Childhood absence epilepsy (CAE) typically begins between 3 and 8 years of age. The major hallmarks are typical 3 Hz spike and wave discharges (SWD) on EEG, along with brief episodes of frequent, unpredictable lapses of consciousness.

Genes associated with CAE include ligand-gated and calcium channel genes, solute carrier transporters genes, and other genes. Animal models have partly supported mutation’s effect on refining neural circuits. For example, greater dentate granule cell density and volume in the hippocampus (Richards et al., 2013), increased densities of GABA neurons, and inhibitory neuron subpopulations in the cortex have been observed in the absence of epilepsy-naive GABAA γ2R43Q mice (Wimmer et al., 2015), which strongly support mutation is most likely altering neural architecture in the early developmental period. Polygenic absence seizure mice also showed increased GABA neurons in ventral posterolateral thalamic nuclei (Cavdar et al., 2014). The alterations of these structures might cause deviations of normal thalamocortical circuit function and enhance susceptibility to absence seizures. For instance, McCafferty et al. (2018) discovered that top-down cortical excitation and thalamic reticular nucleus (TRN) mediated inhibition is essential for synchronous, seizure-perpetuating output of the thalamocortical neurons in the polygenic absence seizure model.

Human neuroimaging also revealed potential brain structural alterations which might be influenced by genetic factors. For example, in newly diagnosed CAE, the patients had significantly smaller frontal and temporal lobe volumes, thicker posterior cingulate and left medial occipital gyrus, and damaged white matter in the posterior corpus callosum and cingulum (Hutchinson et al., 2010; Kim et al., 2020). These differences were also found in non-newly diagnosed patients but were independent of disease duration, seizure frequency, and other epilepsy characteristics (Qiu et al., 2016; Drenthen et al., 2019). Consistent with the genetic rodent model, human EEG-fMRI elucidated the SWD production and maintenance in the thalamocortical network (Tyvaert et al., 2009; Tangwiriyasakul et al., 2018). However, these studies showed different pre-ictal cortical locations in various CAE populations (Bai et al., 2010; Moeller et al., 2010). These alterations preferentially in different cortical areas might partly relate to diverse mutations.

Clinically, the 5-year remission rate of anti-seizure medicines in CAE is 56%–65% (Camfield et al., 2014). The seizures gradually decrease to remission with age, which completely presents the whole process of the abnormal epileptic network from the development to collapse and provides an entry to study its disintegration. Therefore, longitudinal tracking of network change, particularly integrating neuroimaging and genetic alterations during the network collapse process, might provide a potentially deep understanding and new targets for CAE.

Juvenile myoclonic epilepsy (JME) develops predominantly between 12 and 18 years, characterized by myoclonic jerks and typical high amplitude diffuse and bilateral polyspike-waves on EEG. Altered cognitive dysfunctions mainly include executive and memory impairment in most cases (Iqbal et al., 2015).

To date, mutations inGABRA1, GABRD, BRD2, CASR, and gene encoding ICK (intestinal cell kinase) have been found in JME families (Santos et al., 2017). Functional expression studies of the ICK variants in progenitor cells showed impaired proliferation and mitosis, suggesting abnormal migration of cortical progenitors (Bailey et al., 2018). Using the rodent model, McCarthy et al. (2020) reported that BRD2 gene haploinsufficiency led to GABA neuron deficits in the striatum and primary motor neocortex as early as P15, and seizure susceptibility increased from the P30 but not in younger animals (corresponds to the human JME onset pattern). Therefore, these variants are supposed to drive the abnormal functions of neural progenitor cells, induce microarchitecture changes and lead to epilepsy in specific periods through brain maturation. Nonetheless, the causality and bioinformation of these mutations require further ascertainment.

Transition issue also occurs in CAE and JME (Camfield et al., 2014). Autosomal dominant mutations in the GABRA1 are associated with both CAE and JME. Arain et al. (2015) reported that haploinsufficient and dominant-negative mutations led to absence seizures at P35 and persisted at P120 in mutant mice. However, more frequent spontaneous and evoked polyspike-wave discharges and myoclonic seizures were observed at P120, suggesting that shared mutations might lead the disease to evolve from CAE to JME.

For neuroimaging studies of new-onset JME patients, reduced thalamic volume and abnormal white matter values were found (Ekmekci et al., 2016; Perani et al., 2018). Two prospective longitudinal experiments supported dynamic network reorganization compared to normal development, including less modular cortical network and greater frontoparietotemporal volume and thickness (Lin et al., 2014; Garcia-Ramos et al., 2018). Moreover, structural and functional imaging characteristics, especially the frontal and temporal lobes, have been identified as endophenotypes (Caciagli et al., 2019, 2020), suggesting that imaging abnormalities in patients with JME might partly be influenced by genetic factors.

From a developmental perspective, regions of the temporal and frontal lobes exhibited the greatest variations in thinning across the maturation period. Imaging genetics has suggested that cortical thinning both in late childhood and adolescence is closely tied to the expression levels of spine and dendrite gene panels, which contain several CAE and JME genes (Parker et al., 2020). In a word, the cortical and subcortical regions may follow an abnormal developmental trajectory in CAE and JME, in which genetic factors probably play a significant role. Clarifying the precise spatiotemporal modulation of gene expression profiles on imaging attributes in the future will benefit the investigation of the mechanism of seizure onset, cognitive dysfunction, and prognosis.

Juvenile absence epilepsy (JAE) onset occurs typically between 9 and 13 years. Mutations in ion channel genes and INHA may be involved in a subset of patients (Thakran et al., 2020). Network parameters showed increased node efficiency in the JAE bilateral supplementary motor area, independent of clinical variabilities (Zhang T. et al., 2021). In contrast to CAE, subtle structural alterations in JAE are mainly in frontal nodes but not in the thalamus (Tondelli et al., 2016). The brain network differences might relate to environmental and genetic factors and associate with different medication responses in different epilepsy syndromes.

Epilepsy with generalized tonic-clonic seizures (EGTCS) commences in teens and young adults with generalized tonic-clonic seizures that occur primarily within 2 h after awakening. The genetic architecture of EGTCS is complex, and no genetic variants are found in most cases. Gray matter reductions were found most pronounced in the medial frontal gyrus and the ventral thalamic nuclei, which showed no correlation with the clinical characteristics (Ciumas and Savic, 2006). However, although JAE and EGTCS are common phenotypes of GGE, the role of genetics is still highly elusive.

Compared to CAE with two-thirds remit at adolescence, a high proportion of adolescent-onset GGE patients require lifelong treatment because withdrawal could cause a relapse even in people who have been seizure-free for many years with appropriate drugs (Vorderwülbecke et al., 2017). The different disease processes might relate to changed neural plasticity and compensatory capacity during brain development (Helmstaedter and Witt, 2012). The collapse of the normal brain network and an insufficient recovery or compensation might occur in adolescent-onset GGEs. However, these abnormalities in brain networks that are associated with seizures and comorbid behavioral changes might be influenced by various environmental and genetic factors. Therefore, prospective studies comparing the mutation-related abnormal network in subtypes of drug-naive GGEs will help distinguish the pathogenesis and benefit precise treatment.

Benign adult familial myoclonic epilepsy (BAFME), also known as familial cortical myoclonic tremor with epilepsy and autosomal dominant cortical myoclonus and epilepsy, is a syndrome with clinical and genetic heterogeneity and is not yet officially recognized. Pentanucleotide expansions of TTTCA and TTTTA in different genes (BAFME 1: SAMD12, BAFME 2: STARD7, BAFME 3: MARCH6, BAFME4: YEATS2) have been recognized to play an essential role in its pathogenesis (Latorre et al., 2020). Although exact mechanisms remain unclear, non-coding repeat expansions irrespective of the genes revealed that RNA toxicity is the candidate pathomechanism (Cen et al., 2018; Ishiura et al., 2018). These genes are highly expressed in the cerebellum and can cause related local imaging alterations. Multimodal imaging studies have investigated abnormal metabolite patterns, local fluctuation, and atrophy in the BAFME cerebellum (Striano et al., 2009; Buijink et al., 2013, 2016; Long et al., 2015; Wang et al., 2020).

However, BAFME is clinically characterized by cortical tremor and myoclonus with giant somatosensory-evoked potential and long-latency cortical reflex, suggesting that the movements are generated by abnormal sensorimotor discharges (Latorre et al., 2020). The inconsistency between cortical hyperexcitability and cerebellum pathology raises the hypothesis that abnormal cerebellar inputs may induce the disinhibition of cortical neurons and cause cortical tremors/myoclonus in BAFME. This hypothesis was further confirmed by imaging studies. Long et al. (2016) observed altered cerebellar-cerebral functional connectivity in the default network, dorsal attention network, and control network. Wang et al. (2020) showed increased functional connectivity between the cerebellum and the precentral gyrus. The cortical dysfunctions and altered cerebellar-cerebral loop suggest that the main disease mechanism is a network impairment and highlight that noncoding-repeat expansions may cause the disruption of the neural network and induce cortical seizures in BAFME (Latorre et al., 2020).

Sleep-related hyper motor epilepsy (SHE), previously known as nocturnal frontal lobe epilepsy, is a heterogeneous syndrome with acquired injuries and genetic causes. The autosomal dominant pattern of SHE (ADSHE) has been associated with mutations in several genes, including genes encoding nAChR (mainly in CHRNA4, CHRNB2, CHRNA2), DEPDC5, NPRL2, NPRL3, KCNT1, and CRH (Perucca et al., 2020). In addition, these mutations have also been reported in some sporadic cases (Chen et al., 2009; Liu et al., 2011).

Approximately 12% of the pedigrees carry mutations in genes encoding nAChR. Fukuyama and Okada (2020) revealed the age-dependent and event-related epileptogenesis in the S286L-mutant CHRNA4 mice model (corresponding to the human S284L-mutant CHRNA4 gene). The mutant impairs the activation of GABAergic transmission from the TRN to thalamic nuclei, weakening Erk signaling inhibition and gently increasing connexin 43 expression and glutamatergic transmission in the thalamocortical pathways. At the critical seizure onset period, factors (such as physiological sleep spindle and pathological interictal discharges) cause persistent hemichannel opening and excitatory neurotransmitter release, leading to the disruption of several homeostasis systems and orbitofrontal epileptic seizures.

In ADSHE patients with an identified nAChR gene mutation, Picard et al. (2006) found decreased nAChR density in patients’ prefrontal cortex, which explained why mutations in receptors presented in the entire brain result in focal frontal lobe epilepsy. Fedi et al. (2008) also showed reduced D1 receptor binding at the striatum, suggesting that mutation might modify a wide range of other functions. In more common sporadic SHE patients, metabolic alterations were observed in the dorsolateral prefrontal cortex (Wang W. et al., 2021). Although the pathomechanism is partially well revealed in the CHRNA4 mice model, many unsolved challenges exist. For instance, what is the neuroimaging foundation for patients with different variants? The mechanism is essential since patients with various variants in the nAChR gene showed good or poor drug responses (Fukuyama and Okada, 2020). Moreover, genetic underlying in sporadic patients is largely unclear, suggesting neural mechanisms are complicated and need detailed exploration.

Familial focal epilepsy with variable foci (FFEVF) is a focal seizure that begins at any time from infancy to adulthood. The variable foci mean that seizures emanate from various cortical locations in different family members. However, for each individual, a single focus remains constant throughout their lifetime (Picard et al., 2000). FFEVF is the paradigmatic phenotype of GATOR1 related Mendelian focal epilepsy, which has mutations in the GATOR1 protein complex (DEPDC5, NPRL3, and NPRL2, Weckhuysen et al., 2016). GATOR1 protein complex plays a pivotal role in regulating mTOR signaling and several downstream effects in brain development, such as proliferation, differentiation, migration, dendrite formation, and plasticity. Animal model and resected human brain tissue suggest mutations in DEPDC5, NPRL3, and NPRL2 increase mTOR pathway activation, which could disrupt neuronal morphogenesis and cortical lamination, increase neuronal excitability, and might be implicated in epileptogenesis by altering the formation of neural circuits (Weckhuysen et al., 2016; Hsieh et al., 2020; Klofas et al., 2020).

Detailed brain imaging indicated that a subset of patients with GATOR1 mutations had structural brain abnormalities, particularly focal cortical dysplasia (Weckhuysen et al., 2016; Baldassari et al., 2019). The mTOR pathway had been thought to play a significant role in focal cortical dysplasia as a ubiquitous regulator. One interesting question is that instead of causing mTOR hyperactivation in the whole brain neurons, heterozygous germline mutation normally leads to local mTOR activation and cortical dysplasia. Animal model studies suggested that biallelic inactivation was necessary to cause mTOR hyperactivation, shape neuronal morphology, and generate epilepsy with focal cortical dysplasia (Ribierre et al., 2018). Resected human tissue specimens demonstrated somatic mutations plus the germline mutation caused a functionally homozygous condition (Baldassari et al., 2019). The somatic mutational load was higher in the epileptogenic zone with increased concentrations of dysplastic cells, whereas lower or absent in the surrounding areas. These studies provide strong evidence that the somatic “second hit” mutations in cortical neurons rather than heterozygous germline mutation alone cause mTOR signaling hyperactivation and focal abnormal neurodevelopment.

Another clinical challenge is the phenotypic heterogeneity related to mutations in the mTOR signaling pathway. For example, mutations in GATOR1 could induce a range of epilepsy syndromes. In addition to FFEVF, other Mendelian (ADSHE, familial mesial temporal lobe epilepsy) and sporadic focal epilepsies have been reported (Baldassari et al., 2019). Mutations in other mTOR pathway genes, such as TSC1, TSC2, AKT3, MTOR, and PIK3CA, are linked to some epilepsy-related multisystem disorders (e.g., tuberous sclerosis complex; Nguyen and Bordey, 2021). The “second hit” mechanism might involve region- and age-specific effects of the mTOR pathway genes, which provide a plausible explanation for the wide range of phenotypic variability (Rivière et al., 2012; Ribierre et al., 2018; Goswami and Hsieh, 2019). Further combining genetics and neuroimaging would be explorable to decipher these spatiotemporal effects.

Autosomal dominant lateral temporal lobe epilepsy (ADLTE), also known as autosomal dominant epilepsy with auditory features, displays autosomal dominant inheritance with reduced penetrance. The most common gene is LGI1, of which variants account for around 30% of families (Michelucci et al., 2013). RELN and MICAL-1 have been recently identified as susceptibility genes, which together may account for another 30% of pedigrees (Dazzo et al., 2015, 2018). Variants in LGI1 are related to altered synaptic function and neuronal morphology in the hippocampus (Zhou et al., 2009; Hivert et al., 2019). Reelin (encoded by RELN) co-localizes with LGI1 in the temporal and hippocampus, suggesting a convergent signaling pathway (Michelucci et al., 2017). In addition, the MICAL-1 variants cause dysregulation of the actin cytoskeleton dynamics and could also serve abnormal cell maturation and migration (Dazzo et al., 2018). Therefore, MRI of patients carrying mutations could be expected to show even minor changes. Indeed, dysplastic features and abnormal white matter values in the temporal gyrus were found in patients with genetic mutations (Tessa et al., 2007).

Familial mesial temporal lobe epilepsy (FMTLE) is a heterogeneous syndrome with unclear genetic architecture. Although mutations in mTOR pathway genes have been identified, they might not be the frequent cause of FMTLE (Striano et al., 2015; Wang Y. et al., 2021). The most common neuroimaging findings of FMTLE are HS and focal cortical dysplasia (Dührsen et al., 2018), but whether HS is affected by genetic factors remains unclear. MRI studies for both FMTLE patients and their asymptomatic siblings showed similar but mild HS in siblings (Alhusaini et al., 2016). However, no genes have yet been strongly associated with this imaging feature.

Childhood epilepsy with centrotemporal spikes (CECT), previously known as benign childhood epilepsy with centrotemporal spikes or rolandic epilepsy, is a common focal syndrome with complex inheritance patterns. The most relevant gene for CECT is GRIN2A. Other related genes include ELP4, BDNF, KCNQ2, KCNQ3, RBFOX1, and DEPDC5 but require further investigation (Xiong and Zhou, 2017). GRIN2A encodes the GluN2A subunit of the NMDA receptor and transmits signals involved in brain development and synaptic plasticity (Mota Vieira et al., 2020). Recently, neuroimaging has ascertained the developmental delay in CECT, which is associated with neuroanatomic changes in the Rolandic regions (Zhang Q. et al., 2021). These results suggest that CECT is related to abnormal brain development during childhood.

Moreover, the GRIN2A-related epilepsy-aphasia spectrum is broad and ranges from the mild end, such as CECT, to the severe end, such as Landau-Kleffner Syndrome (LKS) and Continuous Spike and Wave During Sleep (CSWS) in DEE (Carvill et al., 2013; McTague et al., 2016). Recent studies have suggested that the clinical heterogeneity might relate to diverse mutations in GRIN2A (Strehlow et al., 2019). Nonetheless, there are still many unclear mechanisms. For example, what are the neuroimaging alterations during the developmental period? How do these variants lead to different brain changes and epilepsy-aphasia phenotypes? What is the genetic basis of patients without the GRIN2A mutations?

Collectively, genetic factors may be involved in the development of some abnormalities in focal epilepsies. For patients with a relatively clear cause, gene mutations affect one or combined developmental steps, including neuronal maturation, proliferation, migration, and layer organization, and might lead to seizures in various periods. However, when and where the brain abnormalities emerge and where changes in the brain fit on the pathway from molecular dysfunction need further investigation. It is essential since early intervention before patients present with symptoms might reverse neuronal abnormalities and seizures (Zhang et al., 2020; Nguyen et al., 2021). most sporadic focal epilepsies, the underlying mechanisms are unclear. Association analysis with whole-genome sequencing to search for causative common and rare variants might reveal the genetic basis of epilepsy more effectively.

Post-hoc GWAS have recently emerged that permit integration of functional genomic annotations with GWAS summary statistics and translation into explicable biological mechanisms and models.

Imaging genetics also adopts multivariate techniques to mine the complex association between genes and quantitative imaging traits (e.g., epistatic effect, phenotypic pleiotropy, and genetic heterogeneity; Shen and Thompson, 2020). Several data-driven multivariate approaches have been proposed for imaging genetics, such as sparse reduced rank regression (sRRR), sparse canonical correlation analysis (sCCA), and parallel independent component analysis (pICA). These methods are designed to extract latent variables from both genetic and imaging data and maximize the links between latent genotype-phenotype pairs. For example, sRRR minimizes the error of the multivariate regression by reducing the rank of the regression matrix. sCCA maximizes the correlation between two modalities, while pICA extracts maximally independent components from each modality as well as simultaneously optimizes the correlation. Since multimodal imaging data may carry complementary information, novel multimodal sCCA models have been proposed to combine multimodal imaging data with genetic data (Du et al., 2020). In addition, extensions of multivariate analysis, which incorporate prior knowledge such as genes, pathways, and known trait-associated variants, can simplify model complexity and offer meaningful biological insights for data-driven estimation (Bi et al., 2020).

It is now appreciated that epilepsy is highly heterogeneous at the phenotypic and genomic levels and that they continue to evolve over time. Thus, the epilepsy field requires not only in vitro imaging genetic biomarkers but also spatially and temporally resolved biomarkers. The value of neuroimaging genetics in epilepsy stems from the fact that patients can acquire multiple times imaging during various developmental and disease stages, thus making it possible to understand the roles of genetic factors on progressive brain changes. Several models have incorporated longitudinal imaging data into the imaging genetic framework. For example, temporal multi-task sCCA models (T-MTSCCA) make full use of the complementary information carried by imaging QTs at different time points through learning a canonical weight matrix for SNPs, where each column corresponds to one SCCA task between each longitudinal imaging modality and SNPs. In addition, T-MTSCCA maximizes the canonical correlation for each time point separately while do not require the genetic component to maintain the same across all the time points. This means this method takes brain heterogeneous progressive patterns (different SNPs at different time points may contribute to regional variations) and longitudinally correlated SCCA tasks into account and is able to identify a trajectory of progressive and time-dependent imaging genetic patterns (Du et al., 2019). Future integration of longitudinal transcriptome and epigenome data in statistical models might help uncover the dynamic interaction pattern from imaging and genomic levels and offer biomarkers for disease progression.

Brain-wide gene-expression atlases open opportunities to investigate the mechanisms between gene expression and brain variations. Allen Human Brain Atlas offers an anatomically comprehensive expression atlas (more than 20,000 genes in 3,702 regions; Hawrylycz et al., 2012). By comparing a spatial imaging phenotype (quantifies age-associated changes or case-control differences) to the spatial expression pattern of each gene, transcriptomic imaging studies can shed light on the molecular correlates of brain changes in both normal and aberrant neurodevelopment. In addition, detailed solution for challenges in transcriptomic imaging data (e.g., intrinsic spatial autocorrelation, specificity of identified associations, false-positive bias in enrichment) enables more robust and rigorous inference (Fulcher et al., 2021; Markello and Misic, 2021). As the field develops, spatiotemporal transcriptome and other functional genomics atlases covering the entire human lifespan will provide molecular correlates of disease-related brain changes from a developmental perspective (Wang et al., 2018; Eze et al., 2021; Song et al., 2021). In addition to human-specific atlases, detailed brain spatiotemporal expression atlas can be measured in animal models such as mice, for which specimens are more readily available (Han et al., 2018). Along with dual in utero electroporation tools, studies can even manipulate the gene expression in a spatially and temporally regulated manner (Zhang et al., 2022). Thus, these techniques allow for mutation discovery and their region- and age-specific effects investigation.

Collectivity, the rapid growth of statistical and machine learning methods offer the possibility of studying epilepsy imaging genetic mechanisms at multiple spatial and temporal scales, which can effectively identify both common and rare genetic variants related to imaging and ultimately help complete the causal line of genome-transcriptome-endophenotype-phenotype.