Box 1. Illustrative clinical vignette
A 6-month-old boy is referred to a pediatric neurology clinic with infantile spasms, and genetic testing confirms a pathogenic TSC2 variant. His seizures are controlled within 2 years, but global developmental delay becomes increasingly apparent. By age 3 years, he is diagnosed with autism spectrum disorder, and his parents struggle with his severe crying, aggression, self-injury, and disturbed sleep. His father and older brother are suspected to have tuberous sclerosis complex but have not been formally diagnosed. The brother, who has no history of seizures, is treated for learning difficulties and attention-deficit/hyperactivity disorder.
This scenario is common in tuberous sclerosis complex (TSC) clinics and reframes the current disease burden. TSC is a multisystem disorder caused by pathogenic variants in TSC1 or TSC2, which lead to constitutive activation of the mechanistic target of rapamycin (mTOR) pathway and the development of hamartomas in multiple organ systems [1]. Clinical care has traditionally focused on seizure control and surveillance of hamartomas in the brain, kidneys, lungs, and skin. Yet for this child, as for many others, seizures are controlled while cognitive and behavioral difficulties remain the principal challenges—areas that can be particularly difficult to address in busy clinics with limited multidisciplinary support.
These difficulties, termed TSC-associated neuropsychiatric disorders (TAND), affect up to 90% of individuals with TSC and are a leading contributor to reduced quality of life [2,3]. They are increasingly understood as core neurodevelopmental manifestations associated with disrupted mTOR signaling rather than solely as secondary consequences of seizures [4]. Nevertheless, fewer than 20% of affected individuals receive adequate evaluation, partly because the phenotype is broad and variable and because a shared clinical framework has not been consistently implemented [5].
This review traces the emergence and evolution of the TAND concept, examines the evidence supporting the natural clusters, and summarizes current knowledge of prevalence, developmental change across the lifespan, and major risk markers, including genotype, intellectual ability, and epilepsy. It then outlines the biological rationale for targeting mTOR signaling before returning to the vignette to illustrate how the 2023 consensus recommendations can be applied in practice.
The Historical Evolution of the TAND Concept
TSC was first described as a brain disorder in 1880, and Vogt later reframed it as a multisystem condition characterized by seizures, intellectual disability, and facial angiofibromas [5]. Intellectual disability was recognized early, but broader behavioral and psychiatric features, including autism-like traits, were not reported until the 1930s [6]. The first diagnostic criteria, established in 1967 [7], focused on physical findings and omitted neuropsychiatric features. International consensus updates in the 1990s refined these criteria [8], but recommendations did not address neurodevelopmental features until 1999, when age-appropriate screening at diagnosis was first advised [9].
The TAND concept was introduced in 2012 after a neuropsychiatry panel noted that most neuropsychiatric manifestations were unrecognized and untreated and that the field lacked a shared vocabulary [10]. The panel proposed TAND as an umbrella term encompassing the full range of manifestations and organized them into six levels: behavioral, psychiatric, intellectual, academic, neuropsychological, and psychosocial [10]. Annual screening was recommended for all individuals with TSC [10]. This structure helped clinicians assess the difficulties systematically and reduced the likelihood that important areas would be overlooked [10].
In practice, the six-level framework had an important limitation: listing features within the six levels did not indicate which problems should be prioritized or where intervention should begin. Because TAND presentations vary widely among individuals and across development, clinicians often faced what Leclezio and de Vries [11] termed ‘treatment paralysis.’ Researchers therefore used data-driven methods to identify natural groupings of TAND features—clusters of manifestations that tend to co-occur [12,13]. First described in 2018 and subsequently replicated, these clusters shifted the focus from cataloguing isolated features to recognizing an individual's overall TAND profile as a basis for more personalized, cluster-informed care [12,13].
These advances were consolidated through the TANDem project, an international initiative launched in 2019 to narrow the gap between identification and treatment of TAND [14]. The 2021 revision of the international TSC recommendations added TAND-specific guidance, including use of the TAND checklist, early identification and treatment of manifestations, and family support [15]. This work culminated in the first international consensus recommendations devoted specifically to TAND, published in 2023 [14] and organized around the natural clusters. Table 1 summarizes the key milestones in the evolution of the TAND concept.
From Levels to Clusters: Identifying Natural Groupings of TAND
An initial feasibility study of 56 individuals from two countries showed that data-driven methods could extract clinically meaningful clusters from TAND checklist data and yielded six provisional clusters [12]. These clusters were reproduced in an independent sample of 85 individuals from seven countries, supporting their stability [16]. A larger analysis of 453 individuals across six international sites then refined and confirmed the structure, establishing the seven natural TAND clusters used today [13]. The term ‘natural’ refers to their empirical derivation rather than their status as diagnostic categories [13]. The 2021 analysis used 29 observable TAND checklist items—items reportable without specialist diagnosis or formal testing—to construct a correlation matrix and apply Ward hierarchical cluster analysis [13]. Reliability was evaluated with three methods: 1,000-fold bootstrapping, exploratory factor analysis, and Cronbach's alpha [13]. All three supported the same seven clusters: scholastic, neuropsychological, autism spectrum disorder (ASD)-like, dysregulated behavior, overactive/impulsive, mood/anxiety, and eat/sleep (Table 2) [13]. The eat/sleep cluster was newly identified in the larger sample and emerged as a distinct grouping rather than combining with other domains [13].
Although the seven clusters are distinct, they are not entirely independent [13]. For example, delayed or absent language initially fell within the dysregulated behavior cluster but was unstable there, appearing in only 3%–5% of bootstrap samples, and instead co-occurred more consistently with ASD-like and scholastic items [13]. It was therefore reassigned to the ASD-like cluster on statistical and clinical grounds [13]. Clinically, language delay in a child with TSC should not automatically be attributed to global developmental delay; it may indicate an emerging autism-like profile that warrants further evaluation [13]. Attention and hyperactivity provide another example. Overactivity and impulsivity formed their own cluster, whereas attentional difficulties grouped with neuropsychological items, suggesting that attention problems and hyperactivity in TSC may follow partly distinct pathways [13]. Cluster strength may also have practical implications. The scholastic cluster had the highest bootstrap support and internal consistency, indicating that its four items—reading, writing, spelling, and mathematics—were tightly linked [13]. Difficulty in one academic area should therefore prompt assessment of the others [13].
The clusters were derived solely from TAND checklist data and were not confirmed by clinical evaluation or formal neuropsychological assessment. External validation against expert clinical measures remains an important next step [13]. The dataset was also cross-sectional. Longitudinal changes in individual cluster profiles are being investigated, but the results of that work are not yet available [17]. Cluster-based scores have been used as outcomes in treatment trials [18], but whether organizing clinical care around the clusters improves patient outcomes remains unknown.
The Clinical and Biological Landscape of TAND
1. Clinical landscape
One of the largest sources of evidence on TAND is TuberOus SClerosis registry to increase disease Awareness (TOSCA), an international registry that included more than 2,000 individuals with TSC from over 30 countries [2]. Its scale confirmed that TAND affects approximately 90% of individuals across the lifespan and provided a broad view of how manifestations vary with age and genotype [2]. High levels of missing or unreported TAND data, even in expert centers, also underscored how often these features remain undocumented in routine care [2].
1) Lifespan pattern
A key TOSCA finding was that TAND does not simply disappear with age but changes in presentation [2]. Overactivity and impulsivity were common in childhood, whereas anxiety, depressed mood, and mood swings were more frequent in adults [2]. Other difficulties were relatively stable: sleep problems occurred in approximately 44% of both children and adults, severe aggression in 23%–27%, and self-injury was slightly more common in adults [2]. The mixed pattern argues against assuming that childhood difficulties will be outgrown and supports continued monitoring and annual screening for every individual with TSC [14].
2) Epilepsy
Epilepsy affects 80%–90% of individuals with TSC and is strongly associated with neurodevelopmental outcomes [19]. Pediatric-onset epilepsy is an independent risk factor for adverse neurodevelopmental outcomes in TSC, as it is in many other pediatric neurological disorders [19]. Detailed evidence comes from the North American TSC Natural History Database, a large longitudinal cohort in which epilepsy occurred in 88% of 1,657 individuals and was more frequent and began earlier among those with TSC2 variants [19]. Epilepsy, particularly onset before age two, was strongly associated with intellectual disability and autism, whereas anxiety and depression were not; age at seizure onset was more informative than seizure duration [19]. The investigators found no independent association of genotype with intellectual disability or autism and proposed that the higher rates of both conditions in TSC2 reflected the association of TSC2 with early-onset epilepsy rather than an independent genotype effect [19]. A limitation is that the study examined individual diagnoses rather than the broader TAND construct [19].
An earlier report from the larger international TOSCA registry described a similar genotype pattern [20]. Approximately 47% of individuals with a TSC2 variant developed infantile spasms, compared with 23% of those with TSC1 [20], and both infantile spasms and focal seizures began earlier in the TSC2 group [20]. A parallel pattern was observed electrophysiologically: infants with a pathogenic TSC2 variant developed epileptiform electroencephalographic discharges earlier and more frequently across multiple brain regions than those with TSC1 variants [21]. Earlier discharges predicted lower cognitive, language, and motor scores at 24 months [21]. Among seizure types, infantile spasms may be particularly consequential. In a controlled study that assessed the same children before and after spasm onset, intelligence quotient (IQ) declined after infantile spasms, and the magnitude of decline was associated with spasm duration; other seizure types of comparable overall severity showed no such effect [22].
Smaller studies have examined epilepsy severity in relation to TAND more directly. On the TAND checklist, children with drug-resistant epilepsy had poorer intellectual, behavioral, neuropsychological, and academic outcomes and higher rates of ASD and attention-deficit/hyperactivity disorder than seizure-free children [23,24]. A study using a standardized epilepsy-severity scale that incorporated age at onset, seizure type, frequency, and treatment response found that greater severity was broadly associated with poorer outcomes across most TAND domains [23]. Another cohort reported a similar pattern, with the strongest association in the ASD-like cluster: children with drug-resistant epilepsy scored higher in this cluster than in other TAND domains [24]. However, seizure control does not eliminate neurodevelopmental risk. In TOSCA, seizure control was associated with a lower rate of intellectual disability but did not prevent it [20], and TAND remained common in seizure-free children [24].
3) Genotype and intellectual ability
Genotype was initially viewed as a strong predictor of TAND. TSC2 variants generally produce a more severe phenotype than TSC1: in TOSCA, fewer than half of individuals with TSC2 had normal intellectual ability versus about two-thirds with TSC1, and TSC2 was associated with higher rates of ASD, self-injury, and academic and neuropsychological difficulties [2]. TSC2 was therefore considered a marker of more severe TAND [2]. The TOSCA authors cautioned, however, that because these manifestations are strongly tied to intellectual ability, the apparent genotype effect required re-evaluation [2].
A later analysis examined 894 participants—40% of the registry—with formal IQ data and stratified them into three levels of intellectual ability [25]. Intellectual ability was associated with most TAND manifestations and was the most consistent risk marker [25]. After adjustment for intellectual ability, nearly all genotype–TAND associations were no longer significant: the previously reported association between TSC2 and ASD disappeared, and self-injury was the only manifestation that remained associated with TSC2 [25]. These findings suggest that observed genotype associations may operate largely through intellectual ability [25]. The interpretation has two important limitations: only approximately 40% of registry participants had formal IQ data, and the analysis did not account for epilepsy, which is closely related to both intellectual ability and TAND [25].
These findings raise a further question: how does genotype shape intellectual ability? TSC1 and TSC2 encode hamartin and tuberin, respectively, which form a complex that normally restrains mechanistic target of rapamycin complex 1 (mTORC1) activity by acting as a GTPase-activating protein (GAP) for Rheb, thereby regulating the rate of protein synthesis required for neurodevelopment [4]. Molecular studies of TSC1 and TSC2 mutations suggest that cognitive outcomes depend less on which gene is affected than on the extent to which a mutation disrupts assembly or function of the hamartin–tuberin complex [26,27]. In a retrospective study of 137 patients, mutations in the tuberin-interaction domain of TSC1 and protein-truncating or hamartin-interaction domain mutations in TSC2 were associated with lower cognitive scores [26]. Because these regions form the binding interface required for complex assembly, disruption at these sites is likely to produce the most severe cognitive impairment [26]. By contrast, missense mutations and small in-frame deletions in TSC2 were associated with better cognitive outcomes [26]. A second study of 100 patients proposed that IQ is inversely correlated with the predicted expression level and biochemical severity of the mutant protein [27]. Within TSC1, IQ decreased as the aberrant C-terminal tail produced by frameshift mutations became longer [27]. The authors argued that, in these cases, the mutant protein persists in the cell rather than being cleared and interferes with the normal complex [27]. TSC2 showed the opposite, although nonsignificant, trend: longer tails were associated with slightly higher IQ, suggesting that more severely damaged proteins may be cleared more efficiently [27]. Together, these studies indicate that mutation type, location, and biochemical severity—factors that directly affect hamartin–tuberin complex function—are more consequential than the specific gene involved. This interpretation is consistent with earlier findings that intellectual ability, rather than genotype, is the stronger predictor of TAND, implying that genotype influences TAND primarily through its effect on intellectual ability [25]. Even so, both studies reported lower mean IQ and more frequent intellectual disability in patients with TSC2 mutations [26,27], likely reflecting the role of tuberin’s GAP domain in directly restraining mTORC1 activity. Severe TSC2 mutations disrupt this restraint, whereas TSC1 contributes mainly to complex stability [27].
2. Biological landscape
Neurocognitive features of TSC were long attributed to structural brain abnormalities, such as cortical tubers, or to seizures. In 2007, de Vries and Howe [4] proposed a broader model: because the TSC1–TSC2 protein complex is a global regulator and integrator of multiple physiological processes—summarized as ‘GRIPP’—neurodevelopmental features may arise directly from disrupted mTOR signaling rather than only as secondary effects of tubers or seizures. From this perspective, TAND is a core biological feature of TSC and may occur without a high tuber burden or severe epilepsy [4]. Tubers and seizures remain important contributors, but neither alone explains the full range of neurocognitive manifestations.
Loss-of-function mutations in either gene result in constitutive mTORC1 activation, disrupting the regulation of protein synthesis that supports synaptic pruning, plasticity, and interneuron maturation [4,28]. The hamartin–tuberin complex also integrates several other pathways central to brain development, including phosphoinositide 3-kinase (PI3K)–Akt, extracellular signal-regulated kinase (ERK), and AMP-activated protein kinase (AMPK), and therefore functions as a broader signaling hub [4]. Disruption of this complex can consequently affect multiple developmental processes at once, including forebrain and thalamic patterning, myelination, dendritic spine morphology, synaptic plasticity underlying learning and memory, and cellular responses to metabolic stress [4]. Dendritic spine abnormalities can arise from haploinsufficiency alone, without requiring a second hit [29], indicating that these synaptic changes occur diffusely rather than being confined to tuber-containing regions. This finding is consistent with the GRIPP model’s view that TAND does not depend on tubers [4].
The development of inhibitory γ-aminobutyric acid (GABA)ergic circuits illustrates this broader mechanism and is closely linked to epileptogenesis. mTORC1 signaling guides normal development of parvalbumin-expressing interneurons; in mice, deletion of Tsc1 from these cells disrupts connectivity and impairs social behavior [30]. Rapamycin can reverse both deficits when administered within a defined sensitive period, whereas the deficits become more difficult to correct after that window closes [30]. A parallel delay in GABA-signaling maturation, linked to altered expression of chloride transporters associated with mTORC1 hyperactivity, shifts neural circuits toward excitation [31], consistent with a proposed model of ASD [32]. Mice carrying a single mutant allele of either gene also show learning and memory deficits [33] and impaired social interaction [34], even without a structural lesion or seizure. Together, these preclinical findings support mTORC1 dysregulation as a potential shared upstream contributor to both epilepsy and TAND rather than establishing that epilepsy alone causes TAND.
This rationale made the mTOR pathway an appealing treatment target for both epilepsy and TAND. Although mTOR inhibitors are effective against seizures [35], randomized trials targeting TAND outcomes have been disappointing. Six months of everolimus did not significantly improve cognition or behavior in children and adolescents aged 6–21 years [36], and a separate twelve-month trial in children aged 4–17 years found no benefit for IQ, autism, or neuropsychological function [37]. Two explanations have been proposed for the difference between these results and findings in animal models: treatment may have begun after a critical developmental window, and the trials excluded children with active or intractable epilepsy, who might differ in their potential to benefit. Neither explanation has been established, and the exclusion criteria also limit generalizability to broader clinical populations. In an earlier trial involving patients with refractory epilepsy, seizure improvement during everolimus treatment was accompanied by improved behavior and quality of life [38]. Because mTOR inhibitors are already approved for epilepsy in TSC, optimizing seizure control may also influence TAND-related outcomes. However, the close relationships among mTOR signaling, epilepsy, and TAND make it difficult to distinguish indirect effects mediated by seizure control from direct neuropsychiatric effects.
These considerations have shifted attention toward early presymptomatic treatment. Targeting seizures alone during this period, however, has shown clear limitations. Two trials of preventive vigabatrin reduced the risk of infantile spasms but did not significantly improve neurodevelopmental outcomes at 24 months [39,40]. This finding raises the possibility that direct intervention on the mTOR pathway during this period may be needed, although that hypothesis remains unproven. In an open-label phase 1 study of preventive sirolimus, none of five treated infants had cognitive or language scores below the normal range at 24 months—a more favorable profile than that of untreated historical cohorts, but one based on a very small, nonrandomized sample [41]. Larger trials, including PROTECT, are testing mTOR inhibition in the first months of life and should clarify whether intervention before symptom onset can alter developmental trajectories [42].
3. Applying the landscape to the vignette
The child in Box 1 illustrates this landscape. His TSC2 variant and early infantile spasms align with the genotype–epilepsy pattern described above, and his TAND manifestations continued to worsen after seizure control. This trajectory is compatible with the GRIPP model, although early seizure-related injury cannot be excluded. His father and older brother, who are suspected to carry the same variant but have not been formally diagnosed, have no history of seizures; the brother has attention-deficit/hyperactivity disorder and learning difficulties. This resembles findings from mouse studies in which a single mutant allele produced learning and social deficits without a second hit or seizures [33,34]. Why the child developed infantile spasms while his relatives did not remains unknown. A single illustrative family cannot establish causality, but the pattern is consistent with the broader evidence that seizures may add to TAND burden while some neuropsychiatric manifestations may arise independently of seizures.
From Recommendations to Practice: Applying the 2023 Consensus Recommendations
1. The 2023 consensus recommendations
In 2023, the TANDem consortium published the first international consensus recommendations devoted entirely to TAND [14]. Unlike earlier guidance, which addressed TAND within broader TSC management, these recommendations were organized around the seven natural clusters. They comprise ten core principles and cluster-specific recommendations, together with psychosocial recommendations that apply across the clusters (Fig. 1) [14].
Several core principles translate the evidence reviewed above into clinical practice. Because every individual with TSC is at risk for TAND and manifestations may shift rather than resolve across the lifespan, the recommendations call for lifelong monitoring, annual screening, and early identification and intervention rather than a ‘wait-and-see’ approach [14]. The principle that ‘clusters cluster together’ reminds clinicians that difficulties in one cluster should prompt assessment of others and that co-occurring problems should be prioritized and addressed in a coordinated manner [14]. Another principle recognizes that physical problems and their treatments, such as a growing subependymal giant cell astrocytoma or antiseizure medications, can cause or worsen TAND manifestations; when manifestations emerge or change abruptly, clinicians should first evaluate possible physical causes [14]. The remaining principles describe care delivery: intervention should use a biopsychosocial, evidence-informed approach developed in partnership with caregivers as experts through lived experience, with optimal function and quality of life rather than symptom control alone as the ultimate goals [14]. Together, the principles are summarized as ‘screen, act, repeat’: screen at least annually, act on concerns through appropriate evaluation and treatment, and repeat the cycle so that emerging problems are addressed early (Fig. 2) [14].
2. Applying the recommendations to the vignette
Returning to the vignette illustrates how this framework may guide practice. Annual screening with the TAND Self-Report Quantified Checklist (TAND-SQ) would identify concerns in several clusters—ASD-like, dysregulated behavior, and eat/sleep—consistent with the principle that clusters often co-occur. Because the TAND-SQ provides quantitative scores, it can help identify the most prominent cluster and, when repeated, indicate whether difficulties are improving or worsening [43]. Rather than presenting an undifferentiated list of concerns, the clusters organize the response and help determine where to begin. The TAND-SQ is freely available for clinical use from the TAND Consortium (https://tandconsortium.org/checklists/), although an authorized Korean translation is not yet available.
The child's dysregulated behavior—severe crying, aggression, and self-injury—is a major source of family distress. If such behavior appears or changes abruptly, the recommendations first call for evaluation of potential medical drivers because a growing subependymal giant cell astrocytoma, subclinical seizures, pain, or medication effects can contribute [14]. After medical causes have been evaluated, the next steps are not TSC-specific: a functional behavioral assessment to clarify what the behavior communicates, followed by the same evidence-based nonpharmacological strategies used in the broader population [14]. The same logic applies to the child's ASD-like manifestations, which warrant formal diagnostic assessment and established autism interventions, and to sleep difficulties, which can exacerbate or result from other clusters and therefore require direct assessment [14]. The wraparound psychosocial recommendations apply throughout. The child's mother reports severe parenting stress and also needs support [14]. The recommendations emphasize monitoring the well-being of the entire family, recognizing the heightened vulnerability of children with intellectual disability, and identifying and supporting relatives who may themselves have TSC, such as the child's father and brother [14].
With annual rescreening, the child's developmental trajectory and response to each intervention can be tracked over time [14]. In this way, an overwhelming presentation becomes a set of defined, actionable problems, replacing the ‘treatment paralysis’ described by families and clinicians with a structured and individualized plan of care.
What practical challenges might affect annual TAND screening and management in Korean clinical practice? A study using the original TAND checklist reported lower psychiatric-diagnosis rates than those observed internationally, a finding attributed largely to limited awareness of TAND among patients and families [44]. Once specific difficulties were identified; however, families often sought support [44]. A greater obstacle may be access to multidisciplinary care. Coordinated services linking neurology with child psychiatry, rehabilitation medicine, and developmental programs remain limited, even at tertiary referral centers. A nationwide multicenter cohort initiative launched in 2025 across 18 university hospitals aims to establish the first systematic TSC registry in Korea. This effort is expected to generate the data needed to identify gaps in care and guide strategies to address them.
Conclusion
TAND has evolved from a scattered set of behavioral observations into a defined and increasingly actionable component of TSC care. It is now recognized as a core feature of TSC that affects most patients rather than solely as a secondary consequence of seizures. Its many manifestations group into a small number of natural clusters, and severity appears to be associated more consistently with intellectual ability and early-onset epilepsy than with genotype alone. The biological rationale for targeting mTOR signaling is strong, but trials in older children have not demonstrated clear neuropsychiatric benefit, and whether early presymptomatic treatment can alter developmental trajectories remains an open question. For clinicians, the practical message is more immediate: every individual with TSC should be screened for TAND at least annually, concerns should prompt early assessment and intervention rather than passive observation, and the cluster framework can convert an overwhelming presentation into defined and manageable problems. The next task is implementation—ensuring that screening, diagnosis, intervention, and family support reach patients for whom TAND remains a major and neglected component of TSC burden.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Cited
