Genetic drivers of Cushing’s disease: frequency and associated phenotypes

Abstract

Purpose:

Cushing’s disease (CD) is often explained by a single somatic sequence change. Germline defects, however, often go unrecognized. We aimed to determine the frequency and associated phenotypes of genetic drivers of CD in a large cohort.

Methods:

We studied 245 unrelated CD patients (139 females, 56.7%), including 230 pediatric (93.9%) and 15 adult patients (6.1%). Germline exome sequencing (ES) was performed in 184 patients; tumor ES was also done in 27 of them. Forty-three germline samples and 92 tumor samples underwent Sanger sequencing of specific genes. Rare variants of uncertain significance (VUS), likely pathogenic (LP) or pathogenic variants in CD-associated genes were identified.

Results:

Germline variants (13 VUS, 8 LP, and 11 pathogenic) were found in 8/19 (42.1%) patients with positive family history and in 23/226 (10.2%) sporadic patients. Somatic variants (one LP and seven pathogenic) were found in 20 out of 119 tested individuals (16.8%); one of them had a coexistent germline defect. Altogether, variants of interest were identified at the germline level in 12.2% of patients, at the somatic level in 7.8%, and coexisting germline and somatic variants in 0.4%, accounting for one-fifth of the cohort.

Conclusions:

We report an estimate of the contribution of multiple germline and somatic genetic defects underlying CD in a single cohort.

Introduction.

Neuroendocrine neoplasms (NENs) are among the most frequently inherited tumors. As with other NENs, pituitary tumors exhibit highly variable clinical courses, ranging from incidentalomas to florid syndromes of hormonal hypersecretion and/or tumors invading surrounding structures, exceptionally metastasizing.¹ Familial pituitary tumors are often more aggressive and arise at an earlier age, compared to their sporadic counterparts; yet, it is estimated that only 5% of them occur on a familial basis.² In these individuals, genetic diagnosis has the potential for modifying the natural history of the disease, by leading to a tailored care of patients and to targeted clinical screening.

At the somatic level, hotspot (hs) variants of the USP8 gene explain around half of corticotropinomas.^3-7 Recurrent somatic variants have also been described in hs of BRAF and USP48, as well as in the ATRX and TP53 genes.^8-11 Germline genetic defects leading to Cushing’s disease (CD), however, go largely unrecognized, because their phenotypic variability and variable penetrance preclude the application of extensive genetic testing. Indeed, their frequency in the general population remains unknown, because corticotropinomas are uncommon and are among the less frequently reported types of inherited pituitary tumors.¹² There are no cohort studies available for most germline causes of CD, thus most available data come from case reports and small series.

We aimed to determine the frequency of known germline and somatic genetic drivers of CD in a large cohort of mainly pediatric patients and to characterize their clinical presentation.

Materials and methods.

Patients and samples

We studied 245 unrelated CD patients who were evaluated at the outpatient clinic and/or admitted for clinical workup and treatment (n=242), or whose DNA samples were referred for study (n=3) at the National Institutes of Health Clinical Research Center (NIH-CRC) between 1997-2018. All individuals and their parents or guardians provided informed assent or consent for the study, including genetic testing, and were recruited under protocol 97-CH-0076 (ClinicalTrials.gov: NCT00001595), approved by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Institutional Review Board; none were excluded. When available, the probands’ parents and siblings were also recruited. Clinical data were obtained from the individuals and/or their clinical records. Germline and tumor DNA samples were extracted as previously described.¹³ The diagnosis of CD was established following established guidelines and our previously published protocol.¹⁴

Some individuals have been previously screened for germline defects in AIP, CDKN2C, MEN1, and PRKAR1A (n=59), CABLES1 (n=140), CDKN1B (n=211), DICER1 (n=182), somatic and/or germline GPR101 variants (n=34), and somatic CDKN1B (n=27), and USP8 variants (n= 44) using Sanger sequencing (and multiplex ligation dependent probe amplification for AIP and MEN1).^{6 13 15-18} A case series of six multiple endocrine neoplasia type 1 (MEN1) patients, individual reports of patients with PRKAR1A and USP8 germline defects, and somatic copy number variation (CNV) analyses in 27 patients included here have also been published.^19-22 Clinical TSC2 Sanger sequencing was performed in one patient (Case 31 in Supplemental table 1) based on his phenotype; a pathogenic variant was found, as previously published.^{23 15} The report, however, was not on file in our Center, the patient could not be located, and there was not enough remaining DNA to repeat the screening. Also, due to low quality or insufficient DNA, eighteen patients did not undergo any genetic testing at the germline level, although six of them underwent somatic testing for one or more hs defects by Sanger sequencing.

Exome sequencing

Exome sequencing (ES) of germline DNA samples from 91 patients and tumor DNA samples from 27 of them was performed at the University of Minnesota Genomics Center (UMGC). ES for 91 additional germline samples was done at Novogene (Beijing, China). Illumina HiSeq 2000 (100 bp paired-end reads, UMGC germline) and 2500 (150 bp paired-end reads, Novogene, and 125 bp paired-end reads, UMGC tumor) platforms were used.¹⁷ ANNOVAR was used to determine the effect of coding variants using both the RefSeq and UCSC gene sets, and nonsynonymous variants were annotated using information from the dbNSFP v4.0 database.^{24 25} All variants were annotated for their frequency in public databases and internal ES datasets totaling >10,000 samples. Two additional patients underwent clinical ES at GeneDx (Gaithersburg, MD, USA). Rare variants (present in <1% of any variant collection) in known pituitary tumor-associated genes (AIP, ATRX, BRAF, CABLES1, CDKN1B, DICER1, GNAS, GPR101, MEN1, NR0B1, NR3C1, PRKAR1A, SDHx, TP53, TSC1, TSC2, USP8 and USP48) were queried in the study population (hereafter “genes of interest”, GOI). For all samples, a manual screen of ES raw data for the variants of interest was performed, using the Integrative Genomics Viewer 2.3.72 platform (Broad Institute).²⁶

Sanger sequencing

Sanger sequencing was used to screen tumor tissues for hs variants (whenever possible due to sample availability), to confirm ES findings, to screen family members for specific variants, and to investigate loss of heterozygosity (LOH) in tumors. Amplification was carried out by end-point PCR (GoTaq Green Master Mix, Promega M7123). Primers for hs screening were: 5’-TGCTTGCTCTGATAGGAAAATG-3’ and 5’-AGCATCTCAGGGCCAAAAAT-3’ for BRAF, 5’-TGACCCAATCACTGGAACCT-3’ and 5’-TGGCTTCCTCTTCTCTTCCTC-3’ for USP8, and 5’-GCCCCGCTAAAGAATAAACA-3’ and CATTCTAAAACATTTGCCTGCT-3’ for USP48; sequences for other primers are available upon request. Amplicons were purified using ExoSAP-IT express (Applied Biosystems 75001.4X.1.ML) and subjected to direct bidirectional sequencing (BigDye Terminator 3.1 Cycle Sequencing Kit, Applied Biosystems 4337456) in a 3500xL Genetic Analyzer (Applied Biosystems). Sequences were analyzed using the SeqMan Pro 17.1.1.120 (DNASTAR) software.

In silico analyses and variant classification

The VarSome v.10.1 and PolyPhen-2 v2.2.2 (r398) platforms were used for the annotation and in silico analysis of the filtered variants; VarSome integrates multiple in silico prediction tools via the dbNSFP v4.0 database.^{27 28} Pathogenic associations were searched in ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/), and COSMIC,²⁹ and previous clinical and experimental data were searched using Genomenon Mastermind (https://www.genomenon.com/mastermind/). Frequency in the general population was searched in gnomAD v2.1.1.³⁰ Functional in vitro assays were performed in some cases.^{13 17} On these bases, variants were classified according to the American College of Medical Genetics and Genomics and Association for Molecular Pathology (ACMG/AMP) guidelines.³¹ Pathogenic and likely pathogenic (LP) variants, as well as variants of uncertain significance (VUS) whose pathogenicity could not be resolved after careful literature review, in vitro and/or in silico analyses were considered as “potential genetic drivers”. The rest of variants are listed in Supplemental table 2.

Immunohistochemistry

Immunohistochemistry (IHC) in corticotropinomas was performed using primary antibodies against SDHA (Abcam ab14715, 1:500), SDHB (Abcam ab14714, 1:500), SDHC (Abcam ab155999, 1:500), SDHD (Sigma-Aldrich HPA045727, 1:50) and TSC2 (Cell Signalling Technology #4308, 1:800), 1:1000 biotinylated secondary antibodies (Jackson ImmunoResearch Laboratories 115-065-062 or 111-065-144, as appropriate) and 1:500 peroxidase streptavidin (Jackson ImmunoResearch Laboratories 016-030-084); the full protocol has been detailed elsewhere.¹³ Hematoxylin-eosin and additional immunohistochemistry slides were retrieved from the NIH-CRC Pathology archives.

Statistical analyses

All analyses were carried out using the Prism 9.2 software (GraphPad Software). Data distribution was analyzed using the Shapiro-Wilk test. Parametric data are presented as mean ± standard deviation and non-parametric data are presented as median and interquartile range. Contingency tables were compared using the chi squared or the Fisher’s exact test, as appropriate. Clinical parameters were compared among groups using the Mann-Whitney test. P<0.05 was considered statistically significant.

Results

Study cohort

The cohort included 139 female (56.7%) and 106 male (43.3%) patients, with median age of 10 years (8-12) at disease onset and 13.5 years (10.8-15.8) at diagnosis, including 230 pediatric (≤18 years at disease onset, 93.9%) and 15 adult patients (6.1%, Figure 1a-b); two-thirds were White, not Hispanic or Latino (Figure 1c). The median maximum tumor diameter was 5 mm (4-8); thirty-two patients (13.1%) had a macroadenoma (tumor ≥10 mm), 206 patients (84.1%) had a microadenoma, and five patients (2%) had two tumors (Figure 1d). Thirty-five patients (14.9%) had extrasellar extension, out of 235 with available data (Figure 1e). An ACTH-producing pituitary tumor was confirmed in 181 cases (Figure 1f); fourteen of them (13 pediatric and one adult) were diagnosed with Crooke’s cell tumor, a rare variant of corticotropinoma with aggressive clinical behavior. The median age of those with Crooke’s cell tumor was 10 years (7.7-15.7) at disease onset and 13.7 years (10.7-16.5) at diagnosis, and the maximum tumor diameter was 6.5 mm (5-22.8), which were not significantly different than for the rest of patients.

Figure 1. Study cohort and phenotype associated with somatic USP8 hs variants.

a) Sex distribution for the whole cohort and the pediatric and adult subsets. b) Median age was 10 (8-12) years at disease onset and 13.5 (10.8-15.8) years at diagnosis for the whole cohort. Pediatric patients had a median age of 10 (8-13) years at disease onset and 13.2 (10.5-15.5) years at diagnosis and adult patients had a mean age of 41.3 (±10.7) years at disease onset and 44.9 (±12.7) years at diagnosis. c) White, not Hispanic or Latino patients accounted for 164 cases. There were 27 Hispanic or Latino patients of mixed or not specified race, one Native Hawaiian or Other Pacific Islander, Hispanic or Latino, 15 White, Hispanic or Latino, 9 not Hispanic or Latino of mixed not specified race, 12 Asian, not Hispanic or Latino, and 14 Black or African American, not Hispanic or Latino. d) Median maximum tumor diameter was 5 (4-8) mm for the whole cohort and for pediatric patients, and 10 (5-17) mm for adult patients. e) The most common site of extrasellar extension was the left cavernous sinus (16 cases, 45.7% of those with extrasellar extension). f) A corticotropinoma was confirmed in 165 patients (including one with coexisting prolactinoma), 14 had Crooke’s cell tumors, and two had ACTH-positive plurihormonal tumors. g) Sex distribution was not significantly different between pediatric patients with somatic USP8 hs variants and pediatric patients with reference allele (P=0.1078). h) Median age at disease onset was 12.5 years (11-14.2) for pediatric patients with USP8 hs variants and 9.9 years (8-12.6) for cases with reference allele (P=0.0018). Median age at diagnosis was 15.3 years (12.9-16.3) for patients with USP8 hs variants and 13 years (10.5-15.2) for cases with reference allele (P=0.0090). i) Median maximum tumor diameter was 7 mm (5.8-10.3) for pediatric patients with USP8 hs variants and 6 mm (5-8) for pediatric patients with reference allele (P=0.1736). Error bars in b, d, h, and i represent interquartile ranges. ns, not specified; ref, reference allele.

Most patients (92.2%, n=226) presented as apparently sporadic CD. Ten (52.6%) out of the 19 individuals with familial presentation had a family history compatible with familial isolated pituitary adenoma (FIPA). Seven (36.8%) were members of MEN1 kindreds. One patient had a family history compatible with pheochromocytoma, paraganglioma and pituitary adenoma (3PA) syndrome and one was part of a kindred with Li-Fraumeni-like phenotype. At CD diagnostic evaluation, four individuals presented MEN1-associated tumors; one of them had apparently sporadic presentation. Another patient had a personal history of tuberous sclerosis complex (TSC). One more patient had a complex multisystemic syndrome including developmental delay, endocrine, cardiological, respiratory and dermatological manifestations. No other patients had syndromic features that would prompt routine genetic tests.

Somatic variants.

In addition to 27 corticotropinomas screened for somatic variants by ES, 71 were screened for BRAF, 92 for USP8, and 67 for USP48 hs variants using Sanger sequencing. Six pathogenic, previously reported, USP8 variants were found in eighteen corticotropinomas (15.1% of 119 tested). Among pediatric cases, tumors with USP8 variants arose and were diagnosed at a significantly older age, compared with those without USP8 variants (Figure 1g-i). The USP48 LP variant c.1243A>G, p.M415V was identified in one tumor (1.1% of 94 tested). The BRAF p.V600E variant was not identified in any screened tumor. In addition, the pathogenic frameshift TSC2 variant c.4815_4816del, p.Q1605Hfs*8, not reported before, was found in one 8 mm corticotropinoma with no extrasellar extension (Case 30).

All somatic hs variants were found in heterozygosis in apparently sporadic and non-syndromic patients (Supplemental table 1). No other potential genetic drivers were identified at the somatic level in the 27 tumors analyzed by ES. Their median maximum diameter was 7 mm (4.8-9.3), which was slightly larger than for cases not analyzed by ES (5 mm [4-8]); this difference did not reach statistical significance (P=0.0518). The results of an analysis of large CNVs in these tumors have been reported elsewhere.²²

Germline variants.

Thirty-two heterozygous potential driver variants in 12 GOI (AIP, CABLES1, CDKN1B, DICER1, MEN1, NR3C1, PRKAR1A, SDHA, SDHD, TSC2, TP53 and USP8) were identified in 31 different individuals, including one adult and 30 pediatric patients (Supplemental table 1). Thirteen variants were VUS, the rest were either pathogenic (n=11) or LP (n=8, Figure 2a). There were no significant differences in sex distribution, age at disease onset, age at diagnosis, or maximum tumor diameter between individuals with germline variants of interest and the rest, neither for pediatric patients nor for the whole cohort. The screened samples per gene are presented in Figure 2b-c.

Figure 2. Germline potential genetic drivers.

a) Distribution of variant types for each ACMG/AMP category. b) and c) Summary of samples included in screening for the genes of interest at the germline (b) and somatic (c) levels. d) Representative SDHA, SDHB, SDHC, and SDHD IHC images in corticotropinoma samples from Cases 22, 11, and 24. Case 19 (negative for SDHx variants, carrying a germline LP NR3C1 variant) is shown for comparison. SDHC and SDHD IHC were no performed for Case 11 due to insufficient slides. e) Representative images for TSC2 IHC in corticotropinoma samples from Cases 27, 29, and 31. Case 46 (case negative for TSC2 variants, carrying a somatic pathogenic USP8 variant) is shown for comparison. All slides were counterstained with hematoxylin. All images were acquired using a Keyence BZ-X710 microscope and the BZ-X Viewer v01.03.01.01 software (Keyence), with a magnification of 60x. Scale bar= 20 μm. VOI, variants of interest.

One patient harbored a deletion spanning two exons of MEN1; the rest of variants were exonic single nucleotide substitutions or short insertions or deletions. Nine variants were of maternal origin, an equal number were paternal, and a de novo variant was confirmed in one case (Case 40). The origin of the rest of variants could not be determined. LOH with loss of the normal allele was demonstrated in Cases 20, 22, and 29, carrying variants in PRKAR1A, SDHA, and TSC2, respectively. Only the reference allele was detected in two cases and somatic heterozygosity was conserved in 11 patients. LOH was not expected in Case 40, who had a germline USP8 variant (considered prooncogenic). Case 11 carried coexistent VUS in DICER1, SDHA and TSC2, and one SDHD VUS was detected in two patients (Cases 24 and 25). The remaining germline variants were found in individual cases. Only Case 7, which has been previously reported, had potential disease drivers at both the germline (CDKN1B) and somatic (USP8) levels.¹⁷

The clinical presentation was familial in eight patients and apparently sporadic in the remaining 23. Five patients with familial presentation (Cases 12, 15, 16, and 17) had MEN1 variants. Patient 15 also carried the ATRX (NM_000489.5) LP germline variant c.6149T>C, p.I2050T (rs122445110), whose origin could not be determined. The patient is female and has no personal or family history of manifestations within the spectrum of the ATRX loss-of-function phenotypes.³² Since X-chromosome inactivation affecting the variant allele is likely, the ATRX variant most probably does not act as a CD driver in this case. One more patient with a MEN1 VUS (Case 13) had a relative affected with a prolactinoma. That individual belonged to the maternal branch of the family, while the variant was of paternal origin, and was therefore considered a phenocopy. A similar presentation was found in Case 25, a patient with a paternally inherited SDHD VUS and a maternal uncle with a non-functional pituitary tumor (previously reported).³³ The last patient with familial presentation (Case 26) was a member of a kindred with a Li-Fraumeni-like phenotype and carried a LP TP53 variant.

Out of 23 patients with sporadic presentation, three had pathogenic or LP CDKN1B variants (Cases 5, 6, and 8) and three had one pathogenic variant in MEN1, PRKAR1A, and SDHA each (Cases 14, 20, and 22, respectively). Since pituitary tumors are part of the multiple endocrine neoplasia type 4 (MEN4, caused by loss-of-function, LOF, CDKN1B variants), MEN1, Carney complex (CNC, LOF PRKAR1A variants) and 3PA (LOF variants of SDHx genes) syndromes, these patients can be considered true simplex cases of such entities. One patient with clinical diagnosis of TSC and a pathogenic TSC2 variant was also a simplex case (Case 31). CD was an unexpected finding in this context, although a small number of cases of pituitary tumors in TSC patients have been reported.^{34 35}

In contrast, two additional patients carried LP TSC2 variants (Cases 28 and 29) but did not develop the neurological phenotype. They were not accounted for as simplex TSC cases because pituitary tumors are not currently considered as a component of TSC. In addition, two patients had LP CABLES1 variants (Cases 3 and 4), and one had a USP8 pathogenic variant (Case 40). These are germline genetic causes of CD that we have recently reported that have not been associated with familial disease. Interestingly, one patient harbored a LP NR3C1 variant (Case 19) in the region encoding the ligand binding domain, but no data of generalized glucocorticoid deficiency were documented during her diagnostic workup.

Twelve different patients carried the 13 VUS identified in the study. Case 11, with coexistent VUS in three genes, did not have a phenotype or a family history compatible with any of the syndromes caused by LOF of DICER1, SDHA, or TSC2. We did not find data in the literature to accurately classify the SDHD p.A18V VUS either, which was found in two patients. To better understand the contribution of these VUS of SDHx genes, corticotropinomas from Cases 22 (pathogenic SDHA variant), 11 (SDHA VUS), and 24 (SDHD VUS) were evaluated using IHC (Figure 2d). All three cases displayed reduced SDHB staining and Cases 22 and 11 had reduced SDHA staining, supporting variant pathogenicity.³⁶ SDHD and SDHC immunoreactivity was lost in Cases 22 and 24, respectively, probably due to SDH complex destabilization. Retention of SDHD staining with diffuse cytoplasmic pattern (Case 24) has been reported before in the presence of SDHD LOF.³⁷

In addition to Case 11, one more patient carried a TSC2 VUS (Case 27). We performed IHC for TSC2 in corticotropinoma samples from three patients, obtaining mixed results that did not contribute to variant classification. While Case 29, carrying a LP variant, displayed positive strong cytoplasmic immunostaining in scattered cells, Cases 27 and 31 (carrying a VUS and a pathogenic variant, respectively) had negative IHC (Figure 2e). The value of this method as a predictor of TSC2 LOF has not been evaluated. An accurate classification of these variants will require collection of evidence from other patient cohorts, as well as additional functional experiments.

We have previously analyzed the CDKN1B VUS p.I119T and p.D136G (Cases 7 and 9). In our functional studies, these variants seemed to have a deleterious effect on protein function.¹⁷ The patients did not have a full MEN4 phenotype, but this syndrome displays variable penetrance and clinical presentation. In silico analyses and previous reported interpretations, however, are contradictory. There was a coexistent somatic USP8 pathogenic variant in the corticotropinoma from Case 7. Although the coexistence of germline and somatic drivers is possible, the presence of the somatic variant further complicates the classification of the germline variant.

Another patient (Case 1) carried an AIP VUS that has rendered variable results in functional studies reported in the literature. There was not a FIPA phenotype in this case; however, given the characteristic low penetrance of AIP LOF-associated FIPA, it is complicated to fully rule out a pathogenic role for this variant. In contrast with the MEN4 and FIPA syndromes, the DICER1 syndrome has a high penetrance. Therefore, the two DICER1 VUS found in the study might not be actual drivers, but instead disease modifiers since the patients (Cases 10 and 11) had neither pituitary blastomas nor personal or family history compatible with DICER1 syndrome. Finally, in vitro functional assays are required to accurately classify the CABLES1 and MEN1 VUS found in Cases 2 and 13. Like previously reported cases with CABLES1 LOF variants, Case 2 had a large and aggressive tumor, although at an older age. In contrast, Case 13 has not developed a full MEN1 phenotype.

Contribution of individual potential genetic drivers.

Thirty patients (12.2% of the cohort, n=245, 13.2% of individuals with any type of germline screening, n=227) carried potential genetic drivers at the germline level, while somatic defects were identified in 19 patients (7.8% of the cohort, n=245, 16% of those tested, n=119). One patient (0.4% of the cohort, n=245) had coexistent germline and somatic defects (Figure 3a). Altogether, we identified potential genetic drivers in 50 patients (20.4% of the cohort, n=245). The genes with the most variants identified were MEN1 and USP8, which accounted for seven variants each, out of the 40 variants detected in the study (Figure 3b).

Figure 3. Contribution of potential genetic drivers to the pathogenesis of CD in the study cohort.

a) Potential genetic drivers of CD were found at the germline level in 30 patients (12.2%), and at the somatic level in 19 patients (7.8%); only one patient (0.4%) had concurrent variants in GOI at the germline and somatic levels. b) Variants per gene, expressed in percentages, out of the 40 different variants identified in the study. c) Cases per potential genetic driver. The percentages are based on the total number of patients in the cohort. d) Results of genetic screening after subtracting VUS. e) Results of genetic screening considering predicted additional cases.

Based on our data, 7.3% of cases in this cohort were caused by USP8 hs variants (6.9% somatic and 0.4% germline), 2.9% were associated with MEN1, 2% with TSC2 (1.6% germline and 0.4% somatic), 1.6% with CDKN1B, 1.2% with CABLES1, 1.2% with SDHA, 0.8% with SDHD, and 0.8% with two or three coexistent variants. Individual cases had defects in AIP, DICER1, NR3C1, PRKAR1A, TP53, and USP48, each accounting for 0.4% of the cohort (Figure 3c). If the VUS (found only at the germline level) are excluded from the total, we report 19 patients with germline defects (7.8%) and 20 (8.2%) with somatic defects, for a total of 16% of patients with identified genetic defects (Figure 3d).

Discussion

In our large cohort of mostly pediatric CD patients, we found potential genetic drivers of disease in one-fifth of cases. The contribution was estimated in 12.2% (7.8% without VUS) for germline variants and in 7.8% for somatic variants; only one patient (0.4%) harbored variants at both levels. Since not all patients underwent the same screening strategy, our results should be taken as bona fide estimates and not as absolute calculations of the contribution of these genetic defects to the pathogenesis of CD.

Additional calculations were performed to determine possible over or underestimations in our results. If VUS are excluded, the frequency of genetic defects is 16%. On the other hand, it is possible that we have missed some of the genetic defects existing in the cohort (Figure 3e). For instance, a germline variant was not identified for eleven out of the 19 cases with familial presentation; these patients represent an additional 4.5% of the cohort. At the somatic level, out of the 119 patients tested, 18 were positive for somatic USP8 hs variants. Extrapolating this result to the whole cohort, we would expect additional 19 patients to also test positive (7.8% of the cohort). This would result in 41 patients with proven or potential germline defects (16.7%), 38 patients with somatic defects (15.6%), and one patient with coexistent germline and somatic defects (0.4%), for a total of 80 CD patients (32.7%) carrying potential genetic drivers. These, however, are mere estimations and should be taken cautiously.

As we have previously reported, CD can be the first manifestation of MEN syndromes.^{17 20} Given that young patients might not present with a full syndromic phenotype, a thorough investigation of family history becomes essential. Still, we identified seven true simplex cases of 3PA, MEN1, MEN4, CNC, and TSC, meaning that it would not have been possible to predict all germline variants based solely on the clinical history. Other than syndromic presentations, there are no clinical features strongly suggestive of specific germline defects in CD, rendering single-gene screening impractical in most cases. While we do not suggest routine testing in all CD cases, our findings indicate that gene panel-based tests might be beneficial in all pediatric patients, although their cost-effectiveness requires validation with long-term studies.

The frequency of somatic USP8 hs variants in our cohort was low (15.1%), compared with the 23-62% reported in previous series.^3-7 The inclusion of mostly pediatric patients here could explain the differences.⁴ Along these lines, we identified only one case with a somatic USP48 hs variant, and none with BRAF hs variants. Studies reporting recurrent variants in those genes have been limited to almost exclusively adult patients.^{8 9} While the possibility of methodological limitations leading to low detection rate cannot be ruled out, we believe that our results reflect the actual frequency of these somatic drivers among pediatric patients. A single patient carried coexisting germline and somatic defects in CDKN1B and USP8, respectively. There was no LOH at the CDKN1B locus; thus, the USP8 variant likely acted as a second hit. In our experience, somatic LOH is often absent in pituitary tumors, and here it was found in only 3/19 tumors tested. We can conclude that somatic USP8 hs variants are infrequent among patients with inheritable causes of CD; since LOH is also infrequent, different second hits are probably present.

It was surprising to find 14 (5.7%) cases of the very infrequent Crooke’s cell tumors in our cohort.³⁸ Four of them (28.6%) were associated with identifiable genetic defects: a pathogenic PRKAR1A variant (Case 20), a somatic USP8 hs variant (Case 34), a CABLES1 VUS (Case 2), and coexistent DICER1, SDHA, and TSC2 VUS (Case 11). The tumor in Case 34 displayed chromosomal instability, as did four other corticotropinomas (Case 10 and three cases without identifiable genetic drivers).²² There was no information in the literature regarding the frequency, behavior, and genetic causes of Crooke’s cell tumors in pediatric patients, for comparison. From our findings, it appears that these tumors are not particularly rare among pediatric patients.

A limitation of the study is our cohort’s inherent selection bias, given that ours is a reference center for rare endocrine conditions. While our results might not reflect the picture of the general population, it would be extremely difficult to put together such a large cohort of patients with this rare condition in a different setting. Because individuals were recruited over 20 years, the availability and quality of DNA greatly varied among patients. All suitable samples underwent ES, but it was not possible to apply a homogeneous genetic testing strategy for all samples. Likewise, the availability of clinical information and of samples from the patients’ parents were also variable. The finding of multiple cases of Crooke’s cell tumors in the cohort deserves a separate study, designed ad hoc. Finally, the identification of novel genetic causes of Cushing’s disease was beyond the aim of this study. Further research to identify and validate possible novel genetic drivers is, however, underway.

In conclusion, we report an estimate of the contribution of multiple germline and somatic genetic variants as genetic drivers of corticotroph tumorigenesis in a large, single-center, predominantly pediatric cohort. We identified potential genetic drivers of disease in approximately one-fifth of the cases; approximately half of them at the germline level. Our findings serve as a basis for understanding the actual prevalence of rare genetic conditions among CD patients, designing strategies for genetic testing and counselling, and identifying novel therapeutic targets.

Data availability statement

Deidentified research data are available upon request by contacting J.L.M. or C.A.S.