OUP user menu

Neuro‐epileptic determinants of autism spectrum disorders in tuberous sclerosis complex

Patrick F. Bolton, Rebecca J. Park, J. Nicholas P. Higgins, Paul D. Griffiths, Andrew Pickles
DOI: http://dx.doi.org/10.1093/brain/awf124 1247-1255 First published online: 1 June 2002

Summary

Tuberous sclerosis is one of the few established medical causes of autism spectrum disorder and is a unique neurogenetic model for testing theories about the brain basis of the syndrome. We conducted a retrospective case study of the neuro‐epileptic risk factors predisposing to autism spectrum disorder in individuals with tuberous sclerosis to test current neurobiological theories of autism spectrum disorder. We found that an autism spectrum disorder diagnosis was associated with the presence of cortical tubers in the temporal but not other lobes of the brain. Indeed, the presence of tubers in the temporal lobes appeared to be a necessary but not sufficient risk factor for the development of an autism spectrum disorder. However, contrary to the predictions of some theories, the location of tubers in specific regions of the temporal lobe, such as the superior temporal gyrus or the right temporal lobe, did not determine which individuals with temporal lobe tubers developed an autism spectrum disorder. Instead, outcome was associated with various indices of epileptic activity including evidence of temporal lobe epileptiform discharges on EEG, the age to onset of seizures in the first 3 years of life and a history of infantile spasms. The results indicated that individuals with tuberous sclerosis are at very high risk of developing an autism spectrum disorder when temporal lobe tubers are present and associated with temporal lobe epileptiform discharges and early‐onset, persistent spasm‐like seizures. These risk markers constitute useful clinical indicators of prognosis, but further research is required to identify the neurobiological mechanisms responsible for their association with outcome. Most especially, it will be important to test whether, as the findings suggest, there is a critical early stage of brain maturation during which temporal lobe epilepsy perturbs the development of brain systems that underpin ‘social intelligence’ and possibly other cognitive skills, thereby inducing an autism spectrum disorder.

  • Keywords: tuberous sclerosis; epilepsy; brain; neuroimaging; autism
  • Abbreviations: ASD = autism spectrum disorder; IQR = inter‐quantile range; OR = odds ratio; PDD = pervasive developmental disorder; TLT = temporal lobe tubers; TSC = tuberous sclerosis complex

Introduction

Tuberous sclerosis complex (TSC) is a genetic disorder with a prevalence in children of around 1 in 10 000–20 000 (Shepherd, 1999). Mutation in either of two genes gives rise to the condition [TSC1 located on chromosome 9q34 and TSC2, on chromosome 16p13.3 (Cheadle et al., 2000)]. Their protein products, termed hamartin and tuberin respectively, appear to act as tumour suppressors (Cheadle et al., 2000). Mutation in one copy of either gene does not seem to give rise to any marked abnormalities, but somatic mutation events during embryogenesis are thought to knock out the remaining normal copy allele and produce cell lines that divide and migrate abnormally. The resulting clonal abnormalities produce developmental anomalies in many organs. The principal brain abnormalities are subependymal nodules and cortical tubers.

Cortical tubers vary in number and location, but most individuals with TSC have several (Shepherd et al., 1995). They can act as foci for epileptic discharges, producing multiple types of difficult to control, often multi‐focal, seizures. Epilepsy commonly begins in infancy as infantile spasms and/or partial seizures. Seizure onset becomes earlier as the number of cortical tubers increases. It has been suggested that the spasm‐like seizures (often denoted as atypical infantile spasms) represent a form of (possibly multi‐focal) focal epilepsy with secondary generalization in the immature, myelinating brain (Curatolo et al., 2001), but the mechanisms that give rise to infantile spasms remain poorly understood (Chugani, 1994).

Epidemiological studies (Harrison and Bolton, 1997) have shown that ∼50–60% of individuals with TSC have mental retardation and 43–86% an autistic‐like pervasive developmental disorder (PDD)—referred to as an autism spectrum disorder in this paper. Similarly, epidemiological studies of children with autism have reported that TSC occurs on average in ∼1% of cases (Harrison and Bolton, 1997). Either way, therefore, there is a strong association between TSC and autism spectrum disorders (Harrison and Bolton, 1997). The basis for this association is not well understood, although elucidation of the mechanisms would throw light on the brain basis of idiopathic autism where there is strong evidence for a genetically determined neurobiological abnormality, but little understanding of its nature (Bailey et al., 1996).

Autistic spectrum disorders are characterized by a complex mix of deficits in the development of social, communication and play skills. What kind of brain abnormalities might give rise to this multi‐faceted syndrome? One of the more recent and influential theories proposes that the social deficits stem from an inability to read other people’s minds (Baron‐Cohen, 1995). This ‘mind blindness’ is hypothesized to result from developmental abnormalities in a putative neural network involving the orbitofrontal cortex, temporal lobes and amygdala. The network is thought to underpin the basic skills required for the development of ‘social intelligence’ or mind reading ability (Brothers and Ring, 1993). One of the key functions of the temporal lobes within the neural network is their role in processing information from faces such as lip‐reading (Haxby et al., 2000), the direction of other people’s eye gaze, the focus of another person’s attention and their emotional expressions (Perrett et al., 1985; Allison et al., 2000). In support of this theory, a variety of abnormalities in processing information from faces have been identified in children with autism (Leekam et al., 1997; Schultz et al., 2000). Moreover, neuroimaging and neuropathological studies have revealed structural abnormalities in the temporal lobes (Bauman and Kemper, 1994; Bailey et al., 1998; Saitoh et al., 2001), as well as abnormal activation patterns in the fusiform gyrus and amygdala during face processing tasks (Baron‐Cohen et al., 1999; Schultz et al., 2000).

Various findings suggest that there may be a critical period in brain maturation when the foundations of ‘social intelligence’ are established. During this critical period, extreme forms of deprivation or insults to the neural network underpinning social intelligence can lead to pronounced autistic‐like abnormalities in social behaviour and long‐lasting deficits in social cognition. For example, Romanian orphans who were exposed to extreme deprivation during early infancy and prior to adoption have been reported to later exhibit quasi autistic‐like symptoms (Rutter et al., 1999). Similarly, medial temporal lobe ablation in primates result in autistic‐like impairments in social function, especially if the lesions are made in the neonatal period rather than later in development (Bachevalier and Merjanian, 1994). Also, children with congenital cataracts that were removed in infancy/early childhood nevertheless continue to show persistent and specific impairments on configural face recognition tests of the kind reported in children with autism, despite years of corrected normal vision (Le Grand et al., 2001).

Could it be that the abnormalities in brain development found in tuberous sclerosis disrupt the functional development of the neural networks supporting ‘social intelligence’ and related cognitive abilities during this critical period of early brain maturation and thereby induce an autism spectrum disorder?

Several reports have indicated that the children with TSC who get infantile spasms are most at risk for developing an autism spectrum disorder (Gutierrez et al., 1998). However, the findings also indicate that infantile spasms are neither a necessary nor a sufficient cause of autism spectrum disorders in TSC (Hunt and Dennis, 1987). Rather, the spasms appear simply to index the risk for the disorder, suggesting that the neurobiological mechanisms that lead to infantile spasms may also give rise, in part, to autism spectrum disorders. EEG studies have indicated that the risk for autism spectrum disorders in children with infantile spasms is increased when an epileptogenic temporal lobe focus is evident (Riikonen and Amnell, 1981). In addition, PET studies in children with infantile spasms have indicated that bilateral temporal lobe hypometabolism is predictive of an autistic outcome (Chugani et al., 1996). Our early findings (Bolton and Griffiths, 1997) indicated that, in children with TSC, the presence of tubers in the temporal lobes was associated with an increased risk for autism spectrum disorders. Others have subsequently shown that temporal lobe tubers in individuals with TSC and autism spectrum disorders are associated with auditory evoked response abnormalities (Seri et al., 1999).

Taken together, the findings raise the possibility that the increased risk for autism spectrum disorders in children with TSC may stem from early disruption to the functional development of the neural systems that support social cognitive representations within the temporal lobes. The anomalies in functional development could arise in two ways. First, the abnormalities in neuronal migration and differentiation that characterize TSC and give rise to temporal lobe cortical tubers could interfere with the proper development of key structures and connections within the temporal lobes. Current neuropsychological theories and data make two testable predictions. Baron‐Cohen’s ‘mind blindness’ theory proposes that abnormality in the ability to detect the direction of eye gaze is a core deficit that leads to the development of autism. Single‐cell recording and functional neuroimaging studies have shown that the direction of eye gaze is principally analysed by neurones in the superior temporal gyrus (Perrett et al., 1985; Allison et al., 2000). As such, the risk for autism spectrum disorders in TSC should be dependent on the presence of tubers in the superior temporal gyrus. In addition, some investigators have proposed that the risk for autism spectrum disorders is dependent on the presence of right hemisphere involvement (Taylor et al., 1999). This potentially ties in with the findings from numerous studies demonstrating a right hemisphere bias in facial information processing.

Secondly, epileptiform discharges associated with the presence of temporal lobe tubers could perturb the development of connections within these same structures or the proper emergence of social cognitive representations. Clearly, if tuber‐related temporal lobe electrophysiological abnormalities are primarily responsible, then patterning to the distribution of tubers within the temporal lobes might not be seen. Rather, the risk should be related to indices of epileptic activity such as the presence of epileptiform discharges on EEG and the duration and severity of seizures. Moreover, these electrophysiological abnormalities would only have an adverse effect during a sensitive period of development, because temporal lobe epilepsy beginning in middle childhood does not carry a risk for the development of autism spectrum disorders. The period of greatest risk would be in the first year of life when key social behaviours, such as attachment behaviours, joint attention and pointing first emerge and when parents often report the onset of autistic symptoms. The period of risk would end by 3 years of age, after which time autism is hardly ever reported to develop. This study aimed to test these propositions.

Methods

Samples

The sample comprised a consecutive series of clinic cases from our original report (n = 19) (Bolton and Griffiths, 1997) and cases recruited from new referrals to the clinic or through an ongoing epidemiological study of children with TSC in the eastern UK (n = 53). Subjects met established diagnostic criteria for TSC (Roach et al., 1998). Cases were excluded (n = 12) if a low mental age precluded confident diagnosis of an autism spectrum disorder. There were 60 diagnosed cases in the sample.

Assessment

Estimates of the children’s abilities were made using standardized cognitive tests appropriate for age and/or ability (Mullens scales, the Wechsler scales, the Ravens Coloured Matrices and the British Picture Vocabulary Scale) and adaptive behaviour assessed with the Vineland Adaptive Behaviour Scales (Survey form) or, when assessment was not feasible, from previous clinical evaluations. We classified individuals into groups corresponding to 15 intelligence quotient (IQ) point bands (0 = >130, 1 = 115–129, 2 = 100–114, 3 = 85–99, 4 = 70–84, 5 = 55–69, 6 = 40–55, 7 = 40–25, 8 = <25).

Clinical and research interviews with parents were used to screen for autism spectrum disorders. Children with a possible autism spectrum disorder were assessed using the autism Diagnostic Interview‐Revised (ADI‐R) and Autism Diagnostic Observation Schedule‐Generic (ADOS‐G) (Lord et al., 2000). Inter‐rater agreement for ADOS‐G diagnoses of autism spectrum disorders was excellent (κ = 0.8). Developmental histories, ADI‐R/ADOS‐G algorithm scores and cognitive test results were reviewed by two psychiatrists and diagnoses were assigned according to ICD‐10 (International classification of diseases–10th Edition) criteria and without reference to the epilepsy or brain scan findings. Children with autism, atypical autism, Asperger’s syndrome or pervasive developmental disorder not otherwise specified (PDD‐NOS), were deemed to have an autism spectrum disorder (ASD).

Case vignettes summarizing parents’ accounts of the children’s seizure histories were compiled as were vignettes describing details from contemporaneous medical reports. These vignettes were separately and independently rated by two psychiatrists, who were blind to the patient’s identity, clinical diagnosis and brain scan findings. The agreement between parent reports and the contemporaneous case notes for the type and age of onset of epilepsy was excellent (κ = 0.8 for a diagnosis of infantile spasms; Spearman’s ρ for age of onset of seizures = 0.79). All available EEG reports were obtained on 57 cases. Several individuals had more than one EEG investigation (31 out of 57 cases) and so each report was examined separately and coded for the presence of a possible/probable temporal lobe focus (defined as spike/sharp wave discharges possibly/probably arising from temporal lobe regions). For example, if discharges were reported as arising from the temporal lobes, this was recorded as a probable temporal lobe focus. If, however, discharges were reported in fronto‐temporal or occipito‐temporal regions, a possible temporal lobe focus was recorded. Brain scan films were obtained, but when none was available, MRI scans were acquired whenever possible, (n = 13) on a 1.5 T super‐conducting system (Signa, General Electric Medical Systems, Milwaukee, WI, USA) using standard Dual Echo/Fast Spin Echo and Fast FLAIR sequences. Neuroimaging data were available for 54 cases (nine CT scans and 45 MRI scans). Two neuroradiologists (P.D.G and N.H.) independently rated the scans, blind to clinical details, for the presence and location of tubers according to a pre‐specified coding scheme based on a detailed cross‐sectional anatomical atlas (Duvernoy, 1999). Inter‐rater agreement for temporal lobe involvement was excellent (κ = 0.81) and the correlations between raters for the number of tubers in temporal (ρ = 0.90) regions, as well as the whole brain (ρ = 0. 93) were very high. In view of the poor sensitivity of CT scans in identifying the abnormalities in tubers sclerosis, tubers were rated as definitely present, possibly absent (if CT scan was negative) and definitely absent (if MRI was negative).

We confirmed the validity of the EEG data by examining its relationship with brain scan data. A possible/probable temporal lobe EEG focus was associated with the presence of temporal lobe tubers (Fisher’s exact test, P = 0.001). Combined imaging/EEG data were available for 53 individuals and full clinical/imaging data for 52 individuals (details on early seizures in one adopted child were unknown).

Data analysis and ethical approval

Data were analysed in STATA6 (StataCorp, 1999) using Cox’s survival analysis and exact methods including Fisher’s exact and Mann–Whitney U tests. Exact logistic regression analyses were performed using LogXact4 (Mehta and Patel, 2000). All significance levels represent two‐tailed tests. Ethical approval for the study was obtained from the Cambridge local research ethics committee and informed written consent obtained from parents and, whenever possible, children.

Results

Autism spectrum disorders were present in 19 out of 53 individuals (14 autism, four atypical autism and one PDD‐NOS). Three cases with a PDD unspecified diagnosis (signifying that past history was inconsistent with current observations) were deemed not to have an autism spectrum disorder. Autism spectrum disorders were strongly associated with the severity of intellectual impairment (Z = –4.67; P < 0.0001).

The number of tubers ranged from 0 to 29 and was significantly correlated with rank intelligence quotient (IQ) (Spearman’s ρ = 0.34, P = 0.01). There was a non‐significant tendency for individuals with an autism spectrum diagnosis to have more tubers (ASD+ median = 12, IQR = 8; ASD– median = 7.5, IQR = 10; Mann–Whitney U‐test: Z = –1.35, P = 0.2]. The presence of temporal lobe tubers (TLT) was associated with an increased number of tubers in non‐temporal lobe regions (median number of non‐temporal tubers if TLT– = 3, IQR 7; median number of non‐temporal tubers if TLT+ = 9, IQR = 11; Mann–Whitney U‐test: P = 0.002).

The presence of cortical tubers in the temporal lobes was associated with an autism spectrum disorder outcome in both the original [odds ratio (OR) = 17.66, confidence interval (CI) = 2.00–∞, exact mid P = 0.004] and the new sample (OR = 5.07, CI = 0.8–58.08, exact mid P = 0.05). In the whole sample, two of 19 individuals with an autism spectrum disorder were reported not to have temporal lobe tubers. One had a CT scan reported to be negative, but a probable temporal lobe focus on EEG. The other had a standard clinical MRI scan but without a FLAIR sequence and an EEG recording indicating a possible temporal lobe focus. (Subsequent review of the MRI revealed a possible, previously undetected, temporal lobe tuber.) By contrast, 14 of the remaining 34 individuals without an autism spectrum disorder had temporal lobe tubers (Mann–Whitney U‐test: P = 0.0006 for the association between no/possible/definite temporal lobe tubers and an autism spectrum disorder diagnosis). Table 1 shows the results of an exact logistic regression analysis of MRI data to test the specificity of the association between temporal tuber location and clinical diagnosis. Only the presence of tubers in the temporal lobes was associated with an ASD outcome. In addition, exact logistic regression showed that the association remained strong after taking into account the small differences in tuber number, either by stratifying the analysis by tuber number (three approximately equal‐sized strata groups with a few, some and many tubers: OR for temporal lobe tuber = 10.93, CI = 1.73–128.91, P = 0.006) or by including total tuber count as a covariate in the regression model (OR for temporal lobe tuber = 16.47, CI = 1.57–903.61, P = 0.011).

View this table:
Table 1

The likelihood of developing an autism spectrum disorder according to the location of cortical tubers

Lobar location oftubersOdds ratio95% confidence intervalExact P (Scores test)
Frontal1.050.068–∞ns
Parietal0.640.031–12.06ns
Occipital1.110.13–9.57ns
Temporal9.91.05–504.700.023

ns = not significant (P > 0.1).

Our results indicated that the presence of a temporal lobe tuber was most probably a necessary but not sufficient risk factor for the development of an autism spectrum disorder. We therefore conducted further analyses to test whether the extent and distribution of temporal lobe structural abnormalities determined outcome.

Table 2 summarizes details of the location and number of temporal lobe tubers according to the presence or absence of an autism spectrum disorder diagnosis in those children with temporal lobe tubers. There was no association between an autism spectrum disorder diagnosis and the number of temporal tubers (Mann–Whitney U‐test: P = 0.8) or the number of tubers overall (Mann–Whitney U‐test: P = 0.2). Indeed, in both instances the children with an autism spectrum disorder and temporal lobe tubers had fewer tubers than the children with temporal tubers but no autism spectrum disorder. There were insufficient numbers of individuals (four with unilateral right and four with unilateral left sided tubers) to test whether hemispheric side of involvement was associated with outcome. However, the large majority (see Table 2) of children without an autism spectrum disorder had left or right temporal lobe tubers and nine out of 14 had bilateral temporal tubers. It was also evident that the children without autism spectrum disorders often had abnormalities in the three principal gyri of the temporal lobes and frequent involvement of the superior temporal gyrus (see Table 2). It was clear, therefore, that neither the extent of involvement nor the position of tubers in key, theoretically based anatomical locations within the temporal lobes could explain why some children with temporal tubers developed autism spectrum disorders whilst others did not.

View this table:
Table 2

The location and number of tubers in individuals with temporal lobe tubers according to autism spectrum disorder diagnosis

Tuber characteristicsAutism spectrum disorder
YesNo
Number of cases1614
Total number of cortical tubers
 Median1117
 (IQR)(9)(14)
 Range2–281–29
Number of temporal tubers
 Median23
 (IQR)(3)(3)
 Range1–81–6
Hemispheric side of tubers (n cases)
 Left1511
 Right1412
Location of temporal tubers (n cases)
 STG1310
 MTG910
 ITG129

STG = superior temporal gyrus; MTG = middle temporal gyrus; ITG = inferior temporal gyrus.

We turned, therefore, to examine whether indices of electrophysiological differences were related to outcome. The rate of autism spectrum disorders in those without a temporal lobe EEG focus was two out of 21 compared with five out of 14 and 12 out of 18 in those with a possible or probable EEG focus, respectively (Fisher’s exact text: P = 0.001). The small numbers of cases with clearly unilateral temporal lobe discharges precluded any meaningful test of the potential influence of hemispheric lateralization of discharges on outcome.

Temporal lobe epileptiform discharges on EEG were also associated with an autism spectrum disorder diagnosis in the individuals with temporal lobe tubers (OR probable temporal lobe discharges = 9.28, CI = 0.77–180.99; OR possible temporal lobe discharges = 1.00, CI = 0.087–15.65; exact mid P for probable and possible discharges = 0.02). The results therefore indicated that although temporal lobe epileptiform activity marked the presence of temporal tubers; the occurrence of probable temporal lobe epileptiform discharges also provided further predictive information about outcome. Accordingly, we constructed a variable that indexed the likely presence of tuber‐related temporal lobe epileptiform discharges (very low  = no temporal tuber on MRI and no temporal lobe EEG focus; possible = temporal lobe tuber present/possible EEG focus; high = temporal lobe tuber present and probable EEG focus). Exact logistic regression confirmed that this measure was a significantly better predictor of the liability for an autism spectrum disorder than the simple presence of a temporal tuber (exact Scores test = 10.8, exact mid P = 0.002). In the presence of the combined EEG/brain scan measure, the presence of a temporal lobe tuber on scan was no longer associated with outcome (exact Scores test = 0.34; exact mid P = 0.5)

Our theoretical considerations predicted that if temporal lobe electrophysiological abnormalities were a risk factor for the developmental of autism spectrum disorders, their effects should be developmentally constrained. That is, the adverse outcome should be dependent on early age of onset of seizures, as late onset temporal lobe electrophysiological disturbances are not known to carry a risk for autism spectrum disorders. Indeed, as autism by definition has an onset within the first 3 years of life, our theory predicted that age of seizure onset during this period should be associated with outcome. Survival analysis of age to onset of seizures (Cox’s regression; seizure onset censored at 36 months for eight individuals) showed that seizures started increasingly early as the number of tubers increased (tuber number categorized into three groups corresponding to having a few, some, or many tubers; log rank test χ2 = 12.77; P = 0.002). Importantly though, age of onset of seizures was not associated with the presence of temporal lobe tubers (log rank test χ2 = 0.38; P = 0.5) or the presence of temporal lobe epileptiform discharges (log rank test χ2 = 0.9; P = 0.6) following stratification by tuber number. In keeping with our predictions, however, children with autism spectrum disorders had a significantly earlier age of seizure onset even after stratification by tuber number (hazard ratio = 2.03, CI = 1.02–4.06; log rank test χ2 = 4.61; P = 0.03).

Table 3 summarizes the details of age to seizure onset (censored at 36 months) according to clinical diagnosis and the evidence for a temporal lobe epileptiform focus. The table shows how the likelihood of an autism spectrum disorder increased as the evidence for a temporal lobe epileptiform focus on EEG mounted. In addition, it shows that all children with an autism spectrum disorder had very early onset epilepsy, with seizures starting before 15 months of age in all 18 children (early epilepsy data were missing in one adopted child). Exact logistic regression (evidence for temporal lobe epileptiform focus entered as dummy variable) confirmed that, jointly, evidence for a temporal lobe epileptiform focus (OR probable focus = 33.42, CI = 4.19–∞, exact P = 0.0003; OR possible focus = 3.71, CI = 0.53–∞, exact P = 0.2) and age to seizure onset (months) within the first 3 years of life (OR = 1.17, CI = 1.03–1.4, exact P = 0.01) were predictive of developmental outcome.

View this table:
Table 3

Age of onset of seizures in first 3 years according to evidence for a temporal lobe epileptiform focus and an autism spectrum disorder diagnosis

Evidence for a temporallobe epileptiform focusAutism spectrum disorder
NoYes
Little
 Number of cases130
 Mean age onset15.85 months
 SD13.6
 Age range0.5–36 months
Some
 Number of cases197
 Mean age onset15.47 months7 months
 SD13.873.83
 Age range1–36 months3–14 months
Substantial
 Number of cases211
 Mean age onset19 months5.64 months
 SD1.413.64
 Age range18–20 months2–13 months

As it was still possible that age to seizure onset and the presence of temporal lobe discharges on EEG were simply markers of the extent of underlying temporal lobe structural abnormality, we tested whether they continued to predict outcome after stratifying the regression analyses by tuber number (three approximately equal‐sized strata representing individuals with a few, some and many tubers) and after including tuber count as a covariate in the regression model. These analyses were performed in the sample as a whole and in the subset of cases with temporal lobe involvement. The results confirmed that both age of seizure onset in the first 3 years of life and evidence of a temporal lobe EEG focus were still independently associated with developmental outcome, after taking account of total and temporal lobe tuber number. This was so both in the sample as a whole and in the subset of cases with temporal lobe involvement (results available from P.F.B).

In the whole sample, exact tests also confirmed that both predictors remained associated with outcome following stratification by IQ (three approximately equal‐sized strata corresponding to no, little, moderate and marked intellectual impairment; exact Scores test for both predictors = 10.52, exact mid P = 0.0062) and following the inclusion of IQ as a covariate (exact Scores test for both predictors = 9.11, exact mid P = 0.018). The analyses thus showed that the IQ differences between children with and without autism spectrum disorders were not confounding our results. The same analyses in the subset of cases with temporal lobe involvement led to the same conclusions (test of association with predictors and outcome following stratification by IQ: exact Scores test for both predictors = 6.82, mid P = 0.021; following inclusion of IQ as a covariate: exact Scores test for both predictors = 6.55, mid P = 0.03).

Infantile spasms were also associated with an autism spectrum disorder diagnosis (four out of 29 without spasms had an autism spectrum disorder compared with 14 out of 23 with spasms; Fisher’s exact test: P = 0.001). However, in the presence of the two neuro‐epileptic predictors, the association between a history of spasms and developmental outcome was no longer significant (exact Scores test = 3.3, mid P = 0.1), whereas the association with the neuro‐epileptic predictors remained strong (exact Scores test for both predictors = 19.62, mid P < 0.0001).

Discussion

This was a retrospective case study with all the attendant shortcomings. We attempted to minimize these by careful attention to methods and ensuring that our measures were reliable, valid and made independently of each other. Our principal findings were clear and strongly significant and identified some important neuro‐epileptic correlates of an autism spectrum disorder in individuals with tuberous sclerosis. These will be useful in predicting prognosis early in the natural history of the disease and well before the frank manifestation of autistic spectrum disorders are ordinarily detectable.

The findings confirm and extend our earlier results showing an association between the presence of temporal lobe tubers and autism spectrum disorder in individuals with tuberous sclerosis (Bolton and Griffiths, 1997) and indicate that the association is specific for the location of tubers in the temporal cortex. However, although tubers within the temporal lobes appeared to be a necessary condition for the emergence of an autism spectrum disorder, their presence was not sufficient to produce the syndrome. We therefore investigated whether the extent or location of the temporal lobe structural abnormalities was related to outcome. Neither the number of tubers within the temporal lobes nor their presence in the superior temporal gyrus predicted outcome. Similarly, outcome was not associated with the hemispheric side of abnormality. Instead, outcome was associated with various indices of epileptic activity, including the presence of a temporal lobe epileptiform focus on EEG and the age of onset of seizures within the first 3 years of life.

In addition and in keeping with previous studies, a history of infantile spasms was found to be associated with the development of an autism spectrum disorder (Hunt and Dennis, 1987). However, it was also evident that by no means all children with an autism spectrum disorder had a history of infantile spasms and, conversely, that several individuals with a history of spasms did not develop an autism spectrum disorder. This suggests that spasms were a marker of the liability for autism spectrum disorders rather than a risk factor. In keeping with this view, the association between autism spectrum disorders and a history of spasms was no longer evident when age to seizure onset and evidence for a temporal lobe epileptiform focus were included in an exact logistic regression model.

Intellectual impairments were an associated feature of autism spectrum disorder in the TSC children, as is the case in children with idiopathic forms of autism spectrum disorder. However, temporal lobe epileptiform discharges and age to onset of seizures in the first 3 years of life still predicted an autism spectrum disorder outcome after taking account of IQ differences in the multivariate analyses. Thus, our findings could not be accounted for by any diagnostic mis‐specification, resulting from the difficulties of assessing children with severe handicap. The reasons for the associated intellectual handicaps remains unclear, but it may reflect the operation of distinct but correlated pathophysiological processes with, for example, intellectual impairments relating to the total number of tubers and the development of infantile spasms (F. O’Callaghan, C. Joinson, P. Bolton, M. Noakes, D. Presdee and S. Renowden, unpublished results).

Recent neuroimaging findings in tuberous sclerosis complex suggest that neuro‐anatomical abnormalities may be widespread and involve grey and white matter regions outside the location of cortical tubers (Ridler et al., 2001). As such, we cannot rule out the possibility in this study that some unmeasured structural abnormality within the temporal lobes underlay the increased risk for autism spectrum disorders in the children with TSC. Although this remains a possibility, several findings were more consistent with the notion that aspects of temporal lobe epilepsy determined outcome. First, amongst those with temporal tubers, the numbers of tubers within the temporal lobes and the numbers of tubers in other brain regions were, if anything, less in the individuals with autism spectrum disorders than they were in those without this diagnosis. Moreover, our results indicated that indices of early epileptiform activity were still predictive of outcome following analyses that attempted to take account of the degree of structural abnormality, albeit by using tuber count to index the extent of anomaly. As the current evidence indicates that the extent of more widespread structural abnormality is correlated with the number of tubers (Ridler et al., 2001), our results are not consistent with the notion that the extent of temporal lobe structural abnormality determines an autism spectrum disorder outcome. Secondly, those individuals with temporal lobe tubers that did not develop an autism spectrum disorder had quite extensive involvement of the temporal lobes and this frequently included areas that have been implicated in the pathogenesis of autism, such as the superior temporal gyrus and the right hemisphere. Thus, it appears that there may be considerable plasticity in the neural encoding of social cognitive representations within the temporal lobes, and structural abnormalities per se need not disrupt their proper or at least partial emergence. Thirdly, developmental outcome was clearly associated with several indices of epilepsy such as the age of onset and type of epilepsy as well as the presence of temporal lobe epileptiform discharges on EEG. Furthermore, the association between the risk of an autism spectrum disorder and the age of onset of seizures was not simply a reflection of a correlation between age of seizure onset and the presence of temporal lobe abnormality or epilepsy. Rather, individuals with possible or probable temporal lobe epilepsy were only at increased risk for an autism spectrum disorder if their seizures started early. In addition, early seizure onset only seemed to be associated with an increased risk for autism spectrum disorders if there was also evidence for a temporal lobe focus. It seems likely, therefore, that tuber‐related temporal lobe electrophysiological disturbances somehow disrupted the establishment of social cognitive representations and this led to the developmental of autism spectrum disorders. Even so, future research will have to employ more advanced image analysis techniques to investigate the relationship between temporal lobe structural abnormalities and developmental outcome in more depth.

A preliminary report linking better epilepsy control to improved developmental outcome following treatment with vigabatrin in children with TSC (Jambaque et al., 2000) lends further support to the notion that epilepsy may shape developmental outcome. However, it must be stressed that the demonstration of an association between epilepsy control and outcome does not prove that the seizures were causally involved. Nevertheless, the study by Jambaque and colleagues highlights the fact that a clarification of the pathophysiological processes may have important therapeutic implications. Together the findings suggest that prompt, effective treatment of seizures may lead to an improved developmental outcome. However, it is important to emphasize that the seizures that develop in children with tuberous sclerosis are often extremely resistant to treatment. There has been a lot of interest in reports suggesting that Vigabatrin is a more effective and safer treatment for infantile spasms, especially in children with TSC (Chiron et al., 1997; Vigevano and Cilio, 1997; Hancock and Osborne, 1999). However, it has recently been found that permanent visual field defects may develop as a complication of Vigabatrin treatment (Hardus et al., 2001) and that Vigabatrin may not be as effective as originally thought (Hancock et al., 2001; Riikonen, 2001). As such, there is more uncertainty over the optimum approach to the treatment of early seizures in TSC. In some instances, brain surgery to remove tubers and control seizures has been employed, with some success. However, the fact that the seizures in TSC are frequently multi‐focal in origin means that identification of a principal major epileptic focus is often difficult. Moreover, outcome is more uncertain as new foci may emerge. Until the development of new and better drugs, it will be difficult to test rigorously whether more effective treatment of early onset seizures reduces the likelihood of an autism spectrum disorder outcome.

If tuber‐related temporal lobe epileptiform discharges/epilepsy interfere with the structural and/or functional development of the brain in some way, what mechanisms might be involved? Various studies have shown that properly patterned neural activity is necessary for normal structural and functional brain development (Sur and Leamey, 2001), so it is possible that the discharges/epilepsy disrupt the synchronization and patterning of neural activity, thereby leading to structural and functional deficits.

Another possibility is that the epileptiform discharges in the temporal lobes indexed more widespread abnormality in brain electrophysiology, which occurred, for example, during sleep. Abnormalities in the architecture of sleep electrophysiology have been implicated in the aetiology of the acquired epileptic aphasia of childhood known as Landau–Kleffner syndrome. In this syndrome, fluctuating aphasia is associated with seizures and temporal lobe epileptiform discharges, especially during sleep (Mantovani, 2000). Typically, the seizures begin after the age of 2–3 years and the disorder is not usually characterized by autistic‐like social deficits, although it has been speculated that very early onset forms of the syndrome may give rise to autism spectrum disorders (Mantovani, 2000). The duration of electrical status epilepticus in slow wave sleep is reported to be associated with outcome, suggesting that the sleep discharges may have an adverse effect on development (Robinson et al., 2001). Although nocturnal seizures are very common in children with TSC, there are no data relating sleep EEG investigations to developmental course.

It is also noteworthy that both the age at which seizures became clearly overt and the type of the seizure disorder were associated with outcome. These findings suggest that the factors that determine whether and how temporal lobe epileptiform discharges generalize may be important in shaping outcome. This potentially implicates factors such as the frequency and severity of discharges, the developmental status of brain myelination and the electrophysiological pathways involved during secondary generalization.

The study findings are of interest not only for demonstrating a possible link between temporal lobe epilepsy and the development of an autism spectrum disorder in children with TSC, but also in indicating that the developmental timing of seizures may be important in determining outcome. Our results provide further support to the idea that there may be a critical period in brain development when the foundations of social intelligence, and possibly other cognitive skills involved in the pathophysiology of autism spectrum disorders, are established. It is notable that the key risk period spanned the first year of life, as this corresponds to the peak period of dendritic arborization and synaptogenesis in the developing auditory and visual cortex (Huttenlocher, 1994), as well as the phase when early social behaviours first emerge and become established. It is also interesting that the risk was reduced if seizures started in the second year of life. This is consistent with evidence indicating that those individuals’ with temporal lobe or other epilepsy syndromes (e.g. Landau–Kleffner syndrome) are not at increased risk for developing autism spectrum disorders when seizures begin after 18–24 months of age. It is also consistent with parent reports of the onset of symptoms in idiopathic autism, as onset is rare after the age of 2 years.

In conclusion, the findings reported here provide some important new leads to further our understanding of the neurobiological basis of autism spectrum disorders. The questions raised will need to be explored in prospective longitudinal and intervention studies, using new and advanced structural and functional imaging techniques.

Acknowledgements

We wish to thank to Rowan Crawley and Petrus De Vries for their help in data collection, colleagues in Oxford and at Great Ormond Street Hospital (especially Brian Neville) for providing clinical details and the UK Tuberous Sclerosis Association for help identifying families. Most of all we wish to thank the children with TSC and their families for their unstinting willingness to assist us in this work. This research was supported by grants to Patrick Bolton from the Anglia and Oxford NHS Research and Development Fund, and from the UK Tuberous Sclerosis Association.

References

View Abstract