Population Modeling of Factor IX Activity Following Administration of Fidanacogene Elaparvovec Gene Therapy in Participants with Hemophilia B

Simulations were performed with a virtual population of 100,000 individuals with hemophilia B. Parameters were randomly drawn from IIV distributions as described by the final model. Required demographic characteristics were randomly sampled from distributions representative of the adult analysis populations from the combined clinical studies. FIX activity was predicted for up to 15 years for each simulated individual following administration of fidanacogene elaparvovec (nominal dose: 5 × 10¹¹ vector genomes per kilogram of body weight [vg/kg], using the commercial manufacturing process [process 3]). The median and 90% prediction intervals (PIs) were calculated for the following summary metrics: FIX activity over time; peak FIX activity; time to peak FIX activity; area under the curve for the follow-up interval (i.e., 15 years; \({\text{AUC}}_{\tau }\)); time to 50, 60, 70, 80, 90, 95, and 99% of peak FIX activity; and time within 50, 60, 70, 80, 90, 95, and 99% of peak FIX activity.

A total of 337 participants and 8931 observations across the nonacog alfa (n = 274) and fidanacogene elaparvovec (n = 63) trials were included in the analysis. Baseline demographics of the pooled study population are summarized in Table 1. The median age of all participants was 21.9 (range < 0.1–64.7) years. The nonacog alfa studies included 141 pediatric and 133 adult participants (median age: 17.1 [< 0.1–64.7] years). The fidanacogene elaparvovec studies included adult participants (median age: 29.0 [18.0–62.0] years). At the time of analysis, most participants in the fidanacogene elaparvovec studies received a dose of 5 × 10¹¹ vg/kg. The proportion of observations BLQ in the analysis dataset was 8.4%.

FIX activity following administration of nonacog alfa was best described by a three-compartment model with IIV on \(BASE\), additive RUV (using a log-transformed both sides approach), and fixed allometric scaling on disposition parameters (referenced to a 70-kg individual). Study-dependent differences in \(BASE\), \(FIXA\), and RUV were likely due to baseline FIX activity enrollment criteria, differences in study inclusion/exclusion criteria, and differences in historical FIX activity assays. Key steps in structural model development for nonacog alfa are described in Table S3 (see ESM).

FIX activity following administration of fidanacogene elaparvovec was best described by first-order FIX synthesis, first-order transgene degradation, and fixed allometric scaling on the rate constants \({k}_{\text{syn}}\) and \({k}_{\text{deg}}\) (referenced to a 70-kg individual). EHL replacement therapy contributions to FIX activity were adequately accounted for by a previously published population PK model [19]. No changes in the disposition and elimination component of the model (described above) were required upon the addition of FIX activity data from fidanacogene elaparvovec studies.

A random effect parameter for IIV on \({k}_{\tau }\) was included, and a Box-Cox transformation was applied to alleviate negative skewness:where \({\eta }_{{k}_{\tau }},t\) is the Box–Cox transformed random effect for \({k}_{\tau }\), \({\eta }_{{k}_{\tau }}\) is the random effect for \({k}_{\tau }\), and \({\theta }_{BXCXK}\) is the estimable parameter for the Box-Cox transformation. \({\theta }_{BXCXK}\) takes on positive values to address positive skewness and negative values for negative skewness.

Two FIX activity time-course phenotypes were observed following the administration of fidanacogene elaparvovec. Individuals with the constant FIX production phenotype achieved peak FIX activity followed by a slow decline, best characterized by a monoexponential profile. Individuals with the transient FIX production phenotype achieved peak FIX activity followed by a temporary sharp decrease then a slow decline, best characterized by a biexponential profile.

A mixture model was used to describe the two FIX activity time-course phenotypes. The probability of being an individual with constant FIX production was estimated with a fixed effect parameter (logit transformed to ensure probabilities were constrained between 0 and 1):where \({P}_{{NRS}_{t}}\) is the log-odds, \({\theta }_{BASEPNRS}\) is the fixed effect parameter for the log-odds, and \({P}_{NRS}\) is the probability of exhibiting the constant FIX production phenotype. Individuals were assigned to either subpopulation 1 (constant FIX production) or subpopulation 2 (transient FIX production):where \({P}_{1}\) and \({P}_{2}\) are the probabilities of being assigned to subpopulation 1 and subpopulation 2, respectively.

For individuals assigned to subpopulation 2 (transient FIX production), changes in FIX production were described by a two-compartment model quantifying the transition of the transgene from ‘productive’ to ‘non-productive’ states, described by first-order rate constants \({k}_{tr,d}\) (transition from productive to non-productive) and \({k}_{tr,p}\) (transition from non-productive to productive). For individuals assigned to subpopulation 1 (constant FIX production), \({k}_{tr,d}\) and \({k}_{tr,p}\) were fixed to 0.

Key steps in developing the final structural model (Fig. 1a) are described in Table S3 (see ESM) and are represented by the following equations:

The final model described the contributions of both fidanacogene elaparvovec and FIX replacement therapy (Fig. 1b), as well as the impact of different FIX activity phenotypes (constant vs transient FIX production; Fig. 1c), to overall FIX activity.

Based on results of univariate testing (Table S4, see ESM), the following covariates met the inclusion criteria (improved model fit at a p-value < 0.01 based on LRT, reduced net IIV, not a subset of or highly correlated with a better performing covariate) and were carried forward for multivariate analyses: effect of age on \({k}_{\text{syn}}\), effect of age on \({k}_{\text{deg}}\), effect of age on \(FIXA\), effect of manufacturing process on \({k}_{\text{syn}}\), and effect of body weight on \({k}_{\tau }\) (referenced to 70 kg).

The effect of concomitant steroids on \({k}_{tr,d}\) was statistically significant (p < 0.01) and indicated that receiving corticosteroids was associated with a 4.97-fold faster transition of the transgene from a productive to non-productive state, representing a faster transient decrease in FIX production. Because changes in the transgene’s ability to produce FIX are likely immune-related and corticosteroids were administered (per study protocol) in response to suspected immune-related increases in liver enzymes, this covariate was considered descriptive with limited predictive utility and not carried forward for multivariate analyses.

Following forward inclusion, the effect of age on \({k}_{\text{syn}}\), \({k}_{\text{deg}}\), and \(FIXA\), the effect of manufacturing process on \({k}_{\text{syn}}\), and the effect of body weight on \({k}_{\tau }\) were retained in the model.

Diagnostic plots demonstrated model goodness-of-fit (Fig. S1, see ESM), and the final model’s predictions overlaid observed FIX activity with good agreement (Fig. 2). Final model parameters (Table 2) estimated that 72% (95% CI 59.3–82.2) of the fidanacogene elaparvovec population was best described by the constant FIX production phenotype (subpopulation 1) and 28% (17.8–40.7) were best described by the transient FIX production phenotype (subpopulation 2). For subpopulation 1, the first-order rate constant parameter for transgene degradation estimated a gradual decline in transgene-produced FIX activity with a half-life of 423 weeks. For subpopulation 2, there was a more rapid transition of the transgene to a non-productive state (estimated half-life of 4.9 weeks) and a slower return to the productive state (estimated half-life of 62.6 weeks).

Bild vergrößern

Age, body weight, and fidanacogene elaparvovec manufacturing process accounted for variability in FIX activity (Table 3 and Fig. 3). Body weight primarily affected peak FIX activity. Compared with the reference scenario (86.1-kg individual), lower body weight (56.5 kg [5th percentile of the analysis population]) was associated with a 34.8% decrease in peak FIX activity, while higher body weight (108 kg [95th percentile]) was associated with a 28% increase in peak FIX activity (Fig. 3a).

Bild vergrößern

Age primarily affected the time to reach peak FIX activity. Younger age (18.8 years [5th percentile]) was associated with a 36.8% earlier time to peak, and older age (56.5 years [95th percentile]) was associated with a 43% later time to peak compared with the reference scenario (35-year-old individual; Fig. 3b). Manufacturing process also affected the time to reach peak FIX activity, with processes 1 and 2 both associated with an earlier time to peak (19.1% and 30.5%, respectively) compared with process 3 (Fig. 3b).

Lower body weight and younger age were both associated with a decrease in \({\text{AUC}}_{\tau }\) over a 15-year period (30.8% and 33.6%, respectively); higher body weight and older age were associated with an increase in \({\text{AUC}}_{\tau }\)(24% and 21%, respectively; Fig. 3c). Unlike body weight and age, manufacturing process did not impact \({\text{AUC}}_{\tau }\).

The time-course of FIX activity following administration of fidanacogene elaparvovec was predicted for a simulated population of 100,000 individuals with hemophilia B. Following administration of fidanacogene elaparvovec, the time-course of FIX activity exhibited a shallow peak within the first 65 weeks before reaching a steady state (Fig. 4a). At Year 1 post-dose, median (90% PI) FIX activity was 11.0 (1.84–40.1) IU/dL. Extrapolation of FIX activity up to 15 years post-dose predicted a gradual longer-term decline, with median FIX activity of 7.16 (1.50–27.9) IU/dL at Year 6 and 4.11 (1.15–17.6) IU/dL at Year 15 post-dose (Fig. 4b). Over the entire 15-year period, the median (90% PI) predicted \({\text{AUC}}_{\tau }\) was 105 (23.4–394) IU·year/dL.

Bild vergrößern

Median (90% PI) peak FIX activity was 13.5 (3.12–41.3) IU/dL, with a median time to peak of 25.3 (6.70–48.7) weeks. Individuals reached 50% of their peak FIX activity within a median of 3.00 (0.000–6.00) weeks and 90% of their peak FIX activity within a median of 10.8 (3.33–20.0) weeks (Fig. 4c). Individuals were predicted to remain within 90% of their peak FIX activity for a median of 1.35 (0.135–4.40) years and within 50% for a median of 8.67 (0.411–15.0) years (Fig. 4d).

The structural model described empirically quantifies the physiological and biological processes that would typically contribute to FIX activity following gene therapy. The proposed interpretations of the model parameters are hypothetical, and other mechanisms may also be responsible for observed changes in FIX activity. A compartmental model for gene and protein expression dynamics provided a semi-mechanistic framework for modeling FIX protein synthesis and transgene degradation following gene therapy administration [25]. While FIX synthesis was dependent on the amount of transgene available, transgene degradation was described by a separate first-order process. Transgene degradation or loss may occur as a result of the slow turnover of cells expressing the transgene (i.e., end-differentiated hepatocytes) or due to immune responses directed at the AAV vector [7, 30, 31]. These processes could influence the long-term durability of FIX activity following fidanacogene elaparvovec administration. Depending on which mechanism of transgene loss predominates, the time-course of FIX activity exhibited at least two distinct profiles (constant or transient FIX production; see Fig. 1c).

Slow turnover of the transgene was quantified by the estimation of a first-order rate constant, \({k}_{\text{deg}}\). This produced a FIX activity time-course with a sustained increase in activity followed by a gradual decline (i.e., constant FIX production). Alternatively, immune-related changes to the transgene’s ability to produce FIX were accounted for with a mixture model, with an additional compartment to describe the transgene transition between productive and non-productive states (Fig. 1a). This resulted in a FIX activity time-course with a transient decrease in activity occurring approximately 8 weeks after gene transfer and partial loss of FIX activity (i.e., transient FIX production).

In a simulated population of individuals with hemophilia B, model-based predictions of FIX activity appeared to be at a steady state at 15 months after fidanacogene elaparvovec administration (Fig. 4a), while longer-term predictions suggested a gradual decline (Fig. 4b). The model’s ability to characterize this longer-term FIX activity was based on extended (≤ 6 years’) follow-up of individuals who received fidanacogene elaparvovec in an initial 52-week phase I/IIa study (ClinicalTrials.gov identifier: NCT02484092) [9] and then rolled over into the study’s phase II long-term follow-up (NCT03307980) [14]. The model adequately described observed FIX activity up to 6 years after gene transfer (Fig. 2b), and extrapolations beyond 6 years assumed that transgene degradation continued in a monoexponential manner.

To quantitatively describe the distribution and clearance of FIX following its synthesis, the model also leveraged FIX activity data from studies of nonacog alfa. The addition of a three-compartment disposition component allowed the model to systematically discriminate between variability in FIX synthesis and FIX clearance. The disposition and clearance component of the model also accounted for the contribution of any FIX replacement therapy to overall FIX activity. This eliminated the need to exclude observations from participants in the fidanacogene elaparvovec studies who received FIX replacement therapy. Although models of FIX activity following the administration of nonacog alfa have been published previously [15, 20], individual-level data from nonacog alfa studies reported here were leveraged to support the estimation of IIV and analysis of covariates given the comparatively smaller sample size of the fidanacogene elaparvovec studies.

Covariate analysis results demonstrated that body weight, age, and manufacturing process explained variability in FIX activity (Table 3, Fig. 3 and Table S4 [see ESM]). The impact of body weight was included in the model a priori on all clearance, volume, and rate parameters. It was anticipated that the effects of body weight could not be precisely estimated given the limited body weight range of the analysis population of adult males. Furthermore, the incorporation of allometric scaling principles allows for future pediatric extrapolation and dose selection assessments. Studies have shown body weight to be inversely related to metabolic rates [26, 32], so it was assumed that higher body weight would result in decreased degradation of the transgene, transgene-produced FIX, and clearance of FIX.

The parameter \({k}_{\tau }\) (describing the transcription and translation of transgene to FIX, thereby influencing peak FIX activity) was not assumed to exhibit an inverse relationship with body weight (as described by an exponent of − 0.25). Instead, the model-estimated exponent for the relationship of body weight on \({k}_{\tau }\) was 1.08, suggesting that translation processes are directly proportional to body weight. This is consistent with prior work demonstrating that higher body weight is associated with increased liver size (the site of transfection and FIX synthesis) [33]. Simulations demonstrated that body weight impacted peak FIX activity, time to peak, and \({\text{AUC}}_{\tau }\) over a 15-year period (Fig. 3); however, these results could be influenced by the weight-based dosing of fidanacogene elaparvovec in which individuals with higher body weights received a higher dose.

The effect of age was evaluated as a covariate on multiple parameters in the model and was retained on \({k}_{\text{syn}}\), \({k}_{\text{deg}}\), and \(FIXA\). However, in a population of adult male patients with hemophilia B, there may be limited information to quantify the relationship between age and overall FIX activity. Simulations demonstrated that age accounted for variability in the time to peak and \({\text{AUC}}_{\tau }\) over a 15-year period. Finally, manufacturing process accounted for variability in the time to peak (Fig. 3). However, most individuals (n = 45) received fidanacogene elaparvovec from manufacturing process 3.

While it was evaluated as a covariate, the effect of concomitant corticosteroids was not retained in the final model. Per protocol, corticosteroids were administered in reaction to elevated liver enzymes (alanine aminotransferase [ALT] and/or aspartate aminotransferase [AST]). An immune response that impairs the transgene’s ability to produce FIX is likely to also produce immune-related increases in liver enzymes. Thus, corticosteroids would be descriptive of an underlying immune response to the AAV vector, with limited predictive utility. Similarly, liver enzymes (e.g., ALT, AST, bilirubin) or other proteins produced by the liver (e.g., albumin) were not evaluated as covariates because they exhibit time-dependent changes following administration of fidanacogene elaparvovec and are expected to have limited predictive utility. The analysis accounts for immune-related changes in FIX production with a mixture model that assumes transient decreases in FIX production are random at the individual level.

Other models of FIX activity following administration of an AAV-based gene therapy in participants with hemophilia B have been published. Shah et al. [34] used a linear mixed-effects model to predict the time-course of log-transformed FIX activity using data starting from 6 months after administration of etranacogene dezaparvovec (Hemgenix^® [35], a recombinant AAV serotype 5 vector encoding the high-activity FIX-R338L variant) up to 2.5 years post-infusion. The analysis population consisted of 55 participants with hemophilia B treated with a single dose of etranacogene dezaparvovec (pooled from phase IIb [NCT03489291] [36, 37] and phase III [NCT03569891] [38] trials). FIX activity was assessed by SynthASil one-stage clotting assay. The study assumed that all individuals had stable FIX activity 6 months after infusion. This 6-month timepoint was used as the analysis baseline and earlier observations were not included. FIX activity measurements taken within five half-lives of FIX replacement therapy administration were also excluded. Frequentist and Bayesian approaches were applied to estimate model parameters and extrapolated FIX activity levels up to 25.5 years.

Here, nonlinear mixed-effects modeling was used to characterize the increase in FIX activity (as assessed by Actin^® FSL assay) following administration of fidanacogene elaparvovec and the disposition following achievement of peak activity. All available data informed the model; observations during the initial 6 months after infusion were not excluded, and long-term FIX activity data up to 6 years post-infusion were available. The model also accounted for the contribution of FIX replacement therapy to FIX activity. Thus, observations within five half-lives of any FIX replacement therapy during the fidanacogene elaparvovec studies did not need to be excluded. Additionally, the present analysis identified that approximately 28% of the population showed decline from their initial FIX activity response by Week 12 (i.e., 3 months post-infusion) and so could be categorized as ‘transient FIX production’; these individuals appeared to have stable FIX activity at later timepoints. If a linear mixed-effects modeling approach were to have been employed with omission of data from this initial time period, then individual slope parameters for these individuals would have likely tended towards 0 and overestimated the durability of their FIX activity responses. The nonlinear mixed-effects model presented here provides a more realistic assessment of the durability of response following administration of gene therapy.

FIX activity predictions up to 25 years were also performed using the present model [39]. However, as in Shah et al. [34], long-term model predictions should be viewed with caution. Here, FIX activity was measured with Actin^® FSL versus SynthASil assays in Shah et al. [34]. Differences in FIX activity measurements have been reported between these assays, with FIX activity values being consistently higher with SynthASil (particularly for the high-activity FIX-R338L variant) [40]. Therefore, direct comparison of predicted FIX activity values across studies employing different assays should be avoided.

The final model adequately described the observed data (Fig. S1, see ESM) using a unified approach that accounted for all sources of FIX activity, rather than excluding data. The model assumed that disposition and clearance mechanisms of transgene-produced FIX are similar to plasma-derived or SHL recombinant FIX, only separately accounting for the contribution of EHL products.

To assess the impact of this assumption on the analysis, model development was repeated with only transgene-produced FIX activity data (Fig. S2, see ESM). FIX activity observations from the nonacog alfa population were excluded as well as any observations attributed to exogenous FIX. The truncated dataset contained approximately 29% of the observations from the combined analysis dataset. A simplified model structure was developed providing a similar description of the data. Specifically, characterization of FIX disposition and elimination was modified from a three- to a one-compartment model (excluding parameters for peripheral and extravascular compartments). Parameters for clearance and central compartment volume were not estimated. Rather, a first-order rate constant for elimination of FIX from the central compartment was included. This simpler model structure could be appropriate in the absence of exogenous FIX activity data; however, the model based on the truncated analysis dataset performed worse at capturing the observed data compared with the combined modeling analysis (Fig. S2a). Both models produced similar time-courses of predicted FIX activity following administration of fidanacogene elaparvovec. Compared with the truncated analysis dataset, the combined model using all available data also generated a more conservative prediction of FIX activity in the first 6 years (Figs. S2b and S2c, see ESM).

The model also assumed dose proportionality (via first-order FIX production and first-order transgene degradation). Due to weight-based dosing of fidanacogene elaparvovec (and dose-adjustment for individuals with a body mass index > 30 kg/m²), there was variability in the total number of vector genomes administered across individuals. The model includes the total number of vector genomes administered as dosing input. No nonlinearity in FIX activity following administration of different doses of fidanacogene elaparvovec has been observed in the clinical studies, but a larger sample size would be required to detect appreciable nonlinearity between doses given the high variability in FIX activity observed between individuals.

Model estimates of FIX activity were based on the Actin^® FSL one-stage clotting assay. Differences between assays in FIX activity measurements (e.g., higher with SythASil compared with Actin^® FSL assay) [40] limit the ability to directly compare FIX activity values across competitor studies. Finally, the model did not assess the impact of immunogenicity on the FIX activity time-course. Individuals who experienced immune responses following fidanacogene elaparvovec administration were not excluded from the analysis and their data contributed to the overall variability in FIX activity over time (via the translation constant, \({k}_{\tau }\), and the time-course for subpopulation 2 in the mixture model).