Genome-wide screen for modifiers of Na+/K+ATPase alleles identifies critical genetic loci

Background Mutations affecting the Na+/ K+ATPase (a.k.a. the sodium-potassium pump) genes cause conditional locomotor phenotypes in flies and three distinct complex neurological diseases in humans. More than 50 mutations have been identified affecting the human ATP1A2 and ATP1A3 genes that are known to cause rapid-onset Dystonia Parkinsonism, familial hemiplegic migraine, alternating hemiplegia of childhood, and variants of familial hemiplegic migraine with neurological complications including seizures and various mood disorders. In flies, mutations affecting the ATPalpha gene have dramatic phenotypes including altered longevity, neural dysfunction, neurodegeneration, myodegeneration, and striking locomotor impairment. Locomotor defects can manifest as conditional bang-sensitive (BS) or temperature-sensitive (TS) paralysis: phenotypes well-suited for genetic screening. Results We performed a genome-wide deficiency screen using three distinct missense alleles of ATPalpha and conditional locomotor function assays to identify novel modifier loci. A secondary screen confirmed allele-specificity of the interactions and many of the interactions were mapped to single genes and subsequently validated. We successfully identified 64 modifier loci and used classical mutations and RNAi to confirm 50 single gene interactions. The genes identified include those with known function, several with unknown function or that were otherwise uncharacterized, and many loci with no described association with locomotor or Na+/K+ ATPase function. Conclusions We used an unbiased genome-wide screen to find regions of the genome containing elements important for genetic modulation of ATPalpha dysfunction. We have identified many critical regions and narrowed several of these to single genes. These data demonstrate there are many loci capable of modifying ATPalpha dysfunction, which may provide the basis for modifying migraine, locomotor and seizure dysfunction in animals. Electronic supplementary material The online version of this article (doi:10.1186/s13041-014-0089-3) contains supplementary material, which is available to authorized users.


Background
In many organisms, highly conserved Na + /K + ATPases are responsible for maintaining ion gradients across the plasma membrane through ATP-dependent asymmetric translocation of Na + and K + ions. These ion gradients maintain the resting potential of cells, which facilitates neural signaling and many essential secondary processes. Mature Na + /K + ATPase complexes are heteromultimers of alpha, beta, and gamma subunits in mammals. Flies express only the alpha and beta subunits, the former of which is known as ATPalpha. Like its mammalian homologue, ATPalpha contains ten transmembrane domains and has the ATP-dependent catalytic activity essential for pump function [1][2][3].
Mutations affecting the alpha subunit of the Na + /K + ATPase in humans are associated with at least three human diseases: Rapid-onset Dystonia Parkinsonism (RDP), Familial Hemiplegic Migraine (FHM), and Alternating Hemiplegia of Childhood (AHC; [4]). RDP is a severe DOPA non-responsive form of dystonia the etiology of which is poorly understood [5]. FHM, possibly the most severe form of migraine, is associated with a debilitating partial paralysis, and currently is largely untreatable [6]. AHC is a severe childhood locomotor disease associated with recurring acute bouts of paralysis and muscle weakness, and general developmental delays (reviewed by [7]). Recently Sasaki and colleagues have described several children who seem to have a disease intermediate to AHC and RDP [8]. All of these diseases are complex neuromuscular conditions associated with marked locomotor dysfunction and for which the underlying pathogenesis is poorly understood.
It has previously been shown that mutations in ATPalpha result in profound neural and locomotor dysfunction in Drosophila [23,[36][37][38][39][40]. Hypomorphic ATPalpha alleles, such as ATPalpha 2206 , display BS paralysis and phenocopy injection of the selective Na + /K + ATPase inhibitor, ouabain [39]. The ATPalpha DTS1 mutation is a dominant, conditional, gain-of-function, missense mutation [23]. The mutation results in an E982K substitution near the protein's C-terminus (short isoform numbering). ATPalpha DTS1 heterozygotes exhibit rapid paralysis at 38°C with complete penetrance. This is thought to be a result of conditional neuronal hyperexcitability caused by the mutation [23]. ATPalpha CJ5 and ATPalpha CJ10 are also dominant missense mutations affecting evolutionarily conserved amino acids [36]. However, they each exhibit unique locomotor phenotypes. ATPalpha CJ5 behaves like a loss-of-function allele of ATPalpha, exhibiting haploinsufficiency and BS paralysis [36]. ATPalpha CJ10 exhibits BS and progressive TS phenotypes, suggesting this is a loss-of-function allele that exhibits weak gain-offunction features, which are uncovered with age [36]. Thus, ATPalpha DTS1 , ATPalpha CJ5 , and ATPalpha CJ10 are all dominant, phenotypically well-characterized, and possibly functionally distinct, conditional locomotor mutants. Such alleles are ideally suited for a modifier screen. Using multiple alleles of ATPalpha increases the power of the screen and affords the likelihood of identifying allelespecific modifiers. Furthermore, to our knowledge, this is the first report of a genome wide genetic screen in any animal system using three distinct alleles of the same gene in parallel to identify allele-specific interactions.
Deficiency screens have been effectively used for elucidating novel gene interactions in Drosophila using various phenotypes [41][42][43]. Deficiency (Df) strains each have a unique deletion of a segment of the genome. Phenotypically screening for genetic interactions between defined point mutations and an individual defined deficiency is an efficient way to identify modifier loci. Using a collection of Dfs covering a high percentage of the genome (95-98%), one can identify critical modifier loci anywhere in the genome. This provides an efficient yet powerful and unbiased forward genetic approach. Critical loci can often be narrowed to single genes using smaller deficiencies and single gene disruptions. We have performed such a screen using ATPalpha DTS1 , ATPalpha CJ5 and ATPalpha CJ10 , identified 64 critical modifier intervals, and successfully confirmed 50 single-gene modifiers, including numerous novel loci of interest. These data suggest the existence of many susceptibility loci capable of modifying migraine, locomotor and seizure dysfunction in animals and provide a rich data set from which new targets for anti-migraine or anti-epileptic drugs could be drawn.

Primary genetic modifier screen
To identify new genes that interact with ATPalpha we performed a deficiency screen using three characterized alleles: ATPalpha DTS1 , ATPalpha CJ5 and ATPalpha CJ10 . We used the Bloomington Stock Center deficiency (Df) kit that covers approximately 98% of the Drosophila genome. All  of the 467 Df strains we received were tested with at least one ATPalpha mutant allele and the vast majority of strains were tested with multiple alleles (see Table 1). Each of the three ATPalpha mutants was mated to each Df line. F 1 progeny bearing ATPalpha DTS1 and each deficiency were subjected to TS assays while progeny bearing ATPalpha CJ5 or ATPalpha CJ10 and each Df were assayed for BS. The data from the primary screen were organized graphically by average time to recovery or paralysis for each double mutant ( Figure 1). In each case, the resulting data formed a largely normal distribution. Double mutants that deviated significantly from the mean were termed putative enhancers or suppressors and were tested again in a verification screen. The workflow for the genetic screen is depicted in Figure 2. In the primary screen, 1137 interactions were examined for the three conditional locomotor mutants identifying 117 putative enhancer, suppressor, or synthetic lethal regions. These interactions were examined further in the verification screen.

Verification screen
To mitigate the effect of false positives and confirm that interactions were reproducible before pursuing them further, we performed a verification screen (an independent experiment) with the putative modifiers. We began the verification with 89 ATPalpha DTS1 modifiers ( Figure 1A), 69 ATPalpha CJ5 modifiers ( Figure 1B), and 78 ATPalpha CJ10 modifiers ( Figure 1C). After verification, we took advantage of having two data sets (primary and verification screen) and created a formula to determine the reproducibility of each putative genetic interaction (see Materials and Methods). We calculated a reproducibility index (RI) and used it to help us identify the most promising critical intervals. Dfs with the highest RIs were prioritized for mapping and secondary screening. This approach yielded 7 putative ATPalpha DTS1 enhancers, 12 suppressors, and five synthetic (See figure on previous page.) Figure 1 Distribution of phenotypic modifiers identified through a deficiency screen. A-C) ATPalpha mutant animals also bearing individual unique chromosomal deficiencies (Df) were assayed for conditional locomotor function to identify modifiers. The data reveal a largely normal distribution centered around a typical response (blue) for each mutant. Those deviating from the typical response were termed putative enhancers (yellow) or suppressors (red). A) ATPalpha CJ5 , Df double mutants and B) ATPalpha CJ10 , Df double mutants were assayed for recovery from mechanical stress at adult day 15. C) ATPalpha DTS1 , Df double mutants were assayed for time to TS paralysis on adult day 1. A-C) The mean response is shown as a dashed green line. +/− 0.5 Std. Dev. are indicated by gray shading.

Figure 2
Schematic of the deficiency screen workflow. Using the Bloomington deficiency kit, 1137 initial interactions were screened using ATPalpha CJ5 , ATPalpha CJ10 , or ATPalpha DTS1 . Putative enhancers and suppressors were selected for verification with a larger sample size. Any verified interacting deficiencies were deemed critical intervals. Once critical intervals were selected a screen for single gene modifiers from within the intervals was performed using available classical mutants and transgenic RNAi strains. If a modifier was found it was retested with other ATPalpha alleles to determine whether the interaction was allele-specific. lethal (enhanced to lethality) combinations ( Table 2). The ATPalpha CJ5 screen yielded 13 enhancers, 10 suppressors, and four synthetic lethal combinations ( Table 3). The ATPalpha CJ10 primary screen yielded 17 enhancers, 11 suppressors, and one synthetic lethal (Table 4).

Single gene identification and testing
After the verification of critical intervals, genes contained within these intervals were selected for testing. Where practical large intervals were narrowed using smaller Dfs. We obtained classical alleles for integral genes from Bloomington, when possible. Each single gene mutant was mated to the ATPalpha allele it putatively modified and to w 1118 . All single gene mutants displayed no BS or TS phenotype as heterozygotes (data not shown). Heterozygous double mutants were again assayed for TS or BS with age matched controls. Significant interacting single gene mutants were also tested with the other ATPalpha alleles ( Figure 2). Twenty single gene interactions were found using classical mutants for ATPalpha DTS1 including 19 single gene enhancers and one single gene suppressor. Ten single gene suppressors were found for ATPalpha CJ5 . Twenty-four single gene interactions were found with ATPalpha CJ10 , all but one of which showed suppression of the mutant phenotypes. In total, 35 single gene interactions were found and, importantly, 14 different genes had effects with more than one ATPalpha allele (Table 5).
Gal4 driven RNAi strains result in a loss-of-function phenotype and are well-suited to confirm the hypomorphic effect of a heterozygous Df. RNAi knockdown was driven with da-Gal4 in ATPalpha mutant backgrounds. Daughterless transcripts are stably expressed throughout the life of a fly and are detectable in every tissue by the FlyAtlas affymetrix array analysis [72,73]. We used this driver to ubiquitously express the RNAi constructs and mimic the effect observed with the Df. RNAi mediated knockdown of candidate genes was compared to age matched controls lacking the UAS-RNAi construct. Twenty-five different genes showed interactions using this method, including 10 genes already identified in the classical mutant screen. Fourteen interactions were identified with ATPalpha DTS1 , with nine enhancers and five suppressors. Seventeen interactions, with two enhancers and 15 suppressors, were identified for ATPalpha CJ5 . Thirteen interactions, all suppressors, were confirmed with ATPalpha CJ10 . In total 15 different genes showed a genetic interaction with two or more ATPalpha alleles (Table 6). In total we have identified 50 genes that interact with ATPalpha, 25 of which were confirmed to interact with at least two independent alleles.

Discussion
The Na + /K + ATPase is central to maintaining cytosolic ion homeostasis suggesting that many of the genes identified in our screen would encode proteins that affect cytosolic ion concentrations and, indeed, this was the case ( Figure 3A). Nearly 25% of the genes we identified encode proteins with a known function in ion transport. In our search for single gene modifiers we selected genes known to be expressed in the nervous system. Unsurprisingly,~50% of our hits are known to cause some neuronal defect when knocked out ( Figure 3B). For example, most of the cell adhesion and paracrine signaling molecules we found, such as Galectin (Tables 5 and 6, Figure 4), Glt (Tables 5 and 6), and Sema-1a (Table 5) were previously known to cause malformed or improperly targeted synapses [45,[48][49][50]. However, about half of our genes were not previously linked to neuronal function. Additionally, many genes we identified encode proteins implicated in signaling pathways. In particular we found proteins involved in developmental signaling pathways, such as Wingless and Hedgehog (rasp (Tables 5 and 6) and slmb (Table 5)), and neuronal growth and survival pathways (spz5 (Tables 5 and 6)). Spz5 ( Figure 5) is especially interesting because it has recently been identified as a Drosophila neurotrophin that signals through a Toll receptor [61,62]. Both Slmb and Cact (Table 5) were also identified by our screen and both may function downstream of Spz5. In mammals and flies, Toll signaling activates NF-κB transcription factors, typically through the degradation of an inhibitor of NF-κB (I-κB), such as Cact. Phosphorylated I-κB is targeted for degradation, allowing NF-κB-like transcription factors to translocate to the nucleus. Slmb and its mammalian homolog β-TrCP regulate phospho-I-κB. β-TrCP, and likely Slmb, target an E3 ubiquitin ligase complex to phospho-I-κB and mediate its degradation via ubiquitin proteasome system [68]. Interestingly, we have also identified Uch-L5 (Table 6) in our screen, a member of the 26S regulatory complex which is likely responsible for the deubiquitylation of proteins as they enter the 26S proteasome [81].
Previously published studies of Slmb, and Spz5 show that they play an important role in neural development. Slmb is involved in pruning dendrites and axons during   [84] and Spz5 is a neurotrophic signal and axon guidance cue in the embryonic nervous system [61]. Interestingly, animals heterozygous for a loss of function allele of either gene displayed no phenotype in neurons [61,85].
In contrast, our screen examined heterozygous double mutants and found large effects, suggesting ATPalpha mutants are sensitive to otherwise inconsequential changes in neuronal development or another unappreciated function of these proteins. Furthermore, a seemingly insignificant disruption of neuronal survival signals early in development may have dramatic phenotypic effects for ATPalpha mutants since heterozygosity of Slmb, or Spz5 suppressed the loss-offunction ATPalpha phenotype. Additionally, numerous developmental genes were identified implying that neurodevelopmental changes may profoundly affect Na + /K + ATPase function or this is a general and potent mechanism to modulate locomotor function. Another interesting possibility is that loss-of-function ATPalpha mutations are disrupting neuronal development through alterations in NF-κB signaling. It has been shown that sub-inhibitory concentrations of ouabain activate NF-κB via an Na + /K + ATPase dependent mechanism in rat kidney cells. The effect is mediated by slow, inositol triphosphate-dependent, calcium oscillations likely caused by shifting electrochemical gradients [86]. More recently, agrin, a protein involved in synapse formation at NMJs and in the CNS, has been shown to bind to and inhibit the mammalian Na + /K + ATPase α3 isoform. Furthermore, agrin seems to bind at the same site as ouabain because a protein fragment can prevent ouabain inhibition of the Na + /K + ATPase [87]. Thus it is possible that agrin exerts its effects through NF-κB. If a similar pathway exists in flies it would likely be constitutively active in our loss-offunction mutants and its dysregulation could cause developmental changes, which might increase seizure susceptibility. This is consistent with our finding that knockdown of proteins required for NF-κB activation suppresses seizures in our loss-of-function mutants. NF-κB activation may be caused by calcium oscillations [86], making it possible that some of the calcium channels we found also play a role in this pathway. FKBP59 ( Figure 6) is particularly interesting because it inhibits an inositol triphosphate sensitive, non-specific calcium channel, TrpL [53]. Inhibition of calcium channels would likely be required in calcium oscillations. The preponderance of hits related to the NF-κB pathway suggests a possible role for this pathway in seizure pathogenesis.
In most cases the ATPalpha CJ5 and ATPalpha CJ10 mutant phenotypes were modified in the same direction (enhancement or suppression) and they never had opposite phenotypes in our screen. This is consistent with the finding that both exhibit loss-of-function characteristics. The ATPalpha DTS1 phenotype, however, usually contrasted with the phenotypes of ATPalpha CJ5 and ATPalpha CJ10 . This is intriguing as ATPalpha DTS1 is a gain-of-function mutation that can be reverted by a second site mutation to give the characteristic ATPalpha loss-of-function phenotype [23]. In accord with this fact the vast majority (~80%) of the single gene interactions with ATPalpha DTS1 modified the loss-offunction alleles in the opposite direction or not at all. Reduction of Ppk11, Ppk21, and Ppk24 function all suppressed the phenotypes of ATPalpha DTS1 and another allele. All three are predicted epithelial sodium channels (DEG/eNaCs) that function in nociception, mechanosensation, gustation and other sensory functions (Reviewed in [88] and [89]). Thus, it is possible that altered sensory function may underlie the ATPalpha DTS1 paralysis phenotype and that a reduction in the ability of Many genes had an interaction with more than one allele, although some appear to be allele specific. Double mutants were compared to ATPalpha * /+ and heterozygous classical mutant controls. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001. # Function per flybase.org and / or listed citation. the double mutant animals to sense the elevated temperature is sufficient to suppress the TS paralysis. This possibility is consistent with the kinetics of recovery after animals are returned to the permissive temperature. This is also intriguing as the locomotor dysfunction resulting in hemiparalysis in FHM patients has been reported to be associated with sensory dysfunction and FHM patients report having prolonged visual auras [90][91][92].

Conclusions
FHM, RDP, and AHC are complex human neurological diseases associated with mutations affecting the catalytic alpha subunit of the Na + /K + ATPase [4][5][6]. Currently, there is no cure or effective treatment for these diseases. Using three Drosophila strains with different missense mutations in ATPalpha we have performed a large-scale deficiency screen to identify novel genes that interact with the gene encoding the Na + /K + ATPase alpha subunit. In total, we have identified 50 genes that interact with ATPalpha, 25 of which were demonstrated to interact with at least two independent alleles. We have also implicated 50 critical intervals/deficiency regions for which we have yet to identify individual genes that interact with ATPalpha (Tables 2, 3 and 4). Modifier loci that encode proteins expressed in the adult, especially those that phenotypically suppress ATPalpha dysfunction, provide proteins/pathways that could be viable targets for the development of new migraine or anti-epileptic drugs. Additionally, studies of these loci and how they modify ATPalpha dysfunction will help us understand epilepsy, hemiplegia and migraine disease pathogenesis in animals.

Drosophila strains
Flies were maintained on standard cornmeal-molasses agar medium at 21-22°C. Chromosomal deficiencies were obtained from the Bloomington Deficiency Kit from the Bloomington Stock Center (order date January 2010). The Df Kit we received contained 467 stocks with deletions spanning 97.8% of the Drosophila genome. Three Na + /K + ATPase alpha subunit mutants were used: ATPalpha DTS1 [23], ATPalpha CJ5 and ATPalpha CJ10 [36]. The other Drosophila strains used were obtained from the Vienna Drosophila RNAi Center (VDRC) or Bloomington Stock Center.
Locomotor assays F 1 offspring heterozygous for an ATPalpha allele and each individual Df were collected upon eclosion (day 0) and aged at 25°C on cornmeal-molasses medium. Temperature sensitivity (TS) was assayed on day 1 and bang sensitivity (BS) was assayed on day 15 as described previously [23]. Aged flies were moved to an empty vial in groups of 5 or fewer using an aspirator. For TS, the vial was submerged in a water bath at 38°C such that the flies were restricted to space in the vial below the waterline. A timer was started when the vial was submerged and time to paralysis was recorded for each fly. For BS, the vial was mechanically shaken using a standard lab Many genes had an interaction with more than one allele, although some appear to be allele specific. RNAi knockdowns were compared with ATPalpha*, daGal4/+ controls.  Vortex Genie 2 (Daigger, IL) on the highest setting for 20 seconds. Time to recovery for each fly was recorded. Both conditional locomotor assays were stopped after 300 seconds.
Each of the 467 Dfs we received was tested with at least one ATPalpha allele, the vast majority were tested with multiple alleles and >55% were tested with all three alleles. Assays were performed as described above.

Verification screen
Putative modifier Df strains identified in the initial screen were retested in an independent experiment to verify the findings and reduce the rate of false positives. In selecting Df stains to test again, we favored Dfs that suppressed ATPalpha mutant phenotypes and/or interacted with more than one ATPalpha allele.

Single gene identification
We developed an analysis called the Reproducibility Index (RI) in order to guide our search for single gene modifiers of the ATPalpha alleles. The goal of this index was to rank the most promising Df intervals based on the magnitude and reproducibility with which they modified an ATPalpha allele phenotype. To this end, we first calculated the number of standard deviations of the Df, ATPalpha* double mutant mean from the total mean of the primary screen of each ATPalpha mutant using: where StdDev total is the standard deviation of all deficiencies in the primary screen, Mean total is the mean of all deficiencies in the primary screen, and Mean Df is the mean response of a Df/ATPalpha double mutant. Num.Std.Dev (#SD) was calculated for the mean response of a Df double mutant in the primary (#SD prim ) and verification (#SD veri ) screen. We reasoned that these values provide a normalized metric of how much a Df modified an ATPalpha phenotype in each trial. We used these values to calculate the RI: where Absolute Varience AV ð Þ¼ #SD prim −#SD veri The RI increases for Dfs that were further from the total mean and decreases for Dfs that varied more across the two trials. Thus, a high RI suggests that a region is more likely to contain a gene that interacts with and modifies an ATPalpha allele in a reproducible manner.
In some intervals we were able to use small Dfs to narrow the interval further. We, again, prioritized strongly suppressing intervals over enhancing intervals and intervals that interacted with multiple alleles. Single genes were selected from critical intervals using the G-Browse feature (an annotated genome) of flybase.org. In some very small intervals all genes in the region were tested. In large intervals we necessarily focused on genes with described expression within the nervous and or muscular systems, introducing a noted bias. Many of the alleles chosen were P-element or classical mutations reported to knockout the genes of interest. The stocks of interest were ordered from the Bloomington Stock Center.

RNAi analysis
When classical mutants were unavailable for certain loci or to confirm an interaction found using a classical mutant, RNAi analysis was used to examine the gene in question. RNAi stocks were ordered from the VDRC. The RNAi transgenes were driven using daughterless Gal4 strains (daGal4) in each ATPalpha mutant background. RNAi male flies were mated to ATPalpha, daGal4 virgin females. Progeny were raised at 25°C, and TS and BS tests were performed as described previously.
reported in Table 5. The effect of RNAi transgenes was analyzed using a Student's t-test to determine if single gene knockdowns significantly modified the phenotype of ATPalpha*, daGal4 controls. Significant interactions are reported in Table 6.

Additional file
Additional file 1: Data from the primary and verification Df screens.