HMGCS1 variants cause rigid spine syndrome amenable to mevalonic acid treatment in an animal model

Abstract

Rigid spine syndrome is a rare childhood-onset myopathy characterized by slowly progressive or non-progressive scoliosis, neck and spine contractures, hypotonia and respiratory insufficiency. Biallelic variants in SELENON account for most cases of rigid spine syndrome, however, the underlying genetic cause in some patients remains unexplained. We used exome and genome sequencing to investigate the genetic basis of rigid spine syndrome in patients without a genetic diagnosis.

In five patients from four unrelated families, we identified biallelic variants in HMGCS1 (3-hydroxy-3-methylglutaryl-coenzyme A synthase). These included six missense variants and one frameshift variant distributed throughout HMGCS1. All patients presented with spinal rigidity primarily affecting the cervical and dorso-lumbar regions, scoliosis and respiratory insufficiency. Creatine kinase levels were variably elevated. The clinical course worsened with intercurrent disease or certain drugs in some patients; one patient died from respiratory failure following infection. Muscle biopsies revealed irregularities in oxidative enzyme staining with occasional internal nuclei and rimmed vacuoles.

HMGCS1 encodes a critical enzyme of the mevalonate pathway and has not yet been associated with disease. Notably, biallelic hypomorphic variants in downstream enzymes including HMGCR and GGPS1 are associated with muscular dystrophy resembling our cohort’s presentation. Analyses of recombinant human HMGCS1 protein and four variants (p.S447P, p.Q29L, p.M70T, p.C268S) showed that all mutants maintained their dimerization state. Three of the four mutants exhibited reduced thermal stability, and two mutants showed subtle changes in enzymatic activity compared to the wildtype.

Hmgcs1 mutant zebrafish displayed severe early defects, including immobility at 2 days and death by Day 3 post-fertilisation and were rescued by HMGCS1 mRNA. We demonstrate that the four variants tested (S447P, Q29L, M70T and C268S) have reduced function compared to wild-type HMGCS1 in zebrafish rescue assays. Additionally, we demonstrate the potential for mevalonic acid supplementation to reduce phenotypic severity in mutant zebrafish. Overall, our analyses suggest that these missense variants in HMGCS1 act through a hypomorphic mechanism.

Here, we report an additional component of the mevalonate pathway associated with disease and suggest biallelic variants in HMGCS1 should be considered in patients presenting with an unresolved rigid spine myopathy phenotype. Additionally, we highlight mevalonoic acid supplementation as a potential treatment for patients with HMGCS1-related disease.

Dofash et al. describe a muscle disorder associated with variants in HMGCS1, which encodes a critical enzyme of the mevalonate pathway. In a zebrafish model, supplementation with mevalonic acid reduces phenotypic severity and improves survival, suggesting that supplementation could be a potential treatment for HMGCS1-related disease.

Introduction

Inherited myopathies are a clinically and genetically heterogeneous group of debilitating disorders characterized by skeletal muscle dysfunction and weakness.¹ Traditional subgroups include congenital myopathies, muscular dystrophies, mitochondrial myopathies and metabolic myopathies. However, the clinical overlap between these entities can make it difficult to attain an accurate diagnosis.^1,2

Rigid spine syndrome is a distinctive manifestation seen in various myopathies.^3-5 Key features include contractures of the limb and spinal joints, limitation of flexion of the neck and trunk, weakness of the cervical and dorso-lumbar spine muscles, and progressive scoliosis.^3-10 Muscle pathology is variable and can include multiminicores, central cores, rods, rimmed vacuoles, and fibre-type predominance.^3,4,11 The disorder typically manifests during the first decade of life and may be severe, slowly progressive, or non-progressive. Serum creatine kinase (CK) levels may be elevated.² Patients may be susceptible to recurrent infections and respiratory insufficiency, which can be fatal.^3,6-10 For the purposes of this manuscript, the term ‘rigid spine syndrome’ will be used as an umbrella description for disorders associated with the rigid spine features described above.

The genetic mechanisms and pathways that underlie rigid spine syndrome are diverse.^5,7,10,12,13 Biallelic variants in SELENON, encoding selenoprotein N, are the most frequently reported cause of rigid spine manifestations.^8,10,14 Other associated genes include RYR1,¹⁵TOR1AIP1⁷TPM3,¹⁶ACTA1,⁹TTN,¹⁷FHL1¹⁸ and more recently, GGPS1.¹¹ However, the genetic cause in some patients remains unknown despite screening known disease-associated genes.¹⁹ This suggests that additional genes are yet to be associated with rigid spine syndrome.

In this study, we investigated the genetic cause of rigid spine syndrome in patients remaining without a genetic diagnosis, using exome or genome sequencing and functional genomic approaches. In five patients from four unrelated families, we identified biallelic variants in HMGCS1, a gene of the mevalonate pathway encoding 3-hydroxy-3-methylglutaryl-coenzyme A synthase. To our knowledge, HMGCS1 has not been previously associated with disease, though overlapping myopathic manifestations have been associated with both acquired and inherited disorders of the mevalonate pathway.^11,20,21 Here, we report biallelic variants in HMGCS1 as a novel cause of early onset myopathy with rigid spine, which we have termed HMGCS1-myopathy. This report expands the associations of the mevalonate pathway with inherited skeletal muscle disease and encourages screening of HMGCS1 in patients with unsolved rigid spine syndrome. Furthermore, we demonstrate that mevalonate supplementation can increase survival and reduce phenotypic severity in a zebrafish model of the disorder.

Materials and methods

All investigations were approved by the Human Research Ethics Committee of the recruiting centres [University of Western Australia Human Research Ethics Office, National Center of Neurology and Psychiatry in Japan Human Research Ethics Office, Gaslini Children’s Hospital (Comitato Etico della Regione Liguria 163/2018) and the Mass General Brigham Institutional Review Board (Protocol No. 2016P001422)]. All families participating in this study provided informed consent. Zebrafish experiments and maintenance were approved by Monash Animal Research Precinct 3 committee.

Clinical investigations

Patient details

We have clinically characterized five patients from four unrelated families originating from Spain (sib pair), Japan, Italy and the USA, annotated here as SPA1, JPN1, ITA1 and USA1, respectively. Patients presented with rigid spine syndrome from childhood. Serum CK levels were measured in all patients including during acute episodes following intercurrent disease in some patients. Patients were also evaluated for pulmonary function and cardiac involvement. Muscle MRI was performed in some cases.

Muscle pathology

Muscle biopsies were obtained from some patients as part of routine diagnostic investigations. The samples were frozen and processed for routine histological and histochemical analyses, as outlined previously.²² Electron microscopy was performed following standard methods,²² as previously described.²³

Genetic investigations

Whole exome or genome sequencing were performed from genomic DNA isolated from peripheral blood of the patients and their relatives, where available. Detailed methods for each family are described in the Supplementary material, ‘Methods’ section.

HMGCS1 expression and abundance

HMGCS1 expression was assessed in healthy control tissue and cell lines as well as in muscle biopsy from Patient P3 (Family JPN1) using quantitative PCR or bulk RNA sequencing (RNA-seq) following methods detailed in the Supplementary material. HMGCS1 abundance in healthy control tissue and cell lines, as well as in muscle biopsy from Patient P1 (Family SPA1) was assessed by western blotting or mass spectrometry, as detailed in the Supplementary material, ‘Methods’ section.

Functional investigations of recombinant HMGCS1

Recombinant wild-type and mutant HMGCS1 were synthesized using a bacterial system (methods detailed in the Supplementary material) to functionally characterize four of six missense variants identified in patients: p.Ser447Pro (HMGCS1^S447P), p.Gln29Leu (HMGCS1^Q29L), p.Met70Thr (HMGCS1^M70T) or p.Cys268Ser (HMGCS1^C268S). Functional investigations included size exclusion chromatography, thermal shift assays, circular dichroism, and enzyme assays to assess protein dimerization, stability, secondary structure and activity, respectively, as described in the Supplementary material, ‘Methods’ section.

Zebrafish hmgcs1 experiments

Generation of zebrafish hmgcs1 mutant line

The zebrafish hmgcs1 null mutant line was generated by injecting one-cell stage embryos with a mix containing two pre-complexed guideRNA’s (Integrated DNA Technologies, Alt-R™) targeting exon 3 (5′-GCTTGGTCAACGTACTGAGAtgg-3′ and 5′-GACACCACTAACGCCTGCTAtgg-3′), Cas9 protein (HiFi v3 from Integrated DNA Technologies), 300 mM KCl and 0.005% phenol red. Injected embryos were raised to adulthood and crossed to wild-type fish, and embryos screened for deletions using the following genotyping primers: hmgcs1_ex3_F 5′-CCATGTGGCCCAAAGATGTG-3′, hmgcs1_ex3_R 5′-TGGACTCCACCCAGTTGAC-3′. A founder passing on a 302 bp deletion in exon3 was identified.

HMGCS1 rescue experiments

Human wild-type and variant HMGCS1 was codon-optimized for zebrafish using CodonZ²⁴ and synthesized with an Sp6 transcriptional start site and an afp 3′UTR by Genscript™. Capped mRNA for injection was synthesized using the mMessage mMachine Sp6 kit (Thermofisher Scientific). One-cell stage embryos from hmgcs1 heterozygote in-crosses were injected with 1.8 nl of a mix containing HMGCS1 mRNA at 18 ng/μl for overexpression or 3 ng/μl for variant functional studies, 300 mM KCl, 0.005% Phenol red and Cascade blue-labelled dextran (Molecular Probes). The control-injection mix did not contain mRNA. Embryo survival was recorded daily until 6 days post-fertilization (dpf) for the overexpression assay or 4 dpf for the variant rescue assay. Remaining embryos were genotyped to determine the number of homozygous mutants alive. Touch-evoked escape response was performed as described in Sztal et al.²⁵

Mevalonic acid treatment, survival and locomotion assays

Embryos from mated hmgcs1^+/− fish were treated with 50 μM, 500 μM or 1000 μM mevalonic acid (Sigma-Aldrich M4667 dissolved in ethanol) or ethanol as vehicle control from ∼6 h post-fertilization (hpf) onwards until 6 dpf. Methods for the survival assays and locomotion assays are detailed in the Supplementary material, ‘Methods’ section.

Electron microscopy and mitochondrial quantification

Embryos at 48 hpf were processed for electron microscopy to investigate skeletal muscle ultrastructure and mitochondrial membrane integrity and density, as detailed in the Supplementary material, ‘Methods’ section.

Results

Clinical findings

We report five patients from four families presenting with a clinical picture of rigid spine syndrome (Fig. 1, Table 1 and Supplementary material). Age-of-onset ranged from birth to 4 years of age. There was no relevant family history of disease in any of the families. Features included spinal rigidity primarily affecting the cervical (n = 5/5) and dorso-lumbar (n = 4/5) regions and scoliosis (n = 5/5; Fig. 1B). All patients had restrictive pulmonary function with a forced vital capacity ranging between 10–56%. Some patients also presented with weakness of the proximal (n = 4/5) and distal (n = 2/5) limbs, scapular winging (n = 3/5) and bulbar weakness (n = 1/5; Table 1). Two patients had a high-arched palate (Table 1). Asymmetric atrophy of the sternocleidomastoid muscle was observed in one patient (Patient P1, Family SPA1). None of the patients presented with facial weakness, ophthalmoparesis or ptosis. No voice abnormalities were observed. Baseline CK levels were variably elevated in all cases, ranging between 190–7890 IU/l (normal ∼70–170 IU/l²⁶). In three patients (Patients P1, P3 and P5), episodes of muscle deterioration and respiratory crisis were triggered by intercurrent disease or certain drugs. Patient P2 (SPA1) died at age 15 from such an episode after developing a respiratory infection followed by a rapid decline in muscle strength and respiratory function leading to respiratory failure. Patient P3 (Family JPN1) was also reported to experience such episodes and elevated CK levels during febrile illness. The episode of decline for Patient P5 (Family USA1) was triggered after starting megestrol and slowly improved after discontinuing the drug.

Figure 1.

HMGCS1-myopathy cohort. (A) Pedigrees of the four families with biallelic variants in HMGCS1 segregating with rigid spine syndrome. (B) Rigid spine presentation demonstrated in images of Patients P1 and P2 (Family SPA1), P3 (Family JPN1) and P4 (Family ITA1). Features include spinal rigidity affecting the cervical and dorso-lumbar regions, limited neck flexion, scoliosis, scapular winging and hypotrophy.

Table 1.

Clinical summary of the HMGCS1-myopathy cohort

Family	SPA1		JPN1	ITA1	USA1
Patient	P1	P2	P3	P4	P5
Sex	M	F	F	M	F
Age at onset, yo	2	2	2	4	0
Age at death, yo	Alive	15	36	Alive	Alive
Cause of death	n/a	Respiratory failure	Acute exacerbation of psychosis with delusion	n/a	n/a
Age at last follow-up, yo	27	15	36	38	45
Proximal limb weakness	Yes	No	Yes	Yes	Yes
Distal limb weakness	No	No	Yes (mild)	Yes (mild)	Yes (mild)
Facial weakness	No	No	No	No	No
Ptosis	No	No	No	No	No
Bulbar weakness	No	No	No	No	Yes
High arched palate	No	Yes	Yes	No	No
Rigid spine, cervical	Yes	Yes	Yes	Yes	Yes
Rigid spine, dorsolumbar	Yes	Yes	Yes	Yes	Yes
Scoliosis	Yes	Yes	Yes	Yes	Yes
Scapular winging	Yes	Yes	Yes	Yes	No
Muscle MRI	Yes	No	Yes	Yes	No
CK, IU/l	190–7890	430–4100	400–2000	274–939	200–5000
Cardiac involvement	No	No	LVEF 57%	LVEF 55–60%	No
Pulmonary function test	Restrictive FVC, 1.51 L (37%)	Restrictive FVC, (56%)	Restrictive FVC, 0.44 L (16%)	Restrictive FVC, 1.11 L (26%)	Restrictive FVC, (10% at 26 yo)
Episodic worsening during intercurrent disease	No	Yes	Yes	No	No

CK = creatine kinase; F = female; FVC = forced vital capacity; LVEF = left ventricular ejection fraction; M = male; n/a = not applicable; yo = years old.

Muscle MRI performed for two of five patients (Patients P1 and P4) showed variability in T₁ signalling primarily in the posterior muscles of the thigh, showing fat replacement predominantly involving the most central areas (near the femur) of muscles of the posterior compartment of the thigh, as well as the vastus intermedious [Fig. 2A(ii, iv and v)]. The pattern observed in the thighs resembles the ‘inside to outside’ pattern described in POGLUT1 mutations.²⁷ Oedema was also observed in the posterior compartment of the thighs on STIR (Short Tau Inversion Recovery) imaging of Patient P1 [Fig. 2A(iii)]. Patient P1 (Family SPA1) had a cervical spine MRI showing complete fat replacement of paraspinal muscles at this level [Fig. 2A(i)]. Fat replacement was observed in the lower leg of Patient P4 (Family ITA1), involving the soleus [Fig. 2A(vi)]. Lower leg muscle MRI of Patient P1 was normal (not shown).

Figure 2.

Muscle MRI and pathology associated with HMGCS1-related myopathy. (A) Muscle MRI performed for Patient P1 (Family SPA1) at 23 years of age [A(i–iii)] and Patient P4 (Family ITA1) at 36 years of age [A(iv–vi)]. Variation in T₁ signal indicative of fatty replacement in the posterior muscles of the cervical paraspinal muscles (stars) [A(i)], the posterior compartment of the thigh (arrows) [A(ii, iv and v)] and the posterior compartment of the leg (arrows) [A(vi)]. Evidence of oedema in the posterior compartment of the thighs in Patient P1 shown by STIR (arrows) [A(iii)]. Vastus intermedius is also affected to a lesser extent (arrow heads). (B) Light microscopy staining of muscle biopsies from Patients P1 (Family SPA1) at age 13 years (top), P3 (Family JPN1) at age 5 years and at 20 years (middle) and P4 (Family ITA1) at 4 years (bottom). Haematoxylin and eosin (H&E) and Gömöri trichrome (GT) staining show occasional internal nuclei and rimmed vacuoles in some patients. Succinate dehydrogenase (SDH) staining of SPA1 and nicotinamide adenine dinucleotide (NADH) staining of Families JPN1 and ITA1 patient biopsies reveal irregularities in staining including moth-eaten appearances and core-like regions (arrows). Scale bars = 20 μm or 50 μm, as indicated. yo = years old.

Four patients (Patients P1, P3, P4 and P5) underwent muscle biopsy (age range 4–20 years). The most noticeable findings were irregularities on oxidative enzyme staining, including core-like regions in some myofibres as shown by SDH and NADH staining (Fig. 2B). There were also occasional internal nuclei and rimmed vacuoles observed on haematoxylin and eosin and Gömöri Trichrome staining in some patients (Fig. 2B). The muscle pathology of Patient P3 worsened with age, showing the presence of rimmed vacuoles and dystrophic features at 20 years of age, which were absent at 5 years of age (Fig. 2B). Muscle biopsy for Patient P5 was indicative of a chronic myopathy with marked macrophage infiltration. There were also vacuoles and atrophic myofibres present (images unavailable). Analysis of inflammatory markers (MHCI, CD3 and MAC) in Patient P1 biopsy was unremarkable (Supplementary Fig. 1). There was no indication of myositis in any of the biopsies. Electron microscopy of muscle from Patient P1 showed some regions with small, sparse or dilated mitochondria, as well as one rare region of sarcomeric disorganization (Supplementary Fig. 1). Electron microscopy of muscle from Patient P4 showed areas of sarcomeric disruption and excess intermyofibrillar and subsarcolemmal glycogen (Supplementary Fig. 1).

Genetic investigations

We identified seven novel or rare biallelic variants in HMGCS1 (ENSG00000112972) co-segregating with rigid spine syndrome in four families (Figs 1A and 3A). This included six missense variants and one frameshift variant. In the index family (Family SPA1), we identified a homozygous missense variant in HMGCS1, c.1339T>C, p.(Ser447Pro) in the two affected siblings (Patients P1, P2). The variant was absent in gnomAD data (v2.1.1) and was heterozygous in the unaffected parents, as confirmed by Sanger sequencing. Additional rigid spine syndrome cases with biallelic HMGCS1 variants were subsequently identified via international collaborations (Family JPN1) and gene-matching platforms within Matchmaker Exchange,^28,29 including GeneMatcher³⁰ accessed via seqr³¹ (Families ITA1 and USA1). In Family JPN1, exome data for Patient P3 revealed a maternally inherited missense variant [c.86A>T, p.(Gln29Leu)] in trans with a paternally inherited frameshift variant [c.344_345del, p.(S115Wfs*12)], both variants were absent in gnomAD data (v4.0.0). The two unaffected siblings were identified as carriers for the frameshift variant (Fig. 1A). In Family ITA1, the affected proband (Patient P4) was identified with compound heterozygous missense variants [c.890G>T, p.(Gly297Val) maternal allele (absent in gnomAD v4.0.0), in trans with c.1289G>A, p.(Arg430Lys), paternal allele]. The p.Arg430Lys variant was present in gnomAD in the heterozygous state in 4 of 1 459 150 alleles (all from the European non-Finnish population, frequency 2.74 × 10⁻⁶). In Family USA1, Patient P5 had compound heterozygous missense variants: c.209T>C, p.(Met70Thr), maternal allele, in trans with c.803G>C, p.(Cys268Ser), paternal allele. The Met70Thr variant was present in the heterozygous state in 4 of 781 040 alleles (all from the European non-Finnish population, frequency 5.12 × 10⁻⁶). The p.Cys268Ser variant was present in the heterozygous state in 6 of 1 460 170 alleles (all from the European non-Finnish population, frequency 4.11 × 10⁻⁶). Apart from the M70 residue, all substituted residues in this cohort are conserved in multiple species, as well as in human HMGCS2, the mitochondrial paralogue (Fig. 3B). Further details including variant effect predictions are included in Supplementary Table 1.

Figure 3.

Localization of HMGCS1 variants within the gene and protein and their impact on HMGCS1 expression and function. (A) Schematic representations of the HMGCS1 gene (top) and protein (bottom) linear architectures. The HMGCS1 gene includes untranslated regions (empty boxes), coding exons (filled boxes) and introns (horizontal line). Variants identified in patients with rigid spine syndrome are labelled, including both missense (circle) and truncating (triangle) variants. The variants are coloured according to the families they have been identified in (red: Family JPN1; blue: Family USA1; green: Family ITA1; orange: Family SPA1). Letters ‘A’, ‘B’ and ‘s’ within the protein structure (bottom) represent residues of the active site (E95, C129, H264), co-enzyme A binding site (N167, S221, K269, K273) and salt bridge (D119, E121, R194, D208, K239, K461 and H462), respectively. Diagram not to scale. (B) Conservation of the six missense variant positions in various species and in HMGCS2; the mitochondrial paralogue. Asterisks indicate complete conservation of residues in the analysed species. (C) RNA-sequencing fold-change in HMGCS1 expression in Patient P3 (Family JPN1) compared to three unaffected controls (left). Error bars correspond to standard deviation. RNA-seq reads indicate virtually monoallelic expression of HMGCS1 c.86A>T, p.(Q29L) suggesting the c.344_345del, p.(S115Wfs*12) variant undergoes nonsense mediated decay (right). (D) Visualization of the HMGCS1 dimer (PDB: 2P8U) using PyMOL. HMGCS1 chain A (right) represented by a ribbon and chain B (left) shown as a cartoon. Ac-CoA represented as dark grey spheres. Variant positions (represented by sticks) coloured by family. Close-up representation showing the minimum distance between S447 and Q29 is ∼4.3 Å (dashed box).

The c.209T>C, p.(Met70Thr) variant was also identified via gene matching in an additional proband (in the USA, Family USA2; Supplementary Table 1) presenting with a similar unresolved rigid spine phenotype. This variant was heterozygous in the proband and inherited from the unaffected father. However, whole genome sequencing and RNA-seq of patient fibroblasts did not identify a second likely pathogenic HMGCS1 variant in trans for this patient, suggesting a possible more complex underlying genetic cause, such as a genomic rearrangement. However, it remains inconclusive whether HMGCS1 is the underlying genetic cause of rigid spine syndrome for this case.

The compound heterozygous frameshift variant [c.344_345del, p.(Ser115Trpfs*12)] in Patient P3 (Family JPN1) was predicted to result in nonsense mediated decay of the transcript. Indeed, RNA-seq of patient muscle showed only one read (out of 44) contained the frameshift variant while all reads (41/41) spanning the c.86 position contained the c.86A>T, p.(Gln29Leu) variant (Fig. 3C). In further support, DEseq³² analyses indicated that HMGCS1 transcript levels were reduced to 53% in the patient relative to three controls (Fig. 3C). Collectively, this suggests that at a protein level, in Patient P3, the c.86A>T, p.(Gln29Leu) variant is almost exclusively present.

To determine whether HMGCS1 dysfunction is associated with aberrant expression of other genes of the mevalonate pathway, we analysed RNA-seq for Patient P3 against 129 individuals by OUTRIDER.³³ We identified eight expression outliers in Patient P3 (Supplementary Fig. 2). However, no candidates showed any remarkable associations with the mevalonate pathway or other shared pathways according to gene ontology enrichment analyses.

Protein studies

The variants identified in the patients were in several exons in HMGCS1 (Fig. 3A). These include exon 3 [p.(Q29L), p.(M70T), p.(S115Wfs*12)], exon 6 [p.(C268S), p.(G297V)], exon 9 [p.(R430K)] and exon 10 [p.(S447P)]. The C268S substitution is adjacent to a co-enzyme A (CoA) binding residue, K269 (Fig. 3A). Positions of the remaining substitutions were distant from residues of the CoA active sites (E95, C129, H264), binding sites (S221, K269, K273) and residues of the dimer salt bridge (D119, Glu121, R194, D208, K239, K461 and H462). Three substituted residues (Q29, M70 and S447) appear to cluster towards the surface of HMGCS1 (Fig. 3D). Residues Q29 and S447 appear <5 Å apart in 3D space according to the PDB model (Fig. 3D).

HMGCS1 is expressed in various tissues, including developing and adult skeletal muscle

Human RNA-seq data from the NCBI BioProject (PRJEB4337)³⁴ and the GTEx portal³⁵ indicated that HMGCS1 is enriched in brain and liver and is expressed at lower levels in other tissues, including skeletal muscle. Similarly, qPCR analyses indicated enriched expression of HMGCS1 in human cortex, and relatively lower expression in cells and tissues, including myotubes and adult skeletal muscle (Fig. 4A). Of the myogenic samples analysed in the FANTOM5³⁶ database, HMGCS1 expression was highest in skeletal muscle satellite cells [235–462 transcripts per million (TPM)] relative to fetal-derived skeletal muscle cells (3–326 TPM), myotubes (0–240 TPM) and myoblasts (31–50 TPM).

Figure 4.

HMGCS1 is expressed in various tissues including patient skeletal muscle. (A) Quantitative real-time PCR analysis using cDNA from healthy human cell lines and tissues. Expression data from human embryonic kidney cells (HEK293FT cell line), fibroblasts, myoblasts, myotubes at D2, D4, D6, D8 of differentiation, cortex, skeletal muscle controls from in vitro contractile testing and fetal muscle. Transcript levels were normalized to the geometric mean of EEF2 and TBP using the delta Ct method. Data presented as mean ± standard error of the mean (n = 2–7 biological). (B) Graphical presentation of HMGCS1 peptides detected by quantitative mass spectrometry from myoblasts (MB) and myotubes at Day 2 (D2) and Day 8 (D8) of differentiation. (C–F) Western blotting for HMGCS1 (top) in (C) primary human (Cook Myosite) myoblasts and myotubes from Days 0–12 of differentiation, in (D) control human tissue and cell lines, and in (E) Patient P1 (Family SPA1) skeletal muscle alongside healthy skeletal muscle controls. Recombinant wild-type HMGCS1 (Rec. HMGCS1) used as a control for antibody specificity. (F) Western blotting for HMGCS1 in mouse extensor digitorum longus (EDL) and soleus muscles (top). Western blots for GAPDH and gel-stained myosin heavy chain (MHC) bands are shown to demonstrate comparable loading of total and muscle proteins.

Western blotting for HMGCS1 in human tissues and cells revealed a band of the expected size (∼57 kDa; Fig. 4C–F), which matched the product size of the recombinant HMGCS1 protein (Fig. 4E). We show that HMGCS1 is enriched in the cortex (Fig. 4D), consistent with the qPCR data and the proteomic data from the Human Protein Atlas.³⁷ In addition, HMGCS1 was detected in adult human skeletal muscle but was relatively less abundant in human fetal skeletal muscle (Fig. 4D), consistent with the qPCR studies. HMGCS1 was also detected in cultured human myoblasts and myotubes from Days 0 to 12 of differentiation by mass spectrometry (Fig. 4B) and/or western blotting (Fig. 4C). There was comparable abundance of HMGCS1 in skeletal muscle of Patient P1 (Family SPA1) and control skeletal muscle biopsies (Fig. 4E) suggesting the p.(S447P) substitution does not significantly impact protein stability or abundance. Similar levels of Hmgcs1 were detected in mouse EDL (extensor digitorum longus) and soleus muscles (Fig. 4F), suggesting that Hmgcs1 is similarly abundant in fast- and slow-twitch myofibres in mice.

Functional investigations of recombinant HMGCS1

HMGCS1 has been reported to exist as a dimer.^38,39 To investigate whether substitutions in HMGCS1 affect dimerization, the recombinant wild-type (HMGCS1^WT) and mutant proteins (HMGCS1^S447P, HMGCS1^Q29L, HMGCS1^M70T and HMGCS1^C268S) were analysed by size exclusion chromatography to evaluate monomer and dimer content. All proteins eluted at equivalent points corresponding to the expected molecular weight for dimerized HMGCS1 (Supplementary Fig. 3A and B). This suggested that the substitutions do not interfere with HMGCS1 dimerization. There were second peaks detected for HMGCS1^WT and HMGCS1^Q29L (Supplementary Fig. 3A); however, their elution volumes did not correspond to the expected molecular weight of the monomer (Supplementary Fig. 3B).

We performed thermal shift assays to investigate whether missense substitutions in HMGCS1 reduce protein thermal stability, given that thermal shift assays performed by Morales-Rosado et al.⁴⁰ showed that missense substitutions in HMGCR reduce protein thermal stability.⁴⁰ Three of the four HMGCS1 substitutions investigated (S447P, M70T and C268S) showed reduced melting temperatures compared to the wild-type (Fig. 5A and B). The differences ranged between 1.4–3.5°C compared to the wild-type and were statistically significant by one-way ANOVA (Fig. 5B).

Figure 5.

Thermal shift and circular dichroism analyses of recombinant wild-type and mutant HMGCS1. (A) Thermal shift assay first derivative fluorescence curves measured for HMGCS1 wild-type and four mutants (12.5 μg) over increasing temperature. The minimum point of the first derivative curves was estimated as the protein melting temperature (T_m). Thermal shift assays were conducted three independent times using four replicates per assay. The data plotted in (B) represent the T_m averages from each assay (n = 3) and are presented as mean ± standard error of the mean. Significance was determined by an ordinary one-way ANOVA followed by Dunnett’s multiple comparison test. (C–F) Circular dichroic spectra scans (195–260 nm) of wild-type and four mutant HMGCS1 (0.1 mg/ml) measured at (C) 15°C, (D) 25°C, (E) 55°C and (F) 95°C in buffer containing 10 mM sodium phosphate buffer pH 7.4, (including 1% of original protein buffer: 150 mM NaCl, 50 mM Tris and 10% glycerol).

Circular dichroism (CD) spectroscopy was performed to investigate whether the HMGCS1 variants affected protein secondary structure. Overall, the wild-type and all mutants investigated appeared structurally stable, showing considerable overlap in the CD spectra (190–260 nm) throughout the melting range (15–95°C; Fig. 5C–F). The majority of the protein remained intact even at high temperatures (>60°C), showing marginal changes in CD signal, suggesting the proteins have a thermally stable core (Fig. 5C–F). Owing to the small changes in CD signal and the variability between experiments, there were no statistically significant differences in protein structure or stability detected by circular dichroism between the mutants and the wild-type (one-way ANOVA; Supplementary Fig. 4B).

Analyses of HMGCS1 enzymatic activity indicated that all four variants tested were active within a similar order of magnitude as the wild-type when assayed with 25 μM AcAc-CoA and 200 μM Ac-CoA (Supplementary Fig. 4C and D). Interestingly, the Q29L mutant consistently showed an increase in activity during the initial (fast) phase of the reaction (Supplementary Fig. 4C and D). Analysis of the initial rate constant (K_fast) across four independent assays (n = 4/4) indicated a significant increase in Q29L K_fast compared to the wild-type (P < 0.001; Supplementary Fig. 4D). Additionally, the K_fast of the C268S mutant was significantly reduced compared to the wild-type (P = 0.016; Supplementary Fig. 4D). During the second (slow) phase of the reaction, no differences in K_slow were observed between the wild-type and any variant (Supplementary Fig. 4D).

Hmgcs1^−/− zebrafish present severe movement defects, mitochondrial abnormalities and an early lethal phenotype

To evaluate the effect of loss of HMGCS1 function in an animal model during developmental stages, we generated a hmgcs1 mutant zebrafish, containing a 302 bp deletion in exon 3, which results in a frameshift after 35 amino acids (aa) (Supplementary Fig. 5A). Loss of Hmgcs1 in zebrafish embryos results in severe early developmental defects, with hmgcs1^−/− embryos displaying blood pooling in the brain, reduced pigmentation (Fig. 6A), absence of movement at 48 hpf (Fig. 6B), and lethality at 3 dpf (Fig. 6C). Further investigation of the movement defect identified that hmgcs1^−/− embryos completely lack a touch-evoked escape response²⁵ compared to hmgcs1^+/+ and hmgcs1^+/− siblings.

Figure 6.

hmgcs1^−/− zebrafish have severe mitochondrial abnormalities and an early lethal phenotype. (A) Brightfield images of wild-type and hmgcs1^−/− embryos at 48 h post-fertilization (hpf). hmgcs1^−/− embryos have developmental delay, reduced pigmentation and blood pooling in the brain. (B) Quantification of movement to a touch-evoke assay. hmgcs1^−/− embryos lack a movement response to touch, unlike wild-type and heterozygous siblings. Injection of mRNA encoding human HMGCS1 rescues the movement defects in hmgcs1^−/− embryos (Student’s t-test). Twenty-four injected embryos were used for each assay. (C) Survival assay on the progeny of a hmgcs1^+/− in-cross. In control injected fish only 79% (n = 19) survived until Day 4 and no hmgcs1^−/− fish were detected at this time, as indicated in brackets, demonstrating that loss of Hmgcs1 results in early embryo lethality. Injection of HMGCS1 mRNA rescues the lethality in hmgcs1^−/− embryos, with 100% (n = 24) of injected fish surviving until Day 4, and subsequent genotyping indicating 33% (n = 8) of the fish were hmgcs1^−/−. Twenty-four embryos from the hmgcs1^+/− in-cross were used for each assay. (D) There is a significant reduction in the survival of embryos resulting from a hmgcs1^+/− in-cross injected with HMGCS1 variant mRNA compared to injection of wild-type mRNA. P-values are listed next to the figure legend. Statistics were performed in SPSS using a generalized linear model. (E) Scanning electron microscopy images of longitudinal muscle sections at 48 hpf, demonstrating mitochondria associated with skeletal muscle. hmgcs1^−/− embryos have an increased proportion of mitochondria with detached membranes. Magenta arrowheads indicate mitochondria, cyan arrowheads indicate myofibres. Scale bar = 1 μm. Three samples were analysed for each genotype. (F) Mitochondrial density is not affected in hmgcs1^−/− embryo skeletal muscle (Student’s t-test). (G) The proportion of mitochondria with detached membranes is higher in hmgcs1^−/− embryos compared to wild-type (Fisher’s exact test). Statistics were performed in Prism 9 (GraphPad). Statistically significant differences are indicated on the graphs by the P-value.

Injection of wild-type HMGCS1 mRNA (18 ng/μl) was able to rescue early lethality in hmgcs1^−/− embryos (Fig. 6C), as well as the movement defects in response to touch (Fig. 6B), demonstrating that these defects specifically result from the loss of Hmgcs1 function. Injection of wild-type HMGCS1 RNA at a lower concentration (3 ng/μl) was also able to rescue hmgcs1^−/− lethality at 4 dpf (Fig. 6D). All four variants tested (S447P, Q29L, M70T and C268S) increased the proportion of fish alive at Day 4 compared to a no-RNA injection control, but did not restore hmgcs1^−/− survival to the same level as wild-type (Fig. 6D and Supplementary Fig. 5B). This demonstrates that these missense substitutions are hypomorphic.

Electron microscopy to investigate the muscle ultrastructure revealed that hmgcs1^−/− embryos exhibited significant mitochondrial abnormalities within the skeletal muscle at 48 hpf, presenting as a detachment or invagination of the inner membrane (Fig. 6E and G), without affecting mitochondrial density (Fig. 6F and Supplementary Fig. 5C), or the proportion of elongated mitochondria (Supplementary Fig. 5D and E). Interestingly, the myofibril pattern in hmgcs1^−/− animals was indistinguishable from wild-type siblings, indicating that the muscle structure was not affected [Fig. 6E (cyan arrowheads) and Supplementary Fig. 5F]. These results indicate that absence of Hmgcs1 results in damage to the integrity of the mitochondrial membrane and loss of muscle function.

Mevalonic acid treatment improves survival and swimming performance in hmgcs1^−/− zebrafish

Given previous demonstration that mevalonic acid treatment was able to treat the myopathy phenotype in a HMGCR1 knockout mouse⁴¹ and gave promising results in a case of HMGCR limb girdle muscular dystrophy,²¹ we decided to test mevalonic acid treatment on the hmgcs1 mutant zebrafish. We treated the progeny of an hmgcs1^+/− in-cross daily from 6 hpf to 6 dpf monitoring survival, and at 6 dpf conducted a swimming analysis on the remaining fish. Treatment with 50, 500 or 1000 μM mevalonic acid all significantly increased the survival of hmgcs1^−/− fish compared to vehicle treated control hmgcs1^−/− fish (P < 0.0001, Fig. 7A). None of the treatments had any effect on the survival of hmgcs1^+/− or hmgcs1^+/+ animals (Supplementary Fig. 7A and B). Treatment with 500 or 1000 μM extended survival of the hmgcs1^−/− to 6 dpf. Hmgcs1^−/− animals often demonstrated severe oedema and head deformity at 6 dpf, although the phenotype was less severe in the hmgcs1^−/− 1000 μM treated fish (Fig. 6B). We analysed the swimming performance of all surviving fish at 6 dpf (Supplementary Fig. 7C). Analysis of the data using a linear mixed model identified significant effects of genotype (P < 0.001) and treatment (P < 0.001), and an interaction between the two (P < 0.001). Follow-up analysis identified that in the surviving hmgcs1^−/− fish, the 1000 μM treated fish swam significantly further than then 500 μM treated fish (P < 0.01), which were barely motile (Fig. 7C). None of the mevalonic acid treatments had any effect on the swimming performance of the hmgcs1^+/+ fish, indicating no apparent toxicity (Supplementary Fig. 7D). Therefore, there was a dose-dependent improvement in the phenotype of the hmgcs1^−/− animals; 50 μM mevalonic acid was sufficient to extend survival by 24 h, 500 μM to 6 dpf, and 1000 μM to 6 dpf and retain some—minimal—swimming capacity. We proceeded to examine the skeletal muscle and mitochondrial phenotype in the 1000 μM treated fish by electron microscopy (Fig. 7D). As before, we identified no effect of genotype (or treatment) on sarcomeric structure (Supplementary Fig. 7E), mitochondrial density (Fig. 7) or the proportion of elongated mitochondrial networks (Supplementary Fig. 7F). However, we identified a significant interaction between genotype and treatment in the proportion of mitochondria with detached membranes (P = 0.043) with a significant reduction in the prevalence of detached membranes in mevalonic acid-treated fish compared to vehicle-treated controls (P = 0.039).

Figure 7.

Mevalonic acid treatment improves hmgcs1^−/− zebrafish survival and reduces phenotypic severity. (A) Survival analysis of hmgcs1^−/− zebrafish embryos treated with 50, 500 and 1000 μM mevalonic acid, E3 control and vehicle control (ethanol) until 6 days post-fertilization (dpf). Control hmgcs1^−/− embryos do not survive past 2 dpf. Treatment with 50 μM mevalonic acid significantly extends survival until 3 dpf, whereas 500 and 1000 μM treatments extended survival until 6 dpf, at which point the experiment was ceased. Statistically significant differences are indicated on the graphs by the P-value next to the figure legend. (B) Phenotypic severity of hmgcs1^−/− embryos rescued with mevalonic acid treatment at 6 dpf. Treatment with 1000 μM mevalonic acid further reduces the phenotypic severity of hmgcs1 loss-of-function compared to 500 μM. (C) Swimming activity in hmgcs1^−/− embryos at 6 dpf treated with mevalonic acid. hmgcs1^−/− embryos treated with 1000 μM mevalonic acid (n = 66) swam significantly further than embryos treated with the 500 μM treatment (n = 40). Rep = biological replicate. Statistics were performed in SPSS using a linear mixed model. (D) Transmission electron microscopy images of longitudinal muscle sections of vehicle control and 1000 μM mevalonic acid treated embryos at 48 hpf, demonstrating mitochondria associated with skeletal muscle. hmgcs1^−/− embryos have an increased proportion of mitochondria with detached membranes. Magenta arrowheads indicate mitochondria, cyan arrowheads indicate myofibres. Scale bar = 1 μm. Three samples were analysed for each genotype. (E) Mitochondrial density is not affected by mevalonic acid treatment. (F) The percentage of mitochondria with detached membranes is significantly reduced in mevalonic acid treated hmgcs1^−/− embryos compared to vehicle-treated controls. Statistics were performed in in SPSS using a two-way ANOVA. For E and F, the graphs show the estimated means and 95% confidence intervals together with the raw data.

Discussion

We report the identification of seven biallelic variants in HMGCS1 in four unrelated families with rigid spine syndrome. Although the cohort thus far includes five patients, the distinct rigid spine phenotype shared by the patients suggests a common genetic basis of disease, which we define here as HMGCS1-related myopathy. There is notable clinical resemblance between patients with SELENON-related and HMGCS1-related myopathy. Both disorders are characterized by early onset rigid spine and scoliosis, muscle hypotrophy and predominantly proximal and axial muscle weakness.⁴² Both disorders also show respiratory involvement and susceptibility to recurrent respiratory infections, which can be fatal, and the respiratory insufficiency developing during the first or second decades of life seems out of proportion for the degree of muscle weakness. Fluctuations or acute deteriorations of muscle strength and respiratory function during intercurrent disease has not been described in patients with SELENON-related disease and may be a clue to suggest the presence of HMGCS1 pathogenic variants. Additional descriptions of HMGCS1-related disease may further clarify identifying clinical features of this newly described entity.

HMGCS1 encodes the cytosolic HMG-CoA synthase; a key enzyme of the mevalonate pathway, which catalyses the condensation reaction between AcAc-CoA and Ac-CoA to form HMG-CoA.³⁸ Its mitochondrial paralogue, HMGCS2 catalyses the same reaction for the ketogenesis pathway. Biallelic variants in HMGCS2 are associated with HMG-CoA synthase deficiency (OMIM# 605911) characterized by episodic metabolic distress, febrile crisis, hepatomegaly and respiratory distress.^43-45 HMGCS1 and HMGCS2 share ∼67% sequence similarity and five of the six substituted HMGCS1 residues presented here (Q29, C268, G297, R430 and S447) are conserved in HMGCS2 (Q66, C305, G334, R467 and S484, respectively) suggesting that they are of functional importance. None of the HMGCS1 substitutions observed in this study overlap with the reported HMGCS2 pathogenic substitutions. Nevertheless, the association of biallelic variants in HMGCS2 with disease suggests that biallelic variants in HMGCS1 may be similarly deleterious. This is further supported by gnomAD (v4.0.0) constraint scores, which indicate that HMGCS1 (pLI = 1; Z = 4.12) is substantially more intolerant to loss-of-function and missense variation compared to HMGCS2 (pLI = 0; Z = 0.53).

Myopathic and rigid spine phenotypes have been associated with several inborn and acquired errors of the mevalonate pathway.^{11,20,21,46-54} This includes biallelic variants in HMGCR (HMG-CoA reductase) immediately downstream of HMGCS1 in the pathway (Supplementary Fig. 8), which have recently been associated with limb-girdle muscular dystrophy (OMIM# 620375).^21,40 Inhibition of HMGCR by statins and autoantibodies is also associated with acquired muscle disorders including statin-induced myopathy²⁰ and anti-HMGCR myopathy.^49,51,55,56 Further downstream, inhibition of FDPS (farnesyl diphosphate synthase) by bisphosphonate medications has been associated with profound muscle weakness with fever and flu-like symptoms.^57,58 Moreover, biallelic variants in GGPS1 have been associated with a muscular dystrophy (OMIM# 619518).^11,59 Of note, GGPS1-related muscular dystrophy includes several overlapping phenotypes including scoliosis, joint contractures, hypotonia, elevated CK, as well as episodic worsening of muscle weakness and respiratory insufficiency during intercurrent disease.^11,59 Overall, the myopathic phenotypes implicated in the mevalonate pathway support HMGCS1 as an underlying genetic cause for our patients’ disorders.¹¹

The mevalonate pathway branches into several critical other pathways including the cholesterol biosynthesis pathway, the isoprenoid pathway and the ubiquinone pathway, which are essential for regulating cell proliferation, maturation and maintenance.⁶⁰ There are various hypotheses that suggest how disruptions to the mevalonate and/or downstream pathways may cause skeletal muscle disease.^20,61 These include alterations in the cholesterol abundance and fluidity of the sarcolemma, impaired mitochondrial function, and dysregulated isopentylation of selenocysteine-tRNA, which is required for SELENON expression.^20,62,63 Disruption of the mevalonate pathway, whether through statin inhibition⁶⁴ or mutation,⁶⁵ also leads to mitochondrial abnormalities, consistent with the findings observed in hmgcs1^−/− zebrafish mutants and in one patient.

Knockout of most genes of the mevalonate pathway is associated with complete pre-weaning lethality in mice (Supplementary Fig. 8).^11,66 This is consistent with the severe phenotype observed in our studies of hmgcs1^−/− zebrafish, which included developmental delay, immobility by 2 dpf, and fatality during 3 dpf. We suspect that biallelic loss-of-function variants in humans would cause prenatal lethality consistent with the prior animal studies⁶⁷ and our zebrafish loss-of-function model. There are no individuals in gnomAD (v4.0.0) homozygous for loss-of-function HMGCS1 variants, further supporting that HMGCS1 function is critical to life.

We functionally investigated four of six HMGCS1 substitutions (S447P, Q29L, M70T and C268S) identified in our cohort. We demonstrate reduced function for all tested HMGCS1 variants (S447P, Q29L, M70T and C268S) in a hmgcs1^−/− zebrafish rescue assay. Three variants (S447P, M70T and C268S) also demonstrate reduced thermal stability compared to the wild-type. A similar reduction in thermal stability has been associated with missense substitutions in HMGCR.⁴⁰ The only variant in which we detected reduced enzyme activity in our assays was C268S, in which the substitution neighbours the active site (H264) and CoA binding site (K269). Surprisingly, Q29L showed a 2-fold increase in K_fast compared to the wild-type. Although the changes detected in the activity of C268S and Q29L were partial, notably, hypomorphic activity for a p.R505Q substitution in HMGCS2 (p.R468 in HMGCS1) has been associated with disease.⁶⁷ Additionally, partial or no changes in mutant enzyme activity were identified by recombinant GGPS1 assays conducted for mutants associated with GGPS1-related muscular dystrophy.¹¹ Although the Q29L variant did not have reduced thermal stability in our assay, and even had increased enzymatic activity, it is the only variant in our patient cohort present in combination with a null allele (p.S115Wfs*12). Collectively, our data demonstrate that all the missense variants examined are hypomorphic.

Partial changes in temperature sensitivity and/or enzyme activity have been proposed as a disease mechanism for other enzymes of the mevalonate pathway, including MVK-related disease (OMIM# 260920)⁶⁸ and NSDHL-related disease (OMIM# 300831).⁴⁶ Minor elevations in physiological temperature are suggested to affect metabolic flux, leading to a temporary deficiency or accumulation of isoprenoid products. This may cause cellular toxicity or dysregulate the activity of other enzymes and lead to metabolic stress, subsequently triggering disease.⁶⁸ We suggest a similar hypomorphic mechanism for HMGCS1-related disease, particularly given the rapid deterioration and worsening of disease reported in some patients during episodic disease and febrile attacks. Of note, HMGCS1 has been suggested to serve as a flux-controlling point in the mevalonate pathway,⁶⁰ further supporting that subtle functional changes may have significant downstream metabolic effects.

Following promising results from mevalonic acid treatment in a single HMGCR LGMD patient,²¹ we tested mevalonic acid treatment on the hmgcs1^−/− zebrafish and identified significant improvements in both survival and swimming ability. Given that the missense variants characterized result in a partial loss-of-function, compared to the complete loss in our animal model, we would anticipate patients may be more amenable to treatment.

Conclusion

We report a novel autosomal recessive rigid spine syndrome we have termed HMGCS1-myopathy. This report expands the genetic causes of rigid spine syndrome and we encourage screening of HMGCS1 in patients with molecularly unresolved rigid spine syndrome. In addition, we demonstrate in an animal model that mevalonic acid treatment can reduce phenotypic severity, highlighting a potential treatment for these patients.

Data availability

The data and constructs generated by this study are available upon request. Zebrafish experiment data are available from the Monash University Bridges data repository (https://doi.org/10.26180/c.7354282).

Funding

L.N.H.D. is supported by an Australian Government Research Training Program (RTP) Scholarship at the University of Western Australia. G.R. (Investigator Grant, APP2007769), N.G.L. (Fellowship APP1117510) and D.S. (Investigator Grant, APP2009732) are supported by the Australian National Health and Medical Research Council. This work is also supported by an National Health and Medical Research Council Ideas Grant (APP2002640). Sequencing and analysis of Family JPN1 were supported partly by Intramural Research Grant (2–5 and 5–6) for Neurological and Psychiatric Disorders of NCNP, and Japan Agency for Medical Research and Development under Grant Numbers 22ek0109490h0003. Sequencing and analysis of Families USA1 and USA2 were provided by the Broad Institute of MIT and Harvard Center for Mendelian Genomics (Broad CMG) and were funded by the National Human Genome Research Institute grants UM1 HG008900 and R01 HG009141 (with additional support from the National Eye Institute, and the National Heart, Lung and Blood Institute), U01 HG0011755, and in part by grant number 2020-224274 from the Chan Zuckerberg Initiative grant DAF2019-199278 (https://doi.org/10.37921/236582yuakxy, an advised fund of Silicon Valley Community Foundation (funder DOI 10.13039/100014989). This study makes use of data shared through the seqr platform with funding provided by National Institutes of Health grants R01HG009141 and UM1HG008900. The work in C.G. Bönnemann’s laboratory is supported by intramural funds from the National Institutes of Health National Institute of Neurological Disorders and Stroke. The electron microscopy has been made possible in part by CZI grant DAF2021-225399 and grant DOI 10.37921/334038myxhsa from the Chan Zuckerberg Initiative DAF, an advised fund of Silicon Valley Community Foundation (funder DOI 10.13039/100014989).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.