1. BRAF
Protein kinase is the most frequently shared domain among cancer-associated proteins, therefore representing a particularly attractive therapeutic target
[1]. The isoforms A, B, and C of the highly conserved serine/threonine protein kinase rapidly accelerated fibrosarcoma (RAF) proteins, encoded in humans by three different genes, are among such proteins. C-RAF (also known as RAF-1) was first described in 1985, while A-RAF was discovered in 1986, and B-RAF in 1988
[2][3][4]. The latter is encoded by
BRAF (7q34, RefSeq NM_001354609.2), a proto-oncogene with preferential expression in neural tissues, and is the most potent activator of the RAS-GTPase (RAS)-RAF-MAPK and ERK kinase (MEK)-extracellular signal-regulated kinase (ERK) signaling pathway (RAS-RAF-MEK-ERK pathway)
[5][6][7] (
Figure 1). This phosphorylation cascade is involved in the physiological regulation of cellular processes such as proliferation, survival, differentiation, apoptosis, and motility
[6].
Germline activating variants affecting either
BRAF or other members of the RAS-RAF-MEK-ERK pathway are associated with a group of developmental syndromes collectively known as RASopathies
[8]. In contrast, the upregulation of this pathway via various mechanisms contributes to tumorigenesis in one-third of human cancers
[9]. Specifically, somatic missense activating variants in the glycine-rich loop or the activation segment of the BRAF catalytic domain occur in about 7% of all cancers. At least 90% of such cases, however, are explained by a single defect: c.1799T>A, p.V600E
[10][11]. This variant is found in two-thirds of malignant melanomas and papillary thyroid carcinomas (PTCs) and less frequently in colorectal, ovarian, and other types of cancer
[10][11][12].
The phosphorylation of residues T599 and S602 (UniProt P15056), which flank the variant, is required for BRAF to be recruited to the cell membrane and folded into its active conformation. The p.V600E change destabilizes the inactive conformation of BRAF and promotes its active state, thereby acting as a
phosphomimetic [13]. This way,
BRAF p.V600E results in an abnormally active RAS-independent kinase that induces cell proliferation and transformation in vitro and in vivo
[10][14][15]. Indeed,
BRAF variants seem to be mutually exclusive with oncogenic
RAS defects
[10]. In addition to the phosphorylation of the well-known downstream effectors MEK1/2,
BRAF p.V600E activates NFKB and prevents apoptosis
[14][16]. In colorectal cancer,
BRAF p.V600E has been associated with poor clinical prognosis and chemoresistance, increased microsatellite instability, and a higher mutational load
[17]. In addition, quantitation of
BRAF p.V600E by droplet digital polymerase chain reaction (ddPCR) has been used as a marker for measurable residual disease in hairy cell leukemia
[18].
Individual case reports of PCP treatment with drugs targeting BRAF and/or other RAS-RAF-MEK-ERK components have shown encouraging results
[19]. Very recently, a phase 2 clinical trial of combined vemurafenib/cobimetinib treatment in PCP showed a response in 94% of participants, with a median tumor reduction of 91% at 22 months, for progression-free survival of 87 and 58% at 12 and 24 months, respectively
[20]. In contrast, BRAF inhibitors have not been evaluated as therapeutic agents for CD in clinical trials. There are, however, three single-case reports of
BRAF p.V600E positive posterior pituitary tumors (two with confirmed NKX2-1-expression) treated with dabrafenib, either alone
[21] or combined with cobimetinib
[22] or trametinib
[23]. All tumors had recurred after one or more surgeries plus radiotherapy. One patient developed stable disease
[23] and two experienced significant tumor regression
[21], although the combined therapy resulted in dermatological toxicity.
Figure 1. The RAS-RAF-MEK-ERK signaling pathway in corticotroph cells. Under physiological conditions, this pathway is activated in response to the interaction of extracellular ligands such as growth factors, hormones, or cytokines with a tyrosine kinase receptor. The receptor-like growth factor receptor-binding protein 2 (GRB2) binds to the activated receptor and interacts with the proline-rich sequence at the C-terminus of the son of sevenless (SOS) protein to form the receptor-GRB2-SOS complex, which in turn promotes the GTP-mediated activation of RAS. Activated RAS protein binds to and recruits BRAF to the inner side of the cell membrane, where it is phosphorylated by tyrosine kinases. The C-terminal catalytic domain of BRAF interacts with and phosphorylates MEK1 and 2 into their catalytic VIII subregion. In turn, MEK1 and 2 phosphorylate and thus activate ERK1 and 2 (also known as mitogen-activated protein kinases (MAPK) 3 and 1). In addition to phosphorylating cytoplasmic targets, active ERK1 and 2 enter the nucleus and phosphorylate multiple transcription factors, such as ELK1, ETS, FOS, JUN, and MYC, thereby inducing the expression of their target genes. Via the phosphorylation of RPS6KA1, ERK1 and 2 also activate the transcription factor cAMP response element-binding protein (CREB). The activation of this pathway leads to tissue-specific molecular consequences, although in the pituitary gland and in many other tissues it results in increased cell proliferation and survival
[6][11][24][25]. In corticotroph cells, this pathway also activates
POMC transcription, although the membrane receptor triggering this response in physiological conditions and in corticotropinomas remains unclear
[26]. The
BRAF p.V600E variant leads to the overactivation of this signaling pathway.
2. GNAS
At least ~100 human genes are subjected to genomic imprinting, an epigenetic mechanism that controls gene expression in a parent-of-origin and tissue-specific manner
[27]. Using differentially imprinted promoters, one of these genes,
GNAS (locus of the GNAS complex, 20q13.32), ultimately translates into multiple proteins, namely XLαs, ALEX, NESP55, and Gsα
[28][29]. The latter, encoded by a 13-exon reference transcript (NM_000516.7), accounts for the 394-amino-acid α subunit of the heterotrimeric stimulating G protein (P63092-1)
[30]. Gsα is translated from the maternal allele in the pituitary, thyroid, and gonads, but depends on biallelic expression in other tissues
[31].
At the molecular level, guanine nucleotide-binding proteins (G proteins) function as information transducers between the cell-membrane-bound G-protein-coupled receptors (GPCRs) and their effectors, thereby regulating the production of second messengers
[32]. G proteins are composed of α, β, and γ subunits (encoded by different genes) and form a complex that binds GPCRs
[33]. Gsα is made of a C-terminal RAS-like guanosine triphosphatase (GTPase) that also functions as an interaction site for the β and γ subunits and an N-terminal helicoidal domain
[34]. A nucleotide binding cleft exists in between those two domains, which binds guanosine diphosphate (GDP) while the GPCR is inactive. Following GPCR activation through ligand binding, Gsα exchanges GDP for guanosine triphosphate (GTP) and dissociates from the βγ dimer and the receptor, thereby allowing for the GNAS-dependent activation of adenylyl cyclases (ACs)
[35][36]. ACs in turn catalyze the synthesis of cyclic 3′,5′-adenosine monophosphate (cAMP), which then activates downstream signaling pathways
[33]. This activation cycle is negatively regulated by the intrinsic GTPase activity of Gsα, which prevents the continued activation of downstream effectors
[35] (
Figure 2). The effects of multiple hormones greatly depend on cAMP, and the specificity of the cellular responses elicited by this second messenger is determined in a tissue-specific manner
[37].
Missense
GNAS variants affecting residues R201 (namely p.R201C, p.R201S, and p.R201H), and G227 (p.G227R, p.G227L, and p.G227K) of GNAS have been described in endocrine tumors and other human neoplasms. They have been found as somatic changes in somatotropinomas (4.4–59.5%), non-functioning PitNETs (7–10%), corticotropinomas (6%), autonomous thyroid adenomas (5%) and thyroid cancer (13% of PTC and up to 4% of follicular tumors), and occasionally, in ovarian and testicular Leydig cell tumors, prolactinomas, adrenocortical adenomas, pheochromocytomas, paragangliomas, parathyroid adenomas, and in patients with multiple endocrine tumors
[38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54][55][56][57][58][59][60][61][62][63][64][65][66]. These variants have also been found in non-endocrine malignant neoplasms, such as pancreatic, colorectal, and lung adenocarcinomas, as well as in hepatocellular carcinomas
[67][68][69][70].
GNAS variants also underlie the McCune–Albright syndrome (MAS, MIM #174800), a rare condition with sporadic presentation characterized by genetic mosaicism due to early postzygotic
GNAS hotspot defects
[71][72]. The diagnosis is established in the presence of two or more of the classic MAS features: polyostotic fibrous dysplasia,
café-au-lait skin spots, and endocrine hyperfunction (gonadotropin-independent precocious puberty, hyperthyroidism, early-onset Cushing’s syndrome, and PitNETs, usually GH or GH and prolactin-secreting, among others)
[73]. Ninety-five percent of MAS cases are due to variants in R201, while only 5% are caused by variants in Q227
[74][75][76]. The phenotype is determined by genomic imprinting and the disease severity correlates with the degree of mosaicism, meaning that the clinical presentation depends on the time of appearance of the
GNAS variant during embryogenesis
[77].
GNAS hotspot variants cause the loss of protein function that results in increased activity of the cAMP signaling pathway, by (1) stabilizing Gsα in its active conformation, thereby mimicking the effect of extracellular growth factors by stimulating ACs, and (2) inhibiting GTPase activity and causing a constitutive activation of ACs
[38][78]. For these reasons, these
GNAS defects are often referred to as activating variants or
gsp oncogene
[38]. Restoring the GTPase activity of
GNAS is an attractive therapeutic target, although drugs with this specific effect have not been reported yet. In contrast, non-hotspot loss-of-function (LOF)
GNAS variants cause Albright’s hereditary osteodystrophy
[79].
The clinical consequences of
GNAS variants have been thoroughly studied in somatotropinomas. Some studies have defined a particular
GNAS-associated phenotype, with patients usually being older and presenting significantly smaller tumors associated with low serum GH or IGF1 levels
[39][57][64][65]. Other studies have described
GNAS-driven tumors as having a slow growth rate and a better response to pharmacological or surgical treatment compared with wild-type tumors
[64][66][80]. These tumors are usually of the densely granulated subtype at the histopathological examination
[81]. Differences in age, sex, and other clinical characteristics have been suggested by some studies
[57][82][83]. At the molecular level,
GNAS hotspot variants define a distinctive subgroup of somatotropinomas that display hypomethylation, limited chromosomal alterations, and activation of the GPCR pathway, although results vary among studies
[66][84][85]. In both sporadic and MAS-related somatotropinomas,
GNAS variants almost always affect the maternal allele
[86]. While wild-type somatotropinomas often display relaxation of the paternal imprinting, this phenomenon is infrequent in tumors carrying
GNAS variants
[65][87][88]. The relaxation of
GNAS imprinting correlates with lower
GNAS,
SSTR2, and
AIP expression, suggesting a possibly reduced response to somatostatin receptor ligands
[65].
Figure 2. The cAMP pathway in somatotroph cells. G proteins are composed of three subunits, and the α subunit contains high-affinity binding sites for guanine nucleotides. The GDP-bound form binds tightly to βγ and is inactive, whereas the GTP-bound form dissociates from βγ and is the active form. GPCRs cause the activation of G proteins by facilitating the exchange of GTP for GDP on the α subunit, which in turn activates ACs. These enzymes use ATP as a substrate to produce cAMP. The latter binds to the regulatory subunits (R) of PKA, allowing for the release of the catalytic subunits (C). Active PKA catalyzes the serine/threonine phosphorylation of target molecules, including the transcription factors CREB, CRE modulator (CREM), and activating transcription factor 1 (ATF1). In complex with co-activators such as CREB-binding protein (CBP) and members of the cAMP-regulated transcriptional co-activators (CRTC), these transcription factors bind the 8 bp palindromic sequence known as cAMP response element (CRE) in the promoter region of target genes to increase their transcription. In somatotrophs, the GH-releasing hormone receptor (GHRHR) is the main GPCR activating this pathway, promoting both cell proliferation and
GH transcription
[33][34][35][36][89].
GNAS hotspot variants result in the constitutive activation of this pathway.
4. DICER1
The DICER1 syndrome (MIM #601200) is an autosomal dominant condition of tumor predisposition that encompasses otherwise infrequent dysembryonic tumors, such as pleuropulmonary blastoma (PPB), cystic nephroma (CN), ovarian sex cord stromal tumor, nasal chondromesenchymal hamartoma, ciliary body and cerebral medulloepitheliomas, anaplastic kidney sarcoma, pineoblastoma, embryonal rhabdomyosarcoma (ERMS), and pituitary blastoma (PitB)
[90]. Other associated neoplasms are Wilms tumor (WT), juvenile hamartomatous intestinal polyps, and differentiated thyroid carcinoma, as well as benign lesions such as multinodular goiter and pulmonary cysts.
This syndrome presents usually at an early age and occasionally in young adults and is caused in most cases by germline heterozygous LOF
DICER1 (14q32.13) variants that appear de novo in 10–20% of cases
[91][92][93][94]. Ten percent of cases are due to somatic mosaicism for
DICER1 variants, which has been associated with earlier disease onset, more
DICER1-associated tumors, and a distinctive presentation known as GLOW syndrome (global developmental delay, lung cysts, overgrowth, and Wilms tumor)
[93][95][96].
The 29-exon
DICER1 canonical transcript (NM_030621.4) encodes a widely expressed 1922 amino acid cytoplasmic enzyme (Q9UPY3-1) composed, from N- to C-terminal, of a helicase domain, a domain of unknown function (DUF283), a platform domain, a P-element-induced whimpy tested (PIWI)-Argonaute (AGO)-Zwille (PAZ) domain, a connector domain, the class 3 ribonuclease (RNase III) a and b domains, and a double-stranded RNA (dsRNA)-binding domain
[97]. DICER1 plays a crucial role in the processing of small RNAs, which are the RNA species involved in gene silencing. It first cleaves pre-miRNAs and long dsRNA substrates into mature microRNAs (miRNAs) and small interfering RNAs (siRNAs), respectively
[98][99]. Then, DICER1 participates in the loading of siRNAs and miRNAs onto the RNA-induced silencing complex (RISC), composed of DICER1, an AGO protein, and the RISC-loading complex subunit transactivating response RNA-binding protein (TARBP2)
[100]. The AGO protein selects a strand of the small RNA as a guide, which in turn directs the small RNA-bound RISC complex toward complementary messenger RNA (mRNA) sequences. The mRNA targets are then either cleaved by AGO (RNA interference) or translationally repressed and directed to degradation (miRNA-mediated gene silencing); the latter mechanism predominates in mammalian cells
[101] (
Figure 3).
Most individuals carrying germline
DICER1 variants also harbor somatic second hits, which in most cases are missense changes and rarely loss of heterozygosity (LOH)
[96]. Moreover, somatic deleterious
DICER1 variants have been reported in the presence or absence of germline defects in patients with PPB, CN, WT, non-epithelial ovarian tumors, cervical ERMS, PitB, prostate carcinoma, pineoblastoma, differentiated thyroid carcinoma, and testicular germ cell tumors
[102][103][104][105][106][107][108][109][110][111][112]. Different to germline variants, which are usually truncating and are not clustered in hotspots, most mosaic and somatic variants occurring isolated or as second hits are missense and located within the RNase IIIb domain
[96][109][113].
Nineteen out of the twenty PitBs genotyped so far were due to LOF
DICER1 variants, although it is not clear if any cases were caused by somatic defects
[94][114][115][116]. These tumors usually affect neonates or infants, but one case diagnosed in childhood and one presenting in young adulthood have been reported
[114][115][116]. These extremely rare and poorly differentiated anterior pituitary neoplasms with a so-called
oncofetal molecular signature usually express ACTH and may present clinically silently or as CD
[109][117][118]. Nine of these patients died during infancy or childhood due to tumor-related complications
[115][116]. Because PitB is considered a pathognomonic lesion of the DICER1 syndrome, its diagnosis should prompt germline
DICER1 screening and genetic counseling
[94].
RNAse IIIb variants affect metal ion binding and adjacent amino acids, specifically 1705, 1709, 1809, 1810, or 1813, which are therefore considered missense hotspots
[90][99]. Second somatic variants outside the hotspot as well as LOH have also been described in patients with somatic mosaicism for RNAse IIIb variants
[93][112]. The abnormal RNase IIIb cleaves 5′-derived miRNAs from the pre-miRNA hairpin loops inefficiently, causing retention of pre-miRNA loop sequences and leading to reduced expression of 5′-derived mature miRNAs and predominance of 3′-derived pre-miRNAs
[112]. The oncogenic capacity of the biased pre-miRNA repertoire seems to depend on the cellular and developmental setting
[113].
Aside from its role as a tumor driver, reduced
DICER1 expression due to haploinsufficiency or other mechanisms correlates with bad outcomes in multiple types of cancer
[99]. In these tumors, unprocessed pre-miRNAs are degraded by the endonuclease complex TSN-TSNAX. Pharmacological or shRNA-mediated inhibition of this complex facilitates the restoration of miRNA levels by DICER1 in vitro, making it a potential therapeutic target
[119][120]. This strategy, however, has not yet been explored in tumors carrying
DICER1 hotspot variants.
Figure 3. RNA processing pathways involved in PitNETs. (
a) Biogenesis of small RNAs. In the nucleus, RNA polymerases II and III (RNA Pol II and III) generate primary miRNA transcripts (pri-miRNAs) from miRNA-encoding genes, which are then processed by the microprocessor complex, including the DROSHA RNaseIII. This initial step renders ~60-nucleotide-long hairpin-folded pre-miRNAs, which are in turn exported to the cytoplasm via exportin 5 (XPO5)/Ran-GTP. In the cytoplasm, DICER1 cleaves pre-miRNAs and long dsRNAs into mature miRNAs and siRNAs, respectively, both of which are 20–22 nucleotide-long double-stranded RNAs. The DICER1-dsRNA complex is then bound by a member of the AGO protein family (AGO2 is the best characterized of them) and TARBP2 to form the RISC-loading complex. This complex in turn loads dsRNAs into the RISC, which is required to produce single-stranded small RNAs that serve as a guide to recognize complementary RNA sequences (located in the 3′ untranslated region of mRNAs). The small RNA-loaded RISC can either block translation and promote degradation or directly cleave (via AGO proteins) target mRNAs. Additional roles for DICER1 in the responses to DNA damage (nuclear) and viral infections (cytoplasmic) have recently been described. In PitBs, this abnormal repertoire of small RNAs results in
PRAME dysregulation, among other transcriptional alterations
[99][101][121][122][123].
DICER1 variants result in abnormal processing of small RNAs, thereby impairing their ability to regulate gene expression. (
b) Processing of mRNAs by the spliceosome. The spliceosome is a large complex of snRNPs and other proteins that carries out the removal of introns and the ligation of exons from mRNA precursors (pre-mRNAs), rendering mature mRNAs. Two types of spliceosomes, U2-dependent and U12-dependent, are recognized in eukaryotes, the former being the predominant one. The U2-dependent spliceosome is composed of U1, U2, U5, and U4/U6 snRNP, as well as other proteins. This process beings when the U1 snRNP binds to the 5′ SS to form the E complex. Then, the non-ribonucleoprotein complex components SF1, U2AF2, and U2AF1 bind the BS (18–40 nucleotides upstream from the 3′ SS), the polypyrimidine tract (a sequence immediately downstream from the BS), and an AG dinucleotide at the intron-exon junction, respectively. The U2 snRNP in turn replaces SF1, forming the A complex, and the U5, and U4/U6 snRNPs are then recruited to form the precatalytic B complex. Rearrangements in RNA–RNA and RNA–protein interactions ultimately lead to dissociation of the U1 and U4 snRNPs, thereby producing the active B complex. The latter is activated by the pre-mRNA-splicing factor ATP-dependent RNA helicase DHX16, thereby generating the B∗ complex, which catalyzes the first step of splicing. The C complex is then formed, triggering the second step of splicing. Finally, the spliceosome is removed and recycled.
SF3B1 hotspot variants lead to the use of cryptic pre-mRNA 3′ SSs, and aberrantly spliced mRNAs are degraded via NMD
[124][125][126][127][128][129][130]. The repertoire of aberrantly spliced mRNAs involved in lactotroph tumorigenesis remains unknown.
3. SF3B1
Using genome sequencing in 21 patients and targeted genotyping by ddPCR in the rest, a recurrent missense somatic variant (c.1874G>A, p.R625H) in the splicing factor 3B subunit 1 gene (
SF3B1, 2q33.1, NM_012433.4) was identified in 20% of prolactinomas of a single cohort of 227 cases
[131]. When 154 PitNETs of other types were tested, this variant was only found in 6% of cases, all of them staining positive for prolactin. Individuals carrying
SF3B1 p.R625H displayed significantly higher prolactin levels and a shorter progression-free survival, compared with
SF3B1 wild-type cases. A recent Sanger sequencing-based study identified the same variant and an additional missense variant in the same residue (c.1873C>T, p.R625C) in 7 out of 282 prolactinomas analyzed (2.5%)
[132]. Interestingly, 50% of metastatic prolactinomas carried
SF3B1 hotspot defects. In line with the earlier findings,
SF3B1 variants were associated with a larger tumor size and increased mortality, but also with a higher Ki67 index and a need for more therapeutic interventions.
SF3B1 encodes a component of the U2 small nuclear ribonucleoprotein (snRNP) complex and is therefore a component of the pre-mRNA splicing machinery. SF3B1 is involved in 3′ acceptor splice site (SS) recognition, as well as in recruiting other U2 snRNP subunits to the branch point (BP) of pre-mRNAs via interaction with the BP and U2AF2
[133] (
Figure 3). The canonical form of SF3B1 (O75533-1) is a 1304-amino-acid protein containing an unstructured N-terminal region, while the C-terminal two-thirds of the protein constitute a huntingtin, elongation factor 3, regulatory A subunit of protein phosphatase 2A, and TOR1 (HEAT) domain, composed of 20 tandem repeats
[124].
Recurrent somatic variants in hotspots within the fifth and ninth HEAT repeats have been found in myelodysplastic syndrome, chronic myelomonocytic leukemia, acute myeloid leukemia, myeloproliferative neoplasms, primary myelofibrosis, chronic lymphocytic leukemia, breast cancer, pancreatic ductal adenocarcinoma, uveal, mucosal, and cutaneous melanoma, and prostate cancer
[125][134][135][136][137][138][139][140][141]. Aberrant splicing is a well-known tumorigenic mechanism, and indeed, abnormal splicing patterns have been demonstrated in neoplasms carrying
SF3B1 hotspot variants in some
[134][135], although not all studies
[139]. In prolactinomas, p.R625H (in the fifth HEAT repeat) leads to aberrant splicing of estrogen-related receptor gamma (
ESRRG) mRNA, resulting in stronger interaction with the pituitary-specific positive transcription factor 1 (POU1F1) and excessive prolactin secretion
[131]. This variant also causes aberrant splicing and downregulation of
DLG1 in human prolactinomas and rat somatotropinoma GH3 cells. In the latter, the variant causes an epithelial–mesenchymal transition phenotype
[142].