The Splicing Factor PTBP1 Represses TP63 γ Isoform Production in Squamous Cell Carcinoma

The TP63 gene encodes the p63 transcription factor. It is frequently amplified or overexpressed in squamous cell carcinomas. Owing to alternative splicing, p63 has multiple isoforms called α, β, γ, and δ. The regulatory functions of p63 are isoform specific. The α isoform inhibits the epithelial-to-mesenchymal transition (EMT) and controls apoptosis, while the γ isoform promotes EMT. Using The Cancer Genome Atlas data, we observed that a higher proportion of the TP63γ isoform is a detrimental factor for the survival of patients with head and neck squamous cell carcinoma (HNSCC) and is accompanied by the downregulation of desmosomal genes. By a correlation-based approach, we investigated the regulation of the production of the TP63γ isoform. According to our analysis of GTEx data, the expression of the RNA-binding protein PTBP1 (polypyrimidine tract binding protein 1) is negatively correlated with the abundance of TP63γ in several tissues. Accordingly, we demonstrated that PTBP1 depletion in HNSCC cell lines, keratinocyte or Xenopus embryos leads to an increase in TP63γ isoform abundance. By RNA immunoprecipitation and in vitro interaction assays, we showed that PTBP1 directly binds to TP63 pre-mRNA in close proximity to the TP63γ-specific exon. Intronic regions around the TP63γ-specific exon were sufficient to elicit a PTBP1-dependent regulation of alternative splicing in a splice reporter minigene assay. Together, these results identify TP63γ as an unfavorable prognostic marker in HNSCC, and identify PTBP1 as the first direct splicing regulator of TP63γ production and a potential route toward TP63 isoform control. Significance: Quantifying TP63γ isoforms in patients’ tumors could allow for the early detection of patients with HNSCC with an early loss in desmosomal gene expression and poor prognostic. The identification of PTBP1 as a transacting factor controlling TP63γ production may allow to control TP63γ expression.


Introduction
The TP gene encodes a conserved transcription factor, p63, controlling epithelial development and homeostasis (1,2). In humans, heterozygous mutations in TP lead to developmental syndromes affecting ectodermal tissue derivatives (3). Overexpression or amplification of TP is observed in multiple cancer types generally from epithelial origin (4) and p63 staining therefore serves as a classification biomarker in some malignancies including skin can-Univ Rennes, CNRS, IGDR (Institut de Genetique et Developpement de Rennes) -UMR 6290, F-35000 Rennes, France. a transcriptional inhibitory domain (14), and a regulatory sumoylation region (15). The shortest mRNA isoform TPγ encodes only the oligomerization and DNA-binding domains. TPγ mRNA is generated by the use of an internal terminal exon and thus has a different 3 untranslated region to the other C-terminal isoforms. Because all p63 isoforms contain the oligomerization domain, they may all interfere with each other's functions. Most described functions of the p63α isoform concern its requirement for epithelial development and homeostasis, whereas the p63γ isoform alone is unable to support proper epithelial development (16). Instead, p63γ promotes the onset of an epithelial-to-mesenchymal (EMT) transition when overexpressed in keratinocytes (17), promotes osteoblastic differentiation (18) and favors terminal myogenesis (19). Metastatic potential is a major parameter of the severity of HNSCC and is enhanced by the acquisition of EMT properties (20). Differential splicing regulation altering the abundance of C-terminal p63 isoforms could therefore promote acquisition of metastatic properties and be detrimental to patients.
The functions of p63 are therefore strongly dependent on the nature of the isoforms and on the regulation of the production of the different 5 and 3 isoforms in a tissue-specific manner. Despite the acknowledged importance of the differential role of the p63 C-terminal isoforms, the regulation of the multiple alternative splicing events involved in TP pre-mRNA maturation has remained unaddressed since the cloning of the gene in the late 90s (21).
Regulation of alternative splicing events requires the assembly of the spliceosome machinery onto splice sites [noted 5 splice site (5 SS) and 3 splice site (3 SS) with regards to the introns] and/or the definition of the cleavage and polyadenylation sites (CPA; ref. 22). The tissue-specific choices among different alternative splicing and polyadenylation events are generally controlled by a combination of regulatory sequence elements, silencers or enhancers, and their cognate trans-acting factors, generally RNA-binding proteins (RBP). The differential tissue distribution of these RBPs will often account for the tissuespecific regulation of alternative splicing (23). We reasoned that by analyzing relative abundance of RBPs and TP alternative splicing events in RNAsequencing (RNA-seq) data from many tissues we could prioritize the RBPs most likely to intervene in TP splicing.
Here, by combining exploration of The Cancer Genome Atlas (TCGA) and GTEX data, RNA/protein biochemistry, and reverse genetics experiments, we show that higher inclusion of the γ exon is associated with poorer survival of patients with HNSCC and we identify the RBP PTBP1 as a direct inhibitor of the inclusion of the γ exon in HNSCC cell lines. Furthermore, patients' primary tumors are characterized by a decreased expression of desmosomal gene that are known actors of tumors suppression (24), suggesting a loss of epithelial characteristics. These results identify TPγ as an unfavorable prognostic marker in HNSCC, and identify PTBP1 as the first direct splicing regulator of TPγ production and a potential route toward p63 isoforms control.

Cell Lines
The panel of HNSCC cell lines (TCP-1012:A253, FaDU, SCC-25, SCC9) was acquired from ATCC in July 2019 and authenticated by the provider by short tandem repeat profiling. Authenticated HaCaT cells were obtained from Cell Lines Service in September 2020 (CLS GmbH). All cell lines used for the experiments were at least three passages and at most 20 passages from thawing, at which point they were systematically discarded. Cells were not tested for Mycoplasma infection. HaCaT, FaDu, HeLa, and Detroit 562 cells were cultured in DMEM (Gibco), A253 cells were cultured in Mc Coy's 5A (Gibco), SCC-9 and SCC-25 cells were cultured in DMEM/F12 (Gibco) supplemented with 400 ng/mL hydroxycortisone (H0888, Sigma-Aldrich). For siRNA depletion, cells were seeded on 6-well plates at 150,000 cells per well. After seeding, a jetPRIME transfection mix (Polyplus) was added, containing 100 pmol of target siRNA or negative controls.

Xenopus Embryos Microinjection
Xenopus laevis eggs were obtained from wild-type (WT) females and fertilized using standard procedures and following national regulation under protocol (APAFIS#7109; ref. 25). When indicated, 30 ng of MoPtbp1 (26) or 30 ng of control morpholino (GeneTools) were injected into each blastomere of two-cell embryos in a volume of 13.8 nL, using a Nanoject II (Drummond). For rescue experiments, 1 fmol of mRNA-encoding morpholino-resistant Ptbp1-V5R was coinjected. Embryos were allowed to develop at 22°C and were collected at stage 26 according to Nieuwkoop and Faber stages (27). Total RNA was extracted using RNeasy columns (Qiagen) from pools of three embryos. RNA was analyzed by RT-PCR for 25 cycles using radiolabeled PCR primers.

Minigene Construction and Analysis
The TP exon 12 minigene and its mutated versions were constructed by Gibson assembly of β-globin and TP gene fragments obtained by PCR amplification of HaCaT genomic DNA using oligonucleotides pairs hGbE1_E2, hTP63_gamma, hGb_E3 and cloned into ApaI, NheI digested pmirglo (Promega). Mutations of the 3 SS and/or CPA were conducted using geneblocks (IDT) overlapping the regions to be mutated. Plasmids were controlled by sequencing. HaCaT cells were transfected with minigene plasmid and selected using G418 at 800 μg/mL until naive cell died then maintained in 400 μg/mL (Gibco). Expression and splicing of the minigene was assessed by qRT-PCR. All oligonulceotides are presented in the reagent table.

RNA Extraction, Reverse Transcription, and qRT-PCR
Total RNA was extracted from cells using nucleospin RNA kits (Macherey Nagel, #740955.50) following the manufacturer's instructions. A total of 1 μg of total RNA was reverse transcribed with superscript II enzyme (Invitrogen) and random primers (Invitrogen #58875). Dilutions of cDNA (1:20) were amplified with PowerSybr Green Master mix (Applied Biosystems, #4367659) using a Quantstudio (TM) 7 flex real-time PCR 384-well system. Two reference genes were used for qRT-PCR, RPLP, and BTUB.
siRNA, oligonucleotides, antibodies, and additional details on reagents and cell lines are listed in Supplementary Table S1.

Statistical and Data Analyses
TP isoforms and splicing in HNSCC patient samples were quantified using clinical and expression data obtained from firebrowse.org and splicing quantification from TCGA spliceSEQ (28). The percent-spliced-in (PSI) metric is computed as the ratio of normalized reads supporting inclusion of an exon to the total normalized reads for that event. All expression, splicing, and survival analysis were conducted in R. For Kaplan-Meier estimation, the survival package was used and a log-rank test (29) applied. Prognostic marker genes were obtained from https://www.proteinatlas.org. Gene Ontology (GO) enrichment analysis was conducted using the topGO package.
To identify RBPs with tissue expression correlated to TP tissue-specific junction usage, we analyzed RNA-seq data from GTEx Analysis 8.

Fluorescent RNA
Transcription templates were synthesized by PCR (GOTAQ G2, Promega) using a hybrid T7promoter/sequence-specific forward primer and a sequencespecific reverse primer. The PCR templates were either the WT minigene or the mutated ones. PCR products were checked on native agarose gel and purified on columns (DC5, Zymoresearch). Fluorescent RNAs were transcribed (37°C,

Protein Production and Electrophoretic Mobility Shift Assay
The DNA encoding hsPTBP1 (NM_002819.5) was obtained from source biosciences (IRAUp969B052D). The PTBP1 ORF was PCR amplified and subcloned by Gibson cloning into pET21A+ (Novagen) digested by XhoI and NheI.
The 6xHis-tagged hsPTBP1 was expressed in E.Coli (BL21) after induction by IPTG (1 mmol/L), and purified on nickel column using standard procedures. PTBP1 was eluted with 250 mmol/L Imidazole and concentrated on vivaspin  One-tenth of the immunoprecipited samples were used for analysis of immunoprecipitated proteins by SDS-PAGE and Western blotting. Coimmunoprecipitated RNAs were isolated by proteinase K treatment (30 minutes, 37°C in 150 μL of PK buffer (50 mmol/L Tris pH 7.4, 1 mol/L NaCl, 0.5% NP40, 0.5 mmol/L EDTA, 0.1% SDS, 2 mol/L urea, 1% deoxycholate) containing 50 μg tRNA and 12 U proteinase K (Thermo Fisher Scientific) followed by phenol-chloroform extraction and ethanol precipitation. RNAs were analyzed by RT-PCR or qRT-PCR.

Data Availability
All scripts are available from https://gitlab.com/YannAudic/tp63_tcga_gtex. Plasmids and transgenic cell lines are available upon request. No difference could be observed between tumors and normal sample for these splicing events. We sought to determine whether differential exon usage in tumors could be associated with difference in survival probability for patients (Fig. 1D). The exon 12 or exon 14+15 inclusion measurements in tumor samples could discriminate patient survival rates.

Higher TP63 γ Exon Inclusion is Associated with Poor Outcome in Patients with HNSCC
Higher inclusion (PSI ≥ 4.07%) of the exon 12 was associated with a lower survival of patients (median half-life = 862 days compared with 2,319 days,

Higher TP63 γ Exon Inclusion is Associated with Loss of Epithelial Markers in Primary Tumors of Patient with HNSCC
On the basis of the results shown in Fig. 1A, we split the patients with HN-SCC into two groups: those with a low percentage of TPγ and a better prognosis, and those with a higher percentage of TPγ and a worst prognosis. We identified the differentially expressed genes (DEG) between these two groups of patients and we highlighted the HNSCC prognostic marker genes.
We observed that all prognostic marker genes overexpressed in the high TPγ tumors are exclusively unfavorable (red dots, Fig. 2A). Conversely, prognostic genes overexpressed in the low TPγ tumors are mainly favorable (blue dots, Fig. 2A). This is coherent with the decreased survival probability of high TPγ patients. Ontology enrichment analysis of the DEG highlighted nucleoplasm genes, desmosomal genes, and SWI/SNF complex genes as statistically enriched ( Fig. 2B; Supplementary Tables S3 and S4). The most significantly enriched term "nucleoplasm" relates to nuclear components that are affecting chromatin epigenetic modification, transcriptional regulation or replication. Most notably, eight of 24 genes encoding proteins of the desmosomal cell-to-cell adhesion complexes are less expressed in primary tumors of patients with high level of TPγ (Fig. 2C). This is in line with the potential loss of epithelial characteristics that has been attributed to TP63γ (17,34). Seven of 24 genes encoding proteins of the SWI/SNF complex involved in the extensive remodeling of the chromatin are downregulated in the same primary tumors (Fig. 2D), in accor-dance with a loss of function of the SWI/SNF complex that is often associated with carcinogenesis (35). This obviously raises the question of the regulators of TP splicing in different patients.

Candidate RBPs Controlling TP63 γ Exon Inclusion
Because tumor samples may be highly heterogeneous due to tumor type, location, and/or intermixed tissues, we sought to identify potential regulators of TP splicing using data from normal tissues. We searched for RBPs whose expression correlates to TP junction usages. RNA-seq junction quantification was obtained from the GTEX consortium (36). They represent quantifications from 54 different tissues or conditions. The correspondence between GTEX and TCGA nomenclatures for TP is shown in Supplementary Fig. S1A and a table ascribing the different transcripts relative to the ENSEMBL annotation is in Supplementary Fig. S1C. We calculated the expression of TP in each sample and for all tissues (see Materials and Methods). Among the 54 tissues, 10 (10/54, 18.5%) were selected as expressing TP (Fig. 3A). These tissues are mainly of epithelial origin, as expected for TP, but Epstein-Barr Virus (EBV)transformed lymphocytes and skeletal muscles also showed a notable TP expression.
For each tissue and sample, we quantified the usage of the 21 individual TP junctions ( Supplementary Fig. S3A). The junction usage is strikingly different between tissues ( Supplementary Fig. S3B). We can distinguish three groups of tissues: muscle, epithelia, and EBV-transformed lymphocytes. Eleven junctions are used at similar levels across tissues. Among those, four are very weakly used (J4-6, J5-5 , J5'-6, and J13-16) and seven are constitutive splicing events included at high levels in all tissues. Ten junctions are differentially used across tissues. Junctions (J1-2, J2-3, J3-5, J4-5) are representative of differential promoter usage between epithelial tissue ( N isoforms favored) and muscle and EBVtransformed lymphocytes (TA isoforms favored). Muscle specific differences (J11-12 and J11-13, red boxes) are visible. These differences are not observed in EBV-transformed lymphocytes. The competition between these two mutually exclusive splicing reactions (J11-12, J11-13) sharing an identical 5 SS allows for the production of the γ or (α, β, δ) terminal isoforms. The quantification of these two reactions (Fig. 3B) clearly shows that they are oppositely regulated: epithelial tissues favor the exclusion of the γ exon, while oppositely about 85% of the TP transcripts include the γ terminal exon in skeletal muscle ( Supplementary  Fig. S4).
We hypothesized that correlations between RBP expression and junction usage could allow us to pinpoint RBPs acting as splicing enhancers or splicing silencers affecting the tissue-specific inclusion of the γ exon. Using the RBPs represented in the RBPDB, we first determined their expression in our selected TP-expressing tissues. We were able to analyze the expression levels of 386 different RBPs in these tissues (Fig. 3C). On the basis of individual samples, we computed the correlations between RBP expression and junction usage relevant to the γ exon (J11-12 and J11-13). Among the 10 most correlated RBPs (Fig. 3D), eight were positively correlated and two were negatively correlated to γ exon inclusion (J11-12). As expected, the correlation was inverted when considering the groups selected for having high or low levels of γ exon which are respectively associated with lower or higher patient survival (Fig. 1C). Of the 10 RBPs tested, only three (SFRS, PTBP, GRSF) are differentially expressed at the RNA level between tumors from patients with high or low γ exon inclusion (Fig. 3E, Wilcoxon, P < 0.05). SFRS expression is higher in samples with high γ exon inclusion (Wilcoxon, P = 0.0385), which is not in accordance with the GTEX data where SRSF expression is negatively correlated to γ exon inclusion (Fig. 3D). PTBP and GRSF are respectively low and high in patients with high TPγ (Wilcoxon, P = 0.0094 and P = 0.0389; Fig. 3E). The same RBPs are negatively and positively correlated, respectively, with γ exon usage in GTEX data (Fig. 3D). Therefore, PTBP1 and GRSF1 appear to be plausible candidates for γ exon regulation. However, because GRSF1 is mainly a mitochondrial RBP and is weakly expressed in epithelial cells (37) whereas PTBP1 is a well-defined alternative splicing regulator ex-  AACRJournals.org Cancer Res Commun; 2(12) December 2022

PTBP1 Represses Endogenous TP63 γ Exon Inclusion in Squamous Carcinoma Cell Lines
Using two different antibodies, we analyzed total p63 and p63α expression by Western blot analysis (Fig. 4A) in a panel of HNSCC cell lines. Both antibodies detected a major isoform around 70 kDa, coherent with the major Np63α isoform (MW 66 kDa). In addition to the 70 kDa protein, the pan-p63 antibody (4A4) detected lower molecular weight bands undetected using the α-specific antibody (D2K8X). This suggests that isoforms other than Np63α are present at lower levels in the tested cells. The identity of these isoforms cannot be inferred from these experiments. In parallel, we used qRT-PCR to evaluate the abundance at the RNA level of total TP. We estimated that global TP expression levels are highest in SCC25, then A253 and SCC9, then FaDu, Detroit 562, and premalignant keratinocytes HaCaT. As expected from the human protein atlas and the Western blot analysis (Fig. 4A), TP is only faintly detectable in HeLa cells used as a negative control (Fig. 4B). We used isoform-specific qRT-PCR to quantify mRNAs encoding the three predominant (α, β, γ) Cterminal isoforms. Figure 4C shows the proportions of each isoform, relative to total TP. In all cell types tested, the α isoform was the most abundant ranging from 87% to 95% of total TP mRNA. Detroit 562 cells have the lowest α proportion. Accordingly, β and γ isoforms are in higher proportions in Detroit 562 cells with 8% and 2.7% of each, respectively. The abundances of the β and γ isoforms are variable among the cell lines with the Detroit 562 and the SCC25 cells having the highest relative proportion of TPγ (2.8% and 1.5%, respectively) while other cell lines have all less than 1% of TPγ mRNA (Fig. 4C).
We reduced PTBP1 levels in the five HNSCC cell lines using two different siR-NAs. The siPTBP1a was more efficient to deplete PTBP1 as can be seen from the Western blot analysis (Fig. 4D, top), the qRT-PCR (middle), and from the increased expression of PTBP which is known to be under PTBP1 repression (ref. 38; top, Supplementary Fig. S5). Yet, presumably because of an existing posttranscriptional regulatory loop controlling PTBP1 protein levels (39), achieving full depletion of PTBP1 was not possible. In all five cell lines, depletion of PTBP1 led to a significant increase in TPγ (Fig. 4D) and a decrease in TPβ abundance (bottom, Supplementary Fig. S5) without affecting total TP or TPα levels (middle, Supplementary Fig. S5). The TPγ levels are therefore inversely related to PTBP1 levels. This demonstrates that PTBP1 represses the accumulation of the TPγ isoform in all five HNSCC cell lines tested.

PTBP1 Binds to TP63 Pre-mRNAs Sequences In Vivo on the γ Exon 3 SS and CPAs
The control of TP splicing by PTBP1 could be direct or indirect. Direct regulation of TP splicing would require binding of PTBP1 to regulatory elements located in the pre-mRNA of interest. In silico prediction by RBPmap (40) revealed several putative PTBP1 binding sites within TP pre-mRNA (Fig. 5A,   top). Interestingly, these sites are enriched within regions that are evolutionary conserved in vertebrates (Fig. 5A, bottom). This prompted us to examine whether the potential regulation of TP by PTBP1 is also evolutionarily conserved, which would highlight its fundamental importance. The exon/intron structure of the TP locus is well conserved between Xenopus laevis and human with an internal terminal exon corresponding to the human γ exon present in both species (Supplementary Fig. S6A). In frog embryos (Xenopus laevis), the depletion of Ptbp1 leads to an increase of tpγ mRNA as shown by RNAseq ( Supplementary Fig. 6B) and qRT-PCR ( Supplementary Fig. S6C), and this effect is reversed by reintroducing Ptbp1 (Supplementary Fig. S6C). Hence, the PTBP1-dependent regulation of TPγ production is a conserved phenomenon that was already present in the common ancestor of Mammals and Amphibian roughly 360 million years ago (41).
To confirm in vivo that PTBP1 binds to TP pre-mRNAs, we immunoprecipitated specifically PTBP1/RNA complexes from keratinocytes (HaCaT) or squamous cell carcinomas cell lines (SCC9) cell extracts and analyzed the pre-mRNA content of the complexes by RT-PCR ( Fig. 5B and C). The presence of the TP pre-mRNA was specifically detected in input and PTBP1 immunoprecipitates obtained from both cell lines, but not using control Ig or beads only. To determine whether PTBP1 was preferentially associated to TP pre-mRNA or mRNA, we ran qRT-PCR with primers specific to TP mRNAs (ex8-ex9, ex11-12) or pre-mRNA (int10-ex11, int11-ex12, int12-ex13) in HaCaT cells. As shown in Fig. 5D, the most significant and robust enrichment was observed with the primer pair specific to the pre-mRNA region closest to the γ exon (int11-ex12). Control RT-qPCR performed in absence of reverse transcriptase did not show PCR amplification (Supplementary Fig. S7A). Therefore, on TP pre-mRNA, PTBP1 is preferentially bound to the region that is upregulated following PTBP1 depletion.
To determine the affinity of PTBP1 for TP63 pre-mRNA sequences, we performed EMSA using recombinant PTBP1 (Supplementary Fig. S7) and several fluorescently labeled RNA molecules that cover TP pre-mRNA in the vicinity of exon γ (see Fig. 5A for the location of the RNA fragments). Representative EMSA experiments are shown in Supplementary Fig. S7B, and the binding curves for at least four replicate experiments per RNA are presented in Fig. 5E.
This allowed us to determine the Kd of PTBP1 for these RNAs (32). As presented in Fig. 5E, RNA fragment 1 (FRAGT1) is composed solely of β-globin RNA sequences, FRAGT2 is composed in part of β-globin RNA sequences and in part of the TP intronic region, FRAGT5 is located entirely in the intronic region upstream of the γ exon, FRAGT7 is fully located within the γ exon and In the CPA2 RNA, we altered the PTBP1 motif to a UGUU motif rather than CCCC to keep the UG richness required for the activity of the downstream element of the polyadenylation signals (CPA2MUT). This mutation provoked a 2-fold decrease in the affinity of PTBP1 for this sequence (Fig. 5E).
Altogether, this demonstrates the direct and sequence-specific association of PTBP1 with regulatory regions surrounding the γ exon 3 SS and CPA on the TP pre-mRNA in vivo in both a noncancerous keratinocyte cell lines (HaCaT) and a HNSCC cell line (SCC9).

PTBP1 Represses TP63 γ Exon Inclusion in a Minigene in Epithelial Cells
If PTBP1 can indeed control TPγ exon inclusion by directly binding to ciselements within the pre-mRNA, then the splicing of a minigene composed of the  the PTBP1-dependent regulation we discovered in HNSCC cell lines was acting in HaCaT cells on endogenous TP. We independently depleted PTBP1 in HaCaT cells with two different siRNAs as above. PTBP1 depletion strongly increases the abundance of TP γ exon and decreases levels of TPβ in HaCaT cells without affecting global TP (TP_all) or TPα mRNA levels ( Fig. 6A and B) as we previously showed in HNSCC cell lines (Fig. 4D). This demonstrates that, like in HNSCC cell lines, the accumulation of the endogenous TPγ isoform is repressed by PTBP1 in HaCaT keratinocytes.
We generated four HaCaT cell lines stably expressing TP minigenes in WT and mutated versions. The mutated variants of the minigenes harbor the muta-tions depicted in the gel shift experiment (Fig. 5E) on the 3 SS, the CPA or both (3ssCPA). The structure of the minigene is shown in Fig. 7A. When expressed in the cells, measuring splicing of the junction of exon E2 to exon γ (E2γ) accounts for the production of the γ isoform. We decreased PTBP1 levels (RNA levels are shown in Fig. 7B and protein levels in Supplementary Fig. S8A) in these minigene-expressing HaCaT cell lines. We quantified endogenous TP isoforms ( Supplementary Fig. S8B) and TP minigene splicing by qRT-PCR (Fig. 7B). It is worth noting that the PTBP1-dependent regulation of the endogenous TP RNA is identical in all four HaCaT cell lines ( Supplementary   Fig. S8B). The WT TP minigene produces about 30% of E2γ RNA as quantified by qRT-PCR (Fig. 7B, PSI). Mutations of the PTBP1 binding sites close to the 3 SS completely abolish the splicing between E2 and the γ terminal exon (Fig. 7B, 3SS, 3SSCPA) compared with the WT or CPA minigene. This probably indicates that the replacement of all the UCU/UCUU motifs by CCC/CCCC sequences drastically affect the recognition of the 3 SS signal of the terminal γ exon. In the siCTRL or siLUC conditions, the minigene harboring mutations of PTBP1 binding sites on the CPA alone produced 40% of E2γ RNA, to compare to the 30% inclusion from the WT minigene.
Upon depletion of PTBP1, the 3 SS and 3SSCPA mutant remained unaffected, in agreement with the mutated 3 SS being inoperative. Both the WT and the CPA mutant are responsive to PTBP1 depletion. Following PTBP1 depletion, the splicing between exon E2 and the γ terminal exon increased about 2-fold in both the WT and the CPA mutant minigenes (Fig. 7B). To better assess the amplitude of the PTBP1-dependent regulation, we plotted a Delta PSI value calculated as the fold difference of exon γ inclusion between treatment conditions and the reference condition (siCTRL). We then tested whether there were any significant differences between the Delta PSI of the WT minigene and that of the mutated minigene. As can be observed on Fig. 7C, both siRNAs targeting PTBP1 promoted efficient inclusion of the γ exon on the WT whereas the control siLuc treatment barely changed inclusion of E2γ for the different minigenes.
Interestingly, a decreased dynamic of exon γ inclusion on the CPA minigene could be observed in the most potent conditions of PTBP1 depletion (Fig. 7C,   SiP1A). This indicates that the PTBP1-binding site close to the CPA contributes to PTBP1-mediated repression of exon γ inclusion.
These experiments demonstrate that the genomic regions surrounding the γ terminal exon are sufficient to allow for its PTBP1-dependent regulation. Mutational analysis demonstrates that the PTBP1 binding sites are intermingled with elements that define the 3 SS and that mutations over the CPA region that decrease the affinity of this region for PTBP1 can promote a better inclusion of the γ exon in control conditions. Mutations that partially reduce the affinity of PTBP1 for the CPA render the minigene less sensitive to PTBP1 depletion.
As summarized in Fig. 8, these data reveal that in HNSCC, PTBP1 controls TPγ exon inclusion by binding to discrete and conserved regulatory elements located in this pre-mRNA region. In normal and cancerous epithelial cell lines where PTBP1 is abundantly expressed, it reduces the inclusion of the γ exon and therefore decreases the abundance of the TP63γ isoform. The increase of TPγ isoform is correlated to a repression of some epithelial marker genes and to a decreased survival probability of patients with HNSCC.

Discussion
Alternative splicing misregulation is emerging as a major theme in the alteration of gene expression observed during the cancerous process (43) and offers new routes of tumor classification or treatment (44). Here, we identify for the first time an RBP that controls the balance between different splice isoforms of TP, a key gene involved in squamous cell carcinoma and especially HNSCC. We demonstrate using a variety of approaches that PTBP1 inhibits the synthesis of the TPγ isoform. PTBP1 is generally known as an inhibitor of the splicing or polyadenylation reactions. The multiple RNA-recognition motifs of PTBP1 and its possible multimerisation on its targets offer multiple modes of interaction (45). PTBP1 can cause internal exon skipping by binding to the pre-mRNA on either side of the regulated exon, creating a loop that impairs spliceosome recognition (46) PTBP1 also regulates tandem CPA usage, which corresponds to the situations where several consecutive CPAs are present on the same terminal exon (47). Here, we reveal a third type of splicing event that is controlled by PTBP1, where a terminal exon (exon γ), defined by a 3 SS and a CPA, is in competition with a downstream classical internal exon defined by a 5 SS and a 3 SS. We show that PTBP1-binding regions are present on either side of exon γ. Interactions between splicing factors involved in the recognition of the 3 SS and the CPA bound factors are central in the efficient processing of terminal exon (48). We therefore propose that by binding to both the 3 SS and the CPA, PTBP1 interferes with the ability of trans-acting factors to interact with their target sequences and therefore impedes their ability to define the γ exon.
The regulation of TP splicing, which is mediated by PTBP1, has strong consequences in HNSCC. We show by analysis of TCGA data that TPγ is associated with poor HNSCC patient outcome. This suggests that the different isoforms of p63 can have different functions in cancer cells. The genomic structure of the TP gene is evolutionarily conserved. It is also at least partly the case for the regulation of TP splicing, as we show here that PTBP1 similarly controls TPγ production in humans and Xenopus. However, only 15/37 (40.5%) amino acids of the γ-specific peptides are conserved between these two species, while the overall identity of Np63α proteins is 85%. This strong divergence makes it unlikely that the γ-specific peptide has any conserved functions by itself.
Rather, the putative specific functions of the p63γ isoform probably rely on the lack of the 231 amino acids that are present in the C-terminal region of p63α. p63 forms homotetramers or heterotetramers composed of various p63 or p73 proteins (49). We can hypothesize that the p63γ isoform, because it is devoid of the TA, SAM and TID domains, affects the activity of the p63 tetramer in a dominant-negative manner (50).
Even in patients classified as expressing "high" levels of TPγ, the proportion of that isoform relative to the other isoforms and especially TPα is only a few percent. How then can this low amount act as dominant negative? First, the ability of an alternative isoform to affect the function of a protein complex generally increases with the number of monomers composing the complex (51). Second, the moderate amount of TPγ RNA does not necessarily imply such a low p63γ protein level. Expression experiments showed that Np63γ is the most stable p63 isoform in Hep3B cells (50). This may be due to Np63γ being deprived of the Fbw7 E3 ligase PD and of some of the lysine residues (K494, K505) involved in ubiquitinylation and destabilization of p63α (12,52). Being able to quantify the γ isoform at the protein level would therefore be of high interest. Unfortunately, no γ-specific antibody is currently available. State-of-the-art mass spectroscopy may allow for isoform-specific quantification of protein in the near future and solve the important question of the amount of p63γ in cancer cells (53). In patients, an interesting alternative could be the detection of potential isoform-specific autoantibodies, as recently shown for another TP family member, TP, in patients with colorectal carcinoma (54).
As a likely dominant negative of a transcription complex, p63γ very probably shapes the transcriptome of cancer cells. This prediction will have to be tested by transcriptomic approaches following manipulations of p63γ levels in cultured cells. However, correlation analyses in TCGA datasets already revealed putative targets of p63γ, such as genes encoding component of the chromatin remodeling complex SWI/SNF, or even direct p63 target genes as the ones encoding components of the desmosomes. Mutations in nine different SWI/SNF genes are collectively found in 25% of all cancer types (35). These mutations are frequently postulated to be loss-of-function mutations (55). We observe that, in patients' tumors with a high proportion of TPγ isoform, six SWI/SNF genes belonging to the ones found mutated in HNSCC are repressed. This suggests that downregulation of SWI/SNF genes can be part of the disease process in patients with a high proportion of TPγ and a poor prognosis, with a potential for therapeutic opportunities (reviewed in ref. 56). The expression of most desmosomal genes is also reduced in primary tumors of patients with high TPγ. Finding that p63γ potentially represses the expression of desmosomal genes is in agreement with nonfunctional p63 variants where mutations are located mainly in the C-terminal region of p63 and are responsible for Ankyloblepharon Ectodermal Defects Cleft lip/palate associated with desmosomal gene repression (57). Loss of desmosomes is an indication of the loss of epithelial characteristics of these primary tumors. This is consistent with the impact of p63γ on promotion of a more mesenchymal state, as demonstrated in a breast cancer cell line (34) and in a three-dimensional organotypic model of invasion (17). EGFR inhibitors such as cetuximab are used in patients with late-stage HNSCC. They stimulate the reexpression of desmosomal genes (58). However, only a subset of patient with HNSCC respond to EGFR inhibitors and no specific feature can predict which ones (59). Together, these and our observations AACRJournals.org Cancer Res Commun; 2(12) December 2022 make it tempting to speculate that the TPγ isoform levels may allow for the definition of patients that could benefit from an early use of EGFR inhibitors.
While our work identified regulatory elements and factor controlling TPγ production, much work remains needed to fully address the transcriptional functions and physiopathologic consequences of this conserved C-terminally truncated isoform of TP63.