Delineating clinical and developmental outcomes in STXBP1-related disorders

Abstract STXBP1-related disorders are among the most common genetic epilepsies and neurodevelopmental disorders. However, the longitudinal epilepsy course and developmental end points, have not yet been described in detail, which is a critical prerequisite for clinical trial readiness. Here, we assessed 1281 cumulative patient-years of seizure and developmental histories in 162 individuals with STXBP1-related disorders and established a natural history framework. STXBP1-related disorders are characterized by a dynamic pattern of seizures in the first year of life and high variability in neurodevelopmental trajectories in early childhood. Epilepsy onset differed across seizure types, with 90% cumulative onset for infantile spasms by 6 months and focal-onset seizures by 27 months of life. Epilepsy histories diverged between variant subgroups in the first 2 years of life, when individuals with protein-truncating variants and deletions in STXBP1 (n = 39) were more likely to have infantile spasms between 5 and 6 months followed by seizure remission, while individuals with missense variants (n = 30) had an increased risk for focal seizures and ongoing seizures after the first year. Developmental outcomes were mapped using milestone acquisition data in addition to standardized assessments including the Gross Motor Function Measure-66 Item Set and the Grasping and Visual-Motor Integration subsets of the Peabody Developmental Motor Scales. Quantification of end points revealed high variability during the first 5 years of life, with emerging stratification between clinical subgroups. An earlier epilepsy onset was associated with lower developmental abilities, most prominently when assessing gross motor development and expressive communication. We found that individuals with neonatal seizures or early infantile seizures followed by seizure offset by 12 months of life had more predictable seizure trajectories in early to late childhood compared to individuals with more severe seizure presentations, including individuals with refractory epilepsy throughout the first year. Characterization of anti-seizure medication response revealed age-dependent response over time, with phenobarbital, levetiracetam, topiramate and adrenocorticotropic hormone effective in reducing seizures in the first year of life, while clobazam and the ketogenic diet were effective in long-term seizure management. Virtual clinical trials using seizure frequency as the primary outcome resulted in wide range of trial success probabilities across the age span, with the highest probability in early childhood between 1 year and 3.5 years. In summary, we delineated epilepsy and developmental trajectories in STXBP1-related disorders using standardized measures, providing a foundation to interpret future therapeutic strategies and inform rational trial design.


Supplementary Methods: Longitudinal seizure frequency forecasting and quantification of unpredictability
We developed a framework for longitudinal seizure frequency forecasting to characterize epilepsy progression and predictability.Standardized seizure histories were captured using a previously published scale derived from the Pediatric Epilepsy Learning Health System (PELHS)-championed framework for seizure severity, and seizure frequencies were grouped in the following categories: multiple daily seizures (>5 per day, SF score = 5), several daily seizures (2-5 per day, SF score = 4), daily seizures (SF score = 3), weekly seizures (SF score = 2), monthly seizures (SF score = 1), and no seizures (SF score = 0).All individuals with complete seizure histories in the first year of life (n=78 individuals) were included in the seizure frequency forecasting analysis.
We compared seizure frequencies across monthly time intervals between all combinations of individual patient pairs during the first 12 months of life, when seizures are the most prominent in STXBP1.Based on the framework for semantic similarity analysis to measure phenotypic resemblance using clinical features, 1,2 we derived a complementary method to measure phenotypic similarity based on epilepsy histories, using monthly seizure frequency captured in the PELHS framework as comparisons (see below).
First, we calculated the information content (IC) of each seizure frequency (SF) category in each month (e.g., SF score = 3 during Month 8 of life).The IC was defined as the -log2 of the frequency of individuals in the SF category during the respective month.For example, if 20% of the cohort had at least daily seizures during the 8th month of life, then the IC for SF score = 3 was defined as -log2(0.20)= 2.32.Accordingly, the higher the frequency of individuals with a certain seizure frequency in a specific month, the less informative that seizure frequency for the respective month.In contrast, if only 5% of individuals had daily seizures at 7 years of life, an SF score = 3 would be more informative, and thus weighed more when deriving phenotypic resemblance.
To quantify phenotypic resemblance between two individuals, we summed the minimum IC overlap of each month across all overlapping months.As in the semantic similarity analysis, the rationale was to derive a higher similarity measure for individuals with more rare and thus more informative, or distinguishing seizure frequencies.For example, if two individuals had a less frequent seizures across many months, the derived similarity measure would be higher than two individuals with the same seizure frequency across fewer months and higher than two individuals with a seizure frequency that was more common in the overall cohort.To correct for varying observation times between patient pairs, we adjusted the cumulative IC score by dividing by the number of overlapping months of comparison for each respective patient pair.
After deriving phenotypic similarity scores for all patient pairs in the first year of life, for each individual, we then identified 10 distinct individuals in the cohort that were the most similar phenotypically based on seizure frequencies across the age span, which was defined as that individual's reference cohort.
Then, with the known epilepsy histories in the reference cohort, including seizures after 12 months of life, we predicted each individuals' epilepsy trajectory after the first year, taking the median of the distribution of seizure frequencies for each month in the reference cohort, representing a distribution of likelihood of possible seizure frequency outcomes, as the predicted trajectory.
The predicted (example shown below in green) and actual seizure trajectories (example shown below in blue) after the first year of life was compared and assessing the cumulative difference enabled us to characterize subgroups defined by a high or a low difference between the forecasted and actual seizure frequencies, which we defined as a measure of epilepsy unpredictability.The grouping was performed using the k-means algorithm.For each individual, forecasted seizure frequencies were compared to a distribution of randomly generated seizure frequencies and permutation testing of 100,000 for each individual estimation allowed us to evaluate whether the predicted trajectory was better than chance.

Supplementary Methods: Comparative effectiveness analysis and treatment response
We measured medication efficacy through short-term treatment response in addition to long-term treatment response in three analyses: 1.For the first analysis, we analyzed the relative effectiveness of treatment strategies in seizure reduction, indicated by a decrease in seizure frequency by any degree.We compared months in which periods of seizure reduction coincided with certain ASMs compared to months of no response, which was defined as seizures worsening indicated by an increase in seizure frequency by any degree, or continuous active seizures with the same frequency.We did not include periods of seizure freedom in the category of no response to treatment (i.e., the seizure frequency score had to be above 0).
2. For the second analysis, we analyzed the relative effectiveness of treatment strategies in either seizure reduction or maintaining seizure freedom.We compared seizure reduction or having consecutive months of being seizure-free versus worsening of seizures, indicated by an increase in seizure frequency.
3. For the third analysis, we analyzed the relative effectiveness of treatment strategies in maintaining seizure freedom only, comparing periods of seizure freedom indicated by consecutive months of no seizures versus worsening of seizures as no response to treatment.We did not include periods of seizure reduction in this analysis.
Associations are presented as odds ratios with 95% confidence intervals, correcting for multiple comparisons using a False Discovery Rate (FDR) of 5%.

Supplementary Methods: Framework for virtual clinical trials
Given the heterogeneity of seizures in STXBP1-related disorders, we aimed to identify time windows during which a treatment effect would have the highest probability of being detected in a clinical trial.We derived a framework for virtual clinical trials, randomly sampling 20 individuals with ongoing seizures and simulated a 6-month and 12-month period of 10%, 15%, and 20% seizure reduction.
The percent reduction in seizures was calculated based on the cumulative sum of seizure frequencies at the start of the window for the sampled cohort, and the simulated treatment effect across each trial window was performed by decreasing seizure frequencies across the trial period for this cohort to create a synthetic treatment cohort.
We used the synthetic control method to evaluate the significance of the simulated effect, comparing the distribution of seizure frequencies following simulated reduction of seizures against the observed, natural history distribution of frequencies in the sampled individuals using the Wilcoxon rank sum test.
We ran 1,000 simulated trials for each month across the age span, shown above an example 6-month trial starting at 18 months of life.For each trial window, the Observed Frequency of Trial Success (OFTS) was defined as the proportion of trials out of 1,000 in which a significant effect could be detected.For analyses across seizure types, we chose to include the trial duration (6-month versus 12-month period) and targeted seizure reduction (10%, 15%, 20%, or 50%) based on the distribution of OFTS that resulted in the widest range, spanning from poor probability of trial success to high probability across the age span.This approach enabled us to identify optimal windows during which a treatment response would most likely be observed in a real-world trial when using seizure frequency as the primary outcome measure.

Table 1 Demographics of the Children's Hospital of Philadelphia (CHOP) cohort and Ciitizen Natural History Registry
In our study, clinical documentation analyzed from the Ciitizen Natural History Registry only included milestone acquisition data.The Ciitizen database includes curated clinical documentation from medical records across multiple hospitals and providers where a patient had been seen and received clinical care.In brief, this data was reviewed and verified by trained medical curators, mapped to standardized clinical terminology, and de-identified curated data was then made available to academic researchers.b Protein-truncating variants (PTV)/dels included splice sites, frameshifts, and whole and partial gene deletions. a

Table 2 Correlations between age at epilepsy onset and severity of developmental outcome Developmental measure a Individuals assessed/available data Association/correlation b
Distribution of epilepsy onset in individuals stratified by milestone acquisition (achieved versus not achieved) was performed using Wilcoxon Rank Sum test.Age distribution at which milestones were achieved was assessed using Pearson Correlation Coefficient.Correlation between epilepsy onset and developmental outcomes as measured via validated scales was performed using the Spearman Correlation Coefficient.c Only 8 of 69 individuals did not achieve the milestone of rolling over, limiting interpretation of correlation.d We were unable to assess correlation due to the limited sample size, with all individuals assessed for the specific milestone having neonatal seizure onset.e Only individuals of at least one year of age was assessed.