Seizures, behavioral deficits, and adverse drug responses in two new genetic mouse models of HCN1 epileptic encephalopathy

De novo mutations in voltage- and ligand-gated channels have been associated with an increasing number of cases of developmental and epileptic encephalopathies, which often fail to respond to classic antiseizure medications. Here, we examine two knock-in mouse models replicating de novo sequence variations in the human HCN1 voltage-gated channel gene, p.G391D and p.M153I (Hcn1G380D/+ and Hcn1M142I/+ in mouse), associated with severe drug-resistant neonatal- and childhood-onset epilepsy, respectively. Heterozygous mice from both lines displayed spontaneous generalized tonic–clonic seizures. Animals replicating the p.G391D variant had an overall more severe phenotype, with pronounced alterations in the levels and distribution of HCN1 protein, including disrupted targeting to the axon terminals of basket cell interneurons. In line with clinical reports from patients with pathogenic HCN1 sequence variations, administration of the antiepileptic Na+ channel antagonists lamotrigine and phenytoin resulted in the paradoxical induction of seizures in both mouse lines, consistent with an impairment in inhibitory neuron function. We also show that these variants can render HCN1 channels unresponsive to classic antagonists, indicating the need to screen mutated channels to identify novel compounds with diverse mechanism of action. Our results underscore the necessity of tailoring effective therapies for specific channel gene variants, and how strongly validated animal models may provide an invaluable tool toward reaching this objective.


Introduction
Developmental and epileptic encephalopathies (DEE) are a devastating group of diseases, often with poor response to pharmacological treatment, resulting in a lifelong burden of seizures, developmental delay, and intellectual disability. Since the original discovery over twenty years ago that de novo mutations in voltage-gated ion channels can directly cause early childhood epilepsy syndromes (Singh et al., 1998), the number of genes and gene variants associated with DEE has grown exponentially. Current estimates indicate that ~30% of DEE patients carry at least one pathogenic variant in the top 100 known gene candidates for the disease, with about a third of the affected genes falling into the category of voltage-and ligand-gated ion channels (Noebels, 2017;Oyrer et al., 2018;Wang and Frankel, 2021). Despite the wealth of genetic data available and recent efforts to model the effects of the mutations in genetically modified mice, we understand very little about the mechanisms that underlie the brain and neuronal circuit alterations responsible for seizures, or the reasons for their drug resistance.
Here we focus on seizures and the pharmacological response profile associated with mutations in the HCN1 gene, which encodes a hyperpolarization-activated cyclic nucleotide-regulated non-selective cation channel expressed prominently in brain (Santoro and Shah, 2020). To date a total of 40 different missense variants in the HCN1 gene have been reported in patients affected by epilepsy and/or neurodevelopmental disorders (source: HGMD Professional 2020.4). Among these, at least 12 different de novo HCN1 variants have been linked to DEE (Nava et al., 2014;Butler et al., 2017;Marini et al., 2018;Wang et al., 2019).
The functional role of HCN1 channels is tied closely to their distinct subcellular localization in the two classes of neurons where HCN1 protein is primarily expressed, pyramidal neurons in cortex and hippocampus, and parvalbumin positive (PV+) interneurons across the brain (Notomi and Shigemoto, 2004). Within these neuronal classes, the subcellular localization of HCN1 channel subunits is tightly regulated. In excitatory (pyramidal) neurons, the channel is targeted to the distal portion of the apical dendrites, where it constrains the dendritic integration of excitatory postsynaptic potentials, limiting excitability (Tsay et al., 2007;George et al., 2009). In inhibitory (PV+) neurons, HCN1 channels are localized exclusively to axonal terminals, where they facilitate rapid action potential propagation (Roth and Hu, 2020) and GABA release (Southan et al., 2000;Aponte et al., 2006).
In this study we examine two new genetic mouse models that reproduce de novo HCN1 variants previously shown to be associated with severe forms of neonatal-and childhood-onset epilepsy (Parrini et al., 2017;Marini et al., 2018;Fernández-Marmiesse et al., 2019). A prior attempt at modeling HCN1-linked DEE in mice (Bleakley et al., 2021) resulted in a relatively mild phenotype, wherein spontaneous seizures could not be systematically documented, despite the presence of epileptiform spikes on electrocorticography recordings (ECoG). This prompts the question whether there may be some general limitation to modeling HCN1-linked DEE in mice. To examine this question and probe further whether and how HCN1 mutations may give rise to DEE, we generated mouse models for two other HCN1 variants linked to DEE in humans.
We examined variants c.1172G>A/p.Gly391Asp and c.459G>C/p.Met153Ile based on three criteria: severity of disease; occurrence in at least two independent patients with similar epilepsy phenotypes; and differing biophysical effects on channel function when tested in heterologous expression systems (Marini et al., 2018). Moreover, clinical reports suggest that seizures are poorly controlled with standard antiseizure medications, and that certain anticonvulsants in fact exacerbate seizures in these patients (Marini et al., 2018). Our results show that both mutations lead to spontaneous seizures in mice, reproduce the paradoxical pharmacological responses seen in patients, and provide new insights into the mechanistic basis for the limitations of current drug therapies.

Results
Pathological mutations orthologous to human HCN1 variants p.G391D and p.M153I were generated using the CRISPR/Cas9 approach in the Hcn1 gene of inbred C57BL/6J mice (see Materials and Methods;Tröder et al., 2018). The resulting lines, Hcn1 G380D/+ and Hcn1 M142I/+ , were maintained using a male heterozygote x female wildtype (WT) breeding scheme and heterozygous mutant animals were compared to WT littermate controls.

General phenotypic features and gross brain anatomy
No overt differences were noted in pup development, except for consistently smaller body weights at weaning in Hcn1 G380D/+ mice of both sexes ( Figure 1A). Smaller body weights persisted for the life of Hcn1 G380D/+ animals, and were accompanied by an ~12% reduction in adult brain weight with gross brain anatomy otherwise normal ( Figure   1B). Brain area measurements showed a similar reduction. Thus, both cerebellum and brainstem areas were significantly smaller in Hcn1 G380D/+ mice compared to WT ( Figure   1 -Figure supplement 1) with the cerebellum more affected than the brainstem, consistent with the higher expression of HCN1 protein in this region (Notomi and Shigemoto, 2004). Reduced mean body weight was also observed in Hcn1 M142I/+ males, but no decrease in brain size was noted in either sex for this line ( Figure 1A,B).
Survival curves revealed a significant number of premature deaths among mutant animals in both lines ( Figure 1C). In many instances, animals were found dead in the cage without having shown prior signs of distress, suggesting sudden death. The survival curves revealed some notable differences in the pattern of death observed between the two lines, both with regards to timing and sex specificity. Thus, while most deaths among Hcn1 G380D/+ animals occurred before the age of three months, with no difference between the sexes ( Figure 1C, left), there was a clear split in Hcn1 M142I/+ animals with females more affected than males and most deaths occurring at an age older than three months ( Figure 1C, right). These findings suggest important differences in the effects of the two variants on brain function and physiology, perhaps reflective of the divergent effects of the mutations on channel biophysical properties (Marini et al., 2018) and HCN1 protein expression (see below).
Given the high expression of HCN1 subunits in cerebellum, both in Purkinje neurons and basket cell axon terminals (Southan et al., 2000;Luján et al., 2005;Rinaldi et al., 2013), as well as the motor phenotype observed in HCN1 global knockout animals (Nolan et al., 2003;Massella et al., 2009) In summary, these results reveal a complex effect of the two variants on phenotype, with an overall more severe impact on the general fitness of Hcn1 G380D/+ compared to Hcn1 M142I/+ animals, reflected in smaller body size, reduced brain size, as well as altered gait and locomotion.

Hcn1 G380D/+ and Hcn1 M142I/+ mice show spontaneous convulsive seizures
The sudden death phenotype observed in the two lines, along with the epilepsy syndromes seen in G391D and M153I HCN1 patients, suggests the animals may be undergoing generalized tonic-clonic seizures (GTCS), which in mice often result in death when escalating into tonic hindlimb extension. One such death event was indeed observed in a female from the Hcn1 M142I/+ line, and several convulsive seizure events were witnessed in mice from both lines during routine cage inspections. We therefore set out to rigorously quantify the occurrence, frequency and severity of seizures in Hcn1 G380D/+ and Hcn1 M142I/+ mice by using chronic video-ECoG recordings for periods of 1-2 weeks at a time ( Figure 3). Adult mice from both lines and sexes were found to display spontaneous GTCS's with distinct electrographic signature on ECoG traces ( Figure 3A,C) and behavioral manifestation on video recordings (Videos 1 and 2).
We recorded from a total of n = 8 Hcn1 G380D/+ animals (2M, 6F) 56-91 days old to avoid selecting against animals with early death; and a total of n = 12 Hcn1 M142I/+ animals (6M, 6F) 80-130 days old to capture the period during which sudden death is most frequently seen in females. Notably, despite the higher occurrence of death in females, seizures were observed with similar frequency in males and females from the Hcn1 M142I/+ line ( Figure 3D). Seizure events were comparatively infrequent, ranging from 0.6 ˗ 4.1 seizures per week in Hcn1 G380D/+ mice (with 2/8 individuals failing to show seizures during the time recorded) and from 0.6 ˗ 5.1 seizures per week in Hcn1 M142I/+ mice (with 2/12 individuals displaying no seizures during the recorded period). Despite the overall more severe phenotype of Hcn1 G380D/+ animals, based on body weight, brain size and general appearance, when rated on a modified Racine scale (see Materials and Methods) we found that animals from the Hcn1 G380D/+ line had on average lower maximum grade seizures compared to Hcn1 M142I/+ animals ( Figure 3B,D).
As GTCS in mice usually reflect involvement of the limbic system, including the hippocampus, we sought to additionally characterize the epilepsy phenotype of Hcn1 G380D/+ and Hcn1 M142I/+ mice by performing morphological analysis and immunohistochemical labeling for known hippocampal markers of epilepsy. In line with our video-ECoG findings, we observed upregulation of neuropeptide Y in dentate gyrus mossy fibers, alterations in granule cell layer morphology and extracellular matrix deposition in the dentate gyrus, as well as increased hippocampal gliosis in a majority of mutant animals from both lines (Figure 3 -figure supplement 1). Such findings are consistent with the presence of spontaneous seizures and altered excitability within the hippocampal circuit (Houser, 1990;Marksteiner et al., 1990;Stringer, 1996;Magagna-Poveda et al., 2017). While the site of seizure onset may differ between species, with a more prevalent role of neocortex in human compared to mouse, these results confirm that Hcn1 G380D/+ and Hcn1 M142I/+ mice are an overall appropriate model for human HCN1associated DEE, and that the two mutations are causal to the epilepsy phenotype.

Effects of mutations on HCN1 expression and the intrinsic properties of hippocampal CA1 pyramidal neurons
Prior in vitro studies of the human G391D and M153I variants using heterologous expression systems have demonstrated the functional impact of these mutations on HCN1 channels (Marini et al., 2018). HCN1 subunits with the G391D mutation fail to express current on their own, but when mixed with WT subunits (as occurs in patients and heterozygote Hcn1 G380D/+ mice) yield, in the ~20% fraction of cells that show measurable currents, a "leaky" channel with significantly decreased current density but a greatly increased voltage-and time-independent inward, depolarizing current component. In contrast, the M153I mutation only slightly reduces current density, but accelerates the opening kinetics and shifts the voltage dependence of the HCN1 channel to more depolarized potentials (midpoint of activation is shifted by +36 mV for homomeric mutant channels, and by +12 mV for heteromeric M153I/WT channels). However, it is not known how these mutations affect HCN1 expression and neuronal physiology in a native environment, which includes both the HCN channel auxiliary subunit TRIP8b and other channel modulators. To address this question, we performed immunohistochemical labeling of HCN1 protein and patch-clamp recordings from CA1 pyramidal neurons in Hcn1 G380D/+ and Hcn1 M142I/+ mice and compared their properties with WT littermate controls.
HCN1 antibody labeling of the hippocampus in WT animals showed the typical distribution of HCN1 subunits in CA1 pyramidal neurons ( Figure 4A), where the channel is present in a gradient of increasing expression along the somatodendritic axis of the apical dendrite (Lörincz et al., 2002). This gradient of expression was preserved in brains from both mutant lines, although labeling revealed a substantial decrease in overall HCN1 protein levels in Hcn1 G380D/+ brains, with a smaller but still significant decrease observed in Hcn1 M142I/+ brains ( Figure 4B). In whole-cell patch-clamp recordings, such results were matched by a substantial reduction in the voltage sag observed in response to hyperpolarizing current injections, a hallmark of I h activity in neurons ( Figure 4C,D). The voltage sag was absent in Hcn1 G380D/+ CA1 pyramidal neurons and significantly reduced in Hcn1 M142I/+ . The lack of voltage sag in Hcn1 G380D/+ neurons could be due to decreased HCN1 protein expression, a dominant negative effect of HCN1-G380D subunits on channel function and/or the increased instantaneous component of HCN channels containing HCN1-G380D subunits, which would cause the channel to conduct an inward Na + "leak" current with no voltage dependence (Marini et al., 2018).
As wild-type HCN1 channels are partially open at the resting membrane potential (RMP) of most cells, including CA1 pyramidal neurons, they exert a well-characterized action to shift the RMP to more positive potentials and decrease the input resistance.
Surprisingly, despite the loss of HCN1 protein expression, CA1 pyramidal neurons of Hcn1 G380D/+ mice displayed a significant depolarization in the RMP; in contrast, there was no change in RMP in Hcn1 M142I/+ neurons ( Figure 4E). The more positive RMP of Hcn1 G380D/+ mice may reflect the depolarizing action of the increased voltageindependent "leak" current seen with these channels and/or secondary changes in the function of other neuronal conductances (Bleakley et al., 2021). Input resistance was unaltered in both mutant lines (G380D: WT = 137.8 ± 7.1 MΩ, n = 20 cells, vs Hcn1 G380D/+ = 153.7 ± 7.8 MΩ, n = 21; M142I: WT = 128.4 ± 9.5 MΩ, n = 25, vs Hcn1 M142I/+ = 125.5 ± 7.9 MΩ, n = 38). For Hcn1 G380D/+ mice this may again reflect the offsetting actions of increased "leak" current and decreased channel expression, or alternatively, offsetting effects due to alterations in the expression of other types of channels.
The possibility that the expression of mutant HCN1 channels may cause secondary changes in CA1 pyramidal neuron function is consistent with our finding that CA1 cells in both mouse lines show an impaired ability to fire action potentials, a phenotype not observed with acute application of the HCN channel specific blocker ZD7288 (Gasparini and DiFrancesco, 1997;and  Hcn1 G380D/+ vs 37% Hcn1 M142I/+ vs 2% in the combined set of WT littermate neurons). In contrast, rheobase was unaltered in both lines, when measured from a holding potential of -70 mV (G380D: WT = 100 ± 9.6 pA, n = 20, vs Hcn1 G380D/+ = 114.3 ± 8.0 pA, n = 21; M142I: WT = 101.0 ± 6.8 pA, n = 25 vs Hcn1 M142I/+ = 98.6 ± 6.4 pA, n = 38). No differences were observed between males and females for any of the measures assessed in either mouse line, hence all reported data were pooled across sexes.
Overall, the observed changes reflect subtle alterations in the intrinsic properties of pyramidal neurons, which do not immediately explain the hyperexcitability and epilepsy phenotype observed in either mutant mouse line. The reduced ability of Hcn1 G380D/+ pyramidal neurons to maintain sustained firing may, in fact, act to limit circuit hyperexcitability in these animals -as reflected in the lower average maximum seizure grade reached by Hcn1 G380D/+ mutants both during spontaneous and drug-induced seizure events ( Figure 3B,D; and see Figure 7C below), compared to Hcn1 M142I/+ animals.
These results suggest that altered activity in non-pyramidal neurons may also play a role in the epilepsy phenotype of the mutants.
Severe impairment in the axonal localization of HCN1 protein in Hcn1 G380D/+ PV+ interneurons As HCN1 subunits are highly expressed in PV+ interneurons in the brain, in addition to their expression in pyramidal neurons, we examined whether this expression is altered by the seizure-causing mutations. Whereas the highest site of expression of HCN1 subunits is at the distal portion of apical dendrites of pyramidal neurons, as detailed above, HCN1 is most strongly enriched in the axonal terminals of PV+ interneurons. The exquisite targeting of HCN1 protein to axonal terminals in PV+ interneurons can be particularly appreciated in basket cells from the cerebellar cortex ( Figure 5). Here, HCN1 protein accumulates in a specialized structure, named the "pinceau", representing the axonal termination of basket cells onto the axon initial segment of Purkinje neurons. Surprisingly, at this location, we found a complete loss of HCN1 protein staining in Hcn1 G380D/+ animals, while the distribution in Hcn1 M142I/+ appeared relatively normal ( Figure 5, top row). This result is unexpected, as levels of HCN1 protein expressed from the WT allele should be normal, providing at least a half dosage in Hcn1 G380D/+ cells, and suggests that the mutation may exert a dominant negative effect by coassembling with WT HCN1 subunits, resulting in improper trafficking and/or premature degradation.
Accumulation of misfolded protein in the endoplasmic reticulum or Golgi apparatus may also impair the trafficking of proteins other than HCN1. As several other proteins and channels are known to specifically localize to the pinceau, we used labeling for one such channel (i.e., K v 1.2) to assess whether the normal pinceau channel architecture is disrupted in Hcn1 G380D/+ brains. As shown in Figure 5 (bottom row), K v 1.2 labeling of the pinceau was also completely lost, similar to HCN1 protein labeling, in Hcn1 G380D/+ but not Hcn1 M142I/+ brains. The downregulation of K v 1.2 does not appear to be due to co-regulation with HCN1, as normal K v 1.2 labeling is still present in the cerebellum of HCN1 knockout mice (Figure 5 -figure supplement 1) and the expression of the two channels at the cerebellar pinceau was shown to be independently regulated (Kole et al., 2015). As HCN1 channels in PV+ axonal terminals are thought to functionally interact with the Na + /K + pump (Roth and Hu, 2020), we further sought to test for the presence of ATP1a3 expression at the pinceau. As shown in Figure 5 (middle row), similar to HCN1 and K v 1.2, there was a complete loss of ATP1a3 labeling in Hcn1 G380D/+ but not Hcn1 M142I/+ animals, compared to WT controls.
The profound disruption of normal pinceau architecture in cerebellar basket cells suggests that, in the case of HCN1-G380D, misfolding of the mutated protein may be causing cellular stress (presumably endoplasmic reticulum stress, due to impaired trafficking and/or increased protein degradation). Such toxicity effects may explain the higher severity of the Hcn1 G380D/+ phenotype, which includes altered anatomy and behaviors beyond the occurrence of seizures and epilepsy.
Similar to what was observed in the apical dendrites of CA1 pyramidal neurons ( Figure 4A,B), immunolabeling of HCN1 protein in the axon terminals of hippocampal PV+ interneurons was strongly reduced in Hcn1 G380D/+ animals, although not completely absent ( Figure 6A). Since endogenous protein labeling is unable to distinguish between WT and mutant subunits, we sought to further probe for the trafficking of mutant HCN1 subunits in hippocampal PV+ interneurons using a viral transduction approach. To this end, we used mice that express Cre recombinase specifically in PV+ interneurons (Pvalb-Cre line) and stereotaxically injected into the hippocampal CA1 area an adeno-associated virus (AAV) driving Cre-dependent expression of hemagglutinin (HA)-tagged HCN1 subunits (AAV-DIO-HA-HCN1). As shown in Figure 6B Figure 6B). These results further underscore the adverse effect of the latter variant on the biosynthesis of the HCN1 protein, and suggest that disruption of PV+ interneuron function, due to HCN1 protein misfolding or altered trafficking, may represent an important component of neuronal circuit dysfunction in HCN1-linked epilepsy.

Paradoxical response of Hcn1 G380D/+ and Hcn1 M142I/+ mice to antiseizure medications
Clinical reports from at least four different patients with HCN1-linked DEE (carrying mutations M305L, G391D, and I380F; Marini et al., 2018;Bleakley et al., 2021;C. Marini, personal communication, March 2021) consistently revealed worsening of seizures after administration of either lamotrigine or phenytoin. A particularly striking example is provided by the G391D mutation, identified in two unrelated patients with similar course of disease. Both G391D patients had neonatal seizure onset, characterized by daily asymmetric tonic seizures with apnea and cyanosis, severe developmental delay, and died between 14-15 months of age. Both patients also showed a paradoxical response to phenytoin, which caused the induction of status epilepticus. Given this precedent, we hypothesized there may be systematic similarities between the adverse pharmacological response profile of HCN1 patients and Hcn1 G380D/+ and Hcn1 M142I/+ mutant mice, which would further endorse their use as models with construct, face and potentially predictive validity.
Consistent with our hypothesis, we found that administration of lamotrigine (23 mg/kg, i.p.) induced convulsive seizures, defined as grade 3 or higher on a modified Racine scale (see Materials and Methods), within 90 min from injection in 11/15 HCN1 G380D/+ animals ( Figure 7A,C, left). Several animals had multiple seizure bouts, ranging from rearing with forelimb clonus to wild running and jumping seizures, during the period of observation. Video-ECoG recordings were performed on a subset of animals (2/15 mutant and 2 WT control animals) for a total of 24 h after drug injection, confirming the presence of GTCS-like activity on ECoG in both mutants ( Figure 7A) but no seizure activity in control littermates (data not shown). Similar outcomes were seen in Hcn1 M142I/+ animals, where convulsive seizures were observed in 6/9 mutant mice in response to lamotrigine ( Figure 7A,C, right). Of note, four of the six Hcn1 M142I/+ animals with adverse response had grade 6 seizures, again higher than the maximum seizure grade observed in Hcn1 G380D/+ mice. ( Figure 7C). Video-ECoG recordings performed on three Hcn1 M142I/+ mutants confirmed the presence of GTCS electrographic activity in 2/3 animals ( Figure 7A). In both lines, all seizure events recorded through video-ECoG were reflected in behavioral manifestations, implying that data collected through observation in non-implanted animals are likely to have captured all seizure events generated in response to lamotrigine injection.
Results with phenytoin (30 mg/kg PE, i.p.) were even more striking, as 14/18 mutant animals tested from the Hcn1 G380D/+ line developed convulsive seizures within 2.5 h of injection (this number includes 11/14 animals with no prior drug exposure and 3/4 animals previously exposed to lamotrigine; Figure 7B,C, left). Latency to first seizure after injection was longer than observed with lamotrigine, in line with the slower time to peak plasma concentration for phenytoin. In addition, multiple seizures were observed during a period up to six hours after injection, again consistent with the longer half-life of phenytoin compared to lamotrigine (Markowitz et al., 2010;Hawkins et al., 2017).
All the Hcn1 G380D/+ or Hcn1 M142I/+ animals tested, either in response to lamotrigine or phenytoin, were randomized to receive a control vehicle injection either one week before or one week after the drug administration session. None had convulsive seizures in response to vehicle injection during the period of observation (4 h for behavioral experiments, and 24 h for video-ECoG experiments), further confirming that the seizures observed in the period following drug administration were a direct consequence of the pharmacological challenge ( Figure 7C). As an additional control, we tested the effects of a third anticonvulsant, namely sodium valproate (VPA), which has not been reported to worsen seizures in any of the HCN1 patients in our database.
Administration of sodium valproate (250 mg/kg, i.p.) did not result in the occurrence of convulsive seizures in any of the Hcn1 G380D/+ or Hcn1 M142I/+ animals tested ( Figure 7C).
However, use of 250 mg/kg VPA resulted in profound sedation of Hcn1 G380D/+ mice, which was not seen in Hcn1 M142I/+ mice or WT littermate controls, with 9/10 Hcn1 G380D/+ animals failing to respond to gentle touch or passive moving of the tail 45 min after drug injection vs 2/8 Hcn1 M142I/+ and 0/8 WT animals (note that tail pinching still elicited a response in all mice tested).
What may be the mechanism by which lamotrigine and phenytoin induce seizures in HCN1 mutant mice? Although both drugs act as Na + channel blockers, lamotrigine is also widely considered an HCN channel activator, a property thought to contribute to its antiepileptic effects (Poolos et al., 2002). Despite multiple reports on its action on I h in neurons, showing somewhat inconsistent outcomes (Peng et al., 2010;Huang et al., 2016), no studies have been conducted to directly test the effect of lamotrigine on isolated HCN channels expressed in heterologous systems. To obtain further clarification on the mechanism of action of lamotrigine, we tested the drug on wildtype HCN1 and HCN2 channels after transient transfection in HEK293T cells. Surprisingly, we found that lamotrigine had no effects on either the midpoint voltage of activation or maximal tail current density of either HCN1 or HCN2 channels. Lamotrigine also had no effect when HCN1 was coexpressed with its auxiliary subunit TRIP8b ( Figure 8). In contrast, under the same conditions, lamotrigine exerted its expected inhibitory effect on Na v 1.5 sodium channel currents (Figure 8 -figure supplement 1; Qiao et al., 2014). These results clearly show that lamotrigine is not a direct activator of HCN channels, and suggest that any effects reported in neurons are the result of indirect regulation of I h , perhaps secondary to Na + channel block, and likely contingent on cell-specific, state-dependent signal transduction pathways. Thus, it is more likely the effects of lamotrigine on Hcn1 G380D/+ and Hcn1 M142I/+ animals are due to its action to directly block Na + channelssimilar to the effects of phenytoin -rather than an enhancement of HCN currents.
While we have at present no available indication that either of the two M153I patients showed adverse effects to lamotrigine or phenytoin treatment, our results in mice suggest that despite the clear differences in their phenotypes, there may be some fundamental similarities in the disease mechanisms at play both in patients and mouse models for HCN1-linked DEE.

Reduced efficacy of HCN/I h -blocking compounds on HCN1-G391D channels and Hcn1 G380D/+ neurons
The observation that several of the known HCN1 variants, when tested for their properties in vitro (Nava et al., 2014;Marini et al., 2018) or in vivo (Bleakley et al., 2021), exhibit an increase in "leak" current suggests that use of HCN channel blockers may be beneficial in the treatment of certain HCN1 patients. Several such blockers exist, although at present none is able to efficiently cross the blood brain barrier. More importantly, all currently available inhibitors that are specific for HCN channels, such as ZD7288, ivabradine, zatebradine, cilobradine and their derivatives (Melchiorre et al., 2010;Del Lungo et al., 2012), are thought to act through the same mechanism, namely as pore blockers by interacting with residues located within the pore cavity. At the same time, many of the most severe HCN1 variants, including G391D, are located in the S6 transmembrane domain of the channel, which forms the inner lining of the pore. Such observations raise the question whether pathogenic mutations in S6 may decrease the efficacy of HCN pore-blocking compounds.
The structural models presented in Figure 9F illustrate the proximity of the G391D mutation to the known binding site of ZD7288 in the HCN pore cavity, wherein the residue immediately preceding G391 (V390 in human HCN1, corresponding to V379 in mouse) was shown to critically affect the sensitivity of the channel to the drug (Cheng et al., 2007). The presence of charged aspartate residues in G391D-containing subunits leads to a disruption in the symmetry of the pore in heteromeric channels containing at least one WT subunit, and consequently in the relative distance between the V390 side chains facing the cavity. To test whether this predicted disruption in the ZD7288 binding site leads to a reduced efficacy of the drug, we expressed heteromeric HCN1 channels containing WT and G391D mutant subunits in HEK cells. As illustrated in Figure 9A,B, compared to the effect of ZD7288 to reduce the current expressed by WT HCN1 channels by ~67%, the drug was minimally effective in blocking either the instantaneous or the time-dependent current component generated by mutant WT/G391D channels (~15% inhibition). Addition of Cs + , another HCN channel blocker (albeit a less specific one, compared to ZD7288) with a different mechanism of action and binding site (DiFrancesco, 1982), led to a reduction in the current expressed by both WT and mutant channels, confirming that the observed currents were indeed generated by HCN1 WT/G391D heteromeric channels.
Parallel experiments in brain slices, where bath application of ZD7288 is the most common method used to isolate the contribution of I h to neuronal physiology, demonstrate the additional difficulties posed by the limited efficacy of the drug in the mutant. As shown in Figure 9C-E, application of the compound did not modify either the RMP or input resistance in Hcn1 G380D/+ neurons, while appropriately eliminating voltage sag, hyperpolarizing the RMP, and increasing input resistance in WT neurons. These results leave unanswered the question whether the positive shift in RMP in Hcn1 G380D/+ neurons is directly due to "leaky" HCN1 channels or is a secondary response to changes in other conductances, further underscoring the need for a radically different approach in dealing both with the therapeutic treatment and experimental investigation of HCN1linked DEEs.

Discussion
The speed and precision of next generation sequencing (NGS) allows the efficient genetic assessment of children that present with early onset epilepsies. The overarching goal of such precision medicine is to identify the specific molecular and network mechanisms that link distinct epilepsy mutations in individual genes to particular disease profiles, resulting in more effective gene-or mutation-specific therapeutic strategies. Despite the fact that dysfunction in ion channels carrying missense mutations can be promptly modeled in heterologous expression systems, such assessments do not sufficiently reflect the dysfunction caused by the mutations in vivo, preventing the direct translation of such datasets into applicable clinical strategies. In our study, we set out to fill this information gap by designing new preclinical models, which reproduce key disease features of HCN1-linked DEE. The Hcn1 G380D/+ and Hcn1 M142I/+ knock-in mice we generated display not only an appropriate and robust epilepsy phenotype, along with comorbid behavioral abnormalities, but illustrate the difficulties intrinsic to the treatment of genetic developmental epilepsies, including the adverse response of individuals with HCN1 mutations to conventional antiseizure medications, and the potential for altered drug sensitivity of mutant channels. These results underscore how mutations in voltage-gated ion channels can fundamentally alter brain development and the response of neuronal circuits to pharmacological challenge. In this scenario, our models provide a useful system for testing mechanistic hypotheses and developing new therapeutic approaches.
A notable feature of our models and patients with HCN1 mutations is the paradoxical induction of seizures by lamotrigine and phenytoin. A similar phenomenon has been reported both in Dravet syndrome patients carrying SCN1A mutations (Perucca and Perucca, 2019) and in mice modeling Scn1A haploinsufficiency (Scn1A +/- ;Hawkins et al., 2017). In Dravet syndrome, loss-of-function mutations in the SCN1A sodium channel gene are thought to promote seizures by exerting a net effect to reduce the excitability of inhibitory interneurons (Catterall, 2018). This has led to the suggestion that certain anticonvulsant drugs acting as Na + channel blockers paradoxically exacerbate seizures in this syndrome as a result of a synergistic action to further suppress inhibition.
Interestingly, the axon-specific expression of HCN1 channels in PV+ interneurons matches the similarly polarized subcellular distribution of voltage-gated Na + channels, wherein Na v 1.1 (encoded by SCN1A) is the predominant subtype expressed in the axon of PV+ interneurons (Ogiwara et al., 2007;Dutton et al., 2013;Ogiwara et al., 2013;Hedrich et al., 2014;Hu and Jonas, 2014). To maintain ionic homeostasis during repetitive firing, PV+ interneuron axons also strongly express Na + /K + -ATPases (Peng et al., 1997), which will activate in response to Na + entry leading to membrane hyperpolarization due to electrogenic pump activity. Recent pharmacological studies in vitro have proposed that axonal expression of HCN channels can counteract the hyperpolarizing Na + pump current during repetitive firing, allowing for fast action potential firing and propagation (Roth and Hu, 2020). Thus, blockade of HCN channel activity in PV+ interneuron axon terminals results in a failure in action potential propagation and reduced perisomatic inhibition onto target excitatory neurons (Southan et al., 2000;Aponte et al., 2006;Roth and Hu, 2020). Since the only HCN subtype expressed in PV+ interneuron terminals is HCN1, diminished GABAergic input onto excitatory neurons as a result of HCN1 protein dysfunction could contribute to excitation/inhibition imbalance and epilepsy. Such an effect could be exacerbated by further inhibition of Na + channel activity by antiseizure medications, similar to what is thought to occur in Dravet syndrome. HCN1 subunits also substantially contribute to regulating the excitability of a second class of inhibitory neurons, namely somatostatinpositive interneurons, which target the dendrites of pyramidal cells (Matt et al., 2011).
Here, HCN1 acts along with HCN2 to set the somatic resting potential and modulate the spontaneous activity of this class of interneurons. Future studies aimed at investigating the physiological effects of HCN1 mutations in identified interneuron populations, along with conditional knock-in mouse lines which limit allele expression to select classes of inhibitory neurons, may provide a more definitive answer to the question of how different variants may affect perisomatic and/or dendritic inhibition in HCN1-linked epilepsy.
When tested in heterologous expression systems, epilepsy-linked HCN1 mutations have been shown to affect the channel in several different ways. These actions include loss-of-function due to decreased HCN1 channel activation, decreased protein levels, and/or impaired targeting to the plasma membrane. Some of the variants exert a dominant negative effect when a mutant subunit assembles into a heteromeric channel with WT subunits. Conversely, gain-of-function effects include accelerated activation kinetics, depolarized midpoint voltage of activation (as in M153I), and an increased voltage-independent component of the HCN current, i.e., generation of a depolarizing "leak" current (as in G391D). Why do HCN1 mutations that in heterologous expression systems have divergent effects on channel function equally result in epilepsy, and a similar paradoxical response to certain Na + channel blockers? Our results in Hcn1 G380D/+ and Hcn1 M142I/+ mice show that both these mutations ultimately led to a net decrease in I h -dependent voltage sag, and a significant decrease in protein expression levels ( Figure   4), despite the M153I variant yielding an apparent gain-of-function channel upon heterologous expression (Marini et al., 2018). This suggests that misfolding and impaired biosynthesis of mutant HCN1 subunits may be a prevalent consequence likely to affect the function of neurons in multiple ways. An increase in cellular stress, for example, would be expected to affect PV+ neurons in particular as they are known to be extremely sensitive to metabolic and oxidative stress (Kann, 2016). Such a mechanism may explain why two variants with seemingly divergent biophysical alterations in channel functional properties ultimately cause overlapping phenotypes.
The divergent characteristics of the two variants do, however, lead to discernible phenotypes, as shown by the overall greater severity of the Hcn1 G380D/+ DEE phenotype, and the remarkable sex-dependence of mortality in Hcn1 M142I/+ mice (Figure 1). This suggests that, while some fundamental mechanisms are affected in similar ways, there are distinctions in the manner in which developmental trajectories are perturbed by different HCN1 mutations. As for other known ion channel linked syndromes, such distinctions may explain the variety of disease phenotypes observed in patients carrying HCN1 pathogenic variants, which range from epileptic encephalopathies to genetic generalized epilepsies and genetic epilepsy with febrile seizures plus (Bonzanni et al., 2018;Marini et al., 2018). An analogous picture emerges from the study of a third HCN1 variant recently modeled in mice (Hcn1 M294L/+ , corresponding to human p.Met305Leu; Bleakley et al., 2021). Similar to HCN1-G391D, this variant generates a channel with a significant "leak" or voltage-independent current component, although protein levels in the Hcn1 M294L/+ mouse brain are less dramatically affected. The epileptic phenotype of Hcn1 M294L/+ mice is indeed considerably milder than either Hcn1 G380D/+ or Hcn1 M142I/+ animals.
Intriguingly, however, Hcn1 M294L/+ mice are also adversely affected by lamotrigine, which caused the induction of seizures, similar to mutant animals from the two lines presented here. From the therapeutic point of view, the observed mechanistic convergence does offer some hope that patients with genetic HCN1 channel dysfunction can be treated following similar principles or strategies across different variants. A larger convergence in treatment strategies, centered on the degree of interneuronal dysfunction across multiple epilepsies, may also be envisioned to emerge in the future.
Finally, what may such strategies look like? Our data once again stress the difficulty of devising targeted therapeutic approaches for ion channel-associated DEE syndromes using existing tools. As stated above, several of the identified epilepsyassociated HCN1 variants generate channels with gain of function and/or gain of aberrant function properties, implying that HCN channel blockers may be potentially helpful. At the same time, several of the most severe mutations in HCN1 epilepsy patients are located in the S6 segment, which forms the lining of the pore cavity, or in the immediate vicinity to the channel's intracellular gate. These include G391D (Figure 9) and two more variants at the same position, G391C and G391S; as well as M379R, I380F, A387S, F389S, I397L, S399P, and D401H (Nava et al., 2014;Lucariello et al., 2016;Marini et al., 2018;Wang et al., 2019). While we do not have complete information available about the effects of known HCN inhibitors on the other variants, results presented here using ZD7288 on G391D-containing HCN1 channels strongly suggest that the mutations may decrease the efficacy of the pore-blocking compounds, reducing their ability to suppress both the time-dependent and time-independent current component generated by the mutant channels (Marini et al., 2018). There is therefore a need to develop new small molecule compounds that target the HCN1 channels at sites outside the pore cavity.
The available 3D structures of HCN1, combined with molecular dynamics simulations, coarse-grained modeling and molecular docking, suggest that there are many opportunities for such a pharmacological-targeting approach to succeed (Lee and MacKinnon, 2017;Gross et al., 2018;Porro et al., 2019). The structures have indeed revealed unique intracellular regulatory modules, including both the HCN domain and Clinker/CNBD, which may allow for the targeting of HCN channels through allosteric modulation of pore gating properties. Furthermore, the recently determined structures of HCN4, the main cardiac isoform of HCN channels, have demonstrated critical differences in the relative arrangement of such regulatory modules (Saponaro et al., 2021), which may be exploited towards the development of isoform-specific drugs.
Compound library screenings for small molecules with HCN1-subtype specificity have already yielded some initial and promising results (McClure et al., 2011;Harde et al., 2019). Other approaches include the design of cell-penetrant peptides, which interfere with the interaction between HCN channels and their regulatory auxiliary subunit, TRIP8b (Han et al., 2015;Saponaro et al., 2018), as well as antisense oligonucleotidebased strategies aimed at HCN1 mRNA downregulation. Of note, the latter strategy may be particularly effective in light of the hypothesis that protein misfolding may be an important component of the pathology in HCN1-linked DEE, together with the observation that Hcn1 null mice do not have spontaneous seizures (Huang et al., 2009;Santoro et al., 2010).
In conclusion, the availability of strong preclinical models, such as the Hcn1 G380D/+ and Hcn1 M142I/+ mouse lines presented here, provides an ideal platform for further investigation of mechanisms underlying epileptic encephalopathies, and for testing new and paradigm-shifting therapeutic strategies.

Animals:
Mouse colonies were maintained both at the University of Cologne and at Columbia University in New York. For animals housed in Cologne, mice were kept in type II long plastic cages under standard housing conditions (21±2°C, 50% relative humidity, food The following commercially available mouse lines were used: B6.129P2-Pvalb tm1(cre)Arbr /J (Jackson Laboratories stock 017320; Bar Harbor, ME) and B6.129S-Hcn1 tm2Kndl /J (Jackson Laboratories, stock 016566). All animals were maintained on a C57BL/6J background.

Mouse genome editing:
We employed CRISPR/Cas9 genome editing using the Easy Electroporation of Zygotes (EEZy) approach in order to generate Hcn1 G380D (C57BL/6J-Hcn1 em1(G380D)Cecad , further referred to as Hcn1 G380D ) and Hcn1 M142I (C57BL/6J-Hcn1 em2(M142I)Cecad , further referred to as Hcn1 M142I ) mice as previously described (Tröder et al., 2018). gRNAs were selected using CRISPOR (Haeussler et al., 2016). Hcn1 G380D mice were generated using the crRNA sequence 5'-ACTGGATCAAAGCTGTGGCA-3' and the ssODN sequence 5' -CCCAAGCCCCTGTCAGCATGTCTGACCTCTGGATTACCATGCTGAGCATGATT GTGGGCGCCACCTGCTACGCAATGTTTGTTGATCATGCCACAGCTTTGATCCA GTCTTTGGACTCTTCAAGGAG -3', containing a new Bcl1 restriction site. Hcn1 M142I mice were generated using the crRNA sequence 5' -ATCATGCTTATAATGATGGT - were anesthetized with 0.5 -4% isoflurane in 100% oxygen, and kept at 0.8 -1.5% isoflurane throughout the surgery. Body temperature was maintained at 36.5°C using a homeothermic heating pad (Stoelting, Germany). Mice were placed into a stereotaxic device (Kopf instruments, CA, USA), a midline skin incision was made above the skull and the periosteum was denatured by a short treatment with 10% H 2 O 2 , followed by rinsing with 0.9% NaCl solution, drying of the skull and application of dental cement (OptiBond™, Kerr Dental, Germany) to harden the skull surface. The transmitter body was implanted subcutaneously in a pouch made in the loose skin of the back. The two lead wires were tunneled subcutaneously through the incision on the skull. Using a dentral drill, a small hole was drilled above the hippocampus (2 mm posterior from Bregma, 2 mm from the midline, always on the right side) for the recording wire, and another hole was made above the cerebellum for the reference wire. The wires were placed into the holes so as to touch the dura and fixed with dental cement. The skin was subsequently closed with tissue glue (GLUture ® , World Precision Instruments, WPI, USA) and animals were allowed to recover for 5 days before data acquisition.
Data acquisition and analysis: Recordings were performed by placing the animal's home cage onto a receiver board that detected the ECoG and movement activity signals, and together with the synchronized video recordings data were digitally stored using Ponemah ( The Netherlands) mice had to run on an enclosed walkway on a glass plate. Run duration variation was set to 0.5˗20 sec, with a maximum speed variation of 60%, and a minimum of ten consecutive steps. A minimum of eight runs per mouse was collected. Runs were classified with the Catwalk XT software (Noldus, Wageningen, The Netherlands).
Footprints were detected automatically by the software and manually corrected by visual inspection. The following parameters were used for analysis: running speed, stand (duration in seconds of contact of a paw with the glass plate), stride length (distance between successive placements of the same paw in centimeter), step cycle (the time in seconds between two consecutive paw placements), base of support (BOS, the average width between the hind paws), step sequence (contains information on the order in which four paws are placed), and regularity index (expresses the number of normal step sequence patterns relative to the total number of paw placements in percent).

Stereotaxic virus injection
Mice were anaesthetized using isoflurane (Covetrus, Portland, ME) and provided analgesics (Carprofen, Zoetis, Troy Hills, NJ). A craniotomy was performed above the target region and a glass pipette was stereotaxically lowered to the desired depth.  To assess HCN channel activation curves, different voltage-clamp protocols were applied depending on the HCN subtype: for HCN1 and HCN1 coexpressed with TRIP8b, holding potential was -20 mV (1 s), with steps from -30 mV to -120 mV (-10 mV increments, 3.5 s) and tail currents recorded at -40 mV (3.5 s); for HCN2, holding potential was -20 mV (1 s), with steps from ˗40 mV to ˗130 mV (˗15 mV increments, 5 s) and tail currents recorded at -40 mV (5 s). Patch-clamp currents were acquired with a sampling rate of 5 kHz and lowpass filtered at 2.5 kHz. Na v 1.5 recordings were obtained from holding potentials of ˗80 mV or ˗130 mV, as indicated. Currents were elicited by stepping to ˗30 mV for 50 ms followed by a step to the next holding potential (-60 or -130 mV for 5 s) in a cycling manner. To measure the inactivation curves a voltage step protocol was used starting from a holding potential of -100 mV for 150 ms followed by a series of inactivating pulses (from -30 to -130 mV, for 500 ms); the fraction of channels which remain available after each inactivating pulse were assessed by the peak currents during the following short test pulse at 0 mV for 50 ms. LTG was applied to the (

Data analysis:
Sample size estimation was based on prior studies and experience (Festing, 2018). For behavioral experiments in Figure 2, sample size was calculated using a power analysis assuming an effect size of 0.5, a power of 0.8 and an α error of 0.05, based on previous experience (G*Power 3.1.9.2, University Düsseldorf, http://www.gpower.hhu.de). No data were excluded after analysis. Statistical analysis was performed using GraphPad Prism (Version 9.0.1, Graphpad, CA, USA). Parameters were assessed for normality using the D'Agostino & Pearson test. For normally distributed data, means were compared using a two-tailed Students t tests (assuming equal variances between genotypes) and paired data was analysed with a paired t test. For repeated measurements, a two-way repeated measurement (RM) ANOVA was performed, and, when appropriate, a post hoc Šídák's multiple comparisons test followed. For non-parametric data, medians were compared using the Mann Whitney U test, paired data were analysed with a Wilcoxon matched-pairs signed rank test, and in case of two grouping factors a Kruskal-Wallis test with post hoc Dunn's multiple comparisons, when appropriate, was performed. All tests were two-tailed, and statistical significance accepted at P < 0.05. All parameters were assessed for a sex difference with a mixed-effects analysis, having genotype and sex as grouping factors. Since we did not find sex-specific differences (except for parameters shown in Figures 1A and 1C), data from female and male subjects were pooled. Unless otherwise stated, data represent mean ± SEM.         The HCN blocker ZD7288 (10 µM) abolished sag (C) significantly hyperpolarized RMP (D) and increased input resistance (E) in WT, but did not alter any of these parameters Hcn1 G380D/+ neurons. (** P < 0.01, *** P < 0.001, paired t test; number of recorded cells is indicated in the bar graphs; number of animals used: WT n = 4, WT+ZD7288 n = 3, Hcn1 G380D/+ n = 3, Hcn1 G380D/+ +ZD7288 n = 3; see Figure 9 -source data 1 for numerical values) F) Snapshots of the human HCN1 pore region structure at the end of a 100 ns Molecular Dynamics (MD) simulation run, with or without introduction of the G391D mutation in two or four subunits.        A) The inhibition of Nav1.5 channel activity by lamotrigine (LTG) was tested at different holding potentials. Representative sodium currents recorded from HEK293T cells transiently expressing Nav1.5 channels before (control), upon 100µM LTG bath application (LTG) and after drug wash out (wash out) are shown. The cell was held at −80 mV (top) or -130 mV (bottom), then stepped to a test potential of -30 mV for 20 ms every 5 seconds. After the test potential the cell was stepped to -60 mV or -130 mV to keep the channel in the inactivated or resting state (top and bottom, respectively).
Marked current inhibition can be noted in the traces shown on the top, consistent with the preferential binding of LTG to the inactivated state of sodium channels. B) LTG binding shifts the inactivation curve of Nav1.5 channel. Inactivating currents were elicited every 10 seconds using the voltage step protocol shown. The holding potential between each series of steps was set to -60 mV in order to keep the channel in the inactivated state. The current recorded at the test pulse of 0 mV was plotted against the voltage of the inactivating pulse to get the inactivation curve in control solution, upon bath application of 100 µM LTG, and after drug washout (traces from representative cell are shown at the bottom