FormalPara Key Summary Points

Why carry out this study?

Trofinetide is the first approved treatment for Rett syndrome, a rare, genetic neurodevelopmental disorder.

In the 12-week LAVENDER study, trofinetide was administered using weight-based dosing to females aged 5–20 years with Rett syndrome and showed significant improvements in the coprimary efficacy endpoints, the key secondary efficacy endpoint, and a secondary endpoint related to nonverbal communication, compared with placebo.

This exposure–response modeling study investigated the relationship between trofinetide exposure levels and efficacy (or response) and whether exposures within the target range identified in the phase 3 LAVENDER study are associated with optimal efficacy.

What was learned from the study?

Higher trofinetide exposure was associated with significant improvements in scores for the Rett Syndrome Behaviour Questionnaire, Communication and Symbolic Behavior Scales Developmental Profile™ Infant–Toddler Checklist Social Composite, and Rett Syndrome Clinician Rating of Ability to Communicate Choices.

Predicted exposure–response efficacy model findings reflect observations in the LAVENDER study and support the approved body weight-banded dosing regimens for trofinetide.

Introduction

Trofinetide is the first US Food and Drug Administration-approved treatment for Rett syndrome (RTT), a rare, genetic neurodevelopmental disorder. In almost all cases, RTT is caused by loss-of-function mutations in the gene encoding methyl-CpG-binding protein 2 (MeCP2), whose deficiency results in abnormal neuronal maturation and plasticity [1,2,3]. In an exploratory phase 2 study (Neu-2566-Rett-001; ClinicalTrials.gov identifier NCT01703533) in adolescent and adult females with RTT, trofinetide (70 mg/kg twice daily [BID]0 provided evidence of a clinical benefit in caregiver- and clinician-assessed measures, including the Clinical Global Impression–Improvement (CGI-I) scale, after 4 weeks’ treatment [4]. In a subsequent phase 2 study of trofinetide in pediatric and adolescent females with RTT (Neu-2566-Rett-002; ClinicalTrials.gov identifier NCT02715115), statistically significant and clinically relevant improvements in caregiver (Rett Syndrome Behaviour Questionnaire [RSBQ]) and clinical (CGI-I) assessments were evident with the highest dose (200 mg/kg BID) versus placebo [5], and all the doses in this study (50, 100, and 200 mg/kg BID) were well tolerated. Post trial it was discovered that when dosed on milligrams per kilogram participants with lower body weights had lower than expected exposure to trofinetide than those with a higher body weight. Given that RTT manifests in children and throughout adulthood, there are a wide range of body weights to consider, which is challenging if dosing regimens are based on milligrams per kilogram.

A pharmacokinetic (PK) analysis showed linear PK across the tested dose ranges, and an exploratory PK-pharmacodynamic analysis suggested a correlation between trofinetide exposure and the magnitude of response in RSBQ and CGI­I [5]. This was subsequently confirmed using exposure–response (E–R) modeling of RSBQ and CGI-I from these phase 2 studies. The individual exposure parameters were predicted using an initial population PK (popPK) model based on pooled PK data from nine clinical studies (MS-007 Study; Unpublished data) and Bayesian estimates. The E–R models for CGI-I and RSBQ scores revealed a statistically significant relationship with trofinetide exposure measures (maximum drug concentration [Cmax] and area under the concentration-time curve for the dosing interval 0 to 12 h [AUC0–12]). After 42 days of treatment the RSBQ model indicated that participants treated with trofinetide, who achieved an AUC0–12, at steady state (AUC0-12,ss) of 800 μg·h/mL, would be expected to experience a 5.1 times greater improvement from baseline compared with placebo, while the CGI-I model suggested that participants treated with trofinetide, who obtained an average steady-state Cmax of 150 μg/mL, were approximately five times more likely to exhibit “much improvement” from baseline compared with placebo. The RSBQ model was used to identify the target exposure which guided the proposed body weight-banded dosing regimen selection for the phase 3 LAVENDER study. The popPK model and stochastic simulations were used to generate the dosing regimens to achieve a target exposure ranging between 800 and 1200 μg·h/mL covering the weight for individuals with RTT aged 2 years and older. In the aforementioned phase 3 study (LAVENDER; ClinicalTrials.gov identifier NCT04181723) of trofinetide in females aged 5–20 years with RTT [6], the proposed dosing regimens were used to achieve the target exposure. Trofinetide was given BID orally or by gastrostomy tube at 30 mL (6 g) for participants weighing ≥ 12 to < 20 kg, 40 mL (8 g) for participants weighing ≥ 20 to < 35 kg, 50 mL (10 g) for participants weighing ≥ 35 to < 50 kg, or 60 mL (12 g) for participants weighing ≥ 50 kg. Change from baseline to week 12 in RSBQ total score (− 4.9 vs − 1.7) and CGI-I score at week 12 (3.5 vs 3.8), the coprimary endpoints, and change from baseline to week 12 in the Communication and Symbolic Behavior Scales Developmental Profile™ Infant–Toddler Checklist (CSBS-DP-IT) Social Composite score (− 0.1 vs − 1.1), the key secondary endpoint, were statistically significantly improved with trofinetide treatment versus placebo, respectively [7]. An additional secondary endpoint that measured nonverbal communication (Rett Syndrome Clinician Rating of Ability to Communicate Choices [RTT-COMC]) by assessing the ability of participants to select pictures or objects using modalities such as eye contact or gestures was also significantly improved with trofinetide (− 0.4) over placebo (0.0) on the basis of the change from baseline to week 12 [8]. Trofinetide was generally well tolerated in LAVENDER. Diarrhea (80.6% and 19.1%) and vomiting (26.9% and 9.6%) were the most common treatment-emergent adverse events (TEAEs) reported with trofinetide and placebo, respectively. A separate E–R safety analysis showed that the TEAEs of diarrhea and vomiting were exposure-dependent (MS-009 E–R safety analyses; Unpublished data).

An updated popPK model [9] using data from 13 clinical studies, including three studies in individuals with RTT (two phase 2 studies [4, 5] and the phase 3 LAVENDER study [7]), was conducted. This analysis was used to estimate the individual systemic exposure of participants with RTT in the LAVENDER study and confirmed that applying the recommended body weight-banded dosing regimens achieved the steady state target exposure range (AUC0–12,ss = 800–1200 μg·h/mL). Significant covariates identified in the popPK model that influenced clearance and volume parameters included age, body weight, and RTT disease status but none were considered to have a clinically relevant impact on exposure when receiving the proposed dosing regimen.

The objectives of these E–R efficacy analyses were to update the previous E–R model that was developed using data from the phase 2 studies [4, 5] with additional efficacy data from the phase 3 LAVENDER study [7] to characterize the relationship between trofinetide exposure and the coprimary efficacy endpoints in LAVENDER (RSBQ and CGI-I), and to develop E–R models to characterize the relationships between trofinetide exposure and the key secondary efficacy endpoint (CSBS-DP-IT Social Composite) and another secondary efficacy endpoint (RTT-COMC) in the LAVENDER study.

Methods

Study Design and Response Endpoints

Efficacy data from three studies in females ≥ 5 years of age with RTT were pooled for the E–R analyses: Neu-2566-Rett-001 (NCT01703533) [4], Neu-2566-Rett-002 (NCT02715115) [5], and LAVENDER (NCT04181723) [7]. The results presented in this manuscript are based on published studies. All procedures performed in those studies involving human participants were conducted in accordance with the ethical standards of the local institutional review boards for each site and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Informed consent was obtained by the parent or legal guardian on behalf of the study participants included in publication of the phase 2 studies and the phase 3 LAVENDER study.

The four efficacy response endpoints included in the E–R efficacy model were based on scores for the RSBQ, CGI-I, CSBS-DP-IT Social Composite, and RTT-COMC [6]:

  1. 1.

    The RSBQ is a 45-item caregiver-rated scale with eight subscales (1. General mood, 2. Breathing problems, 3. Hand behaviors, 4. Repetitive face movements, 5. Body rocking and expressionless face, 6. Night-time behaviors, 7. Fear/anxiety, and 8. Walking/standing). Each item is rated by the caregiver as 0 = not true, 1 = somewhat/sometimes true, or 2 = very true [10]. The RSBQ data used in E–R analyses were the RSBQ total scores (maximum of 90); lower RSBQ scores denote improvement in symptoms.

  2. 2.

    The CGI-I scale is used by clinicians to rate change in an individual’s illness relative to the symptoms at a baseline state using a seven-point Likert scale (1 = very much improved, 7 = very much worse) [11]. Note that CGI-I is not assessed at baseline.

  3. 3.

    The CSBS-DP-IT Social Composite score comprises 13 items (maximum score of 26) for assessing social communication skills and is derived from the CSBS-DP, a standardized screening scale for assessing communication and prelinguistic skills in young children 12–24 months of age [12] and can also be used with older children with developmental delay [13, 14]. Higher CSBS-DP-IT Social Composite scores indicate better social communication development.

  4. 4.

    The RTT-COMC is a clinician-completed assessment of the individual’s ability to communicate their choices or preferences using an eight-point Likert scale (0–7). This is an important measure of nonverbal communication; lower scores indicate more normal functioning [6].

Permission to use the CSBS-DP-IT Checklist was obtained prior to the initiation of the LAVENDER study. A summary of the studies included in the efficacy analyses and additional information on endpoints and sampling strategies are shown in the Supplementary Material, Tables S1 and S2.

Analysis Datasets and Populations

All participants enrolled in the studies used in the E–R modeling were females diagnosed with RTT. Key eligibility criteria for the phase 2 studies and LAVENDER are described in the Supplementary Material, Table S1. Demographic characteristics, including age, body weight, and body mass index, were collected in all of these studies. Baseline values were used to assess covariate effects in these E–R analyses. Data utilized in the creation of the analysis datasets included dosing information (amount, timing), treatment assignment, demographic data, clinical covariates, efficacy endpoint measurements, and timing of collection. The analysis population for each E–R model included individuals who received either placebo or trofinetide and had available trofinetide exposure measures.

Dataset Creation

The E–R efficacy analyses used predicted trofinetide exposure measures based on individual PK parameters estimated from the previously developed popPK model [9] that included 13 trofinetide clinical studies: eight phase 1 studies in healthy participants (including a food effect study [15] and mass balance study [16]), two phase 2 studies in participants with RTT (Neu-2566-Rett-001 [4] and Rett-002 [5]), one phase 2 study in fragile X syndrome [17], one phase 2 study in traumatic brain injury, and the phase 3 LAVENDER study [7]. Exposure measures for each participant and the efficacy endpoints were combined with demographics and covariates in a time-ordered sequence to create the analysis-ready datasets.

The measures of trofinetide exposure that were evaluated included average daily consecutive between-visit exposure estimates of average drug concentration, Cmax, and AUC0–12. In all E–R analyses, trofinetide exposure measures were set to 0 for participants given placebo. If the baseline measure for the RSBQ, CSBS-DP-IT Social Composite, or RTT-COMC score was missing for an individual, the individual was excluded from all analyses evaluating that pharmacodynamic endpoint.

Model Development

Nonlinear mixed effects modeling was used to describe the time course and E–R relationships of trofinetide in the efficacy models for the RSBQ total scores, CSBS-DP-IT Social Composite scores, and RTT-COMC scores. A proportional odds model was used to describe the ordered categorical CGI-I endpoint data. The overall procedure for the development of the E–R efficacy models for RSBQ, CGI-I, CSBS-DP-IT Social Composite, and RTT-COMC scores was:

  1. 1.

    Generation of individual estimates of exposure based on the popPK model.

  2. 2.

    Exploratory data analysis.

  3. 3.

    Base structural model development.

  4. 4.

    Evaluation of covariate effects.

  5. 5.

    Final model refinement; and

  6. 6.

    Model evaluation.

The first-order conditional estimation with interaction method was used during all stages of the model development process. NONMEM computed the value of the objective function (VOF), a statistic that is proportional to minus twice the log likelihood of the data. For hierarchical models, the change in the VOF produced by the inclusion of a parameter is asymptotically χ2 distributed, with the number of degrees of freedom equal to the number of parameters added to or deleted from the model.

Goodness-of-fit was assessed according to the following criteria and/or considerations: convergence of the estimation and covariance routines, estimation of three or more significant digits for each fixed and random effect parameter, size of gradients associated with each parameter at the final iteration of estimation, reasonable parameter estimates based upon the expected relationship, adequate precision of parameter estimates as measured by the relative standard error expressed as a percentage (%RSE = standard error/parameter estimate × 100), agreement in scatterplots of measured versus predicted and individual predicted observations assessed visually, lack of trend or pattern in scatterplots of conditional weighted residuals versus predicted observations and time assessed visually, lack of trend or pattern in scatterplots of individual weighted residuals versus individual predicted observations assessed visually, and estimates of interindividual variability (IIV) and residual variability (RV) for the specified model versus comparator models.

All exploratory data analyses and presentations of data were performed using SAS Version 9.4 (SAS Institute, Cary, NC, USA) and KIWI Version 4 202,111 (Cognigen division of Simulations Plus, Inc., Buffalo, NY, USA). E–R modeling was performed using the computer program NONMEM, Version 7, Level 3.0 (ICON Development Solutions, Hanover, MD, USA). NONMEM analyses were performed on an Intel cluster with the Linux operating system. Further details on statistical methods and model development are shown in the Supplementary Material, Methods for Exposure–Response Efficacy Model Development.

Covariates

The following covariates were evaluated for their ability to explain variability in the E–R model parameters for efficacy: age (years), body weight (kg), body mass index (kg/m2), and baseline efficacy endpoint (RSBQ, CSBS-DP-IT Social Composite, and RTT-COMC scores).

Covariates were considered significant and met the criteria for inclusion in the model if during forward selection there was a statistically significant reduction in the minimum VOF (α = 0.01) and a 5% reduction in IIV for the parameter of interest. On the basis of α = 0.01, a covariate contributing a change in the minimum VOF of ≥ 6.64 (p < 0.01, with one degree of freedom) was considered significant. After forward selection and any model refinement of the full model was completed, backward elimination of covariates was performed (α = 0.001).

If the baseline value for any covariate treated as stationary was missing, the value was imputed from the screening visit value. If both the screening and baseline values of a covariate were missing, the value was imputed on the basis of the age- and sex-specific median baseline value for all participants contributing such information, otherwise the median for the population was used. If the percentage of missing data was greater than 10% for a particular covariate, no imputations were made (the values were set to missing) and the variable was evaluated only in exploratory graphical displays.

Further details of the functional forms for covariate models (continuous and categorical variables) and the forward selection followed by backward elimination approach for covariate evaluation in the efficacy models are shown in the Supplementary Material, Methods for Exposure–Response Efficacy Model Development.

Model Evaluation

For the E–R efficacy models, assuming that uncertainty in the final model parameters was small relative to other sources of variability, the adequacy of the final models was evaluated using a simulation-based, visual predictive check (VPC) method [18] to assess concordance between the model-based simulated data and the observed data.

Utilizing NONMEM, the final model was used to simulate 500 replicates of the analysis dataset sufficient to achieve at least 10,000 participants overall, or 10,000 participants per stratum if the VPC was stratified.

Results

Baseline demographics and baseline scores (except CGI-I) in each E–R analysis set are shown in Table 1.

Table 1 Baseline demographics and efficacy assessments for participants in each exposure–response efficacy analysis dataset
Full size table

E–R Modeling of RSBQ Total Scores

The E–R analysis of RSBQ total scores included 264 participants (placebo n = 117, trofinetide n = 147) and 1022 RSBQ total scores collected for a maximum of 12 weeks following randomization to study therapy. The E–R model for RSBQ response was a linear time-course model that included parameters estimating the baseline RSBQ total scores and the slope for time (Eq. 1). A linear function described the relationship between the trofinetide AUC0–12 and slope, whereby higher exposure was predictive of a reduction (improvement) in RSBQ total scores (Eq. 2).

$${{\text{RSBQ}}}_{ij}=41.61+{{\text{SLOPE}}}_{i}\times {{\text{DAY}}}_{ij}$$
(1)
$$\begin{aligned} {\text{SLOPE}_{i}} & = \left( -0.00908-0.00004147 \times {{\text{AUC}}}_{ij} \right. \\ & \left. -0.002369 \times(\text{WTKG}_{i}-26 )\right)\end{aligned}$$
(2)

where \({{\text{RSBQ}}}_{ij}\) is the model-predicted RSBQ total score in the ith participant at the jth day; \({{\text{SLOPE}}}_{ i}\) is the model-predicted slope for time in the ith participant; \({{\text{DAY}}}_{ ij}\) is the day corresponding to the RSBQ total score in the ith participant at the jth day; \({{\text{AUC}}}_{ij}\) is the average daily trofinetide AUC0–12 between visits at the time of RSBQ total score in the ith participant at the jth day; and \({{\text{WTKG}}}_{ i}\) is the observed baseline body weight (kg) in the ith participant.

All model parameters were estimated with good precision (< 29%RSE), except for the IIV on slope (53.52%RSE; Table 2). The estimated IIV standard deviation in baseline RSBQ total scores and the slope were 10.74 and 0.05211, respectively, reflective of the variability between individuals in observed RSBQ time-course profiles.

Table 2 Parameter estimates and standard errors from the final exposure–response models for RSBQ, CSBS-DP-IT Social Composite, and RTT-COMC scores
Full size table

Of the tested covariates, only baseline body weight as a linear function of slope for time met the criteria for inclusion in the model, whereby heavier individuals had a greater response in RSBQ total scores (Fig. 1a). Assuming the median body weight and median trofinetide AUC0–12 for each age group at week 12, the model-predicted reductions in RSBQ total score were 3.1, 6.1, and 7.1 for individuals administered trofinetide aged < 12 years old, 12–17 years old, and ≥ 17 years old, respectively, compared with reductions of 0.3, 3.1, and 3.1, respectively, for individuals treated with placebo. These age categories reflected the age stratification categories that were used in LAVENDER for the subgroup analysis of age in the coprimary endpoints.

Fig. 1
figure 1

Scatterplot of RSBQ total scores versus baseline weight (a); scatterplot and model-predicted change in RSBQ total scores from baseline to end of treatment versus trofinetide average AUC0–12 (b, c); model-predicted change in RSBQ scores from baseline versus week for each dose level (assuming median trofinetide AUC0–12) (d). In panel a, the line represents a smoothing spline fit to the data. In panels b and c, the solid line represents the model-predicted change for the final exposure–response model; one placebo outlier (RSBQ score = 40) was excluded for graphical purposes. Vertical lines indicate the target trofinetide AUC0–12 range of 800 and 1200 μg·h/mL. In panel b, data plotted at zero trofinetide AUC0–12 concentration represents the placebo values. In panel d, dose regimens of 50, 100, and 200 mg/kg BID were from the phase 2 study (Neu-2566-Rett-002), and doses of 6, 8, 10, and 12 g BID were from the LAVENDER study. ACP-2566–003 LAVENDER study, AUC0–12 area under the concentration–time curve for the dosing interval 0–12 h. BID twice daily, Rett-002 phase 2 study, RSBQ Rett Syndrome Behaviour Questionnaire

Full size image

For the goodness-of-fit plots, the plots of residuals and conditional weighted residuals versus either days or population predictions show a homogeneous scatter of data above and below the 0 reference line, indicative of a relatively unbiased fit for the entire range of model-predicted RSBQ total scores. Plots of individual predicted versus measured RSBQ total scores revealed a reasonably tight clustering of data points about the line of unity at all values, signifying that the E–R model described the data well at the individual participant level (Fig. 2a). For the prediction-corrected VPC plots, the majority of the observed data fell within the prediction interval in each plot, with an appropriate amount and similar distribution of observed data points falling below the 5th and above the 95th percentiles of simulated data (Fig. 3a). After accounting for placebo response, the linear function of trofinetide AUC0–12 on the slope indicated that a typical individual would experience a further decrease from placebo in RSBQ total score with an increase in trofinetide AUC0–12. Assuming trofinetide AUC0–12 values of 800–1200 μg·h/mL, the reductions (improvements) in model-predicted RSBQ total scores at week 12 were 3.55 and 4.94, respectively, compared with a reduction of 0.76 for placebo (Fig. 1b, c). Based on the dose regimen used in the phase 2 study and the LAVENDER study, the model-predicted change in RSBQ total scores from baseline increased in a linear and dose-proportional manner (Fig. 1d).

Fig. 2
figure 2

Goodness-of-fit plots for the final exposure–response models for RSBQ total scores (a) and CSBS-DP-IT Social Composite scores (b). CSBS-DP-IT Communication and Symbolic Behavior Scales Developmental Profile™ Infant–Toddler Checklist; |IWRES| absolute value of the individual weighted residuals, RSBQ Rett Syndrome Behaviour Questionnaire

Full size image
Fig. 3
figure 3

Prediction-corrected VPC plots versus week for the final exposure–response models by treatment (days since first dose) and versus trofinetide exposure (AUC0–12 or Cmax) for RSBQ total scores (a) and CSBS-DP-IT Social Composite scores (b). Median (solid lines) and percentiles (dashed lines) are plotted at the midpoint of each days-since-first-dose interval or each trofinetide average AUC0–12 or Cmax interval. AUC0–12 area under the concentration–time curve for the dosing interval 0–12 h, CI confidence interval, Cmax maximum drug concentration, CSBS-DP-IT Communication and Symbolic Behavior Scales Developmental Profile™ Infant–Toddler Checklist, RSBQ Rett Syndrome Behaviour Questionnaire, VPC visual predictive check

Full size image

E–R Modeling of CGI-I Scores

The E–R analysis of CGI-I scores included data from 989 records from 316 participants collected for a maximum of 12 weeks following randomization to study therapy. There was no clear relationship between trofinetide exposure and CGI-I scores, and none of the exposure measures resulted in an acceptable model fit; thus, modeling was completed, and no E–R relationship was concluded.

E–R Modeling of CSBS-DP-IT Social Composite Scores

The E–R analysis of CSBS-DP-IT Social Composite scores included 182 participants and a total of 679 CSBS-DP-IT Social Composite scores collected for a maximum of 12 weeks following randomization to study therapy. The E–R model for CSBS-DP-IT Social Composite scores was an exponential time-course model that included parameters estimating the baseline CSBS-DP-IT Social Composite score and the rate for time; a linear function described the relationship between the trofinetide Cmax and rate.

$${{\text{CSBS}}}_{ij}=8.67\times {e}^{\left({{\text{RATE}}}_{i} \times {{\text{DAY}}}_{ij}\right)}$$
$${{\text{RATE}}}_{i}=\left(-0.0016+0.00000777\times {C{\text{max}}}_{ij}\right)$$

where \({{\text{CSBS}}}_{ij}\) is the model-predicted CSBS-DP-IT Social Composite score in the ith participant at the jth day; \({{\text{RATE}}}_{ i}\) is the model-predicted rate of change in CSBS-DP-IT Social Composite scores over time in the ith participant; \({{\text{DAY}}}_{ ij}\) is the day corresponding to the CSBS-DP-IT Social Composite score in the ith participant at the jth day; and \({C{\text{max}}}_{ij}\) is the average daily trofinetide Cmax between visits at the time of CSBS-DP-IT Social Composite score in the ith participant at the jth day.

All model parameters were estimated with good precision (< 32%RSE; Table 2). None of the tested covariates during forward selection met the criteria for inclusion in the model. Goodness-of-fit plots (Fig. 2b) and the prediction-corrected VPC plots (Fig. 3b) demonstrated that the CSBS-DP-IT Social Composite score model fit was reasonable, essentially unbiased, and had no significant trends or signs of substantial misfit evident in any of the plots. After accounting for placebo response, the linear function of trofinetide Cmax on the rate indicated that a typical individual would experience an increase from placebo in CSBS-DP-IT Social Composite scores with an increase in trofinetide Cmax. Assuming the median trofinetide Cmax at each week for each dose level, at the median trofinetide Cmax value of 147 μg/mL, the reduction in model-predicted CSBS-DP-IT Social Composite scores at week 12 was 0.33, smaller than the reduction of 1.09 for placebo, indicating treatment with trofinetide resulted in less deterioration of CSBS-DP-IT Social Composite response compared with placebo treatment (Fig. 4a, b). Based on the dosing regimen in the LAVENDER study, the model-predicted change in CSBS-DP-IT Social Composite scores from baseline followed a linear decline over time and was similar with each trofinetide dose (Fig. 4c).

Fig. 4
figure 4

Scatterplot and model-predicted change in CSBS-DP-IT Social Composite scores from baseline to end of treatment (week 12) versus trofinetide Cmax (a, b), and model-predicted change in CSBS-DP-IT Social Composite scores from baseline versus day based on the approved LAVENDER dosing regimen (c) for the final exposure–response model. In panel a and b, the solid line represents the model-predicted CSBS-DP-IT Social Composite scores at week 12; the dashed horizontal line represents a reference line at no change in CSBS-DP-IT Social Composite score. In panel a, the dashed vertical line represents the median Cmax of 147 μg/mL. In panel c, the dashed horizontal line represents a reference line at no change in CSBS-DP-IT Social Composite score. BID twice daily, Cmax maximum drug concentration, CSBS-DP-IT Communication and Symbolic Behavior Scales Developmental Profile™ Infant–Toddler Checklist

Full size image

E–R Modeling of RTT-COMC Scores

The E–R analysis of RTT-COMC scores included 181 participants and 672 RTT-COMC score records collected for a maximum of 12 weeks following randomization to study therapy. The E–R model for RTT-COMC scores was a proportional odds model with two additive components on the logit scale: baseline RTT-COMC score and the drug effect. An exponential function described the relationship between trofinetide Cmax and RTT-COMC scores: as trofinetide Cmax increased, there was a higher probability of lower (improved) RTT-COMC scores.

$${\text{logit}}\left[P\left({{\text{RTT}}-{\text{COMC}}}_{ij}\le m\right)\right]=\sum_{K = 1}^{m}{B}_{k}+{f}_{{{\text{drug}}}_{ij}}+{\eta }_{i}$$
$${f}_{{{\text{drug}}}_{ij}}={e}^{\left(0.00366 \times {C{\text{max}}}_{ij}\right)}$$

where \(P\) is the probability; \({{\text{RTT}}-{\text{COMC}}}_{ij}\) is the model-predicted RTT-COMC score in the ith participant at the jth trofinetide average Cmax; \({B}_{k}\) is the population mean estimated intercept for the logit representing the baseline probability for the different RTT-COMC scores where k = 1, 2, 3, 4, 5, or 6; \({f}_{{{\text{drug}}}_{ij}}\) is the estimated trofinetide drug effect in the ith participant at the jth week; \({\eta }_{i}\) is the interindividual random effect with 0 mean and variance ω2; and \({C{\text{max}}}_{ij}\) is the trofinetide average Cmax in the ith participant at the jth week.

All model parameters were estimated precisely (< 26%RSE) (Table 2). None of the tested covariates during forward selection met the criteria for inclusion in the model. VPC plots demonstrated that the model fit was reasonable, essentially unbiased, and had no significant trends or signs of substantial misfit evident in any of the plots. There was good concordance between the observed and model-predicted proportion of participants in all categories of RTT-COMC scores across the range of trofinetide Cmax (Fig. 5). Across the range of trofinetide Cmax (0–270 μg/mL), increasing trofinetide exposure resulted in a higher probability of lower RTT-COMC scores compared with placebo treatment, indicating an improvement in response. The model-predicted cumulative probability of RTT-COMC score ≤ 3 was 0.55 for the median trofinetide Cmax of 147 μg/mL compared with 0.49 for placebo (Fig. 6).

Fig. 5
figure 5

VPC plots versus trofinetide average Cmax for the final exposure–response model for RTT-COMC scores. Proportions are plotted at the midpoint of each trofinetide average Cmax interval. CI confidence interval, Cmax maximum drug concentration, DV observed RTT-COMC scores, RTT-COMC Rett Syndrome Clinician Rating of Ability to Communicate Choices, VPC visual predictive check

Full size image
Fig. 6
figure 6

Model-predicted cumulative percentage of RTT-COMC scores versus trofinetide Cmax for the final exposure–response model for RTT-COMC scores. The vertical lines represent the 25th and 75th percentiles of Cmax for the target dose. Cmax maximum drug concentration, P probability, RTT-COMC Rett Syndrome Clinician Rating of Ability to Communicate Choices

Full size image

Discussion

This study describes the E–R efficacy analyses of trofinetide data from pediatric and adult females with RTT (≥ 5 years of age) using trofinetide PK exposure measures estimated from the popPK model. E–R models were developed for efficacy endpoints that are relevant in RTT studies, including the LAVENDER study (RSBQ, CSBS-DP-IT Social Composite, and RTT-COMC scores).

E–R modeling of RSBQ, CSBS-DP-IT Social Composite, and RTT-COMC scores suggested that trofinetide is efficacious compared with placebo following the body weight-banded dosing regimens administered in LAVENDER [6, 7]. At trofinetide target AUC0–12 values of 800–1200 μg·h/mL, the reductions (improvements) in model-predicted RSBQ total scores at week 12 were approximately five- to seven-fold higher than placebo. Similar scores were reported in the LAVENDER study, in which treatment differences were associated with medium effect sizes (Cohen’s d), which can be interpreted as clinically meaningful in the context of a rare disease with a high burden for affected individuals and families [7]. In the LAVENDER study, all eight subscores or domains of the RSBQ (i.e., general mood, breathing problems, hand behaviors, repetitive face movements, body rocking and expressionless face, nighttime behaviors, fear/anxiety, and walking/standing) were also directionally in favor of trofinetide, suggesting broad improvement across the core symptoms of RTT [7].

It is unclear as to why heavier individuals were predicted to experience greater reductions in the RSBQ score. Although exposure to trofinetide was previously shown to be influenced by body weight (i.e., lower exposure in lower body weight) in the phase 2 study, the use of weight-based dosing bands ensured that exposure was within the target range for all individuals in LAVENDER. In fact, participants falling in the lowest body weight band (receiving 6 g BID) had slightly higher values for the AUC0–12,ss in the popPK model [9].

At the median trofinetide Cmax value of 147 μg/mL, the reduction in model-predicted CSBS-DP-IT Social Composite scores at week 12 indicated that treatment with trofinetide resulted in less deterioration of CSBS-DP-IT Social Composite response compared with placebo treatment. The model-predicted cumulative probability of RTT-COMC scores indicated that individuals treated with trofinetide were more likely to achieve a score ≤ 3, assuming the median trofinetide Cmax of 147 μg/mL, compared with placebo (55% vs 49%). Scores of ≤ 3 on the RTT-COMC suggest individuals are at the very least capable of making a forced choice between two photographs of objects [6]. Other than baseline body weight on RSBQ scores, none of the other covariates that were tested significantly influenced efficacy endpoints. Although body weight was identified as a covariate on RSBQ scores, body weight-based dosing in LAVENDER ensured that individuals with body weights as low as 12 kg achieved the same target exposure as those weighing > 50 kg. Modeling was not completed for the endpoint of CGI-I scores, as there was no clear relationship between trofinetide exposure and CGI-I scores, and none of the exposure measures resulted in an acceptable model fit. Although the model was insufficiently sensitive to detect improvements in the CGI-I scores relative to trofinetide exposure, there was a statistically significant improvement with trofinetide over placebo on the CGI-I scores at week 12 in the LAVENDER study [7].

Pharmacokinetic-pharmacodynamic (PK-PD) modeling and simulation are core techniques that have been successfully applied in the drug development process. The use of a non-linear mixed effect model (NONMEM) allows modeling of many individuals with various physiologies from different populations to be conducted simultaneously. This allows for the accurate prediction of response based on exposure, and the prediction of an optimal dose regimen, while accounting for variations in age, body weight, etc. The fact that the E–R models demonstrate a positive correlation between exposure levels and response supports the clinical findings and confirms that the benefit seen in the clinical studies is driven by the presence of trofinetide. Furthermore, the efficacy endpoints included in the modeling are clinically important assessment tools. The RSBQ shows correlations with functioning, has been evaluated across a range of ages (2–47 years) in RTT [19,20,21], and is the most widely used assessment tool in RTT studies. Communication is a key concern for caregivers [22], and aspects of social communication and nonverbal communication are captured by the CSBS-DP-IT Social Composite and RTT-COMC, respectively. Verbal communication skills are generally absent in individuals with RTT [23], but many individuals are capable of limited nonverbal communication [14]. The RTT-COMC is therefore an important and relevant secondary endpoint in LAVENDER and was developed on the basis of previous research related to communication skills in RTT [24,25,26] with anchors based on existing, validated measures of early language development and social communication skills [27,28,29]. Despite the merits of these assessment tools there is an unmet need in RTT clinical studies for more objective outcome measures since the reliance on observer-reported outcomes means there is a higher potential of placebo response. Study limitations include the lack of an effect with the CGI-I score, which contrasts with the initial E–R model that showed a relationship between exposure and response. It is sometimes difficult to find an E–R relationship with this type of endpoint since CGI itself is an ordered categorical endpoint. The lack of a significant E–R relationship for this endpoint implies that the E–R model is not sensitive enough to detect the incremental increase in response relative to the exposure.

Conclusion

Significant E–R relationships were evident for the efficacy endpoints in LAVENDER (RSBQ, CSBS-DP-IT Social Composite, and RTT-COMC scores), and higher trofinetide exposure was associated with improved RSBQ, CSBS-DP-IT Social Composite, and RTT-COMC scores. The significant E–R relationships in the model are generally reflective of the efficacy results in the clinical setting based on the LAVENDER study in which trofinetide treatment was associated with significant improvements in the RSBQ, CGI-I, CSBS-DP-IT Social Composite, and RTT-COMC scores compared with placebo. Trofinetide was efficacious within the target AUC0–12 exposure range of 800–1200 μg·h/mL that was achieved using the body weight-based dosing regimen in LAVENDER. The findings of these E–R efficacy analyses support the approved body weight-based dosing regimen for trofinetide.