A recurrent de novo mutation in KCNC1 causes progressive myoclonus epilepsy

Abstract

Progressive myoclonus epilepsies (PMEs) are a group of rare, inherited disorders manifesting with action myoclonus, tonic-clonic seizures, and ataxia. We exome-sequenced 84 unrelated PME patients of unknown cause and molecularly solved 26 cases (31%). Remarkably, a recurrent de novo mutation c.959G>A (p.Arg320His) in KCNC1 was identified as a novel major cause for PME. Eleven unrelated exome-sequenced (13%) and two patients in a secondary cohort (7%) had this mutation. KCNC1 encodes K_V3.1, a subunit of the K_V3 voltage-gated K⁺ channels, major determinants of high-frequency neuronal firing. Functional analysis of the p.Arg320His mutant channel revealed a dominant-negative loss-of-function effect. Ten patients had pathogenic mutations in known PME-associated genes (NEU1, NHLRC1, AFG3L2, EPM2A, CLN6, SERPINI1). Identification of mutations in PRNP, SACS, and TBC1D24 expand their phenotypic spectrum to PME. These findings provide important insights into the molecular genetic basis of PME and reveal the role of de novo mutations in this disease entity.

PMEs are among the most devastating forms of epilepsy. They are clinically and genetically heterogeneous, characterized by core features of action myoclonus, tonic-clonic seizures, and progressive neurological decline¹. Most molecularly characterized PMEs are autosomal recessively inherited with rare cases showing autosomal dominant or mitochondrial inheritance^2,3. Unverricht-Lundborg disease (ULD; OMIM 254800) is the commonest form of PME in most series and an important clinical feature is preservation of cognition^1-3. ULD is caused by mutations in CSTB⁴, and mutations in recently discovered genes including SCARB2^5,6 and GOSR2⁷ also contribute to cases of PME with preserved cognition. Other PMEs may have additional features, particularly dementia. PME-associated genes encode a variety of proteins, many of them being associated with endosomal and lysosomal function^8,9, but the associated disease mechanisms are generally poorly understood.

The precise clinical diagnosis of specific forms of PME is challenging due to their genetic heterogeneity, phenotypic similarities and overlap of symptoms with other epileptic and neurodegenerative diseases. In many cases, there are no distinguishing clinical features or biomarkers. Consequently, a substantial proportion of PME cases remain without a molecular diagnosis³.

Here, we aimed to identify the causative genes for unsolved PME cases by employing exome sequencing in unrelated patients assembled from multiple centers in Europe, North America, Asia, and Australia over a 25-year period. The extent of previous molecular studies varied, but all cases were negative for mutations in the CSTB gene and approximately half were negative for SCARB2 and GOSR2 mutations. The patient cohort was thus enriched for novel gene findings and potential atypical phenotypes associated with known disease genes.

Results

We exome sequenced 84 unrelated PME cases, of which 70 were sporadic and 14 were from families with pedigrees suggestive of either dominant or recessive inheritance. We therefore analyzed the data seeking pathogenic autosomal recessive or dominant/de novo, sex-linked, and mitochondrial DNA (mtDNA) variants (Fig. 1, Online Methods). On average 4.15 Gb of sequence was produced within the exome bait regions with an average coverage of 81 reads per nucleotide (Supplementary Table 1). After filtering the raw variant data for possible pathogenic variants, we first analyzed the data for mutations in known PME, epilepsy, or neurodegenerative disease genes followed by analysis for novel genes under both the recessive and dominant/de novo models. Analysis of the filtered variant data under the recessive model (Fig. 1a,b) revealed pathogenic or probably pathogenic mutations (see Online Methods for classification criteria) in known disease-causing genes in 12 patients. Analysis under the dominant/de novo model (Fig. 1a,b) led to discovery of a novel PME-associated gene, KCNC1, with, remarkably, the same recurrent de novo mutation in 11 patients and, in addition, revealed pathogenic mutations in known disease genes in three patients. We did not identify any obvious pathogenic mutations in mtDNA. In total, we identified pathogenic or probably pathogenic mutations in 26 of 84 cases (31.0%).

Figure 1.

Analysis of the PME exomes. (a) A simplified flow chart of the exome data analysis strategy under autosomal and sex-linked recessive and dominant/de novo inheritance models, and summary of results. The average total number of variants per exome (36528) refers to single nucleotide variants and indels passing quality filtering within the exome bait regions. Numbers in parentheses after gene names indicate numbers of patients with mutations in the gene. Gene and patient counts under ‘Potential novel PME genes’ are derived after excluding patients with pathogenic or probably pathogenic mutations in known disease genes or with the pathogenic mutation in KCNC1 identified in this study. (b) Numbers and proportions of patients without obvious pathogenic mutations or with pathogenic or probably pathogenic mutations of variable patterns of inheritance. The three patients with probably pathogenic mutations are included under sporadic cases with recessive mutations. Case PME84-1 with the recurrent de novo KCNC1 mutation is included under cases with autosomal dominant mutations with a dominant family history (Fig. 2a).

Identification of a recurrent mutation in KCNC1

To identify novel pathogenic mutations under the dominant/de novo model, we analyzed the data for potentially deleterious heterozygous variants absent in three variant databases. In addition, we did not consider variants in dbSNP except for those with a clinical association in the ClinVar database.

The highest number of novel heterozygous variants occurred in KCNC1 together with TTN (Supplementary Table 2). TTN encodes a large muscle protein and has a high mutational load¹⁰ due to its large coding region and was thus not considered further. We identified 11 (13.1%) unrelated patients with a novel heterozygous c.959G>A mutation (Supplementary Fig. 1; see Accession codes for the transcript IDs) in KCNC1 encoding the potassium voltage-gated channel (K_V) subfamily C member 1 (KCNC1, also known as K_V3.1). The c.959G>A mutation causes substitution of histidine for arginine at codon 320 of the K_V3.1 protein (p.Arg320His).

In all 11 cases, the mutation was confirmed by Sanger sequencing. Parents of all c.959G>A positive patients were unaffected and segregation analysis in eight patients where both parental DNA samples were available showed that in each case the mutation occurred de novo (Fig. 2a and Supplementary Fig. 2). Genotyping a set of microsatellite markers in five trios with sufficient DNA available confirmed that the pedigrees were correct. One of the index patients (PME84-1) with the c.959G>A mutation has an affected sibling and two affected children, who are also each heterozygous for the mutation (Fig. 2a). The parents of PME84-1 and her unaffected brother were both negative for the mutation. The presence of two mutation-positive and clinically affected children from mutation-negative unaffected parents suggests the occurrence of mosaicism in one of the parents. A restriction fragment length assay designed to detect both the normal and mutant alleles was carried out on the peripheral blood DNA from both parents but showed no indication of mosaicism (data not shown).

Figure 2.

The recurrent c.959G>A mutation in KCNC1. (a) Pedigrees of 13 unrelated patients demonstrate the de novo occurrence of the c.959G>A mutation in KCNC1. Patient IDs of the exome-sequenced cases are marked with asterisks. Triangle symbol indicates miscarriage at 8 weeks. m, c.959G>A; +, wild-type; n.d., not determined. (b) A cartoon showing the domain structure of a single K_V3.1 subunit. The positively charged arginine residues (marked with “+”) in the S4 segment detect changes in voltage. The p.Arg320His changes one of the arginine residues to a histidine (red “+”and arrow). (c) ClustalX amino acid sequence comparison of the voltage-sensing S4 segment shows full conservation of the arginine 320 (indicated by an arrow) across different species. The four positively charged arginine residues (Arg311, Arg314, Arg317, Arg320) occurring every third position are highlighted in blue. The K_V3.3 residues Arg420, Arg423, Thr428, mutated in spinocerebellar ataxia^24,25,27, are in red boxes. Asterisks, colons, and periods indicate fully conserved, strongly similar, and weakly similar residues, respectively.

We screened the KCNC1 c.959G>A mutation in a secondary cohort of 28 PME patients and identified two (7.1%) additional unrelated cases heterozygous for the mutation. Sanger sequencing of the available parental samples of one of these patients confirmed de novo occurrence of the mutation (Fig. 2a). Together, we identified 16 patients (13 unrelated) with the c.959G>A mutation. We did not identify any other mutations in KCNC1 in the exome data. As exon 1 was not sufficiently covered, we Sanger sequenced it in the 73 exome-sequenced c.959G>A negative patients, but no potentially deleterious variants were found.

The c.959G>A mutation in KCNC1 affects a highly evolutionarily conserved arginine residue in segment S4 constituting the main voltage-sensing domain of the channel (Fig. 2b,c). The mutation is predicted to be deleterious by all four in silico methods used (Supplementary Table 3).

To assess the frequency of PME due to KCNC1 c.959G>A, we used a newly published mutational model¹⁰, which takes into account both the local sequence context of mutation site and regional factors such as divergence between humans and macaques, to estimate the rate of this specific mutation. A rate of 1.75×10⁻⁷ per person was obtained, indicating that the mutation should occur in 1/5,700,000 conceptions. Additionally, we examined three other potential mutations encoding changes of the conserved voltage-sensing arginines to histidines in segment S4 of K_V3.1 (p.Arg311His, p.Arg314His, p.Arg317His). Given that the sequence context of the four potential mutations is the same (they all occur in highly mutable CGC codons), the same estimated rate applied to all of them. If all four caused PME with equal penetrance, the probability of seeing only one of four possible mutations in 13 of 13 independent cases is P = 4 × (1/4)¹³ = 6.0 × 10⁻⁸ strongly suggesting the effect of p.Arg320His is surprisingly specific for PME.

Functional analysis of p.Arg320His in K_V3.1 channel

To assess the effects of the p.Arg320His substitution on channel function, we used the Xenopus laevis oocyte expression system and automated two-electrode voltage clamp. Potassium currents were recorded from oocytes injected with identical amounts of wild-type or mutant cRNA of the human K_V3.1 channel. Whereas the wild-type channel produced robust currents upon membrane depolarization, the currents observed with the mutant channel were barely detectable (Fig. 3a,b). A western blot analysis using an antibody to DDK-tag revealed similar protein levels in total lysates of oocytes expressing wild-type or mutant subunits, indicating that the mutant protein has a similar stability as the wild-type (Fig. 3c). To examine for a possible interaction of mutant with wild-type subunits, we co-expressed them in a 1:1 ratio. These experiments showed an approximately 80% reduction in the expected current amplitude recorded at +60 mV, indicating a dominant-negative effect (Fig. 3d,e). The diminished currents produced by the combination of mutant and wild-type subunits revealed altered gating properties, with a significant hyperpolarizing shift of the activation curve (Fig. 3f).

Figure 3.

Functional analysis of p.Arg320His in K_V3.1. (a) Representative traces of whole-cell currents recorded in Xenopus laevis oocytes, injected with the same amount of cRNA encoding K_V3.1 wild-type (WT) or Arg320His, during 0.5 s voltage steps ranging from −60 mV to +60 mV. (b) Current amplitudes analyzed at the end of the voltage step to +60 mV and normalized to the mean current amplitude of WT (n = 49) recorded on the same day, revealed significantly smaller potassium currents for the Arg320His (n = 50) (Mann-Whitney U test; ***P < 0.001). (c) Western blot analysis of Xenopus laevis oocytes injected with either WT or Arg320His mutant cRNA using a mouse antibody to DDK-tag showed similar bands in total cell lysates in the two representative experiments. H₂O-injected oocytes were used as a negative control and actin served as a loading control. The presence of two bands in the blot is likely to be due to N-glycosylation that the K_V3.1 protein (expected protein size ~56kDa) is subjected to in vivo and in heterologous expression systems^50,51. (d),(e) A dominant-negative effect of Arg320His mutant on WT channels was determined when a constant amount of the WT cRNA was injected with either H₂O or the same amount of Arg320His cRNA in a 1:1 ratio. (d) Representative whole-cell currents recorded as in a were decreased for the co-expression of WT and Arg320His mutant channels. (e) Analysis of current amplitudes at the end of a 0.5-s pulse to +60 mV, followed by normalization to the WT recorded on the same day, revealed that the co-expression with the Arg320His mutant (n = 27) reduced currents by about 4-fold compared to the WT (n = 32) (Student’s t-test; ***P < 0.001). (f) Current-voltage relationships of WT and co-expression of WT with Arg320His channels revealed a leftward shift for the latter, resulting in a significant increase of normalized tail current amplitudes in almost the whole range of analyzed potentials (Student’s t-test; P < 0.001 for −60 mV to +50 mV). Lines represent fits of a Boltzmann function. The V_0.5 was 19.6 ± 0.5 mV (n = 19) and 8.0 ± 1.3 mV (n = 16) (Student’s t-test; P < 0.001) and the slope factor k 12.9 ± 0.2 and 10.2 ± 0.3 (Student’s t-test; P < 0.001) for WT and WT/Arg320His co-expression, respectively. Data in b,e,f are presented as means ± s.e.m.

The clinical phenotype of patients with the KCNC1 mutation

We obtained detailed clinical data for 15 of the 16 KCNC1 mutation-positive patients (Table 1). Their clinical phenotype was similar and, at disease onset, resembled classic ULD. On a background of usually normal development, the first symptom in the majority of patients was myoclonus (sometimes reported as tremor) at ages of 6–14 years. Ataxia developed early in one patient (PME17-1) but otherwise it was overshadowed by myoclonus as the major motor impediment. There were infrequent tonic-clonic seizures in all patients. During adolescence myoclonus generally became very severe, limiting ambulation; a walking aid or wheelchair was needed by mid to late teens. Learning disability prior to seizure onset was noted in some cases; in particular the nuclear family of PME84-1. There was mild cognitive decline in seven subjects in early adolescence, but this was difficult to quantitate due to the severe motor disability. Early death was not observed. Electroencephalogram recordings showed generalized epileptiform discharges with photosensitivity in some cases. MRI scans had no specific features; and were regarded as normal or showed cerebellar atrophy. The clinical picture in the family with four affected members (PME84) was milder; the two older sisters were ambulant in the fourth decade. Similarly, the secondary cohort case 1 had a less severe course than the majority of cases.

Table 1. Clinical features of patients with the c.959G>A mutation (p.Arg320His) in KCNC1.

Individual ID	Ancestry	Gender	Onset age	Initial symptom	Seizures		Learning disability	Cognitive decline	Age and outcome
					Myoclonus	Tonic-clonic
Exome sequenced patients
PME8-1	Italian	M	12y	Myoclonus	+++	+	No	Yes	38y; Wheelchair at 27y
PME10-1	Italian	M	6y	Myoclonus	+++	++	No	Possible	34y; Wheelchair at 17y
PME17-1	French	M	<5y	Ataxia	+++	++	No	No	40y; Wheelchair at 16y
PME18-1	Norwegian	F	10y§	Myoclonus	+++	+	No	Yes	36y; Wheelchair at 15y
PME29-1	German	M	9y	Tremor/Myoclonus	++	+	Yes	No	24y; Cautiously ambulant
PME36-1	American (Italian)	F	7y	Tremor	+++	+	No	No	22y; Wheelchair at 14y
PME37-1	Italian	F	10y	Myoclonus	+++	+	No	No	19y; Wheelchair at 17y
PME44-1	French Canadian	F	12y	Myoclonus	++	+	No	Yes	24y; Walker at 17y
PME63-1	Moldavian☼	F	9y	Myoclonus	+++	++	No	Possible¶	15y; Wheelchair at 13y
PME65-1	Portuguese	F	9y	Myoclonus	+++	+	No	Yes	25y; Wheelchair at 19y
PME84-1	Israeli (Sephardic)	F	10y	Tremor	+	+	Yes	No	42y; Ambulant
Sanger sequenced patients
PME84-sister	Israeli (Sephardic)	F	13y	Myoclonus	++	++	Yes	Yes	37y; Cautiously ambulant
PME84-son	Israeli (Sephardic)	M	12y	Ataxia	+	+	Yes	Yes	19y; Cautiously ambulant
PME84-daughter	Israeli (Sephardic)	F	14y	Tonic-clonic seizure	−	+	Yes	No	16y; Ambulant
Secondary 1	Danish	M	10y‡	Myoclonus	++	+	No	Yes	18y; Cautiously ambulant
Secondary 2†	Israeli	M	n/a	n/a	n/a	n/a	n/a	n/a	n/a

+ (mild), ++ (moderate), +++ (severe), − (not observed)

^§earlier absence seizures

^☼ascertained in Italy

^¶severe myoclonus and language barrier prevented good assessment

^‡simple febrile seizures at 6 - 12 months

^†no detailed clinical data available

Mutations in known disease genes

Analysis for mutations in known PME, epilepsy, and neurodegenerative disease genes revealed either pathogenic or probably pathogenic mutations in 15 out of the 84 exome sequenced patients (Table 2; see Supplementary Table 4 for clinical details as well as summaries of the genetic findings, Supplementary Fig. 3 for pedigrees with segregation data, and Supplementary Figs. 4 and 5 for conservation of the novel mutation sites). These fell into three groups.

Table 2. Pathogenic or probably pathogenic mutations in known PME, epilepsy or neurodegenerative disease associated genes.

		Patient ID, gender	Gene	Zygosity	Coding DNA change§	Protein change§	CADD, SIFT, PolyPhen, Mutation taster prediction‡	Allele frequency in 1000G/EVS☼ (%)	Previously reported pathogenic mutation (reference)	Disease associated to the gene (OMIM gene number)
Recessive	Pathogenic	PME23-1, F	NEU1	het	c.1208delG	p.Ser403ThrfsTer85	32,NA,NA,NA	0/0	Yes52	PME: sialidosis (608272)
				het	c.982G>A	p.Gly328Ser	32,D,D,D	0/0	Yes53
		PME51-1, F	NEU1	het	c.982G>A	p.Gly328Ser	32,D,D,D	0/0	Yes53
				het	c.679G>A	p.Gly227Arg	28.5,D,D,D	0/0	Yes53
		PME87-1, F	NEU1	het	c.914G>A	p.Arg305His	9.4,B,PD,D	0/0	No, but p.Arg305Cys is reported54
				het	c.625delG	p.Glu209SerfsTer94	26.1,NA,NA,NA	0/0	Yes53
		PME35-1, F	NHLRC1	hom	c.436G>A	p.Asp146Asn	15.75,B,PD,D	0/0	Yes55	PME: Lafora disease (608072)
		PME81-1, M	NHLRC1	hom	c.830C>A	p.Ala277Glu	12.72,D,D,D	0/0	No
		PME82-1, M	EPM2A	hom	c.590A>T	p.Asp197Val	27.4,D,D,D	0/0	No	PME: Lafora disease (607566)
		PME33-1, M	CLN6	hom	c.509A>G	p.Tyr170Cys	17.11,D,D,D	0/0	No	PME: Neuronal ceroid lipofuscinosis, late infantile and adult variants (606725)
		PME62-1, M	AFG3L2	hom	c.1875G>A	p.Met625Ile	30,B,PD,D	0/0	No	Spinocerebellar ataxia 28 and spastic ataxia with PME (604581)
		PME32-1, F	AFG3L2	hom	c.1875G>A	p.Met625Ile	30,B,PD,D	0/0	No
	Probably pathogenic	PME14-1, F	TBC1D24	hom	c.1079G>T	p.Arg360Leu	19.83,D,D,D	0/0	No	Multiple neurological diseases including familial infantile myoclonic epilepsy (613577). PME not previously described
		PME15-1, F	SACS	het	c.8393C>A	p.Pro2798Gln	25.7,D,D,D	0.09%/0.29%	Possible13¶	Autosomal Recessive Spastic Ataxia of the Charlevoix Saguenay (604490). PME not previously described
				het	c.2996T>C	p.Ile999Thr	17.35,D,B,D	0/0.015%	No
		PME75-1, F	SACS	het	c.8393C>A	p.Pro2798Gln	25.7,D,D,D	0.09%/0.29%	Possible13¶
				het	c.1373C>T	p.Thr458Ile	17.45,D,D,D	0.14%/0.24%	Possible15¶ and yes14
Dominant/de novo	Pathogenic	PME42-1, F	SERPINI1	het	c.1175G>A	p.Gly392Glu	22.5,D,D,D	0/0	Yes11	PME: Familial encephalopathy with neuroserpin inclusion bodies (604218)
		PME86-1, F	PRNP	het	c.305C>T	p.Pro102Leu	19.12,D,D,D	0/0	Yes12	Gerstmann-Sträussler-Scheinker disease (176640). PME not previously described
		PME20-1, F	SCN1A	het	c.677C>T	p.Thr226Met	20.5,D,D,D	0/0	Yes21	Dravet syndrome (182389)

^§Genomic positions and more detailed annotations of the mutations are presented in Supplementary Table 3. Mutation coding DNA nucleotide and amino acid position numbers are based on Ensembl transcripts listed under Accession codes.

^‡CADD scores of >15 indicate deleteriousness of the variant (see Online Methods). B=benign (not considered deleterious by the method), PD=possibly deleterious (applies to PolyPhen only), D=deleterious, NA=not available.

^☼1000G, 1000 Genomes project; EVS, Exome Variant Server of NHLBI Exome Sequencing Project.

^¶See Supplementary Note for discussion of the pathogenicity of the previously reported SACS mutations.

First, ten patients had recessively inherited or a de novo heterozygous mutation in established PME genes, seven with atypical clinical presentations. The ten comprised three cases of Lafora disease (NHLRC1 and EPM2A), three cases of sialidosis (NEU1), and one of neuronal ceroid lipofuscinosis (CLN6), all well-recognized PME genes. Two Italian cases, not known to be related, had the same novel homozygous missense AFG3L2 mutation (c.1875G>A, p.Met625Ile) in the proteolytic domain of the protein (Supplementary Fig. 4e). Exome data showed that they shared a ~1.75 megabase run of identical homozygous polymorphisms flanking the mutation, indicating that the mutation is identical by descent. The tenth case had a de novo previously described¹¹ pathogenic heterozygous missense mutation (c.1175G>A, p.Gly392Glu) in the SERPINI1 coding neuroserpin. Mutations in both AFG3L2 and SERPINI1 are very rare known causes of PME.

Second, four patients had mutations in known genes where PME has not been reported as a key part of the presentation. One case with an adult-onset PME and a father who had died of a similar disease was found to have a previously described¹² pathogenic heterozygous missense mutation (c.305C>T, p.Pro102Leu) in the PRNP gene encoding prion protein. The mutation is a known cause of Gerstmann-Sträussler-Scheinker disease, one of the inherited prion diseases. Two unrelated patients had probably pathogenic, rare compound heterozygous missense mutations in the SACS gene (c.8393C>A, p.Pro2798Gln in both patients; c.1373C>T, p.Thr458Ile and c.2996T>C, p.Ile999Thr in one patient each). Two of the substitutions (p.Thr458Ile, p.Pro2798Gln) have been reported in spastic ataxia patients^13-15. One patient had a novel probably pathogenic, homozygous missense mutation (c.1079G>T, p.Arg360Leu) in TBC1D24, where recessive mutations cause variable neurological disorders including familial infantile myoclonic epilepsy^16-20.

Third, in one patient we identified a previously seen²¹ pathogenic heterozygous mutation (c.677C>T, p.Thr226Met) in SCN1A, the major gene underlying severe myoclonic epilepsy of infancy (Dravet syndrome)²². The eventual clinical evolution of this case was typical of Dravet syndrome, and not that of PME. Initial consideration of PME as a clinical diagnosis was due to the predominence of myoclonic seizures in the early disease course.

All variants passing the filtering in the known disease genes, including the ones not considered pathogenic or probably pathogenic, are listed in Supplementary Tables 5 (recessive) and 6 (dominant/de novo).

Search for additional novel genes

We analyzed the exome dataset for potential novel PME genes, other than KCNC1, using both recessive and dominant/de novo models. We did not identify other likely novel genes using our criteria of observing putative mutations in a gene in at least two unsolved patients under the recessive model or in at least four patients using the dominant/de novo model (Supplementary Note, Supplementary Tables 2 and 7). However, genes of interest in single cases that would warrant further exploration include ALG10 and APOA1BP, which harbor homozygous loss-of-function mutations (Supplementary Table 8).

Discussion

Using an exome sequencing approach in a clinically heterogeneous cohort of 84 unrelated PME patients without a specific cause, we reached a genetic diagnosis in 31% of the cases. Importantly, we identified a recurrent mutation in KCNC1 as a novel cause of PME explaining a substantial proportion of cases. KCNC1 encodes K_V3.1 functioning as a highly conserved²³ K⁺ channel subunit of the K_V3 subfamily of voltage-gated tetrameric K⁺ channels. While KCNC1 mutations have not been associated with human disease to date, autosomal dominantly inherited or de novo missense mutations in KCNC3 (K_V3.3) cause spinocerebellar ataxia^24-27.

K_V3 channel subunits consist of six membrane-spanning segments (S1–S6), have overlapping expression patterns, and can form heterotetramers²⁸. S4 constitutes the main voltage-sensor where specific positively-charged arginines contribute to the gating charge^29,30.

Electrophysiological analysis showed that the PME-causing p.Arg320His substitution, which affects one of the voltage-sensing residues, has a prominent loss of function with a dominant-negative effect on wild-type K_V3.1 channels. We also observed altered gating properties but the physiological consequence of this finding is questionable as its overall contribution on K_V3.1-mediated current is minor. Similar biophysical properties have been reported for an ataxia-causing alteration in K_V3.3, p.Arg423His³¹, occurring in the position analogous to p.Arg320His in K_V3.1. Mutations affecting positively charged residues in S4 of voltage-gated cation channels may contribute to pathogenicity by generating leak currents through the gating pore³². For example, in Shaker K_V channel the change analogous to K_V3.1 p.Arg320His increases proton permeability³³. However, gating pore currents are not detectable in the analogous K_V3.3 mutant³¹, questioning the importance of the phenomenon in the context of K_V3.1. Given the ability of K_V3 subunits to form heterotetramers, the dominant-negative p.Arg320His in K_V3.1 is likely to disrupt all K_V3-mediated currents of the neurons in which it is expressed. Reflecting functional redundancy, Kcnc1 and Kcnc3 knockout mice have relatively mild phenotypes, whereas double mutant mice show myoclonus, tremor, and gait ataxia^34-36. Thus, the dominant-negative K_V3.1 and K_V3.3 patient mutations seem to have an effect comparable to the double knockout.

The K_V3 subfamily is distinguished from other K_V channels by showing a more positively shifted voltage-dependent activation and faster activation and deactivation rates. This makes K_V3 channels major determinants of high-frequency firing in several types of central nervous system (CNS) neurons²⁸. Studies using mutant or pharmacologically suppressed K_V3 channels have demonstrated that loss of K_V3 function disrupts firing properties of fast-spiking neurons^28,37,38, impacts neurotransmitter release^28,39, and induces cell death⁴⁰. The expression of K_V3.1 is limited to the CNS with the exception of a subpopulation of T lymphocytes^41,42. It is preferentially expressed in specific subsets of fast-spiking neurons with prominent expression in inhibitory GABAergic interneurons^28,41. Therefore it is likely that the p.Arg320His substitution in K_V3.1 results in disinhibition due to impaired firing of fast-spiking GABAergic interneurons. This mechanism is likely to contribute particularly to myoclonus and tonic-clonic seizures. Furthermore, dysfunction and/or degeneration of cerebellar neurons, where K_V3.1 is expressed⁴¹, is likely to contribute to the motor impairment. Modulation of K_V3 channel function may provide a possibility for pharmacological intervention in KCNC1 patients. However, while drugs with anticonvulsant effects activating other K_V channels exist⁴³, there is no activator of K_V3 channels currently available.

The initial clinical presentation and evolution of ULD⁴⁴ and the disorder in patients with the KCNC1 mutation, designated here as “MEAK” (Myoclonus Epilepsy and Ataxia due to Potassium channel mutation), are similar. They have overlapping age of onset and moderate to severe incapacitating myoclonus, infrequent tonic-clonic seizures, and mild if any cognitive decline. Differences emerge as the clinical course for MEAK is generally more severe. ULD is caused by mutations in CSTB encoding cystatin B implicated in oxidative stress and inflammation^45,46. Evidence from the mouse model for ULD suggests that altered GABAergic signaling contributes to the latent hyperexcitability^47,48, implying a possible convergent pathway for ULD and MEAK.

The recurrence of the KCNC1 mutation is likely due to its location in a CpG dinucleotide, which are mutation hotspots⁴⁹. We estimated that the mutation occurs in 1/5,700,000 conceptions, thus potentially affecting hundreds of patients globally. Observation of four KCNC1 patients ascertained in a multicenter clinical collaboration³ in Italy supports this estimate assuming that probably not all existing cases were ascertained, and that the mutation may reduce lifespan. We did not observe any other mutations in KCNC1, which is among the top 1% most constrained genes¹⁰. For example, the estimated mutation rate of the three other potential arginine to histidine substitutions affecting the voltage-sensing residues is equal to p.Arg320His and thus should have been observed in our cohort if the phenotypic consequence was the same. This suggests that Arg320 is biophysically special. Indeed, codon specific consequences of S4 mutations have been demonstrated, for example, in K_V3.3³¹.

Identification of mutations in previously established disease genes expands the phenotypic and genotypic spectrum of PME. Highlighting the utility of exome sequencing as a diagnostic tool in a heterogeneous patient cohort previously subjected to molecular analyses, we identified pathogenic mutations in known PME genes in ten individuals, of whom the majority had atypical symptoms (Supplementary Note and Supplementary Table 4). Importantly, we identified mutations in three known disease genes (PRNP, SACS, TBC1D24), where PME has not been appreciated as part of the clinical spectrum. These cases are discussed in Supplementary Note. The majority of all solved cases also had variants in other known disease genes (Supplementary Tables 5 and 6). These variants, however, did not fulfill our criteria for pathogenicity. It is possible that they modify the clinical outcome, thus contributing to some atypical presentations.

The genetic basis remained unknown in over two thirds of cases. Besides KCNC1, there were no other novel PME genes definitively identified (Supplementary Note and Supplementary Tables 2 and 7), however, of the nine genes with homozygous loss-of-function mutations in single patients some are interesting candidates (Supplementary Table 8). In light of establishing de novo mutations as an important cause of PME, exome sequencing in trio setting could be pursued to further dissect this heterogeneous cohort. Also, the role of CNVs and epistatic mutations in PMEs should be assessed. Our findings, especially the discovery of MEAK, will aid in molecular diagnostics and potential therapeutic interventions in PME and the exome data will facilitate further gene identification.

Online Methods

Study subjects

The cohort consisted of 84 unrelated patients with a clinical presentation of PME, collected from multiple centers in Europe, North America, Asia, and Australia over a 25-year period for molecular study, where a specific diagnosis had not hitherto been made. Seventy-three patients were of European origin, seven of Western Asian, three of Southern Asian, and one of Chinese. The extent of previous molecular investigations varied; mutations in the CSTB gene had been excluded in all and 43 patients were tested negative for mutations in SCARB2 and 39 negative for GOSR2. A variable set of other genes was also screened in some patients. Informed consent for DNA analysis was obtained from patients in line with local institutional review board requirements at time of collection. Family members, when available, were recruited for segregation analysis done by Sanger sequencing.

Seventy cases were sporadic. Three probands had either an affected parent or affected offspring. Ten probands had at least one affected sibling and one had an affected cousin with known parental consanguinity. Fifteen of the patients were reported to be the result of a consanguineous union. Based on inbreeding coefficients obtained by FEstim⁵⁶ the number is eighteen. No cryptic relatedness (PIHAT > 0.05) between patients was detected by Plink⁵⁷ IBD analysis.

A secondary cohort of 28 cases with PME or possible PME was Sanger sequenced for the KCNC1 recurrent mutation. These cases were excluded from our original cohort for exome sequencing due to insufficient DNA or inadequate clinical data.

Exome sequencing

Exome sequencing was carried out at the Wellcome Trust Sanger Institute (Hinxton, Cambridge, UK). Genomic DNA (approximately 1 μg) extracted from peripheral blood for each sample was fragmented to an average size of 150 bp and subjected to DNA library creation using established Illumina paired-end protocols. Adapter-ligated libraries were amplified and indexed via PCR. A portion of each library was used to create an equimolar pool comprising 4–8 indexed libraries. Each pool was hybridized to SureSelect Human All Exon 50Mb V3 RNA baits (Agilent Technologies) and sequence targets were captured and amplified in accordance with manufacturer’s recommendations. Enriched libraries were subjected to 75 base paired-end sequencing (HiSeq 2000; Illumina) following manufacturer’s instructions. Each pool of indexed samples was sequenced twice on two different flow cells.

Sequence read alignment and processing

Alignment of the sequenced DNA fragments to Human Reference Genome was performed using the Burrows-Wheeler Alignment Tool (BWA)⁵⁸. The reference sequence used was the 1000 Genomes Phase II reference (hs37d5) which is based on GRCh37 and consists of chromosomes 1–22, X, Y, as well as mitochondrial genome (rCRS mitochondrial sequence NC_012920), unlocalized and unplaced contigs, Human herpesvirus 4 type 1 (NC_007605), and decoy sequences derived from HuRef, Human Bac and Fosmid clones and NA12878.

Duplicate reads were marked using the Picard toolkit. Sequence reads were further processed using GATK software (version 2.8.1)^59,60 based on GATK Best Practices. Specifically, local realignment was performed around known insertion or deletion (indel) locations, and base quality scores were re-calibrated.

Variant calling

Single nucleotide variants (SNVs) and indels were called for all exomes jointly using GATK HaplotypeCaller. Called variants with quality score >30 were accepted. Variant quality scores of the PME exomes were recalibrated jointly with GATK VariantRecalibrator. HapMap 3.3 and Omni 2.5M SNP chip array from 1000 genomes project were used as truth sites training resources in variant quality score recalibration (VQSR) of SNVs. ‘Mills and 1000 genomes gold standard’ indels were used as truth sites in indel VQSR. A truth sensitivity cutoff of 99.0% was used for both SNVs and indels. Additional filtering was applied to exclude variant sites where >80% of the samples have a <5× read depth. Individual variant calls with a read depth below 5× were excluded.

Variant filtering under recessive and dominant/de novo inheritance models

Reflecting the different possible inheritance patterns of the underlying mutations in the study subjects, we aimed to identify autosomal or sex-linked recessive and dominant (including de novo) mutations as well as pathogenic changes in the mitochondrial genome (Fig. 1). We considered coding variants with the following Variant Effect Predictor⁶¹ (Ensembl release 75) annotated variant consequences based on Sequence Ontology nomenclature: missense variant, initiator codon variant, splice donor or acceptor variant, stop lost, stop gained, inframe insertion or deletion (inframe indels in tandem repeat regions⁶² obtained from UCSC Genome Browser⁶³ were excluded), and frameshift variant. If there were multiple Ensembl gene transcripts overlapping the variant site, the transcript having the most severe consequence annotation for the variant was selected while prioritizing Ensembl database canonical transcripts. Only variants within the genes included in the December 2013 version (release 15) of Consensus CDS (CCDS) project, implemented in Ensembl release 75, were considered. To exclude likely benign amino acid changes, missense variants were considered only if two or more of the four in silico methods predicted the variant to be deleterious: CADD⁶⁴ (PHRED-like scaled C-score >15, which is a suggested threshold for the deleteriousness by the authors), PolyPhen-2 HumVar⁶⁵ (“possibly” or “probably damaging”), SIFT⁶⁶ (“deleterious”), MutationTaster⁶⁷ (“disease causing”, the MutationTaster prediction scores were obtained using dbNSFP⁶⁸ and correspond to the converted score suggested by the authors). In rare cases, when not all of the four prediction methods could be applied, we accepted missense mutations predicted to be deleterious by at least 50% of the available methods. We also analyzed the exome data using only CADD to evaluate the deleteriousness of silent or noncoding mutations (Supplementary Note).

In the recessive filtering, we included homozygous, hemizygous, or compound heterozygous variants with allele frequencies <1% in the three following variant databases: 1092 exomes from the phase I release of the 1000 genomes project⁶⁹, the 6503 exomes of the Exome Variant Server (EVS) NHLBI GO Exome Sequencing Project (version 0.0.25), and 3268 Finnish exomes (Sequencing Initiative Suomi project, SISu; only autosomal SNVs, epilepsy patients removed). Variants present as homozygous or hemizygous in any of the three variant databases were excluded. As we did not exome sequence parental samples, we used three methods for phasing genotypes to facilitate identification of compound heterozygous variants. First, physical phasing of SNV calls, based on the presence of heterozygous variants in the same sequencing reads, was performed using GATK ReadBackedPhasing. Since ReadBackedPhasing method handles SNVs only, heterozygous indels present in the same sequencing reads were removed manually. Second, to detect rare variants segregating together in populations (likely to be inherited from one parent only), Beagle⁷⁰ was applied to phase genotypes using 1000 genomes phase I data as a reference. Third, we used genotype information from the 3268 Finnish exomes to detect heterozygous variants likely to be in the same allele. After filtering the variant data using recessive hypothesis prior to taking advantage of the phasing information, we visualized the sequencing reads surrounding the variants passing the filtering using Integrative genomics viewer⁷¹ (IGV) to detect likely false positive calls raising mainly within segmental duplication regions where mapping of reads is more challenging. As a technical comparison set in variant visualization, we used 43 in-house exomes, which had been processed in the same sequencing, read processing and variant calling pipelines as exomes in this study. After excluding low quality variants, we implemented the phasing information to the cleaned data and repeated the recessive filtering.

In the dominant/de novo filtering strategy, we assumed full penetrance of the mutations and, since the great majority of patients have disease onset before adulthood (10/84 cases with onset at 18 years or later, maximum onset at 26 years), we included heterozygous variants absent from the three variant databases. Additionally, dbSNP build 138 variants were excluded except those with clinical association in NCBI ClinVar. After applying the filtering criteria, we excluded low quality variants based on IGV visualization of sequencing reads as described in the filtering of recessive variants.

Analysis of filtered variant data under recessive and dominant/de novo strategies

The exome data were analyzed both for variants in genes previously associated with PME, epilepsy and/or neurodegenerative disorders and for variants in novel genes. The association of genes to Mendelian diseases was annotated based on the Online Mendelian Inheritance in Man, OMIM^® database⁷². Additionally, genes involved in monogenic forms of epilepsies and neurodegenerative diseases were retrieved based on diagnostic panels developed by Lemke and colleagues⁷³. Genes involved in PME were also retrieved from literature.

To identify novel recessive or dominant/de novo genetic causes, genes with variants surviving the filtering strategy were ranked based on the number of PME patients with variants in the gene. Genes that are highly polymorphic in human populations and thus are likely to rank high in the data analysis only due to their high and benign mutational load were given less priority if at least one of the following criteria was fulfilled: 1) the gene has a Residual Variation Intolerance Score (RVIS)⁷⁴ (estimates the toleration of genes for functional variation) in the most tolerant 10^th percentile (this criteria used only in dominant/de novo model) and 2) gene was identified as hyperpolymorphic by Fuentes Fajardo and colleagues⁷⁵. Other measures used in prioritization of candidate genes and variants were known disease association, function (Uniprot database⁷⁶ and literature) and expression pattern (GTEx database⁴² and literature) of the gene.

Analysis of mitochondrial DNA variants

Since PME may also be caused by mutations in the mitochondrial genome⁷⁷, we analyzed the mtDNA for possible disease-associated mutations. Even though mtDNA is not included in the SureSelect exome capture kit, we obtained an average of 32.7× sequencing coverage per sample in the mitochondrial genome, due to the abundance of mitochondrial DNA in the cells. We called mtDNA variants using GATK UnifiedGenotyper. At the sites for the most common reported mutations associated with diseases showing myoclonus as part of the clinical presentation, namely myoclonic epilepsy with ragged red fibers (MERRF) and mitochondrial encephalomyopathy, lactic acidosis, and stroke-like episodes (MELAS), the average coverage was at least 24.8×. We searched for novel mitochondrial mutations by excluding known mtDNA polymorphisms in the MITOMAP database. The same database was also used to obtain positions for known disease-associated mtDNA mutations.

Classification criteria for mutations in known disease genes as well as for novel disease genes

We used a three-tier scale to classify mutations passing the variant filtering in known PME, epilepsy, or neurodegenerative disease genes: 1) pathogenic, 2) probably pathogenic, and 3) unlikely pathogenic. The classification criteria were the following.

Pathogenic

Mutation is in a known PME gene, i.e., where mutations cause a disease with PME as a key part of the presentation, or mutation is in a known epilepsy or neurodegenerative disease gene and the mutation has previously been reported as pathogenic and the phenotype of the patient is compatible with the previous cases. In addition, mutations were required to occur in a conserved domain where other pathogenic mutations have previously been reported. Segregation data, when available, had to comply with the expected inheritance in the family and the mode of inheritance had to concur with previously reported cases. If no segregation data were available, the phenotype had to present substantial overlap with previously reported cases, and in the case of previously reported pathogenic compound heterozygous mutations, they should not be occurring in cis in the original reports.

Probably pathogenic

Mutations occur in only one patient in a previously established disease gene with phenotypic overlap with PME or mutations occur in two or more patients in a gene with a previously established association to neurological disease, with little overlap with PME. In addition, mutations are required to occur in a conserved domain where other pathogenic mutations have been reported previously. Segregation data of the mutations had to comply with the expected inheritance in the family and the mode of inheritance had to concur with previously reported cases.

Unlikely pathogenic

Mutations not fulfilling the above listed criteria.

To consider novel disease genes as candidates for PME, we required them to harbor mutations in at least two patients in the recessive variant filtering strategy and in at least four patients in the dominant/de novo strategy. The higher threshold in the dominant/de novo strategy was set to limit the number of candidates to a feasible level to do follow-up, as exome variant data of the parents was not available for filtering inherited variants in data analysis.

Variant validation and segregation analysis

Candidate mutations in the known and novel disease genes were confirmed and segregation of the variants was analyzed, where DNA from parents or siblings was available, by bidirectional Sanger sequencing (ABI BigDye 3.1, Applied Biosystems) on ABI 3730xl DNA Analyzer. Primers (available from authors upon request) were designed with Primer-Blast⁷⁸. The sequences were analyzed using Sequencher (Gene Codes Corporation) and illustrated using 4Peaks (Nucleobytes Inc.). Evaluation of the quality of sequence reads over the recurrent c.959G>A mutation in KCNC1 is described in Supplementary Note.

Parental testing of the cases with the c.959G>A mutation in KCNC1

Parental testing was carried out for five cases (those with sufficient amount of parental DNA available) with the de novo c.959G>A mutation in KCNC1 mutation to exclude false paternity and inadvertent sample substitution. Biparental testing was performed using twelve highly polymorphic microsatellite markers: D3S3680, D4S418, D6S289, D7S2560, D8S281, D13S175, D13S221, D15S117, D19S1150, DXS1113, DXS1036, and DXS7423. PCRs were done using the Qiagen Multiplex PCR kit according to the manufacturer’s instructions. The reverse primer of each pair was labeled with either HEX or FAM. Products were analyzed on an ABI 3131 Genetic Analyzer.

Analysis of parental mosaicism of c.959G>A mutation in KCNC1 in PME84-1 family

The KCNC1 exon 2 PCR product was amplified using the same primers and conditions as for sequencing and 10 μl of PCR reaction was digested with 2 units of HpyCH4V (New England Biolabs) for 2 hours at 37°C under the conditions recommended by the manufacturer. Fragments were visualized by electrophoresis of 10 μl of the digest on a 2% agarose gel in TBE. Gels were stained with RedSafe™ (INtRON Biotechnology). PCRs and digests were performed in duplicate.

Evaluation of the expected rate of mutations encoding arginine to histidine changes in the S4 segment of K_V3.1

The expected rate of the four possible mutations encoding arginine to histidine changes (CGC>CAC) in the voltage-sensing arginine residues of the K_V3.1 S4 segment (p.Arg311His, p.Arg314His, p.Arg317His, p.Arg320His, see Fig. 2b,c) was established using a recently developed statistical framework¹⁰, which takes into account both the local sequence context of mutation site and regional factors such as divergence between humans and macaques.

Functional analysis of p.Arg320His substitution in K_V3.1

Mutagenesis and RNA preparation

We used the Quick Change kit (Stratagene) to engineer the missense mutation c.959G>A (p.Arg320His) in the human KCNC1 cDNA (NM_004976, this construct corresponds to the K_V3.1a isoform (511 aa), which has a shorter cytoplasmic C-terminal domain but identical biophysical properties compared to the longer K_V3.1b (585 aa) isoform⁷⁹) cloned in a pCMV-Entry vector obtained from OriGene Technologies. This clone contains a C-terminal Myc-DDK-tag. Insertion of the mutation was confirmed and additional mutations excluded by Sanger sequencing. Primers are available upon request. cRNA was prepared using the T7 mMessage mMachine kit from Ambion.

Oocyte preparation and injection

The use of animals and all experimental procedures were approved by local authorities (Regierungspraesidium Tübingen, Tübingen, Germany). Extracted ovary pieces (Xenopus laevis) obtained from the Institute of Physiology I, Tübingen were treated with collagenase (1 mg ml⁻¹ of type CLS II collagenase, Biochrom KG) in OR-2 solution (mM: 82.5 NaCl, 2.5 KCl, 1 MgCl₂ and 5 Hepes, pH 7.6), followed by thorough washing and storing at 16 °C in Barth solution (mM: 88 NaCl, 2.4 NaHCO₃, 1 KCl, 0.33 Ca(NO₃)₂, 0.41 CaCl₂, 0.82 MgSO₄ and 5 Tris/HCl, pH 7.4 with NaOH) supplemented with 50 μg ml⁻¹ gentamicin (Biochrom KG). The equal volumes (50 nl) of cRNAs with the concentrations adjusted to 2 μg μl⁻¹ were injected using Robooinject® (Multi Channel Systems) in the same batch of oocytes plated in 96 well-plates. Recordings were performed in parallel at days 2–3 after injection. Amplitudes of interest for all currents recorded on the same day were normalized to the mean value obtained for the K_V3.1 wild-type on that day, so that the normalized data from different experiments could be pooled.

Automated oocyte two-microelectrode voltage clamp

Potassium currents in oocytes were recorded at room temperature (20–22°C) on Roboocyte2® (Multi Channel Systems) using prepulled and prepositioned intracellular glass microelectrodes with a resistance of 0.3–1 MΩ when filled with 1 M KCl/1.5 M KAc. The bath solution was ND96 (mM: 93.5 NaCl, 2 KCl, 1.8 CaCl₂, 2 MgCl₂ and 5 Hepes; pH 7.5). Currents were sampled at 5 kHz. For the analysis of channel activation, we kept cells at the holding potential of −90 mV and used 0.5^−s depolarizing steps (Δ 10 mV) from −60 mV to +60 mV, followed by a step to −90 mV for 0.5 s to analyze tail currents.

Data analysis

Voltage clamp recordings were analyzed using Roboocyte2+ (Multi Channel Systems), Clampfit (pClamp 8.2, Axon Instruments), Excel (Microsoft), and Origin (OriginLab Corp.) software. The voltage-dependence of channel activation was derived from tail current amplitudes recorded at −90 mV. A Boltzmann function was fit to the current-voltage relationships, I(V) − I_max/(1 − exp[(V − V_0.5)/k]) + C, where I_max is the maximum tail current amplitude at test potential V, V_0.5 the half-maximal activation potential, k a slope factor reflecting characteristics of voltage-dependent channel gating and C a constant. Maximum current amplitudes were compared at the end of a 0.5-s test pulse to −60 mV. All data are shown as mean values ± s.e.m. Statistical analysis was performed with GraphPad Software, and significant differences (P < 0.05) determined using Student’s t-test or Mann-Whitney U test.

Western blot

For Western blots, injected Xenopus oocytes were lysed in the buffer containing in mM: 20 Tris, 100 NaCl, 1 EDTA, 0.5% Triton X-100, and 10% glycerol with protease inhibitors cOmplete (Roche). Upon determining the protein concentrations (BCA system, Thermo Fisher Scientific), 20 μg of protein was separated by SDS-PAGE on 8% polyacrylamide gels. The proteins were transferred onto nitrocellulose membrane (Whatman, GE Healthcare Europe) and Western blotting performed using a mouse monoclonal antibody to DDK-tag (1:1500; OriGene Technologies, TA50011). Chemiluminescence detection was done according to the manufacturer’s protocol (ECL Western Detection Kit; Amersham Pharmacia Biotech Europe). Actin was used as a loading control.