Burden and characteristic features of TSC1 and TSC2 in a cohort
In order to fulfil the aims of this study, the characterisation of the patient cohort was first conducted. To do this, examination of several key variables was necessary which included the prevalence and type of each specific TSC mutation, ratio of males to females, and diagnostic data of the brain, kidney, and heart to further delve into the very essence of TSC as a multisystem heterogenous genetic disorder.
TSC1 and TSC2 mutations in the patient cohort were identified using NGS and SS methods. TSC2 variants represented the majority of the cohort, constituting approximately 72% (N = 68) of all 95 genetically tested TSC patients (Fig. 2A). In contrast, only 25% of tested patients had TSC1 mutations (N = 24), and no mutation was identified (NMI) in the remainder (N = 3) (Fig. 2A). This means that the ratio of TSC1 to TSC2 patients for this cohort is approximately 1:2.8. A further breakdown of the mutation variant types of the 92 patients with an identified mutation is summarised in Fig. 2B, where 63% of all changes corresponded to small variants (SV) such as small deletions or insertions, duplications, or point mutations. There was also a further group of 13 patients, where no mutation data was available on hospital databases beyond unspecified ‘TSC1 mutation’ or ‘TSC2 mutation’, who were therefore classed into an ‘unknown’ category. The data presented in Fig. 3B was then further explored in Fig. 2C where the proportion of each type of variant in TSC1 vs TSC2 was compared and statistically evaluated. There was no significant difference between TSC1 and TSC2 groups in the case of nonsense mutations (p = 0.059), missense mutations (p = 0.051), intronic mutations (p = 0.105) and large deletions (p = 1.00). However, a significant difference in the proportion of frameshift mutations was identified (p = 0.0018) comprising 33% and 6% of the TSC1 vs TSC2 groups, respectively.
Next, a meta-analysis was conducted of the male to female ratio (M:F) for this study in addition to ten other previous studies of TSC patient cohorts (Fig. 2D). Within the CAV UHB cohort, there were 50 male patients and 42 females, resulting in a M:F ratio of 1.19:1 (Fig. 3D). The 95% confidence interval (CI) intersects with the line of no effect (x = 1), and thus, at the given level of confidence, the M:F ratio does not differ from equal ratio and is therefore statistically insignificant (Fig. 3D). Explicitly, TSC burdens male and female patients equally in this particular cohort. Out of the other ten studies inspected, a further eight were also synonymous with this result and also showed equal burden of TSC in male and female patients. The total M:F ratio across all studies was calculated at 1.04:1 with a slim confidence interval of 0.97 to 1.10 and therefore did not differ from an equal ratio. Cochrane’s Q test revealed no significant heterogeneity between studies (Q = 8.14, df = 11, p = 0.61), which was further confirmed with a more powerful quantity I2 that established 0% of total variation across studies was due to heterogeneity instead of chance. Therefore, male and female patients are burdened by TSC at equal rates and any variance of the M:F ratio is due to statistical chance alone.
Patient ages in the whole cohort (as of October 2020) ranges between < 0 (prenatal) and 75 years of age, with median ages of 17 and 20 years old between TSC1 and TSC2 groups, respectively. The age distribution between TSC1 and TSC2 groups is very similar, with median ages of first symptom detection in both groups being at < 1 year of age with interquartile ranges of 2.25 and 2 years respectively. The standard deviation of age at first inclusion was 1.91 in the TSC1 group, and 8.01 in the TSC2 group indicating a greater variety of detection age in the latter (Additional file 1: Table S1).
Prevalence of CHD
The overall aim of this investigation is to characterise the nature of the relationship between CHDs, NDDs and KDs in TSC1 and TSC2 patients. To establish the association between the three conditions, it is important to first investigate each manifestation at its individual organ level. Firstly, CHD prevalence in TSC1 and TSC2 patients in this cohort were determined, with 57% and 75% of patients respectively receiving a diagnosis of either single or multiple CR at some point during their lifetime, with no significant difference in rhabdomyoma prevalence between the two populations found (z = −1.56, p = 0.119) (Fig. 3A).
The location of rhabdomyoma within the heart was also compared between TSC1 and TSC2 patients to evaluate whether their genotype impacted the preferential location CR development within different heart structures. The position of CRs within the heart was categorised into five primary locations; interventricular septum (IVS), left ventricle (LV), right ventricle (RV), left atrium (LA) and right atrium (RA) (Fig. 3B). Cumulatively, it was found that 98% of all TSC patients with CR had at least one tumour located in either ventricle or the IVS (Fig. 3B). Despite a marked difference between the prevalence of CR in the ventricles versus the atria in all patients, no significant difference of CR prevalence in each structure was found when comparing TSC1 vs TSC2 hearts (Fig. 3B), implying that the tumours appear at the same rate in each heart structure no matter the genotype (IVS, z = −0.162, p = 0.873; LV, z = 0.695, p = 0.490; RV, z = 0.969, p = 0.332; LA, z = −1.02, p = 0.308; RA, z = 0.681, p = 0.497).
In addition to the structural location of CR, the number of individual tumours is also a considerable factor in the overall cardiovascular health of TSC patients. The proportion of individuals with multiple CRs as opposed to an isolated tumour stands at 86% of TSC1 patients and 89% of TSC2 patients with cardiac manifestations (Fig. 3C).
Overall, most patients with multiple CR had either two or three tumours, representing 43% and 65% of the patients in TSC1 and TSC2 cohorts, respectively (Fig. 3C). There was no significant difference between number of tumours observed between TSC1 and TSC2 patients; therefore, we hypothesise that the number of tumours follow the same distribution, no matter the genotype (1 tumour, z = 0.268, p = 0.787; 2–3 tumours, z = –1.10, p = 0.271; 4+ tumours, z = 1.01, p = 0.313). The diameter of the largest cardiac rhabdomyoma in TSC1 and TSC2 patients was also investigated and ranged from 5 mm up to 34 mm, with the majority (22%) ranging from 10 to 15mm (Fig. 3D). Both frequency distributions followed a normal Gaussian distribution as confirmed by a non-significant Shapiro-Wilk test result (TSC1, W = 0.84, df = 6, p = 0.099; TSC2, W = 0.94, df = 6, p = 0.61). The two distributions returned a z-test statistic of 1.59, and a p value of 0.11 indicating there was no significant difference between the largest CR diameter of TSC1 and TSC2 patients.
Whilst CRs are thought to be present as early as 20–30 weeks of gestation [17], in this cohort, only 43% of rhabdomyoma patients were diagnosed in utero (Fig. 3E). After birth and onset of tumour regression, once a CR has completely shrunk or remains stable in size, it is said to be ‘resolved’. An exponential graph of resolution of tumours over time (Fig. 3F) revealed an average CR resolution age of 6.8 years old, with an average resolution age of 6.7 years old in TSC1 patients, and 7.2 years old in TSC2 patients. However, no further tumour regression was observed in either TSC1 or TSC2 patients beyond 15.5 years of age, meaning that for approximately 16% of patients, CR persisted into adulthood.
Prevalence of NDD
In addition to structural abnormalities within the brain, often seizures and delayed developmental milestones will become apparent during infancy and childhood, potentially leading to a diagnosis of some form of the neurodevelopmental disorder (NDD) which lie under the umbrella term of TSC-associated neuropsychiatric disorders (TAND). The areas of cerebral cortical dysplasia in the form of cortical tubers (CT) were prevalent in 42% and 80% of TSC1 and TSC2 patients, respectively (Fig. 4A). By employing a Z-score test for two population proportions, this marked difference between the two groups was deemed statistically significant (Z = −3.0, df = 1, p = 0.0026). Subependymal nodules (SEN) appear deeper within the brain, affecting 68% and 78% of TSC1 and 2 patients in their respective groups, and subependymal giant cell astrocytoma (SEGA) were detected with even smaller prevalence at 32% and 26%, respectively (Fig. 4A). Neither of these lesion types differed significantly between TSC1 and TSC2 (TSC1, Z = 0.85, df = 1, p = 0.40; TSC2, Z = 0.49, df = 1, p = 0.62).
The presence of NDDs in the cohort was evaluated from results of regular TAND screening, and the comorbidity of epilepsy was also quantified. 75% of TSC1 patients and 81% of TSC2 patients experienced epilepsy in the form of a range of seizure types (Fig. 4B). Statistical analysis revealed that epilepsy occurs with equal prevalence between TSC1 and TSC2 groups (Z = −0.53, df = 1, p = 0.60). Most of the other TAND assessed such as motor disorders, anxiety, depression and global developmental delay (GDD) also showed no significant difference in prevalence between TSC1 and TSC2 patient groups.
Autism spectrum disorder (ASD) and intellectual disability (ID) are thought to affect around half of all TSC patients; however, within this cohort, the ASD prevalence ranged from 14 to 25% and ID prevalence from 25 to 36% (Fig. 4B), possibly a direct result of an observed assessment gap for TAND in the UK [40]. Furthermore, a significant difference was found between the presentation of communication disorders in TSC1 and TSC2 at a prevalence of 6% and 50%, respectively (Fig. 4B) (Z = −3.17, df = 1, p = 0.0015).
The comorbidity between different structural brain abnormalities and NDDs was explored in this cohort. Individual instances of each type of NDD were counted for each comorbidity with either CT, SEN, or SEGA and expressed as a percentage of all NDD diagnoses for each brain lesion type (Fig. 4C). Differences in proportions of separate NDDs were unremarkable when comparing by both lesion type or genotype, and appeared in almost equal proportions as the findings from Fig. 4B, most likely a result of the technicality that often brain malformations appear together, and thus, it is difficult to quantify TAND prevalence as a comorbidity of each type of lesion separately.
Finally, the age at which patients received a brain magnetic resonance imaging (MRI) screening which returned consistent with TSC-associated brain abnormalities was graphed for both TSC1 and TSC2 patient groups (Fig. 4D). Both graphs of brain lesion prevalence within the cohort follow a logarithmic curve, with a plateau indicating the final prevalence, settling at 79% prevalence in the TSC1 patient group and 92% in the TSC2 patient group. The mean age of brain lesion detection varied slightly between the groups, with TSC1 patients on average receiving a diagnosis at 2.5 months old, which was extended to 6.4 months old for TSC2 patients. Both models were assessed for normal Gaussian distribution with Shapiro-Wilk test, both of which demonstrated non-normal distribution (TSC1, W = 0.62, df = 30, p < 0.0001; TSC2, W = 0.86, df = 30, p = 0.0008. A two-tailed Wilcoxon matched-pairs signed-rank test was performed and the outcome revealed no significant difference between the brain abnormality detection pattern observed between TSC1 and TSC2 groups (W = 24, df = 30, p = 0.82).
Prevalence of KD
Unlike CRs and brain malformations, detection of kidney lesions is very rare at initial presentation [14]. A stark difference was observed between the prevalence of AMLs between TSC1 and TSC2 groups at 32% and 68%, respectively (Fig. 5A). Statistical quantification of this difference with the Z-score test for two population proportions confirmed statistical significance between these groups (Z = −2.9, df = 1, p = 0.0035). In contrast, this same difference is not observed in the case of renal cysts, where prevalence amongst TSC1 and TSC2 patients is almost equal at 27% and 29%, respectively (Fig. 6A), an insignificant difference at p < 0.05 (Z = −0.18, df = 1, p = 0.86).
The diameter of the largest AML in each patient was extracted from renal imaging data and categorised into small groups at 5mm increments to evaluate the distribution of tumour sizes amongst the cohort (Fig. 5B). Across all patients, an individuals’ largest observed AML measured 10 mm or smaller in 44% of cases, with an even split between tumours ≤5mm and 5–10mm in diameter. All four TSC1 patients with recorded AML sizes also fell within the 5–10mm category. The median largest AML size across patients was determined as 15 mm with an interquartile range of 25 mm.
Renal lesion localisation across both kidneys was investigated. TSC2 patients demonstrated a pronounced localisation within both kidneys concurrently (78%), where in contrast only a quarter of investigated TSC1 patients had lesions within both kidneys at any one time (Fig. 5C). Statistical testing with a Z-score test for two population proportions revealed that this difference was indeed statistically significant (Z = −3.0, df = 1, p = 0.0030). Furthermore, lesion presence had a significantly higher degree of localisation to the left kidney in TSC1 patients than TSC2 patients as confirmed with Fisher’s exact test (p = 0.026); however, this was not the case within the right kidney where instead lesions had an equal prevalence across both groups (p = 0.33).
Finally, the age at which patients received a USS or MRI screening which exhibited consistency with TSC-associated renal abnormalities was graphed for both TSC1 and TSC2 patient groups (Fig. 5D). The choice of imaging technique reflected clinical practise and pragmatic resource availability. Patients were frequently screened with ultrasound (USS), and MRI would be requested for confirmation of growing AML. Both time graphs of kidney lesion prevalence within the cohort follow a logarithmic curve, with a plateau indicating the final prevalence of all kidney lesions within each group, settling at 56% prevalence in the TSC1 patient group and 73% in the TSC2 patient group. The mean age of kidney lesion detection was identical between the groups, with both TSC1 and TSC2 patients on average receiving a diagnosis of renal involvement at 4.7 years of age. Both models were then assessed for normal Gaussian distribution using Shapiro-Wilk test, both of which demonstrated non-normal distribution (TSC1, W = 0.87, df = 161, p < 0.0001; TSC2, W = 0.87, df = 161, p < 0.0001). A two-tailed Wilcoxon matched-pairs signed-rank test was performed and the outcome revealed a significant difference between the kidney lesion detection pattern observed between TSC1 and TSC2 groups (W = 13019, df = 161, p < 0.0001). A median difference of 12.2% prevalence was observed between TSC1 and TSC2 groups at any one point in time, with the prevalence of kidney lesions within the TSC1 group consistently trailing behind those in their counterpart genotype throughout life.
Co-occurrence of CHD, NDD and KD
With organ system involvement of TSC having been characterised for this cohort on a separate basis, it is now crucial to consider the quintessence of TSC as a multisystem disorder where disease manifests simultaneously in parallel systems throughout the whole body. Firstly, the prevalence and co-occurrence of organ manifestations of TSC were calculated for CHD, NDD and KD and results were separated into either a combined association for both TSC1 and TSC2 patients cumulatively, as well as separate genotypes (Fig. 6A). With TSC1 and TSC2 patients combined, the prevalence of organ involvement stood at 70% of patients with a CHD, 88% with an NDD and 63% with renal disorders. Separately, only 57% of TSC1 patients exhibited some form of cardiac involvement, as opposed to 75% of TSC2 patients, and a further TSC1/2 variation of 42% and 71%, respectively, was observed in the case of renal involvement. The prevalence of NDDs remained consistent between both groups at 83% and 90%, respectively.
Relative co-occurrence of different organ system manifestations was determined for the group. A consistent pattern is clear across both TSC1 and TSC2 patients where the order of co-occurrences from least probable to most probable in Boolean operator terms is P(CHD ∩ KD), P(CHD ∩ NDD) and P(KD ∩ NDD) (Additional file 1: Tables S2 and S3). Therefore, the most frequently observed co-morbidity throughout the whole dataset is the presence of one or more NDD and co-occurring renal involvement.
However, static data only holds a limited degree of predictive power. A temporal variable was added to the comorbidity data which enabled ‘tracking’ of a patients’ disease trajectory throughout their lifetime. Probability trees of disease trajectory throughout organ systems were generated for both TSC1 patients (Fig. 6B) and TSC2 patients (Fig. 6C). Initial organ presentation is characterised at the base of each tree. From an initial presentation at the ‘starter organ’, disease trajectory branches out to different organs with variable transitional probabilities. The most probable disease trajectory out of all possible disease outcome scenarios calculated for TSC2 patients was CHD → NDD at 0.85, with a further 0.74 probability of renal involvement detected afterward (Fig. 6C). Within the TSC1 patient group, the most probable trajectory was CHD → NDD at a probability of 0.70, with no further organ involvement likely at 0.86 probability (Fig. 6C). Altogether, this figure provides disease trajectories and probability for 18 different possible disease outcomes across both TSC1 and TSC2 patient groups.