Although syndrome‐specific DEE models often appear to be characterized by a good face and predictive validity, there can be limitations with an incomplete penetrance of seizure phenotypes in several models of DEEs (Figure 3). ¹⁴⁷ As discussed intensely by Bertocchi et al., ¹⁴⁷ mice with a Cdkl5 knockout or Cdkl5 gene variant and mouse models of glutamate ionotropic receptor N‐methyl‐d‐aspartate (NMDA) (GRIN)‐related encephalopathies carrying mutations in genes encoding one of the NMDA receptor subunits represent examples of DEE models, which do not exhibit relevant seizure activity. Use of these and other models without a seizure phenotype for drug testing implies that chemical or noise triggers are necessary to induce seizure activity. As these triggers may affect the mechanisms of ictogenesis, these procedures can be theoretically associated with a potential bias and impact on predictive validity. Along this line, phenotypes are sometimes only evident in a homozygous situation as compared to heterozygosity in patients. ¹⁴⁷ This difference can also have implications for the network situation and its responsiveness to ASMs and drug candidates.

On the other hand, phenotypes can be rather severe and associated with high mortality rates resulting in ethical issues and high animal numbers needed. ¹⁴⁷ Moreover, the models are challenging to work with due to the intense care necessary to minimize mortality rates.

The time‐consuming nature of the work with genetic mouse models with spontaneous recurrent seizures is also related to the need for the continuous video‐electroencephalogram (EEG) monitoring necessary to assess the impact on seizure frequencies and seizure parameters. Although a spectrum of non‐seizure symptoms is observed in genetic DEE mouse models comprising possible alterations in mood, affective state, cognition, and alertness, ³ , ¹²⁸ , ¹⁴⁷ there are remaining uncertainties when it comes to interpretation of behavioral patterns and translation to the clinical phenotypic spectrum. The same applies to the assessment of motor function, with readout parameters in mice as quadrupeds being less sensitive to syndrome‐related changes than bipeds. Finally, sleep analysis requires a thorough and very time‐consuming assessment based on continuous EEG/ electromyography (EMG) monitoring.

As evident from data summarized in Table 4, there is still limited knowledge concerning pharmacology and the predictive validity of the models and the testing scenarios. For example, several ASMs licensed for management of Dravet syndrome fail to exert relevant effects on hyperthermia‐induced seizures in mouse models of this DEE. ¹³⁰ It remains unclear whether hyperthermia‐induced seizures are characterized by a different pharmacology than recurrent seizures or whether the differences in efficacy are related to acute vs. chronic dosing regimens. Considering that one study in young mice with a genetic Scn1a deficiency demonstrated protection from hyperthermia‐induced seizures, ¹²⁹ further efforts seem necessary to determine factors (such as age, genetic variant, and so on) that might influence the response of hyperthermia‐induced seizure in different Dravet models to ASMs.

For DEEs other than Dravet, only very limited data from genetic rodent models are available (Table 4; Tables S2–S7), thus not providing a robust and solid basis for a retrospective evaluation of the predictive validity of these models.

In general, it is of particular interest to test for an association between time windows for therapy initiation and long‐term outcomes concerning potential disease‐modifying or preventive effects as well as differences in efficacy and tolerability during brain development and disease phases. The age‐dependent efficacy of stiripentol on hyperthermia‐induced seizures has already been discussed as a relevant practical example. ¹⁷ , ¹²⁹ However, there are species‐related differences in the sequence of events during brain development that pose a challenge to direct translation of findings in mice to patients.

In this context, it is important to note that preclinical testing of drug candidates is often performed in young adult and adult animals, not recapitulating the initiation of therapy in infants and young children. This could lead to selection bias during drug development, implying that compounds with a favorable profile for neuropediatric therapy might be overlooked.

Finally, the time‐consuming nature of the preclinical studies in genetic models with spontaneous recurrent seizures often results in a limited heterogeneity or lack of heterogeneity in the testing scenarios. In this context it needs to be kept in mind that assessment in one genetic model with one selected genetic variant and a rather standardized genetic background and environment does not recapitulate the clinical variance related to gene variants, genetic modifiers, and numerous external factors. Although it would be highly informative in terms of robustness and generalizability of findings to test various doses in different models and laboratories, there are obvious limitations in terms of cost, time, and capacity. Thus testing scenarios in models of DEEs often at least partially neglect common recommendations to increase rigor and reproducibility and inform about the robustness and generalizability of preclinical data. ¹⁵⁴ , ¹⁵⁵

In comparison with mouse models, zebrafish models can allow a higher throughput; however, they have even more pronounced limitations when it comes to interspecies differences concerning the entire syndromic spectrum of symptoms or the developmental time course. ¹²⁸ , ¹⁵⁶ Nevertheless, concerning efficacy, available datasets support a relevant predictive validity of zebrafish models of Dravet syndrome confirming efficacy of ASMs in clinical use in patients with Dravet syndrome and lack of efficacy of ASMs that are contraindicated in this DEE. ¹²⁸ , ¹³¹ , ¹³⁹ , ¹⁵⁷

Because various specific disease‐targeting approaches are in the drug development pipeline, it will be of particular interest to compare clinical outcomes with preclinical findings. First, clinical data from patients with SCN1A‐related Dravet syndrome seem to confirm a relevant efficacy of the antisense oligonucleotide zorevunersen (former STK001), ¹⁵⁸ , ¹⁵⁹ which mediates a targeted augmentation of nuclear gene output with increased production of the productive SCN1A and functional messenger RNA (mRNA) by modulation of a splicing event. These data, which resulted in the U.S. Food and Drug Administration (FDA) classification as a breakthrough product, are in line with earlier reports from a genetic mouse model of Dravet syndrome. ⁷ An adeno‐associated viral vector‐based approach (ETX101) with cell‐selective expression of a transcription factor upregulating SCN1A expression has entered clinical testing in patients with Dravet syndrome in 2024. First reports at the company's website are limited to statements on tolerability (NCT06112275, NCT06283212, NCT05419492, https://encoded.com/press‐releases/encoded‐therapeutics‐reports‐clinical‐progress‐of‐etx101‐gene‐therapy‐for‐dravet‐syndrome‐recaps‐2024‐corporate‐achievements‐and‐provides‐2025‐outlook/). Once efficacy data become available, it will be interesting to compare them with preclinical data from a genetic mouse model of Dravet syndrome. ⁸ Encouraged by promising preclinical data from a SCN2A‐GOF mouse model, the gapmer antisense oligonucleotide elsunersen (former PRAX‐222) has entered clinical testing in patients with the corresponding DEE. Recent clinical data have been published providing first evidence for successful translation to clinical use. ¹⁶⁰ , ¹⁶¹ As conclusions about the efficacy of these and many other approaches can only be drawn once more comprehensive clinical datasets are available, it is too early to conclude about the predictive validity of DEE models for efficacy and tolerability testing of disease‐targeting approaches.

In this context, it is important to keep in mind that species differences might imply the need for a species‐specific design of candidates when it comes to ATMPs, including GTMPs, thereby increasing the effort and eventually limiting the translational value of the findings.