Tuberous sclerosis complex (TSC) is a rare genetic disease causing multisystem growth of benign tumours and other hamartomatous lesions, which leads to diverse and debilitating clinical symptoms. Patients are born with TSC1 or TSC2 mutations, and somatic inactivation of wild-type alleles drives MTOR activation; however, second hits to TSC1/TSC2 are not always observed. Here, we present the genomic landscape of TSC hamartomas. We determine that TSC lesions contain a low somatic mutational burden relative to carcinomas, a subset feature large-scale chromosomal aberrations, and highly conserved molecular signatures for each type exist. Analysis of the molecular signatures coupled with computational approaches reveals unique aspects of cellular heterogeneity and cell origin. Using immune data sets, we identify significant neuroinflammation in TSC-associated brain tumours. Taken together, this molecular catalogue of TSC serves as a resource into the origin of these hamartomas and provides a framework that unifies genomic and transcriptomic dimensions for complex tumours.
Tuberous sclerosis complex (TSC) is a neurocutaneous, autosomal dominant genetic disease affecting ∼1 in 6,000 to 10,000 live births1,2,3,4. TSC causes highly variable, multisystem growth of benign tumours and other hamartomatous lesions that cause diverse clinical problems5. Abnormal brain growths are one of the most common features of TSC and lead to epilepsy, developmental delay, cognitive impairment, autism, behavioural problems and hydrocephalus. Most prevalent of these are cortical tubers, largely static malformations of the cerebral cortex that are present at birth and associated with seizure activity3,6. Approximately 80% of patients develop subependymal nodules (SENs) on the lateral ventricle walls, which can progress into subependymal giant cell astrocytomas (SEGAs), larger well-circumscribed tumours near the foramen of Monro. Several studies have suggested an association between brain lesions and neurological symptoms in TSC patients, underscoring the need to understand and reduce these growths to improve quality of life7,8,9,10.
Other major organs affected by TSC lesions are the skin, kidney, lung and heart. Skin lesions include hypomelanotic macules and facial angiofibromas that are important diagnostic features of TSC and affect nearly all TSC patients. Renal angiomyolipomas (RAs) affect more than 70% of patients and are typically benign lesions that can cause kidney dysfunction and require treatment if significantly large, abundant or susceptible to bleeding11. In fact, RAs are the most common cause of mortality in adult TSC patients12. Finally, heart tumours called cardiac rhabdomyomas (CRMs) are another major diagnostic feature of TSC, as they can be detected prenatally and are most common in infants.
While TSC may be inherited (familial), it is more often the result of de novo (sporadic) germline mutations in one of two tumour suppressor genes, TSC1 (encoding TSC1, also known as hamartin) and TSC2 (encoding TSC2 or tuberin)13,14,15. Purely heterozygous germline mutations as well as mosaic mutations have been identified in TSC patients16,17,18,19. Along with TBC1D7, TSC1 and TSC2 form a physical complex that supports the GTPase-activating protein (GAP) activity of TSC2 towards the small GTPase, RHEB, a direct and positive regulator of MTOR (specifically, MTOR complex I or MTORC1)20. MTORC1 integrates signals from growth factors, amino acids and energy to promote cell growth, division and survival. Accordingly, loss-of-function mutations in TSC1 or TSC2 lead to constitutive MTORC1 activation that is uncoupled from upstream signalling inputs. This molecular insight led to the evaluation of MTOR inhibitors in clinical trials and U.S. Food and Drug Administration (FDA) approval for therapeutic use in TSC patients21,22,23,24,25. Despite considerable promise, MTOR inhibitors are not universally effective across the TSC population, fail to maintain tumour reduction following cessation of treatment, and may be associated with undesirable side effects23,25,26. Therefore, a critical need remains to develop additional therapeutic options for TSC, including those that target tumour growth.
While TSC lesions may develop by somatic inactivation of TSC1/TSC2, second hits (mutations) are not always observed, especially in brain lesions, suggesting that additional mechanisms may contribute to their growth27,28,29,30,31,32. Moreover, the collective molecular changes underlying TSC tumour growth are unknown, yet essential to understanding disease aetiology and developing therapies. To address this, we implement a comprehensive genomics study to characterize the molecular landscape of TSC. We evaluate 111 TSC-associated tissues for TSC1/TSC2 status, DNA mutations, copy number aberrations, differential gene expression and DNA methylation patterns. We find that unlike a majority of RAs and SEN/SEGAs, only one-third of cortical tubers are driven by somatic TSC1/TSC2 inactivation, suggesting monoallelic mutation may be sufficient to cause cortical malformation. Further, we discover that most TSC lesions have a low somatic mutational burden, in contrast to malignant tumours. Instead, large arm-level chromosomal aberrations are found in tumours from a subset of patients (11%). We uncover conserved gene expression signatures for each lesion type, and use computational cell sorting to identify individual components of pleiotropic tumours. Moreover, we identify a substantial immune expression signature in TSC-associated brain tumours, particularly SEN/SEGAs, which is supported by immunohistochemistry. Taken together, this study provides a comprehensive genomic landscape of TSC, knowledge around cell-of-origin, and unifies the molecular signatures of these complex tumours.
TSC1 and TSC2 mutational spectrum
Genomic DNA and total RNA were isolated from 78 fresh-frozen TSC lesions, including 31 cortical tubers (TUB), 20 RAs, 20 SEN/SEGA (2 SEN and 18 SEGA, which represent a continuum of the same tumour), 5 CRMs and 2 skin lesions. In addition, 33 TSC-associated non-tumour tissues and 16 non-TSC (normal) brain and kidney tissues were included. As genetic material permitted, samples were assayed on the following platforms: whole-exome sequencing (WES), Illumina Infinium Omni2.5 single-nucleotide polymorphism (SNP) arrays, Illumina Infinium HumanMethylation450 (HM450) BeadArrays, targeted-deep TSC1/TSC2 sequencing and mRNA sequencing (RNAseq). Sample name prefixes correspond to patient identifiers and suffixes indicate individual tissue samples (for example, 01-RA1 denotes RA sample 1 from patient 01). All available sample information is presented in Supplementary Data 1.
Our first objective was to characterize the mutational spectrum of TSC1 and TSC2 (refs 2, 14). For this, we used WES to define point mutations (single-nucleotide variants or SNVs) and small insertions or deletions (INDELs), and high-resolution SNP arrays to identify large deletions and regions of copy-neutral loss-of-heterozygosity (CN-LOH). We also suspected that a small fraction of TSC1/TSC2 mutations may be missed by these two platforms including systemic mosaic or subclonal somatic mutations occurring at low allelic frequencies, medium-sized deletions beyond the capture of WES and absent from greater scale copy number segmentation based on SNP arrays, and mutations found in regions of poor coverage (for example, splicing mutations within introns). To address this, we implemented targeted-deep sequencing of the entire TSC1 and TSC2 loci—including upstream and downstream elements, introns and exons—to better detect such mutations. Collectively, we identified 57 unique DNA mutations (SNVs and INDELs) in TSC2 and eight in TSC1 from both normal and lesion tissues (Fig. 1a and Supplementary Data 2). Most mutations were observed in only a single individual with the exception of two TSC2 mutations, which were shared by two or more unrelated patients. Mutations were distributed across each locus with no enrichment in specific domains or hotspots33. Targeted sequencing identified point mutations in non-tumour tissue from three patients that were not detected by WES. These included a heterozygous germline splicing mutation at an exon–intron boundary in TSC1 (62-UG1), and two apparently mosaic mutations found at <5% allelic frequency (in patients 57 and 74). We also identified two low-frequency somatic TSC2 mutations in tumour tissues: a frameshift mutation in 18-RA1 and nonsense mutation in 06-RA1. The latter of these co-occurred with a somatic TSC2 mutation already identified by WES, suggesting independent second hits drove subclonal growth within the tumour. This approach also allowed fine-mapping of deletions first observed by SNP array, and enabled detection of intragenic deletions below the limits of detection by SNP array including a deletion spanning a single exon in patient 43 (Supplementary Fig. 2). Using SNP arrays, we found larger (>100 bp) deletions in TSC2 in eleven patients, ranging in size from 462 bp to 4.8 Mb, with a median size of 48.6 kb (Fig. 1b–d and Supplementary Data 2). Furthermore, we used SNP arrays to detect regions of CN-LOH, identified as areas in which B-allele frequencies (BAF) diverge from the heterozygous state while copy numbers (log-R ratios) remain stable (Fig. 1e,f and Supplementary Data 2). These affected TSC2 (chromosome 16p13) in 18 lesions, and TSC1 (chromosome 9q34) in 4. Last, we used HM450 arrays to assess genome-wide DNA methylation profiles with a focus on the promoters and gene bodies of TSC1/TSC2. We did not find evidence of epigenetic silencing of TSC1 or TSC2 in any tissue (Supplementary Fig. 1).
Taken together, we identified TSC1/TSC2 mutations in 64 of 66 (97%) patients (84.9% TSC2 and 12.1% TSC1), leaving two patients (3%) with no mutation identified (NMI), a smaller percentage than previous estimates based on conventional molecular testing15,33 (Fig. 1g). We found no mutations in other MTOR pathway genes in these two NMI patients, nor did we find any genes with both germline and somatic variants, supporting the hypothesis that a third TSC-causative locus (‘TSC3’) does not exist.
As TSC lesions are thought to arise by Knudson’s two-hit model of tumorigenesis, we next specifically investigated the germline and somatic origin of the TSC1/TSC2 mutations occurring in these tissues. For lesions lacking patient-matched normal samples, we made predictions for the somatic or germline origin of mutations (indicated by asterisks in Supplementary Data 2; see Methods) based primarily on allele frequencies. We discovered that roughly two-thirds of hamartomas from TSC1/TSC2 patients harboured two TSC hits, including most RAs and SEN/SEGAs, while second hits were found in only 35% of cortical tubers (Fig. 1h,i). Frameshift INDELs and splicing mutations rarely occurred somatically, despite representing over half of germline mutations. Instead, CN-LOH events, which arise from errors in mitotic recombination, were the most common type of second hit and nearly the exclusive somatic event in SEN/SEGAs (Fig. 1j,k). For both TSC1 and TSC2, lesions with single point mutations or point mutations in combination with CN-LOH were most common, although TSC2 lesions with two point mutations and combinations involving large deletions were also found, in contrast to TSC1 (Fig. 1l). Finally, we wanted to determine whether TSC1 and TSC2 expression was decreased in lesions with mutations predicted to decrease or truncate transcripts. Despite considerable heterogeneity, tumours with one or two truncating mutations in TSC2 showed reduced levels of TSC2 mRNA transcripts compared to non-TSC tissues (pair-wise Welch’s t-tests; FDR-adjusted P=0.01) (Fig. 1m). Similarly, tumours with two truncating TSC1 mutations showed a lower level of TSC1 mRNA compared to non-TSC tissue (pair-wise Welch’s t-tests; FDR-adjusted P=0.03) (Fig. 1n).
Coding mutational landscape of TSC tumours is quiet
In addition to TSC1/TSC2, we hypothesized that lesions may acquire mutations in other genes, including those that affect tumour growth. To test this, we profiled the coding genome of 42 lesions paired with normal samples using WES. We uncovered a median somatic mutation rate of 0.31 mutations per megabase (Mb) of DNA (range: 0.16–3.8 mutations per Mb), including silent and non-silent SNVs and small INDELs, with a median variant allelic fraction (VAF) of 0.13 (Fig. 2a and Supplementary Data 3). This mutation rate is substantially lower than almost all malignant tumour types, with the exception of acute myeloid leukaemia (AML) (Fig. 2b). We found that 10 of 42 (24%) tumours contained at least one somatic mutation in a candidate or high-confidence tumour driver gene34, although there was no enrichment in tumours lacking somatic TSC1/TSC2 inactivation (that is, tumours with less than two TSC1/TSC2 mutations) (Supplementary Data 3). Moreover, no specific mutations recurred across patients and only two genes were somatically mutated in more than one patient. Importantly, we also failed to find somatic mutations in any other MTOR pathway gene.
Subset of TSC tumours harbour large chromosomal aberrations
Given this low mutational burden, our next objective was to determine whether large chromosomal copy number aberrations (CNAs) exist that may play a role in tumour development. Aside from deletions and CN-LOH events involving TSC1/TSC2, we discovered that nine lesions from eight TSC patients harboured large (arm or whole chromosome level) CNAs at other chromosomal locations (Fig. 2c,d and Supplementary Datas 4 and 5). This included chromosome 1 and chromosome 12 CNAs in four tumours each, and chromosomes 5, 7, 11, 17 and 19 CNAs in two tumours each. The remaining CNAs were not shared across multiple tumours. These CNAs were found in each of the major lesion types studied (RA, TUB, SEN/SEGA), as well as CRM, and not found in any normal (non-lesion) tissues. Five of these CNA-bearing tumours also showed TSC1/TSC2 CN-LOH, and in all cases, a larger fraction of DNA was affected by the CN-LOH event than these CNAs, suggesting they occurred subsequently to a driving LOH event. Importantly, we used fluorescent in situ hybridization (FISH) on fresh-frozen tumour sections to confirm 24 of 25 (96%) molecularly-detected arm-level events (Fig. 2e and Supplementary Table 1).
RAs display adipose and PEComa features
Our next goal was to define the molecular signatures of each TSC hamartomatous lesion type using genome-wide DNA methylation and transcript profiling. Unsupervised clustering of DNA methylation array data revealed lesions of each type clustered with one another and away from normal (non-TSC) tissue counterparts, suggesting a high degree of molecular conservation within each (Supplementary Fig. 3). To investigate tumour-specific methylation, we calculated the hypermethylation fraction of each lesion as the fraction of probes methylated that lack methylation in a panel of normal tissues. Among TSC lesions, RAs had the highest hypermethylation fraction (Fig. 3a). Although this level was just a fraction of the hypermethylation observed in malignant tumours (Supplementary Fig. 4a), we identified 240 CpG probes, mapping to 149 genes, with enriched methylation in RAs compared to non-TSC normal kidneys (Fig. 3b and Supplementary Data 6). Several methylated genes—including WT1, SIX2, SLIT2, EMX2 and OSR1—are known to play roles in kidney development35,36. To determine whether this methylation is associated with differences (that is, decreases) in gene expression, we cross-referenced them with relative transcript levels determined by RNAseq. Of the 127 methylated genes detected in our RNAseq assay, 13 were significantly differentially expressed in RAs, with all but one specifically showing reduced expression in tumours (Supplementary Fig. 4b). While we did not detect a statistically significant association between hypermethylation and differential expression across all genes (χ2(1, n=16,408)=1.873, P=0.17), the decreased expression of 12 of 13 (92%) genes both hypermethylated and differentially expressed in RAs is consistent with methylation-induced silencing. Four RAs lacked the methylation signature shared by the bulk of RAs, two of which (from patient 05) may be attributed to a somatic DNMT3A-V716F mutation predicted to affect methyltransferase activity (Fig. 3b)37.
Next, we interrogated the RA transcriptome to establish whether gene expression patterns could provide insight to their development. For this effort, we used RNAseq data from a panel of non-TSC normal kidneys and 11 RA samples to identify 1,395 differentially expressed genes (DEGs; defined by log2 fold-change +/− >2 and limma moderated t statistic FDR-adjusted P<0.001) (Fig. 3c and Supplementary Data 7)38. Genes with the most substantial decrease in expression included those with roles in normal kidney function, such as NPHS2 (Fig. 3d), reflecting the loss of normal kidney tissue. Consistently, the top significantly enriched biological processes among genes decreased in RAs were primarily related to normal kidney development and function (Table 1). RAs are classified as PEComas, tumours arising from perivascular epithelioid cells (PECs) that co-express markers of melanocytes, bone, cartilage and smooth muscle, likely reflecting a neural crest origin39. Consistent with this, the two most highly expressed RA genes were CTSK, which has been proposed as a robust PEComa biomarker, and PMEL, which encodes a melanocyte-specific premelanosome protein (Fig. 3e,f)40. In fact, PMEL encodes the protein target of HMB-45 (gp100), a diagnostic antibody used to identify RAs and other PEComas clinically41.
To estimate the relative proportion of different cell types in RAs, we employed CIBERSORT, a computational framework for virtually sorting complex cell mixtures using gene expression data42. We created a custom gene signature differentiating cell types we suspected comprise RAs: (a) adipose tissue, smooth muscle and blood vessel, which histologically define RAs; (b) adult and fetal kidney, with the hypothesis that RAs may bear more resemblance to fetal than adult kidney; and (c) leukocytes, which frequently infiltrate tumour microenvironments. As expected, CIBERSORT predicted non-TSC kidney samples to be comprised exclusively of normal adult kidney and similarly, the two RAs with DEG signatures least similar to the other RAs (01-RA1 and 10-RA1) were also predicted to contain a significant amount of normal tissue (Fig. 3g). The remainder of RAs resembled mixtures of adipose tissue, smooth muscle, blood vessels, leukocytes and fetal kidney tissue, with most showing a striking enrichment in adipose tissue (Fig. 3g). The loss of normal kidney tissue and the presence of the three known RA components, including the lipoma-like phenotype of many, were supported by histology (Fig. 3h).
Brain lesions show evidence of significant neuroinflammation
Analogous to the approach we took for RAs, we next performed differential gene expression analysis of the two main classes of TSC-associated brain lesions, cortical tuber (n=15) and SEN/SEGA (n=15), using normal non-TSC brain tissues as negative controls. We identified 3,692 DEGs (log2 fold-change +/− >2; limma moderated t statistic FDR-adjusted P<0.001) in SEN/SEGAs and 297 DEGs in cortical tubers (Fig. 4a,b and Supplementary Datas 8 and 9). Almost all genes with decreased expression in cortical tubers were also decreased in SEN/SEGAs, with both lesion types showing decreased expression of genes related to synaptic transmission (Fig. 4c and Table 2). SEN/SEGAs showed a large number of uniquely decreased genes, which were associated with other normal nervous system processes (Table 2).
We found that genes most significantly increased in expression in both brain lesion types were related to the immune system and inflammation (Table 2). Antigen processing and presentation—specifically, major histocompatibility (MHC) class II—was a major process significantly enriched among increased SEN/SEGA DEGs (Table 2), which we highlighted by colour-coding a molecular map of this network according to average fold-changes in SEN/SEGAs (Fig. 4d). Most genes increased in expression in tubers were also increased in expression in SEN/SEGA, reflecting this shared immune signature (Fig. 4c). To identify genes with discordant expression between the two brain lesion types, we performed a final differential gene expression analysis between cortical tuber and SEN/SEGA. While genes uniquely decreased in expression in SEN/SEGA again mapped to normal nervous system processes, we found receptor-mediated endocytosis and angiogenesis were enriched among genes increased in SEN/SEGA compared to cortical tuber (Supplementary Table 2). This angiogenesis signature may contribute to the known vascularized nature of SEGAs. It is worth noting that while MTOR-related signalling was not identified as enriched in this analysis, we were able to detect an enrichment of MTORC1 networks in SEN/SEGA (but not TUB or RA) using a second pathway enrichment analysis (MetaCore) with reduced stringency of our analysis (Supplementary Table 3).
Finally, we wanted to explore the neuroinflammation phenotype further which we began by employing CIBERSORT to estimate the proportion of cell types constituting these lesions. Cortical tubers showed a relatively equal mixture of adult neuron and astrocyte, similar to normal non-TSC brain tissue, along with a small fraction of leukocytes (∼1%) (Fig. 5a). Meanwhile, SEN/SEGAs were estimated to be enriched in less differentiated neurons and astrocytes, a result substantiated by the decreased expression of known neuronal differentiation markers (Fig. 5b). In addition, SEN/SEGAs showed evidence of substantial leukocyte levels (12.3% mean) (Fig. 5a). To predict the relative abundance of individual immune cell components found in the leukocyte fraction of SEN/SEGA samples, we utilized a gene signature distinguishing 22 immune cell types42. We identified three immune cell types with relative fractions differing more than threefold between non-TSC brain and SEN/SEGAs (two-tailed Student’s t-tests; FDR-adjusted P<0.05) (Supplementary Data 10). When compared to normal brain tissue, SEN/SEGAs were predicted to be enriched in monocytes, and harbour a population of macrophages switched from a resting (M0) an activated (M2) state (Fig. 5c). Using immunohistochemistry (IHC) on sections from a fresh-frozen patient-derived SEGA, we confirmed the presence of this activated macrophage (microglia) population using CD68 (macrophage marker) and HLA-DR (a MHC-class II antigen), as well as AIF1/IBA1 (microglia marker) (Fig. 5d).
We have presented the most complete molecular portrait of TSC to date, adding genomic information beyond the well-described TSC1 and TSC2 loci. The genomes of TSC-associated lesions are relatively simple, with somatic mutation rates lower than most malignant tumours. The mutational burden of TSC lesions suggests a low mitotic index, consistent with their slow-growing nature and lack of exposure to genotoxic therapies. Instead, the most remarkable DNA feature of TSC genomes was whole or arm-level chromosome gains and losses, which were observed in four different tumour types from just over 10% of patients in our study. These generally (seven of nine tumours) co-occurred with large aberrations to TSC1/TSC2 (large deletions or CN-LOH), suggesting certain genomes may be less structurally stable. Future studies will be required to establish the role of these CNAs in TSC tumour growth.
By integrating targeted-deep sequencing that spanned introns and exons with high-resolution SNP arrays, we were able to identify pathogenic TSC1/TSC2 mutations in almost 94% of patients, leaving just two classified as NMI. The remaining cases may be explained by (a) mosaicism, in which only a portion of cells (and therefore, DNA) is affected by a mutation; (b) a third TSC locus (TSC3); or (c) TSC1/TSC2 mutations that have not yet been attributed pathogenicity, for example, intronic mutations that may affect splicing. The latter of these seems most plausible as our integrated platforms proved sensitive at detecting low-frequency mutations, including low-level mosaicism, and our WES analysis failed to provide evidence for TSC3. Moreover, we did identify rare, intronic TSC2 mutations of unknown significance in one of the NMI patients. Genetic testing of biological parents and biochemical evaluation of these mutants will resolve whether one of these variants is indeed pathogenic. Overall, our data is consistent with a recent thorough evaluation of 53 NMI patients by targeted-deep sequencing that concluded 85% of cases could be explained by low-frequency mosaic mutations or mutations in introns17.
In addition to finding germline mutations, we also provided a detailed description of the second hit landscape across TSC tumours. Our observation of widespread somatic TSC inactivation in RA and less common second hits in cortical tubers is consistent with previous studies27,28,29,30,31,32,43. Epigenetic silencing of TSC1/TSC2 has been postulated to explain a portion of 1-hit tumours and in fact, there has been some evidence that TSC1/TSC2 are subject to methylation44,45. However, we found no evidence of promoter methylation in 63 TSC-associated tissues analysed, reducing the likelihood that this mechanism contributes significantly to TSC1/TSC2 inactivation in TSC.
Although our second hit rate in cortical tubers (35%) was higher than most previous studies, somatic TSC inactivation in cortical tubers is clearly a less frequent and more sporadic event. This suggests that either only a small portion of the tuber is affected by a second hit (for example, one cellular component, such as giant cells), hindering its identification or that monoallelic inactivation of TSC1/TSC2 is sufficient for cortical malformation. The latter is supported by the fact that cortical tubers form prenatally, are found in a majority of TSC patients, and despite some evidence of proliferation46, lack appreciable growth in size or number over time. These features are consistent with a developmental origin rather than neoplastic formation via the sporadic acquisition of somatic TSC1/TSC2 mutations over time. This concept of haploinsufficiency is consistent with other features of this disease, such as cognitive and behavioural impairments, and to some degree, epilepsy47,48,49. The second hits found in a minority of cortical tubers may contribute to tuber pathology, although they are unlikely to represent a requirement for their formation.
Interestingly, two of the somatic TSC mutations in cortical tubers (44-TUB1 and 50-TUB1) were unusual and appeared to involve the loss of TSC1 or TSC2 introns. The deleted introns were continuous and breakpoints appeared to be precisely at exon–intron boundaries, raising the possibility that they are the consequence of somatic retroduplication events where reverse-transcribed copies of genes lacking introns are integrated into the genome, forming a processed pseudogene. Such events are widely present in human germline evolution but also have recently been reported to occur in cancers50,51,52. An added layer of intrigue stems from the fact that one of these events affected TSC1 but was found in a patient with a pathogenic germline TSC2 mutation. A similar case was previously reported in which a low-frequency somatic TSC1 mutation was identified in the periungual fibroma from a mosaic TSC2 patient53. While the authors suggested it was unlikely that a monoallelic mutation in TSC1 could cooperate with a germline TSC2 mutation to drive MTOR activation, we feel that together, our reports with similar observations from unique patients supports the idea that trans-heterozygous TSC1/TSC2 mutations may contribute to tumorigenesis in TSC. Interestingly, Tsc1+/−;Tsc2+/− compound heterozygous mice show increased numbers of hippocampal GFAP-positive astrocytes compared to either single heterozygote mice, suggesting potential epistatic interaction between monoallelic mutation of the two genes54.
Despite generally stable DNA genomes, TSC lesions were defined by marked and uniform changes in gene expression. In fact, the DEGs we identified in SEN/SEGAs constitute nearly 20% of all genes identified in our RNAseq assay. These RNA signatures were shared by tumours regardless of TSC1/TSC2 mutational status or presence of second hits. While enriched MTORC1 signalling was observed in SEN/SEGA, it was only significant when the pathway analysis stringency was reduced and it was not detected in RA or TUB. This result may be a consequence of tumour heterogeneity (that is, if only a portion of a tumour or specific cells—such as giant cells—bear second hits and are strongly driven by MTORC1 signalling) and is also consistent with the molecular role that MTOR regulation plays in translational control (versus transcription)55. Moreover, this dramatic expression signature also suggests that a strong (and common) cell-of-origin expression signature may be dominating over additional molecular signalling changes. This is best supported by the RA expression signature, which showed features of several cell types known to be derived from the neural crest, the proposed cell-of-origin for this lesion56. In addition, the SEN/SEGA gene expression signature showed evidence of less differentiated neurons and glia, consistent with their proposed derivation from neural stem progenitor cells (NSPCs), early and shared precursors of both of these cell types57,58.
Throughout this study, we employed CIBERSORT to generate computational estimates for the relative proportion of cell types in TSC lesions using gene expression data. While the original report of this methodology used microarray data and focused on immune cell types, we have extended its use to RNAseq data and a host of additional cell types. It is worth noting that CIBERSORT can only generate predictions using input cell types; therefore, additional cellular components of these tumours beyond those we tested may exist. We used these molecular tools to detect neuroinflammation associated with TSC brain lesions. Inflammation has been previously documented in TSC lesions, both in the brains of TSC animal models and in patient tissues59,60,61,62. Work from Zhang et al.62 suggested that this inflammation is directly related to hyperactive MTORC1 signalling and is not merely a result of seizure activity. Inflammation has even been detected in prenatal TSC brain lesions, suggesting it is an early, and sustained, feature of TSC pathology63. In our study, we uncovered inflammation in both cortical tubers and SEN/SEGAs, although the extent of inflammation appeared much more substantial in SEN/SEGAs. In fact, computational sorting of RNAseq data estimated as much as 20% of the mRNA fraction of SEN/SEGA tissues was associated with leukocytes, with a specific enrichment of activated macrophages. We postulate that this macrophage signature largely reflects the activation of brain-resident microglia, the primary immune cell component of the central nervous system (CNS) and known mediator of neuroinflammation, which is supported by positive AIF1/IBA1 staining in SEGA tissue. Reactive astrocytes, known key mediators of innate immunity in the CNS and neuroinflammation, also likely contribute to the immune signature we detected. While inflammation may serve a protective role in response to acute brain injury, triggering angiogenesis and promoting tissue repair, chronic neuroinflammation may instead be destructive and contribute to neuronal damage, as is the case in CNS pathologies like Alzheimer’s and Parkinson’s disease. Important future work should focus on defining the relationship between neuroinflammation and neurological symptoms of TSC, including seizure activity and cognitive impairment, as well as evaluation of anti-inflammatory agents in the treatment of TSC.
Patients and samples
Samples from TSC patients or non-TSC organ donors were acquired from the NIH NeuroBioBank’s Brain and Tissue Repository at the University of Maryland, Houston-McGovern Medical School at the University of Texas, Cincinnati Children’s Hospital Medical Center, New York University School of Medicine and Helen DeVos Children’s Hospital. All tissues used in this study were fresh-frozen and collected at the time of surgery or procedure or post-mortem (see Supplementary Data 1 for details). This study was approved by the Van Andel Research Institute (VARI) Institutional Review Board (IRB). Written informed consent was obtained from all human participants providing samples. Samples were reviewed by a certified clinical pathologist to confirm tissue type and assess integrity, whenever possible (samples with inconsistent, unlikely to be consistent or unclear diagnoses were excluded from the study). Samples were also excluded if they failed to produce usable data on two of three DNA platforms (WES, SNP array and targeted TSC sequencing), with the exception of one non-tumour tissue sample in which the germline mutation was identified in the completed platform (eliminating the need for the additional platforms to be completed).
For immunohistochemistry, 5 μm fresh-frozen tissue sections were fixed and stained with primary antibodies (CD68: 1:100; HLA-DR: 1:40; AIF1/IBA1: 1:500), secondary antibodies (Ultramap anti-mouse HRP multimer) and detection reagent (Ventana Chromomap DAB). Slides were processed on the Discovery Ultra platform (Ventana) and imaged using the ScanScope XT digital pathology slide scanner (Aperio).
DNA and RNA isolation
The specific method for DNA and RNA isolations is indicated in Supplementary Data 1. For majority of frozen tissues, DNA and RNA was simultaneously isolated using a modified version of the method described in Pena-Llopis and Brugarolas64. Briefly, tissues were lysed and homogenized using mirVana kit lysis buffer (Ambion), a micropestle and QIAshredder columns (Qiagen). DNA was isolated using AllPrep columns (Qiagen) while flow-throughs were used to isolate RNA using an acid phenol–chloroform extraction and the mirVana kit (Ambion). DNA integrity was confirmed by agarose gel electrophoresis and RNA integrity was confirmed using a BioAnalyzer 2100 (Agilent). DNA and RNA concentrations were determined using a Qubit 2.0 fluorometer (Invitrogen).
DNA sequencing was completed at the HudsonAlpha Institute for Biotechnology (HAIB) Genomic Services Laboratory (GSL) or Beijing Genomics Institute (BGI) at the Philadelphia Children’s Hospital. Briefly, exonic DNA was enriched using a SeqCap EZ Human Exome Library v3.0 (NimbleGen) or SureSelect Human All Exon capture kit (Agilent) from genomic DNA. Libraries were pooled and clustered at 16–18 pM on the HiSeq 2500 or HiSeq 2000 with high output flowcells and sequenced at 100PE according to Illumina protocols. Fastq files were generated using Illumina software, aligned to the hg19 genome with BWA-MEM and variants called using Haplotype Caller in GATK. Filtered variants were annotated with Variant Effect Predictor (VEP) and imported to GEMINI. Detailed methods can be found in Supplementary Methods.
For the TSC1/TSC2 expression analysis, pair-wise Welch’s t-tests (in GraphPad Prism 6 for Windows, version 6.07) of 5 groups of data (for Fig. 1m: non-TSC tissue; NMI and TSC1 tumours; 0, 1 or 2 truncating TSC2 mutations; for Fig. 1n: non-TSC tissue; NMI and TSC2 tumours; 0, 1 or 2 truncating TSC1 mutations) were followed by false discovery rate (FDR) correction (in R) to generate corrected P values. This approach was taken because samples failed Bartlett’s test for homogeneity of variances, ruling out ANOVA as an option. Truncating mutations included nonsense, frameshift, splicing and large deletions. Tumours with truncating germline mutations and CN-LOH were classified as harbouring two truncating mutations (because CN-LOH duplicates the germline mutant allele). As a priority for visualization, only non-TSC tissue and 1 or 2 truncating mutation groups were shown in Fig. 1m,n although all were included in the statistical analysis. For immune cell type analysis by CIBERSORT (Fig. 5c), the relative fraction of each cell type in SEN/SEGA was divided by the fraction in non-TSC brain. Individual two-tailed student’s t-test P values were adjusted via the FDR method using R. Those cell types with +/− >threefold changes and FDR-adjusted P<0.05 were included in the panel. To test the association between hypermethylation and differential expression, we identified genes covered by both HM450 and our RNAseq assay and categorized each as being hypermethylated and a DEG (13) or not a DEG (114), or not hypermethylated and a DEG (1,156) or not a DEG (15,125). These values were entered into a 2 × 2 contingency table and a χ2 test performed in GraphPad Prism 6.
Targeted TSC1/TSC2 sequencing
We designed a custom targeted enrichment kit (SeqCap EZ Choice Library, NimbleGen) with comprehensive coverage of TSC1 and TSC2, including upstream and downstream elements (including PKD1), exons and introns. Samples were multiplexed (9–10 per library hybridization) and sequenced similar to above at the HAIB GSL using Illumina reagents and the HiSeq 2500. Alignment and variant calling and annotation were performed similar to WES. In addition, we explored mutations present at low allele frequencies down to 0.5% in the deep sequencing experiment by recalling mutations using LoFreq65 and VarDict66. Detailed information can be found in Supplementary Methods.
TSC1/TSC2 mutation calling
To be included in Supplementary Data 2, TSC1/TSC2 variants (>10 × total read-depth) were required to be either published in the tuberous sclerosis Leiden Open Variation Database (LOVD; www.LOVD.nl/TSC2; www.LOVD.nl/TSC1) (v2.0 Build 36) as pathogenic or probably pathogenic, or if not present in LOVD (or ‘unknown’ pathogenicity in LOVD), determined to be rare (not present in 1000 genomes database) and impactful to gene function (medium/moderate or high impact SNV or INDEL). All large deletions and CN-LOH events affecting TSC1 and TSC2 were assumed detrimental to gene function and included. For non-tumour (normal) tissues (n=33) or tumours that were paired with non-tumour tissue from the same patient (n=42), the germline or somatic origin of mutations could be absolutely determined. We then used information from these samples to establish features of germline and somatically derived mutations to predict the origin of mutations in unpaired tumours (n=36), described in detail in Supplementary Methods.
Somatic mutation analysis
Somatic SNVs were identified using MuTect using default settings and annotated using VEP and GEMINI67. INDELs were detected and characterized in both tumour and matched normal samples using Pindel68. To call a somatic INDEL, we required >5 × coverage in both tumour and normal samples, >5 reads supporting the variant allele in the tumour with 0 reads in the matched normal sample, and a variant allelic fraction of >0.10 in the tumour sample. Somatic mutation rates were determined by normalizing the combined number of somatic SNVs and INDELs by the total number of bases with >5 × read-depth in both tumour and normal samples. We excluded reads with base quality <20 at each mutation locus. The mutation rates for cancers presented in Fig. 2b were obtained from Kandoth et al.34. Supplementary Data 3 includes only SNVs and INDELs passing more stringent criteria: SNVs required >10 × read-depth at the variant position and 0 variant reads in the normal samples; all somatic INDELs were manually inspected in the Integrative Genomics Viewer (Broad Institute) and clear artifacts were excluded.
RNA sequencing and differential gene expression analysis
RNA sequencing was completed at the HAIB GSL. Briefly, messenger RNA (mRNA) libraries were prepared using NEBNext reagents (New England BioLabs), and samples underwent directional sequencing on the Illumina HiSeq 2500 using100 bp paired end reads. Quality-filtered reads were aligned to the hg19 genome using Subread. Raw read counts obtained using FeatureCounts were imported into R for differential expression analysis via limma38 and counts per million (CPM) calculated and log2-transformed using voom69 followed by trimmed mean of M-values (TMM) normalization. GeneAnalytics (LifeMap Sciences; geneanalytics.genecards.org) was used for primary gene set enrichment analysis70. A maximum of 300 gene symbols were used and up to 10 GO biological processes with medium or high matching scores (FDR-adjusted P<0.05) were included in the results. Only processes with at least 10 matched genes were shown in Tables 1 and 2. A follow-up enrichment analysis to search for MTORC1-related signatures was completed using MetaCore. For this, gene-level fold changes and adjusted P values were imported into MetaCore version 6.29 build 68613 (Thomson Reuters) for pathway analysis. Pathway analysis was performed using the Pathway Maps One-Click Analysis on genes with an absolute log-fold change >1 and FDR-adjusted P-value<0.001. Pathway Maps with a FDR-adjusted P-value <0.05 were considered significant. RNAseq variant calling was conducted using GATK (v3.0) using the suggested Best Practices parameters and with a two-pass STAR (v 2.4.2a) alignment method to the hg19 genome. DNA variants identified in RNAseq are indicated in Supplementary Data 2.
CIBERSORT42 was used to estimate the relative fraction of cell types. Publically available RNA sequencing data was downloaded from the NCBI Short Read Archive (http://www.ncbi.nlm.nih.gov/sra) (see Supplementary Methods for detailed information). The values for the iPSC neurons were duplicated into two columns to meet CIBERSORT input requirements. Read quality was assessed using FASTQC v. 0.11.3 (http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc/). Reads were aligned to the hg19 genome using Subread (v1.4.5) with default parameters. Raw read counts were obtained as described for RNAseq. For immune cell types, the LM22 gene signature was used41. Only samples with estimates yielding P values<0.05 were reported.
SNP arrays and copy number analysis
Copy number analysis was performed using Infinium HumanOmni2.5S Arrays (Illumina) at the HAIB GSL. Briefly, genotypes were called and total copy number, log-R ratio (LRR) and B-allele frequency (BAF) estimated for each SNP using IDAT files in GenomeStudio (v2011.1, Illumina). Total genome-wide copy number estimates were refined using tangent normalization and individual copy number estimates underwent segmentation per-sample arm-level and gene-level copy ratios were identified from segmented data using GISTIC. Purity and ploidy estimates and allelic integer copy number (including regions of copy-neutral loss-of-heterozygosity) were calculated from LRRs and BAFs using ASCAT. Arm-level copy number events determined by GISTIC 2.0 were visually validated in genome-wide LRR and BAF plots generated by ASCAT 2.4. Chromosomes 9 and 16, as well as the region in chromosome 9q containing TSC1 and the region in chromosome 16p containing TSC2, were visually inspected using genoCN to validate loci with copy-neutral loss-of-heterozygosity and focal deletions as reported by ASCAT 2.4 and/or GISTIC. Copy number events detected only visually because of low tumour purity or low signal were also reported.
Array-based DNA methylation assay
DNA methylation profiling was completed using Infinium HumanMethylation450 BeadChips (Illumina) at the University of Southern California Epigenome Center to obtain DNA methylation profiles, which were analysed using the same pipeline used for The Cancer Genome Atlas (TCGA) project. Briefly, bisulfite conversion of genomic DNA was performed with the EZ-96 DNA Methylation Kit (Zymo Research). After quality control measures, bisulfite-converted DNA was whole genome amplified and fragmented prior to hybridization to BeadArrays, which were scanned using the Illumina iScan technology. IDAT files were used to extract the intensities and calculate beta values for each probe and sample with the R-based methylumi package. A P value comparing the intensity of each probe to the background level was calculated and data points with detection P values >0.05 were deemed not significantly different from background measurements. Detailed information can be found in Supplementary Methods.
Fluorescent in situ hybridization
FISH probes were prepared from purified BAC clones (BACPAC Resource Center; bacpac.chori.org); see Supplementary Methods for specific BAC probes and detailed information. Briefly, each clone was labelled with Green-dUTP, Orange-dUTP or Red-dUTP by nick translation. Tumour touch preparations were made on glass slides, which were fixed, dried, aged, digested and washed. Slides were placed in 1% formaldehyde, washed and dehydrated in an ethanol series. Slides were then denatured, washed and air-dried. FISH probes were denatured probe was applied to each sample slide. Coverslips were adhered and slides hybridized overnight in a ThermoBrite hybridization system (Abbott Molecular). Post-hybridization, slides were washed with 2 × SSC and briefly rinsed with water. Slides were dried and counterstained with VectaShield mounting medium with 4′-6-diamidino-2-phenylindole (DAPI). Image acquisition was performed at × 600 or × 1,000 system magnification with a COOL-1300 SpectraCube camera (Applied Spectral Imaging-ASI) mounted on an Olympus BX43 microscope. Images were analysed using FISHView v7 software (ASI) and at least 200 interphase nuclei were scored for each sample.
All raw data has been deposited in the Database of Genotypes and Phenotypes (dbGaP) under the accession code phs001357.v1.p1. All other remaining data are available within the Article and Supplementary Files, or available from the authors upon request.
How to cite this article: Martin, K. R. et al. The genomic landscape of tuberous sclerosis complex. Nat. Commun. 8, 15816 doi: 10.1038/ncomms15816 (2017).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Osborne, J. P., Fryer, A. & Webb, D. Epidemiology of tuberous sclerosis. Ann. N. Y. Acad. Sci. 615, 125–127 (1991).
European Chromosome 16 Tuberous Sclerosis Consortium. Identification and characterization of the tuberous sclerosis gene on chromosome 16. Cell 75, 1305–1315 (1993).
Gomez, M. R. Criteria for diagnosis. in Tuberous Sclerosis ed. Gomez M. R. 10–23Raven Press (1988).
Roach, E. S., Gomez, M. R. & Northrup, H. Tuberous sclerosis complex consensus conference: revised clinical diagnostic criteria. J. Child Neurol. 13, 624–628 (1998).
Krueger, D. A. & Northrup, H. International Tuberous Sclerosis Complex Consensus Group. Tuberous sclerosis complex surveillance and management: recommendations of the 2012 International Tuberous Sclerosis Complex Consensus Conference. Pediatr. Neurol. 49, 255–265 (2013).
Crino, P. B., Nathanson, K. L. & Henske, E. P. The tuberous sclerosis complex. N. Engl. J. Med. 355, 1345–1356 (2006).
Gallagher, A. et al. Decreased language laterality in tuberous sclerosis complex: a relationship between language dominance and tuber location as well as history of epilepsy. Epilepsy Behav. 25, 36–41 (2012).
Goodman, M. et al. Cortical tuber count: a biomarker indicating neurologic severity of tuberous sclerosis complex. J. Child Neurol. 12, 85–90 (1997).
Kassiri, J., Snyder, T. J., Bhargava, R., Wheatley, B. M. & Sinclair, D. B. Cortical tubers, cognition, and epilepsy in tuberous sclerosis. Pediatr. Neurol. 44, 328–332 (2011).
Kothare, S. V. et al. Severity of manifestations in tuberous sclerosis complex in relation to genotype. Epilepsia 55, 1025–1029 (2014).
Bernstein, J. & Robbins, T. O. Renal involvement in tuberous sclerosis. Ann. N. Y. Acad. Sci. 615, 36–49 (1991).
Shepherd, C. W., Gomez, M. R., Lie, J. T. & Crowson, C. S. Causes of death in patients with tuberous sclerosis. Mayo Clin. Proc. 66, 792–796 (1991).
Dabora, S. L. et al. Mutational analysis in a cohort of 224 tuberous sclerosis patients indicates increased severity of TSC2, compared with TSC1, disease in multiple organs. Am. J. Hum. Genet. 68, 64–80 (2001).
van Slegtenhorst, M. et al. Identification of the tuberous sclerosis gene TSC1 on chromosome 9q34. Science 277, 805–808 (1997).
Sancak, O. et al. Mutational analysis of the TSC1 and TSC2 genes in a diagnostic setting: genotype–phenotype correlations and comparison of diagnostic DNA techniques in tuberous sclerosis complex. Eur. J. Hum. Genet. 13, 731–741 (2005).
Verhoef, S. et al. High rate of mosaicism in tuberous sclerosis complex. Am. J. Hum. Genet. 64, 1632–1637 (1999).
Tyburczy, M. E. et al. Mosaic and intronic mutations in TSC1/TSC2 explain the majority of TSC patients with no mutation identified by conventional testing. PLoS Genet. 11, e1005637 (2015).
Kozlowski, P. et al. Identification of 54 large deletions/duplications in TSC1 and TSC2 using MLPA, and genotype-phenotype correlations. Hum. Genet. 121, 389–400 (2007).
Kwiatkowska, J., Wigowska-Sowinska, J., Napierala, D., Slomski, R. & Kwiatkowski, D. J. Mosaicism in tuberous sclerosis as a potential cause of the failure of molecular diagnosis. N. Engl. J. Med. 340, 703–707 (1999).
Dibble, C. C. et al. TBC1D7 is a third subunit of the TSC1-TSC2 complex upstream of mTORC1. Mol. Cell 47, 535–546 (2012).
Franz, D. N. et al. Efficacy and safety of everolimus for subependymal giant cell astrocytomas associated with tuberous sclerosis complex (EXIST-1): a multicentre, randomised, placebo-controlled phase 3 trial. Lancet 381, 125–132 (2013).
Krueger, D. A. et al. Everolimus for subependymal giant-cell astrocytomas in tuberous sclerosis. N. Engl. J. Med. 363, 1801–1811 (2010).
Bissler, J. J. et al. Sirolimus for angiomyolipoma in tuberous sclerosis complex or lymphangioleiomyomatosis. N. Engl. J. Med. 358, 140–151 (2008).
Bissler, J. J. et al. Everolimus for angiomyolipoma associated with tuberous sclerosis complex or sporadic lymphangioleiomyomatosis (EXIST-2): a multicentre, randomised, double-blind, placebo-controlled trial. Lancet 381, 817–824 (2013).
McCormack, F. X. et al. Efficacy and safety of sirolimus in lymphangioleiomyomatosis. N. Engl. J. Med. 364, 1595–1606 (2011).
Krueger, D. A. et al. Long-term treatment of epilepsy with everolimus in tuberous sclerosis. Neurology 87, 2408–2415 (2016).
Green, A. J., Johnson, P. H. & Yates, J. R. The tuberous sclerosis gene on chromosome 9q34 acts as a growth suppressor. Hum. Mol. Genet. 3, 1833–1834 (1994).
Sepp, T., Yates, J. R. & Green, A. J. Loss of heterozygosity in tuberous sclerosis hamartomas. J. Med. Genet. 33, 962–964 (1996).
Chan, J. A. et al. Pathogenesis of tuberous sclerosis subependymal giant cell astrocytomas: biallelic inactivation of TSC1 or TSC2 leads to mTOR activation. J. Neuropathol. Exp. Neurol. 63, 1236–1242 (2004).
Henske, E. P. et al. Allelic loss is frequent in tuberous sclerosis kidney lesions but rare in brain lesions. Am. J. Hum. Genet. 59, 400–406 (1996).
Niida, Y. et al. Survey of somatic mutations in tuberous sclerosis complex (TSC) hamartomas suggests different genetic mechanisms for pathogenesis of TSC lesions. Am. J. Hum. Genet. 69, 493–503 (2001).
Knudson, A. G. Jr Mutation and cancer: statistical study of retinoblastoma. Proc. Natl Acad. Sci. USA 68, 820–823 (1971).
Au, K. S. et al. Genotype/phenotype correlation in 325 individuals referred for a diagnosis of tuberous sclerosis complex in the United States. Genet. Med. 9, 88–100 (2007).
Tamborero, D. et al. Comprehensive identification of mutational cancer driver genes across 12 tumor types. Sci. Rep. 3, 2650 (2013).
Patel, S. R. & Dressler, G. R. The genetics and epigenetics of kidney development. Semin. Nephrol. 33, 314–326 (2013).
Brunskill, E. W. et al. Atlas of gene expression in the developing kidney at microanatomic resolution. Dev. Cell 15, 781–791 (2008).
Hollink, I. H. et al. Low frequency of DNMT3A mutations in pediatric AML, and the identification of the OCI-AML3 cell line as an in vitro model. Leukemia 26, 371–373 (2012).
Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47 (2015).
Fernandez-Flores, A. Evidence on the neural crest origin of PEComas. Rom. J. Morphol. Embryol. 52, 7–13 (2011).
Martignoni, G. et al. Cathepsin K expression in the spectrum of perivascular epithelioid cell (PEC) lesions of the kidney. Mod. Pathol. 25, 100–111 (2012).
Bonetti, F. P. et al. The perivascular epithelioid cell and related lesions. Adv. Anat. Pathol. 4, 343–358 (1997).
Newman, A. M. et al. Robust enumeration of cell subsets from tissue expression profiles. Nat. Methods 12, 453–457 (2015).
Giannikou, K. et al. Whole exome sequencing identifies TSC1/TSC2 biallelic loss as the primary and sufficient driver event for renal angiomyolipoma development. PLoS Genet. 12, e1006242 (2016).
Lesma, E. et al. The methylation of the TSC2 promoter underlies the abnormal growth of TSC2 angiomyolipoma-derived smooth muscle cells. Am. J. Pathol. 174, 2150–2159 (2009).
Jiang, W. G. et al. Tuberin and hamartin are aberrantly expressed and linked to clinical outcome in human breast cancer: the role of promoter methylation of TSC genes. Eur. J. Cancer 41, 1628–1636 (2005).
Lee, A. et al. Markers of cellular proliferation are expressed in cortical tubers. Ann. Neurol. 53, 668–673 (2003).
Lozovaya, N. et al. Selective suppression of excessive GluN2C expression rescues early epilepsy in a tuberous sclerosis murine model. Nat. Commun. 5, 4563 (2014).
Ehninger, D. et al. Reversal of learning deficits in a Tsc2+/− mouse model of tuberous sclerosis. Nat. Med. 14, 843–848 (2008).
Goorden, S. M., van Woerden, G. M., van der Weerd, L., Cheadle, J. P. & Elgersma, Y. Cognitive deficits in Tsc1+/− mice in the absence of cerebral lesions and seizures. Ann. Neurol. 62, 648–655 (2007).
Cooke, S. L. et al. Processed pseudogenes acquired somatically during cancer development. Nat. Commun. 5, 3644 (2014).
Kazazian, H. H. Jr Processed pseudogene insertions in somatic cells. Mob. DNA 5, 20 (2014).
Ewing, A. D. et al. Retrotransposition of gene transcripts leads to structural variation in mammalian genomes. Genome Biol. 14, R22 (2013).
Tyburczy, M. E. et al. Sun exposure causes somatic second-hit mutations and angiofibroma development in tuberous sclerosis complex. Hum. Mol. Genet. 23, 2023–2029 (2014).
Uhlmann, E. J. et al. Heterozygosity for the tuberous sclerosis complex (TSC) gene products results in increased astrocyte numbers and decreased p27-Kip1 expression in TSC2+/− cells. Oncogene 21, 4050–4059 (2002).
Ma, X. M. & Blenis, J. Molecular mechanisms of mTOR-mediated translational control. Nat. Rev. Mol. Cell Biol. 10, 307–318 (2009).
Delaney, S. P., Julian, L. M. & Stanford, W. L. The neural crest lineage as a driver of disease heterogeneity in tuberous sclerosis complex and lymphangioleiomyomatosis. Front. Cell Dev. Biol. 2, 69 (2014).
Zhou, J. et al. Tsc1 mutant neural stem/progenitor cells exhibit migration deficits and give rise to subependymal lesions in the lateral ventricle. Genes Dev. 25, 1595–1600 (2011).
Magri, L. et al. Sustained activation of mTOR pathway in embryonic neural stem cells leads to development of tuberous sclerosis complex-associated lesions. Cell Stem Cell 9, 447–462 (2011).
Maldonado, M. et al. Expression of ICAM-1, TNF-alpha, NF kappa B, and MAP kinase in tubers of the tuberous sclerosis complex. Neurobiol. Dis. 14, 279–290 (2003).
Boer, K. et al. Inflammatory processes in cortical tubers and subependymal giant cell tumors of tuberous sclerosis complex. Epilepsy Res. 78, 7–21 (2008).
Boer, K. et al. Gene expression analysis of tuberous sclerosis complex cortical tubers reveals increased expression of adhesion and inflammatory factors. Brain Pathol. 20, 704–719 (2010).
Zhang, B., Zou, J., Rensing, N. R., Yang, M. & Wong, M. Inflammatory mechanisms contribute to the neurological manifestations of tuberous sclerosis complex. Neurobiol. Dis. 80, 70–79 (2015).
Prabowo, A. S. et al. Fetal brain lesions in tuberous sclerosis complex: TORC1 activation and inflammation. Brain Pathol. 23, 45–59 (2013).
Pena-Llopis, S. & Brugarolas, J. Simultaneous isolation of high-quality DNA, RNA, miRNA and proteins from tissues for genomic applications. Nat. Protoc. 8, 2240–2255 (2013).
Wilm, A. et al. LoFreq: a sequence-quality aware, ultra-sensitive variant caller for uncovering cell-population heterogeneity from high-throughput sequencing datasets. Nucleic Acids Res. 40, 11189–11201 (2012).
Lai, Z. et al. VarDict: a novel and versatile variant caller for next-generation sequencing in cancer research. Nucleic Acids Res. 44, e108 (2016).
Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013).
Ye, K., Schulz, M. H., Long, Q., Apweiler, R. & Ning, Z. Pindel: a pattern growth approach to detect break points of large deletions and medium sized insertions from paired-end short reads. Bioinformatics 25, 2865–2871 (2009).
Law, C. W., Chen, Y., Shi, W. & Smyth, G. K. voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 15, R29 (2014).
Ben-Ari Fuchs, S. et al. GeneAnalytics: an integrative gene set analysis tool for next generation sequencing, RNAseq and microarray data. OMICS 20, 139–151 (2016).
We are grateful to all the patients and families who contributed to this study. A portion of human tissue was obtained from the NIH NeuroBioBank’s Brain and Tissue Repository at the University of Maryland (Baltimore, MD). We thank members of the MacKeigan laboratory and TSC Pathway of Hope External Advisory Board (Peter Laird, Peter Saltonstall and Len Post) for critical discussions and feedback. We also thank Nicole Doppel for project management; Jennifer Webb, Sarah Nota, Molly Griffith and Maxwell Mays for clinical coordination; Alejandro Salah for clinical information; and Braden Boone, Lisa Turner and Jennifer Kordich for technical assistance and experimental insights. This work was supported by grants and funding from the Michigan Strategic Fund, Van Andel Research Institute, Tuberous Sclerosis Alliance, Blue Cross Blue Shield of Michigan Foundation, Great Lakes Scrip, Rockford Construction, Colliers International, Team Hannah for TSC and individual donors. J.P.M. has research support from the NIH National Cancer Institute (R01CA197398) and the Tuberous Sclerosis Alliance. D.A.K. has research support from the NIH National Institute of Neurological Disorders and Stroke (U01-NS082320, U54-NS092090 and P20-NS080199) and the Tuberous Sclerosis Alliance.
The authors declare no competing financial interests.
Supplementary figures, supplementary tables, supplementary methods and supplementary references. (PDF 1214 kb)
Information about individual patients and tissue samples. RA: renal angiomyolipoma; SEGA: subependymal giant cell astrocytoma; SEN: subependymal nodule; TUB: cortical tuber; CRM: cardiac rhabdomyoma; SK: skin lesion; MG: matched germline; UG: unmatched germline; MB: matched TSC brain; NK: normal non-TSC kidney; NB: normal nonTSC brain; MK: matched TSC kidney; Univ of MD: NIH NeuroBioBank's Brain and Tissue Repository at the University of Maryland; CCHMC: Cincinnati Children's Hospital Medical Center; Univ of TX: Houston-McGovern Medical School at the University of Texas; ND: not determined; NA: not applicable; any tissues reviewed by histopathology and not receiving a diagnosis confirmation as definite, consistent, or probable were excluded from study. Yes indicates assay completed and generated usable data; No indicates assay not attempted or data not usable. (XLSX 24 kb)
TSC1 and TSC2 mutations in TSC patient samples. Mutations were detected using genomic DNA (i.e., protein level changes are predicted, not tested) and annotated as follows: genomic events (“g.” prefix) are described on chr16 (TSC2) or chr9 (TSC1) of hg19; cDNA events (“c.” prefix) are described on NM_000548.3 (TSC2) or NM_000368.4 (TSC1); protein events ("p." prefix) are described on NP_000593.2 (TSC2) or NP_000359.1 (TSC1). If LOVD entry indicated "no known pathogenicity" or "probably no known pathogenicity", these were excluded and only occurred in samples with other pathogenic mutations. For mutations found on multiple platforms, that with the greatest support (i.e., depth) included. Read depth is depth at which variant confidently called; not necessarily total depth at that loci (i.e., in targeted sequencing). TSC (LoFreq) indicates a mutation called by LoFreq (and VarDict) from targeted sequencing data (see Methods). Germline and somatic events predicted in unmatched tumors indicated with * (see Methods). Variants unsupported by 2nd platform not included (i.e., WES mutation not called confidently by targeted TSCseq removed). All somatic mutations (determined to be absent in germline and specific to tumor) included, regardless of LOVD pathogenicity (e.g., 01-RA1 missense predicted to be deleterious but "unknown" in LOVD). Abbreviations are as follows: N/A or “---“ = Not Applicable; N/D (visual) = CN-LOH event visually detected, but exact coordinates could not be determined; Y = Yes (assay completed successfully); N = No (assay not attempted, sample or data failed QC); VAF = variant allelic fraction (number of variant-containing reads divided by the total number of reads at that position); Ref = reference allele; Alt = Alternate (variant) allele; WES = whole exome sequencing; RNA RNAseq; SNP = SNP array; TSC-seq = targeted TSC1/TSC2 sequencing; bp = basepair (DNA); LOVD = Leiden Open Variation Database; HGVS = Human Genome Variation Society (annotation indicated according to HGVS guidelines). (XLSX 45 kb)
Somatic mutations. Somatic SNVs and INDELs, called by MuTect and Pindel, respectively, and detailed. Only SNVs with at least 10 total reads and 0 variant reads in normal (non-tumor) samples included. Only INDELs passing manual (visual) inspection are included. See Methods. AA = amino acid; AAF 1kg All = minor allele frequencies in the 1000 genomes database (all populations); COSMIC = Catalogue of Somatic Mutations in Cancer identifier; CADD = combined annotation dependent deletion score; Ref = reference (nonvariant) allele; Alt = alternate (variant) allele; VAF = variant allelic fraction (variant reads divided by total reads, expressed as a percentage); chr: chromosome; High confidence and candidate drivers are as classified in Tamborero et al., 2013. MTOR signaling genes as classified per KEGG pathway hsa04150. (XLSX 60 kb)
Copy number segment data. Segment data generated from SNP arrays and fed into GISTIC 2.0 to generate arm and whole chromosome level calls (see Table S5). See Methods. (XLSX 566 kb)
GISTIC arm-level copy number aberrations. Arm-level calls made by an Ziggurat Deconstruction within GISTIC 2.0 (longest arm-level copy level on each arm above 50% of the length of the arm is reported). (XLSX 18 kb)
RA-specific methylation. CpG probes methylated specifically in RAs are included. Genes that CpG probes map to, if any, are included ("---" indicates no linked genes). See Methods. (XLSX 103 kb)
Differentially expressed genes in RAs compared to non-TSC kidneys. Differentially expressed genes included with log2 fold-change (RA / NK) +/- > 2 with FDR-adjusted p < 0.001 are included. T = t-statistic; B = b-statistic. See Methods. (XLSX 552 kb)
Differentially expressed genes in SEN/SEGAs compared to non-TSC brain. Differentially expressed genes included with log2 fold-change (SEN/SEGA versus NB) +/- > 2 with FDR-adjusted p < 0.001 are included. T = t-statistic; B = b-statistic. See Methods. (XLSX 1836 kb)
Differentially expressed genes in cortical tuber compared to non-TSC brain. Differentially expressed genes included with log2 fold-change (TUB versus NB) +/- > 2 with FDR-adjusted p < 0.001 are included. T = t-statistic; B = b-statistic. See Methods. (XLSX 171 kb)
Analysis of immune cell fractions in SEN/SEGA by CIBERSORT. CIBERSORT estimated fractions of individual immune cell types. Individual two-tailed student's t-tests were calculated for each cell type (SEN/SEGA versus non-TSC brain) and then adjusted in R by the FDR method. See Methods. (XLSX 20 kb)
About this article
Cite this article
Martin, K., Zhou, W., Bowman, M. et al. The genomic landscape of tuberous sclerosis complex. Nat Commun 8, 15816 (2017). https://doi.org/10.1038/ncomms15816
Insight into developmental mechanisms of global and focal migration disorders of cortical development
Current Opinion in Neurobiology (2021)
Genetic pathogenesis of the epileptogenic lesions in Tuberous Sclerosis Complex: Therapeutic targeting of the mTOR pathway
Epilepsy & Behavior (2021)
MITF is a driver oncogene and potential therapeutic target in kidney angiomyolipoma tumors through transcriptional regulation of CYR61
CKMT1B is a potential prognostic biomarker and associated with immune infiltration in Lower-grade glioma
PLOS ONE (2021)
Journal of Genetics and Genomics (2021)