Autism spectrum disorder (ASD) has emerged as a major public health concern worldwide, with a prevalence of 1 in 68 children and a growing adult population with an urgent, unmet therapeutic need. Although autism is highly heritable, the identification of candidate genes has been complicated by extreme genetic heterogeneity, polygenic inheritance, occurrence of de novo mutations and rare variants that individually do not contribute to more than 1% of cases. Furthermore, most genes remain candidates because of a lack of functional validation. Among these candidates is the endosomal Na+/H+ exchanger NHE9 (SLC9A9), also named as autism susceptibility candidate 16 (AUTS16) (OMIM: 613410). Homozygosity mapping in autism pedigrees with shared ancestry identified a large deletion upstream from the 5’ end of NHE9. Further analysis of NHE9 sequence in non-consanguineous autism patients revealed a nonsense mutation and 6 rare, missense variants not found in controls. The nonsense change in NHE9 was phenotypically similar to an adjacent nonsense mutation in a related isoform, NHE1 that is causal to slow wave epilepsy in mouse. However, in the absence of functional evaluation, it was not possible to distinguish whether the remaining variants were benign polymorphisms or pathological mutations causal to autism phenotypes. Previously, loss of function was demonstrated in three of these variants (V176I, L236S and S438P); here, we evaluated the remaining three variants (P117T, D496N, and Q609K) reported in patients with autism and epilepsy. When expressed in HEK293 cells, all three were similar to wild type in endosomal expression and function. We discuss the implications of our findings on future diagnosis and treatment of autism patients.

Given the importance of functional evaluation of genetic variants to establish causal disease associations, we sought to evaluate three previously uncharacterized NHE9 missense substitutions that were identified in patients with autism and epilepsy.

NHE9 missense substitutions that were identified in patients with autism and epilepsy localize throughout the coding frame, including the membrane embedded NHE-homology domain and the C-terminal cytoplasmic tail (Fig. 1A). Of these, P117T localizes to an extracellular loop in NHE9 that corresponds to a region between transmembrane helices I and II in E. coli NhaA crystal structure. The other 2 variants, D496N and Q609K, are localized to the C-terminal tail. As a starting point, we calculated evolutionary conservation (Consurf) score for these three residues, using a scale ranging from 1 (highly variable) to 9 (invariant). Whereas Gln609 is poorly conserved with a Consurf score of 1, Pro117 and Asp496 are moderately conserved with score of 4 and 6, respectively (Suppl. Table 1). We then evaluated the physiochemical effects of these substitutions and predicted the functional consequences using several standard bioinformatic approaches (Suppl. Table 1). The P117T substitution exhibits a shift in polarity from non-polar to polar and displays an increase in Kyte-Doolittle hydrophobicity from -1.6 to -0.7. The D496N substitution exhibits a shift in polarity from negatively charged to polar and displays no change in hydrophobicity (-3.5). The Q609K variation exhibits a shift in polarity from polar to positively charged and displays a decrease in hydrophobicity from -3.5 to -3.9. The outcome predictions from bioinformatic efforts varied from being neutral to disease causing (Suppl. Table 1). To experimentally test these predictions on endosomal function, we expressed NHE9 variants with C-terminal DsRed tag in HEK293 cells and evaluated protein expression by western analysis, using an anti-DsRed antibody. NHE9-DsRed migrated as prominent high molecular weight smear (>200 kDa) that was resolved into discrete band(s) with dithiothreitol (DTT) treatment, suggesting oligomerization by disulfide bonds (Fig. 1B). Densitometric scan of immunoblot normalized to tubulin showed that the DsRed-tagged variants, P117T, D496N, and Q609K, were expressed at levels ~60–90% of wild-type (WT) NHE9, suggesting that autism variants might result in modest changes in expression (Fig. 1C). To compare subcellular localization of NHE9 variants, we determined DsRed fluorescence overlap with Alexa Flour 488 labelled transferrin (TFN), following 60 min incubation with live cells (Fig. 1D). NHE9 (WT and variants) showed characteristic distribution to vesicular compartments overlapping with TFN as previously described. Fractional colocalization determined using Manders’ coefficient revealed no significant differences (WT: 0.83±0.03, P117T: 0.80±0.04, D496N: 0.88±0.03, Q609K: 0.87±0.03), suggesting no deleterious effects of these autism-associated NHE9 variants on endosomal trafficking. We took advantage of the excellent overlap of NHE9 (WT and variants) with TFN to determine endosomal pH as a measure of Na+/H+ exchange activity. Endosomal pH was quantitatively measured by flow cytometry using pH-sensitive fluorescein (FITC) tagged TFN. Uptake of TFN into endosomes was normalized using pH-insensitive Alexa Fluor 633 tag and calibrated in four buffers with fixed-pH for a ratiometric determination of lumen pH (R2, square of the Pearson correlation coefficient=0.969; Fig. 1E). Consistent with proton-leak function of NHE9, we documented alkalization of endosomal pH from 6.10 ± 0.09 in vector-transformed control, to 7.05 ± 0.04 in cells expressing WT NHE9 (mean ± S.D.; n= 3 biological replicates; ***,P<0.001; two-tailed t test; Fig. 1F). To rule out nonspecific overexpression effects, we examined an autism with epilepsy associated S438P mutation within the membrane-embedded transporter domain of NHE9 (Fig. 1A), which was previously shown to result in loss of function. NHE9 carrying the S438P mutation failed to elevate endosomal pH in HEK293 cells, confirming loss of function as predicted using a structure-function approach (mean ± S.D.; n= 3 biological replicates; NS, not significant; P=0.155; two-tailed t test; Fig. 1F). We used this gain-of-function phenotype to score NHE9 variants. In contrast to previously characterized NHE9 autism variants, we observed significant alkalization of endosomes with expression of P117T, D496N, and Q609K substitutions, relative to empty vector transfection (mean ± S.D.; n= 3 biological replicates; **,P<0.01; ***,P<0.001; two-tailed t test; Fig. 1G). No significant difference in endosomal pH was noted in cells expressing either WT or the 3 NHE9 variants (Fig. 1G). Steady-state accumulation of TFN is dependent on endosomal NHE function: cells expressing NHE9 were shown to accumulate more TFN relative to vector-transformed control. We used flow cytometry of ∼10,000 cells per sample gated for NHE9 positivity in biological triplicates to quantify endocytosed TFN in NHE9 variants. All three variants significantly increased TFN accumulation relative to levels comparable to WT NHE9 (mean ± S.D.; n= 3 biological replicates; ***,P<0.001; two-tailed t test; Fig. 1H). Taken together, our findings indicate that all three autism-associated variants were not deleterious to ion transport function of NHE9.

Our findings may be interpreted in two ways. First, the variants could be benign polymorphisms and not causal to autism phenotype. This is especially likely for residues with low conservation scores, such as Q609. Second, the variants could be involved in regulatory functions not detectable in this cell culture model. This may be relevant for variants that map to the C-terminal tail as only the membrane-spanning helices are thought to be required for ion transport. All NHE isoforms have an extra-membranous C-terminal domain known to bind regulatory proteins. In silico prediction revealed the presence of disordered elements in NHE9 C-terminal tail, consistent with experimental observations in NHE1 (Fig. 1I). The D496N and Q609K variants lying partly within disordered regions might alter binding to the CHP or RACK1 proteins, previously shown to interact with NHE9 tail. These possibilities remain be investigated further.

As sequencing technology improves and costs decline, more rare variants will be identified that may be scored as pathogenic because they occur in autism patients. For instance, in another autism related gene, NHE6, a potentially harmless Ala9Ser variant was reported in a male with Angelman-like phenotype. However, his sister also presented with intellectual disability, but did not carry the Ala9Ser variant, strongly suggesting that the underlying neuropathology might be due to an unrelated brain defect or a mutation in a different autism gene. This is recently well appreciated in the field of cancer biology, wherein about a third of variants identified in clinically actionable cancer genes in patient tumors were false positives or passenger mutations. Functional evaluation of disease-associated NHE9 variants as we described here will therefore be essential to predict clinical outcome, as a prerequisite to personalized therapy in patients with autism and other neurological diseases.

3 autism-associated variants in NHE9, P117T, D496N, and Q609K, show no loss of function defects in a cell culture model and may well be benign polymorphisms, although we cannot rule out cell type specific, or subtle, regulatory function. Our findings highlight the need for functional analysis of all variants in autism-associated genes to distinguish between harmless polymorphisms and disease-associated mutations. We argue that performing genetic tests on autism patients without careful cellular studies of function could lead to inaccurate false positive diagnoses and poor treatment decisions resulting in unnecessary tests and unneeded therapies.

The functional analysis of NHE9 variants was conducted using a gain-of-function scoring system performed in cell culture. Although no functional deficits were observed for these variants in HEK293 cells, we cannot rule out cell type specific effects that may alter CNS function e.g., in neurons or astrocytes.

Although our data suggest that the variants P117T, D496N, Q609K tested here do not significantly affect transporter function of NHE9, it remains to be determined if these variants are involved in transporter regulation. It will be interesting to determine if NHE9 tail has a disordered structure as we predict here and whether autism variants in the C-terminal tail alter protein binding and regulation. For example, C-terminal NHE9 variants in a rodent model of ADHD appear to show increased binding to CHP, although functional consequences were not examined. Further characterization of these variants in a neurobiological model might also be warranted.

Cell culture

HEK293T cells were obtained from ATCC and grown in DMEM/high glucose/sodium pyruvate (Invitrogen) containing 10% FBS. Culture conditions were in a 5% CO2 incubator at 37°C. Cell viability was measured using the trypan blue exclusion method.

Plasmids and transfection

NHE9-DsRed was cloned into FUGW-lentiviral vector into the BamHI site. QuikChange Lightning site-directed mutagenesis kit (#21051, Agilent Technologies) was used to engineer P117T, D496N, Q609K, and S438P missense substitutions into NHE9. Resulting plasmids were confirmed by sequencing. HEK293T cells were transfected using Lipofectamine 2000 reagent, as per the manufacturer's instructions. Empty vector transfection was used as control. Transfection efficiency was determined by visualizing DsRed fluorescence.

Endosomal pH measurement

Endosomal pH was measured using flow cytometry, as previously described. Briefly, cells were rinsed and incubated in serum-free medium for 30 min, to remove residual transferrin and then incubated with pH-sensitive FITC-Transferrin (#T2871, Thermo Fisher Scientific) (75 μg/ml) together with pH non-sensitive Alexafluor 633-Transferrin (#T23362, Thermo Fisher Scientific) (25 μg/ml) at 37°C for 55 min. Transferrin uptake was stopped by placing the cells on ice. Excess transferrin was removed by washing with ice-cold serum-free DMEM and PBS, whereas bound transferrin was removed by washing with ice-cold pH 5.0 PBS and pH 7.0 PBS. Cells were trypsinized and pH was determined by flow cytometry analysis of ∼10,000 cells gated for NHE9 positivity in biological triplicates using the FACSAria instrument (BD Biosciences). A 4–point calibration curve with different pH values (4.5, 5.5, 6.5 and 7.5) was generated using Intracellular pH Calibration Buffer Kit (#P35379, Thermo Fisher Scientific) in the presence of 10 μM K+/H+ ionophore nigericin and 10 μM K+ ionophore valinomycin.

Transferrin uptake

Subcellular localization and functional evaluation of variants was performed using steady state transferrin uptake assay. Briefly, cells were rinsed and incubated in serum-free medium for 30 min, to remove residual transferrin and then incubated with Alexa Fluor 633 or 488-Transferrin (Thermo Fisher Scientific) (100 μg/ml) at 37°C for 60 min. Transferrin uptake stopped by placing the cells on ice. Excess transferrin was removed by washing with ice-cold serum-free DMEM and PBS, whereas bound transferrin was removed by 2× washing with ice-cold pH 5.0 PBS and pH 7.0 PBS.

Cells were trypsinized for flow cytometry transferrin uptake analysis of ∼10,000 cells gated for NHE9 positivity in biological triplicates using the FACSAria instrument (BD Biosciences). Alternatively, cells were fixed with a solution of 4% paraformaldehyde, for confocal imaging using the LSM 700 Confocal microscope (Zeiss). Expression of NHE9 (WT and variants) was detected using the DsRed fluorescence. Nuclei were stained with DAPI. Fractional colocalization was determined from Manders' coefficient using the JACoP ImageJ plugin that measures the direct overlap of pixels in the confocal section, and represented it as mean ± S.E.

Western blot

Cells were lysed using 1% Nonidet P-40 (Sigma) supplemented with protease inhibitor cocktail (Roche). Cells were sonicated and then centrifuged for 15 min at 18,000× g at 4°C. Protein concentration was determined using the BCA assay. Dithiothreitol (DTT) (NuPAGE) was used to reduce protein samples for protein gel electrophoresis. Equal amounts of protein were separated by polyacrylamide gel (NuPAGE Novex) and then electrophoretically transferred onto nitrocellulose membranes (Bio-Rad). Next, the membranes were treated with the blocking buffer containing 5% milk, followed by overnight incubation with primary Anti-DsRed antibody (#ab62341, Abcam, 1:1,000) and 1 h incubation with HRP-conjugated secondary antibodies (GE Healthcare, 1:1,000). SuperSignal West Pico substrate was used for detection. Fujifilm LAS 3000 imaging system was used to capture images. Tubulin (Sigma T9026, 1:5,000) or GAPDH (Sigma G9295, 1:10,000) was used as a loading control. Densitometry analyses was performed with ImageJ software.

Bioinformatics

Evolutionary conservation scores were calculated using the Bayesian method using the Consurf web server (http://consurf.tau.ac.il/) that analyses the evolutionary pattern of the amino acids to reveal functionally important regions. The physiochemical and functional consequences of NHE9 substitutions was predicted using tools from the Human Gene Mutation Database (http://www.hgmd.cf.ac.uk/). In silico intrinsic disorder with annotated protein-binding activity was predicted for the NHE9 C-terminal tail using the DISOPRED3 program (http://bioinf.cs.ucl.ac.uk/psipred/).

Statistical Analysis

Data were analyzed statistically by Student's t test using GraphPad Prism and presented as mean ± SD.

This work was supported by a grant from the National Institutes of Health R01 DK108304 to R.R. H.P. was a Fulbright Fellow supported by the International Fulbright Science and Technology Award.

All work was performed according to the Johns Hopkins University regulations regarding use of recombinant DNA and cell lines.