Functional correlates of clinical phenotype and severity in recurrent SCN2A variants

In SCN2A-related disorders, there is an urgent demand to establish efficient methods for determining the gain- (GoF) or loss-of-function (LoF) character of variants, to identify suitable candidates for precision therapies. Here we classify clinical phenotypes of 179 individuals with 38 recurrent SCN2A variants as early-infantile or later-onset epilepsy, or intellectual disability/autism spectrum disorder (ID/ASD) and assess the functional impact of 13 variants using dynamic action potential clamp (DAPC) and voltage clamp. Results show that 36/38 variants are associated with only one phenotypic group (30 early-infantile, 5 later-onset, 1 ID/ASD). Unexpectedly, we revealed major differences in outcome severity between individuals with the same variant for 40% of early-infantile variants studied. DAPC was superior to voltage clamp in predicting the impact of mutations on neuronal excitability and confirmed GoF produces early-infantile phenotypes and LoF later-onset phenotypes. For one early-infantile variant, the co-expression of the α1 and β2 subunits of the Nav1.2 channel was needed to unveil functional impact, confirming the prediction of 3D molecular modeling. Neither DAPC nor voltage clamp reliably predicted phenotypic severity of early-infantile variants. Genotype, phenotypic group and DAPC are accurate predictors of the biophysical impact of SCN2A variants, but other approaches are needed to predict severity.


Supplementary Figure 2.
Steady-state inactivation of Nav1.2 channel variants. Representative Na + current availability traces, recorded at −5 or −10 mV, immediately after 100-ms pre-pulses to voltages between −120 and 0 mV, applied every 10 s from a holding potential of −120 mV (inset voltage protocol). The membrane potential for half-maximal inactivation (V0.5,inact) values and their statistical evaluation are shown in Supplementary Table 3. For each variant, the current trace recorded after the 100-ms pre-pulse at −50 mV is highlighted in light grey (WT, wild-type) or black (mutants). Current traces of individual variants were normalized to the same amplitude; note inset time scale bar. The voltage dependence of inactivation of WT and mutant Nav1.2 channel variants is shown in Figure 2 (main text). Relative to WT channels, the V261L, Q383E, E1321K, and Q1531K variants recovered faster, corresponding to gain-of function (GoF), whereas E999K, K905N, E1211K, and D195G showed slower recoveries, consistent with loss-of-function (LoF); the recoveries of R1629L, R856Q, A263V, V1325I, and R1329Q were unchanged. Data shown are mean ± SEM; * P < 0.05 (one-way ANOVA, followed by Dunnett's post-hoc test); NS, statistically not significantly different relative to WT; see the P values in Supplementary Table 3. Figure 6. Dynamic action potential clamp (DAPC) experiments implementing external Nav1.2 currents from Chinese hamster ovary (CHO) cells expressing wild-type (WT), early-infantile (EI) severe/variable/benign, or later-onset (LO) Nav1.2 variants. In the axon initial segment compartment model, the Nav1.6 conductance (gNav1.6) was set to zero, whereas the potassium channel conductance (gKv) was set to 100% (gNav1.6 = 0, gKv = 1, respectively). See also the effect of a gKv setting of 2 in Figure 3 (main text). a Representative examples of action potential firing in response to 4 and 10 pA step current stimuli. b Input−output relationships. Note the overall lower AP firing frequencies compared to data shown in Figure 3. Data shown are mean ± SEM; Two-way ANOVA, followed by Dunnett's post-hoc test, was used to compare the action potential firing frequencies elicited by step stimuli in the presence of Nav1.2 variants (WT, black solid circle; mutants, colored open circles); asterisks indicate P < 0.05; n values, the number of independent experiments, are shown in parentheses. Figure 7. 3D structures of the functionally studied Nav1.2 channel variants, shown as cartoon (PDB accession no. 6J8E) 1 . The variants are localized in channel domains or segments associated with gating or ion permeation (see details in the main text) 2 . a Side and intracellular views of the wild-type channel highlighting the residues affected by missense mutations (red sticks). The four domains, DI-IV, are colour-coded; modeling was performed in PyMOL (Schrödinger LLC, New York, USA). b Zoomed-in views of structures in Nav1.2, highlighting residues before (left) and after (right) in silico mutagenesis. All residues within 5Å distance from the mutated residue are shown in stick representation (blue: nitrogen, red: oxygen). D195G (associated with later-onset (LO) phenotype): negatively charged Asp residue mutated to uncharged Gly residue in segment 5 of domain I (S5DI); the polar interaction between D195 and N192, shown in wild-type channel, cannot be formed in the mutant variant. V261L (earlyinfantile (EI)-variable): hydrophobic Val residue mutated to hydrophobic Leu residue in S5DI; the longer side chain of Leu may affect the hydrophobic interactions within DI. A263V (EI-variable): hydrophobic Ala residue mutated to hydrophobic Val residue in S5DI; it is likely that this mutation affects hydrophobic side chain interactions within DI. Q383E (EI-variable): polar uncharged Gln residue mutated to negative Glu residue in P1-P2 of DI; Gln383 is adjacent to the key Glu384 residue in the DEKA selectivity filter. Within the DEKA motif, the positively charged Lys and the carboxylate from Glu are responsible for maintaining an ionic permeability ratio of 0.03:0.075 for K + over Na + 3 . Monte Carlo simulations suggest that the selectivity filter of the wild-type Nav1.2 channel has an electrostatic balance of 2 positive (K1422 and the Na + ion) and 2 negative (D384 and E942) 4 . In the wild-type channel, Q383 contributes to stabilizing the selectivity filter by donating a H-bond to the backbone carbonyl of L380 (and/or R379). It has been suggested that the Q383E channel may be Ca 2+ permeable because E383 provides an additional negative charge to the DEKA motif (3 negative: E942, D384, and E383), which can be balanced with 3 positive charges carried by K1422 and Ca 2+ 4 . However, this hypothesis has not yet been tested experimentally. R856Q (EI-severe): Positive Arg mutated to Gln (with polar uncharged side chain) neutralizing mutation in S4DII. In the wild-type channel, the gating charge-carrying Arg residues in segment 4 (S4) and several negative residues in S1 and S2 of the voltage sensor domain are involved in channel state-dependent interactions, resulting in a network of ionic and hydrogen-bonding interactions 5 . In the R856Q variant, some of the above interactions, including the polar contact between positions R856 and R853, and the cation- interaction 6 between the sidechain of R856 and the aromatic sidechain of F802 are disrupted. E999K (EI-severe): Negative Glu residue mutated to positive Lys in the intracellular DII-DIII linker. The structure of DII-DIII linker is currently unresolved. Sequence alignment of the proximal DII-DIII linker of Nav1.2, Nav1.3, Nav1.1, and Nav1.6 channels, performed with CLC Sequence Viewer 7.7 (QIAGEN Aarhus, Denmark). The residue colours correspond to polarity. The green-shaded boxed area highlights the intracellular terminal of S6DII. Note that the E999 residue of Nav1.2 is conserved across the brain sodium channels. Various physiological roles have been attributed to the DII-DIII linker, including the regulation of clustering of sodium channels at the axonal initial segment 7 , ankyrin binding 8 , and gating 9 . E1211K (LO): Negative Glu residue mutated to positive Lys in S1DIII; E1211 is highly conserved in across voltagegated sodium and calcium channels. E1211 forms polar contacts with V1215 and I1214. In the mutant channel, additional interactions may be formed between K1211 and E1206, and/or K1211 and N1208. R1319Q (EI-variable): Positive Arg residue mutated to polar uncharged Gln residue in S4DIII. This mutation is likely to affect the movement of the voltage sensor 10 . E1321K (EI-benign): Negative Glu residue mutated to positive Lys in the intracellular S4-S5 linker in DIII (S4-5DIII). Mutations in the in the S4-S5 linker are likely to disrupt the various coupling interactions between the voltage sensor and the pore 11 . V1325I (EIvariable): hydrophobic Val residue mutated to hydrophobic Ile residue in S4-5DIII. See the role of S4-S5 linker above (E1321K). Note the proximity of residue F1489, a key IFM hydrophobic sequence motif involved in fast inactivation 2 . Q1531K (EI-benign): Glu residue with polar uncharged side chain mutated to positive Lys in S1DIV. Q1531 of Nav1.2 is a conserved residue across human sodium channels. R1629L (EI-severe): Positive Arg residue mutated to hydrophobic Leu residue in S4DIV. In the wild-type channel, the R1 to R4 gating charges reside above the hydrophobic constriction site, when VSD4 is in an activated conformation. R1629 forms polar interactions with Q268 (S5DI) and E1551 (S4DIV), and cation- interaction with the aromatic sidechain of F1625 (S1DIV) 6 . These interactions cannot be formed in the R1629L variant. Dashed lines indicate polar interactions with a number denoting the distance between atoms in Å. Two individuals with mildly delayed developmental outcome had seizure onset at ages 3 and 6 months, and one also had episodic ataxia. These individuals had Q383E and D343 variants.

Supplementary
Seizure onset between age 3-24 months seen in at least one individual with 10/30 variants associated with early-infantile phenotype. 8/10 also seen in at least one individual with seizure onset <3 months (including D343G and Q383E) and 2/10 (V208E, Y1589C) with at least one individual with seizure onset at age 3 months.  Statistically significant differences between the wild-type (WT) and mutant channels were determined using one-way ANOVA, followed by Dunnett's post-hoc test ( * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001); NS, statistically not significant difference compared to WT. Data are represented as mean  SEM. Abbreviations: INa, sodium current; INa-P, persistent sodium current; n, number of independent experiments; f, fast time constant (of the inactivation);  recovery, time constant of recovery from fast inactivation; V0.5,act, membrane potential for half-maximal activation;V0.5,inact, membrane potential for halfmaximal inactivation; VC, voltage clamp. Early-infantile Nav1.2 variants were allocated into benign, variable, and severe phenotypic groups. The clinico-electrophysiological severity score of Nav1.2 variants (CESSNa + score) was calculated as described according to Lauxmann et al 12 (see Methods). Briefly, the maximum effect of all 13 variants was determined relative to control for each assessed biophysical property, and divided in three thirds corresponding to large, medium, and small changes, associated with high (3), medium (2) and low (1) scores. Additional points were scored for the maximal (strongest) change of a selected biophysical property (e.g., persistent sodium current and recovery from fast inactivation) and if multiple (at least three) biophysical properties were affected by a mutation for a given variant. Opposing effect, such as the shifts of the V0.5,act and V0.5,inact in the same direction were considered as 'neutralising'. A high CESSNa + score was correlated with pronounced severity 12 . Two sets of CESSNa+ scores were generated, by considering or omitting the neutralising effect. . Mean CESSNa + scores in the benign, variable, and severe phenotypic groups were compared using one-way ANOVA and the P value determined. Mean CESSNa + scores in the early-infantile-severe, early-infantile-variable, and early-infantile-benign groups were statistically not significantly different (P = 0.46; one-way ANOVA). These results suggest that, for this relatively small dataset, the CESSNa + score cannot differentiate earlyinfantile-severe or early-infantile-variable groups from early-infantile-benign. CESSNa + scores should be regarded as an indication of variant severity rather than an unequivocal prediction of variant severity 12 .

Early-infantile variant
Statistically significant differences between the wild-type (WT) and mutant channels were determined using two-way ANOVA, followed by Dunnett's post-hoc test ( * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001). Data are mean SEM. Abbreviations: n, number of experiments; NF, no firing; NS, statistically not significant difference compared to WT.  Table 6. Action potential characteristics of the axon initial segment neuronal model incorporating wild-type or mutant Nav1.2 channels in dynamic action potential clamp experiments.
The first action potential elicited by a current step 2 pA above rheobase was analysed. Firing was elicited by depolarizing step current stimuli in 2-pA increments. In the axon initial segment model, the virtual Nav1.6 channel conductance (gNav1.6) and the virtual potassium channel conductance (gKv) values were set to gNav1.6 = 0 and gKv = 2. Statistically significant differences between the action potential characteristics of the wild-type (WT) and mutant channels were determined using one-way ANOVA followed by Dunnett's post-hoc test ( * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001). Data are mean SEM. Abbreviations: n, number of independent experiments; NS, statistically not significant difference compared to WT.  Table 7. Action potential firing activity during synaptic current stimulation in dynamic action potential clamp experiments implementing Nav1.2 variant current and virtual conductance settings of gNav1.6 = 0/gKv = 1. Statistically significant differences between the wild-type (WT) and mutant channels were determined using two-way ANOVA, followed by Dunnett's post-hoc test ( * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001). # P < 0.05 at ge/gi  4 (see Fig. 4b). Data are mean SEM. Abbreviations: n, number of independent experiments; NF, no firing; NS, statistically not significant difference compared to WT. Abbreviations: VC, voltage clamp; DAPC, dynamic action potential clamp; GoF, gain-of-function; LoF, loss-of-function; *, missense variants characterized in our previous study 13 ;  , persistent current is present at membrane voltages more positive than −30 mV; §, machine learning-based statistical model, trained on selected protein features of published LoF and GoF variants 14 ;  , unreliable functional prediction, because variant is part of the training data 14 ; N/A, not applicable; NT, not tested; # Biophysical characteristics and DAPC prediction relative to control variant (wild-type α1 subunit) co-expressed with β2 subunit.

Effects of scaled excitatory to inhibitory conductance ratios (ge/gi) on firing frequency (Hz
Supplementary Table 9. The effect of β2 subunit co-expression on the biophysical characteristics of the wild-type and K905N Nav1.2 channel variants in voltage clamp experiments. WT*, CHO cells transfected with wild-type α1 subunit alone; WT + β2, CHO cells transfected with wildtype α1 and β2 subunits; K905N + β2, CHO cells transfected with K905N α1 and β2 subunits. Statistically significant differences between WT + β2 and K905N + β2 or WT* channels were determined using one-way ANOVA, followed by Dunnett's post-hoc test ( * P < 0.05, ** P < 0.01, and **** P < 0.0001); NS, statistically not significant difference compared to WT + β2. Data are represented as mean SEM. Abbreviations: n, number of independent experiments; INa, sodium current; V0.5,act, membrane potential for half-maximal activation; V0.5,inact, membrane potential for half-maximal inactivation;  recovery, time constant of recovery from fast inactivation.
Dynamic action potential clamp experiments were performed with virtual conductance settings of gNav1.6 = 0/gKv = 2. The input-output relationships of these experiments are shown in Figure 5f (main manuscript). Differences in firing rates at each step current stimulus were determined relative to WT + β2 using multiple comparisons in two-way ANOVA, followed by Dunnett's post-hoc test ( * P < 0.05, ** P < 0.01, and, **** P < 0.0001); NS, statistically not significant difference compared to WT + β2. Data are represented as mean SEM; n, number of experiments.  Table 11. The effect of β2 subunit co-expression on action potential characteristics in dynamic action potential clamp experiments implementing wild-type or K905N Nav1.2 channel variants.
Dynamic action potential clamp experiments were performed with virtual conductance settings of gNav1.6 = 0/gKv = 2. Statistically significant differences between the WT* or K905N + β2 channels and WT + β2 were determined using one-way ANOVA, followed by Dunnett's post-hoc test ( * P < 0.05); NS, statistically not significant difference compared to WT + β2. Data are represented as mean  SEM. Abbreviations: n, number of independent experiments; WT*, CHO cells transfected with wild-type α1 subunit alone; WT + β2, CHO cells transfected with wild-type α1 and β2 subunits; K905N + β2, CHO cells transfected with K905N α1 and β2 subunits.