frequency at informative SNPs. GISTIC2.0 was used to assess the significance of recurrent CNVs as described previously[29].

Sanger sequencing of genomic DNA

Direct bidirectional Sanger sequencing of PRKACA Leu206, exon 3 of CTNNB1 and GNAS (Arg201 and Gln227) from genomic DNA from tumors and matched normal samples was performed following PCR amplification using specific primers. Primer sequences are listed in Supplementary Table 7.

Whole transcriptome (mRNA-Seq) profiling

Ribosomal RNA (rRNA) was removed from total RNA using the Ribo-Zero rRNA Removal Kit (Epicenter) according to manufacturer’s instructions. cDNA was synthesized and libraries prepared using the Illumina RNA-Seq kit, followed by sequencing on Illumina HiSeq instrument. RNAseq data for normal adrenal gland was obtained from the Illumina Human Body Map 2.0 Project[63]. Sequences were aligned using TopHat version 2.08 and assembled into transcripts using Cufflinks version 2.0.2[64,65].

Clones and mutagenesis

Expression plasmids encoding human PRKACA and PRKAR1A were obtained from Origene (RC210332, NM_002730; RC203828, NM_002734). The Leu206Arg mutation was introduced into PRKACA using the QuikChange site-directed mutagenesis system (Stratagene), and validated by sequencing the complete coding region. Plasmids were prepared using the HiSpeed Plasmid Maxi Kit (Qiagen). The FLAG tag was deleted in the PRKAR1A construct for the IP experiment using EcoRV and PmeI restriction enzymes (New England Biolabs).

Transient transfection and IP

HEK293T cells (ECACC) were cultured in DMEM (Gibco) supplemented with 10% fetal bovine serum, penicillin (100 units/ml) and streptomycin (10mg/ml). Cells were transfected with expression plasmids using Lipofectamine 2000 (Life Technologies) according to manufacturer’s instructions, followed by incubation for 48h. PRKACA and PRKAR1A were co-transfected at a molar ratio of 1:1. Cells were then washed in cold PBS and lysed at 4°C in lysis buffer as previously described[66]. Protein concentrations were quantified using a BCA Protein Assay Kit (Pierce), incubated with Anti-FLAG M2 Affinity gel (Sigma) at 4°C, washed and eluted with 3X FLAG peptide.

Western blotting

Protein extraction and Western Blotting were performed as previously described[66]. Lysates and immunoprecipitates were fractionated on 4–20% polyacrylamide gels, transferred to nitrocellulose membranes, and immunoblotted with primary antibodies: rabbit anti-FLAG (Sigma, 1:5,000), anti-CREB (CST, 1:1000), anti-Phospho-CREB (CST, 1:1000), anti-ATF1 (Millipore, 1:1000) or mouse anti-PRKAR1A (BD Biosciences, 1:500). Signals were visualized by chemiluminscence after incubation with peroxidase-conjugated secondary antibody.

Phospho-CREB staining

IHC was performed on slides from FFPE samples using a rabbit polyclonal antibody to Phospho-CREB (CST, 1:600). To determine specificity of antibody binding, blocking peptides specific to the antibody were used (CST). All slides were stripped of paraffin in xylene before rinsing in decreasing concentrations of ethanol, prior to antigen retrieval in boiling 10 mM Sodium Citrate + 0.05% Tween 20 at pH 6.0. Cells were permeabilized in 0.2% Triton X100 in PBS, then incubated in 3% hydrogen peroxide. Blocking was achieved with 10% BSA in PBS. After incubation with primary and secondary antibodies, slides were stained with DAB and counter-stained with Hematoxylin.

An experienced endocrine pathologist determined the percentage of nuclei stained positive for phosphorylated CREB by selecting most strongly staining area, and at least 100 nuclei were assessed for each tumor. Endothelial cells in the tumor were used as an internal positive control for regions where staining of tumor nuclei was not observed, and the presence of unstained nuclei was used as internal negative controls.

Statistical analysis

The data are summarized as means ± SEM. Two-tailed t tests, Mann-Whitney U tests and Fisher’s exact tests were used for comparison between two groups. The p value for the probability of observed recurrent mutation occurring by chance was calculated as a binomial probability, using the somatic mutation rate of CNV− tumors.

Monte Carlo simulations (1 × 106 iterations) were performed to calculate the probability of observing mutually exclusive mutations in PRKACA, CTNNB1 and GNAS by chance.