Recessive variants in mitochondrial complex I nuclear subunits are an underrated cause of optic atrophy

Abstract

Our understanding of the genetic landscape of inherited optic neuropathies (ION) has grown significantly over the past decades, and it is now known to involve many genes found in both the nuclear and mitochondrial genomes, exhibiting all possible inheritance patterns. Furthermore, pathogenic variants in nuclear genes encoding mitochondrial respiratory complex I (CI) subunits have been identified in some cases of ION, in addition to the more common severe presentation of CI deficiencies, which usually have an early onset.

We conducted next-generation sequencing screening of CI genes to identify potential causative variants in patients with optic atrophy, and performed comprehensive clinical assessments, including neuroimaging (MRI) and neurological evaluations. Detailed molecular structure modelling was performed to better evaluate the damaging effects of both novel and previously reported variants in the relevant CI subunits.

We identified candidate causative variants in 31 patients from 23 unrelated families, with biallelic or hemizygous variants characterized in 11 nuclear CI-related genes encoding polypeptides involved in CI structure, including three core subunits (NDUFS7, NDUFV1, NDUFV2), four accessory subunits (NDUFA1, NDUFA10, NDUFA12, NDUFB11) and four assembly factors (NDUFAF2, NDUFAF3, NDUFAF4, NDUFAF8). Notably, defects in core CI subunits in this cohort led to isolated optic atrophy, while defects in accessory CI subunits and assembly factors resulted in a spectrum of phenotypes, from isolated to syndromic optic atrophy. In 12 cases, the subacute onset of vision loss enabled us to associate or confirm novel genes (NDUFS7, NDUFV1, NDUFAF2, NDUFAF4, NDUFAF8) with the autosomal recessive Leber hereditary optic neuropathy (arLHON) phenotype. Moreover, in the NDUFS7 subunit, partial spatial segregation was observed for missense variants associated with either Leigh syndrome or isolated optic atrophy, suggesting possible disease-specific molecular defects.

Our case series broadens the genetic spectrum of ION, emphasizing the crucial role of nuclear CI genes in pathogenesis. The arLHON phenotype, reportedly associated with an insidious onset of optic atrophy, is linked to numerous nuclear CI genes, and in some cases, the same variant may underlie both phenotypes. Overall, we highlight the possibly underestimated prevalence of CI nuclear subunits in the molecular diagnosis of ION, prompting the inclusion of all CI-related genes in the standard diagnostic screening.

Fiorini et al. identify mutations in 11 nuclear-encoded mitochondrial complex I genes as an under-recognised cause of inherited optic neuropathies. These variants cause both isolated and syndromic forms, including autosomal recessive Leber hereditary optic neuropathy, supporting inclusion of all complex I genes in routine genetic testing.

Introduction

Inherited optic neuropathies (ION) include two major entities: the first is maternally inherited Leber hereditary optic neuropathy [LHON; Mendelian Inheritance in Man (MIM) #535000], typically caused by homoplasmic mitochondrial DNA (mtDNA) point mutations. The second is autosomal dominant optic atrophy (ADOA; MIM #165500), which is predominantly caused by heterozygous pathogenic variants in the OPA1 gene encoded by nuclear DNA (nDNA).^1-3 Both affect respiratory complex I (CI, EC 7.1.1.2), the largest enzymatic complex of the mitochondrial electron transport system, which in humans comprises 45 distinct subunits. Those are assembled into functionally and evolutionarily distinctive modules, namely the NADH dehydrogenase module (N), the quinone-binding module (Q) and the proton-pumping module (P), the latter being further subdivided into two proximal submodules (ND1 and ND2) and two distal submodules (ND4 and ND5).⁴ Indeed, the three most common mtDNA pathogenic variants associated with LHON affect genes (MT-ND1, MT-ND4, MT-ND6) encoding CI core subunits, directly impairing the catalytic function of the complex.¹ In contrast, OPA1 variants associated with ADOA do not directly affect CI, as its enzymatic activity remains normal,⁵ but instead influence CI-driven oxidative phosphorylation efficiency, likely due to disrupted mitochondrial dynamics and cristae morphology.^6,7

In addition to the classic clinical manifestation of isolated optic atrophy (OA), both LHON and ADOA can also manifest with syndromic phenotypes described under the broader LHON and ADOA ‘plus’ groups.^1-3 These phenotypes may arise from more severe and rarer mtDNA pathogenic variants for LHON plus^8-10 or from specific missense pathogenic variants, mostly affecting the GTPase domain of the OPA1 protein in ADOA plus.^11,12

Beyond the well-established associations of mtDNA-LHON and OPA1-ADOA, the genetic landscape of optic neuropathies has expanded to include several other nDNA-encoded genes or loci (also identified as OPA2-16), which underly both autosomal dominant, recessive and X-linked inheritance patterns.^13,14 In this broader context of newly identified ION-related genes, the role of CI dysfunction in the pathogenesis of optic nerve pathology remains of utmost importance.^15-17

Overall, an increasing number of nuclear CI-related genes have recently been recognized as genetic causes of autosomal recessive LHON (arLHON), including core subunits, accessory subunits, and assembly factors (i.e. NDUFS2, NDUFS7, NDUFA12, NDUFAF2 and NDUFAF5).^18-22 Furthermore, a gene involved in CI N-module turnover and quality control, DNAJC30, has emerged as a relatively frequent cause of arLHON (LHONAR1, MIM #619382), indistinguishable from LHON caused by mtDNA pathogenic variants.^23,24 Finally, TMEM126A, responsible for isolated or syndromic OA with recessive inheritance (OPA7, MIM #612989),^25-27 has been found to encode a key factor for CI biogenesis, interacting with the ND4 module.^28-30

Finally, it is important to note that most pathogenic variants in nuclear subunits of CI are typically associated with CI deficiency and severe early-onset phenotypes, mainly attributable to autosomal recessive Leigh syndrome (LS) or the Leigh-like syndrome (LLS) spectrum.³¹ While the clinical presentation of LS is highly variable and complex, visual impairment due to OA is a recurrent feature in ∼15% of patients, although ophthalmic evaluation may not always be performed, especially in very young and severely affected patients.³²

In the current study, we identified a significant role of biallelic defects in CI nuclear-encoded subunits as a mechanism of ION, including arLHON. Such findings have previously been recognized in only a handful of cases.¹⁴ Moreover, in this international cohort, we describe patients with an OA plus or LLS presentation harbouring variants in otherwise scarcely reported CI subunits. Finally, we present evidence that a structural approach, considering subunit location as well as specific amino acid substitutions, can shed light on the phenotypic and genetic diversity observed.

Materials and methods

Clinical evaluation

Clinical assessments included best-corrected visual acuity (BCVA) by Snellen charts extrapolated to decimal values, colour vision tests, slit-lamp biomicroscopy, tonometry, colour fundus photography, optical coherence tomography (OCT; Cirrus, Carl Zeiss Meditec Inc.; DRI Triton, Topcon; or SPECTRALIS spectral-domain-OCT, Heidelberg Engineering), automated visual field test (Humphrey Field Analyzer, SITA Standard 30-2/24-2 protocol, Zeiss; or M-700, Medmont International). OCT protocols included the evaluation of peripapillary retinal nerve fibre layer (pRNFL) thickness (3.4 acquisition protocol).

Genetic investigations

Genetic testing of each patient was approved by the ethical committees of the respective institutes and authorized by informed consent. Nuclear gene sequencing was performed after complete mtDNA sequencing, which excluded the presence of pathogenic variants. Among the institutions, next-generation sequencing (NGS) screenings were performed as follows: target gene panel for mitochondrial disorders at IRCCS Istituto Neurologico Carlo Besta, Milano, Italy (Families B, H, I and L) and Charles University and General University Hospital, Prague, Czechia (Family M); clinical exome at Moorfields Eye Hospital, London, UK (Families C and T) and IRCCS Ospedale Pediatrico Bambino Gesù, Roma, Italy (Families D, G, H and J); whole exome sequencing at IRCCS Istituto delle Scienze Neurologiche di Bologna, Bologna, Italy (Families A, E, K, G, N, O, P, R and S) and TUM Klinikum, Munich, Germany (Family U); and whole-genome sequencing at Moorfields Eye Hospital, London, UK (Families F, Q, V and W).

Methods used in NGS data generation and processing, including variant filtering and prioritization are reported in the Supplementary material, ‘Methods’ section. The details of all NGS screenings performed and quality metrics obtained are reported in Supplementary Table 1.

Variant classification was performed according to American College of Medical Genetics (ACMG) guidelines, using the Rare Exome Variant Ensemble Learner (REVEL) score of meta-prediction for the application of PP3/BP4 criteria, then adapted through the use of targeted modelling (see the ‘In silico prediction of pathogenicity’ section).^33-35

In silico prediction of pathogenicity

To unravel the functional role of newly identified variants associated with OA and to search for possible functional differences between them and those previously reported in the literature as causative for LS/LLS, we conducted a comprehensive in silico analysis. The current literature (updated to February 2025) was searched to produce a list of pathogenic variants in the genes found in the current study. For each article, we collected the PubMed Identifier (PMID), genetic variants, main phenotype and information about the proband and alleles (summarized in Supplementary Table 2).

We considered the impact of missense variants on protein folding by calculating the ΔΔG induced by every variant using DynaMut2³⁶ and DUET,³⁷ with the latter using predictions made by mCSM³⁸ and SDM.³⁹ These tools were chosen because they performed the best among the publicly available web services in benchmarking experiments.⁴⁰ We considered the variants to impact protein folding only when both predictors coherently revealed ΔΔG values exceeding (±) the root mean square error (RMSE) indicated for each predictor, i.e. ±1.02 kcal/mol and ±0.98 kcal/mol for DynaMut2 and DUET, respectively. When the prediction was inconsistent across the two tools, we considered the effect of that variant as uncertain. The structures used to exploit the tools were: Protein Data Bank (PDB) ID 5XTD for human CI⁴¹; and those predicted by AlphaFold 3 (AF3)⁴² for NDUFAF2, NDUFAF3, NDUFAF4 and NDUFAF8. For these predicted structures, we performed the ΔΔG calculation only if the segment affected by the variant had a predicted Local Distance Difference Test (plDDT) score >70, indicating high confidence in the prediction. Lastly, we mapped the amino acids affected by variants on human CI structure and predicted structures, performed in silico mutagenesis, and where possible, investigated their involvement in the formation of protein–protein interactions (PPIs) and the characteristics of the interfaces using the Proteins, Interfaces, Structures and Assemblies tool (PDBePISA).⁴³ Mitochondrial targeting sequences (MTS) were identified using MTSviewer⁴⁴ and variants reported in the 3D structure predicted by AF3. For the NDUFAF3 N-terminal, which is not modelled as an α-helix by AF3, we predicted the secondary structure using JPred4⁴⁵ and generated a structural model of the first 32 amino acids using UCSF ChimeraX 1.8.⁴⁶ The same software was used for the visual inspections and modelling of the variants, using the swapaa tool.

For the adjustment of the ACMG PP3/BP4 criteria of variants described in our cases, we combined the evidence from the two ΔΔG modelling tools with that from two sequence-based tools, namely AlphaMissense⁴⁷ and PrimateAI-3D (PAI-3D),⁴⁸ which predict the structural impact of amino acid changes. The calibrated prediction scores (range: 0–1) and pathogenicity prediction categories (likely benign, ambiguous or likely pathogenic for AM and benign or pathogenic for PAI-3D) are reported in Supplementary Tables 5–7. When all approaches for structural evaluation were concordant in identifying a variant as damaging, we applied the upgraded PP3 criterion reported in Supplementary Table 3. The same algorithm was applied for variants affecting the MTS, substituting this specific model for ΔΔG-based prediction.

Quantitative proteomics

Tandem mass tag (TMT)-labelled mass spectrometry was performed at the BayBioMS core facility of the Technical University of Munich as described by Kopajtich et al.,⁴⁹ with two minor modifications. Peptide fractionation was carried out using high pH reversed phase instead of trimodal mixed-mode chromatography, and TMT-labelling was carried out using TMT 11-plex instead of TMT 10-plex reagent. Quantification of NDUFAF5 protein and CI subunits was performed in a single TMT batch of 11 samples.

Data normalization and expression outlier analysis were conducted using PROTRIDER⁵⁰ in a dataset of n = 815 fibroblast samples.

Results

Variants in multiple complex I nuclear subunits are associated with optic neuropathy

We identified biallelic or hemizygous candidate pathogenic genotypes in 11 genes encoding subunits of respiratory CI in 31 patients from 23 families. These patients had no putative causative variants in the mitochondrial genome or in other nuclear genes associated with OA (Fig. 1 and Table 1). The identified variants were located in three core subunits (NDUFS7, NDUFV1 and NDUFV2), four accessory subunits (NDUFA1, NDUFA10, NDUFA12 and NDUFB11) and four assembly factors (NDUFAF2, NDUFAF3, NDUFAF4 and NDUFAF8). Overall, based on ACMG criteria, we classified 10 variants as pathogenic, six as likely pathogenic and four as variants of unknown significance (Supplementary Table 3). For these latter missense variants, additional evidence from recurrence in independent families and in silico modelling was integrated into the ACMG classification.

Figure 1.

Family trees. For each available individual the mutated (m) and wild-type (+) alleles are reported. The wild-type Y chromosome (y) is reported when the mutated allele is on the X chromosome in males.

Table 1.

Genetic variants

Family: ID	OA	Gene	Subunit type, location (module)	Inher.	Gene MIM#	Genotype	Protein
A: II-1	Plus	NDUFS7	Core, peripheral arm (Q)	AR	601825	NM_024407.5:c.[223C>T];[559_564delinsTAGAT]	NP_077718.3:p.[(Arg75Cys)];[(Ala187*)]
B: II-2	Isolated	NDUFS7	Core, peripheral arm (Q)	AR	601825	NM_024407.5:c.[298C>T];[298C>T]	NP_077718.3:p.[(Arg100Cys)];[(Arg100Cys)]
C: III-1	Isolated	NDUFS7	Core, peripheral arm (Q)	AR	601825	NM_024407.5:c.[298C>T];[298C>T]	NP_077718.3:p.[(Arg100Cys)];[(Arg100Cys)]
D: II-1, II-2	Isolated	NDUFS7	Core, peripheral arm (Q)	AR	601825	NM_024407.5:c.[313C>T];[313C>T]	NP_077718.3:p.[(Arg105Cys)];[(Arg105Cys)]
E: II-1	Plus	NDUFS7	Core, peripheral arm (Q)	AR	601825	NM_024407.5:c.[313C>T];[313C>T]	NP_077718.3:p.[(Arg105Cys)];[(Arg105Cys)]
F: III-2	Isolated	NDUFV1	Core, peripheral arm (Q)	AR	16101561015	NM_007103.4:c.[1156C>A];[1156C>A]	NP_009034.2:p.[(Arg386Ser)];[(Arg386Ser)]
G: II-1	Isolated	NDUFV2	Core, peripheral arm (Q)	AR	600532	NM_021074.5:c.[163T>G];[674G>A]	NP_066552.2:p.[(Phe55Val)];[(Cys225Tyr)]
H: II-1	LLS	NDUFA1	Accessory, membrane arm (PP)	XLR	300078	NM_004541.4:c.[28T>C];[0]	NP_004532.1:p.[(Ser10Pro)];[0]
I: II-1	Plus	NDUFA1	Accessory, membrane arm (PP)	XLR	300078	NM_004541.4:c.[28T>C];[0]	NP_004532.1:p.[(Ser10Pro)];[0]
J: II-1	Plus	NDUFA1	Accessory, membrane arm (PP)	XLR	300078	NM_004541.4:c.[55C>T];[0]	NP_004532.1:p.[(Pro19Ser)];[0]
K: II-1	Plus	NDUFA10	Accessory, membrane arm (PP)	AR	603835	NM_004544.4:c.[233_235del];[233_235del]	NP_004535.1:p.[(Ala78del)];[(Ala78del)]
L: II-1	Plus	NDUFA10	Accessory, membrane arm (PP)	AR	603835	NM_004544.4:c.[233_235del];[233_235del]	NP_004535.1:p.[(Ala78del)];[(Ala78del)]
M: II-2, II-3	Isolated	NDUFA10	Accessory, membrane arm (PP)	AR	603835	NM_004544.4:c.[1009C>T];[1009C>T]	NP_004535.1:[(p.Arg337Cys)];[(p.Arg337Cys)]
N: II-1, II-2	Isolated	NDUFA12	Accessory, peripheral arm (N/Q interface)	AR	614530	NM_018838.5:c.[69T>G];[69T>G]	NP_061326.1:p.[(Tyr23*)];[(Tyr23*)]
O: II-1	Plus	NDUFB11	Accessory, membrane arm (PD)	XLR	300403	NM_019056.7:c.[276_278del];[0]	NP_061929.2:p.[(Phe93del)];[0]
P: II-2, II-4	Plus	NDUFAF2	Assembly, peripheral arm (N/Q interface)	AR	609653	NM_174889.5:c.[95A>G];[148del]	NP_777549.1:p.[(Tyr32Cys)];[(Arg50Glufs*3)]
Q: II-1	Plus	NDUFAF2	Assembly, peripheral arm (N/Q interface)	AR	609653	NM_174889.5:c.[114C>G];[114C>G]	NP_777549.1:p.[(Tyr38*)];[(Tyr38*)]
R: II-1	LLS	NDUFAF3	Assembly, peripheral arm (Q)	AR	612911	NM_199069.2c.[5C>A];[489_490del]	NP_951032.1:p.[(Ala2Asp)];[(Gly164Serfs*29)]
S: II-1	Isolated	NDUFAF4	Assembly, peripheral arm (Q)	AR	611776	NM_014165.4:c.[224del];[413T>C]	NP_054884.1p.[(Pro75Leufs*7)];[(Ile138Thr)]
T: II-2, II-3	Isolated	NDUFAF4	Assembly, peripheral arm (Q)	AR	611776	NM_014165.4:c.[413T>C];[413T>C]	NP_054884.1:p.[(Ile138Thr)];[(Ile138Thr)]
U: II-1, II-2	Isolated	NDUFAF8	Assembly, peripheral arm (Q)	AR	618461	NM_001086521.2:c.[44T>G];[44T>G]	NP_001079990.1:p.[(Leu15Arg)];[(Leu15Arg)]
V: II-1, II-2, II-3	Plus, isolated	NDUFAF8	Assembly, peripheral arm (Q)	AR	618461	NM_001086521.2:c.[195+271C>T];[195+271C>T]	NP_001079990.1:p.?
W: II-2	Isolated	NDUFAF8	Assembly, peripheral arm (Q)	AR	618461	NM_001086521.2:c.[195+271C>T];[195+271C>T]	NP_001079990.1:p.?

AR = autosomal recessive; ID = Individual; Inher. = Inheritance; LLS = Leigh-like syndrome; OA = optic atrophy; XLR = X-linked recessive.

The combined prevalence of CI-related nuclear genes (listed in the Supplementary material) in ION was assessed in a subset of the cohort screened at the Italian and German diagnostic centres. A total of 836 probands underwent whole-exome sequencing, with a 45% yield of cases with genetic findings reported. This cohort included 38 cases harbouring variants in nuclear CI genes, increasing the total diagnostic yield by 5% and representing 10% of all solved cases.

Patients exhibited a broad spectrum of clinical manifestations, ranging from isolated OA to OA ‘plus’, which included features such as nystagmus, Wolff-Parkinson-White (WPW) syndrome, hearing impairment, intellectual disability, seizures, cerebellar ataxia, peripheral axonal neuropathy and LLS with basal ganglia and brainstem abnormalities (Fig. 2, Tables 2 and 3 and Supplementary Table 4). OA onset was either insidious during childhood or presented acutely/subacutely in young adulthood, resembling the clinical course of LHON (Tables 2 and 3 and Supplementary Table 4). Pharmacological treatments for OA included idebenone, CoQ10, riboflavin and ubidecarenone at variable dosages. Visual improvement was variable, with only 3 of 13 patients showing some degree of response to idebenone treatment (Supplementary Table 4).

Figure 2.

Illustrative findings in eye and brain imaging. (A) Optical coherence tomography (OCT) of both eyes from Family D, Patient II:1 and (B) OCT with retinal ganglion cell analysis of both eyes from Family R, Patient II:1. Both cases show bilateral diffuse optic nerve atrophy with partial sparing of the nasal quadrant. (C) Axial and (D) coronal T2-weighted imaging sequences from Family R, Patient II:1, showing hyperintensities, initial atrophy and vacuolar degeneration of putamen and hyperintensity of caudati. (E and F) Coronal long repetition time sequences from Family H, Patient II:1, showing bilateral pallidal and dentati hyperintensity. OD = right eye; OS = left eye; RNFL = retinal nerve fibre layer.

Table 2.

Clinical features

Family: ID	Optic atrophy		Other symptoms	MRI findings
	Type of onset	AOO, years
A: II-1	LHON-like	16	Upbeat nystagmus	ONA, new tiny focus of T2/FLAIR signal abnormality in the left cerebral peduncle
B: II-2	LHON-like	25	–	ONA
C: III-1	Insidious	9–10	Neurogenic bladder	N/A
D: II-1	Insidious	2–6	–	MRI normal
D: II-2	Insidious	2–3	–	MRI normal
E: II-1	LHON-like	10	Wolff-Parkinson-White syndrome	MRI normal
F: III-2	LHON-like	16	–	MRI normal
G: II-1	Insidious	2–12	Hyperintensity of MRI T2 signal of the spinal cord, vitamin B12 deficiency	Hyperintensity of the posterior part of the cervical spinal cord
H: II-1	Insidious	2	Pendular nystagmus, abnormal basal ganglia MRI signal intensity, delayed ability to walk	Bilateral pallidal and dentati hyperintensity in TR sequences
I: II-1	Insidious	3	Mild neurosensory hearing impairment	ONA
J: II-1	Insidious	5	Nystagmus, polyneuropathy, intellectual disability, borderline, mild neurosensory hearing impairment, sensory ataxia	ONA, very mild atrophy of cerebellar vermis
K: II-1	Insidious	0–1	Multiple sclerosis, pendular nystagmus, decreased mean corpuscular volume, CNS demyelination	Demyelinating brain lesions (images and report not available)-2021 (followed elsewhere for MS)
L: II-1	Insidious	5	Cerebellar atrophy, abnormality of extrapyramidal motor function, cerebral cortical atrophy, generalized-onset seizure, intellectual disability, borderline, mild neurosensory hearing impairment, cognitive impairment	Cerebellar and mild cerebral cortical atrophy
M: II-2	Insidious	2–12	–	MRI normal
M: II-3	Insidious	2–12	–	MRI normal
N: II-2	LHON-like	15	–	MRI normal
N: II-3	LHON-like	16	–	MRI normal
O: II-1	Insidious	0–1	Megaloblastic anaemia, abnormality of the gallbladder, folate deficiency	ONA
P: II-2	Insidious	6–12	Cerebral calcification, lactic acidosis, sensory axonal neuropathy, mild neurosensory hearing impairment, hyperintensity of cerebral WM on MRI, multiple meningiomas, vitamin B12 deficiency	Brainstem WM changes; multiple frontal and falx calcified meningiomas; ONA; abnormal lactic acid accumulation at MRS
P: II-4	Insidious	2–6	Lactic acidosis, sensory axonal neuropathy, cerebral aneurysm, abnormal basal ganglia MRI signal intensity	Focal left substantia nigra hyperintensity; left M1 aneurysm; ONA
Q: II-1	LHON-like	2–6	Nystagmus, peripheral axonal neuropathy	MRI normal
R: II-1	Insidious	0–3	Motor delay, abnormal basal ganglia MRI signal intensity	Bilateral striatal necrosis (putaminal bilateral hyperintensity)
S: II-1	Insidious + LHON-like	0–3	–	MRI normal
T: II-2	Insidious	3	–	Mild disproportionate volume loss involving superior vermis and cerebellar hemispheres
T: II-3	Insidious	5	–	Mild disproportionate volume loss involving superior vermis and cerebellar hemispheres
U: II-1	LHON-like	10	–	MRI normal
U: II-2	LHON-like	13	–	MRI normal
V: II-1	LHON-like	3	Strabismus, hyperopia, nystagmus, unsteady gait, mild developmental delay	Non-specific WM lesions
V: II-2	LHON-like	3	Strabismus, hyperopia	Non-specific WM lesions
V: II-3	LHON-like	4	Strabismus, hyperopia	Non-specific WM lesions
W: II-2	Insidious	0–1	–	MRI normal

AOO = age of onset; ID = Individual; LHON = Leber hereditary optic neuropathy; N/A = not available; ONA = optic nerve atrophy; WM = white matter.

Table 3.

Ophthalmic findings

Family: ID	Agea	Fundus oculi description	VA OD	VA OS	OCT RNFL avg OD	OCT RNFL avg OS	Visual field MD OD	Visual field MD OS
A: II-1	30	Bilateral and symmetrical OA	0.01	0.01	56	55	−27.74	−30.84
B: II-2	39	Temporal ON pallor	1	0.6	N/Q, diffuse thinning more pronounced in temporal quadrant		N/A	N/A
C: III-1	25	Bilateral OA, high myopia	0.8	0.4	47	54	−7.23	−9.12
D: II-1	7	Temporal ON pallor	0.8	0.4	49	44	N/A	N/A
D: II-2	5	Temporal ON pallor	0.6	0.9	42	43	N/A	N/A
E: II-1	30	Diffuse OA	0.05	0.05	N/A	N/A	N/Q, cecocentral scotoma
F: III-2	27	Bilateral OA	0.01	0.01	55	46	N/A	N/A
G: II-1	20	Diffuse ON pallor more evident temporally, small ONH	0.1	0.1	46	48	−14.65	−10.92
H: II-1	22	Pale optic disk temporal bilaterally	0.4	0.3/0.4	N/A	N/A	N/A	N/A
I: II-1	17	Temporal ON pallor	0.15	0.1	N/A	N/A	N/A	N/A
J: II-1	45	ON pallor (OS > OD)	0.02	0.02	N/Q, inside normal values		N/Q, central scotoma + enlarged blind spot
K: II-1	37	Diffuse ON pallor	0.1	0.16	N/Q	N/Q	−14.25	−16.6
L: II-1	35	N/A (uncooperative patient)	N/A	N/A	N/A	N/A	N/A	N/A
M: II-2	18	Temporal ON pallor	0.25	0.16	65	64	−2.82b	−6.56b
M: II-3	16	Temporal ON pallor	0.4	0.5	58	56	−3.69b	−3.66b
N: II-2	24	Bilateral OA, marked loss of papillo-macular bundle	0.01	0.01	N/A	N/A	N/Q, generalized depression
N: II-3	18	OD severe OA, OS some preservation of nasal side	0.01	0.02	N/A	N/A	N/A	N/A
O: II-1	25	Temporal pallor	0.32	0.4	67	68	−10.2	−8.25
P: II-2	52	Diffuse ON pallor more evident temporally	0.25	0.2	41	39	−12.63	−12.73
P: II-4	37	Diffuse ON pallor more evident temporally	0.13	0.1	43	37	−16.43	−19.41
Q: II-1	15	Bilateral OA	0.16	0.01	40	43	−20.54	−11.98
R: II-1	7	ONH pallor	0.5	0.3	38	41	N/A	N/A
S: II-1	18	Temporal pallor followed by diffuse pallor after recurrence	0.2	0.25	42	41	−30.08	−32.00
T: II-2	17	Bilateral OA, Left Coats disease	0.5	0.5	56	55	N/A	N/A
T: II-3	13	Bilateral OA	1	1	68	58	−6.64	−13.05
U: II-1	24	N/A	0.015	0.03	40	55	N/A	N/A
U: II-2	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
V: II-1	15	Bilateral OA, tortuous vessels	0.03	0.03	N/Q	N/Q	−24.76	−20.78
V: II-2	11	Bilateral OA, tortuous vessels	0.08	0.063	N/Q	N/Q	−6.21	−6.77
V: II-3	6	Bilateral OA, tortuous vessels	0.063	0.063	N/A	N/A	N/A	N/A
W: II-2	54	Bilateral OA	0.08	0.05	40	43	N/A	N/A

avg = average; ID = Individual; N/A = not available; N/Q = not quantifiable; OA = optic atrophy; OCT = optical coherence tomography; OD = right eye; ON = optic nerve; ONH = optic nerve head; OS = left eye; RNFL = retinal nerve fibre layer; MD = mean deviation; VA = visual acuity.

^aAge at last visit, years.

^bVisual field test with M-700, Medmont International.

We evaluated the structural properties and pathogenicity prediction of the missense variants identified in our cohort, as well as those previously reported in the literature and associated with other phenotypes (Supplementary Table 2). Specifically, we combined structure-based prediction tools such as AlphaMissense and PAI-3D with an estimation of impact on protein folding (ΔΔG) to add a line of pathogenicity evidence to a commonly used meta-prediction tool (REVEL). Full data regarding pathogenicity prediction and ΔΔG values are reported in Supplementary Tables 5–7 and discussed together with mapping and in silico mutagenesis for each affected polypeptide.

Defects in core complex I subunits lead to isolated optic atrophy

Eight patients from seven families (A–G) carried previously unreported variants in NDUFS7, NDUFV1 or NDUFV2, encoding core subunits of CI (Fig. 3A). These patients typically exhibited isolated OA, with either an insidious onset during childhood or an acute/subacute onset similar to LHON in young adulthood (16–25 years range; Table 2). At the last clinical evaluation, no significant new symptoms were reported, except for nystagmus in the proband of Family A and WPW syndrome in the proband of Family E. Additionally, a subclinical hyperintensity of the posterior part of the cervical spinal cord was noted in the proband of Family G, carrying NDUFV2 variants, likely attributable to vitamin B₁₂ deficiency Tables 23

Figure 3.

Modelling of missense variants in core subunits. (A) Localization of the core subunits in the complex I structure considered in this work, with co-factors shown as ‘balls-and-sticks’, coloured according to the atom type. (B–D) Position of the variants described for NDUFS7 (B), NDUFV1 (C) and NDUFV2 (D) subunits. The residues subject to a variation are shown in ‘spheres’, coloured as they are described in the literature (dark cyan), in this work (purple) or in both cases (dark green).

NDUFS7 (NM_024407.5)

The proband from Family A was found to carry both a missense variant c.223C>T p.(R75C) and a small indel c.559_564delinsTAGAT p.(A187*). The homozygous variant c.298C>T p.(R100C) was detected in affected individuals from Families B and C, while the homozygous variant c.313C>T p.(R105C) was found in affected individuals from Families D and E, making NDUFS7 the most commonly affected gene in our case series. The missense variants p.(R100C) and p.(R105C) were predicted as pathogenic by AlphaMissense and PAI-3D, while p.(R75C) was classified as ambiguous by AlphaMissense. The ΔΔG values for these variants were inconsistently predicted by DynaMut2 and DUET, thus being classified as uncertain for their potential impact on subunit folding. The five variants previously reported in the literature were predicted to be likely pathogenic by AlphaMissense, while PAI-3D classified them as pathogenic, except for p.(P144L) and p.(R145H), which nevertheless showed high pathogenicity scores (Supplementary Table 5). Among these variants, p.(L126F), p.(R145H) and p.(R168S), associated with LS, were also predicted to have an impact on protein folding based on their ΔΔG values. The most striking difference between previously reported variants and those described here was their location within the 3D structure of CI. Indeed, the former were grouped close to the Fe-S cluster N2, associated with the ubiquinone (coenzyme Q10, CoQ10) binding site, while the latter were situated in the N-terminal of the subunit (Fig. 3B). Specifically, p.(R100) and p.(R105) were positioned peripherally in an external region of the subunit and involved in the formation of PPI interfaces with NDUFS2, NDUFS8, NDUFA7 and NDUFA12. These were distinct from the remaining variants, which mediated the interactions between NDUFS7 and NDUFS3, NDUFA9 and ND3 (Supplementary Table 8). Notably, p.R100 forms a hydrogen bond with p.E208 of NDUFS2 using its side chain (distance 3.29 Å), meaning this interaction could be disrupted by the variants identified in Families B and C, while the side chain of p.R105 is hydrogen bonded with p.I82 of NDUFS8 (distance 3.89 Å), which may be compromised by the variant found in Families D and E. The only subunit interacting with NDUFS7, using residues from both cohorts, was ND1, which forms a PPI interface involving both p.R75 (Family A) and p.R105 (Families D and E), as well as p.P144, which was also shown to be mutated in a published case of OA.²¹

NDUFV1 (NM_007103.4)

The proband from Family F carried the homozygous variant c.1156C>A p.(R386S) at a known mutational hotspot of NDUFV1, where pathogenic variants p.(R386C) and p.(R386H) have previously been described in several LS patients following the first published cases.^51,52 Among the subunits studied, NDUFV1 harbours the highest number of pathogenic variants, with 28 linked to LS in the literature, which are all predicted as pathogenic by AlphaMissense, while PAI-3D classifies 17/28 variants as pathogenic and 11/28 as benign (Supplementary Table 5). Some of these variants are predicted to impact protein folding and are generally spread across the protein, although they tend to cluster around specific regions, such as the flavin mononucleotide-binding site and the Fe-S clusters N1a and N3.

Notably, p.R386 is located on the subunit surface within the four-helix bundle of the NDUFV1 subunit and plays a role in forming a PPI interface with NDUFS1 (Fig. 3C and Supplementary Table 8). Moreover, it is part of an extended H-bond network involving its guanidino tail, NDUFV1 residues p.G380, p.Q381 and p.T383, and NDUFS1 residue p.Q178. Residue p.S386 also forms a π-stacking interaction with the side chains (Supplementary Fig. 1). These interactions could be disrupted when p.R386 is replaced by a shorter amino acid, such as serine, cysteine or histidine. However, it remains unclear how the specific serine substitution would cause a milder phenotype compared to its substitution with cysteine or histidine. No clear biochemical differences can be inferred from the available data, aside from the proximity of residue 386 to the Fe-S cluster N3, which may undergo positional shifts in the presence of cysteine, potentially affecting its function.

NDUFV2 (NM_021074.5)

The proband from Family G was found to carry compound heterozygous c.163T>G p.(F55V) and c.674G>A p.(C225Y) variants. Both variants, along with the five variants reported in the literature, are predicted to be pathogenic (Fig. 3D). Two are likely to affect protein folding, namely p.(F55V), found in the present cohort, and p.(E194K), reported in literature (Supplementary Table 5). The two OA-associated variants are located at distant sites, with no apparent mutational hotspots when also considering the LS-associated variants (Fig. 3D). NDUFV2 forms several PPIs with NDUFV1, NDUFV3, NDUFS1, NDUFS4 and NDUFS6 (Supplementary Table 8). Among these interactions, only the interface with NDUFV1 contains mutated residues, specifically p.C225 (mutated in Family G) and p.T232 (reported in the literature).

Defects in accessory complex I subunits lead to a spectrum from isolated to syndromic optic atrophy

Ten patients from eight families (Families H–O) carried hemizygous variants in NDUFA1 or NDUFB11 or biallelic variants in NDUFA10 or NDUFA12, encoding accessory subunits of CI. In all patients, OA began during childhood, except for Family N, which exhibited a LHON-like onset. Most patients, except Families M and N, showed signs of syndromic disease, including hearing impairment, intellectual disability, and movement disorders associated with basal ganglia abnormalities, seizures and cerebellar ataxia. Bilateral pallidal and dentate hyperintensity was observed in only one patient (Family H) (Fig. 2E and F). Cerebellar atrophy was reported in two cases (Family J and L). One patient (Family K) also had multiple sclerosis (Tables 2 and 3).

NDUFA1 (NM_004541.4)

The probands from Families H and I both carried the novel missense variant c.28T>C p.(S10P), while the proband of Family J carried the missense variant c.55C>T p.(P19S), which has been reported in two published cases of LS.^53,54 This subunit contains four variants previously reported in the literature, along with two novel variants identified in our cohort, for which pathogenicity predictions by AlphaMissense and PAI-3D, as well as their impact on ΔΔG, were uncertain (Supplementary Table 6). Focusing on the variants identified in this study, p.(P19S) is predicted to be pathogenic and to affect protein folding, while p.(S10P) is generally reported as benign. However, this variant was found in two independent families, with no alleles present in the gnomAD v4.1 dataset, supporting its pathogenicity. When mapping the variants in the 3D structure model, OA-associated variants are located in the N-terminal helix of NDUFA1, where the p.(G8R) variant (reported in the literature) also resides (Fig. 4B). The other LS-associated variants are located in the loop connecting the two main α-helices. Notably, residues from the N-terminal region are involved in forming a PPI with ND1, similar to NDUFS7. We hypothesize that the p.(P19S) variant could result in the formation of a longer transmembrane helix that, in turn, may impact protein folding, as predicted by high variation in the ΔΔG values. However, the effect of the p.(S10P) variant remains difficult to define, even if proline insertion into α-helices is often disruptive.

Figure 4.

Modelling of missense variants in accessory subunits. (A) Localization in the complex I structure of the accessory subunits considered in this work, with co-factors shown in ‘balls-and-sticks’, coloured according to atom type. (B–D) Position of the variants described for NDUFA1 (B), NDUFA10 (C) and NDUFB11 (D) subunits. The residues subject to a variation are shown in ‘spheres’, coloured as they are described in the literature (dark cyan), in this work (purple) or in both cases (dark green). IMM = inner mitochondrial membrane; IMS = intermembrane space.

NDUFA10 (NM_004544.4)

The proband from Families K and L carried the homozygous in-frame deletion c.233_235del p.(A78del). This variant has previously been reported in  trans with c.296G>A p.(G99E) in a case of LS.⁵⁵ Two siblings from Family M harboured the homozygous missense variant c.1009C>T p.(R337C); this genotype has previously been reported in a patient with ataxia, hypotonia and developmental regression.⁵⁶ In total, five missense variants have been reported in this subunit in cases of LS, including the p.(R337C) variant identified in the present cohort, highlighting that other genetic factors likely modulate the clinical phenotype in these patients (Fig. 4C and Supplementary Table 6). While the in-frame deletion of residue p.A78 can be predicted as pathogenic for the lack of the single amino acid, missense variants were variably predicted by AlphaMissense and PAI-3D. However, p.(R337C) shows an impact on protein folding with a significant divergent value of ΔΔG (Supplementary Table 6). Additionally, in the 3D structure, p.G99 and p.R337 are in close proximity and involved in PPI interfaces with NDUFC1, NDUFC2 and ND2 (Fig. 4C and Supplementary Table 8). Specifically, the p.(R337C) variant could induce a reduction in steric bulk, affecting protein folding, while p.(G99E) may form a salt bridge with p.R337, thus impacting the flexibility of the loop containing p.G99.

NDUFA12 (NM_018838.5)

Two siblings from Family N were found to be homozygous for the novel stop-gain variant c.69T>G p.(Y23*). To date, only loss-of-function variants have been identified in this subunit. Nevertheless, a complex phenotypic spectrum characterizes NDUFA12 cases and even the arLHON phenotype as observed in our cases has been reported.^19,57

NDUFB11 (NM_019056.7)

The proband from Family O carried the hemizygous in-frame deletion c.276_278del p.(F93del), which was first described as a recurrent variant in patients with syndromic sideroblastic anemia due to CI deficiency (with myopathy, lactic acidosis, optic atrophy).⁵⁸ Three missense variants in this subunit have been reported in the literature (Supplementary Table 6), and they are distributed along the entire length of the protein without clear hotspots (Fig. 4D). From the protein interaction perspective, NDUFB11 interacts with 11 subunits, including ND4 through the residues p.F93 and p.P110 (Supplementary Table 7), suggesting these residues play a critical role in the assembly of CI.

Defects in assembly complex I factors lead to a spectrum from isolated to syndromic optic atrophy

Thirteen patients from eight families (Families P–W) carried compound heterozygous variants in NDUFAF2, NDUFAF3, NDUFAF4 or NDUFAF8, encoding assembly factors of CI. Six patients out of 13, carrying NDUFAF2, NDUFAF4 or NDUFAF8 variants, showed a LHON-like phenotype (Tables 2 and 3). All three patients from Families P and Q, carrying variants in NDUFAF2, showed OA and peripheral axonal neuropathy, while the single patient harbouring NDUFAF3 variants showed LLS with basal ganglia hyperintensity (Fig. 2C and D) and motor delay. The phenotype associated with variants in NDUFAF4 was characterized by isolated OA in all three patients, as observed also for five of six patients with NDUFAF8 variants; the remaining proband also exhibited cerebellar ataxia and nystagmus. One of the patients carrying NDUFAF4 variants who presented with insidious OA also suffered at the age of 17 years a LHON-like visual loss.⁵⁹ Additionally, basal ganglia/brainstem abnormalities, white matter changes, and cerebellar atrophy were reported in most cases. The in silico analyses were performed for these polypeptides based on the structures generated by AF3 (Fig. 5 and Supplementary Table 7).

Figure 5.

Modelling of missense variants in assembly factors. The position of the variants described for NDUFAF2 (A), NDUFAF3 (B), NDUFAF4 (C) and NDUFAF8 (D) subunits are shown on AlphaFold 3 model structures. The residues subject to a variation are shown in ‘spheres’, coloured as they are described in the literature (dark cyan) or in this work (purple). Ribbons are coloured according to the AlphaFold 3 predicted local distance difference test confidence score (plDDT).

NDUFAF2 (NM_174889.5)

Two affected siblings from Family P carried the novel missense variant c.95A>G p.(Y32C) and the frameshift variant c.148del p.(R50Efs*3), which was recently reported in trans with c.139C>T, p.(R47*) in a patient with an LHON-like phenotype.²⁰ The proband from Family Q carried the homozygous nonsense variant c.114C>G, p.(Y38*), which was known from a patient with LS.⁶⁰ The amino acid change p.Y32C is the first missense pathogenic variant described and lies in a well modelled antiparallel β-sheet (Fig. 5A). This variant is predicted as pathogenic by both AM and PAI-3D, highlighting a possible detrimental effect on protein function.

NDUFAF3 (NM_199069.2)

The proband from Family R carried the novel variant c.5C>A, p.(A2D) and the c.489_490del p.(G164Sfs*29) already reported in a LS case.⁶¹ The p.(A2D) variant is located at the N-terminal of NDUFAF3, in contrast to the five other variants reported in the literature, associated with LS and predicted as pathogenic, affecting amino acids in the well-modelled globular C-terminal portion of the protein (Fig. 5B). The N-terminal (aa 1–31) is predicted by MTSviewer as a canonical MTS but is not well modelled by AF3. However, when the secondary structure is predicted using JPred4, p.(A2D) amino acid change is found to disrupt the positively charged surface of the N-terminal amphipathic α-helix, thus possibly impacting the protein import.

NDUFAF4 (NM_014165.4)

The proband from Family S carried the novel biallelic variants c.224del p.(P75Lfs*7) and c.413T>C p.(I138T), with the latter variant also found in the homozygous state in the two affected siblings from Family T. The p.I138T variant is predicted as pathogenic and is likely to destabilize the protein folding (Supplementary Table 7). This amino acid lies in a well-modelled globular domain of NDUFAF4 (Fig. 5C), clearly distant from those LS-associated reported in literature affecting the N-terminal, which are ambiguous in terms of pathogenicity prediction, but may affect the import of this assembly factor.

NDUFAF8 (NM_001086521.2)

The siblings from Family U carried the homozygous novel missense variant c.44T>G p.(L15R), while the probands from Families V and W carried the homozygous deep intronic variant c.195+271C>T p.? This latter variant was already described in two independent families affected by LS, in compound heterozygosity with pathogenic alleles, and it was confirmed to induce transcript decay.⁶² The AF3 model is moderately accurate for the N-terminal α-helix where the p.L15 residue is located, however the change is predicted to affect the MTS introducing a positive charge on the hydrophobic side of the amphipathic α-helix (Fig. 5D), possibly affecting NDUFAF8 import into the mitochondria.

To validate the impact of the p.L15R variant at the protein level, we performed quantitative proteomics on fibroblasts derived from individual II-4 in a batch of 11 samples. NDUFAF8 protein was not detected in the whole TMT 11-plex batch, not allowing us to directly evaluate the variant effect. However, we observed a strong reduction of NDUFAF5 protein (fold change: 0.18; Supplementary Fig. 2A), another CI assembly factor known to require NDUFAF8 for its stability.⁶² Moreover, a moderate reduction of CI protein level (mean fold change: 0.73) was observed, indicating an impaired assembly pathway (Supplementary Fig. 2B).

Variants of uncertain significance of mtDNA in recessive optic atrophy

It is now established that a clinical phenotype of LS can result from digenic variants in CI subunits encoded in both nuclear and mitochondrial genomes.⁶³ To explore this phenomenon in our cohort, we also investigated the affected individuals’ mtDNA sequence background (Supplementary Table 9). No imbalance in major haplogroup frequency was observed over the 23 families in the cohort, where we observed haplogroups C (n = 1), H (n = 8), J (n = 3), M (n = 2), N (n = 1), R0 (n = 1), U (n = 4), V (n = 1) and X (n = 2). A total of six families harboured missense variants in mitochondrial CI genes, namely MT-ND1, MT-ND2, MT-ND3, MT-ND5 and MT-ND6. Among those, the proband of Family I carried three different variants in CI genes, including m.3394T>C/MT-ND1, already associated with several phenotypes, especially a population-specific LHON.⁶⁴ Two more families carried missense variants in the ATPase complex, namely m.9025G>A/MT-ATP6 in Family A, previously reported in a LS case,⁶⁵ and m.8555T>C/MT-ATP6 in Family N.

Discussion

This study describes a series of 31 affected individuals from 23 families in whom OA was the primary phenotype, with some cases showing additional symptoms and pathogenic variants in nuclear-encoded CI subunits. Overall, we associate for the first time the arLHON phenotype with variants in NDUFAF4 and NDUFAF8, and we consolidate NDUFS7, NDUFV1 and NDUFAF2 variants as causative for arLHON, while previously reported in single families.^20,21 Furthermore, isolated or plus OA was found for the first time in NDUFV2, NDUFA1, NDUFA10 and NDUFAF3 families, for which only a single report of arLHON existed for a patient carrying NDUFA10 variants.⁶⁶

In all patients, previous genetic screening excluded the involvement of mtDNA or other optic neuropathies-related genes. Clinically, a subset of these patients experienced a subacute loss of vision resembling LHON, while others had an insidious onset of OA in childhood. It remains poorly understood what drives the difference in disease onset: the subacute deterioration of visual acuity in adolescence or young adulthood, which hallmarks LHON, or the more insidious onset in childhood, as typically seen in OPA1-dominant OA. However, as previously well described, even in LHON pedigrees with common mtDNA variants, both kinds of disease onset may be seen.⁶⁷ While OA was an isolated feature in about half of the patients, others exhibited broader involvement of the central and peripheral nervous systems, with manifestations ranging from peripheral neuropathy and sensory neural ataxia to nystagmus, neurosensory hearing impairment, basal ganglia lesions and LLS features. Extra-neurological features included megaloblastic anaemia and WPW syndrome, whereas lactic acidosis was a frequent laboratory finding. These findings greatly expand the genetic landscape of recessive OA, including arLHON, suggesting that LS/LLS and OA/arLHON are the two extremes of a phenotypic spectrum with variable severity seen more frequently than previously thought.

To understand the possible genotype-phenotype correlation, we categorized the CI subunits carrying the pathogenic variants according to their function in the complex, with three main groups, including core subunits, accessory subunits and assembly factors. The variants in the CI core subunits (NDUFS7, NDUFV1 and NDUFV2) were associated with isolated OA in six of eight patients, with only one case also having upbeat nystagmus and another one WPW syndrome. Half of these patients can be categorized as arLHON/arLHON plus based on disease onset and progression.

It is important to note that among the 14 core subunits of CI, three are the most affected by primary LHON variants, namely ND1, ND4 and ND6, suggesting that functions carried out by the catalytically competent enzyme core may be phenotypically linked to the subacute vision loss characterizing LHON. Congruently with this observation, the first report describing arLHON found biallelic variants in NDUFS2, which is a core subunit.¹⁷ Indeed, core subunits are responsible for redox activity and proton translocation,³⁰ as well as for the recently proposed function of the mitochondrial Na⁺/H⁺ exchanger,⁶⁸ which seems to be involved in the pathogenetic mechanism of the classic LHON mutation m.11778G>A/MT-ND4.

Defects in CI accessory subunits may also have a pronounced impact on the complex structure and function, similar to that observed for the core subunits.⁶⁹ However, the pathogenic variants affecting accessory subunits (NDUFA1, NDUFA10, NDUFA12, NDUFB11) were more frequently associated with OA plus (5 of 10 patients) and LLS phenotype (one patient). Only two families carrying NDUFA10 and NDUFA12 variants presented with isolated OA. This may be related to the role of accessory subunits, most of which are known to be necessary for human CI assembly and stability.⁶⁹ In this frame, variants in accessory subunits may induce clinical phenotypes which are more similar to those caused by variants in assembly factors. Indeed, in the latter category (NDUFAF2, NDUFAF3, NDUFAF4, NDUFAF8), we observed both isolated and plus OA phenotypes for the 13 patients, including one LLS case. For example, both families harbouring variants in NDUFAF4 showed isolated OA, whereas all three NDUFAF2 patients had multisystemic involvement. Given their role in the CI assembly, predicting their functional impact in silico is challenging due to varying levels of partially assembled sub-complexes and residual CI activity, which should be demonstrated experimentally.

The core subunits described for our cohort are all located in the matrix portion of the complex, specifically in module N (NDUFV1 and NDUFV2) and module Q (NDUFS7), which also interacts with the accessory subunits NDUFA1 and NDUFA12, which are connected to the membrane arm.⁷⁰ The assembly factors in which we identified pathogenic variants are variously involved in CI assembly. In fact, according to complexome profiling studies, NDUFAF3 and NDUFAF4 are necessary for the early connection of the Q module to ND1; NDUFAF2 is involved in the late maturation of the Q module; while NDUFAF8 contributes to early stabilization of NDUFS7 together with NDUFAF5.^62,71 This latter polypeptide, lacking from our cohort, has recently been found mutated in two cases showing OA/arLHON.^22,72 Since NDUFS2 and DNAJC30 are also part of, or involved in, the stability of the peripheral arm of CI, this portion of the complex appears to be a hotspot for recessive OA variants.

Our structural analysis also uncovered a possible segregation of variants linked to LS or OA into different parts of NDUFS7, the subunit most frequently affected in our OA cohort, which is located at the junction between the peripheral and membrane arms of CI and is involved in the formation of the CoQ binding site.⁷³ Notably, the variants identified here clustered around the N-terminal region and formed PPI with subunits such as NDUFS2, NDUFA12 and ND1, which have been commonly linked to OA and LHON/arLHON phenotypes in the literature.

Using up-to-date structural prediction tools, such as AlphaMissense and PrimateAI-3D, together with ΔΔG analyses, we applied more precise criteria for pathogenicity prediction (PP3) of missense variants according to the ACMG. Therefore, for 6 of 13 missense variants, we found evidence supporting a damaging role of the amino acid change, where REVEL meta-prediction was unable to assess the effect. Notably, we identified two variants affecting the MTS properties of NDUFAF3 and NDUFAF8, proving experimentally for the latter that the pathogenic variant impacts NDUFAF8 abundance in the mitochondria and affects CI assembly. The only variant lacking support for a pathogenic role based on structural evidence is p.(S10P) in NDUFA1, which requires further functional investigations. In-depth modelling could therefore help in the improvement of ACMG criteria-based assessment beyond the commonly available prediction and meta-prediction tools’ results.

An important observation to highlight is the clinical variability that can occur with the same pathogenic variant in different individuals. For example, we observed isolated OA in two siblings homozygous for the NDUFA10 p.(R337C) variant, whereas a previously published case with the same amino acid change was reported to have multisystemic involvement.⁵⁶ Even within the same family, we observed discordant phenotypes, as in the case of Family V carrying the recurrent NDUFAF8 c.195+271C>T intronic variant, here associated with either the isolated or plus arLHON phenotype. The phenotypic variability may be influenced by multiple factors, including genetic modifiers in both mitochondrial and nuclear genomes, as well as exposure to environmental factors, as documented for LHON.⁷⁴

We also note the presence of mtDNA variants with possible functional impact but without clear-cut pathogenicity in some patients in our cohort. A plausible hypothesis is that mtDNA variants could act synergistically with nDNA variants to exacerbate disease severity. However, an epistatic interaction might also counterbalance the effect of amino acid changes in nuclear subunits. Predicting the interaction between multiple specific variants remains challenging and would likely require advanced molecular dynamics modelling, at least for those affecting the same protein complexes, or in vitro/in vivo functional validation. For instance, the m.3394T>C/MT-ND1 variant, identified in Family I, was reported to interact with m.11778G>A/MT-ND4, increasing the LHON penetrance of LHON,⁷⁵ and with haplogroup and the environment,⁶⁴ becoming benign or pathogenic, depending on the context.

Moreover, in the Family A proband, we detected both biallelic NDUFS7 variants and the m.9025G>A variant in MT-ATP6, which was reported as pathogenic and supported by yeast validation.⁷⁶ However, it is relatively frequent in healthy controls (i.e. 45/56418 homoplasmic cases in gnomAD v4.1), and its role in disease modulation is unclear. Strikingly, a few reports attributed a pathogenic role to MT-ATP6 variants in LHON patients.^65,77 However, noticeably, these probands were singleton cases without any evidence of maternal inheritance, raising doubt that, as evidenced by the three cases found in this study, the real molecular defect resides in nuclear genes.

In conclusion, our study demonstrates that OA, whether LHON-like or with insidious early onset, is frequently consequent upon biallelic or hemizygous variants in nuclear genes involved in CI structure and assembly. Moreover, OA can be part of the LS/LLS spectrum more commonly known for the nuclear CI genes but can also occur as an isolated feature. Our study prompts a change in the diagnostic algorithm for ION patients, which should now always include the screening of all known CI-related genes, as, although individually rare, their combined prevalence can be comparable to other well-established ION genes, as reported by other studies.^16,17,66 Finally, these patients may benefit from treatment with idebenone, following appropriate dosing and standard protocols similar to those used for LHON. Idebenone, in its reduced form, can bypass CI and transfer electrons from NADH directly to respiratory complex III, thus restoring the downstream electron transport chain and oxidative phosphorylation.⁷⁸ In this respect, any CI defect may be partially corrected through the drug’s mechanism of action.

Data availability

The authors confirm that the data supporting the findings of this study are available within the article and its Supplementary material. Raw data were generated in each participating centre. Derived data supporting the findings of this study are available from the corresponding author on request.

Funding

P.Y.-W.-M. is supported by an Advanced Fellowship Award (NIHR301696) from the UK National Institute of Health and Care Research (NIHR). P.Y.-W.-M. also receives funding from the Rosetrees Trust (PGL23/100048), Fight for Sight UK, the Isaac Newton Trust (UK), Moorfields Eye Charity (GR001376), the Addenbrooke’s Charitable Trust, the National Eye Research Centre (UK), the International Foundation for Optic Nerve Disease (IFOND), the NIHR as part of the Rare Diseases Translational Research Collaboration, the NIHR Cambridge Biomedical Research Centre (NIHR203312), and the NIHR Biomedical Research Centre based at Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, University College London (NIHR203322). This research was supported by LifeArc under grant no. 10748. LifeArc is a charity registered in England and Wales under no. 1015243 and in Scotland under no. SC037861. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. L.C. is supported by the Italian Ministry of Health grant GR-2016-02361449. V.C. is supported by Italian Ministry of Health funding ‘Ricerca Corrente’. V.C. is also supported by NextGenerationEU (NGEU) and funded by the Italian Ministry for University and Research (MUR) National Recovery and Resilience Plan (NRRP), project MNESYS (PE0000006)—A multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022). G.T. and L.I. are supported by the European Union—NextGenerationEU, Mission 4, Component 2, CUP B93D21010860004. P.L. is supported by the Czech Ministry of Health (Ministerstvo Zdravotnictví Ceské Republiky) AZV NU22-07-00614 and by Charles University and the General University Hospital Prague UNCE/24/MED/022 and SVV 2600631. H.P. is supported by the Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie (BMBF) and Horizon 2020 through the EJP RD project 01GM1920A, 01GM1920B, 01GM2404A and 01GM2301 (GENOMIT), by the German Center for Child and Adolescent Health (DZKJ) under the funding code 01GL2406B. The publication of this article was supported by the "Ricerca Corrente" funding from the Italian Ministry of Health.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.