Early role for a Na⁺,K⁺-ATPase (ATP1A3) in brain development

Individuals with ATP1A3 Variants Display Cortical Malformation Phenotypes.

MRI of four unrelated individuals with de novo ATP1A3 variants revealed a range of PMG severity, from extensive bilateral frontoparietal PMG to unilateral PMG (Fig. 1A). Detailed descriptions of each case are included in SI Appendix, Clinical Descriptions.

PMG-Associated ATP1A3 Mutations Are Predicted to Disrupt Function.

ATP1A3 encodes the P-type Na⁺,K⁺-ATPase α3-subunit (NM_152296), an integral membrane protein that relies on ATP hydrolysis to transport ions (Na⁺ and K⁺) via 10 membrane-inserted helices (tTM1 to TM10). This Na⁺,K⁺-ATPase is functionally supported by the β- and FXYD-subunits (7). ATP1A3 has several conserved domains across α-isoform paralogs (α1 to α4) and orthologous genes (18), and is a highly constrained gene, with many fewer missense and LOF variants seen in the general population than predicted (gnomAD z-score of 6.33 and pLI score of 1.0, respectively). Taken together with ATP1A3’s known disease association with AHC, RDP, EIEE, and CAPOS (11), this suggests the identified variants in ATP1A3 are likely disease-causing (SI Appendix, Table S1).

PMG-associated ATP1A3 variants (p.Gln851Arg, p.Arg901Met, p.Leu924Pro, and c.2921+1G > A), and an AHC variant with PMG (p.Leu888Pro) (17), occur at sites that are highly conserved among the α-subunit paralogs (ATP1A1 to ATP1A4) and invariant in other vertebrate ATP1A3 orthologs (SI Appendix, Fig. S1B). In silico pathogenicity predictions for p.Gln851Arg, p.Arg901Met, p.Leu924Pro, and c.2921+1G > A suggest these variants are likely deleterious and disease-causing (SI Appendix, Table S1). Two amino acid substitutions reported here, Leu924Pro and Gln851Arg, have been previously associated with EIEE and AHC without MCD (13, 19), further supporting the wide spectrum of ATP1A3 phenotypes resulting from single amino acid variants (SI Appendix, Table S4). Leu924Pro pathogenicity has been demonstrated to affect ATPase trafficking to the plasma membrane, categorized functionally as LOF, leading to toxic effects in cells (13, 20). Together, the conservation of nucleotides, absence of variants in population databases, and previous LOF description of Leu924Pro (13, 20), suggest that these PMG ATP1A3 alleles are likely pathogenic.

PMG-associated missense ATP1A3 variants (p.Gln851Arg, p.Leu888Pro, p.Arg901Met, p.Leu924Pro) cluster within TM7 to TM8 segments (Fig. 1). The de novo ATP1A3 variant in Case A, NM_152296.4:c.2702G > T, p.Arg901Met (R901M), affects an arginine in the extracellular loop connecting TM7 and TM8 (Fig. 1C and SI Appendix, Fig. S1A), in the same segment as a previously described variant (p.Leu888Pro) occurring in an AHC individual with PMG (17). Additionally, these PMG-associated amino acids, Arg901 and Leu888, are in close proximity to Glu899, Gln904, and Gln905, which interact with the Na,K-ATPase β-subunit (21), suggesting these variants may affect the critical α-β interaction necessary for membrane localization (Fig. 1C). Case C possesses a de novo ATP1A3 variant, NM_152296.4:c.2771T > C, (p.Leu924Pro), altering an amino acid in the TM7 helix just upstream of the loop (Fig. 1C). Case D possesses a de novo ATP1A3 variant, NM_152296.4:c.2552A > G, p.Gln851Arg, altering an amino acid in TM7 (Fig. 1C). The clustering of these missense variants between TM7 and TM8 suggests a disruption of specific protein–protein interactions; on the other hand, the de novo ATP1A3 variant in Case B, NM_152296.4:c.2921+1G > A, abolishes the essential splice donor site of exon 21 (SI Appendix, Fig. S1), and similar ATP1A3 splice site variants are known to cause AHC (22). Therefore, the PMG-associated variants found in Case B and Case C likely represent the most significant lost Na⁺,K⁺-ATPase functionality, with the most severe PMG patient phenotypes likely reflecting the most severe LOF effects (i.e., Case B, splice donor variant likely resulting in haploinsufficiency), compared to other less-severe ATP1A3-associated conditions. Importantly, the four de novo-affected individuals did not possess any alternative variants in known MCD genes or related MCD pathways.

ATP1A3 mRNA Expression in the Human Fetal Neocortex.

To investigate the transcriptional trajectory of ATP1A3 during human brain development underlying the prenatal phenotypes described in this study, we first contrasted the temporal expression patterns of ATP1A1 to -3 across brain regions (ATP1A4 does not maintain any significant level of brain expression), mining bulk neocortical transcriptome data from the Allen Brain Atlas (23), ranging from 12 wk postconception (wpc) to adulthood (40-y-old) (Fig. 2B). We find that prenatal expression level of ATP1A3 is higher than that of its paralogs (ATP1A1 and -2), and ATP1A3 persists postnatally (Fig. 2B). To gain further insights into the cell-biological mechanisms underlying the developmental cortical phenotypes described in this study, we characterized the patterns of ATP1A3 expression in the midgestational fetal human neocortex (19 to 21 wpc). This developmental age offers a uniquely informative window into corticogenesis as many and diverse biological processes—including neurogenesis, neuronal migration and differentiation, and cortical folding—co-occur at this age. At this stage, early-born ENs have settled and started to differentiate within the innermost strata of the cortical plate (CP; prospective layers 6a-5), while later-born ENs progressively migrate past their predecessors to occupy the most superficial aspect of the CP (prospective layers 4-2) (24). A large fraction of upper-layer ENs are still being generated in the germinal zones, while others—en route to the CP—are withheld within the SP, a transient developmental layer crucial for corticogenesis, where ENs are primed by thalamic input and form transient synaptic connections with one another and with resident SP neurons (prospective layer 6b) (25).

To investigate ATP1A3 expression in its tissue context, we performed RNA in situ hybridization (ISH) in coronal sections of fetal neocortex at 19 wpc from the perisylvian region, the area most commonly affected by PMG. ATP1A3 mRNA was virtually undetectable in the germinal zones where dividing neural stem cells reside (ventricular and subventricular zone, VZ and SVZ), but abundant in postmitotic neuronal layers, with highest expression in the inner CP and SP (Fig. 2 C and D). Multiplexed fluorescent ISH analysis of ATP1A3 with markers of two populations of cortical NPCs, radial glia, and intermediate progenitor cells, confirmed ATP1A3 has low to zero expression in these dividing cells or their immediate progeny in either the VZ or the SVZ (Fig. 2C).

ATP1A3-associated diseases display a range of cerebellar deficits throughout childhood development (11, 26, 27). We therefore temporally profiled cerebellar ATP1A3 expression from bulk RNA-seq data sampled from human embryonic week 12 to adult cerebellum (23). We show a differential enrichment of ATP1A3 expression to the prenatal period, compared to ATP1A1 and ATP1A2, yet ATP1A3 also persists postnatally (Fig. 2F). To better understand ATP1A3’s developmental profile, we next performed ISH to evaluate ATP1A3 expression in fetal cerebellum tissue (19 wpc), demonstrating an exclusive enrichment of ATP1A3 within the Purkinje cell layer (Fig. 2E).

Fetal Neocortex Single-Cell Analysis Demonstrates Robust ATP1A3 Enrichment to Resident SP Neurons and Rostral Neocortical ATP1A3 Enrichment.

Despite the high degree of compartmentalization of cell types into laminae, biological processes in the developmental neocortex are highly dynamic, and cellular compartments are heterogeneous in cell-type composition. scRNA-seq enables large-scale, simultaneous comparisons of gene expression across thousands of individual cells, thus bridging functional data from human genetics, immunohistochemistry, and cell-type–specific biology (28). Additionally, the limited spatial resolution of bulk RNA data from the Allen dataset does not allow cell-type identification within layers, which can result in miscategorized nonresident cells as they traverse layers (SI Appendix, Fig. S2G) (23). To this end, we performed Drop-seq on 125,942 individual cells from a 21-wpc human neococrtex (Fig. 3A and SI Appendix, Fig. S2A). Given the widespread pattern of PMG phenotypes from the perisylvian to frontal cortex localization described here, we sought to investigate ATP1A3 expression across presumptive neocortical areal domains. To build our single-cell expression dataset, we collected cells from 11 adjacent sections spanning the dorsomedial and rostrocaudal axes of the neocortex, validating and refining the positional identity of these serial samples using the graded expression of cortical patterning genes as benchmarks (Fig. 3A and S2 A–C) (29). Using unsupervised statistical methods (Materials and Methods), cells were partitioned into four transcriptionally distinct clusters, which corresponded to the major neural cell classes in the midgestational neocortex: ENs, INs, NPCs, and glia (Fig. 3 B and C and SI Appendix, Fig. S2D). We excluded from this analysis all nonneural cell types as our scope of inquiry was specifically regarding ATP1A3 dysfunction in neurons, and nonneural expression of ATP1A3 is limited (Fig. 3B and SI Appendix, Fig. S2D). Within each of the main neuronal cell classes we could further resolve transcriptionally distinct cell subtypes and states, which we interpreted by the distinctive enrichment of marker genes in each group (SI Appendix, Fig. S2D). Among ENs, consistent with the developmental landscape of midgestational corticogenesis described above, we could identify newborn and migratory upper-layer ENs at different stages of differentiation (clusters EN-0 to -3), and postmigratory deep-layer ENs expressing marker genes of maturing layer 6-5, including markers of resident SP neurons (EN-4) (Fig. 3C). Across fetal cortical regions, ATP1A3 expression was observed in intermediate progenitor cell clusters, and more uniformly expressed in oligodendrocyte progenitor cells, INs, and ENs (Fig. 3C and SI Appendix, Fig. S2 D and E). Of these 18 cell types analyzed, the EN-4 cluster demonstrated the most significant ATP1A3 enrichment and the focus for detailed analysis.

The EN-4 cluster was composed of cells sourcing from all cortical regions, and was marked by enriched expression of SP-specific genes (CRYM and LPL) (Fig. 3C), as well as by the expression of genes associated with neuronal maturation (e.g., axonal genes NEFL/NEFM and synaptic genes SEMA3E) (Fig. 3H and SI Appendix, Fig. S2E). This is consistent with the notion that the earliest-born neurons (a subset of which will persist in L6b in the adult cortex) settle and mature early in the developing SP (30). Within this cluster, ATP1A3 appeared to be differentially expressed across the cortical region, with highest expression levels found in cells from frontal samples, intermediate levels in cells from medial samples, and lowest levels in cells sampled from more caudal cortical regions (Fig. 3C). In agreement with this, expression of ATP1A3 in the EN-4 cluster was found to be enriched in neurons coexpressing SP markers and frontal cortex markers, but much less so in neurons expressing caudal cortex markers (Fig. 3D and SI Appendix, Figs. S2 and S3).

Cortical development proceeds along a rostral-to-caudal maturation gradient, with asynchronous growth patterns accounting for the differential expansion of frontal and caudal cortical regions (31). This implies that, at any given time, neurons in frontal regions are on average more mature than their caudal counterparts. To investigate whether the observed ATP1A3 enrichment in the EN-4 cluster might reflect a differential maturation state of EN neurons along the rostrocaudal axis, we performed pseudotime analysis of ATP1A3 across all ENs (Fig. 3E). To do this, we used the Monocle3 algorithm, an established analytical tool allowing single-cell transcriptomes to be ordered and connected along a branched topology based on the differential expression of variable genes (32). By selecting an “origin” cell, this algorithm allows projection of a “pseudo-differentiation” trajectory onto the network topology (33). This analysis revealed a positive correlation between ATP1A3 expression and EN differentiation, and may suggest that ATP1A3 may be required during neuronal differentiation, as neurons transition to a mature state, most readily observed in resident SP neurons (Fig. 3E).

Finally, as the Na⁺,K⁺-ATPase is a heteromeric complex, we sought to identify potential isoform-specific and cell-type–specific vulnerabilities during corticogenesis; we compared the expression patterns of the ATP1A1 to -3 paralogs across cell clusters (SI Appendix, Fig. S3A). We found that expression of ATP1A2 is restricted to NPCs (and specifically to radial glia), and is virtually undetectable in neurons, in an almost mutually exclusive pattern to ATP1A3. ATP1A1 expression more closely mimicked ATP1A3 expression, albeit its expression in NPCs was higher, and partially overlapped ATP1A2 (SI Appendix, Fig. S3A). Accordingly, pairwise correlation analysis of the three isoforms across clusters revealed that expression of ATP1A1 and ATP1A3 correlates more strongly with the EN-4–SP cluster, while expression of ATP1A1 and ATP1A2 correlates more highly with NPC clusters (SI Appendix, Fig. S3 B and C). Across all cell-type clusters, pairwise correlation analysis indicated a positive correlation, though moderate, between ATP1A1 and ATP1A3, but weak-to-no correlation between ATP1A1 and ATP1A2, and between ATP1A2 and ATP1A3 (SI Appendix, Fig. S3 B and C). Interestingly, ATP1A3 also seems to have an exclusive enrichment to interneurons, not shared by ATP1A1. Taking these data together, this analysis suggests that single-cell variable expression of Na⁺,K⁺-ATPase isoforms could underlie compensation or vulnerability mechanisms for disease haploinsufficiency or differential penetrance within specific cell populations. Finally, we analyzed a previously published smaller human fetal scRNA-seq data (34) (∼4,300 cells), which lends support to our findings that ATP1A3 is ubiquitously expressed across several cell types (SI Appendix, Fig. S9), with an EN enrichment.

ATP1A3 Is Enriched in PV Interneurons and ENs in Infant Neocortex.

Given that ATP1A3 expression persists in the postnatal neocortex (Fig. 2B), and that several known ATP1A3 mutations show postnatal onset of symptoms, including epilepsy (11, 12), we profiled ATP1A3 expression in the infant neocortex. Using Drop-seq, we profiled 51,878 nuclei from an ∼8-mo-old human cortex, sampled from prefrontal, temporal, occipital, and parietal lobes (Fig. 4A). Unsupervised hierarchical clustering sorted nuclear transcriptomes into 3 major classes – ENs, INs, and glia – subdivided into 24 transcriptionally distinct cell classes (Fig. 4A and SI Appendix, Fig. S4). Within the EN clade, clusters stratified largely by layer and neuronal subtype, rather than cortical area, with several clusters enriched in gene markers of deep layers 5 and 6 (Fig. 4B; see SI Appendix, Fig. S4 for full clading). Within the IN clade, cells stratified by developmental origin (caudal or medial ganglionic eminence) and major “cardinal types,” according to previous studies (35). In the infant neocortex, similar to the fetal neocortex, ATP1A3 was enriched in neurons compared to glia, but unlike in the fetal dataset, ATP1A3 was more highly expressed in INs than in ENs (Fig. 4B and SI Appendix, Fig. S5). Among neuronal clusters, and across cortical regions, ATP1A3 expression was fairly variable, with the lowest expression in V1/occipital cortex cells, and wide range of expression levels within deep and upper layer EN clusters (Fig. 4B). However, no specific enrichment of ATP1A3 expression was found in L6b ENs (i.e., the postnatal remnant of the fetal SP layer), suggesting that fetal enrichment of ATP1A3 in the SP is developmentally transient, with maturation and migration of cells continuing to populate the upper layers. Of note, the lack of specific enrichment in postnatal SP may also reflect the fact that, unlike the fetal dataset, the postnatal dataset was built from nuclear RNA rather than whole-cell RNA, so that some transcripts may be underdetected.

Across IN clusters, expression of ATP1A3 was highest in cells sampled from the frontal lobe (prefrontal BA10), and more strikingly so within PV⁺ INs (Fig. 4B and SI Appendix, Fig. S4). This differential expression pattern raised the hypothesis that ATP1A3 may be required dynamically across neuronal types, cell-biological processes, and across developmental ages. To gain insights into the nature of the cell-biological signature associated with ATP1A3 expression, we compared the top 100 genes correlating with ATP1A3 in EN and IN clusters. Interestingly, we found that 50 of 100 ATP1A3-correlating genes were shared across cell types, suggesting that ATP1A3 partakes in similar transcriptional programs, and thus potentially similar function, in different neuronal types. In both ENs and INs, among the top genes correlating with ATP1A3 was FXDY6, the FXYD-subunit of the ATPase complex (Fig. 4D). We did find, however, that the β-subunit ATP1B1 (β1) correlated with ATP1A3 only in ENs, and ATP1B2 (β2) in INs, suggesting that, at least in part, ATP1A3 may form cell-type–specific heteromeric ATPase complexes in different neuronal classes (Fig. 4D and S3 A and B). Gene ontology (GO) enrichment analysis of the top 100 genes most strongly correlated with ATP1A3 in both ENs and INs revealed significant overrepresentation of gene products associated with TM ionic transport and the somatodendritic compartment, including dendrites and synapses (Fig. 4D and SI Appendix, Table S2). This is consistent with the established role of ATP1A3 in synaptic transmission and membrane potential and suggests that ATP1A3 may partake in a shared transcriptional network for which ATP1A3 variants may perturb aspects of neuronal physiology central to many cell types at different stages of their maturation process.

Given the association of ATP1A3 pathological variants with human disease phenotypes affecting higher-order cognitive functions with no clear counterpart in animal models, it was of interest to investigate the degree to which ATP1A3 enrichment in INs may be conserved among humans, nonhuman primates, and rodents. We took advantage of a recently published scRNA-seq dataset profiling gene expression in homologous IN types in humans, nonhuman primates (macaques and marmosets), and mice (36). We found that relative ATP1A3 expression levels across IN types, both in the neocortex and in the striatum, are largely conserved across species, with shared highest ATP1A3 expression in the fast-spiking PV⁺ INs (SI Appendix, Fig. S6). This suggests that the observed differential expression of ATP1A3 across neuronal types is genetically encoded and evolutionarily conserved.

Human Subjects and Samples.

Individuals presented herein were identified and evaluated in a clinical setting, and biological samples collected after obtaining written informed clinical and research consent. Human subject research was conducted according to protocols approved by the institutional review boards of Boston Children’s Hospital, Beth Israel Deaconess Medical Center, and the Mayo Clinic.

Human Genetics Sequencing and Analysis.

For Cases A and B, WES and data processing were performed by the Genomics Platform at the Broad Institute of MIT and Harvard. Libraries from DNA samples (>250 ng of DNA, at >2 ng/μL) the Human Core Exome Kit from Twist Biosciences was used to capture target regions (∼38-Mb target), and sequencing was performed on the Illumina NovaSEq. 6000 (150-bp paired reads) to cover >96% of targets at 20× and a mean target coverage of >100×. Sample identity quality-assurance checks were performed on each sample. See SI Appendix for additional information.

Phenotypic Assessment.

All affected individuals and clinical data were examined by neurologists and geneticists and radiologists, and PMG was diagnosed using criteria described previously (1, 16). The SI Appendix summarizes their phenotypes and detailed clinical and radiographic evaluations.

Human Brain Tissue Preparation and mRNA ISH.

Human brain tissue preparation and mRNA ISH were performed as previously described by Smith et al. (4). Briefly, following fixation (4% PFA) and cryoprotection (30% sucrose), brains were frozen using isopentane on dry ice. Samples were sectioned at 20- to 30-μm thickness (Leica Cryostat), mounted immediately onto warm charged SuperFrost Plus slides (Fisher), and stored at –80 °C. We followed the manufacturer’s standard protocol for multiplex fluorescent ISH (Multiplex v2 kit, Advanced Cell Diagnostics). See SI Appendix, Supplemental Materials and Methods for more details.

Bulk Human Cortex Gene-Expression Analysis.

The Allen Human Brain Atlas (ABA) publishes a rich dataset of cortical genetic expression across cortical brain regions, from age 8 wpc to adult ages (23). BrainSpan data analysis of ATP1A1 (chr1:116,915,289–116,952,883, GRCh37/hg19), ATP1A2 (chr1:160,085,548–160,113,381, GRCh37/hg19), and ATP1A3 (chr19:42,470,733–42,498,384, GRCh37/hg19) was performed. ATP1A4 is not expressed in the CNS and was not included as the read counts were near zero. See SI Appendix, Supplemental Materials and Methods for more details.

Drop-Seq Isolation and Sequencing.

Single-nucleus and single-cell suspensions were processed for Drop-seq as described in detail (5, 6), with modifications detailed by Krienen et al. (36) (see SI Appendix for additional details). Single-cell cDNA libraries were processed and sequenced using Illumina short-read sequencing on a Nova-seq platform at the Broad Institute. Libraries were sequenced at an average depth of >8 reads per unique molecular identifier (UMI). Sequencing reads were aligned to the hg19 human reference genome. For the fetal dataset, only reads mapping to exons were used; for the postnatal dataset, reads mapping to both exons and introns were used. Software and core computational analysis for alignment and downstream processing of Drop-seq sequencing reads are freely available at https://github.com/broadinstitute/Drop-seq. Cell selection was performed as described previously (63). See SI Appendix, Supplemental Materials and Methods for independent component analysis, clustering, and statistical analysis details.

Graphics.

Graphics were prepared in Adobe Illustrator and using BioRender software.