Variants Within TSC2 Exons 25 and 31 Are Very Unlikely to Cause Clinically Diagnosable Tuberous Sclerosis

ABSTRACT Inactivating mutations in TSC1 and TSC2 cause tuberous sclerosis complex (TSC). The 2012 international consensus meeting on TSC diagnosis and management agreed that the identification of a pathogenic TSC1 or TSC2 variant establishes a diagnosis of TSC, even in the absence of clinical signs. However, exons 25 and 31 of TSC2 are subject to alternative splicing. No variants causing clinically diagnosed TSC have been reported in these exons, raising the possibility that such variants would not cause TSC. We present truncating and in‐frame variants in exons 25 and 31 in three individuals unlikely to fulfil TSC diagnostic criteria and examine the importance of these exons in TSC using different approaches. Amino acid conservation analysis suggests significantly less conservation in these exons compared with the majority of TSC2 exons, and TSC2 expression data demonstrates that the majority of TSC2 transcripts lack exons 25 and/or 31 in many human adult tissues. In vitro assay of both exons shows that neither exon is essential for TSC complex function. Our evidence suggests that variants in TSC2 exons 25 or 31 are very unlikely to cause classical TSC, although a role for these exons in tissue/stage specific development cannot be excluded.


Introduction
Tuberous sclerosis complex (TSC) is an autosomal-dominant disease caused by inactivating mutations in either TSC1 (MIM #605284) or TSC2 (MIM #191092). TSC disease severity is variable with signs and symptoms ranging from hypomelanotic macules, to epilepsy, intellectual disability, autism, and multiple hamartomas in kidney, brain, heart, and lung. In TSC, about 70% of cases are due to new mutations [Sampson et al., 1989;Osborne et al., 1991;Au et al., 2007] and this presents a challenge for molecular diagnostics especially when the variant identified is not obviously disease causing. For example, some missense changes in TSC2 have been associated with TSC in patients diagnosed with definite TSC Hoogeveen-Westerveld et al., 2011], as well as cases in which symptoms are less severe and TSC is more often likely to be familial [Khare et al., 2001;O'Connor et al., 2003;Mayer et al., 2004;Jansen et al., 2006;Wentink et al., 2012].
Although there are well-defined clinical diagnostic criteria for TSC [Northrup et al., 2013], it can still be difficult to establish a clinical diagnosis of TSC, particularly in young patients who do not yet exhibit typical TSC lesions, but may have severe but nonspecific symptoms such as epilepsy, intellectual disability, and/or autism. In these cases, a molecular diagnosis can be helpful. Indeed, the identification of a clearly inactivating TSC1 or TSC2 mutation is considered sufficient evidence for a diagnosis of TSC, even in the absence of clinical signs [Northrup et al., 2013]. Increasingly, next-generation sequencing (NGS) of TSC1, TSC2, and many other genes simultane-ously (gene panels, whole-genome [WGS] or whole-exome [WES] sequencing) is being used in diagnostic settings. NGS is also applied to population genetics in large-scale sequencing studies, such as the NHLBI GO Exome Sequencing Project with data in the Exome Variant Server (EVS) and the Exome Aggregation Consortium (ExAC). In these studies, TSC is not assessed and the available phenotype data are very limited. This reliance on molecular diagnosis makes it even more important that the interpretation of any apparently pathological variant is correct.
The main recognized challenge to the molecular diagnosis of TSC has been (and still is) the "Variants of Uncertain Significance" (VUS), many of these being missense and in-frame indels. The TSC1 and TSC2 genes encode core components of the TSC protein complex that is a critical negative regulator of the mechanistic target of rapamycin (mTOR) complex 1 (TORC1) [Dibble and Manning, 2013]. In vitro assays to determine the effects of TSC1 and TSC2 missense and in-frame indels on TORC1 activity have proven useful in the ascertainment of the pathogenicity of variants previously reported as VUS [Hoogeveen-Westerveld et al., 2011Dunlop et al., 2011]. In these tests, assay results are summarized as the effect of the variants on the "TSC complex function" by which we mean "the inhibition of TORC1 activity." An additional option for evaluating a VUS is to perform conservation analysis of orthologous protein sequences in multiple species. Several computational algorithms (e.g., PolyPhen, SIFT) incorporate protein alignments from multiple species into the evaluation of the effects of amino acid substitutions on protein function [Ng and Henikoff, 2003;Adzhubei et al., 2013]. In our experience, these algorithms were not reliable enough to classify individual VUS with confidence [Hoogeveen-Westerveld et al., 2011]. However, this approach may still be applicable where high-quality alignments of multiple species are utilized. Conservation analysis has been utilized as a secondary line of evidence in the classification of VUS in the BRCA1/2 genes [Eggington et al., 2014].
The TSC1 and TSC2 Leiden Open Variation Databases (TSC LOVD) were made publicly available in 2006 and the TSC2 LOVD now displays 2,232 different variants (November 13, 2015), with a further 300 variants available on querying. Of these 2,532 different variants, about half are considered to be certainly or probably pathogenic. All 41 coding TSC2 exons, except exons 25 and 31, include both obviously truncating variants and fully confirmed missense variants that cause tuberous sclerosis. These two exons (25 and 31) have been shown to undergo alternative splicing in many tissues [Xiao et al., 1995;Xu et al., 1995;Olsson et al., 1996], although the extent and clinical significance of this is unclear. So far, we are not aware of confirmed reports of any pathological variants in these two exons in TSC cases. We have questioned the meaning of this observation over the 10 years of curating the TSC LOVD. To investigate the implications of this finding, we conducted function tests to assay expression constructs that lacked sequences corresponding to whole TSC2 exons, before the cases with truncating variants described below came to our attention.
As curators of the TSC LOVD, we are often asked to assist in classifying new TSC1 and TSC2 variants and we recently encountered a frameshift variant in TSC2 exon 25 and a nonsense variant in TSC2 exon 31, in individuals who had not been diagnosed with TSC and as described below (cases 1 and 2) were unlikely to have the disease. Finding these two stop codons triggered further investigations on the two exons and we gathered existing evidence that might help clarify the significance of exons 25 and 31 in TSC. During our investigations, we became aware of an individual with an in-frame deletion in TSC2 exon 31 (case 3 below). None of our three cases met the diagnostic criteria for TSC. Therefore, to gain insight into the sig-nificance of these variants in TSC, we reviewed all variants mapping to exons 25 and 31 in the TSC2 LOVD (http://www.lovd.nl/TSC2), performed conservation analysis, reviewed TSC2 mRNA expression data, and conducted in vitro functional analysis. We conclude that TSC2 exons 25 and 31 are not required for the TSC complexdependent inhibition of TORC1 and that variants in these exons are unlikely to cause TSC.
Nucleotide numbering corresponds to the TSC2 cDNA sequence (GenBank NM 000548.3, GI:116256351) with +1 as A of the ATG translation initiation codon in the reference sequence, and the initiation codon as codon 1.

Patients and Variants
Case 1: A predicted nonsense variant in TSC2 exon 25 (c.2859dup, p.K954Qfs * 6) was identified in an epilepsy gene panel test performed on DNA from a 10-year-old patient with epilepsy, but no clinical or radiographic findings supporting a diagnosis of TSC. The variant was reported as "pathogenic." Case 2: We analyzed rare (frequency <0.005) TSC1 and TSC2 variants identified in 1,805 individuals who had their whole exome sequenced as part of the UK10K project [ UK10K Consortium et al.,2015]. Study participants consented to be included as controls for diseases that were not directly under study. As part of the search for "incidental findings" that might cause serious inherited disease we received 67 rare variants to assess. These included a predicted nonsense variant in TSC2 exon 31 (c.3837C>G, p.Y1279 * ). The only accessible information on this individual is that they were recruited for a diabetes study. This variant is reported in the UK10K publication [ UK10K Consortium et al., 2015].
Case 3: We identified an African American individual with intractable neonatal onset epilepsy and profound developmental delay, and an in-frame deletion in TSC2 exon 31 (TSC2 c.3846 3855delinsG, p.S1282 G1285delinsR). This individual also carries the TSC2 c.4007C>T (p.S1336L, exon 33) variant that has a minor allele frequency of 0.137%-0.296% in the African population (rs148527903; http://evs.gs.washington.edu/EVS/ and http://exac.broadinstitute.org/). Both variants were reported as "possibly pathogenic." The individual has two café au lait spots, normal renal imaging, normal ophthalmology examination, no hypopigmented macules, and no brain MRI findings characteristic of TSC (no tubers, subependymal nodules, or areas of cortical dysplasia). A possible cardiac rhabdomyoma was detected on an echocardiogram obtained after the genetic test result. The most recent MRI at 4 years 9 months still shows no stigmata of TSC. TSC2 c.3846 3855delinsG was inherited from a parent (32 years old) who has also been seen at the TSC clinic. The parent has a single hypopigmented macule, but no obvious stigmata of TSC. Of note is a different TSC2 exon 31 variant (TSC2 c.3846 3854del) that is predicted to encode an identical protein (p.S1282 G1285delinsR). This variant was reported in a large-scale WES study of individuals with autism spectrum disorder and epilepsy [Schaaf et al., 2011]. HUMAN

Conservation Analysis and Variants in TSC2 Exons 25 and 31
To assess the conservation of TSC2 exons 25 and 31 and review the classification of variants within these two exons, a manual alignment of 163 orthologous vertebrate TSC2 protein sequences was constructed. For the comparison of conservation in exons 25 and 31 to other TSC2 exons, we performed a simple mathematical analysis and a more comprehensive statistical analysis using Scorecons [Valdar, 2002] and Wilcoxon Rank-Sum test [R Core Team, 2015].
The classification of all variants in TSC2 exons 25 and 31 reported to us (http://www.lovd.nl/TSC2) were reviewed.

Functional Analysis
To assess the functional importance of exons 25 and 31, TSC2 expression constructs with in-frame deletions of sequences corresponding to specific exons were derived from the original full-length TSC2 expression construct containing exon 25 but lacking exon 31 . Site-directed mutagenesis (SDM) was used to delete these sequences. Expression constructs lacking exons 3,4,5,6,9,12,19,22, and 41 served as controls for the expression construct lacking exon 25. To derive a TSC2 expression construct containing exon 31 (referred to as +ex31) an Eco47III fragment from a partial cDNA clone containing exon 31 was cloned into the original TSC2 expression construct. The p.R611Q (exon 16), p.S1282 G1285delinsR (exon 31), and p.S1336L (exon 33) variants were derived from the +ex31 expression construct by SDM. For each construct, the entire TSC2 open-reading frame was checked by sequencing to ensure that no additional nucleotide changes were introduced by the SDM procedure. A standard functional assay [Hoogeveen-Westerveld et al., 2011] was applied to determine the effects of the variants on TSC2 protein stability (estimated from the TSC2 protein signal), the TSC1-TSC2 protein interaction (estimated from the TSC1 signal), and on TORC1 activity (estimated from the ratio of T389-phosphorylated p70 S6 kinase (S6K): total S6K).

Conservation Analysis and Pathogenicity of Variants in TSC2 Exons 25 and 31
TSC2 exons 25 and 31 are conserved in vertebrates down to the lamprey (Supp. Fig. S1, A and B). The TSC2 LOVD contains 17 different variants (excluding contiguous exon deletions) mapping to TSC2 exon 25, or to the adjacent intronic sequences; 14 of these (five intronic, two silent, and seven missense variants) are classified as not pathogenic (not causing TSC) and three missense variants (p.K954R, p.S960F, and p.E984Q) are unclassified. One of the three unclassified missense variants (p.K954R) occurs naturally in 24/161 vertebrate species analyzed, including 11/83 mammals, whereas codons 960 (22/161) and 984 (19/161) in the alignment have different substituted amino acids to those found in our database (Supp. Fig. S1A).
The TSC2 LOVD lists 16 different variants (excluding contiguous exon deletions) mapping to exon 31, of which 15 (10 intronic, one silent, one in-frame indel, and three missense variants) are classified as not pathogenic. One intronic variant remains unclassified. Amino acids corresponding to the three missense variants (p.S1276F, p.S1282G, and p.V1291I) occur naturally in one or more of the vertebrates analyzed (Supp. Fig. S1B).
Our simple mathematical analysis showed that exon 25 has an average degree of conservation, whereas exon 31 was among the least conserved. Comparisons of exons 25 and 31 to two other comparatively more conserved exons (7 and 22) are shown (Supp. Fig. S2). A more comprehensive conservation analysis showed agreement with the simple analysis described above. There was a wide range of conservation across the 41 TSC2 exons and the general pattern was similar with or without fish sequences, apart from a few exceptions (Supp. Fig. S3A and B). The amino acids encoded by exon 31 were less conserved on average when compared with those encoded by exon 25 (Supp. Fig. S3A-D), and the statistical analysis showed that the distributions of conservation scores for exons 25 and 31 were significantly lower in value (P < 0.05; Wilcoxon Rank-Sum test) as compared with the distribution of scores in most of the TSC2 exons (Supp . Table S1).

Expression Analysis
The first full-length human TSC2 cDNA reported was from human fetal brain and included exon 25, but not exon 31 [European Chromosome 16 Tuberous Sclerosis Consortium, 1993]. Exon 31 was subsequently reported in rat and confirmed in human cDNA [Xu et al., 1995;Kimura et al., 2006]. We reviewed publications that describe the tissue distribution of TSC2 transcripts with and without exon 25 in rodents and in limited human material [Xiao et al., 1995;Xu et al., 1995;Olsson et al., 1996]. Although exon 25 and the first codon in exon 26 were often absent from TSC2 transcripts in rodents and humans, most tissues expressed some transcripts containing these sequences [Xiao et al., 1995;Xu et al., 1995;Olsson et al., 1996]. TSC2 transcripts lacking exon 25 (-ex25) were predominant in samples from adult human brain and adult human cerebellum [Xu et al., 1995], whereas human fetal muscle had more transcripts containing exon 25 (+ex25) [Xu et al., 1995]. A quantitative study [Olsson et al., 1996] estimated the ratio of +ex25:-ex25 in human fetal brain at 1:1; in three other fetal tissues, the -ex25 transcript was predominant. Human fetal kidney and fetal brain were found to have mainly the transcript lacking exon 31 [Xiao et al., 1995].
We searched two databases with TSC2 transcript information: The Ensembl Genome Browser at http://www.ensembl.org/ and the National Center for Biotechnology Information Gene database at http://www.ncbi.nlm.nih.gov/gene/7249. Six of the 26 different TSC2 transcripts listed in Ensembl were aligned using Clustal Omega to confirm the exact sequences that were absent from each transcript. The remaining 20 transcripts were not analyzed as these are too short to produce a functional protein. In the NCBI database (Homo sapiens annotation release 106), tracks of intron spanning TSC2 RNA-sequencing (RNA-seq) reads generated from alignments of RNA-seq data from the Bodymap2 dataset (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30611) of 16 adult tissue samples were analyzed. Each tissue in the Bodymap2 dataset is derived from a different single individual. To obtain evidence for tissue-specific alternative splicing of TSC2 exons 25 and 31, we examined tracks of intron spanning RNA-seq reads for these exons in each Bodymap2 tissue sample. Analysis of the BodyMap2 tracks showed evidence for the expected sequence joining each pair of adjacent exons in the majority of reads, except those involving exons 25 and 31 where a minority of reads (ß6% and ß11%, respectively) showed a join between adjacent exons, indicating that transcripts encoding exons 25 and/or 31 are relatively scarce. Exon 25 was just detectable in heart, lymph node, skeletal muscle, and thyroid tissue samples, but not detectable in adipose, adrenal, brain, breast, colon, kidney, liver, lung, ovary, prostate, testes, or white blood cells. Exon 31 was detectable in transcripts from adipose, adrenal, brain, breast, colon, kidney, lung, prostate, skeletal muscle, testes, thyroid, and white blood cells, but not in heart, liver, lymph node, or ovary.
A similar analysis was performed for 27 different tissues that are reported in a study of normal adult tissues taken from 95 different individuals , and are added to the TSC2 dataset at NCBI (Homo sapiens Annotation Release 107) [https://goo.gl/Gu0JWq]. Some of the tissues examined in the BodyMap2 dataset are not included in this study, for example, skeletal muscle, breast, and white blood cells. The same tissue type was analyzed from at least two individuals and exons 25/31 are also just detectable in some individuals and not evident at all in others. Aggregate reads show that the inclusion of exons 25 and 31 is 7.8% and ß23% of transcripts, respectively, supporting the abundance of the -ex25 and -ex31 transcripts in adult tissues. Since the tracks in both studies are from short reads, it was not possible to determine whether the observation of both exons, for example, in skeletal muscle and thyroid tissue, indicated that both exons are present in the same transcript in those tissues. No other TSC2 exons showed evidence for alternative splicing in these datasets at NCBI.
Analysis of a third RNA-seq dataset from 175 individuals in the genotype-tissue expression (GTEx) project [GTEx Consortium, 2015] also confirmed the relatively low abundance of TSC2 transcripts containing exons 25 and/or 31 in 51 different tissues from adults (Supp. Fig. S4; Supp. Tables S2 and S3).
In all the three RNA-seq datasets described above, both exons 25 and 31 are present in very low amounts in adult tissues and symptoms restricted to specific cell types cannot be excluded if there is some expression of abnormal +ex25 and/or +ex31 transcripts.

Effects of TSC2 Exon-Specific Deletions and Variants on the Canonical TSC Complex Function
To investigate the functional importance of exons 25 and 31 compared with other TSC2 exons, we derived a full-length TSC2 expression construct containing exon 31 (referred to as +ex31) and utilized existing data from a series of TSC2 expression constructs lacking exons 3, 4, 5, 6, 9, 12, 19, 22, 25, and 41 -Westerveld et al., 2013]) did not affect TSC complex function. The TSC1-TSC2 protein interaction is important for maintaining the activity of the TSC complex, and therefore for effective inhibition of TORC1. Reductions in the TSC1 signals indicate that the TSC1-TSC2 protein interaction is disrupted [Hoogeveen-Westerveld et al., 2011]. It was observed that deletion of exons corresponding to the TSC1 interaction domain of the TSC2 protein, or the introduction of a pathogenic amino acid substitution (TSC2 p.R611Q) in this domain, significantly reduced the TSC1 protein signal (Fig. 1B) and prevented the TSC complex-dependent inhibition of TORC1 activity, as assessed by the T389 phosphorylation status of a S6K reporter (Fig. 1C). In contrast, deletion of TSC2 exon 25 (delex25) or inclusion of exon 31 (+ex31) did not significantly affect TSC1 protein signals (Fig. 1B) or S6K-T389 phosphorylation (Fig. 1C), compared with the original wild-type TSC2 protein. Also, the presence of sequences corresponding to exon 31 did not rescue the activity of the p.R611Q variant. Therefore, we did not obtain evidence for a clear functional effect of exons 25 or 31 on TSC complex activity in our assay. Similarly, we did not obtain evidence that either the p.S1282 G1285delinsR (S1282delins) or p.S1336L variant affected TSC2 protein stability, TSC1-TSC2 protein interaction, or TSC complex-dependent inhibition of TORC1 activity (Fig. 1A-C). Deletion of exon 25 (delex25) did result in a nonsignificant reduction in TSC1 protein signals, a nonsignificant increase in S6K-T389 phosphorylation and a significant decrease in the total S6K signal. The low S6K signal suggests that the transfection efficiency for the delex25 variant was reduced compared with the other variants tested, and it is possible that the deletion of exon 25 sequences might have a subtle effect on TSC complex activity. Nonetheless, compared with the pathogenic R611Q variant, and the negative control transfection (TSC1/S6K only), the delex25 variant was still able to stabilize the TSC1 protein and inhibit TORC1.

Discussion
Clinical diagnosis of TSC can be problematic due to phenotypic variability, even between affected individuals from the same family who have the same TSC-causing variant. Therefore, adding genetic criteria [Northrup et al., 2013] should make TSC diagnosis less ambiguous. However, the increasing application of NGS-based gene panels, WES and WGS in diagnostic settings, as well as the increasing amount of data from large-scale sequencing projects that include subjects who do not have a definite diagnosis of TSC, need to be matched with correct variant interpretation. Two of our cases, neither of whom had a clinical diagnosis of TSC, had stop variants in exon 25 (TSC2 c.2859dup, p.K954Qfs * 6) and exon 31 (TSC2 c.3837C>G, p.Y1279 * ), respectively, that would necessitate a diagnosis of TSC according to the updated TSC diagnostic criteria [Northrup et al., 2013]. This heightened our awareness to the downside of automatically associating stop codons in the alternatively spliced TSC2 exons 25 and 31 with TSC.
Defining pathological as "causing TSC," the only two exons in TSC2 in which no pathogenic variants have been identified to date are exons 25 and 31 (www.lovd.nl/TSC2). Therefore, recent reports of "pathogenic" variants in these alternatively spliced exons led us to review different strands of evidence to try and ascertain the functional and clinical importance of these exons and of the putative "pathogenic" variants reported within them.
First, we examined amino acid sequences in exons 25 and 31 in multiple species to assess the conservation of these exons that may shed some light on their relevance to TSC. Although our simple and statistical analyses suggest that the conservation in these exons is less compared with the majority of the TSC2 exons, the results still point to some importance for these exons, possibly in early development and/or specific tissues.
Next, we investigated the extent to which these exons undergo alternative splicing. As evident from available data on RNA expression in adult human tissues, there are tissue-specific differences, but these support the abundance of the -ex25 and -ex31 transcripts HUMAN MUTATION, Vol. 37, No. 4, 364-370, 2016 Figure 1. Functional assessment. TSC2 expression constructs were derived from the original wild-type construct containing exon 25 but lacking exon 31 (TSC2), indicated in black. Nonitalicized "TSC2" refers to the protein and is used as such throughout this legend. The pathogenic TSC2 p.R611Q (R611Q) control, and the p.S1282_G1285delinsR (S1282delins) and p.S1336L variants identified in case 3 are indicated. In-frame exonspecific deletion constructs are referred to as delex3 (for deletion of exon 3), delex4, and so forth. Expression constructs encoding exon 31 are indicated in gray. TSC2 expression constructs were transfected into HEK 293T cells together with TSC1 and S6K expression constructs. The signals for TSC2, TSC1, total S6K (S6K), and T389-phosphorylated S6K (T389) were determined per variant, relative to the wild-type control (TSC2) in at least three independent transfection experiments by immunoblotting, as described previously [Hoogeveen-Westerveld et al., 2011]. Results for cells transfected with TSC1 and S6K only (TSC1/S6K) are shown for comparison. The mean TSC2 (A), TSC1 (B), and S6K (D) signals and mean T389/S6K ratio (C) are shown. In each case, the dotted line indicates the signal or ratio corresponding to wild-type TSC2 (+exon 25; −exon 31) (TSC2; = 1.0). Error bars represent the standard error of the mean; variants that showed significant differences from the wild-type control (TSC2; +exon 25; −exon 31) are indicated with an asterisk (P < 0.05; Student's t-test). A: Mean signals for the TSC2 variants, relative to wild-type TSC2 (+exon 25; −exon 31). Note that the TSC2 signal is significantly reduced by deletion of exons 3, 4, 5, or 22, or by the pathogenic R611Q substitution. Deletion of sequences corresponding to exon 41 (delex41) removes the epitope recognized by the antibody used for TSC2 detection. B: Mean TSC1 signals in the presence of the TSC2 variants relative to wild-type TSC2 (+exon 25; −exon 31). TSC1 signals were significantly reduced in the presence of the delex3, 4, 5, 6, 9, 12, 19, and 22 variants, and in the presence of the TSC2 R611Q substitution. C: Mean T389/S6K ratios in the presence of the TSC2 variants, relative to wild-type TSC2 (+exon 25; −exon 31). The T389/S6K ratios were significantly increased in the presence of the R611Q, delex3, 4, 5, 6, 9, 12, 19, 22 and 41 variants, as well as in the absence of TSC2 (TSC1/S6K).D: Mean total S6K signals in the presence of the TSC2 variants, relative to wild-type TSC2 (+exon 25; −exon 31).
(Supp. Figs. S4 and S5; Supp. Table S3). It was difficult to find data comparing the same human adult and fetal tissue types. In the mouse, the +ex25 transcript was more abundant in fetal and neonatal brain, whereas adult brain had an abundance of the -ex25 transcript [Xu et al., 1995;Olsson et al., 1996]. This agrees with tissue-specific expression data showing predominantly the -ex25 transcript in adult human brain [https://goo.gl/6V7PpS] and could point to some relevance in stage-specific development. More information on transcript and tissue-specific expression of TSC2 in the human fetus could help clarify the relative importance and specific functions of the different transcripts. Nonetheless, the evidence reviewed here leaves no doubt that TSC2 transcripts lacking exon 25 or 31 exist and suggest that they form the majority of TSC2 transcripts in a wide range of adult human tissues, including those, such as brain, important in TSC.
Finally, we investigated the influence of the amino acid sequences corresponding to exons 25 and 31 on TSC complex function. We used a previously described functional test that assays the effect of the TSC complex on TORC1 activity [Hoogeveen-Westerveld et al., 2011. According to this assay, the amino acids encoded by TSC2 exons 25 and 31 are not required for TSC complex function. Furthermore, we did not obtain any evidence that the p.S1282 G1285delinsR or p.S1336L variants significantly affected TSC complex function. Also, none of the tested variants that map to exon 25 affect TSC complex function [Hoogeveen-Westerveld et al., 2011.
Our analysis indicates that variants within TSC2 exon 25 or 31 do not inactivate the TSC complex and are unlikely to cause TSC. Coincidentally, just prior to submission of this manuscript, a batch of data submitted to the TSC databases was found to contain the same exon 31 variant (TSC2 c.3846 3855delinsG) as in case 3. This in-frame deletion co-occurs with a definite TSC-causing variant in TSC2 exon 5 (TSC2 c.569dup, p.Y190 * ) in an infant who fulfils the diagnostic criteria for TSC (cortical tubers, SEGA, subependymal nodules, hypomelanotic macules, renal angiomyolipoma, epilepsy). The exon 5 and exon 31 variants (TSC2 c.569dup and TSC2 c.3846 3855delinsG) are both inherited from a parent who also has clinical TSC (hypomelanotic macules, facial angiofibromas, cortical tubers, epilepsy, intellectual disability, and renal angiomyolipomas). This new case is consistent with our assertion that TSC2 c.3846 3855delinsG on its own is not the cause of classical TSC. Nonetheless, it is possible that variants mapping to TSC2 exon 25 or 31 might have a significant role in disease. For example, a role for the TSC2 c.2859dup (p.K954Qfs * 6) and c.3846 3855delinsG (p.S1282 G1285delinsR) variants in epilepsy cannot be excluded. The TSC2 c.3846 3855delinsG variant has been identified in several other individuals with infantile epilepsy who are without clinical features of TSC (Panzer, 2015, unpublished data). Exons 25 and 31 encode consensus sites of AKT and AMPK phosphorylation, respectively (http://scansite3.mit.edu) [Yaffe et al., 2001]. The TSC2 protein is inactivated by AKT phosphorylation [Manning et al., 2002] and activated by AMPK phosphorylation [Inoki et al., 2006]. It is possible that these exons alter the activity or response to stimuli of the TSC complex, particularly during fetal development. More sensitive assays of TSC complex activity are required to compare the activities of the different TSC2 protein splice isoforms and variants in more detail.
The findings reported here are specific to TSC2 exons 25 and 31, as stop variants that definitely cause TSC have been identified in all other TSC2 exons. Additional evidence that variants within exon 31 do not cause TSC comes from reports of three truncating variants (two frameshift and one nonsense) in the EVS and ExAC population databases (Supp . Table S4). Although phenotype information is not available, the population frequencies are quite high (1:532-1:1,443 individuals), which supports our conclusion that these truncating variants do not cause TSC. The exon 25 and 31 cases described here show the need to exercise caution when assigning pathogenicity, especially if the variant found has never been shown to cause TSC in a patient with a firm clinical diagnosis. This caution is probably reasonable in many other diseases.

Acknowledgments
We wish to thank Johan den Dunnen, Peter E.M. Taschner, and Ivo Fokkema (LOVD Team at Leiden) for hosting and maintaining the TSC1 and TSC2 variation databases. Our thanks also goes to Terence Murphy (NCBI) for help with the RNA-sequence track reads, to Matthew Hurles (Wellcome Trust Sanger Institute) for help with accessible information on the UK10K study, and to Elizabeth Shephard (UCL) and Christine Orengo (UCL) for useful discussions. This study makes use of data generated by the UK10K Consortium. A full list of the investigators who contributed to the generation of the data is available from http://www.UK10K.org.

Patient consent
Written informed consent was obtained for genetic testing and the case reports comply with the regulations of the respective local research ethics committees.