Inhibition of OATP1B1/3 Rather Than UGT1A1 May Be the Major Cause of the Bilirubin Elevation After Atazanavir Administration

Abstract

Atazanavir has been reported to increase total serum bilirubin level up to ninefold. It is widely believed that the observed total bilirubin elevation is primarily due to UGT1A1 inhibition. However, UGT enzymes are well‐known as a low‐affinity and high‐capacity system, and the observed drug–drug interaction mediated by UGTs is usually less than twofold. There were discrepancies in the explanation of total bilirubin elevation due to UGT1A1 inhibition alone, suggesting the contribution of other mechanism(s) to the interaction. As atazanavir is a potent OATP1B1/3 inhibitor and the hepatic uptake of both unconjugated and conjugated bilirubin are mediated by OATP1B1/3, these transporters could be involved in the bilirubin–atazanavir interaction. To better understand the roles of UGT1A1 and OATP1B1/3 in this interaction, it would be useful to characterize the contribution of each individual pathway to the interaction. As multiple compounds, pathways, and potentially UGT1A1 polymorphism are involved, a thorough physiologically‐based pharmacokinetic (PBPK) analysis was utilized to integrate the information from various relevant in vitro and clinical studies to quantitatively estimate the contribution of UGT1A1 and OATP1B1/3 inhibition to the interaction between bilirubin and atazanavir. The PBPK analysis indicated that UGT1A1 inhibition plays a modest role in bilirubin and atazanavir interaction contributing less than 33%. The results also suggested that unconjugated bilirubin is less sensitive than raltegravir upon UGT1A1 inhibition, therefore, unconjugated bilirubin may not be a useful endogenous biomarker for UGT1A1 inhibition. The analysis demonstrated that the metabolism of unconjugated bilirubin shares common features of other UGT enzyme‐mediated reactions.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Atazanavir has been reported to increase total bilirubin levels up to ninefold. It is widely believed that the observed total bilirubin elevation is primarily due to UGT1A1 inhibition. Pharmacogenetics Implementation Consortium (CPIC) recommended prescribing atazanavir based on UGT1A1 genetic status and considering an alternative agent for PM. However, the UGT enzyme system is well‐known for its low‐affinity and high‐capacity. The clinically observed drug–drug interaction mediated by UGT enzymes is generally less than twofold.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Is bilirubin a special UGT1A1 substrate with ultra‐sensitivity upon UGT1A1 inhibition? How much do UGT1A1 and OATP1B1/3 pathways contribute to the bilirubin–atazanavir interaction? Is unconjugated bilirubin a sensitive and useful endogenous biomarker of UGT1A1 inhibition?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

UGT1A1 plays a secondary role in bilirubin and atazanavir interaction as the contribution of UGT1A1 inhibition to the interaction is less than 33%. Unconjugated bilirubin has less sensitivity than raltegravir upon UGT1A1 inhibition, whereas the sensitivity of conjugated bilirubin may be comparable to that of pravastatin and greater than that of simvastatin acid upon OATP1B1/3 inhibition.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Bilirubin metabolism might not be an exception contradicting the common feature of reactions mediated by UGT enzyme systems as low‐affinity and high‐capacity systems. Unconjugated bilirubin may not be a useful endogenous biomarker for UGT1A1 inhibition. It was also demonstrated that PBPK analysis is a useful tool to decipher complex interactions involving multiple compounds and pathways.

Enzymes of the uridine diphosphate glucuronosyltransferase (UGT) 1A and 2B sub‐families contribute the most to glucuronidation of certain drugs (e.g., dolutegravir, raltegravir, SN‐38) and endogenous compounds (e.g., unconjugated bilirubin). They are the second most important metabolic pathways for small‐molecule drugs. ¹ UGT enzymes are typically known to be low‐affinity and high‐capacity. Their drug–drug interaction (DDI) risk is generally considered to be low according to the International Council for Harmonization. ² The observed magnitude of DDI shown as the ratio of the area under the curve (AUC) with and without concomitant administration of UGT inhibitors is usually less than twofold. ¹ , ³

Total serum bilirubin mostly consists of unconjugated bilirubin with less than 5% present as conjugated bilirubin in healthy subjects. ⁴ Unconjugated bilirubin is metabolized solely by UGT1A1 to conjugated bilirubin after hepatic uptake by OATP1B1/3 (Figure 1 a ). ¹ Conjugated bilirubin is a mixture of bilirubin mono‐ (15%) and di‐glucuronide (85%), ⁵ and is primarily eliminated by biliary excretion mediated by MRP2. It is also a substrate of OATP1B1/3 as well as MRP3. OATP1B1/3 in downstream hepatocytes may re‐uptake conjugated bilirubin refluxed by MRP3 to serum back to hepatocytes. Total bilirubin level may be altered by drugs that inhibit UGT1A1 and/or OATP1B1/3, ¹ , ⁶ as well as polymorphism of UGT1A1, OATP1B1/3, and MRP2. Individuals with decreased activities in these enzymes/transporters may have Gilbert's syndrome (decreased UGT1A1 activity), Rotor syndrome (simultaneous absence of OATP1B1 and OATP1B3), or Dubin‐Johnson syndrome (absence or deficiency of MRP2). ⁷

Figure 1.

(a) Diagrams of unconjugated and conjugated bilirubin disposition. U‐BIL, unconjugated bilirubin; C‐BIL, conjugated bilirubin; UGT, UDP‐glucuronosyltransferases; OATP, organic‐anion‐transporting polypeptides; MRP, multidrug resistance‐associated protein. (b) Overall PBPK modeling strategy. (c) Overview of PBPK modeling validation workflow. [Correction added on 04 June 2025, after first online publication: Figure 1A has been corrected in this version.]

Atazanavir is an HIV‐1 protease inhibitor that is prescribed as 400 mg once daily (q.d.) without a pharmacokinetic (PK) booster or, more commonly, as 300 mg q.d. with either ritonavir (100 mg) or cobicistat (150 mg) as boosters. Auto‐inhibition is observed in the PK of atazanavir as it is both a substrate and a mechanism‐based inhibitor of CYP3A4/5. The commonly co‐prescribed boosters, ritonavir and cobicistat, are more potent mechanism‐based CYP3A4/5 inhibitors. Atazanavir's AUC at steady state is ~90% higher when it is co‐administered with ritonavir or cobicistat than when administered alone. Atazanavir is also a well‐known UGT1A1 inhibitor. It was reported that the AUCs of dolutegravir (primarily metabolized by UGT1A1 and a minor substrate of CYP3A4, a substrate of P‐gp and BCRP ³ ) and raltegravir (mainly metabolized by UGT1A1 ³ ) were increased by ~90% ⁸ and ~70%, ⁹ respectively, when co‐administered with atazanavir alone. Furthermore, atazanavir is an inhibitor of P‐gp, BCRP, and OATP1B1/3. The AUC of atorvastatin (a substrate of CYP3A4, P‐gp, BCRP, and OATP1B1/3) was increased by ~800% when co‐administered with atazanavir/cobicistat (atazanavir/c). ¹⁰

Atazanavir is generally safe and well‐tolerated. However, total bilirubin elevation is common after atazanavir administration and is manifested in virtually every patient treated with atazanavir/ritonavir (atazanavir/r). Grade ≥3 and 4 bilirubin elevation occurred with a frequency of ~40% and ~4–8%, respectively, with atazanavir/r treatment. ¹¹ Up to ~9‐fold total bilirubin increase was reported when administering 300 mg atazanavir alone twice daily (b.d.). ¹² Anti‐HIV regimens containing atazanavir/r were downgraded to alternative status based on a large comparative trial showing a greater rate of toxicity‐related (primarily hyperbilirubinemia) discontinuation than those with either darunavir/ritonavir (darunavir/r) or raltegravir. ¹¹ As the risk for bilirubin‐related discontinuation is highest among UGT1A1 poor metabolizers (PM), the Clinical Pharmacogenetics Implementation Consortium (CPIC) recommended prescribing atazanavir based on UGT1A1 genetic status and considering an alternative agent for PM. ¹¹

The hyperbilirubinemia after atazanavir administration does not indicate liver injury. ¹¹ It is widely believed that the observed total bilirubin elevation is primarily due to UGT1A1 inhibition as it is mainly driven by the elevation of unconjugated bilirubin level. ¹ Indeed, the percentage of conjugated bilirubin in total bilirubin was reduced in atazanavir/r treatments. ¹³ The observed large bilirubin elevation in atazanavir treatments also suggested that bilirubin appears to be more sensitive to UGT1A1 inhibition than other substrates, even though UGT1A1 is generally a low‐affinity and high‐capacity enzyme.

Because multiple steps and pathways are involved in bilirubin disposition, it was suggested to interpret the observed changes in bilirubin levels cautiously. ¹⁴ In fact, discrepancies exist in observed bilirubin interactions if they were attributed to only UGT1A1 inhibition. For example, sorafenib is one of the strongest UGT1A1 inhibitors, which increased the AUC of SN‐38 as a UGT1A1 substrate 1.67–2.20‐fold. ¹ As sorafenib and atazanavir have comparable inhibitory potency (Appendix A , Table A2), therefore, it is expected that the frequency of hyperbilirubinemia would be similar in patients taking sorafenib and the ones taking atazanavir. However, less than 10% of patients with differentiated thyroid carcinoma treated with sorafenib had grades 1 and 2 bilirubin elevations and none had grades 3 and 4 bilirubin elevations, ¹⁵ whereas ~40% of patients treated with atazanavir had grade 3 or higher bilirubin elevations. ¹¹ Also, Peer et al. ¹⁶ reported minimal total bilirubin elevation in UGT1A1 extensive metabolizers (EM) and ~0.1 mg/dL elevation in PM after sorafenib treatment, which was much lower than those observed in atazanavir treatments.

Furthermore, total bilirubin elevation is atazanavir exposure dependent, whereas the inhibition effect of atazanavir on dolutegravir is not (Appendix A , Table A3). Both dolutegravir and total bilirubin exposure were measured in a dolutegravir DDI study with atazanavir. ⁸ The AUC ratio of dolutegravir was 1.91 (90% confidence interval (CI): 1.80–2.03) with atazanavir alone (400 mg q.d.) and 1.62 (90% CI: 1.50–1.74) with atazanavir/r (300 mg/100 mg q.d.), although the atazanavir exposure was higher after atazanavir/r administration. The larger magnitude of dolutegravir DDI with atazanavir alone compared with atazanavir/r was believed to be due to a UGT1A1 induction effect by ritonavir. ⁸ If total bilirubin is more sensitive to the change in UGT1A1 activity, the change in total bilirubin would be expected to reflect a more pronounced induction effect by ritonavir than that observed with the change in dolutegravir exposure. However, the increase in total bilirubin was of greater magnitude with atazanavir/r than with atazanavir alone (~7.8‐fold vs. 3.5‐fold) without showing any UGT1A1 induction effect by ritonavir. ⁸

As mentioned above, atazanavir is a potent inhibitor of OATP1B1/3 as well as UGT1A1 whereas both unconjugated and conjugated bilirubin are OATP1B1/3 substrates. It is possible that OATP1B1/3 inhibition plays a more important role than UGT1A1 inhibition in the bilirubin elevation. To better assess UGT1A1 as an enzyme system with low‐affinity and high‐capacity, it is warranted to understand the role of UGT1A1 in the bilirubin and atazanavir interaction. The bilirubin–atazanavir interaction is complex due to the involvement of multiple compounds (unconjugated and conjugated bilirubin, atazanavir, ritonavir, or cobicistat), pathways (CYP3A4/5, UGT1A1, and OATP1B1/3), and potentially UGT1A1 polymorphism as indicated by the CPIC. PBPK modeling was implemented as a tool to quantitatively understand the contribution of UGT1A1 and OATP1B1/3 to bilirubin elevation with atazanavir treatment. The main aims of this study were to (i) determine the contribution of UGT1A1 and OATP1B1/3 inhibition to bilirubin elevation with atazanavir treatment; (ii) compare the sensitivity of bilirubin upon UGT1A1 or OATP1B1/3 inhibition with those of other sensitive substrates.

METHODS

Software

PBPK modeling was performed with Simcyp (version 21, Simcyp Limited, Sheffield UK). Graphics were produced in R (version 4.2.0, The R Foundation for Statistical Computing, Vienna, Austria). Data digitization from literature was performed with Engauge Digitizer version 12.1.

Overall modeling strategy

The overall PBPK modeling strategy is shown in Figure 1 b,c (see modeling details in the appendices). First, individual PBPK models of unconjugated and conjugated bilirubin, and interacting drugs were developed and validated. Briefly, unconjugated and conjugated bilirubin models (Appendix B ) were developed and validated. OATP1B1/3 kinetics were validated with clinical interaction with rifampicin and cyclosporine, and UGT1A1 kinetics were validated with the observed baseline levels in subjects with UGT1A1 variants. The atazanavir model (Appendix C ) was modified from the built‐in Simcyp model to predict its exposure at a steady state after multiple dose administration of atazanavir alone, atazanavir/r, or atazanavir/c. The inhibition parameters of atazanavir against UGT1A1 in the built‐in Simcyp model were maintained the same in the modified model. Atazanavir inhibition effect on UGT1A1 was validated with clinical data of the DDI with raltegravir. Atazanavir inhibition effect on OATP1B1/3 was optimized with clinical DDI data of atorvastatin after atazanavir/c administration, assisted by darunavir and cobicistat models. The darunavir model (Appendix D ) was constructed based on publications. ¹⁷ , ¹⁸ The cobicistat model (Appendix E ) was developed and validated with its clinical PK and the DDI with midazolam. The OATP1B1/3 inhibition effects of darunavir and cobicistat were validated with clinical data of the DDI with atorvastatin. The ritonavir model (Appendix F ) was modified from the built‐in Simcyp models. Induction of UGT1A1 and OATP1B1/3 inhibition by ritonavir were validated with the clinical DDI data with raltegravir and pravastatin, respectively.

Second, the established models in the first step were further validated with observed total, unconjugated, and/or conjugated bilirubin elevations after administration of atazanavir alone, atazanavir/r, or atazanavir/c in healthy subjects or subjects with UGT1A1 variants.

Finally, the validated models were applied to (i) assess the contribution of UGT1A1 and OATP1B1/3 pathways to bilirubin elevation after atazanavir administration; (ii) compare the sensitivity of bilirubin upon UGT1A1 or OATP1B1/3 inhibition with other sensitive substrates.

Assessment of the contribution of UGT1A1 and OATP1B1/3 pathways to bilirubin elevation by atazanavir administration

The bilirubin–atazanavir (with or without ritonavir) interactions were simulated without either UGT1A1 or OATP1B1/3 inhibition in UGT1A1 EM, intermediate metabolizers (IM), and PM. The simulation results were then compared with the “control” results obtained by incorporating all interacting pathways.

As UGT1A1 induction was incorporated into the validated ritonavir model, bilirubin and atazanavir/r interaction were simulated with and without a UGT1A1 induction effect by ritonavir in UGT1A1 EM, IM, and PM to assess the contribution of incorporating the induction effect.

Comparison of the sensitivity of bilirubin with raltegravir and statins upon UGT1A1 or OATP1B1/3 inhibition

The sensitivities of unconjugated and total bilirubin and raltegravir upon UGT1A1 inhibition were compared by simulating their interactions with atazanavir alone (400 mg q.d.). UGT1A1 inhibition was included as the only interacting pathway, whereas all other inhibition effects by atazanavir were ignored except the CYP3A4‐mediated auto‐inhibition. The CYP3A4 auto‐inhibition on the PK of atazanavir itself was kept in the model to simulate the steady‐state exposure of atazanavir without affecting the PK of bilirubin or raltegravir.

Similarly, the sensitivity of unconjugated, conjugated, and total bilirubin, atorvastatin, pravastatin, and simvastatin acid upon OATP1B1/3 inhibition was compared by simulating their interactions with rifampicin (600 mg single oral dose). OATP1B1/3 inhibition was included as the only interacting pathway whereas all other inhibition effects by rifampicin were ignored.

Model evaluation

The predictive performance of each model was evaluated by overlaying the observed mean concentration–time data with the model predicted profile (mean and 90% prediction interval). The PK metrics assessed were the AUC from zero to infinity or at steady state and peak plasma concentration (C _max). The predictive performance of PK metrics (AUC and C _max), and DDI ratios (AUC and C _max with/without perpetrator) was assessed based on a predefined criterion of twofold range of the predicted/observed ratio. The population arithmetic or geometric mean of the concentration–time profiles and geometric mean of PK metrics were predicted and compared with the corresponding observed data.

RESULTS

Bilirubin model validation

There was good agreement between the simulated and observed unconjugated, conjugated, and total bilirubin AUC and C _max change after rifampicin or cyclosporine administration. The observed total and conjugated bilirubin concentration–time profiles were within the 90% prediction interval (Figure 2 a,b ; Appendix B , Figure B1 for additional rifampicin dose levels). The majority of the simulated vs. observed AUC and C _max ratios were within 0.8–1.25 (Figure 2 c,d ).

Figure 2.

Validation of the bilirubin PBPK model. Plasma concentration–time profiles of total (a) and conjugated (b) bilirubin with and without 600 mg oral administration of 600 mg of rifampicin (RIF) (Appendix B , Figure B1 for additional rifampicin dose levels). Closed circles are observed in mean plasma concentration–time profiles of total bilirubin ((a); blue, baseline; pink, with rifampicin) and conjugated bilirubin ((b); green, baseline; yellow, with rifampicin) from the study reported by Mori et al. ²⁵ The continuous line represents the simulated mean plasma concentration–time profiles of total bilirubin ((a); blue, baseline; pink, with rifampicin) and conjugated bilirubin ((b); green, baseline; yellow, with rifampicin); the shaded areas represent the simulated 5th and 95th percentiles of the simulations. Results were from the virtual subjects with simulated total bilirubin baseline levels of ≤1.3 mg/dL. ³³ Summary of simulated vs. observed geometric means ratios of AUC (c) and C _max (d) for total (triangle), unconjugated (square), and conjugated (circle) bilirubin in the presence of rifampicin (blue, 150 mg; yellow, 300 mg; red, 600 mg) or cyclosporine (green, 20 mg; gray 75 mg). The observed AUCs of unconjugated bilirubin in the absence and presence of rifampicin were calculated by deducting the observed AUCs of conjugated bilirubin from total bilirubin with the data obtained in the corresponding administration. Then, the AUC ratios of unconjugated bilirubin were calculated with the corresponding values. The solid line is the line of unity; dashed lines represent the interval of 0.8–1.25; the shaded area represents the interval between 0.5 and 2. (e) Simulated (sim.) vs. observed (obs.) mean total bilirubin concentration in UGT1A1 EM, IM, and PM. Results shown in (a), (b), (c), and (d) were from virtual subjects with simulated total bilirubin baseline levels less than 1.3 mg/dL as the study was performed in healthy subjects. ³³ Results shown in (e) were from virtual subjects with simulated total bilirubin baseline levels of ≤1.3, 1.8, and 3.4 mg/dL for EM, IM, and PM, ¹⁹ , ³³ respectively. The calculated fold of elevation of total, unconjugated, and/or conjugated bilirubin were similar between the results with or without excluding virtual subjects with total bilirubin greater than the cutoff values (Appendix B , Table B3, B4, and B5). AUC, area under the curve; C _max, maximum plasma concentration; EM, extensive metabolizer; IM, intermediate metabolizer; PM, poor metabolizer; obs., observed; sim., simulated.

In addition, the developed bilirubin model reasonably captured the observed total bilirubin baseline levels in UGT1A1 EM, IM, and PM (Figure 2 e ).

Atazanavir model validation

The validation results of the atazanavir, darunavir, cobicistat, and ritonavir models were presented in Appendices [Link] , [Link] , [Link] , and F , respectively. The following paragraphs briefly describe the validation of the atazanavir model with its PK in subjects with CYP3A5 variants, the validation of its inhibition against CYP3A4 and UGT1A1, and the optimization of the inhibition constant (K _i) of atazanavir against OATP1B1/3.

The atazanavir model predicted the PK metrics of atazanavir within 0.8–1.25‐fold of the observed data obtained after administering atazanavir alone and atazanavir/r in both CYP3A5 expressors and non‐expressors (Figure 3 a,b ; Appendix C , Table C2, C3). There was good agreement between the simulated and observed exposure of atazanavir after its administration with various dosing regimens and boosters (Figure 3 c,d ). Also, the majority of simulated vs. observed AUC and C _max ratios of clarithromycin (CYP3A4 substrate and inhibitor) and drospirenone (CYP3A4 substrate) with atazanavir as a CYP3A4 inhibitor were within 0.8–1.25 (Figure 3 e,f ; Appendix C , Table C5).

Figure 3.

Validation of the atazanavir PBPK model. Plasma concentration–time profiles of atazanavir after administration of atazanavir alone ((a), 400 mg q.d.) or atazanavir/r ((b), 300 mg/100 mg q.d.) in CYP3A5 expressors and non‐expressors. Closed circles are observed geometric mean plasma concentration–time profile of atazanavir. The continuous line represents the simulated geometric mean plasma concentration–time profile of atazanavir; the shaded area represents the 90% prediction intervals. Pink, expressor; blue, non‐expressor. Summary of simulated vs. observed AUC (c) and C _max (d) of atazanavir after atazanavir administration with various dosing regimens (Appendix C , Table C2, C3, and C4). (e) and (f) Summary of simulated vs. observed geometric mean ratios of AUC and C _max of clarithromycin, raltegravir, and drospirenone when administered in the presence and absence of atazanavir or atazanavir/r (Appendix C , Table C5). (g) Simulated vs. observed geometric mean ratio of AUC of atorvastatin in the presence and absence of atazanavir/c (300 mg/150 mg q.d.) or darunavir/c (800 mg/150 mg). in vitro K _i, in vitro K _i of atazanavir, darunavir, and cobicistat against OATP1B1/3; K _i of atazanavir optimized, K _i of atazanavir against OATP1B1/3 was optimized; CYP3A4 inh only, simulated without K _i of darunavir or cobicistat against OATP1B1/3 and assumed the interaction was by CYP3A4 inhibition only. In (c), (d), (e), and (f), the solid line is the line of unity; dashed lines represent the interval of 0.8–1.25; the shaded area represents the interval between 0.5 and 2. b.d., twice daily; q.d., once daily; SD, single dose.

The inhibition effect of atazanavir against UGT1A1 was validated with clinical DDI data with raltegravir after multiple doses of atazanavir alone or atazanavir/r administration (Figure 3 e,f ; Appendix C , Table C5). All predicted DDI ratios were within twofold of observed values, and most were within 0.8–1.25‐fold.

The atazanavir model underpredicted the atorvastatin AUC increase after atazanavir/c co‐administration when simulated using the in vitro K _i of atazanavir and cobicistat against OATP1B1/3 (Figure 3 g , simulated vs. observed, 5.89 vs. 9.2). This could be due to the underestimation of the OATP1B1/3 inhibition by atazanavir alone, cobicistat alone, or both. However, the simulation using the in vitro K _i of darunavir and cobicistat against OATP1B1/3 well predicted the AUC ratio of atorvastatin (Figure 3 g , simulated vs. observed, 3.93 vs. 3.90). This indicated that underprediction of the DDI between atorvastatin and atazanavir/c was not caused by the in vitro K _i of cobicistat but rather was primarily due to the underestimation of the OATP1B1/3 inhibition by atazanavir. Therefore, the in vitro K _i of atazanavir against OATP1B1/3 was optimized. The atorvastatin AUC increase due to co‐administration with atazanavir/c was well predicted (Figure 3 g , simulated vs. observed, 10.5 vs. 9.22) with the optimized K _i of atazanavir and the in vitro values of cobicistat.

Atorvastatin is a substrate of CYP3A4, P‐gp, BCRP, and OATP1B1/3, whereas atazanavir is an inhibitor of CYP3A4, P‐gp, BCRP, and OATP1B1/3. The minimal contributions of P‐gp and BCRP pathways to atorvastatin–atazanavir DDI are discussed in Appendix C .

Validation of the bilirubin and atazanavir/darunavir interaction

The developed PBPK models reasonably recovered the observed total/unconjugated bilirubin elevation due to various atazanavir dosing regimens including atazanavir (400 mg q.d. or 300 mg b.d.), atazanavir/r (300 mg/100 mg), and atazanavir/c (300 mg/150 mg) (Table 1 ).

Table 1.

Simulated total or unconjugated bilirubin elevations compared with observed clinical data from studies with atazanavir or darunavir administration

Protease tnhibitor		Booster		Total or unconjugated bilirubin elevationa		References
Drug	Dose	Drug	Dose	Observed	Simulatedb
Atazanavir	400 mg q.d.	—	—	3.54‐foldc	4.07‐fold	8
	300 mg q.d.	Ritonavir	100 mg q.d.	~4.25‐fold	4.93‐fold	21
	300 mg q.d.	Ritonavir	100 mg q.d.	~5.14‐fold	4.78‐fold	28
	300 mg q.d.	Ritonavir	100 mg q.d.	7.85‐foldc	6.21‐fold	8
	300 mg q.d.	Ritonavir	100 mg q.d.	~2.07 mg/dLd	2.38 mg/dL	29
	300 mg q.d.	Cobicistat	150 mg q.d.	~2.73 mg/dLd	2.77 mg/dL
	300 mg b.d. for 3 days	—	—	~9.14‐folde	6.10‐folde	12
	300 mg b.d. for 4 days	—	—	5.70‐foldf	5.56‐fold	30
	300 mg b.d. for 7 days	—	—	4.48 mg/dL (SD 2.15 mg/dL)	3.66 mg/dL (SD 2.87 mg/dL)	31
Darunavir	800 mg q.d.	Ritonavir	100 mg q.d.	~1–1.1‐fold	0.99‐fold	28
	600 mg b.d.	Ritonavir	100 mg b.d.	Comparableg	0.98‐fold	32
	800 mg q.d.	Cobicistat	150 mg q.d.		1.09‐fold

b.d., twice daily; q.d., once daily.

^a Shaded rows indicate unconjugated bilirubin elevation; unshaded rows, total bilirubin elevation. Fold changes are the ratios of the mean of the average steady‐state concentration (C _ss,avg) of total bilirubin after atazanavir or darunavir administration in each virtual subject to the mean of their baseline levels, unless specified otherwise. Similarly, the concentrations of total or unconjugated bilirubin are means of the C _ss,avg in each virtual subject, unless specified otherwise.

^b Results were from virtual subjects with simulated total bilirubin baseline levels of ≤1.3 mg/dL, ³³ unless specified otherwise.

^c The highest mean total bilirubin level was reported. To be consistent with the report, the simulated average highest total bilirubin concentrations with atazanavir administration were calculated and compared with the simulated baseline values.

^d The observed unconjugated bilirubin concentration after treatment was the mean of the mean unconjugated bilirubin concentrations of weeks 2, 24, and 48. The mean unconjugated bilirubin levels were slightly higher in subjects receiving atazanavir/c treatment than in those receiving atazanavir/r treatment. However, the difference was statistically significant only at week 2.

^e Trough concentration of total bilirubin after day 3 of atazanavir administration was compared with the baseline level.

^f According to Dorresteijn et al. ³⁰ per the inclusion criterion, total bilirubin baseline levels of the participants were ≤0.88 mg/dL. Subjects with total bilirubin concentrations of ≥4.68 mg/dL after 2 days of treatment were excluded from the study. The inclusion and exclusion criteria were reflected in the simulation when validating the model with the observed total bilirubin elevation reported by Dorresteijn et al. ³⁰

^g Total bilirubin levels in subjects receiving 600 mg darunavir +100 mg ritonavir b.d. and in subjects receiving 800 mg darunavir +150 mg cobicistat q.d. were 0.75 and 0.70 mg/dL, respectively, whereas baseline values in the two treatment groups were unknown.

Clinically, darunavir/r or darunavir/cobicistat (darunavir/c) has minimal impact (<10%) on bilirubin level. The models predicted the absence of an effect of darunavir administration on bilirubin levels (Table 1 ).

Validation of bilirubin and atazanavir interaction in subjects with UGT1A1 variants

Clinical data of total bilirubin elevation after atazanavir/r administration in UGT1A1 EM, IM, and PM were used to validate the predictability of the developed PBPK models. The models reasonably simulated the total bilirubin concentration in UGT1A1 EM, IM, and PM (Figure 4 a , simulated 2.1, 2.5, 3.6 mg/dL vs. observed 2.1, 2.6, 5.2 mg/dL).

Figure 4.

(a) Summary of simulated vs. observed median total bilirubin levels in UGT1A1 EM, IM, and PM after atazanavir/r administration (300 mg/100 mg q.d.). The error bars represent the observed or simulated first (25th percentile, Q1) and third quartile (75th percentile, Q3). (b) Summary of simulated vs. observed mean unconjugated bilirubin increase in UGT1A1 EM, IM, and PM after administration of atazanavir alone (400 mg q.d.) or atazanavir/r (300 mg/100 mg q.d.). The error bars represent the observed or simulated 95% confidence interval (CI). Results shown in (a) and (b) were from virtual subjects with simulated total bilirubin baseline levels of less than 1.3, 1.8, and 3.4 mg/dL for EM, IM, and PM, ¹⁹ , ³³ respectively. The average steady‐state concentration (C _ss,avg) of total or unconjugated bilirubin in the absence and presence of atazanavir administration was simulated. The calculated fold of elevation of total, unconjugated, and/or conjugated bilirubin was similar between the results with or without excluding virtual subjects with total bilirubin greater than the cutoff values (Appendix B , Tables B9 and B10).

In addition, validation was performed with the clinical data of unconjugated bilirubin elevation after atazanavir alone or atazanavir/r administration in healthy subjects with both CYP3A5 and UGT1A1 genotyped. The proportion of CYP3A5 expressor was kept as reported when simulating the clinical trial. The model reasonably simulated unconjugated bilirubin increase with atazanavir alone administration in UGT1A1 EM, IM, and PM (Figure 4 b , simulated 1.10, 1.61, 1.86 mg/dL vs. observed 1.01, 1.23, 2.38 mg/dL). Similarly, the model reasonably simulated unconjugated bilirubin increase with atazanavir/r administration in UGT1A1 EM, IM, and PM (Figure 4 b , simulated 1.87, 2.90, 3.73 mg/dL vs. observed 1.79, 3.04, 5.00 mg/dL).

Validation of conjugated bilirubin elevation after atazanavir/r administration

The concentration of conjugated bilirubin increased twofold, whereas its percentage in total bilirubin decreased by ~56% when healthy subjects were administered with atazanavir/r. ¹³ The simulated 3.1‐fold elevation of conjugated bilirubin and the 39% reduction of percentage of conjugated bilirubin in total bilirubin were consistent with the observed changes.

Contribution of UGT1A1 and OATP1B1/3 pathways to bilirubin elevation after atazanavir administration

In UGT1A1 EM, the simulated percent change of total bilirubin was 149% with atazanavir alone whereas it decreased to 124% and 19% without UGT1A1 or OATP1B1/3 inhibition, respectively (Figure 5 a ). A more pronounced difference was observed in the simulation with atazanavir/r administration. The simulated percent change of total bilirubin was 471% with atazanavir/ritonavir whereas it decreased to 345% and 37% without UGT1A1 or OATP1B1/3 inhibition, respectively (Figure 5 a ). Similar to the simulations in EM, simulations with atazanavir alone or atazanavir/r administration in IM and PM showed that OATP1B1/3 inhibition has a higher impact on the interaction than UGT1A1 inhibition (Figure 5 a ). The simulated contribution of UGT1A1 inhibition to the interaction was less than 33% after atazanavir/r administration in UGT1A1 PM.

Figure 5.

(a) Comparison of the contribution of UGT1A1 and OATP1B1/3 inhibition to the simulated percent change of total bilirubin after administration of atazanavir alone (left, 400 mg q.d.) or atazanavir/r (right, 300 mg/100 mg q.d.). Control, simulated with all interaction pathways; w/o_UGT1A1_inh, simulated without UGT1A1 interaction by atazanavir or atazanavir/r; w/o_OAT1B1/3_inh simulated without OATP1B1/3 inhibition by atazanavir or atazanavir/r. (b) Comparison of simulated percent change in total bilirubin after administration of atazanavir/r (300 mg/100 mg q.d.) with (w/) and without (w/o) UGT1A1 induction effect by ritonavir in UGT1A1 EM, IM, and PM. Results shown in (a), (b), and (c) were calculated from the simulated C _ss,avg of total bilirubin in the absence and presence of atazanavir administration.

In addition, the simulation of the percent change of total bilirubin after atazanavir/r administration only showed a small difference with and without UGT1A1 induction effect by ritonavir (Figure 5 b ). The results reaffirmed the low risk of underestimating the contribution of UGT1A1 inhibition to the interaction.

Comparison of the sensitivities of bilirubin, raltegravir, and statins upon UGT1A1 or OATP1B1/3 inhibition

The sensitivities of unconjugated and total bilirubin as UGT1A1 substrates were compared with that of raltegravir due to UGT1A1 inhibition by atazanavir alone. Unconjugated and total bilirubin (AUC ratios of 1.26 and 1.25) showed less sensitivities than raltegravir (AUC ratio of 1.78) (Figure 6 a ). The lack of sensitivity of total bilirubin upon UGT1A1 inhibition might be expected as ~10% total bilirubin elevation (0.1 mg/dL elevation with an assumed mean total bilirubin baseline level of 0.93 mg/dL ¹⁹ ) was reported in UGT1A1 PM after sorafenib treatment. ¹⁶

Figure 6.

(a) Comparison of AUC ratios of unconjugated bilirubin, total bilirubin, and raltegravir in the presence and absence of atazanavir after administration of atazanavir alone (400 mg q.d.) with only UGT1A1 effect by atazanavir. All other inhibition effects of atazanavir except its auto‐inhibition via CYP3A4 were excluded in the simulation. (b) Comparison of AUC ratios of unconjugated, conjugated, and total bilirubin, atorvastatin, pravastatin, and simvastatin acid in the presence and absence of rifampicin after oral administration of a single dose of rifampicin (600 mg) with only OATP1B1/3 inhibition effect. All other inhibition effects by rifampicin were excluded in the simulation. Results of all AUC ratios shown in (a) and (b) were from virtual subjects with simulated total bilirubin baseline level of ≤1.3 mg/dL. ³³ The horizontal line in the box indicates the median value. The box edges represent the 1st and 3rd quartiles. The whiskers represent the 1st quartile −1.5 × IQR and the 3rd quartile +1.5 × IQR, where IQR is the interquartile range.

The sensitivities of unconjugated, conjugated, and total bilirubin as OATP1B1/3 substrates were compared with those of atorvastatin, pravastatin, and simvastatin acid due to OATP1B1/3 inhibition by rifampicin. Conjugated bilirubin with a simulated AUC ratio of 2.42 showed higher sensitivity than unconjugated and total bilirubin with AUC ratios of 1.55 and 1.61, respectively (Figure 6 b ). The sensitivity of conjugated bilirubin was comparable to that of pravastatin (AUC ratio of 2.54) and greater than that of simvastatin acid (AUC ratio of 1.76) but less than that of atorvastatin (AUC ratio of 3.24) (Figure 6 b ).

DISCUSSION

UGT is well‐known as a low‐affinity and high‐capacity system with low DDI risk. The observed bilirubin–atazanavir interaction is a special case with unexpectedly large magnitude of interaction that is commonly attributed to UGT1A1. In contrast, our PBPK analysis suggested that UGT1A1 plays a less substantial role in the interaction than has been previously presumed, as the simulated contribution of UGT1A1 inhibition to the interaction was less than 33%.

The lack of sensitivity of bilirubin upon UGT1A1 inhibition and the substantial role of OATP1B1/3 inhibition in bilirubin–atazanavir interaction could reasonably explain the difference between atazanavir and sorafenib in the frequency of clinical hyperbilirubinemia, as both drugs are considered as potent UGT1A1 inhibitors, whereas sorafenib is not a potent OATP1B1/3 inhibitor. Similarly, the discrepancy between dolutegravir and bilirubin increase by atazanavir, with and without ritonavir (Appendix A , Table A3), could also be explained by the secondary role of UGT1A1 inhibition in the bilirubin–atazanavir interaction.

The simulation results suggested that there are minor differences in the percent change of total bilirubin in UGT1A1 EM, IM, and PM with either atazanavir alone or atazanavir/r administration (Figure 5 a ). Similarly, total bilirubin after atazanavir/r administration was observed to increase to ~fivefold over the baseline level ¹⁹ in various UGT1A1 phenotypes ²⁰ (elevated 2.1, 2.6, 5.2 mg/dL vs. baseline 0.48, 0.58, 0.93 mg/dL in EM, IM, and PM, respectively). Despite the similar fold of baseline elevation, the absolute total bilirubin level in PM was higher than those in EM and IM. Therefore, the risk for bilirubin‐related discontinuation due to hyperbilirubinemia is highest in UGT1A1 PM. This finding is consistent with the CPIC recommendation that there is no need to avoid prescribing atazanavir/r in EM or IM, whereas an alternative agent is recommended for PM. ¹¹

One of the reasons why the bilirubin and atazanavir interaction has been commonly believed to be mediated by UGT1A1 inhibition is that the total bilirubin elevation was primarily driven by the unconjugated bilirubin elevation, although the high potency of atazanavir inhibition against OATP1B1/3 has been well recognized. ¹ It is expected that the percentage of conjugated bilirubin increases after atazanavir administration if OATP1B1/3 inhibition has a significant contribution to the interaction as conjugated bilirubin is also a substrate of OATP1B1/3. In fact, ~2‐fold elevation of conjugated bilirubin was observed with atazanavir/r, ¹³ , ²¹ indicating that OATP1B1/3 did contribute to the interaction. The observed data showed that the percentage of conjugated bilirubin in total bilirubin decreased. However, this is likely due to a greater increase in unconjugated bilirubin levels caused by the synergistic effect of simultaneous inhibition of both UGT1A1 and OATP1B1/3, as compared with the increase in conjugated bilirubin with inhibition of only OATP1B1/3. As the increased conjugated bilirubin concentration is still much less than 2 mg/dL and less than 20% of total bilirubin, the clinical symptom is shown as hyperbilirubinemia rather than conjugated hyperbilirubinemia in patients treated with atazanavir.

The simulation results suggested that OATP1B1/3 plays a more important role than UGT1A1 with a contribution of no less than 67% in the bilirubin–atazanavir interactions (Figure 5 a ). Similar to UGT1A1, OATP1B1 is polymorphic. However, to the best of our knowledge, there is a lack of clinical polymorphism data of OATP1B1 in bilirubin–atazanavir interaction. Singkham et al. ²² showed that OATP1B1 polymorphism may affect ritonavir but not atazanavir exposure. As the inhibition against OATP1B1 is primarily due to atazanavir and not ritonavir (Appendix C , Table C6), the OATP1B1/3 inhibition effect after atazanavir/r administration is not expected to be significantly altered in subjects with OATP1B1 polymorphism. Xiang et al. reported the impact of OATP1B1 and UGT1A1 polymorphism on the total, unconjugated, and conjugated bilirubin baseline levels. It appears that OATP1B1 polymorphism itself does not cause a significant bilirubin elevation. ²³ The difference in the impact of UGT1A1 and OATP1B1 polymorphism on bilirubin baseline levels might be because unconjugated bilirubin is solely metabolized by UGT1A1, whereas other hepatic transporters might be involved in the uptake of unconjugated and conjugated bilirubin to hepatocytes. Alternative hepatic uptake pathways may provide a compensatory effect in subjects with poor OATP1B1 function, mitigating the effect of OATP1B1 polymorphism. The different impact of OATP1B1 and UGT1A1 polymorphism on bilirubin baseline levels might explain the recommendation of prescribing atazanavir treatment based on a genetic test result of UGT1A1 rather than OATP1B1. Yet, marked elevation of both unconjugated (~4‐fold) and conjugated (~170‐fold) bilirubin has been observed in patients with Rotor syndrome featuring simultaneous deficiency of OATP1B1 and 1B3 but no change in UGT1A1 activity. ²⁴ It is possible that the risk of hyperbilirubinemia is higher in patients who are receiving atazanavir treatment and have a higher baseline level of bilirubin but normal UGT1A1 activity. Thus, further investigation might be warranted.

Unconjugated and/or total bilirubin were proposed as potential endogenous biomarkers of UGT1A1 inhibition. Our simulation results showed that both unconjugated and total bilirubin were less sensitive than raltegravir (Figure 6 a ). Given the low sensitivity of unconjugated and total bilirubin upon UGT1A1 inhibition, they might not be useful endogenous biomarkers of UGT1A1 metabolism. This might be expected, as there was no correlation between the observed minimal bilirubin elevation after sorafenib treatments and the potent in vitro and clinical UGT1A1 inhibition by sorafenib.

Conjugated bilirubin showed good sensitivity upon OATP1B1/3 inhibition with rifampicin co‐administration, although its sensitivity was lower than that of coproporphyrin I (CP I). ²⁵ The purpose of investigating the OATP1B1/3 inhibition potential of a drug is usually to guide its co‐administration strategy with statins. An endogenous biomarker candidate might provide useful information as long as it is more sensitive than the known statins and is able to identify clinically relevant statin DDIs.

Our simulation results indicated that the sensitivity of conjugated bilirubin was comparable to that of pravastatin and greater than that of simvastatin acid, but less than that of atorvastatin (Figure 6 b ). As a conjugated bilirubin assay might be relatively more accessible and is more frequently monitored in clinical trials or drug treatments, assessing its change might be helpful when guiding the co‐administration of a drug with pravastatin or simvastatin in the absence of CP I data.

The simulation results for assessing the contribution of UGT1A1 and OATP1B1/3 inhibition to bilirubin–atazanavir interaction were based on PBPK models that underwent extensive validation with clinical DDI and/or polymorphism data. However, it has been reported that metabolisms mediated by UGT enzymes may follow non‐Michaelis–Menten kinetics. ²⁶ This could be critical to mechanistically explain the features of UGT enzymes as low‐affinity and high‐capacity enzymes, which are different than CYP enzymes. On the other hand, in Simcyp, the metabolisms and interactions mediated by UGT enzymes were modeled with Michaelis–Menten kinetics, in the same way as those mediated by CYP enzymes. To better describe UGT enzymes as low‐affinity and high‐capacity enzymes in the PBPK models, further investigation of their metabolism and interaction kinetics and including a more mechanistic model may be warranted.

In addition, the K _i value of atazanavir against OATP1B1/3 was optimized to predict the observed atorvastatin and atazanavir/c interaction due to the underprediction with in vitro K _i values. It is not uncommon that clinical DDI mediated by OATP1B1/3 is underpredicted with in vitro inhibition parameters. Novel in vitro assays assessing OATP1B1/3 inhibition are being developed, such as pre‐incubation with inhibitors in physiologically relevant sandwich‐cultured primary human hepatocytes. ²⁷ The K _i values from the novel in vitro assays may improve the in vitro‐in vivo extrapolation of DDI mediated by OATP1B1/3. And, CP I has been identified as an endogenous biomarker to assess OATP1B1 inhibition in clinical studies. However, to the best of our knowledge, there is no reported CP I concentration data after atazanavir administration. The OATP1B1/3 inhibition by atazanavir administration can be further validated with clinical CP I data in the presence and absence of atazanavir.

Overall, our PBPK analysis indicated that UGT1A1 may play only a secondary role in the bilirubin–atazanavir interaction, as the contribution of UGT1A1 inhibition to the interaction is less than 33%. The results indicated that unconjugated bilirubin is less sensitive than raltegravir upon UGT1A1 inhibition. Thus, bilirubin is not ultra‐sensitive upon UGT1A1 inhibition. The metabolism of unconjugated bilirubin shares common features with other reactions with UGT enzymes.

FUNDING

This study was funded by AstraZeneca.

CONFLICT OF INTEREST

J.D., P.S., W.T., D.Z., D.W.B., and M.N. are employees of AstraZeneca and may hold stock ownership, options, and/or interests in the company. All other authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

J.D. and M.S.P. wrote the manuscript. J.D. designed the research. J.D., R.E., and D.C. performed the research. J.D., P.S., R.E., D.C., W.T., D.Z., D.W.B., M.N., and M.S.P. analyzed the data.

Supporting information

Appendix A.

Appendix B.

Appendix C.

Appendix D.

Appendix E.

Appendix F.