Quantitative Systems Toxicology Model Predicts Obeticholic Acid‐Associated Liver Injury in Metabolic Dysfunction‐Associated Steatotic Liver Disease

Abstract

Obeticholic acid (OCA), a synthetic analog of chenodeoxycholic acid, was approved in 2016 for the treatment of primary biliary cholangitis. Early clinical trials revealed elevated liver biomarkers in healthy subjects receiving supratherapeutic OCA doses (100–250 mg). OCA was also evaluated as a treatment for metabolic dysfunction‐associated steatotic liver disease (MASLD) but was not approved by the FDA due to liver safety concerns. In this in silico study, we investigated mechanisms of OCA‐associated liver injury in virtual healthy and MASLD populations receiving supratherapeutic and therapeutic (10–25 mg) doses, respectively. OCA and metabolite exposures in plasma, sinusoidal blood, liver, and gut compartments were simulated using a physiologically based pharmacokinetic model. In the virtual MASLD population, exposures were increased 2‐, 5‐, and 10‐fold in plasma, sinusoidal, and/or liver compartments relative to baseline. Mechanistic parameters relevant to OCA‐mediated liver injury, including bile acid transporter inhibition and mitochondrial dysfunction, were incorporated into the DILIsym model. Predicted liver injury was reported as evaluation of drug‐induced serious hepatotoxicity (eDISH) plots, and elevations in alanine aminotransferase, aspartate aminotransferase, and total hepatic bile acids. DILIsym simulations recapitulated liver biomarker elevations observed at supratherapeutic OCA doses in healthy subjects and predicted biomarker increases in the MASLD population under conditions of 5‐ and 10‐fold increased exposures relevant to this population. Bile acid transporter inhibition alone reproduced simulated biomarker elevations, whereas mitochondrial uncoupling alone predicted increased biomarkers only at the highest exposures. Results suggest that DILIsym modeling would have predicted the liver safety concerns that led to withdrawal of OCA from the US market.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Obeticholic acid (OCA) was approved for primary biliary cholangitis but withdrawn in 2025 due to liver safety concerns. Clinical studies demonstrated elevations in liver injury biomarkers at supratherapeutic doses in some healthy subjects and at therapeutic exposures in some patients with metabolic dysfunction‐associated steatotic liver disease (MASLD). The mechanisms underlying these hepatotoxic effects have not been fully elucidated.

• WHAT QUESTION DID THIS STUDY ADDRESS?

A quantitative systems toxicology (QST) modeling approach using DILIsym was applied in virtual healthy and MASLD populations to recapitulate OCA‐associated hepatotoxicity and identify mechanistic drivers, such as bile acid transporter inhibition or mitochondrial dysfunction, across different exposure scenarios.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

Simulations reproduced observed biomarker elevations and revealed that bile acid transporter inhibition alone could explain most simulated liver injury, whereas mitochondrial dysfunction contributed only at high exposures.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

This study with OCA demonstrates how integrating QST modeling early in development could prospectively predict hepatotoxic risk, reduce costly late‐stage failures, guide mechanistic investigations, and enhance overall drug safety.

Obeticholic acid (OCA) was approved in 2016 as a second‐line therapeutic for primary biliary cholangitis (PBC). ¹ , ² , ³ , ⁴ OCA is a synthetic bile acid analog of the endogenous bile acid chenodeoxycholic acid (CDCA), a well‐known farnesoid X receptor (FXR) agonist. ⁴ , ⁵ , ⁶ FXR is a nuclear receptor that regulates bile acid synthesis and the expression of hepatic transporters. ⁴ FXR activation decreases overall bile acid synthesis, reduces the expression and function of hepatocyte bile acid uptake transporters, and increases the expression and function of bile acid efflux transporters on both the basolateral and canalicular membranes. ⁴ , ⁷ , ⁸ These effects lower the overall concentration of bile acids in hepatocytes in an effort to reduce hepatocyte and bile duct damage. ⁴ , ⁷ , ⁸ OCA is approximately 100 times more potent as an FXR agonist than endogenous CDCA. ⁴ , ⁶ , ⁹ Similar to CDCA, there are two major metabolites of OCA in humans, the glycine conjugate (glyco‐OCA) and the taurine conjugate (tauro‐OCA), which are the predominant forms found in both the systemic circulation and hepatocytes. ⁷ , ¹⁰

OCA has several documented side effects, including pruritus, cholelithiasis, and dyslipidemia, as well as a risk of liver toxicity in some subjects. ⁹ , ¹¹ Of note, adverse hepatic events were observed in healthy subjects during early‐phase clinical trials at supratherapeutic doses of 100 and 250 mg daily. ¹² , ¹³ In clinical trials that enrolled patients with PBC who had an inadequate therapeutic response to the first‐line therapeutic, ursodeoxycholic acid (UDCA), OCA reduced biomarkers of liver injury, including alanine aminotransferase (ALT), alkaline phosphatase (ALP), and γ‐glutamyl transferase (GGT), and improved liver histology. ¹⁴ , ¹⁵ OCA was administered to PBC patients orally once daily at doses ranging from 5 to 50 mg, with side effects occurring most frequently at higher doses. ¹⁴ , ¹⁵ Multiple instances of severe hepatic adverse events in the PBC population led to two black box warnings for OCA. The first black box warning in 2018 emphasized proper dosing to reduce the risk of adverse hepatic events. In 2021, a second warning contraindicated the use of OCA in PBC patients with decompensated cirrhosis or advanced liver disease due to increased hepatotoxic risk. ¹⁶ In 2025, OCA was withdrawn from the US market due to concerns over severe liver injury. ¹⁷

In addition to PBC, OCA was investigated for treatment of metabolic dysfunction‐associated steatotic liver disease (MASLD) at doses of 10 and 25 mg daily. MASLD encompasses several stages of liver disease and is a highly prevalent condition, with more than one‐third of the world's population estimated to have some degree of MASLD. ¹⁸ , ¹⁹ This condition is predicted to affect over half of the world's population by 2040. ¹⁹ There are limited therapeutic options for the treatment of MASLD, with only one medication approved for metabolic dysfunction‐associated steatohepatitis (MASH), the most severe form of MASLD. ²⁰ Despite the need for medications to treat patients with MASLD, and promising initial efficacy in clinical trials, OCA was not approved for this indication due to concerns over liver safety. ²¹

Early clinical trials showed that some patients with MASLD exhibited markedly increased total OCA (OCA and metabolites) concentrations compared to concentrations in healthy subjects, with individual plasma concentrations reaching total OCA concentrations up to 15‐fold higher. ²² On average, systemic total OCA exposure, measured as area under the curve (AUC) and maximum concentrations (C_max), increased approximately 2‐ to 5‐fold in clinical trial patients with MASLD and MASH, albeit with substantial inter‐subject variability. ²² Initial trial results with OCA were promising and met several efficacy endpoints for MASLD. ¹¹ , ²² Liver exposure was not measured via liver biopsy in healthy subjects, and liver biopsy samples in the MASLD population were taken at different times during the course of treatment. Thus, average concentrations in liver biopsies from patients with MASLD are difficult to compare to total OCA liver concentrations predicted for healthy subjects from the physiologically based pharmacokinetic (PBPK) model. However, liver exposure in patients administered therapeutic doses tended to increase as MASLD progressed, and the mean liver exposure in patients with MASLD was predicted to increase compared to the liver exposure predicted based on the PBPK model for healthy subjects. ⁷ , ²² On average, the increase in total OCA exposure in the liver from the early stages of MASLD to MASH was approximately 2‐fold based on dose‐normalized liver biopsy trough concentrations. ²² In general, liver exposure to total OCA was predicted to increase approximately 2‐fold in subjects with hepatic impairment, irrespective of disease etiology. ⁷

Quantitative systems toxicology (QST) modeling is a maturing approach for predicting drug‐induced liver injury (DILI). DILIsym is a well‐established QST modeling platform that integrates hepatic biochemistry, compound pharmacokinetics, and interactions with key hepatotoxic mechanisms, including reactive oxygen species (ROS) production, mitochondrial dysfunction, and bile acid transporter inhibition. Inhibition of bile acid transporters is a well‐characterized mechanism of DILI. ²³ , ²⁴ , ²⁵ Inhibition of efflux transporters such as the canalicular bile salt export pump (BSEP) and the basolateral multidrug resistance‐associated proteins (MRP) 3 and 4 may cause hepatocyte retention of bile acids, bilirubin, and/or xenobiotics, leading to hepatocyte damage. ²³ In contrast, inhibition of uptake transporters such as Na⁺‐taurocholate co‐transporting polypeptide (NTCP) and organic anion transporting polypeptides (OATPs) may decrease bile acid accumulation within hepatocytes and lead to increased plasma bile acid concentrations. ²³ DILIsym modeling also utilizes simulated virtual populations (SimPops) of healthy subjects and patients representing a subset of full MASH pathophysiology, both incorporating intersubject variability in hepatotoxic pathways. These simulations enable the prediction of DILI incidence and the identification of risk factors contributing to simulated hepatotoxicity. DILIsym simulations can also provide insights into DILI susceptibility within specific patient populations, as well as the effects of alternative dosing regimens on liver safety. ²⁴ , ²⁵

This in silico study assessed whether DILIsym simulations could reproduce the increases in hepatotoxic biomarkers observed in healthy subjects at supratherapeutic doses of OCA (>100 mg daily) and whether liver injury liability could have been predicted at therapeutic doses in the MASLD patient population. In addition, simulations were conducted to determine the respective contributions of two OCA‐relevant mechanisms of DILI: bile acid transporter inhibition and mitochondrial uncoupling.

MATERIALS AND METHODS

Concentration vs. time data for OCA were simulated using a previously developed PBPK model transcribed in MATLAB version R2024a. ⁷ The parameters included in the PBPK model are listed in Table S1 . Concentrations of OCA and the two major metabolites, glyco‐OCA and tauro‐OCA, at daily doses ranging from 10 to 250 mg OCA for a 12‐day period, were simulated in plasma, sinusoidal blood, and liver, and total OCA (parent OCA and OCA metabolites) in the gut lumen (Figure S2a ). Interindividual variability in OCA exposures was not included in the current work.

Similarly, concentration vs. time profiles of OCA and OCA metabolites in virtual patients with MASLD were simulated as uniform 2‐, 5‐, or 10‐fold increases in the baseline exposure profile simulated for healthy subjects in the plasma, sinusoidal, and liver compartments, to determine the predicted impact of increased exposures on the MASLD population (Figure S2b–d ). Since the amount of total OCA in the gut lumen was primarily dependent on the oral dose of OCA, this value only changed as a function of dose. In additional simulations, the exposures were increased 5‐fold in the plasma and sinusoidal compartments, and 2‐fold in the liver compartment (Figure S2e ). This was based on the relatively smaller increases in total OCA exposure in the liver compartment predicted in patients with hepatic impairment, and the observed exposure in liver biopsies in patients with MASLD compared to the simulated exposure in healthy subjects. ⁷ , ²²

OCA and metabolite exposure data, along with in vitro parameters for bile acid transporter inhibition (e.g., IC₅₀ values) ¹² and mitochondrial dysfunction (Table 1 ), were imported into a modified version of DILIsym v8A that incorporated the extended bile acid representation available in DILIsym v11. ROS formation was not included as a potential mechanism of OCA‐associated liver injury because OCA was not shown to increase ROS; in some cases, OCA decreased ROS. ²⁶ Endogenous hepatic bile acid transport processes in DILIsym are primarily represented by the bile acid uptake transporter NTCP, canalicular BSEP, and a basolateral efflux transporter modeled to reflect the activity of MRP3 or 4 (Figure 1 ). ¹² Since the OCA and metabolite IC₅₀ values for MRP3 were lower than those for MRP4, the IC₅₀ values for MRP3 were used to represent inhibition of basolateral efflux as a conservative approach for the prediction of hepatotoxicity. An IC₅₀ value was also included for the apical sodium‐dependent bile acid transporter (ASBT), which is a transporter found in the intestine that is responsible for the uptake of bile acids and other substrates. ¹² , ²⁷ Inhibition of bile acid transporters by OCA and metabolites was parameterized as competitive inhibition using the mixed inhibition equation with a large α value. Furthermore, IC₅₀ values for OATP1B1, MRP3, and MRP2 were used to represent inhibition of the uptake, basolateral efflux, and biliary efflux of bilirubin, respectively, as part of the bilirubin disposition representation in DILIsym. ²⁸

Table 1.

Select input parameters for the DILIsym mechanistic toxicity model for the parent compound obeticholic acid (OCA), and two major metabolites, glycine conjugated OCA (glyco‐OCA) and taurine conjugated OCA (tauro‐OCA) ¹²

Mechanism	DILIsym parameter	Unit	OCA	Glyco‐OCA	Tauro‐OCA
Bile acid transporter inhibition 12	NTCP IC50	μM	7.04	6.83	4.99
	BSEP IC50	μM	13.7	6.9	10.5
	MRP3 IC50	μM	5.64	14.6	51
	ASBT IC50	μM	43.1	N/A	N/A
Mitochondrial dysfunction 29	Uncoupler effect Km	μM	1468.4	1468.4	1468.4
	Uncoupler effect Hill coefficient	–	2.25	2.25	2.25
	Uncoupler effect V max	–	196.24	196.24	196.24
Bilirubin transporter inhibition 12	OATP1B1 IC50	μM	2.57	4.05	3.01
	MRP2 IC50	μM	69.4	126	205
	MRP3 IC50	μM	5.64	14.6	51

–, Dimensionless.

ASBT, apical sodium‐dependent bile acid transporter; BSEP, bile salt export pump; IC₅₀, half‐maximal inhibitory concentration; K _m, Michaelis–Menten constant; MRP, multidrug resistance‐associated protein; N/A, not applicable; NTCP, Na⁺‐taurocholate co‐transporting polypeptide; OATP, organic anion transporting polypeptide; V _max, maximal velocity.

Figure 1.

Key input parameters for the DILIsym model. (a) Bile acid transporters (highlighted with red boxes) and bilirubin transporters (indicated in bold) that regulate hepatocyte uptake and efflux of substrates. (b) Representation of mitochondrial uncoupling showing loss of mitochondrial membrane potential leading to decreased ATP production. NTCP, Na⁺‐taurocholate co‐transporting polypeptide; OATPs, organic anion transporting polypeptides; MRP2, 3/4, multidrug resistance‐associated protein 2, 3, or 4, respectively; BSEP, bile salt export pump.

Mitochondrial dysfunction was assumed to be due to mitochondrial uncoupling, based on the similarity between OCA and CDCA (Figure 1 ). ²⁹ CDCA is a known mitochondrial uncoupler, and the values for K _m, V _max, and the Hill coefficient for CDCA were used for OCA and metabolites. Both the bile acid transport inhibition and mitochondrial dysfunction mechanisms were used in all simulations unless noted otherwise.

DILIsym simulated populations, hereafter referred to as SimPops, used in this study were the healthy subject population (Human_ROS_apop_mito_BA_v11A_1, n = 285), and the MASLD patient population (Human_NAFLD_ROS_apop_mito_BA_v11A_4, n = 263) representing patients with varying disease severity, but without fibrosis. Both SimPops incorporated intersubject variability in susceptibility to hepatotoxicity mechanisms (e.g., mitochondrial function disruption, bile acid transport inhibition), while the MASLD SimPops additionally accounted for disease‐related variability in plasma glucose, plasma free fatty acids, lipogenesis, bile acid synthesis, liver glutathione, triglycerides, and lipotoxicity. ²⁵ , ³⁰ , ³¹ , ³² , ³³ The model also included variability in transporter function as a range of V _max values representing bile acid uptake and efflux. Each virtual patient was assigned a value within the defined ranges of parameters for MASLD, therefore incorporating interindividual variability. ³⁰ All simulations used daily oral doses of OCA for 12 days. Simulated data outputs from DILIsym included elevations in ALT and total bilirubin (TBL), including evaluation of drug‐induced serious hepatotoxicity (eDISH) plots showing peak ALT and TBL, as well as elevations in aspartate aminotransferase (AST) and endogenous hepatic bile acid concentrations. Simulated data from the virtual healthy subjects were compared to clinical trial data where ALT and AST were elevated more than two times the upper limit of normal (>2×ULN). The healthy subject SimPops had ALT and AST levels of 30 U/L at baseline, with an ULN of 40 U/L. However, the MASLD virtual population had a wide range of baseline ALT and AST values. Therefore, biomarker data were analyzed and presented as the percentage of individuals with ALT and AST values increased 2‐fold or more from baseline. Data were processed and analyzed using RStudio version 4.4.1, and figures were created using the built‐in DILIsym software or GraphPad Prism 10. RStudio was used to implement a simulated treatment‐discontinuation protocol in which OCA dosing was stopped when ALT exceeded 2× the individual baseline (30 U/L for healthy subjects, variable for virtual patients with MASLD). AST elevations were evaluated at scheduled 24‐hour sampling intervals up to the time of dosing cessation.

RESULTS

ALT and AST elevations in healthy subjects are dependent on OCA dose

The eDISH plots generated by DILIsym in the healthy SimPops administered once daily OCA doses of 10 and 25 mg (Figure S3 ) and 50 mg (Figure 2 ) revealed that all subjects remained close to the baseline ALT and TBL values. Of note, all of the virtual healthy subjects started with the same baseline ALT and TBL values, so data points overlapped. However, at supratherapeutic doses of OCA (100 and 250 mg), elevations in the hepatotoxicity biomarkers were predicted (Figure 2 ). eDISH plots for these supratherapeutic doses showed increases in the number of subjects reaching the >2× ULN and >2× baseline thresholds. This trend matched dose‐dependent findings from clinical trials involving small numbers of healthy subjects (between 7 and 16 healthy subjects in each dosing group, with up to 50% of subjects exhibiting increased ALT and AST in the 250 mg dose group). ¹² The percentage of virtual subjects predicted to exhibit ALT and AST elevations of >2× ULN and >2× baseline is summarized in tabular format in Figure 2 , compared to the observation of hepatotoxicity in healthy subjects at each dose level in the clinic. ¹²

Figure 2.

Evaluation of drug‐induced serious hepatotoxicity (eDISH) plots depicting peak alanine aminotransferase (ALT) and peak total bilirubin (TBL) elevations predicted from the DILIsym model in healthy subjects (n = 285) following once‐daily administration of obeticholic acid (OCA) 10–250 mg (10 and 25 mg plots are shown in Figure S3 ) over the course of 12 days. Baseline ALT and TBL are fixed for virtual healthy subjects; therefore, some points are overlapped. The percentage of subjects with elevations in hepatotoxicity biomarkers (ALT and aspartate aminotransferase (AST)) in virtual healthy subjects (n = 285) based on the OCA dose is summarized in tabular format and compared to clinical observations of hepatotoxicity over this range of doses. ULN, upper limit of normal. *Simulated AST elevations were evaluated at scheduled 24‐hour sampling intervals up to the simulated time at which OCA administration would have been discontinued because ALT exceeded a predefined stopping threshold of either 2× ULN (80 U/L) or 2× individual baseline (60 U/L).

Biomarker elevations in the MASLD population depend on OCA dose and exposure

eDISH plots based on simulations in the virtual MASLD population using total OCA exposures typical of healthy subjects revealed a similar trend to healthy subjects, with significant increases in biomarkers from baseline occurring only at supratherapeutic OCA doses (Figure 3 a ). The DILIsym model predicted greater susceptibility to liver injury when exposures to total OCA increased 2‐ or 5‐fold (Figure S4b,c ) or 10‐fold (Figure 3 b and Figure S4d ) in the plasma, sinusoidal, and liver compartments. When exposures were increased 5‐fold in plasma and sinusoidal compartments and 2‐fold in the liver (Figure 3 c ), the percentage of virtual patients with predicted ALT elevations >2× baseline was higher than in the MASLD population when OCA exposures were typical of healthy subjects (Figure 3 a ), but lower than simulations where the plasma, sinusoidal, and liver concentrations were increased 2‐fold in the virtual MASLD population (Figure S4b ). After a 5‐fold increase in exposures across the plasma, sinusoidal, and liver compartments, the DILIsym model predicted biomarker elevations starting at 25 mg (Figure S4c ), and after a 10‐fold increase in exposures, the DILIsym model predicted elevations in liver biomarkers at all doses tested, from 10 to 250 mg in the virtual MASLD population (Figure 3 b and Figure S4d ).

Figure 3.

Evaluation of drug‐induced serious hepatotoxicity (eDISH) plots depicting peak alanine aminotransferase (ALT) and peak total bilirubin (TBL) elevations achieved in virtual patients with metabolic dysfunction‐associated steatotic liver disease (MASLD; (n = 263)) following once‐daily administration of obeticholic acid (OCA) at doses ranging from 10–250 mg over the course of 12 days. Simulated exposure to total OCA (OCA and OCA metabolites) were (a) typical of healthy subjects, (b) increased 10‐fold in plasma, sinusoidal, and liver compartments, and (c) increased 5‐fold in plasma and sinusoidal, and 2‐fold in liver compartments. The percentage of virtual patients with MASLD (n = 263) exhibiting elevations in hepatotoxicity biomarkers (ALT and aspartate aminotransferase (AST)) based on the OCA dose and simulated exposure scenario is summarized in tabular format. ULN, upper limit of normal. *Simulated AST elevations were evaluated at scheduled 24‐hour sampling intervals up to the simulated time at which OCA administration would have been discontinued because ALT exceeded 2× the individual patient baseline; in MASLD patients, this ALT threshold was patient‐specific.

Transporter inhibition drives most biomarker elevations in both populations

To examine potential mechanisms of hepatotoxicity, transporter inhibition and mitochondrial uncoupling were simulated independently using the 250‐mg daily OCA dosing protocol. Nearly all ALT and AST elevations were attributable to bile acid transporter inhibition in the virtual healthy subjects (Figure 4 a ). This was also observed in the virtual MASLD population at total OCA exposures typical of healthy subjects (Figure S5a ), when total OCA exposures increased 2‐fold (Figure S5b ), and when total OCA exposures increased 5‐fold in the plasma and sinusoidal compartments, and 2‐fold in the liver compartment (Figure 4 b ). Mitochondrial dysfunction alone did not reproduce ALT or AST elevations at these OCA and metabolite exposures. However, at high exposures (5‐ and 10‐fold increase) in the virtual MASLD population, mitochondrial dysfunction alone was sufficient to produce elevations in ALT and AST (Figure 4 c and Figure S5c ). In addition to biomarkers of liver hepatotoxicity, simulated total average endogenous hepatic bile acid concentrations were increased in virtual patients with MASLD after OCA exposure, especially at high doses and/or high exposures (Figure 5 and Figure S6 ). Simulated total average endogenous hepatic bile acid concentrations also increased in virtual healthy subjects at high OCA doses (Figure S6a ). In the absence of a stop protocol, several patients were simulated to die during the course of the 12‐day treatment.

Figure 4.

Evaluation of drug‐induced serious hepatotoxicity (eDISH) plots generated for simulations conducted with only one parameterized DILIsym mechanism of hepatotoxicity (mitochondrial uncoupling or bile acid transporter inhibition). eDISH plots depict peak alanine aminotransferase (ALT) and peak total bilirubin (TBL) elevations achieved following once‐daily administration of 250 mg obeticholic acid (OCA) over the course of 12 days. eDISH plots simulated for (a) virtual healthy subjects (n = 285), (b) virtual patients with metabolic dysfunction‐associated steatotic liver disease (MASLD, n = 263) where total OCA (OCA and OCA metabolites) exposures typical of healthy subjects were increased 5‐fold in plasma and sinusoidal, and 2‐fold in liver compartments, and (c) virtual patients with MASLD where total OCA exposures were increased 10‐fold in plasma, sinusoidal and liver compartments. The percentage of individuals with elevations in hepatotoxicity biomarkers (ALT and aspartate aminotransferase (AST)) in virtual healthy subjects and patients with MASLD based on the mechanism of hepatotoxicity and exposure are summarized in tabular format. ULN, upper limit of normal. *Simulated AST elevations were evaluated at scheduled 24‐hour sampling intervals up to the simulated time at which OCA administration would have been discontinued because ALT exceeded 2× the individual patient baseline; in MASLD patients, this ALT threshold was patient‐specific.

Figure 5.

Simulated total endogenous hepatic bile acid concentrations after once‐daily administration of obeticholic acid (OCA) (10–250 mg) over the course of 12 days in virtual patients with metabolic dysfunction‐associated steatotic liver disease (MASLD) (n = 263). Simulated exposure to total OCA (OCA and OCA metabolites) was increased from the exposure simulated for healthy subjects by (a) 5‐fold in plasma and sinusoidal, and 2‐fold in liver compartments, and (b) 10‐fold in plasma, sinusoidal, and liver compartments. In the absence of a stop protocol, several virtual patients did not survive the 12‐day dosing period (represented by a line that ends before day 12).

DISCUSSION

This QST modeling study used DILIsym to reproduce clinically observed OCA‐associated liver injury in healthy subjects and patients with MASLD, and to investigate potential underlying mechanisms of hepatotoxicity. The model successfully recapitulated reported elevations in ALT and AST in healthy subjects at supratherapeutic OCA doses while predicting no clinically meaningful biomarker elevations at lower doses (Figure 2 and Figure S3 ). These findings are consistent with clinical observations showing no ALT or AST elevations >2× ULN in healthy subjects at doses up to 50 mg, but increased incidence of biomarker elevations at supratherapeutic doses of 100 and 250 mg. ¹²

Surprisingly, DILIsym simulations indicated that virtual patients with MASLD did not exhibit increased susceptibility to liver injury relative to healthy subjects when total OCA exposures were comparable between populations (Figure 3 a and Figure S4a ). Under these conditions, dose‐dependent increases in ALT and AST occurred only at supratherapeutic doses, mirroring the pattern observed in healthy subjects. However, clinical studies in patients with MASLD have demonstrated increased systemic exposure to OCA and OCA metabolites and predicted increases in hepatic exposure. ⁷ , ²² When total OCA exposures were increased 2‐fold in simulations of the virtual MASLD population, ALT elevations emerged at the 50 mg dose group (Figure 3 and Figure S4b ). At 5‐ and 10‐fold increases in exposures, ALT and AST elevations were predicted at therapeutic doses evaluated in MASLD clinical trials (Figure 3 and Figure S4c,d ). These findings support the conclusion that increased exposures, rather than disease‐related susceptibility alone, drive the heightened risk of liver injury in patients with MASLD.

Simulations in which plasma and sinusoidal OCA exposures were increased 5‐fold while hepatic intracellular exposure increased only 2‐fold (Figure 3 c ) were conducted to reflect observed disparities between circulating and hepatic OCA concentrations. Under these conditions, DILIsym predicted lower ALT and AST elevations than in simulations with a uniform 2‐fold exposure increase across all compartments. This is explained by stronger NTCP inhibition at higher sinusoidal OCA concentrations, which reduces endogenous bile acid uptake and limits intracellular bile acid accumulation, which is the principal driver of toxicity in the current OCA representation. In contrast, when sinusoidal and hepatic exposures increase uniformly by 2‐fold, NTCP inhibition is less pronounced, allowing greater bile acid influx and higher intracellular bile acid concentrations. Thus, reduced toxicity in the 5‐fold sinusoidal, 2‐fold hepatic scenario is attributable to decreased hepatic bile acid accumulation.

Mechanistic simulations demonstrated that inhibition of bile acid transporters accounted for the majority of predicted ALT and AST elevations at the 250 mg OCA dose in both virtual healthy subjects and virtual patients with MASLD (Figure 4 and Figure  S5 ). In contrast, mitochondrial dysfunction alone did not produce meaningful biomarker elevations except at extreme OCA exposures. Consistent with this mechanism, simulations predicted increased hepatic total endogenous bile acid concentrations in both healthy subjects and patients with MASLD, particularly at high doses or elevated exposures (Figure 5 and Figure S6 ). Endogenous bile acid accumulation in hepatocytes leads to hepatocyte death or injury and declining ATP levels, which impact overall transport activity of both endogenous bile acids and OCA metabolites. Although inhibition of the bile acid uptake transporter NTCP can partially mitigate intracellular bile acid accumulation, the model suggests that this effect is insufficient to counterbalance OCA‐mediated inhibition of bile acid efflux under the exposure conditions examined.

A limitation of this study is the lack of available IC₅₀ data for multidrug resistance protein 3 (MDR3), the phospholipid floppase, which restricts the ability of DILIsym to predict cholestatic patterns of liver injury. This limitation is relevant given that some cases of OCA‐associated hepatotoxicity in clinical settings exhibited cholestatic features, including elevations in ALP and GGT with ALT/ALP ratios ≤2. ³⁴ Despite this limitation, the model predicted hepatocellular injury at higher OCA exposures, indicating that bile acid‐mediated toxicity can be recapitulated even in the absence of detailed MDR3 inhibition parameters. An additional limitation is the sparsity and timing variability of liver biopsy data in patients with MASLD, which complicates direct comparisons with PBPK‐predicted hepatic concentrations in healthy subjects.

In summary, DILIsym simulations reproduced ALT and AST elevations observed in healthy subjects at supratherapeutic OCA doses and predicted clinically relevant biomarker elevations in patients with MASLD at therapeutic doses under conditions of increased total OCA exposures. These findings suggest that QST modeling could have anticipated liver injury risk following early clinical studies in populations with hepatic impairment or MASLD. Earlier identification of such risks may have reduced the time, cost, and patient burden associated with late‐stage clinical development of OCA. Although OCA has since been withdrawn from the US market, this work highlights the value of early screening for bile acid transporter inhibition, mitochondrial dysfunction, and other mechanisms of DILI to inform risk assessment and guide decision‐making in drug development. Future DILI assessment using a QST modeling approach, such as DILIsym, should be conducted proactively when data are available for small‐molecule novel compounds. In addition to small molecules, novel therapeutics that are structurally similar to bile acids, including synthetic bile acids, would be good candidates for DILIsym modeling, given the demonstrated risk with these compounds. These types of molecules may have complex interactions with transporters, and systemic and liver exposures may vary significantly, especially in patients with liver disease. This type of modeling may strengthen confidence in liver safety or show early warning signs for novel compounds when sufficient pharmacokinetic and in vitro data are available. ²⁴ , ²⁵ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ In the case of OCA, DILIsym modeling after the data from pharmacokinetic studies in healthy subjects were available could have predicted liver injury liability when OCA exposure was increased in patients with MASLD.

FUNDING

This work was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number R35 GM122576.

CONFLICT OF INTEREST

Drs. Beaudoin and Woodhead are employees of Simulations Plus, Inc and hold stock ownership and/or stock options in the company. Dr. Watkins worked in the remote past as a paid consultant for Intercept. All other authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

A.K.M., J.J.B., and K.L.R.B. wrote the manuscript; A.K.M., J.J.B., P.B.W., and K.L.R.B. designed the research; A.K.M. performed the research; A.K.M. and J.J.B. analyzed the data; J.L.W. contributed new reagents/analytical tools.

Supporting information

Table S1.

Figure S3.

Figure S6.