TSC patient clinical and genetic characteristics

We obtained fresh frozen surgical resection frontal/posterior parietal cortical or temporal lobe tuber tissue from three male and three female TSC patients ranging from 7 to 20 years of age with a median age of 9.5 years (Fig. 1a, Tables 1 and 6). All TSC tuber patients had epilepsy and three also had autism (Table 1). Genetic analysis showed that five tuber patients had germline pathogenic TSC2 variants and one patient had a TSC2 variant of unknown significance (Table 2). The identified germline variants included three small deletions of 4–18 bp, two missense variants, and one large deletion encompassing the entire genomic extent of TSC2 (exons 1–41).

Fig. 1

Overview of the experimental conditions and pipeline. a Schematic of the TSC patient tissue and control sample sizes for tubers (upper panel) and SEGAs (lower panel). Control samples that were specific to tuber and SEGA analyses are shown in red. b Schematic of the proteomics and phosphoproteomics methodology and analysis pipeline. PM, postmortem controls, TLE, temporal lobe epilepsy controls. PM control samples were age-matched to either tuber or SEGA tissue

Table 1 Clinical information on tuber specimen TSC patientsTable 2 Genetic information on tuber specimen TSC patients

We obtained fresh frozen surgical resection SEGA tissue from one male and four female TSC patients ranging from 11–25 years of age with a median age of 18 years old (Fig. 1a, Tables 3 and 6). Three of the TSC SEGA patients were reported to have epilepsy (Table 3). Genetic analysis was performed for four of the SEGA patients and revealed that all had germline pathogenic TSC2 variants (Table 4). The TSC2 variants included two small (1 bp) deletions, one small (2 bp) duplication, and one missense variant. Loss of heterozygosity (LOH) for TSC2 was identified in 2 of 4 SEGAs (50%) (Table 4).

Table 3 Clinical information on SEGA specimen TSC patientsTable 4 Genetic information on SEGA specimen TSC patients

As controls we used fresh frozen temporal lobe surgical resection tissue from two male and four female temporal lobe epilepsy (TLE) patients ranging from 29–58 years of age, with a median age of 41.5 years old (Fig. 1a, Tables 5 and 6). In addition, as age-matched controls we used Brodmann area 10 postmortem tissue (PM) cortical tissue from five male and three females who died of non-neurological causes aged between 3–27 years, with a median age of 13 years old (Fig. 1a, Tables 5 and 6).

Table 5 Clinical information on PM and TLE controlsTable 6 Comparison of ages of tuber patients, SEGA patients and controlsProteomics indicates mitochondrial respiration processes are decreased in tubers

To examine the proteomic and phosphoproteomic landscape in tuber tissue, tuber lysates were digested with trypsin, reduced and alkylated followed by TMT labelling. Labelled samples were pooled together and subjected to phosphopeptide-enrichment using sequential metal oxide affinity chromatography (SMOAC) prior to liquid chromatography tandem mass spectrometry (Fig. 1b) [53, 71]. Flow-through material from SMOAC was used for the generation of proteome data (Fig. 1b). Proteomic analysis detected 6,058 proteins while phosphoproteomics detected 17,475 phosphopeptides (Fig. 2a). Principle component analysis (PCA) of proteins and phosphopeptides showed that four of the tuber samples were well separated from the TLE and PM controls, while two tuber samples (A and B) were separated from the other tuber samples, and clustered with the controls by proteomic analysis (Supplemental Fig. 1a) and the TLE controls by phosphoproteomic analysis (Supplemental Fig. 1b). Using a statistical cutoff of p < 0.005, the expression level of 1,073 proteins were significantly different between tuber and PM tissue, while 850 proteins were significantly different between tuber and TLE tissue (Fig. 2a-c; Supplemental Table S1). The expression of 307 proteins was significantly different between tuber tissue and both PM and TLE tissue (Fig. 2a).

Fig. 2

Proteomics and phosphoproteomics of tubers detects reduced TSC1/2 expression but no mTORC1 activation. a Summary of the number of proteins and phosphopeptides detected and the number of significantly different proteins and phosphopeptides. b, c Volcano plots of tuber (TUB) versus PM control (b) and TLE control (c) proteomic data. Blue and purple dots represent proteins with significantly altered levels in tubers versus controls. TSC1 and TSC2 are labelled in red. d, e TSC1 (d) and TSC2 (e) protein expression levels are slightly reduced in tubers. f, g Volcano plots of TUB versus PM control (f) and TLE control (g) phosphoproteomic data. Blue and purple dots represent phosphopeptides with significantly altered levels in tubers versus controls. 4E-BP1/2 (EIF4EBP1/2) and RPS6 phosphopeptides are labelled in red. h, i 4E-BP1 (h) and 4E-BP2 (i) protein expression levels are unchanged in tubers. j-l Phospho-4E-BP1 T70 and phospho-4E-BP2 T45 and T70 peptide expression levels are unchanged or slightly increased in tubers. m RPS6 protein expression levels are unchanged in tubers. n-p Phospho-RPS6 S235, S236 and S240 peptide expression levels are unchanged or decreased in tubers. n = 5 for PM, n = 6 for TLE, n = 6 for TUB. Data are represented as mean ± SEM and were analysed using one-way ANOVA. ns not significant, * p < 0.05, **p < 0.01, ***p < 0.001****p < 0.0001

The proteomic data showed that there was a small but significant decrease in TSC1 expression (6%) between TLE and tuber tissue and a greater decrease in TSC2 expression (17–18%) between tuber and both PM and TLE controls (Fig. 2d, e). We also used the proteomic ruler method, which uses the mass spectrometry signal from histones that is proportional to the amount of DNA in the sample, to quantify absolute TSC1 and TSC2 copy number per cell [97]. The protein ruler method showed that TSC1 and TSC2 had similar copy numbers and that the copy number of both proteins was significantly decreased in tubers compared to TLE tissue (23% reduction in TSC1 copy number in tubers versus TLE and 34% reduction in TSC2 copy number in tubers versus TLE; Supplemental Fig. S1c, d).

To understand the differences in the protein expression profile of tuber tissue we performed gene ontology (GO) analysis using STRING [87, 88]. From the proteomic data, proteins with significantly increased expression in tuber tissue showed enrichment for the biological processes GO term ‘response to cytokine’, but no molecular function or cellular component terms were significantly enriched (Supplemental Fig. S2a). Proteins with significantly decreased expression were dominated by mitochondria-related GO terms including oxidative phosphorylation, aerobic respiration and the electron transport chain (Supplemental Fig. S2b-d).

Phosphoproteomics indicates changes in cytoskeleton organisation and neuronal function in tuber tissue

Phosphoproteomic analysis of tuber tissue showed that the level of 7,533 phosphopeptides were significantly different between tuber and PM tissue, while 1,760 phosphopeptides were significantly different between tuber and TLE tissue (Fig. 2a, f, g; Supplemental Table S2). 912 phosphopeptides were significantly different between tuber tissue and both PM and TLE tissue (Fig. 2a). Proteins with significantly increased phosphorylation in tuber tissue showed enrichment for GO terms involving cytoskeleton organisation and actin filament-based processes (Supplemental Fig. S2 e-g). Proteins with significantly decreased phosphorylation showed enrichment for GO terms involving neuronal functions including neurotransmitter transport, synaptic vesicle cycle and synaptic vesicle membrane (Supplemental Fig. S2h, i). These data suggest that cytoskeleton organisation processes are increased, while oxidative metabolism and neuronal processes are reduced in tuber tissue.

Phosphoproteomics shows no mTORC1 activation in tuber tissue

It was surprising that the GO analysis of the tuber tissue proteomics or phosphoproteomic data did not identify processes or functions relating to the mTOR pathway. Therefore, we next focused on RPS6 and 4E-BP, established targets of mTORC1, whose phosphorylation should be increased in cells that lack TSC1 or TSC2 as a result of activated mTORC1 signalling. Total protein expression levels of 4E-BP1 and 4E-BP2 were unchanged in tubers compared to controls (Fig. 2h, i). Neither 4E-BP1 nor 4E-BP2 phosphorylation was significantly increased in the overall dataset (Fig. 2f, g). When compared directly, there was a small but significant increase in 4E-BP1 T70 phosphorylation in tubers compared to TLE tissue and in 4E-BP2 T45 phosphorylation compared to PM tissue, but no change in 4E-BP2 T70 phosphorylation (Fig. 2j-l). Total protein expression levels of RPS6 were unchanged in tubers compared to controls (Fig. 2m). RPS6 phosphorylation was not significantly increased in the overall dataset (Fig. 2f, g). When compared directly, RPS6 S235 and S236 phosphorylation were significantly decreased in tuber and TLE tissue compared to PM control tissue and RPS6 S240 phosphorylation was unchanged compared to both controls (Fig. 2n-p).

We next compared the tuber phosphoproteomic data with a recently described list of 57 direct mTORC1 substrates consisting of 140 phosphosites (Supplemental Table S3) [6]. Of the phosphosites detected, none were significantly increased in tuber tissue compared to both PM and TLE controls (Supplemental Table S3). Thus, consistent with the small decrease in TSC2 expression, using phosphoproteomics we were unable to detect evidence of mTOR pathway activation in TSC patient tuber tissue.

TSC1 and TSC2 expression are strongly reduced in SEGA tissue

Tubers and SEGAs have distinct anatomical locations and pathology. To understand the proteomic and phosphoproteomic landscape of SEGAs, we used the same experimental pipeline as tuber tissue (Fig. 1b), but samples were digested separately with trypsin and chymotrypsin to increase the depth of coverage of peptides lacking the consensus sequence for trypsin. Proteomic analysis using trypsin detected 6,304 proteins (Fig. 3a, Supplemental Tables S4, S5). PCA of the trypsin proteomic data showed that the SEGAs were well separated from the PM and TLE controls (Fig. 3b).

Fig. 3

TSC1 and TSC2 expression are strongly reduced in SEGA tissue. a Summary of the number of proteins detected and the number of significantly different proteins in the trypsin and chymotrypsin conditions. b PCA plot of postmortem (PM), temporal lobe epilepsy (TLE) and SEGA proteomic data in the trypsin condition. c, d Volcano plots of SEGA versus PM (c) and TLE (d) trypsin proteomic data. Blue and purple dots represent proteins with significantly altered levels in SEGAs versus controls. TSC1, TSC2, GPNMB, ANXA1, S100A11, NPTX1, HLA and ITG proteins are labelled in red. e, f TSC1 (e) and TSC2 (f) protein expression levels in the trypsin condition are strongly decreased. g, h 4E-BP1 (g) and RPS6 (h) protein expression levels in the trypsin condition are increased. i-k ANXA1 (i), GPNMB (j) and S100A11 (k) protein expression levels in the trypsin condition are strongly increased. l NPTX1 protein expression levels in the trypsin condition are strongly decreased. n = 6 for PM, n = 5 for TLE, n = 5 for SEGA. Data are represented as mean ± SEM and were analysed using one-way ANOVA. ns not significant, * p < 0.05, **p < 0.01, ***p < 0.001****p < 0.0001

The expression of TSC1 and TSC2 in SEGA tissue were strongly decreased. With trypsin, TSC1 expression was decreased by 45% and TSC2 expression was decreased by 55% in SEGAs compared to PM controls (Fig. 3c-f). The protein ruler method also showed a strong decrease in TSC1 and TSC2 protein copy number in SEGAs compared to PM and TLE controls (Supplemental Fig. S1e, f). Moreover, the total protein levels of mTORC1 targets 4E-BP2 and RPS6 were significantly increased by 30% and 110% respectively, compared to the PM control, in SEGA tissue (Fig. 3g, h).

Proteomic analysis with chymotrypsin showed similar results to trypsin. With chymotrypsin we detected 4,501 proteins, which were well separated from PM and TLE controls by PCA (Fig. 3a, Supplemental Fig. S3a). TSC1 expression was decreased by 30% and TSC2 expression was decreased by 84% in SEGAs compared to PM controls with chymotrypsin (Supplemental Fig. S3b, c).

Transcriptomic analysis previously identified 50 genes with significantly altered expression in SEGA tissue including ANXA1, GPNMB, and S100A11 and these proteins were also shown to be increased in SEGAs by immunostaining [91]. In our total proteome analysis, ANXA1, GPNMB, and S100A11 were three of the most significantly increased proteins in SEGA tissue (Fig. 3c, d, i-k; Supplemental Table S4). This study also found decreased expression of genes involved in nervous system development, including NPTX1. NPTX1 was one of the proteins with the most significantly decreased expression in our SEGA tissue (Fig. 3c, d, l; Supplemental Table S4).

Together, these proteomic data show that SEGAs exhibit changes in protein expression consistent with previous studies and show a much greater decrease in TSC1 and TSC2 expression compared to tuber tissue.

Proteomic analysis shows SEGA tissue exhibits increased ribosomal biogenesis and an inflammatory response

GO analysis of proteins with significantly increased expression in SEGA tissue compared to both PM and TLE controls, combined from the trypsin and chymotrypsin datasets, showed the greatest enrichment for GO terms around cytoplasmic translation and the ribosome (Fig. 4a-c). In fact, 33 large ribosomal subunits and 27 small ribosomal subunits had increased expression in SEGA tissue, including RPL37 and RPS27L (Fig. 4d, e; Supplemental Tables S4, S5), consistent with the role of mTORC1 in promoting protein translation. GO biological process analysis also showed enrichment for terms antigen processing and presentation (Fig. 4a) and the expression of 12 MHC human leukocyte antigen (HLA) proteins was increased in SEGAs, including HLA-G and HLA-DRB5 that were increased more than tenfold (Fig. 3d, e; Fig. 4f-h, Supplemental Tables S4, S5). These data indicate that, in addition to promoting protein synthesis, SEGA tissue exhibits a neuroinflammatory response.

Fig. 4

Proteomics of SEGA tissue identities increased ribosomal protein expression and a neuroinflammatory response. a-c GO analysis of biological process (a), molecular function (b) and cellular component (c) of proteins with significantly increased expression in SEGAs shows strong enrichment for inflammatory processes. d, e RPL37 and RPS27L expression in the trypsin condition are increased in SEGAs. f Network nodes representation of MHC protein complex (GO: 0042611) proteins with significantly increased expression in SEGAs. g, h HLA-G and HLA-DRB5 expression in the trypsin condition are increased in SEGAs. n = 6 for PM, n = 5 for TLE, n = 5 for SEGA. Data are represented as mean ± SEM and were analysed using one-way ANOVA. ns not significant, * p < 0.05, **p < 0.01, ***p < 0.001****p < 0.0001

Proteins with decreased expression in SEGA tissue were highly enriched for GO classes including chemical synaptic transmission, presynapse, postsynapse and somatodendritic compartment (Supplemental Fig. S3d-f). The enrichment of neuron-related GO terms in the proteins with decreased expression potentially reflects the unique cellular composition of SEGA tissue, compared to the neuron-rich temporal lobe TLE and cortical PM tissue controls.

Phosphoproteomics detects activation of mTORC1 signalling in SEGA tissue

Phosphoproteomic analysis of trypsin and chymotrypsin digested tissue was used to analyse changes in protein phosphorylation in SEGAs. Phosphoproteomic analysis using trypsin detected 24,861 phosphopeptides and 12,248 phosphopeptides with chymotrypsin (Fig. 5a; Supplemental Tables S6, S7). PCA of the phosphoproteomic data showed that the SEGA tissue samples were very well separated from the PM and TLE controls (Fig. 5b, Supplemental Fig. S3g).

Fig. 5

mTORC1 signalling is activated in SEGAs. a Summary of the number of phosphopeptides detected and the number of significantly different proteins in the trypsin and chymotrypsin experiments. b PCA plot of postmortem (PM), temporal lobe epilepsy (TLE) and tuber (SEGA) phosphoproteomic data in the trypsin condition. c, d Volcano plots of SEGA versus PM (c) and TLE (d) trypsin phosphoproteomic data. Blue and purple dots represent phosphopeptides with significantly altered expression in SEGAs versus controls. 4E-BP1/2 (EIF4EBP1/2), RPS6, STAT3 and HNRNPU phosphopeptides are labelled in red. e, f Phospho-4E-BP1 S65 (e) and phospho-4E-BP2 S65 (f) phosphopeptide expression levels in the trypsin condition are increased in SEGAs. g-j Phospho-RPS6 S235 (g), S244 (h), S236* (i) and S240* (j) phosphopeptide expression levels in the trypsin condition are increased in SEGAs. Asterisks indicate peptides with 1 additional phosphorylated residue. k Phospho-STAT3 Y705 phosphopeptide expression levels in the trypsin condition are increased in SEGAs. n = 6 for PM, n = 5 for TLE, n = 5 for SEGA. Data are represented as mean ± SEM and were analysed using one-way ANOVA. ns not significant, * p < 0.05, **p < 0.01, ***p < 0.001****p < 0.0001

In contrast to tuber tissue, the phosphorylation of established targets of mTORC1 was significantly increased in SEGA tissue compared to controls in the overall datasets (Fig. 5c, d). When compared directly, phosphorylation of 4E-BP1 at T37, T46, S65, T70 and 4E-BP2 at S65 were all significantly increased in SEGAs (Fig. 5e, f; Supplemental Tables S6, S7). RPS6 phosphorylation was significantly increased at S235 and S244 (Fig. 5c, d, g, h; Supplemental Table S6), and also at S236 and S240 but only in peptides that were phosphorylated at one additional serine residue (Fig. 5c, d, i, j; Supplemental Table S6), suggesting priming phosphorylation was required for phosphorylation at these residues. Moreover, phosphorylation of STAT3 Y705, which is strongly increased in giant cells of SEGAs [13], was increased in SEGAs on average ninefold compared to PM controls (Fig. 5c, d, k; Supplemental Table S6).

To further validate the SEGA phosphoproteomic data, we combined the trypsin and chymotrypsin phosphoproteomic datasets of significantly increased phosphopeptides compared to controls, and compared this combined SEGA phosphosite dataset with 57 previously reported direct mTORC1 substrates [6]. 40 of 57 (70%) of phosphosites, corresponding to 20 proteins, were significantly increased in the SEGA dataset with a median fold change of 3.06 (Supplemental Table S8). Together with the significant increase in P-4E-BP1/2 and P-RPS6 levels, these data provide strong evidence that phosphoproteomics detects activation of mTORC1 signalling in TSC patient SEGA tissue.

We therefore used log FC ≥ 1 (and p < 0.005) as a cutoff to identify all potential mTORC1 targets in the SEGA phosphosite dataset. Since some previously reported direct mTORC1 substrate phosphosites were significantly increased compared to either the PM or TLE control but not both (Supplemental Table S8), we included phosphosites that were significantly increased (p < 0.005) compared to either control with an FC ≥ 1. Using these criteria, there were 6060 phosphosites within 2154 proteins that were increased in SEGA tissue (p < 0.005, logFC ≥ 1, Supplemental Tables S9 and S10). mTOR has a strong bias towards serine as the phosphoacceptor residue (77% serine, 23% threonine), and a strong bias towards proline at the + 1 position with glutamic acid, phenylalanine, tyrosine, and glutamine also favoured somewhat at + 1 [6, 36]. Analysis of SEGA phosphosites showed that serine was the dominant phosphoacceptor residue (85%), with proline most common (29%) at the + 1 position and some enrichment for glutamic acid, leucine, serine and aspartic acid (Fig. 6a). These data suggest that many of the proteins whose phosphorylation is increased in SEGAs are direct mTORC1 substrates.

Fig. 6

Phosphorylation of proteins regulating RNA metabolic processes is increased in SEGAs. a Residue enrichment motif analysis of the significantly increased SEGA phosphopeptide dataset phosphosites in the trypsin condition, the chymotrypsin condition and both conditions combined. b-d GO analysis of biological process (b), molecular function (c) and cellular component (d) of proteins represented by phosphopeptides with significantly increased expression in SEGAs shows strong enrichment for RNA-related processes. e Reactome pathways of proteins represented by phosphopeptides with significantly increased expression in SEGAs. f Network nodes representation of RNA splicing (GO:0008380) proteins represented by phosphopeptides with significantly increased expression in SEGAs, HNRNP subcluster. g Schematic of the role of hnRNPs in mRNA splicing and translation. h–k phospho-HNRNPU S26 (h), phospho-HNRNPL S52 (i), phospho-HNRNPK S353 (J) and phospho-HNRNPC S247 (k) phosphopeptide levels in either the trypsin or chymotrypsin condition are increased in SEGAs. n = 6 for PM, n = 5 for TLE, n = 5 for SEGA. Data are represented as mean ± SEM and were analysed using one-way ANOVA. ns not significant, * p < 0.05, **p < 0.01, ***p < 0.001****p < 0.0001

Phosphorylation of RNA metabolism proteins is increased in SEGAs

GO analysis of the significantly increased SEGA phosphopeptide dataset (Supplemental Table S10) showed robust enrichment for nucleic acid metabolic process/nucleic acid binding biological processes and molecular functions as well as nucleus-related cellular components (Fig. 6b-d). In particular, RNA metabolic processes, RNA processing and RNA binding GO terms were very highly enriched (Fig. 6b-d). Moreover, Reactome pathways analysis showed enrichment for processing of capped intron-containing pre-mRNA, metabolism of RNA and mRNA splicing (Fig. 6e).

Many phosphopeptides increased in SEGAs were from proteins involved in mRNA splicing, including 132 of the 208 proteins annotated to GO term ‘RNA splicing’ (GO:0008380) (Fig. 6f). mRNA splicing is carried out by the spliceosome, a nuclear-localised multi-megadalton ribonucleoprotein complex involving more than 120 proteins that assembles on pre-mRNA substrates and removes non-coding introns to generate mature mRNAs [96]. Phosphorylation of U2-related spliceosome components (PUF60, SF3B1, SFPQ), pre-mRNA processing factor proteins (e.g. PRPF38A, PRPF38B, PRPF40A), SR-related proteins (SRRM1, SRFBP1, SRPK2) and exon junction complex proteins (ACIN1, RBM8A, CASC3) were all increased in SEGAs (Supplemental Tables S9, S10). Phosphorylation of ACIN1 was increased at residues S240 and S243, which are directly phosphorylated by mTORC1 [6, 76]. Moreover, the phosphorylation of 16 heterogeneous nuclear ribonucleoproteins (hnRNPs) was increased in SEGAs (Fig. 6f, h–k; Supplemental Tables S9, S10). HnRNPs are involved in regulating the maturation of newly formed heterogeneous nuclear RNAs (hnRNAs/pre-mRNAs) into messenger mRNAs, stabilising mRNAs during cellular transport and controlling their translation (Fig. 6g) [24]. Remarkably, the activity of five hnRNPs whose phosphorylation was increased in SEGA tissue has been shown to be regulated through phosphorylation at the same residues (Fig. 5c, d; Fig. 6h-k; Table 7). These data strongly indicate that mTORC1 activation in SEGAs causes concerted changes in the regulation of key proteins involved in mRNA splicing.

Table 7 HnRNPs with functionally validated phosphosites that have increased expression in SEGAs. a Identified in trypsin experiment. b Identified in chymotrypsin experimentEvidence of widespread changes to splicing in SEGAs

Based on the increased phosphorylation of proteins involved in RNA-metabolism and mRNA splicing we hypothesised that splicing of mRNAs would be affected in SEGA tissue. Unfortunately, the SEGA tissue available to us for this study was depleted by our proteomic analyses, and RNA-sequencing (RNA-seq) could not be performed for these same samples. Therefore to test this hypothesis, we analysed previously published SEGA tissue RNA-seq data from 17, mostly paediatric, TSC patients and periventricular tissue from 8 (postmortem tissue) controls [11].

We analysed global changes in splicing at the level of isoforms, splicing events and exons. The isoform approach, using IsoformSwitchAnalyzeR [95], identified 3,139 significant isoform switches from 2,526 genes, showing that a substantial subset (~ 15%) of genes exhibit significant changes in isoform usage in SEGA tissue versus controls (Fig. 7a). The rMATS splicing analysis pipeline is an event-based approach that analyses differences in specific types of splice events including alternative 3’ or 5’ splice sites, mutually exclusive exons, retained introns and skipped exons [80]. rMATS analysis identified a total of 1,572 significant splicing events with a meaningful ΔPSI in SEGA tissue, highlighting both gains and losses in exon inclusion or splice site usage (Fig. 7b, c). Finally, at the exon level we identified 3,817 genes within which differential exon usage was detected, a total of 18,183 exons (Fig. 7d). Together, these analyses show clear evidence that SEGA tissue experiences widespread alternative splicing dysregulation, with many significant changes in exon inclusion/exclusion and other splice site usage.

Fig. 7

Evidence of large-scale alterations in mRNA splicing in SEGAs. a Volcano plot showing significantly different isoforms (blue dots) in SEGA tissue versus controls identified using IsoformSwitchAnalyzeR. b Volcano plot showing significantly different splicing events (blue dots) identified by rMATS in SEGA tissue versus controls. c Number of significant rMATS splicing events of each type in SEGA tissue versus control. d Volcano plot showing genes with significantly different exon usage (blue dots) in SEGAs versus controls. e Venn diagram showing the number of genes identified using three different methods and genes in common using each method. f GO analysis of genes identified using all three methods as having significantly altered splicing in SEGAs versus controls. g, h Splicing changes in HNRNPD (g) and HNRNPDL (h) in SEGAs

We next looked at the overlap between the genes identified using each method of splicing analysis. There was significant overlap between each method, with 1,667 genes identified by at least two approaches (Fig. 7e, Supplemental Table S11). Remarkably, biological process GO analysis of genes that had altered splicing in SEGAs validated by all three analysis methods was dominated by enrichment for GO terms around RNA processing and RNA splicing (Fig. 7f; Supplemental Table S12). Moreover, splicing of six hnRNPs was altered in SEGA tissue, including isoform switching in HNRNPD and HNRPNPDL (Fig.