A subtype of oral, laryngeal, esophageal, and lung, squamous cell carcinoma with high levels of TrkB-T1 neurotrophin receptor mRNA

Squamous cell carcinoma (SCC) is a cancer identified histologically, with the cells resembling, and possibly derived from, the flattened epithelial cells of skin or mucosal epithelium. Nearly 95% of head and neck carcinomas (HNC), 30% of lung cancers, and 50–90% of esophageal cancers, depending on geographic site, are SCCs [1‐3]. The tumors, while histologically similar, have distinct treatments depending on the anatomic site. With head and neck SCCs there are at least 4 tumor subtypes based on mRNA gene expression [4, 5] and tumors of the oral pharynx are different in that they often have HPV infection in their etiology [6, 7]. Molecular characterization can be used to classify head and neckl SCCs and has potential in treatment optimization.

The intact, nonmutated TrkB neurotrophin receptor is thought to play a role in progenitor neural cell migration, survival, and differentiation [8‐10]. In oral, head and neck, ovarian, pancreatic, colon, prostate, and gastric cancers and neuroblastomas, high TRKB protein levels correlate with worse patient outcomes [10‐18]. Rare mutated versions of the NTRK2 gene, which encode proteins consisting of fusions of the TrkB kinase domain with domains of other signaling proteins, are drivers in a number of cancers [17, 19‐22]. The intact NTRK2 gene is also a potential oncogene [12, 13, 23]. The largest form of the NTRKB gene products, TrkB-FL, binds Brain Derived Neurotrophic Factor (BDNF) which is known to activate three signaling pathways via autophosphorylation. This is followed by the recruitment of intermediate signaling proteins that activate the PI3K/AKT signaling pathway, which promotes neural cell outgrowth and cell survival, the ERK/MAPK pathways that control survival, and pathways involving phospholipase C that enhance neuronal plasticity [24‐27]. While most of the processes have been best studied in nervous system cells, there is good evidence TrkB-FL overexpression can activate PI3K/AKT pathways, MEK/ERK cascades, cell proliferation, epithelial mesenchymal transition, and multiple metastasis-promoting properties in tumors [18, 28]. There is a second form of TrkB, TrkB-T1, also a membrane BDNF receptor, with a distinct C-terminus that lacks its own kinase domain and instead has a unique 12-amino acid at the carboxy terminus of its shortened cytoplasmic domain. TrkB-T1 can form heterodimers with kinase active forms, which may at times work as a dominant negative of TrkB-FL [8, 29]. TrkB-T1 can also compete for binding of BDNF, which it then internalizes [29]. TrkB-T1 plays a major role in neurite filopodia outgrowth in the cell and changes in cytoskeleton in glioma cells [8, 9, 30]. Little else is known mechanistically about how TrkB-T1 alters other cell phenotypes in neural cells, such as proliferation or survival [30‐32]. What it does in tumor cells is similarly poorly understood, it has been shown to protect mammary cells from apoptosis in the presence of BDNF and in pancreatic cancer cell lines it can increase proliferation and cell migration [33, 34].

Accurate measurement of NTK2 gene expression can be difficult. The TrkB receptor protein is differentially glycosylated and there are multiple forms produced from differentially spliced NTRK2 gene mRNA [35]. While measurement of TrkB protein levels in tumors may prove useful for prognosis, immunohistochemistry-based quantification is variable among head and neck cancers. The percentage estimates of head and neck SCCS expressing high levels of TrkB vary widely [11, 18, 35‐37]. Antibodies that differentiate the most common forms of the protein are not readily available and the molecular weight of the most often studied TrkB-FL is nominally 92KD though it is often identified at 145 KD on denaturing western analysis, a range that overlaps with other isoforms [18, 38]. In many studies it is not clear which form of NTRK2 gene mRNA or protein has been measured [15].

The work here exclusively focuses on the measurement of the two most common forms of NTRK2 gene RNAs: TrkB-FL and TrkB-T1. This work highlights that TrkB-T1 is the most abundant form of the mRNA in all tumor types examined as has been shown for many normal cell types in the body [35]. The ease of reproducible measurement of TrkB-T1 mRNA, via detection of the terminal exon using DNA microarrays or RNASeq, was used to determine if tumors highly enriched for TrkB-T1 mRNA make up a separate subtype of head and neck SCC and to discern pathways that are enriched in this subset of SCCs.

The Cancer Genome Atlas (TCGA) gene expression (RNAseq V2), RSEM (RNA-Seq by Expectation-Maximization), CNA copy number amplification and clinical data for the 518 lung adenocarcinoma (LUAD), 501 lung SCC (LUSC), 284 oral SCC (OSCC), 96 laryngeal SCC (LASC), 90 esophageal adenocarcinoma (ESAD), and 97 esophageal SCC (ESSC) samples, data originally curated by TCGA Research Network were obtained from the Broad Institute TCGA GDAC Firehose repository via Firebrowse (http://fbdev/api-down.html). The Cancer Proteome Atlas (TCPA) reverse-phase protein array analysis data (V4.2, 2018/07/18) of 353 LUAD, 322 LUSC, 206 OSCC, and 79 LASC was downloaded from the MD Anderson TCPA website [39] https://​www.​tcpaportal.​org/​tcpa/​download.​html. Level 4 replica-based normalization was used to minimize batch effects.

The original intents of this study were to use the accuracy of RNA measurement to characterize SCCs based on NTRK2 gene expression and to develop a reproducible method to identify aggressive tumors. The observation that TrkB-FL makes up only a tiny percentage of total NRTK2 gene mRNA put the focus on TrkB-T1 mRNA (Additional file 1: Figure S1). TrkB protein analysis in many studies has indicated higher levels of TrkB protein are found in a large subset of tumor types, but has not allowed a thorough description of the differences in high versus low TrkB expressing tumors [8, 10, 16, 17]. The work described here showed that TrkB-T1 RNA, which is at high levels and can be measured accurately, can be used to subclassify SCC tumors of different organs as a group which turn out to have some shared properties.

The TrkB-T1 high expresser tumors fall mainly into two of four head and neck SCC subtypes described earlier, atypical and classical [4, 5]. They showed increased expression of.

NFE2l2, SOX2, and PIK3CA associated with both those subclasses of head and neck SCC [4]. Compared to tumors with low-level expression of TrkB-T1, the TrkB-T1 high expressing tumors showed increased levels of neural specific mRNAs (Additional file 3: Table S2). These high TrkB-T1 SCCs would likely be a subset of the neuroendocrine-like tumors or C4 classification of Chen et al. from their TCGA pan-cancer analysis, though they are distinct in that only SCCs are included and they all express TrkB-T1 mRNA at high level [60]. In that TrkB plays a role in neuronal cell behavior in development, it may contribute to regulation of neural-specific gene expression in these tumors. TrkB-T1 expression showed a strong correlation in OSCC, LASC, ESSC, and LUSC tumors groups in this study with the mRNA for the Sox2 neural developmental transcription factor [51, 52, 61]. Other candidates to control the high TrkB-T1 SCC mRNA levels seen in these tumors include Tcfap2a, a transcription factor involved in neural crest formation and/or function [62]. The Tcfap2a transcription factor binding site sequence was overrepresented in the promoters of DE genes in high TrkB-T1 mRNA expressers versus low expressers in all 4 SCCs studied [32, 44].

SCCs, whether laryngeal, oral, lung, or esophageal, that expressed high levels of TrkB-T1 mRNA also showed enrichment of the same KEGG pathways identified by gene set analysis compared to their low TrkB-T1 counterparts. (Fig. 2c). This duplication of results with high TrkB SCCs at different body sites solidifies the association of these pathways with TrkB-T1 mRNA expression and these SCCs as a distinct entity. The enriched pathways included hedgehog, long associated with Basal Cell Carcinoma and normal development [63]. This pathway could play a role in tumor formation and progression in SCCs [64, 65]. Sox2 may serve as a link between TrkB-T1 RNA levels and the hedgehog pathway. Sox2 has been shown to play a role in the function of hedgehog signaling [51, 66‐68]. Specific proteins of the pathway and Sox2 work together to determine cell fate. The mechanism for co-expression of TrkB-T1 and Sox2 in these SCCs is unclear [61]. Analysis of 237 proteins as part of the Cancer Proteome Atlas revealed Tp63 which stimulates expression of sonic hedgehog pathway genes in mammary cancer stem cells was enriched in high TrkB-T1 expresser SCCs (Fig. 4) [69].

A second pathway, retinol metabolism, which includes differentially expressed enzymes, some of which can deactivate retinols, suggests the idea that TrkB-T1 high expresser SCCs would be resistant to retinoids [70], a family of drugs once tested for their curative effects on OSCC, but found to be ineffective overall [71, 72]. Gene set analysis among disease- and drug-related pathways in the MSigDB revealed that a large number of the 450 genes that comprise the Nfe2l2 pathway were enriched in the TrkB-T1 high expressers of each SCC. This occurred to a much lesser degree in the two adenocarcinoma groups (Additional file 4: Figure S2). Nfe2l2 induces transcription of genes of the antioxidant and detoxification pathways [53]. These include glutathione metabolism and xenobiotic/drug metabolism by cyp450, two KEGG pathways associated with high TrkB-T1 mRNA levels in SCC. In non-SCC breast cancer, TrkB kinase activity has been shown to reduce levels of the Keap inhibitor, which increases Nfe2l2 directed transcription [73]. In addition BDNF has been shown to induce Nfe2l2 mRNA in astrocytes via TrkB-T1 activation [74]. Keap1 protein was enriched in high TrkB expressing LASC, OSCC, and LUSC, as was p62/SQSTM1 (Fig. 4). Keap1 is an inhibitor of the Nfe2l2 pathway, while p62/SQSTM1 is an activator [53, 75]. In that many Nfe2l2 targets are enriched in TrkB-T1 high expresser SCCs one might speculate that p62/SQSTM1 is more active in those tumors.

While the correlation between TrkB-T1 mRNA level and PIK3CA copy number was marginal in the SCCs (Additional file 8: Table S4), the functional interaction of these genes in these tumors was further supported by the increase in mutagenic PIK3CA activation in high TrkB expressers in LASC for example (Fig. 6). We note TrkB-FL kinase is known to work via activation of PI3K and downstream AKT [17, 18]. Recent work showed TrkB-T1 can activate ERK and AKT signaling both of which are downstream of Pi3K [76]. These and other findings allow speculation that PI3K activation and increased transcription of TrkB-FL and TrkB-T1 are mutually stimulatory [9, 77, 78]. This study provides some of the first evidence Sonic Hedgehog and Retinol metabolism are related to NTRK gene expression in SCC tumors, and some of the first for the Nfe2l22 pathway. However, a limitation of the study is that it is not clear if TrkB-T1 or, for that matter, TrkB-FL protein contribute to specific patterns of gene expression seen in the TrkB-T1 high expressers.

The observation that high TrkB-T1 LASC patients show poor outcomes, while the opposite is true in high TrkB-T1 LUSCs creates a question (Fig. 6a and c). If TrkB-T1 mRNA enrichment and the set of pathways associated with this enrichment in the analysis are relevant to this TrkB-high SCC tumor subtype then why do TrkB-T1 high expressers still have organ site-specific properties in regard to patient outcomes? Importantly, enrichment of TrkB-T1 mRNA is correlated to varying degrees with enrichment of Nfe2l2 pathway, Pik3CA of the PI3K pathway and hedgehog pathways and Sox2. These are pathways and factors known to have SCC site-specific effects on patient outcomes and on resistance to radio- and chemotherapy in at least some tumor types, [58, 79‐82]. Pik3ca enrichment in head and neck SCC predicts poorer outcomes [80, 81] while amplified PIKC3A in LUSC may have the opposite effect [83](Fig. 6). Likewise, enrichment of factors linked to activated Nfe2l2 are associated with poor survival of head and neck SCC and LUAD, but that is not the case for LUSC and other cancers [58, 59, 82, 84, 85]. Finally, Sox2 has tumor-specific associations with patient outcomes similar to TrkB: negative in LASC, mixed for head and neck SCC overall, and positive in LUSC [79, 86, 87]. So, while SCCs with high levels of TrkB–T1 mRNA may make up a subtype of SCC, where the same suite of pathways is enriched, the tumor behavior continues to depend on how the cell responds to activation of those pathways. This can be due to differences in background gene expression in the high TrkB-T1 SCCs at different sites or can be due to external differences associated with that tumor site, such as the standard treatment for that tumor type.