Since 2016, when the mutations in the ATAD3A gene were identified as causative in the first group of patients [11], a total of 69 patients from 52 families (Supplementary Table 1, Additional file 1, including those in this study, with various types of variants (Supplementary Table 2, Additional file 1) have been reported [10, 13–18]. In this study, we describe 4 new patients and show that the severity of the phenotype of patients with the recessive form of ATAD3A-linked disorder is relative to the impact of the variants.
Three main types of ATAD3A-related disorders can be observed. Two of them are autosomal dominant – i) Harel-Yoon syndrome (p.Arg528Trp variant in 5 cases) [11] or milder hereditary spastic paraplegia (p.Gly355Asp in 2 patients from 1 family) [15], where the patients survive into adulthood; and ii) autosomal dominant neonatal lethal cardiomyopathy (4 types of ATAD3 locus duplications in 22 cases) with only a mild neurological phenotype and strong cardiac impairment with neonatal lethality [13, 16].
The third, iii) autosomal recessive form, leads to pontocerebellar hypoplasia with a range of severity from neonatal lethal to patients living into adulthood. Large deletions encompassing the ATAD3 region or resulting in a generation of a fusion gene between ATAD3B and ATAD3A are the most common variants. Due to the high sequence homology, the fusion cDNA carries only several missense variants, but it is expressed from the ATAD3B promotor on a very low level [10, 12, 13]. Thus far, various combinations of different recessive point mutations, large deletions, or rearrangements have been found in 40 patients from 25 families (including this study).
The clinical picture and severity of the four patients described herein matched strikingly with that of the patient from Family 3 depicted in [12] (Supplementary Table 3, Additional file 1). All five patients carried a loss of function variant on one allele and the variant p.Leu77Val on the second allele. This consistency of the phenotype in patients sharing the same variants in an otherwise heterogeneous disorder led us to the hypothesis that the severity of the phenotype could depend on the type and severity of the pathogenic variants rather than on the influence of external factors or other genetic modifiers.
To follow this rationale, we sorted the recessive variants according to their impact predicted by their type and the severity of the disease in the patients. The details of the reviewed patients and variants are given in Supplementary Tables 1 and 2, respectively. We used the lifespan of the patients with the recessive disease as the marker of disease severity. The most severe form is neonatal lethal, which we labeled as the “severe form”. Cases in which the patient survived the neonatal period but died in early childhood were labeled as the “moderate form”, and, finally, patients with later onset and living into adulthood were labeled as the “mild form”.
As proof of this concept, we gathered all patients with biallelic null variants (large deletions, premature stop codons, and frameshift variants), which were automatically considered as “high impact” variants. If the idea that the severity of phenotype correlates with the severity of the variant impact was correct, we would expect these patients to have the most severe form of the disorder. And indeed, 12 of the 13 patients with biallelic null variants deceased in the first two weeks of life. One of them died at 7 months, however, this patient was incapable of breathing without support, therefore the increased lifespan appears to have been caused by the duration of the intervention, rather than being classified as a milder phenotype [19].
Subsequently, we were able to sort the missense and single amino acid deletion variants according to the severity of the disease, taking into consideration the variant on the other allele, as follows. The variants that were neonatal lethal in the homozygous state or in combination with a null variant were labeled “high impact” (p.Phe50Leu, p.Leu77Arg, p.Leu406Arg). Variants found either compound heterozygous with a null variant (p.Leu77Val, p.Gly236Val) or homozygous (c.384 + 3A > G, p.Arg327Pro) in patients with the moderate severity were labeled as “medium impact”. And finally, variants found homozygous or in combination with a null variant in patients living into adolescence and adulthood (p.Thr53Ile, p.Arg170Trp, p.Lys568del) or with a milder course of the disease with normal mental development in childhood (p.Thr84Met) were considered as “low impact”. Two variants, p.Trp537Arg and p.Arg503Profs*11, were found in a compound heterozygous state in two siblings with the mild form in this combination only. As the latter is predicted to escape nonsense-mediated decay, it is currently not possible to predict their individual impact. Supplementary Table 4, Additional file 1, sums up the sorting process. The lifespan of the patients with the given combinations of alleles is presented in Fig. 3.
The residual activity of the medium and the low-impact hypomorphic variants is most probably a continuum, and various combinations of alleles will result in a continuum of disease severity. Nevertheless, we believe that our stratification (summarized in Fig. 4) will help in better prognosis estimation in further patients to be diagnosed and may also be helpful in future structure-function studies.
The results of the functional studies for some of these variants performed by Yap et al. in Drosophila knock-outs [12] showed that only variants homologous to human p.Leu77Val (labeled “medium impact”) and p.Arg170Trp (“low impact”) rescued the developmental lethality caused by dAtad3a gene loss, indicating some residual functionality. Nevertheless, other variants tested – p.Phe50Leu (“high impact”), p.Arg327Pro (“medium impact”), p.Gly236Val (“medium impact”), and p.Lys568del (“low impact”) – did not rescue the developmental lethality of the dAtad3a knock-out. This suggests that Drosophila is a good conservative model for ATAD3A variant testing, but the capability of a variant to rescue the lethality in Drosophila doesn’t have to correlate with the outcomes of the patients.
Peralta et al. [17] suggested that mutations affecting the C-terminal region with the ATPase domain located in the matrix could be more severe than those affecting the N-terminal region containing two transmembrane domains and the coiled-coil domains involved in ATAD3A interactions with other protein partners outside mitochondria. However, since then several recessive variants have been discovered in both regions, and the genotype-phenotype correlation doesn’t seem to be dependent on the position (Fig. 4).
Nevertheless, the two dominant missense variants discovered so far, p.Arg528Trp and p.Gly355Asp causing HAYOS and spastic paraplegia, respectively, are both located at the ATP-binding site. Glycine 355 is directly in the Walker A motif and arginine 528 is located in the Sensor-II motif, opposite the glycine 355 in the conserved pocket [15]. Moreover, the third type of dominant mutations – various duplications found in the patients with neonatal lethal cardiomyopathy – results in the expression of a fusion gene ATAD3A-ATAD3C, where the N-terminal part corresponds to the 3A gene, whereas the C-terminal ATPase domain comes from the 3C pseudogene. The chimeric A/C gene contains multiple variants, including seven variants in the ATPase domain [13, 16]. Among them, p.Arg466Cys deserves special interest, as arginine 466 is conserved in all multimeric AAA-domain containing ATPases and functions as an arginine finger, a trans-acting residue that binds to the γ-phosphate of ATP in the neighboring monomer [3]. Therefore, it is probable, that the dominant variants act through the incorporation of an ATPase-deficient monomer into ATAD3A hexamers [16]. Different clinical consequences of these three dominant variants, mainly the strong cardiac involvement in the case of A/C chimera expression, however, remain unexplained and underline the involvement of ATAD3A in a number of processes that can differ in various tissues.
Furthermore, the impairment of the respiratory chain in the studied patients seems to be variable, as well. The deficiency of complex IV activity was seen in the fibroblasts of our Patient 1 (Fig. 2A), but in the muscle mitochondria, we detected decreased content not only of complex IV but of complexes I and V, as well (Fig. 2B). The decrease in the subunits of all complexes seen in the patient’s fibroblasts (Fig. 2D) and the decreased mitochondrial proteosynthesis (Fig. 2C) confirm a combined OXPHOS deficiency in this patient. In other published cases with biallelic variants, decreased activities of complexes I, III, or IV were reported, and a decrease in all complexes on the protein level has also been described, similarly to our results [10]. Combined OXPHOS deficiency agrees with what would be expected in the case of impairment of the ATAD3A function. However, multiple patients with biallelic large deletions and the severe phenotype showed only borderline or even normal activities in fibroblasts or in muscle and liver [10], and patients with mild phenotype also had either decreased CI and CIV complexes [10] or normal activities in fibroblasts or muscle cells [11]. This shows that in the case of recessive ATAD3A-related disorder variable complexes can be affected and that impairment of the respiratory chain is not necessarily detected in all tissues or in all patients. Interestingly, in the case of the dominant A/C chimera causing perinatal lethal cardiomyopathy, markedly decreased complex I activity occurred selectively in cardiac muscle [13], underlying the importance of the role of ATAD3A in cardiomyocytes. It is thus possible that OXPHOS activities are more affected in the cells of the nervous system in the case of the recessive variants, where the neurological symptoms are predominant, but this has not been studied yet.
ATAD3A-related disorders are most probably underdiagnosed due to the difficult-to-analyze chromosomal region containing three gene paralogs. Mitochondriopathies are genetically and phenotypically heterogeneous, therefore NGS techniques are the preferred method of choice for DNA diagnostics. However, copy-number variation analysis from WES data is still tricky and the presence of unidentified CNVs may often be a cause of negative analysis results in patients. When a targeted analysis was performed, ATAD3A locus mutations were the fourth most common nuclear-encoded cause of mitochondriopathy in a group of > 500 pediatric-onset cases in Australia and New Zealand [13]. Furthermore, variants such as p.Leu77Val can be filtered out during analysis based on higher population frequency or negative in silico predictions.
Moreover, variants in the ATAD3A gene could also possibly play a modifier role in other diseases. We published a case of a patient with Leigh syndrome caused by mutations in the SURF1 gene several years ago, Patient 1 in [20]. This patient showed an extremely severe clinical course, which was not typical for the SURF1-related Leigh disease. Interestingly, it has now been revealed, that this patient also carried the above-mentioned p.Leu77Val variant on a single allele, in addition to biallelic pathogenic variants in the SURF1 gene. The modifying effect of ATAD3A variants would, certainly, have to be supported by further studies and other cases.