The Phenotypic Continuum of ATP1A3-Related Disorders

Background and Objectives ATP1A3 is associated with a broad spectrum of predominantly neurologic disorders, which continues to expand beyond the initially defined phenotypes of alternating hemiplegia of childhood, rapid-onset dystonia parkinsonism, and cerebellar ataxia, areflexia, pes cavus, optic atrophy, sensorineural hearing loss syndrome. This phenotypic variability makes it challenging to assess the pathogenicity of an ATP1A3 variant found in an undiagnosed patient. We describe the phenotypic features of individuals carrying a pathogenic/likely pathogenic ATP1A3 variant and perform a literature review of all ATP1A3 variants published thus far in association with human neurologic disease. Our aim is to demonstrate the heterogeneous clinical spectrum of the gene and look for phenotypic overlap between patients that will streamline the diagnostic process. Methods Undiagnosed individuals with ATP1A3 variants were identified within the cohort of the Deciphering Developmental Disorders study with additional cases contributed by collaborators internationally. Detailed clinical data were collected with consent through a questionnaire completed by the referring clinicians. PubMed was searched for publications containing the term “ATP1A3” from 2004 to 2021. Results Twenty-four individuals with a previously undiagnosed neurologic phenotype were found to carry 21 ATP1A3 variants. Eight variants have been previously published. Patients experienced on average 2–3 different types of paroxysmal events. Permanent neurologic features were common including microcephaly (7; 29%), ataxia (13; 54%), dystonia (10; 42%), and hypotonia (7; 29%). All patients had cognitive impairment. Neuropsychiatric diagnoses were reported in 16 (66.6%) individuals. Phenotypes were extremely varied, and most individuals did not fit clinical criteria for previously published phenotypes. On review of the literature, 1,108 individuals have been reported carrying 168 different ATP1A3 variants. The most common variants are associated with well-defined phenotypes, while more rare variants often result in very rare symptom correlations, such as are seen in our study. Combined Annotation-Dependent Depletion (CADD) scores of pathogenic and likely pathogenic variants were significantly higher and variants clustered within 6 regions of constraint. Discussion Our study shows that looking for a combination of paroxysmal events, hyperkinesia, neuropsychiatric symptoms, and cognitive impairment and evaluating the CADD score and variant location can help identify an ATP1A3-related condition, rather than applying diagnostic criteria alone.

not fit clinical criteria for previously published phenotypes. On review of the literature, 1,108 individuals have been reported carrying 168 different ATP1A3 variants. The most common variants are associated with well-defined phenotypes, while more rare variants often result in very rare symptom correlations, such as are seen in our study. Combined Annotation-Dependent Depletion (CADD) scores of pathogenic and likely pathogenic variants were significantly higher and variants clustered within 6 regions of constraint.

Discussion
Our study shows that looking for a combination of paroxysmal events, hyperkinesia, neuropsychiatric symptoms, and cognitive impairment and evaluating the CADD score and variant location can help identify an ATP1A3-related condition, rather than applying diagnostic criteria alone.
Throughout the past 20 years, pathogenic variants in ATP1A3 have been discovered to cause an ever-expanding range of rare neurologic phenotypes, affecting both children and adults. ATP1A3 encodes the α3 subunit of a sodiumpotassium-ATPase (NKA) present in excitable (neuronal and cardiac) cells. The α3-subunit has a relatively low Na+ affinity coupled with a high affinity to adenosine triphosphate (ATP). 1,2 Consequently, the NKA carrying the α3-subunit is ideally configured to clear high intraneuronal sodium concentrations occurring after intense neuronal firing, by being able to use the low concentration of ATP that will occur near the neuronal membrane shortly after an energy-demanding task.
In this study, we aim to describe the phenotypic features of a cohort of previously undiagnosed individuals with developmental delay and a neurologic presentation, carrying a pathogenic/likely pathogenic ATP1A3 variant, and examine where they fit within the current spectrum of ATP1A3-related disorders. We also perform a literature review of all ATP1A3 variants published thus far in association with human neurologic disease. Our work clearly demonstrates the heterogeneous clinical spectrum associated with ATP1A3 variants, as well as phenotypic overlap between patients, that will streamline the diagnostic process.

Subject Cohort
An application was made to the Deciphering Developmental Disorders (DDD) study 17 for a Complementary Analysis Project, allowing access to anonymized details of individuals with ATP1A3 variants identified through this study. 18 If variant analysis and phenotypic details made pathogenicity likely, responsible clinicians were contacted to invite patients and their families to study recruitment. Some of the contacted clinicians had further individuals with ATP1A3 variants in their care, which they put forward as potential participants. Genomic diagnosis was reached through trio whole-exome sequencing (WES) for the DDD participants and through either WES or diagnostic gene panels for the other study participants. Phenotypic details were collected using a standardized clinical proforma covering all symptoms GLOSSARY ADHD = attention-deficit and hyperactivity disorder; AHC = alternating hemiplegia of childhood; ASD = autism spectrum disorder; CADD = Combined Annotation-Dependent Depletion; CAPOS = cerebellar ataxia, areflexia, pes cavus, optic atrophy, sensorineural deafness; COS = childhood onset schizophrenia; DDD = Deciphering Developmental Disorders; EIEE = early infancy epileptic encephalopathy; FIPWE = fever-induced paroxysmal weakness and encephalopathy; IP = intermediate phenotype; NKA = sodium-potassium-ATPase; PMG = polymicrogyria; RDP = rapid-onset dystonia parkinsonism; RECA = relapsing encephalopathy with cerebellar ataxia; SE = status epilepticus; SIFT = sorting tolerant from intolerant; VUS = variants of uncertain significance; WES = whole-exome sequencing. We used the UpSetR package in R to visualize intersections of signs and symptoms, trying to identify common phenotypes among individuals. As the clinical proforma included a long list of symptoms and signs due to the phenotypic variability of ATP1A3-related conditions, we decided to also group symptoms into broader categories where possible. We formed 4 categories: 1. Paroxysmal events, including hemiplegic events, dystonic episodes, seizures, abnormal eye movements, apnoea, and autonomic episodes. 2. Persistent movement disorders, including hyperkinetic phenotypes (dystonia and chorea) and ataxia. 3. Cognitive impairment (mild, moderate, and severe). 4. Neuropsychiatric symptoms, including behavioral difficulties, attention-deficit and hyperactivity disorder (ADHD), autism spectrum disorder (ASD), and COS.
To evaluate the pathogenicity of missense variants, in silico prediction tools were used, including sorting tolerant from intolerant (SIFT), 19 polymorphism phenotyping v2, PRO-VEAN, 20 MutationTaster, 21 and CADD scores. 22  We calculated Combined Annotation-Dependent Depletion (CADD) scores 22 for all missense ATP1A3 variants collected from the literature and for all missense variants reported within ClinVar as likely benign and benign and compared them. Unlike other genomic annotations that tend to exploit a single information type (i.e., conservation), CADD is a framework that objectively integrates many diverse annotations into a single, quantitative score. The integrated annotations include conservation metrics, functional genomic data, transcript information, and protein-level scores (Grantham, SIFT, and PolyPhen). CADD calculates a raw score and a "PHRED-scaled" score. PHRED-scaled scores are normalized to all potential ;9 billion SNVs, thus providing a comparable unit for analysis. Hence, a PHRED-scaled score of ≥10 indicates a raw score in the top 10% of all possible reference genome SNVs, a score of ≥20 or greater indicates a raw score in the top 1%, etc. 25 The developers of CADD do not suggest a rigid cutoff to suggest pathogenicity; however, looking at various HGMD molecular categories of 174,183 disease-associated deleterious mutations, one study found mean CADD scores for pathogenic missense variants to be above 20. 26 The tool is freely available on cadd.gs.washington.edu.

Constraint Analysis
Constraint analysis 27 was performed on all missense pathogenic variants in the study cohort and for published cases and compared with the reported benign missense variants in the population database GnomAD. The number of benign missense variants present within every 10 amino acid residues was plotted across the length of the gene, and from this, the missense constraint heat map included in Figure 4 was generated using the following parameters: dark green = >20 variants, light green = 11-20 variants, yellow = 8-10 variants, light red = 1-3 variants, and dark red = 0 variants. Locations of pathogenic missense variants identified in the study cohort were also plotted across the length of the gene, visually highlighting regions of the gene that are tolerant and intolerant to missense variation.

Data Availability
All data relevant to our patient cohort are published within the text and tables of this article. The detailed data extracted from the literature to conduct the literature review may be shared at the request of any qualified investigator for purposes of replicating procedures and results.

Standard Protocol Approvals, Registrations, and Patient Consents
The DDD study has UK Research Ethics Committee approval (10/H0305/83, granted by the Cambridge South REC, and GEN/284/12, granted by the Republic of Ireland REC). The "Natural history in ATP1A3-related disease: a deep phenotyping-genotyping project" has also been granted UK Research Ethics Committee approval (18/LO/ 1169), including for collecting anonymized genotypic and phenotypic data from international collaborators. Informed consent for publication of clinical and molecular data was obtained from all patients included in this manuscript and/ or their guardians.

Subject Cohort
Twenty-seven individuals with ATP1A3 variants in the DDD cohort (nearly 14,000 children recruited with their parents) were identified. In 7 individuals, the ATP1A3 variant was classified as class 2 (likely benign) due to either an incongruent phenotype or high prevalence in healthy populations. For the remaining 20 individuals, 2 clinicians did not report back, 2 clinicians declined participation in the study, and 3 individuals did not consent to study participation. As a result, 13 individuals from the DDD cohort were included in this study. Another 11 individuals with class 4 and 5 (likely pathogenic/ pathogenic) ATP1A3 variants were volunteered by collaborating clinicians and included because they fulfilled the study criteria. Three were family members of 2 separate patients with DDD. Three individuals have been previously published in the literature (patients 3, 5, and 14). Table 2 summarizes the ATP1A3 variants present in our cohort, including 13 novel variants.

Paroxysmal Events
All individuals except for 2 (siblings 19a and 19b) had at least 1 type of paroxysmal event in their clinical history. On average, individuals had 2-3 different types of paroxysmal events. The most common were dystonic events (n = 14), followed by abnormal eye movements (n = 12) and epileptic seizures (n = 12). Hemiplegic episodes were less common in this cohort (n = 9). Autonomic episodes, such as tachycardia, mydriasis, and vomiting, were reported in 9 and apnoeic events in 4 patients.

Epilepsy
Recurrent epileptic seizures were reported in 12 individuals. In 6 (50%), this was supported by epileptiform features on EEG. Seizure types were reported as focal seizures (n = 6), generalized tonic-clonic seizures (n = 5), and absence absence seizures, this is a seizure type seizures (n = 3). Four individuals had 2 seizure types, and 7 individuals had a history of status epilepticus (SE).

Development
Cognitive impairment was reported in all individuals, but severity was extremely varied. Motor delay was reported in 20 (83.3%) individuals, with walking age ranging from normal at 13 months to some individuals not having learnt to walk by 18 years. Language delay was also reported in 20 (83.3%) individuals. Communication skills were very varied, ranging from starting to communicate at 9 months to not having acquired language at age 20 years. Grade of cognitive impairment was reported in 17 individuals and classified as mild in 5 individuals (29.4%), moderate in 5 individuals (29.4%), and severe in 7 individuals (41.2%). In 5 individuals (21%), regression of skills was reported, occurring either steadily over time (n = 3) or in association with a clear trigger (fever, infection, SE) (n = 2).

Symptom Combinations
Looking at the combination of symptoms in our 24 patients, phenotypes were extremely varied. Across all 22 neurologic signs and symptoms we collated, no 2 individuals shared the same combination ( Figure 1A). Looking only at the 11 more common signs and symptoms, again there was little overlap, with only 2 pairs of individuals sharing the same features ( Figure 1B). However, looking at the broader symptom categories, paroxysmal events, movement disorders, cognitive impairment, and neuropsychiatric symptoms, almost half of the cohort (45.8%, n = 11) had at least 1 symptom from each of all 4 categories, while most individuals (91.6%, n = 22) had at least 1 symptom from 3 categories ( Figure 1C).     • Initial concern: hemiplegic episodes a

Genotype-Phenotype Correlation in
• Initial concern: rapid-onset left brachialpredominant dystonia • Dysphagia a • Action tremor • Mild developmental delay: walks independently and talks in 2-word sentences.
• Hemiplegic episodes Yang et al. 33 Holze et al. 35 Hully et al. 34 (siblings) • Age: 27 y • Onset age: 2 mo • Onset age: 9 mo • Onset age: 2 and 4.5 mo • Onset in infancy • Initial concern: abnormal eye movements • Initial concern: apnoeic events (? epileptic) • Initial concern: epileptic seizures • Initial concern: developmental delay a  We attempted to assign phenotypes to all published cases. Mostly, this was already provided by the authors but when not, we used the information available in the manuscript to determine phenotype. If we were unable to assign 1 phenotype (e.g., if the patient had features of several ATP1A3-associated phenotypes), we used the term intermediate phenotype (IP). Although this approach is limited by the quality of published data, it nevertheless provided an overview of the most common ATP1A3 phenotypes and associated genotypes. AHC was by far the most commonly reported phenotype, with 817 reported patients. A further 140 patients with RDP, 53 with CAPOS, 45 with RECA/FIPWE, 31 with EIEE, 13 with IP, 4 with D-DEMØ, 3 with isolated ataxia, and • Age: 43 y One of the 3 most common variants associated with a relatively mild phenotype of AHC (n = 77 patients published) • Onset age: 6-12 mo • Initial concern: moderate developmental delay • Hemiplegic episodes 2 with COS have also been reported ( Figure 2

CADD Scores
The mean CADD score for missense ATP1A3 variants published in the literature as likely pathogenic/pathogenic was 26.5 (SD: 2.04). The mean CADD score for variants reported in ClinVar as likely benign/benign was 7.729 (SD: 5.27). This difference was statistically significant (p < 3.49e-85). The CADD scores of the novel variants identified in our patient cohort were similar to the variants published (mean: 25.98, SD: 1.65) (Figure 3).

Constraint Analysis and Mutation Clusters
The most pathogenic and likely pathogenic variants identified both in the literature and this study lie within 6 clusters that correspond to benign missense variant deserts from the gnomAD database and vice versa (Figure 4). Benign missense constraint analysis identified gene deserts displayed as a heatmap. The 6 regions of constraint in which missense variation leads to pathogenicity are p.123-154, p.264-382, p.578-613, p.706-818, p.854-867, and p.887-955. These regions include key protein domains such as the transmembrane helices and the cytoplasmic P and N domains.

Discussion
In this study, we describe the genetic and clinical features of a heterogeneous cohort of individuals with pathogenic or likely pathogenic variants in the ATP1A3 gene. Previous studies have mostly recruited patients with a specific ATP1A3-related phenotype, such as AHC, RDP, CAPOS, or RECA, resulting in the description of clinically homogeneous cohorts. The starting point of this study differed in that patients were recruited based on the detection of ATP1A3 variants within a WES study aimed at using advanced genomics to diagnose individuals with previously undiagnosed developmental delay. The resulting cohort is clinically very heterogeneous, apart from a degree of cognitive impairment present in all patients. Only 2 individuals in our cohort, that were recruited through a collaborator rather than the DDD study, met Aicardi diagnostic criteria for AHC, and no individual met diagnostic criteria for either RDP or CAPOS. We believe that our data set represents a growing reality commonly faced by clinicians: The increasing phenotypic pleiotropy associated with ATP1A3 variants and identification of variants of uncertain significance poses a number of diagnostic challenges.
In addition to the broad phenotypic spectrum, our cohort was also genetically heterogeneous, with shared variants only seen among the members of the 2 participating families (individuals 19a and 19b and individuals 21a, 21b, and 21c). The genetic heterogeneity of the cohort might be reflective of its phenotypic heterogeneity because some phenotype-genotype association is evident for ATP1A3-related phenotypes. As discussed, all 53 CAPOS individuals published so far are linked to the variant Glu818Lys, and 45 published RECA/ FIPWE cases are linked to variants involving substitution of arginine at residue 756. Within the AHC phenotype, it is also well documented that individuals with variant c.2443G>A (p.Glu815Lys) are more severely affected than individuals with variant c.2401G>A (p.Asp801Asn), while the third most common variant c.2839G>A/C (p.Gly947Arg) results in a milder phenotype. 35,36 The 8 pathogenic variants mostly reported in the literature are associated with 1 specific phenotype ( Figure 1).
Looking at the phenotypes of individuals sharing the same ATP1A3 variant, it seems that while some genotypes are strongly associated with specific phenotypes, there are others that result in more phenotypic variability, such as those associated with c.2116G>A (p.Gly706Arg) and c.2839G>T (p.Gly947Trp). However, this is an observation based on a very small number of individuals, and it might be that given the opportunity to look at larger cohorts of the rarer ATP1A3 variants, mutation-specific phenotypes will arise as they have for c.2452G>A (p.Glu818Lys) and variants at amino acid residue 756. This is important information to gather because it may help clinicians provide families with more accurate prognosis after diagnosis.
Fifty percent of our patients were reported to have a history of epileptic seizures. In half of these, epileptiform features were seen on EEG. The epilepsy phenotype varied among patients, with some having focal seizures, while others had generalized epilepsy. This variability in epilepsy phenotype has been described previously in a study of 51 patients with an AHC phenotype, 37 where 32 individuals had an epilepsy diagnosis, 18 focal, and 11 generalized. Twenty-nine percent of our patients had a history of SE. Epilepsy treatment in this patient group is challenging. Many paroxysmal events are misdiagnosed as epilepsy and treated with antiepileptic medications with no benefit, whereas missing the epilepsy diagnosis in these patients might put them at risk of SE, which has been associated with severe aggravation of symptoms. 38 It is important to correctly classify the events patients are experiencing and have a low threshold to conduct video-EEG investigations if uncertain of the nature of occurring events. This is especially important in individuals with a higher risk of a more complex epilepsy phenotype, such as those carrying variant c.2443G>A (p.Glu815Lys).
Earlier studies of patient cohorts with a clinical phenotype of AHC mostly reported normal MRIs. 39 More recently, however, as the diverse phenotypes associated with ATP1A3 are evolving, several reports of abnormal neuroimaging have also been published including cerebellar atrophy 40 and PMG. [8][9][10] Eleven of our patients also had abnormal MRI, most commonly with cerebellar atrophy, in 1 case proven to be progressive.
A large European cohort study of 155 individuals with a clinical diagnosis of AHC conducted in 2010 reached the conclusion that AHC is a nonprogressive disease. 41 However, recent publications dispute this at least for some patients with an AHC phenotype. 38,42 Little long-term data have been published for patients with other ATP1A3-related phenotypes. In our cohort, 21% (n = 5) of individuals had regression of skills over time.
Although further longitudinal data are required, when counselling families, it may be important to discuss the risk of neurologic regression over time in ATP1A3-related disorders.
Traditionally, patients with an ATP1A3-related phenotype, such as AHC or RDP, have been diagnosed by using clinical diagnostic criteria. 43,44 In recent years with the association of ATP1A3 variants with a broadening clinical spectrum, this approach is not feasible for all patients because many do not fulfil classic phenotypic criteria. In addition, with broad genetic testing (gene panels, WES, and wholegenome sequencing) being brought into the diagnostic process at a much earlier stage, clinicians are often faced with an ATP1A3 variant in an undiagnosed patient, trying to decide whether it is responsible for the phenotype, rather than having already reached a clinical diagnosis and using genetic investigations to confirm or inform it. We compared the phenotypic characteristics of our cohort with the diagnostic criteria published for AHC, RDP, CAPOS, and D-DEMØ (Table 1). Using the classic diagnostic criteria for AHC first introduced by Aicardi 45 and subsequently used as inclusion criteria in clinical studies of patients with AHC, 41 only 2 patients in our cohort and none of our DDD cohort achieve a clinical AHC diagnosis. Very recently, these criteria have been revised to include the presence of an ATP1A3 variant and relaxed to describe a wider clinical spectrum of patients. 43 Applying the new criteria, 1 patient recruited through the DDD study and another 7 patients of our extended cohort are clinically diagnosable with AHC. None of our patients met criteria for either RDP or CAPOS, and 1 patient met criteria for D-DEMØ. Overall, we found that most of our patients cannot be grouped into any of the existing described phenotypes. Rosewich et al. 46 published major and minor criteria to support a diagnosis of an ATP1A3-related condition. The authors identified 5 major and 5 minor criteria for patients with infantile and early childhood onset, 6 major and 6 minor criteria for patients with childhood and adult onset, and 7 major and 7 minor criteria applicable for patients presenting at any age. Hence, for early onset, there are 12 major and 12 minor criteria overall, while for late onset, there are 13 major and 13 minor criteria. No cutoff is given as to how many criteria should be fulfilled to establish an ATP1A3-related condition diagnosis. In our cohort, 22 individuals had an onset in infancy/early childhood and 2 had an onset in later childhood or adulthood. All patients met at least 3 minor criteria. The only criteria met by all individuals in this cohort were cognitive impairment and negative family history or history suggesting autosomal dominant inheritance (both minor criteria). On average, patients met 3.4 major and 5 minor criteria. This approach of defining a spectrum of associated symptoms seems to be preferable for patients with ATP1A3-related disorders, if the threshold for the . We constructed a heatmap representing intolerance to missense changes, ranging from dark green through yellow, orange, and red with increasing intolerance (dark green = >20 variants, light green = 11-20 variants, yellow = 8-10 variants, light red 1-3 variants, dark red 0 variants). All pathogenic mutations from the literature (missense in green, small deletions/duplications in pink, and a sole frameshift mutation at residue 89 in red) and our cohort (all missense in yellow) and their distribution across the protein are shown (bottom section). The grey and white vertical shading represents mutation clusters and deserts, respectively. Highly constrained regions encompass transmembrane domains 2-9 (turquois), the phosphorylation site (orange), and the end of the nucleotide binding site (green), while the actuator domain (red) is situated in a mutation desert.
number of criteria needing to be fulfilled to prompt testing is kept low.
In addition to this, we found that in our cohort, looking for a combination of broad symptom categories, namely paroxysmal symptoms, hyperkinetic symptoms, neuropsychiatric symptoms, and cognitive impairment, rather than specific symptom combinations, was more likely to identify patients with ATP1A3-related disorder. A CADD score above 20 and a variant located within the mutation clusters in regions of constraint further support the diagnosis of an ATP1A3-related disorder.
There are limitations to our study. The phenotypic information was collected in retrospect through patient interview or case note review, rather than prospective evaluation. The phenotypic information available for published cases is variable and sometimes limited; patients are reported at different ages, clinical information is collected retrospectively, and different authors focus on different symptoms.
Two-thirds of all published individuals with ATP1A3-related disorders carry one of the 8 most common variants and display one of the 4 most common phenotypes: AHC, RDP, CAPOS, or RECA/FIPWE. However, the remaining third of individuals carry another 160 ATP1A3 variants, and their phenotypes are very variable forming a phenotypic continuum of paroxysmal, neurologic, developmental, and neuropsychiatric features. Nowadays, clinicians are often faced with novel variants of uncertain significance in genes associated with rare diseases. Looking for a combination of paroxysmal events, hyperkinesia, neuropsychiatric symptoms, and cognitive impairment, as well as evaluating CADD score and variant location, can aid in the diagnosis of an ATP1A3-related disorder.
The ongoing collection of phenotypic information of individuals carrying rarer variants will help us discover further mutation-specific phenotypes and aid in disease prognosis and management.