A Modeling Investigation of the CYP1A Drug Interactions of Riluzole

ABSTRACT

Cytochrome‐P‐450 (CYP)1A2 has been considered the major enzyme metabolizing riluzole since its approval. However, the inhibitor that was used in the original experiments, α‐naphthoflavone, is also a potent inhibitor of CYP1A1. In this work, physiologically based pharmacokinetic (PBPK) modeling investigates the interplay between CYP1A1 and CYP1A2 and the relevance to drug–drug interactions. Following review of clinical and non‐clinical data from literature, the relative contributions of CYP1A1, CYP1A2, and UGT1A8/9 to riluzole metabolism were assigned as 60%, 30%, and 10%, respectively. The model was calibrated on single‐dose pharmacokinetic (PK) data from healthy subjects. The translational potential of the model was verified by predicting riluzole PK in people with amyotrophic lateral sclerosis, spinal muscular atrophy, advanced age, renal impairment, and hepatic impairment, and when administered with a high‐fat meal. The relative contributions of CYP1A1 and CYP1A2 to metabolism were verified through prediction of an observed drug–drug interaction between riluzole and fluvoxamine—a strong CYP1A2 inhibitor and a weak CYP1A1 inhibitor—in children with obsessive–compulsive disorder. Overall, evidence suggests that CYP1A1 is a major enzyme metabolizing riluzole, and that CYP1A2 has similar or lower importance. Only clinically relevant inhibitors of both enzymes may pose a safety concern when administered with riluzole. Strong CYP1A1 inhibitors and strong CYP1A2 inhibitors may be used with caution if they do not significantly modulate the other enzyme. Concomitant use of CYP1A1 inducers may be reconsidered where possible. The enzymatic contributions to riluzole metabolism should be reconsidered after formal drug–drug interaction studies are completed.

Study Highlights.

• What is the current knowledge on the topic?

  • ○
    Riluzole is labeled as primarily metabolized by CYP1A2, and concomitant use of CYP1A2 inhibitors or inducers is typically avoided. However, the original assignment was based on inhibition by α‐naphthoflavone, a non‐selective inhibitor that also strongly inhibits CYP1A1, which has not been formally incorporated into the clinical pharmacology of riluzole.
• What question did this study address?

  • ○
    What are the relative contributions of CYP1A2 and CYP1A1 to riluzole metabolism, and how might this influence drug–drug interaction risk with inhibitors or inducers of either enzyme?
• What does this study add to our knowledge?

  • ○
    Using a validated physiologically based pharmacokinetic (PBPK) model, this study redefines riluzole's metabolic pathways and suggests that CYP1A1 may be the dominant contributor to its clearance. The analysis incorporates a broad set of clinical scenarios and highlights limitations in prior assumptions about enzyme selectivity and f _m assignment.
• How might this change clinical pharmacology or translational science?

  • ○
    This work demonstrates how PBPK modeling can uncover unrecognized contributors to drug metabolism—particularly for legacy therapeutics—and can inform rational DDI risk assessment. It advocates for updated labeling considerations, more selective probe development, and broader inclusion of CYP1A1 in in vitro screening and translational workflows.

1. Introduction

Riluzole is the most prescribed standard‐of‐care therapy for amyotrophic lateral sclerosis (ALS) and is the only approved therapy in Europe for extending life in this ultimately fatal condition. The prescribing information suggests that based on an in vitro evaluation, CYP1A2 is the major enzyme responsible for the metabolism of riluzole. However, the inhibitor that was used to identify CYP1A2 as the major enzyme contributing to the metabolism of riluzole, α‐naphthoflavone, has since been shown to be an equally potent inhibitor of CYP1A1 [1, 2]. Therefore, the historical assignment of CYP1A2 as the dominant enzyme may conflate contributions from CYP1A1, which are not currently reflected in the drug label. A recent review highlights the importance of CYP1A1 as it is expressed in multiple eliminating organs beyond the liver, including the lung and intestines [3]. While CYP1A1 is similar in homology, substrate specificity, and inducibility (e.g., by tobacco smoke) to CYP1A2 [3, 4], there is limited overlap between the clinical inhibitors and inducers of CYP1A1 and CYP1A2 [5, 6]. Therefore, distinction of riluzole metabolism by these two separate pathways is clinically relevant for the management of drug–drug interactions (DDIs) with this standard‐of‐care therapy for ALS.

Drug–drug interactions are typically investigated with gold standard clinical trials in a small number of healthy subjects or patients with intensive pharmacokinetic (PK) sampling following administration of the object drug with and without the precipitant drug(s). In addition to performing clinical studies, physiologically based pharmacokinetic (PBPK) models have earned regulatory confidence for assessing DDIs with mechanistic modeling and simulation [7, 8]. A PBPK model is a system of rate‐based equations that describes the flow and elimination of a drug through organ compartments in a virtual individual according to the actual mechanisms responsible, as far as known. A network of PBPK models for CYP1A2 substrates and inhibitors has been published using the Open Systems Pharmacology Suite software (PK‐Sim) [9].

This study aims to develop a PBPK model for riluzole, verify the contributions of CYP1A2 and CYP1A1 to riluzole metabolism, and re‐evaluate its clinical DDI profile for prescribers. This modeling effort accounts for CYP1A1 expression in extrahepatic tissues (lung and intestine) and addresses interindividual variability in enzyme expression based on gene expression databases.

2. Methods

2.1. Clinical Pharmacokinetics of Riluzole

Riluzole is well absorbed after oral administration. Oral bioavailability is 60% due to a significant first pass effect. It is 96% bound to plasma proteins in the blood yet distributes widely into tissues, including the brain. Riluzole exhibits linear, dose‐independent kinetics at doses studied up to 200 mg/day. The median terminal elimination half‐life is 12 h in healthy subjects but is variable between individuals and in patients with ALS. Phase 1 metabolism is through CYP1A1 and CYP1A2 with a minor contribution of direct glucuronidation by UDP‐glucuronosyltransferase (UGT) 1A8/9 [2]. The drug is a minor substrate for P‐glycoprotein transport [10, 11]. Although riluzole has limited evidence of being a P‐gp substrate in vitro, in vivo studies have demonstrated increased brain exposure following P‐gp inhibition [10, 11]. Of the total administered material, 5% is excreted in feces and 88% or more is excreted in urine, though only 2% of the drug is recovered unchanged. Accordingly, 98% of the administered dose is metabolized.

CYP1A2 is primarily responsible for the formation of the N‐hydroxy‐riluzole metabolite, based on urinary metabolite recovery data [2]. In a single dose study, N‐hydroxy‐riluzole and N‐hydroxy‐riluzole‐O‐glucuronide in urine accounted for 27% of the administered dose of riluzole [12], which is an estimate of the contribution of CYP1A2 to metabolism when corrected for incomplete urinary recovery (f _m,CYP1A2 = 30%). In vitro results are concordant. In human liver microsomes, the CYP1A2 inhibitors caffeine (1 mM) and acetanilide (1 mM) inhibited the biotransformation of riluzole by 37% and 21%, respectively [2]. Although some in vitro inhibition studies used caffeine and acetanilide, the fraction metabolized by CYP1A2 (f _m,CYP1A2 = 30%) was determined from in vivo urinary recovery. In vitro results are cited for support, with recognition of their limited selectivity at high concentrations.

Direct glucuronidation by UGT1A8/9 is a minor process contributing to the metabolism of riluzole in vitro [2]. In a single‐dose study, there was a negligible amount of riluzole glucuronide recovered in urine, but the content in feces was not measured (which may account for 5%–12% of the administered material) [12]. After multiple doses in 14 patients with ALS, the median molar plasma concentration ratio of riluzole glucuronide to unchanged riluzole at the end of the dosing interval was 0.16 [13]. Therefore, the f _m,UGT1A8/9 can be estimated as 10%.

With CYP1A2 and UGT1A8/9 accounting for a combined 40% of riluzole metabolism, CYP1A1 is concluded to be responsible for the remaining 60%. This conclusion is supported by:

1. Riluzole metabolism in human liver microsomes is near‐completely inhibited by α‐naphthoflavone, which is a potent inhibitor of CYP1A enzymes.

2. No other CYP isoform has been identified as contributing to riluzole elimination.

3. Riluzole PK is highly susceptible to factors that are well‐known to modulate CYP1A enzymes (e.g., smoking, Japanese ancestry, sex, and hepatic impairment) and those that are well‐known to not affect CYP1A enzymes (e.g., renal impairment), suggesting that this enzyme subfamily together is responsible for the vast majority of elimination in vivo (~90%).

2.2. Modeling Workflow

A PBPK model for intravenous and oral riluzole was developed using knowledge of its physicochemical and absorption, distribution, metabolism, and elimination (ADME) properties. The model was refined by optimization against single‐dose pharmacokinetic (PK) data from literature. The translational value of the model was verified by predicting PK in clinically heterogeneous scenarios, including hepatic impairment, renal impairment, neuromuscular disease, the elderly, and when administered with a high‐fat meal. The contributions of CYP1A enzymes to riluzole metabolism were verified by predicting the observed DDI with fluvoxamine, a strong CYP1A2 inhibitor and a weak CYP1A1 inhibitor, in children with obsessive–compulsive disorder (OCD) (Figure 1) [14].

FIGURE 1.

Physiologically based pharmacokinetic modeling workflow to evaluate CYP1A metabolism of riluzole.

2.3. Pharmacokinetic Data

Riluzole plasma PK data were collected from literature and segregated for model development and verification as shown in Table 1.

TABLE 1.

Clinical pharmacokinetic data of riluzole used for modeling.

Study	Dose	Form	Cohort	N	Age	Weight
Model development
Le Liboux et al. [15]	50 mg IV inf 30 min	Solution for injection	Healthy Western	16	[18–40]	—
	100 mg SD	Tablet, fasted	Healthy Western	16	[18–40]	—
Tiglutik, FDA Clinical Pharmacology Review	50 mg SD	Solution, fasted	Healthy Western	34	[18–55]	—
	50 mg SD	Tablet, fasted	Healthy Western	34	[18–55]	—
Exservan, FDA Clinical Pharmacology Review	50 mg SD	Tablet, fasted	Healthy Western	30	[18–64]	—
Model verification
Le Liboux et al. [15]	100 mg SD	Tablet, fed	Healthy Western	16	[18–40]	—
Tiglutik, FDA Clinical Pharmacology Review	50 mg SD	Solution, fed	Healthy Western	34	[18–55]	—
Le Liboux et al. [16]	50 mg BID x5d	Tablet, fasted	Healthy Elderly	18	74.6 [70–82]	69.0 [54–84]
Abbara et al. [17]	50 mg QD x5d	Capsule, fasted	Pediatric SMA II and SMA III	13	13.1 [9–17]	29 [17–58]
Groeneveld et al. [18]	50 mg BID	Tablet, fasted	ALS	160	[22–75]	75.3 ± 13.8
Rilutek, Prescribing Information	50 mg SD	Tablet, fasted	Hepatic impairment	—	—	—
Rilutek, Prescribing Information	50 mg SD	Tablet, fasted	Renal impairment	—	—	—
Evaluation of CYP1A metabolism
Grant et al. [14]	50 mg BID	Tablet, fasted	Pediatric OCD	37	14.6 [9–18]	—

Abbreviations: ALS = amyotrophic lateral sclerosis, BID = twice daily, OCD = obsessive–compulsive disorder, QD = once daily, SD = single dose, SMA = spinal muscular atrophy.

2.4. Modeling Software

PBPK modeling and simulation were performed using PK‐Sim v11 as part of the Open Systems Pharmacology Suite (www.open‐systems‐pharmacology.org). The whole‐body structure includes 15 solid organs and two blood organs connected by physiological blood flows. Each solid organ is further divided into plasma, red blood cell, interstitial, and cellular sub‐compartments. The rate and extent of drug distribution across sub‐compartments are described with permeability‐surface area products and partition coefficients. Oral absorption is modeled using the advanced compartmental and transit framework [19]. The foundational anatomical and physiological parameters for virtual individuals in this study are derived from the International Commission on Radiological Protection database, 2002 [20].

2.5. Riluzole Model Development

Table 2 presents the physicochemical and ADME properties of riluzole that were used to construct the model.

TABLE 2.

Physicochemical parameters of riluzole used for model building.

Parameter	Value used in model	References
Physicochemical properties
Molecular weight	234.198 g/mol	Product information, Sanofi
Halogens	F: 3	Product information, Sanofi
Acid–base status	pKa 3.8 (base)	Product information, Sanofi
Lipophilicity (logP)	3.29 (3.48 in vitro)	Parameter identification
Fraction unbound, plasma (f u)	0.04 (human)	Product information, Sanofi
Partition coefficient method	Berezhkovskiy	Berezhkovskiy [21]
Cell permeability	0.0685 cm/min	Parameter identification
Water solubility	0.0395 mg/mL	ALOGPS
Tablet, time to 50% dissolution	15 min	Riluzole tablet USP
Intestinal permeability (apical)	0.000239 cm/min	Parameter identification
ADME
f m,total	0.98	98% metabolized 2% unchanged in urine 0% unchanged in feces
f m,CYP1A1	0.60 a Specific clearance: 5.81 1/min Reference concentration: 1 μM	Literature derivation Parameter identification
f m,CYP1A2	0.30 a Specific clearance: 2.79 1/min Reference concentration: 1.8 μM	Literature derivation Parameter identification
f m,UGT1A8	0.05 a Specific clearance: 0.079 1/min Reference concentration: 1 μM	Literature derivation Parameter identification
f m,UGT1A9	0.05 a Specific clearance: 1.67 1/min Reference concentration: 1 μM	Literature derivation Parameter identification
Glomerular filtration rate (GFR) fraction	1.0	Filtration at the maximal rate
Enterohepatic circulation (EHC) fraction	1.0	DDI network default [9]
P‐gp reference concentration	1.41 μM	DDI network default [9]
P‐gp Michaelis Menten constant (K M)	1 mM	Assumption for linear kinetics
P‐gp catalytic rate constant (k cat)	0.9 mM/min	Parameter identification

^a The enzyme was implemented in the model according to the PK‐Sim gene expression database (RT‐PCR profile), and a first‐order clearance rate was optimized to achieve the observed f _m.

CYP1A1, CYP1A2, UGT1A8, UGT1A9, and P‐gp (as ABCB1) were implemented in organs of virtual individuals according to the PK‐Sim gene expression database. CYP1A1 is predominantly expressed in the lung/liver/intestines, CYP1A2 in the liver, UGT1A8 in the kidney/liver/intestines, UGT1A9 in the kidney, and P‐gp in the liver/kidney/intestines. The expression of CYP1A1 in the intestinal mucosa was increased by a factor of 2 based on a review of data in the Human Protein Atlas (www.proteinatlas.org), and the expression of P‐gp in intestinal mucosa was adjusted according to the standard from the PK‐Sim DDI networks as described elsewhere [22].

PK data collected after single intravenous and oral doses of riluzole were used to calibrate the model (Table 1, model development). The riluzole tablet is modeled with a 15 min dissolution half‐time according to the riluzole tablet USP (Tables 1 and 2). The Tiglutik suspension is modeled as “dissolved” upon administration (solution formulation, Table 1). The capsule administered in the study by Abbara et al. [6] is assumed equivalent to the tablet, in the absence of any other information. Other formulations were not modeled.

The lipophilicity, cell permeabilities, the first order specific clearance for each of the metabolic enzymes, and the catalytic rate constant for transport by P‐gp (k _cat) were optimized. Optimization was carried out in the Parameter Identification Tool in PK‐Sim using a Levenberg–Marquardt approach to exploring the parameter space. Six optimizations were run from randomized start values to ensure identifiability of a unique solution. Goodness of fit was assessed with a visual predictive check and a plot of fitted versus observed concentrations.

The model then was expanded from a reference individual to populations of virtual individuals [23]. One virtual population was generated for each real‐world study population being simulated. These virtual subjects have normally and log‐normally distributed inter‐individual variability in height, weight, organ sizes, blood flows, plasma protein concentrations, gastrointestinal transit, glomerular filtration rates, and metabolic enzyme concentrations as informed by literature [23].

2.6. Riluzole Model Verification

The translational potential of the riluzole PBPK model was verified by predicting PK in a series of clinically diverse populations and scenarios, including administration to adult patients with ALS, hepatic impairment, or renal impairment, elderly volunteers (> 70 years), and pediatric patients with spinal muscular atrophy (SMA) (Table 1, model verification). The oral absorption component (including intestinal metabolism) of the model was verified by predicting the food effect following administration with a high fat breakfast in healthy subjects. The anatomical and physiological parameterization of these populations and scenarios is presented in Table 3. Verification was considered acceptable if the fold‐error between predicted and observed parameters fell within strict bioequivalence limits (0.8‐ to 1.25‐fold).

TABLE 3.

Parameterization of various populations and scenarios for verifying the riluzole model.

Population/scenario	Parameterization	References
ALS	Clearance increased by 20% in smokers (comprising 18% of the study population)	Rilutek, prescribing information
	Clearance decreased by 27% in severe neuromuscular disease	Cleary et al. [24]
Hepatic impairment	Hepatic impairment subjects were used as parameterized by Edginton et al. accounting for changes in organ blood flows, functional liver mass, plasma protein binding, and GFR	Edginton et al. [25]
	CYP1A1 and CYP1A2 expression/activity were decreased by 47% in Child‐Pugh A and 77% in Child‐Pugh B	Johnson et al. [26]
Renal impairment	Renal impairment populations used as default from PK‐Sim accounting for changes in kidney volume, renal blood flow, plasma protein binding, hematocrit, gastric emptying time, and intestinal transit time	Malik et al. [27]
	No change to CYP1A1 and CYP1A2 expression/activity in renal impairment	Tan et al. [28]
Pediatric SMA II and SMA III	Pediatric populations used as default from PK‐Sim accounting for changes in organ size, blood flows, body composition, plasma protein binding, GFR, and metabolic enzyme expression/activity	Edginton et al. [29]
	Undersize for age according to study demographics (weight, height, body surface area)	Abbara et al. [17]
	Clearance decreased by 27% in severe neuromuscular disease (pragmatic physiologic proxy reflecting the cumulative effects of severe neuromuscular disease [e.g., reduced organ size, altered plasma protein levels, impaired tissue perfusion])	Cleary et al. [24]
Healthy elderly	Elderly populations used as default from PK‐Sim accounting for changes in organ size, blood flows, body composition, plasma protein binding, GFR	Schlender et al. [30]
	No change to CYP1A1 and CYP1A2 expression/activity in elderly	Schlender et al. [31]
High fat breakfast	High fat breakfast event used as default from PK‐Sim	Thelen et al. [19]

Abbreviation: GFR = glomerular filtration rate.

2.7. Fluvoxamine Model Verification

Fluvoxamine is a strong competitive CYP1A2 inhibitor (K _I = 2.97 nM) and a weak competitive CYP1A1 inhibitor (K _I = 2.45 μM; 825‐fold greater than the potency against CYP1A2) [5, 32]. The precipitant PBPK model for fluvoxamine was downloaded and used with no deviations from the qualified CYP1A2 DDI network in PK‐Sim v11, except that weak inhibition of CYP1A1 was added.

The precipitant effects on the CYP1A2 pathway have been verified against clinical DDI studies with multiple CYP1A2 substrates (e.g., caffeine and mexiletine). A report of these evaluations is available on the Open Systems Pharmacology website (https://github.com/Open‐Systems‐Pharmacology/OSP‐Qualification‐Reports/tree/master/DDI_Qualification_CYP1A2). To estimate inhibitory potency of fluvoxamine for CYP1A1, the model leveraged a published in vitro potency ratio (CYP1A2 K _I = 40 nM; CYP1A1 K _I = 33 μM, PMID: 9105404) scaled from the verified in vivo CYP1A2 K _I (2.97 nM, PK‐Sim) [5, 32]. This indirect inference assumes similar in vitro‐to‐in vivo scaling and is acknowledged as an approximation.

2.8. Evaluation of Riluzole CYP1A Metabolism

The proposed f _m,CYP1A1, f _m,CYP1A2 and the overall sensitivity of riluzole to CYP1A DDIs were confirmed by simulating an observed DDI between riluzole and fluvoxamine. Observed data were available for the PK of riluzole 50 mg twice daily (BID) with fluvoxamine 100 mg BID in children with OCD [14]. Evaluation was considered acceptable if the fold‐error between predicted and observed parameters fell within strict bioequivalence limits (0.8‐ to 1.25‐fold).

2.9. Sensitivity Analysis

A one‐at‐a‐time sensitivity analysis was conducted by varying the competitive inhibition constants (K _I) for fluvoxamine to assess how uncertain the verification may be if there is uncertainty in the inhibitory potency of the precipitant against CYP1A2 and CYP1A1. Simulations were performed after increasing and decreasing the in vitro‐informed K _I values by a factor of 3.16 ($\sqrt{10}$) to cover a 10‐fold uncertainty margin, and the resulting riluzole DDI ratios in these alternative scenarios were reported.

3. Results

3.1. Riluzole Model Development

Inputs to the PBPK model were the physicochemical properties of riluzole and the fractions metabolized, absorbed, and excreted unchanged in urine. The specific enzyme and transporter rates and distribution/permeability were fitted. Four of the six optimization runs converged to the identifiable solution. The fitted riluzole PBPK model captured the observed PK after single intravenous and oral doses in healthy subjects plus its detailed ADME characteristics (including f _m by each enzyme as assigned). Goodness‐of‐fit assessments are presented in Figure 2. All simulations are available in the Supporting Information 1. The intestinal permeability value reflects basolateral membrane transport across tight epithelial junctions and is expected to be lower than overall cellular permeability. Differences in in vitro permeability models and assay conditions can exceed 100‐fold and are not directly comparable to physiologic estimates used in PBPK.

FIGURE 2.

Goodness‐of‐fit assessment for the riluzole model.

3.2. Riluzole Model Verification

The fitted model was verified by predicting the PK of riluzole in several special populations and following administration with a high‐fat meal. The virtual healthy subjects in the PBPK model were substituted with established special populations from the literature, and metabolic enzyme expression/activity was adjusted for each scenario according to in vivo data for other molecules with similar ADME characteristics (Table 3). Prediction success in these challenging scenarios serves as verification of the translational potential of the model. Since organ weights, blood flows, plasma protein binding, and the expression/activity of the metabolizing enzymes (which each bear a unique profile of expression in different organs) are changing differently in each of these scenarios, the ability of the model to capture PK in all of them lends confidence to the assigned metabolic contributions of each enzyme (e.g., CYP vs. UGT, and CYP1A1 vs. CYP1A2).

3.2.1. Elderly

In a study by Le Liboux et al., 18 healthy elderly subjects between 70 and 82 years of age received riluzole 50 mg BID for 5 days [16]. Plasma drug concentrations were measured after the last dose on the morning of Day 5 and analyzed separately by sex. For PBPK simulation, the default elderly population in PK‐Sim was used as published by Schlender et al. [30] While it has been hypothesized that CYP1A enzymes may display some age dependence in expression/activity [3], another modeling example in the elderly population was able to capture the age‐dependent PK of ciprofloxacin, a CYP1A2 substrate through the inherent changes in liver volume, blood flow, and plasma protein binding that occur with age without adjusting CYP1A2 expression/activity [31]. The observed data for riluzole PK in healthy elderly subjects was generally captured by the PBPK simulation, but C _max was modestly underestimated (Figure 3). The mean C _max was 202 ng/mL (predicted) versus 271 ng/mL (observed), and the mean AUC_0–12h,SS was 1179 ng h/mL (predicted) versus 1029 ng h/mL (observed). The verification was accepted on the basis of AUC with minimal error in C _max (fold‐error was 0.75 for C _max).

FIGURE 3.

Riluzole model performance when predicting pharmacokinetics in clinically diverse scenarios.

3.2.2. Pediatric Spinal Muscular Atrophy

In a study by Abbara et al., 14 (13 evaluable) pediatric patients with SMA between 9 and 17 years of age received riluzole 50 mg once daily (QD) for 5 days [17]. Plasma drug concentrations were measured after the last dose on the morning of Day 5. The pediatric subjects were significantly undersize for age (mean age 13 years, weight 9 kg, height ~1.3 m), which is consistent with the progression of this severe neuromuscular disease. For PBPK simulation, a reference pediatric individual with this age and body size was used, and metabolic clearance was decreased by 27% based on the observed difference in risdiplam clearance versus healthy subjects in this same patient population [24]. The observed data for riluzole PK in pediatric patients with SMA was captured by the PBPK simulation (Figure 3). The mean C _max was 437 ng/mL (predicted) versus 359 (observed), and the mean AUC_0–24h,SS was 2350 ng h/mL (predicted) versus 2257 ng h/mL (observed). The verification was accepted on the basis of C _max and AUC.

3.2.3. Amyotrophic Lateral Sclerosis

In a study by Groeneveld et al., 160 patients with ALS received riluzole 50 mg BID to assess the relationship between PK at steady state and overall survival. For PBPK simulation, a virtual population was created between 40 and 75 years of age (31% female), and metabolic clearance was decreased by 27% based on the observed difference in risdiplam clearance versus healthy subjects in a severe neuromuscular disease (SMA) [24]. The observed data for riluzole PK in patients with ALS was well captured by the PBPK simulation (Figure 3). The median C _max was 227 ng/mL (predicted) versus 183 (observed), the median C _min was 49 ng/mL (predicted) versus 54 ng/mL (observed), and the median AUC_0–12h,SS was 1292 ng h/mL (predicted) versus 1473 ng*h/mL (observed). The verification was accepted on the basis of C _max, C _min, and AUC.

3.2.4. Hepatic Impairment

Hepatic impairment often significantly decreases the clearance of CYP1A substrates, such as caffeine, tizanidine, or riociguat. According to the prescribing information for Rilutek, exposures are increased by 1.7‐fold and 3.0‐fold in patients with Child‐Pugh Class A and Child‐Pugh Class B hepatic impairment, respectively. For PBPK simulation of hepatic impairment, virtual subjects with each stage of hepatic impairment (A and B) were created using the methods published by Edginton et al. [25] CYP1A enzyme expression/activity was decreased by 47% in Child‐Pugh A and 77% in Child‐Pugh B based on the caffeine metabolic ratios observed in these disease stages [26]. This operation assumes that CYP1A1 expression/activity is affected by hepatic impairment similarly to CYP1A2; indeed, the reductions in quantitative levels of CYP1A1 and CYP1A2 protein that occur in Child‐Pugh A and Child‐Pugh B liver samples follow very similar trajectories [33]. UGT1A8 and UGT1A9 are not affected by hepatic impairment [33]. After accounting for the changes in organ blood flows, functional liver mass, plasma protein binding, and GFR, PBPK simulations for the disease state very accurately described the geometric mean AUC_inf ratios in Child‐Pugh A and Child‐Pugh B (Figure 3). This verification suggests that the relative contributions of enzymes affected by hepatic impairment versus those not affected are appropriately captured, and that the balance of hepatic versus non‐hepatic metabolism is adequate.

3.2.5. Renal Impairment

Renal impairment does not often affect the clearance of CYP1A substrates, such as caffeine, tizanidine, or riociguat. According to the prescribing information for Rilutek, there is not a meaningful effect of moderate (stage 3) to severe (stage 4) renal impairment on the PK of riluzole. For PBPK simulation of chronic kidney disease (CKD), virtual populations of otherwise healthy subjects with each stage of renal impairment were created using the methods that are now default in PK‐Sim v11 [27]. Previous analyses have concluded that there is not a significant influence of renal impairment on CYP1A or UGT1A expression/activity [28], and so metabolic clearance was unchanged in the model for this disease state. After accounting for the changes in kidney volume, renal blood flow, plasma protein binding, hematocrit, gastric emptying time, and intestinal transit time that occur in CKD, the greatest geometric mean AUC ratio up to stage 5 CKD was 0.72 (Figure 3), which is consistent with the assessment of “no meaningful effect of renal impairment” from the prescribing information. This verification suggests that the relative contributions of enzymes affected by renal impairment versus those not affected are appropriately captured, and that the balance of renal (e.g., UGT1A8/9) versus non‐renal metabolism is adequate.

3.2.6. Food Effect

The effect of a high fat meal on the PK of the Tiglutik solution was simulated in a healthy reference individual. The high‐fat meal event was used with no deviations from the PK‐Sim default. A high‐fat meal increases gastric pH (with resulting changes to charge‐dependent solubility), adds to the liquid content of the stomach, and prolongs gastric emptying time. The non‐fasting:fasting C _max ratio of riluzole was captured within acceptance criteria by the simulation (0.52 predicted vs. 0.45 observed) and the AUC was minimally affected in accordance with the observed data (Figure 3). This verification suggests that the mechanisms underlying oral absorption and bioavailability are sufficiently characterized in the ACAT sub‐model, with intestinal metabolism by CYP1A1 and UGT1A8 present in the mucosa.

3.3. Evaluation of Riluzole CYP1A Metabolism

The final riluzole model was used to simulate the DDI with fluvoxamine, a strong CYP1A2 inhibitor and a weak CYP1A1 inhibitor, in order to confirm the relative contributions of CYP1A enzymes to riluzole metabolism (i.e., 60% for CYP1A1 and 30% for CYP1A2). Observed PK data for evaluation are available from a study by Grant et al. in children with OCD between the ages of 8.8 and 18.4 years of age [14]. Thirty‐seven children received riluzole at a mean dose of 50 mg BID, and six of them additionally received fluvoxamine at an estimated dose of 100 mg BID based on the regimens from previous studies in children by the same author group [34]. The average plasma concentration at steady state (C _avg,SS) was 79 ng/mL compared with 164 ng/mL in those taking fluvoxamine, providing an observed DDI ratio of 2.08. For PBPK simulation, a virtual population was created with the same age and BMI range as the study population. The simulated C _avg,SS in the fluvoxamine group was 177 ng/mL, and in the riluzole only group, it was 100 ng/mL, providing a DDI ratio of 1.77 and falling within the strict acceptance criteria (Table 4). This evaluation result suggests that the relative contributions of CYP1A enzymes to riluzole metabolism are appropriately captured by the final model.

TABLE 4.

Simulated drug–drug interactions between riluzole and fluvoxamine.

Enzyme(s)	Object	Precipitant	C avg,SS ratio
			Base case		Sensitivity analysis
			Predicted	Observed	K I × 3.16	K I/3.16
CYP1A2 CYP1A1	Riluzole 50 mg BID	Fluvoxamine 100 mg BID	1.77	2.08	1.63	2.01

Abbreviations: BID = twice daily, K _I = competitive inhibition constant.

3.4. Sensitivity Analysis

Increasing the inhibitory potency of fluvoxamine toward CYP1A enzymes improved the prediction of the DDI ratio, and decreasing the inhibitory potency lowered the prediction below the acceptance threshold (Table 4). Most modeling exercises of fluvoxamine in CYP1A2‐mediated DDIs have used inhibitory K _I values less than or equal to the present value (2.97 nM; lower K _I = higher inhibitory potency) [9, 32, 35]. Therefore, the f _m,CYP1A2 of riluzole is most likely around or less than the proposed value in this work (30% or less), and CYP1A1 may be the dominant enzyme responsible for riluzole metabolism (60% or more).

4. Discussion

A riluzole PBPK model was built from knowledge of the physicochemical properties of riluzole, an understanding of fractions metabolized, absorbed, and excreted in urine, and virtual populations from PK‐Sim. The model was calibrated on single‐dose data from healthy subjects and was then extrapolated to predict PK in many clinically diverse scenarios, providing verification of its translational potential. Prediction of the DDI with fluvoxamine, a strong CYP1A2 inhibitor and a weak CYP1A1 inhibitor (observed C _avg,SS ratio = 2.08), verified the relative contributions of CYP1A2 and CYP1A1 enzymes to metabolism (predicted C _avg,SS ratio = 1.77). Overall, CYP1A1 was identified to be a major enzyme metabolizing riluzole in contrast to previous literature and drug labeling. This work highlights how legacy assumptions about drug metabolism—particularly for older medicines such as riluzole—can be revisited using modern PBPK approaches. It underscores the need for CYP1A1‐selective tools and supports renewed attention to drugs with ambiguous elimination pathways when used in vulnerable populations.

Unlike CYP1A2, there are few confirmed in vivo inhibitors of CYP1A1, and most are identified based on in vitro data. As such, clinical DDI risk remains poorly characterized and warrants further investigation. Potent CYP1A1 inhibitors in vitro (K _I < 10 μM) include quinidine [6], quinine [6], amiodarone (and metabolite) [36, 37], ketoconazole [36, 38], tamoxifen [36], cyproterone acetate [36], bicalutamide [36], and abacavir [39]. Since riluzole is approximately 60% metabolized by CYP1A1, exposures may increase to a maximum of 2.5‐fold with a strong CYP1A1 inhibitor if the drug does not also modulate CYP1A2. Therapeutic alternatives to these drugs are preferred when available, but they may be used with caution in view of the wide pharmacokinetic variability of riluzole, and the monitorable and reversible side effect profile [40, 41].

Strong CYP1A2 inhibitors may be used with caution if they do not also significantly modulate CYP1A1, such as fluvoxamine [5]. Mexiletine may be a moderate and equipotent inhibitor of both CYP1A1 and CYP1A2, and should be used with caution [42, 43]. While ciprofloxacin is a clinically relevant CYP1A2 inhibitor, its effects on CYP1A1 are not known, and so it should also be used with caution.

Leflunomide [44], rabeprazole [44, 45], lansoprazole [45], and omeprazole [44, 46] are reported to induce CYP1A1 by more than 30‐fold in vitro. Therapeutic alternatives to these drugs are preferred when available, as they may significantly decrease riluzole plasma concentrations. While omeprazole and related drugs have shown strong CYP1A1 induction in vitro, their in vivo effect remains unconfirmed. These results should be interpreted as hypothesis‐generating and are intended to prompt further investigation of CYP1A1 induction potential in clinical settings. More research is needed, perhaps using real‐world sample collection, to identify whether these drugs have a clinically relevant induction effect on riluzole metabolism.

Each verification scenario in this study was selected to probe a distinct aspect of riluzole disposition relevant to the hypothesized contributions of CYP1A1 and CYP1A2. Although these clinical conditions do not isolate enzyme‐specific metabolism, the ability of the model to accurately reproduce observed pharmacokinetics under each condition supports the physiological plausibility of the assigned f _m values. Verification in hepatic impairment suggests that the relative contributions of enzymes affected by hepatic impairment versus those not affected are appropriately captured, and that the balance of hepatic versus non‐hepatic metabolism is adequate. While the activities of CYP1A enzymes are hypothesized as equally affected by hepatic impairment, hepatic impairment as a physiological state disproportionately reduces CYP1A2‐mediated metabolism of riluzole due to its liver‐specific expression and vulnerability to physiological disruption in functional liver mass and liver blood flow, etc., while the state has less impact on the predominantly extrahepatic‐mediated metabolism of riluzole by CYP1A1. Accurate model prediction of riluzole exposure under hepatic impairment conditions therefore indirectly validates the assigned hepatic versus extrahepatic metabolic contributions. Renal impairment does not affect either CYP1A1 or CYP1A2 expression directly, but alters gastrointestinal transit, renal clearance, and plasma protein binding. Model concordance with observed pharmacokinetics under this scenario suggests that the relative contributions of enzymes affected by renal impairment versus those not affected are appropriately captured, and that the balance of renal (e.g., UGT1A8/9) versus non‐renal metabolism is adequate. Fed/fasted simulations provide a means to assess appropriate translational parameterization of absorption and intestinal metabolism. CYP1A1 is expressed in intestinal mucosa, whereas CYP1A2 is primarily hepatic. Replicating the observed fed‐state increase in riluzole exposure lends further support to the intestinal localization and activity of CYP1A1 in the model. Taken together, these verification scenarios provide orthogonal challenges to the PBPK model and support the translational plausibility of the proposed enzyme contributions, despite the absence of direct enzyme‐selective clinical probes.

The model file with all simulations and the parameter identification will be available on Github and is supplied in the Supporting Information 2; it can be used to simulate CYP1A DDIs with potential new drugs that could become a standard‐of‐care for ALS.

This model was not informed by in vitro f _m assignments but rather by mass balance data and clinical metabolite recovery studies, supported by in vitro inhibition data with appropriate caveats. A limitation to this modeling work is that the inhibitory potency of fluvoxamine against CYP1A1 could not be validated by predicting an observed DDI with a sensitive substrate of CYP1A1 since no data are available for such a substrate. Further, the DDI data between riluzole and fluvoxamine used for evaluation carries some uncertainty; the DDI ratio was calculated using a linear regression of 14 steady state PK samples in six subjects who were taking fluvoxamine with riluzole and 75 samples in 31 subjects who were taking riluzole alone [14]. The enzymatic contributions to riluzole metabolism should be reconsidered after formal drug–drug interaction studies are completed. Finally, it should be noted that the optimized intestinal permeability (~4 × 10E‐06 cm/s) is substantially lower than the Caco‐2 P_app reported for riluzole (85.5 × 10E‐06 cm/s). This discrepancy is expected, as in vitro monolayers generally overestimate in vivo permeability due to looser tight junctions, lack of mucus, and absence of luminal flow.

Author Contributions

P.Ma., P.Mi., J.A., M.R., and S.A.‐D. wrote the manuscript; P.Ma., P.Mi., J.A., M.R., and S.A.‐D. designed the research; P.Ma. and P.Mi. performed the research; P.Ma. analyzed the data.

Conflicts of Interest

P.Ma. is a full‐time employee of Ionis and may hold Ionis stock or options. P.Mi. has no conflicts of interest to declare. J.A. has received research funding to their institution from Alexion, AZTherapies, Amylyx, Biogen, Cytokinetics, Orion, Novartis, MGH Foundation, Ra Pharma, Biohaven, Clene, and Prilenia and consulting fees from AL‐S. Affinia, Amylyx, Apellis, Biogen, Cytokinetics, Denali, Orphazyme, NeuroSense, Novartis, UCB, and Wave Life Sciences. M.R. is a full‐time employee of AbbVie and may hold AbbVie stock or options. S.A.‐D. served on an advisory board for Amylyx and Biogen, received honorarium from MDA, and received research support from Biogen, Amylyx, Alnylam, and MT pharma.

Supporting information

Supporting Information 1: cts70358‐sup‐0001‐Supplementary Materials 1.docx.

Supporting Information 2: cts70358‐sup‐0002‐Supplementary Materials 2.

Malik P., Mian P., Andrews J., Rosebraugh M., and Ajroud‐Driss S., “A Modeling Investigation of the CYP1A Drug Interactions of Riluzole,” Clinical and Translational Science 18, no. 9 (2025): e70358, 10.1111/cts.70358.