Germline homozygous missense DEPDC5 variants cause severe refractory early-onset epilepsy, macrocephaly and bilateral polymicrogyria

Abstract DEPDC5 (DEP Domain-Containing Protein 5) encodes an inhibitory component of the mammalian target of rapamycin (mTOR) pathway and is commonly implicated in sporadic and familial focal epilepsies, both non-lesional and in association with focal cortical dysplasia. Germline pathogenic variants are typically heterozygous and inactivating. We describe a novel phenotype caused by germline biallelic missense variants in DEPDC5. Cases were identified clinically. Available records, including magnetic resonance imaging and electroencephalography, were reviewed. Genetic testing was performed by whole exome and whole-genome sequencing and cascade screening. In addition, immunohistochemistry was performed on skin biopsy. The phenotype was identified in nine children, eight of which are described in detail herein. Six of the children were of Irish Traveller, two of Tunisian and one of Lebanese origin. The Irish Traveller children shared the same DEPDC5 germline homozygous missense variant (p.Thr337Arg), whereas the Lebanese and Tunisian children shared a different germline homozygous variant (p.Arg806Cys). Consistent phenotypic features included extensive bilateral polymicrogyria, congenital macrocephaly and early-onset refractory epilepsy, in keeping with other mTOR-opathies. Eye and cardiac involvement and severe neutropenia were also observed in one or more patients. Five of the children died in infancy or childhood; the other four are currently aged between 5 months and 6 years. Skin biopsy immunohistochemistry was supportive of hyperactivation of the mTOR pathway. The clinical, histopathological and genetic evidence supports a causal role for the homozygous DEPDC5 variants, expanding our understanding of the biology of this gene.


Introduction
The mammalian target of rapamycin (mTOR) pathway is central to many aspects of intracellular function, including the regulation of cellular growth and cell proliferation (1). In the brain, the pathway is active from the early stages of development and is implicated in neuronal differentiation and growth, synaptogenesis and dendrite formation, thus playing a key role in shaping the hexalaminar cytoarchitecture of the cerebral cortex (2).
Disease-causing germline variants described so far in DEPDC5 are heterozygous and in most cases inactivating, including nonsense, frameshift and splice-site variants (10). The most strongly established pathogenic mechanism is haploinsufficiency, with loss of the inhibitory function of the GATOR-1 complex causing hyperactivation of the mTOR pathway (7,11,12), ref lected by the histopathological characteristics of FCD Type II (i.e. disrupted cortical lamination and cytomegalic features).
The pathogenicity of heterozygous missense DEPDC5 variants is less clear due to limited functional evidence (13). Recently, an algorithm for pathogenicity, specific for missense variants in the GATOR-1 genes, was proposed, based on allele frequencies and in silico prediction tools (10).
A two-hit model has been suggested to explain phenotypic gradients, with evidence of a second, somatic, loss-of-function variant present within brain regions involved in FCD, on a background of a germline heterozygous inactivating variant 6 . Compound heterozygous variants (germline plus somatic) in DEPDC5 have been reported, supporting this model (5,11,(14)(15)(16).
Here, we report two germline recessive homozygous missense variants in DEPDC5, identified in nine children from five families with a characteristic severe neurological phenotype and some with additional systemic features. The affected individuals were of Irish Traveller, Lebanese and Tunisian origin. The Irish Travellers are a small nomadic population operating a clan-like structure (17). Such family structuring results in high frequency and familial occurrence of autosomal recessive disorders caused by homozygous variants, due to the presence of founder mutations (18). Both the Lebanese and Tunisian families also showed consanguinity.

Clinical phenotypes
Family trees are shown in Figure 1. The patients' phenotypes can be found in Table 1. The second-degree relative of Patient 5 (9th patient) had an identical phenotype and underwent genetic testing. It was not possible to obtain consent to describe the ninth patient in detail.
Further details on the eight patients' phenotypes can be found in Supplementary Material, S1.

Neuroimaging
Bilateral polymicrogyria and macrocephaly were common shared features in the cohort. Abnormalities of the corpus callosum, pons and basal ganglia were also evident. Magnetic resonance imaging (MRI) findings are illustrated in Figure 2.

Electroencephalography findings
Longitudinal electroencephalographies (EEGs) displayed features of progressive epileptic encephalopathy with increasingly frequent multifocal discharges of variable side emphasis (Fig. 3). Frequent, often long, clinical and electrographic seizures with multifocal onset from either hemisphere were evident (Fig. 4). Seizures were often electrographic only or characterized by subtle features of eye deviation and autonomic disturbance.

Genetic analysis
All eight patients had normal array CGH or SNP array. The parents of each child were heterozygous for the variant. Family segregation analysis showed that one of the unaffected siblings of Patient 4 was heterozygous for the p.(Thr337Arg) variant and the other sibling did not carry it.
Various further analyses, such as data interrogation for shared rare variants, did not identify other variants of interest in Patients 1 and 3-8 who underwent whole-genome sequencing (WGS) or whole-exome sequencing (WES) (see Supplementary Material, Table S1 and Supplementary Material, S2). Specific interrogation of the WES data of Patients 4 and 5 did not identify other rare variants shared by the two boys. Similarly, interrogation of the WES data of Patients 6 and 7 (siblings) did not identify other rare shared variants with a plausible relationship to the siblings' phenotype (see Supplementary Material, Table S2).
Additionally, the SNP array data from Patients 1, 2, 3 and 4 were screened for regions of homozygosity (ROH) sized >3 MB. The only shared ROH was located on chromosome 22q12 and contained six OMIM morbid genes, of which only DEPDC5 was relevant to the patients' shared phenotype (see Supplementary Material,Table  S3).

Protein modelling
Variants were analysed in structures of GATOR-1 bound to Rag GTPases RagA/RagC and/or the Ragulator complex, in three different states of activation (PDB 7t3a, 7t3b and 7t3c) (19). The overall The majority of previously reported pathogenic missense variants in DEPDC5 lie at protein interfaces, either between DEPDC5 and NPRL2 or RagA, or at inter-domain interfaces within DEPDC5 itself (Fig. 5A). In contrast, both Thr337 and Arg806 are completely buried in the DEPDC5 structure; Thr337 lies within the SABA (structural axis for binding arrangement) domain, just below the interface with NPRL2, while Arg806 forms part of the SHEN (steric hindrance for enhancement of nucleotidase activity) domain and lies close to the RagA binding surface.
In the native structure, the Thr337 sidechain lies in close juxtaposition to those of Phe343 and Asp365, while also forming a hydrogen bond to the sidechain of Gln176 (Fig. 5B). This bond was lost in the p.Thr337Arg variant, while the larger sidechain of arginine was predicted to cause steric clashes with those of its near neighbours. The thermodynamic impact of the substitution was calculated using FoldX, which provides a value for ΔΔG, the change in free energy of the variant structure compared with that of the native sequence, where values >3 kcal/mol are generally regarded as severely destabilizing (20,21). In PDB 7t3a, the calculated ΔΔG value for p.Thr337Arg was 6.39 kcal/mol, while in 7t3b and 7t3c the values were 11.07 kcal/mol and 8.06 kcal/mol respectively, indicating that the variant is likely to result in a loss of stability and a reduced level of functional DEPDC5 protein.
Consistent with this, analysis of the p.Thr337Arg variant using Missense3D predicted the variant to cause structural damage as a result of introduction of a buried charged group and breakage of a buried hydrogen bond.
Arg806 lies just below the interface with RagA, and in the native structure provides a link between this interface and the DEPDC5 core by forming hydrogen bonds to the sidechains of His861 and Asp922 (Fig. 5C). These bonds are lost in the p.Arg806Cys variant, while there will also be a loss of non-bonded contacts in the protein core due to the smaller sidechain size of cysteine compared with arginine. Consistent with this, FoldX calculated ΔΔG to be 5.21 kcal/mol in PDB 7t3a, with values of 4.36 and 3.30 kcal/mol in 7t3b and 7t3c, respectively, indicating that this variant is also likely to result in reduced protein stability. Missense3D predicted the p.Arg806Cys variant to be structurally damaging due to replacement of a buried charge and breakage of a buried salt bridge (between Arg806 and Asp922).
To examine whether ΔΔG values have biological relevance in DEPDC5, in silico mutagenesis was used to introduce all variants observed in the gnomAD database within the SABA (structural axis for binding arrangement) and SHEN (steric hindrance for enhancement of nucleotidase activity) domains (residues 166-425, and 721-1010, respectively) structures of DEPDC5 from 7t3a, 7t3b and 7t3c. The total number of variants analysed was 185, which included seven variants which have also been reported in association with disease in the Human Gene Mutation Database (HGMD; http://www.hgmd.cf.ac.uk) (four as class DM, pathogenic, and three as class DM?, possibly pathogenic), and p.Val272Ile, the only variant in the SABA or SHEN domains which has been observed as a homozygote in gnomAD. The average ΔΔG value for each variant was plotted against variant allele frequency; this strongly suggests that there is selection against alleles which are structurally damaging, while there is tolerance of those which are neutral or benign, and notably, p.Val272Ile, the only gnomAD variant observed in the homozygous state, was predicted to have no significant impact on DEPDC5 stability ( G = −0.44 kcal/mol) (Fig. 6). As a group, the average G value for all gnomAD missense variants was 1.08 kcal/mol (standard deviation, 1.76 kcal/mol; median, 0.50 kcal/mol). By comparison, the two missense variants reported here were predicted to be substantially more destabilizing, Cystic changes of the ganglionic eminence are noted (black arrow). There is corresponding restricted diffusion in DWI images (C) shown by the white arrow. The biparietal diameter and head circumference (measurements not shown) correspond to 35 weeks, suggestive of macrocephaly. Midline T2 sagittal image (D) shows a small volume pons (white arrow). The corpus callosum is fully formed (black arrow). The frontal lobes are relatively large in size. Patient 3: MRI at 7w 5d. T2 axial images (A, B) and T1 parasagittal image(C) showing extensive bilateral frontal and and this difference was highly significant (P < 0.0001, Student t-test).
In this context, it is highly likely that structural damage and destabilization caused by the p.Thr337Arg and p.Arg806Cys variants will, in the homozygous state, be sufficient to result in a significant loss of DEPDC5 and GATOR-1 function, consistent with increased mTOR activity. Furthermore, the high values of ΔΔG calculated for some rare gnomAD variants suggest that some of these might also be pathogenic if present in the homozygous state or in trans with a second damaging variant.

Immunohistochemistry
Immunohistochemistry for mTOR pathway effectors in the skin sample of Patient 3 showed prominent positive staining for pS6, as well as positive staining for pEBP1 (Fig. 7) compared with control tissue, suggestive of an overall increase in mTOR activity.

Variant classification
Both variants were absent from the gnomAD population database in heterozygous and homozygous state. The p.Arg806Cys was also absent from the local database of >100 Tunisian individuals of the Center for Integrative Genomics, University of Lausanne. A different amino acid change, p.Thr337Met, in the same position as p.Thr337Arg was found in three heterozygotes, but no homozygotes, in gnomAD.
The p.Thr337Arg variant has been reported in heterozygosity by a single submitter in ClinVar as a variant of uncertain significance. We confirmed with the submitter that the test had been requested for a condition other than epilepsy, which the subject did not have. The p.Arg806Cys variant has been reported in heterozygosity by three submitters in ClinVar as a variant of uncertain significance.   Based on evidence from bioinformatics, modelling, consistent clinical features and segregation analysis, the variants were classified as likely pathogenic (class 4) in the homozygous state, using ACMG and ACGS criteria and the framework proposed by Baldassari et al. (13). The criteria used, in detail, for the p.Thr337Arg and the p.Arg806Cys included PM3_Moderate (the homozygous variants were identified in a total of six and three patients, respectively, thought to originate from at least two apparently unrelated families with clinical features consistent with DEPDC5related disorder), PM2_Supporting (the variants have not been reported in the gnomAD database), PP1_Supporting (the variants co-segregate with disease in multiple affected family members), PP3_Supporting (the variants are highly conserved, and are predicted by SIFT, PolyPhen and AlignGVGD to have a deleterious effect on protein function) and PP4_Supporting (the patients had clinical features compatible with an mTOR pathway-related epilepsy and a skin biopsy in Patient 4 was supportive of an mTOR pathway disorder).

Discussion
We describe a characteristic severe neurologic phenotype caused by two homozygous DEPDC5 missense variants, p.Thr337Arg and p.Arg806Cys, identified in six and three children, respectively. Affected individuals demonstrated extensive bilateral polymicrogyria, early-onset refractory epilepsy and severe developmental delay. Six of the children also exhibited macrocephaly. This novel recessive phenotype differs significantly from the epilepsy (with or without FCD) phenotype previously reported with heterozygous loss-of-function germline DEPDC5 variants (6)(7)(8).
Global developmental delay was a universal feature in our patients. This worsened in severity in Patients 1, 4, 5 and 8 after the onset of seizures. Longitudinal EEGs showed progressive worsening of the encephalopathic features, in conjunction with an increased frequency of multifocal epileptiform discharges and electroclinical seizures. These features are consistent with a phenotype of developmental and epileptic encephalopathy (22).
The children also showed variable systemic extracerebral involvement. While the neurological features were strikingly similar, extracerebral manifestations were less consistent and included eye involvement, severe transient neutropenia, congenital heart defect, constitutionally small kidney and severe hyponatremia and hypo-osmolarity. The presence of multisystemic features is common in other mTORopathies, such as the well-known cardiac, renal and ophthalmological manifestations frequently seen in tuberous sclerosis. However, given the variability of extra-cerebral features in our cohort and the small number of patients so far described, it remains to be seen whether a characteristic extra-neural phenotype emerges that is causally related to DEPDC5.
All eight patients described in detail had strikingly similar findings on brain imaging. These features included bilateral polymicrogyria with anterior predominance (frontal, anterior parietal and perisylvian regions) in all patients. The corpus callosum was dysmorphic in several patients with thickened anterior segments and a posterior drooping morphology. In six patients, pontine volume was less than expected. The basal ganglia were dysmorphic in seven patients, with posterior tapering of the caudate. Macrocephaly was present in seven of the eight patients and most patients had increased frontal lobe volume, which became more conspicuous with age. A squared appearance of the frontal bone was also demonstrated in most. Patient 8 showed symmetrical T2 hyper-intensities with corresponding diffusion restriction along the inferior olivary nuclei and inferior cerebellar peduncles, the cause for which remains unknown.
Macrocephaly in the absence of cerebral ventriculomegaly is a key discriminating feature of mTORopathies, having been described as part of the phenotypic spectrum caused by variants in other genes involved in the regulation of the mTOR pathway, including autosomal dominant variants in PTEN (23) and biallelic variants in TBCK (24), HERC1 (25), TBC1D7 (26) and STRADA (27). Polymicrogyria is one of the most common malformations of cortical development, characterized by abnormal cortical lamination and excessive folding of the cortical surface (28), and while this is most commonly associated with microcephaly or normal head size, an association with macrocephaly (which was prominent in our cohort) is typically related to mTOR-hyperactivating mutations in genes of the PI3K-AKT-mTOR pathway (29)(30)(31).
The skin biopsy immunostaining were consistent with mTOR hyperactivation, providing indirect, albeit non-specific, evidence for a functional impact of the p.Thr337Arg variant at least.
Given the differences in the phenotype observed in our patients compared with that previously reported in DEPDC5associated disease, we made considerable efforts to identify possible candidate genes and variants from patient WES, WGS and SNP array data. The homozygous DEPDC5 variants p.Thr337Arg and p.Arg806Cys were the only shared variants in our cohort and the only plausible candidates remaining after filtering.
On the basis that pathogenic germline variants in DEPDC5 are believed to cause disease as a result of haploinsufficiency, it follows that missense variants, which cause a substantial loss of DEPDC5 activity, could also result in disease when present in the homozygous state or in trans with a second deleterious variant or, rarely, on their own, if they have severe enough consequences. Pathogenic missense variants may cause partial or complete loss of function either by disrupting a critical property of the protein (e.g. catalytic activity, ligand binding or protein-protein interaction), or by causing structural destabilization leading a reduced level of functional protein as a result of misfolding, increased degradation or both (20,21).
Both the p.Thr337Arg and p.Arg806Cys variants were predicted to be severely destabilizing at the molecular level, and with a magnitude which was predicted to be significantly higher than that for missense variants reported in gnomAD. However, whereas structural destabilization is a common mechanism of loss-offunction variants, missense variants may also have a deleterious impact by affecting specific functions such as catalysis, ligand binding or protein-protein interactions. Interestingly, of the eight missense variants currently reported in HGMD as pathogenic, five lie at known interfaces, either for interaction of DEPDC5 with NPRL2 or RagA or at inter-domain interfaces within DEPDC5 itself, and of these, only two are predicted to have a significant impact on protein stability (p.Met181Lys, average G = 1.96 kcal/mol; and p.His214Asp, average G = 2.88 kcal/mol). A further two variants, p.Arg247His, p.Arg997Cys) result in loss of surface charge in a region which could potentially interact with substrates or with other, as yet unknown binding partners. In this context, the p.Thr337Arg and p.Arg806Cys variants reported here are unusual in that they are both predicted to cause severe destabilization at the molecular level. However, based on the absence of phenotype in heterozygous carriers of these variants, we conclude that the two variants result in a partial loss of function, causing disease only in homozygosity, and so given that there appears to be selection against strongly destabilizing variants in the general population (as shown by thermodynamic analysis of gnomAD variants), it is perhaps not surprising that such homozygous variants have not been observed previously. Nevertheless, as genetic analysis becomes more widespread, the identification of novel individual missense variants, or new combinations of rare but potentially damaging variants, is increasingly leading to a broadening of phenotypic spectra, with recessive phenotypes (sometimes of greater severity, or of completely different features) being described for genes previously associated with 'dominant' conditions (32).
In addition to the homozygous DEPDC5 variants in our cohort, another germline homozygous DEPDC5 variant, p.Pro1031His, was recently identified in a 5-year-old girl with FCD and childhoodonset epilepsy (33). The authors proposed a molecular subregional effect, according to which variants closer to the NPRL2/NPRL3 binding site (where DEPDC5 binds to exert its inhibitory effect on the mTOR pathway) may lead to more severe phenotypes featuring cortical dysplasia (34). However, Pro1031 lies in an unstructured region of the protein which is not resolved in experimental structures, and thus the effect of the Pro1031His variant may be mediated by regulatory rather than structural effects. Interestingly, Ser1028 is reported in the PhosphoSitePlus database (https://www.phosphosite.org) to be phosphorylated, and it is possible that this modification is affected by the Pro1031His variant, although the functional consequences are as yet unknown.
The lack of functional validation of the effect of the two variants on protein stability is a significant limitation of this study. Also, the unusual characteristics of our cohort, with a high degree of relatedness among patients and the broad spectrum of extraneurological features, were major challenges for the ascertainment of the pathogenicity of the two DEPDC5 variants. Despite these limitations, the pathogenicity of the two variants is clearly indicated by the combination of evidence presented: the consistent MRI features (well in-keeping with other mTORopathies), the result of the skin biopsy and the findings of the protein modelling. Furthermore, given the differences in the phenotype observed in our patients compared with that previously reported in DEPDC5associated disease, we extensively searched for possible candidate genes and variants from patient WES, WGS and SNP array data. The homozygous DEPDC5 variants p.Thr337Arg and p.Arg806Cys were the only shared variants in our cohort and the only plausible candidates remaining after filtering.
We have, additionally, applied the ClinGen scoring system to quantify the evidence supporting the DEPDC5-phenotype association. The curated evidence included case-level data and experimental data as detailed already in the manuscript. Based on the ClinGen Gene-Disease Validity Standard Operating Procedures, Version 9 (35), the available evidence reached the level of moderate. While more evidence is needed to establish this relationship definitively, no convincing contradictory evidence has emerged.
Animal models have been generated to recapitulate pathological changes underlying DEPDC5-related epileptogenicity and to better understand gene function. Constitutive Depdc5 knockout rodents showed severe in utero growth delay, micro/anophthalmia, heart defects and embryonic lethality (12). Of further interest, a neuron-specific Depdc5 conditional knockout mouse model displayed macrocephaly, increased neuron size and dysplastic features consistent with abnormal mTOR activity and seizure susceptibility (36). Acute knockdown of Depdc5 in cultured neurons leads to mTOR hyperactivation, increased cell body size and increased excitatory (but not inhibitory) synaptic transmission and intrinsic excitability. These models provide functional evidence of the pro-epileptogenic effect of DEPDC5 loss-offunction-related mTOR hyperactivation, and indicate the gradient of phenotype severity described above, in relation to the time of onset and location of DEPDC5 functional loss (12,37). A dose effect is also suggested by the differences between knockdown and knockout models.
Both time of onset and degree of the functional impairment have an impact on tissue distribution of pathology. Thus, hemimegalencephaly and FCD share neuropathological features and common genetic aetiologies: their different spatial extents ref lect the occurrence of similar mutations at distinct developmental timepoints (38). A dose-dependent effect was recently highlighted in megalencephaly syndromes associated with pathogenic variants in PIK3CA, PIK3R2, AKT and MTOR (39). There is also pre-clinical evidence that the final consequence of hyperactivation of the mTOR pathway depends on the stage of brain development at which it occurs: while in early stages of development in a mouse model with a gain-of-function mTOR mutant, hyperactivation causes neuronal apoptosis and microcephaly, hyperactivation of the mTOR pathway in postmitotic neurons results in impaired neuronal migration and cellular hypertrophy, with macrocephaly and abnormal cortical architecture (40).
There is thus increasing evidence supporting the hypothesis that, in mTORopathies, the phenotype relates to three axes: dose effect (related to the type and allelic status of the variant); timing of onset of the effect of the mutation (pre-or postmitotic) and the consequent spatial distribution of tissue alterations. The timing of onset can clearly be determined in model systems; the timing of the occurrence of human brain somatic mutation can only be inferred by the pattern of cortical development.
In the five Irish Traveller children reported here a founder mutation seems likely, proving this is a limitation of our study. The Irish Traveller endogamous population, a nomadic population, operates a clan-like structure. Individuals typically marry within their clan and relationships are frequently consanguineous. Even couples not known to be related may share significant genetic material (18). Lynch et al. (18) published SNP-based data showing that the average level of homozygosity in the Irish Traveller population was found to be 8%, compared with 2% in the general Irish population. In our cohort, detailed family trees for the three Irish families did not support relatedness. However, it is often challenging to refine the degree of relatedness in this population, due to issues of privacy, high mobility, reluctance to seek health advice, poor health literacy and, last but not least, lack of consent to examine possible relatedness.
Taken together, the nature of the brain malformations; the absence of other plausible variants; the segregation and the evidence from immunohistology, pathogenicity predictions and in silico structural analysis pointed to the conclusion that the homozygous DEPDC5 variants were causative of the shared phenotype in our cohort. The novel recessive phenotype described here broadens the spectrum of entities associated with DEPDC5 variants, and, to our knowledge, is the most severe DEPDC5-related condition documented so far.

Materials and Methods
This project was approved by the Great Ormond Street Research and Development and Information Government Offices. Written informed consent for genetic testing, sharing of clinical information and publication was obtained from parents or legal guardians as approved by the relevant institutional review boards.

Case ascertainment
Patients 1-4 were identified at Great Ormond Street Hospital, London, UK from January to June 2019. Patient 5 was identified through personal communication with the referring clinician from Dublin, Ireland. Patients 6 and 7 were identified at the Center for Integrative Genomics, University of Lausanne, Lausanne, Switzerland. Patient 8 was identified at the Department of Clinical Genetics, Oslo University Hospital, Oslo, Norway. The connection between clinicians in the UK, Switzerland and Norway was made through Genematcher (41).

Family 1
Patient 1 was referred to the Clinical Genetics service at the age of 4 weeks, due to epilepsy, macrocephaly and polymicrogyria on brain MRI. He had baseline genetic investigations, including a segmental overgrowth panel (PIK3CA, PTEN, PIK3R2, AKT1, AKT3, CCND2, MTOR) on genomic DNA extracted from cultured skin fibroblasts. He subsequently underwent trio WGS.
Patient 2 was the younger male sibling of Patient 1. Fetal brain MRI, undertaken because of the family history, showed polymicrogyria. Patient 2 died at day 1 from an inoperable cardiac anomaly. After his death, a homozygous variant in the DEPDC5 gene was identified in his brother, Patient 1. Subsequently, targeted genetic testing was undertaken on genomic DNA (previously extracted from amniotic f luid) from Patient 2.
Patient 3 is the daughter of a paternal cousin of Patients 1 and 2 (see Fig. 1). She was referred to the Clinical Genetics service at the age of 3 months, due to epilepsy and polymicrogyria on brain MRI. She underwent trio WGS.

Family 2
Patient 4 was referred to the Clinical Genetics service at the age of 8 months, due to epilepsy, macrocephaly and polymicrogyria on brain MRI. He had normal baseline genetic investigations, including a segmental overgrowth panel (PIK3CA, PTEN, PIK3R2, AKT1, AKT3, CCND2, MTOR) on genomic DNA extracted from peripheral blood. He underwent trio WES.

Family 3
Patient 5 was referred to the Clinical Genetics service at the age of three and a half months due to possible skeletal dysplasia, as he had rhizomelic shortening, and macrocephaly. Robinow syndrome was considered, but testing for ROR2, WNT5A, DVL1, DVL3 and NOG was negative, so he underwent trio WES.

Family 4
Patient 6 was referred to the Clinical Genetics service at the age of 6 years, due to macrocephaly, epilepsy and developmental delay. He underwent duo WES, along with his affected sister, Patient 7.
Patient 7 was the younger female sibling of Patient 6. She was referred to the Clinical Genetics service, due to macrocephaly, along with the family history. At the age of 4 years, she underwent duo WES, as mentioned above.

Family 5
Patient 8 was referred to the Clinical Genetics service in the neonatal period, due to hypotonia and macrocephaly. She was reevaluated due to refractory epilepsy at 15 months, whereupon she underwent trio WES.

Genome and exome sequencing
WGS in Patient 1 was performed on a research basis through the 100 000 Genomes Project (42). Genomic DNA was extracted from cultured skin fibroblasts of Patient 1 and his parents' peripheral blood. Sequencing was performed on a HiSeq2500 (Illumina, San Diego, CA, USA) and alignment was performed by Illumina's Isaac aligner against the reference human genome GRCh37. The length of paired-end reads was 150 bp and the mean depth of coverage was 30×. Clinical genome interpretation was performed using Omicia's Opal platform (43). WGS in Patient 3 was performed by the national UK WGS provider (Illumina) and data analysis and interpretation were carried out by the Genomic Laboratory Hub based at Great Ormond Street Hospital. Genomic DNA samples were obtained from peripheral blood of the patient and her parents. The analysis included interrogation of Tier 1 and Tier 2 variants (i.e. variants in 'green' genes-confirmed clinically relevant genes). If required, Tier 3 variant analysis was restricted to relevant de novo variants and prioritized variants identified by Exomiser (https://www. sanger.ac.uk/tool/exomiser/). WES in Patients 4 and 5 was performed in the Exeter Genomics Laboratory, UK. Genomic DNA samples were obtained from peripheral blood of the patients and their parents. Whole exome libraries were prepared according to the manufacturer's instructions using the Agilent SureSelect All Exon capture kit v6 (Santa Clara, USA) or the Twist Core Human Exome protocol (Twist Bioscience, San Francisco, USA). Paired-end short reads were sequenced on a NextSeq 500 (Illumina, San Diego, CA, USA) and alignment was performed by BWA-MEM (v0.7.12) against the reference human genomes GRCh37. A minimum of 60 million reads with >80X mean coverage and >98% of target bases at ≥20X were generated. A bioinformatics pipeline designed by the Exeter Genomics Laboratory was applied to identify rare nonsynonymous variants and variants affecting conserved splice sites or within −50/+10 base pairs of exon-intron boundaries.
Duo WES in Patients 6 and 7 was performed at the Center for Integrative Genomics, University of Lausanne. Genomic DNA of the affected siblings was purified from blood. WES was performed on gDNA of the affected siblings. The exome was captured using the xGen Exome Research Panel v2 (Integrated DNA Technologies) and sequenced using an Illumina HiSeq4000 platform according to the manufacturers' protocols. The overall mean-depth base coverage was 136-and 125-fold, while on average 93% and 92% of the targeted region was covered at least 20-fold, respectively. Read mapping and variant calling were performed as described in Alfaiz et al. (44) and updated in Mattioli et al. (45). Brief ly, homozygous and heterozygous variants present in both affected siblings in reported ID genes or potential new ID genes with an MAF <1% and <0.1% in the general population (1000genome, EVS, gnomAD), respectively, were retained. Their familial segregation was assessed by Sanger sequencing.
WES in Patient 8 was performed at the Department of Medical Genetics, University of Oslo. Genomic DNA from the patient and her parents was extracted from peripheral blood. Sequencing was performed as described by McKenna et al. (33) and annotation was done using Annovar (http://wannovar.wglab.org) (46). Downstream filtering and analysis were done with Filtus (47) on the variants within coding regions and intron/exon boundaries. A triobased inheritance filtering was used focusing on de novo, recessive or X-linked variants.
Variants were classified according to the American College of Medical Genetics (ACMG) (48) and the Association for Clinical Genomic Science (ACGS) (49) guidelines for variant interpretation.

Protein modelling
Modelling of the DEPDC5 missense variants p.Thr337Arg and p.Arg806Cys was carried out using the FoldX modelling suite (50), which also provides quantitative values for G, the thermodynamic impact of variants on protein stability. Variants were also assessed using the Missense3D prediction tool (http://missense3 d.bc.ic.ac.uk/missense3d), which assesses the structural impact of missense variants on protein structure by a number of objective criteria (51). All structures were visualized in PyMOL (PyMOL Molecular Graphics System, Version 2.0, Schrödinger LLC; New York, NY, USA).

Pathological examination
A skin sample from Patient 4 was used for immunohistochemistry. Post mortem examinations were either not suggested, or declined by the families.

Immunohistochemistry
The phospho-S6 ribosomal protein (PS6; clone Ser235/236, Cell Signalling #2211) and phosphor-4E-BP1 (P4EBP-1; clone Thr37/46, 236B4, Cell Signalling #2855) antibodies were used in the dilution of 1:50 and 1:200, respectively. The HIER 30 ER2 antigen retrieval method was used and Leica Bond-Max automated immunohistochemistry for antigen detection was done as per the manufacturer's protocol. Brief ly, Leica bond detection kit (Leica #DS9800) for both antibodies was used containing the post primary (secondary) antibody-anti-mouse IgG (<10 μg/ml) in 10% (v/v) animal serum in tris-buffered saline/0.09% ProClin™ 950, DAB substrate chromogen and haematoxylin counterstain. A general control tissue micro-array was used for controls, which included a normal skin sample from excision of polydactyly in a newborn (presumed negative control). In addition, the controls for PS6 included a brain sample from FCD in a 5-year-old (positive control).

Funding
The work was supported by the Epilepsy Society. This work was partly carried out at NIHR University College London Hospitals Biomedical Research Centre, which receives a proportion of funding from the UK Department of Health's NIHR Biomedical Research Centres funding scheme. The 100 000 Genomes Project is managed by Genomics England Limited (a wholly owned company of the Department of Health and Social Care UK). The 100 000 Genomes Project is funded by the NIHR and NHS England. The Wellcome Trust, Cancer Research UK and the Medical Research Council have also funded research infrastructure. The 100 000 Genomes Project uses data provided by individuals and collected by the NHS as part of their care and support. This work was supported by grants from the Swiss National Science Foundation (31003A_182632), the Lejeune Foundation (#1838-2019A) and the Blackswan Foundation to AR.

Supplementary Material
Supplementary Material is available at HMG online.