key: cord-283064-ncyhvkwl
authors: Rowland Yeo, Karen; Zhang, Mian; Pan, Xian; Ban Ke, Alice; Jones, Hannah M; Wesche, David; Almond, Lisa M
title: Impact of disease on plasma and lung exposure of chloroquine, hydroxy‐chloroquine and azithromycin: application of PBPK modelling
date: 2020-06-12
journal: Clin Pharmacol Ther
DOI: 10.1002/cpt.1955
sha: 
doc_id: 283064
cord_uid: ncyhvkwl

We use a mechanistic lung model to demonstrate that accumulation of chloroquine (CQ), hydroxychloroquine (HCQ) and azithromycin (AZ) in the lungs is sensitive to changes in lung pH, a parameter that can be affected in patients with COVID‐19. A reduction in pH from 6.7 to 6 in the lung, as observed in respiratory disease, led to 20‐, 4.0‐ and 2.7‐fold increases in lung exposure of CQ, HCQ and AZ, respectively. Simulations indicated that the relatively high concentrations of CQ and HCQ in lung tissue were sustained long after administration of the drugs had stopped. Patients with COVID‐19 often present with kidney failure. Our simulations indicate that renal impairment (plus lung pH reduction) caused 30‐, 8.0‐ and 3.4‐fold increases in lung exposures for CQ, HCQ and AZ, respectively, with relatively small accompanying increases (20 to 30%) in systemic exposure. Although a number of different dosage regimens were assessed, the purpose of our study was not to provide recommendations for a dosing strategy, but to demonstrate the utility of a PBPK modelling approach to estimate lung concentrations. This, used in conjunction with robust in vitro and clinical data, can help in the assessment of COVID‐19 therapeutics going forward.

Coronavirus Disease of 2019 (COVID-19) has rapidly become a global pandemic, since the outbreak was initially identified in Wuhan, China in December, 2019. The virus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), can infect the lower respiratory tract causing fevers, cough, and pneumonia. As new drug candidates are being investigated for treatment of COVID-19, efforts are being made to repurpose existing antimalarial drugs, as they are readily available, and have a known safety profile.

Specfically, it has been reported that chloroquine (CQ) has been successful in treating SARS-CoV-2 infections in China (1) . In vitro studies have since confirmed that hydroxychloroquine (HCQ), an analog of CQ, is a more potent inhibitor of SARS-CoV-2 (5-to 7-fold) (2) . Given that HCQ also has a more favourable safety profile than CQ during chronic dosing, a clinical study was conducted in France to determine whether HCQ (600 mg daily; 465 mg base) could be a more viable option for COVID-19 treatment (3) . Despite its small sample size, the results of the study showed that treatment with HCQ alone or in combination with azithromycin (AZ) shortened the time to resolution of viral shedding. Based on the results of this study, clinicians in many countries have already begun using these medications in clinical practice, and multiple randomized trials are being initiated.

For effective treatment of viral pneumonia in the lungs, it is expected that the concentration of drug at the site of infection should achieve or exceed EC 90 values for inhibition of SARS-CoV-2. Uptake of drugs into the lung is particularly significant for basic amines, such as CQ, HCQ and AZ, with pK a values greater than 8. Most basic amines, including CQ and HQ, are amphophilic, with a large hydrophobic group and a hydrophilic group that is ionized at physiological pH (4). This makes them susceptible to lysosomal trapping, a mechanism that can lead to significant accumulation of a drug in lungs, or other lysosomal rich organs such as the heart (5). Small changes in the pH of the lung as a consequence of a SARS-CoV-2 infection may, therefore, have a significant impact on the accumulation of CQ, HCQ and AZ in lung tissue or the epithelial lining fluid (ELF) relative to plasma.

This article is protected by copyright. All rights reserved

Most of the published articles on COVID-19 have focused on the lungs as the main organ involved in the disease, while few articles have reported SARS-CoV-2 involvement in other organs, including liver and kidneys, both of which are involved in the metabolism and excretion of CQ, HCQ and AZ. Liver injury and an increased incidence of acute renal injury during SARS-CoV-2 infection have been reported (6, 7). This is likely to add to the complication of deciding on the correct therapeutic dose of CQ, HCQ and AZ in patients with COVID-19 and may lead to an increased risk of adverse drug reactions.

Physiologically-based pharmacokinetic (PBPK) modelling allows integration of drugrelated data (absorption, metabolism, plasma protein binding and physicochemical) with relevant physiology, to simulate profiles of drug in plasma as well as other organs and tissues, including the site of action (8). Thus, changes in physiology as a consequence of disease can be assessed using PBPK modelling, including renal (9) and hepatic impairment (10).

Two recent publications have applied PBPK modelling to predict the exposure of HCQ and/or CQ in the lung (2, 11) . In the first of these publications, Yao et al., (2020), applied a lung to plasma partition coefficient (Kp) of 541 for HCQ based on rat data and then compared the significantly enhanced lung exposure with an in vitro derived EC 50 . Arnold and Buckner (2020) reported on the limitations of the former study, mainly focusing on the source of the Kp value and the fact that the lung concentrations were referenced against an EC 50 rather than an EC 90 . The authors assessed the impact of a range of Kp values (including other rat studies -Kp of 220) on the lung exposure for a number of different dosing regimens for HCQ and concluded that improved PK models were required for HCQ in order to have more confidence in the efficacy of HCQ against SARS-

We use previously published PBPK models for CQ, HCQ and AZ, and integrate a more mechanistic permeability-limited lung model (12) , which allows us to investigate the exposure of the drugs in lung tissue and ELF, rather than considering the lung as a homogenous perfusion limited organ as in previous publications (2, 11) . We also assess the impact of COVID-19 (via potential pH changes in the lung and renal impairment as a covariate) on the accumulation of each of the drugs in the plasma and lung. In addition,

we compare the predicted exposure of each drug in plasma and lung with the published EC 50 and EC 90 values. This article is protected by copyright. All rights reserved

A literature search was conducted to collate EC 50 values for CQ, HCQ and AZ determined in vitro. It was noted that the in vitro viral activity assays differed slightly in their methodology where described in detail. However in essence, Vero E6 cells were maintained in medium supplemented with low levels of serum were infected with SARS-CoV-2 virus at a multiplicity of infection (MOI) ranging from 0.001-0.8 for several hours at a temperature of 37°C. These virus cells were then washed with media and treated with medium containing drug over a range of concentrations for 24 or 48 hours. Viral RNA was extracted and analysed by Real-Time PCR. In each of the publications, a sigmoidal hill function was used to estimate EC 50 values (half maximal viral inhibition constant) (2, (13) (14) (15) .

A survey of literature data was performed to identify parameters, such as plasma binding proteins and pH values for the lung tissue and ELF that may influence the predicted systemic and lung drug exposures in a "COVID-19 population". Due to the scarcity of data specific for COVID-19, we expanded our search to other types of viral pneumonia (Supplementary Information).

Simulations using full PBPK and the permeability limited lung model The Simcyp (V19.1) population-based PBPK simulator (Simcyp Ltd, Sheffield, UK) was used to generate the profiles of CQ, HCQ and AZ. PBPK models for all 3 compounds were verified and published previously (2, 16) . Each of these models was refined to include a permeability-limited lung model (12) . Based on its anatomy and physiology, the lung model is described by seven segments representing upper and lower airways (2 segments) and the lobes of the lung (5 segments). Each segment is divided into four compartments representing pulmonary capillary blood, tissue mass, fluid, and alveoli air.

The fluid compartment represents mucus and ELF (pH=6.6), whereas the mass compartment represents the different cell types within the lung (pH=6.7).

This article is protected by copyright. All rights reserved For each compound, Henry's constant was predicted using a quantitative structureactivity relationship (QSAR) approach and used to calculate the air:fluid partition coefficient. Three different in vitro-in vivo extrapolation approaches are available to predict lung permeability. The preferred method is in vitro permeability data obtained from the Calu-3 cell line (12) . However, Caco-2 permeability data (P app ) or physicochemical properties, such as Log D, pKa, hydrogen bond donor count can be used to predict Calu-3 permeability and hence, lung permeability. The in vitro permeability data are corrected by the unionized fraction of compound (calculated at the pH of the in vitro system) as only the unbound, unionized drug is considered to be passively permeable in the lung PBPK model. Due to low protein concentrations in ELF (17), the unbound fraction in the ELF was assumed to be 1 for all 3 drugs. Measured values of unbound fraction in lung tissue homogenate (fu,mass) could not be found for any of the compounds, therefore, fu in lung tissue mass was predicted as reported previously (18, 19) .

After integrating the permeability-limited lung model, simulations were run to assess the impact of key physiological parameters (pH in the lung) on the exposure of each of the drugs in the lung and plasma using a number of different dosage regimens that are currently being assessed in clinical trials. The effect of severe renal impairment (GFR<30 ml/min) was assessed using the population (elevated AAG, reduced albumin levels, elevated serum creatinine) within the Simcyp Simulator. Unless stated otherwise, 100 subjects aged 20 to 80 years (50% male, 50% female) were simulated.

For all of the above scenarios, the simulated concentrations for total lung, unbound lung, total ELF (assumed to be the same for unbound) and total plasma were compared against the lowest and highest reported EC 50 values. In addition, trough lung concentrations (unbound) were compared against a reported or extrapolated EC 90 (9 x EC 50 ).

The input parameters for the CQ model are provided in Supplementary Supporting Material (Tables S1 and S2 ). An in vitro Caco-2 P app value of 21x10 -6 cm/s was used to predict permeability across the lung (20) . Henry's constant was predicted to be 1.08 x 10 -7 Pa m 3 /mol (EPI Suite). Simulations were performed to demonstrate that the updated CQ

This article is protected by copyright. All rights reserved model was able to generate concentration-time profiles that were consistent with observed data (Supplementary Information). Predicted lung exposures relative to plasma over time were generated for a single 300 mg CQ dose and the Kp values were compared with observed rat data. The effect of lung pH (5.5 to 6.7) on the Kp values was also assessed.

Plasma and lung exposures were simulated for different dosage regimens that have traditionally been used for malaria treatment or are currently being investigated for treatment of COVID-19. These include:

(A) 600 mg CQ base on days 1 and 2, 300 mg on day 3 (1.5 g over 3 days) -WHO regimen (B) 600 mg CQ base on day 1 at 9am, followed by 300 mg at 3 pm on day 1, followed by 300 mg at 9 am on days 2 and 3 (1.5 g over 3 days) -FDA regimen (C) 1200 mg CQ base on day 1 followed by 300 mg QD for 9 days (3.9 g over 10 days) (ClinicalTrials.gov Identifier: NCT04328493) (D) 600 mg CQ base BID for 10 days (12 g over 10 days) (ClinicalTrials.gov Identifier: NCT04323527) (E) 300 mg CQ base QD for 4 days, followed by 300 mg weekly out to 30 days (1.65 g over 30 days) (ClinicalTrials.gov Identifier: NCT04333732) (F) 300 mg CQ base QD for 4 days, followed by 150 mg QD out to 30 days (5.1 g over 30 days) (ClinicalTrials.gov Identifier: NCT04333732)

With the exception of a metabolic intrinsic clearance of 21.5 µL/min/mg protein, the input parameters for the HCQ model were as cited previously by Yao et al, (2) . An in vitro Caco-2 P app value of 0.138 x 10 -12 cm/s (50-fold lower than that of CQ based on Ferrari et al, (21) ) was used to predict permeability across the lung. Henry's constant was predicted to be 3.98 x 10 -12 Pa m 3 /mol (EPI Suite) (Table S1 ). Plasma and lung exposures were simulated for the dosage regimen used in the French study; 200 mg HCQ (155 mg base) was given every 8 hours for 10 days (3).

This article is protected by copyright. All rights reserved

With the exception of the B:P ratio which was set at 2.28, the input parameters for the AZ model were as cited previously by Johnson et al., (2016) . An in vitro Caco-2 P app value of 1.67 x 10 -6 cm/s calibrated using a P app of 10.4 x 10 -6 cm/s for propranolol (22) was used to predict permeability across the lung. Henry's constant was predicted to be 5.37 x 10 -24 Pa m 3 /mol (EPI Suite) (Table S1 ). Simulations were performed to demonstrate that the updated AZ model was able to generate exposures of AZ that were consistent with observed data. Plasma and lung exposures were simulated for the dosage regimen used in the French study; 500 mg was given on day 1 followed by 250 mg once daily for 4 days (3).

Depending on the methodology applied, at 48 h, EC 50 values for CQ range between 1.13 and 7.36 µM (Table 1) . At 48 h, EC 50 values for HCQ range between 0.72 and 17.3 µM. A single EC 50 value of 2.12 µM was available for AZ. In each case, these represent total EC 50 values as no protein binding corrections have been applied. However, given that the the fu,p values for CQ, HC and AZ are 0.4, 0.5 and 0.69, it is expected that the protein binding in these in vitro pharmacology experiments will be minimal.

Normal ELF pH is considered to be acidic, averaging ~6.6 (range 5.7 to 7.5) in healthy airways compared to a blood and interstitial pH of 7.4. There are indications that in airway diseases, such as cystic fibrosis and pneumonia, ELF pH is more acidic, suggesting breakdowns in the pH regulatory mechanisms (23). Using a pH electrode/probe, the most acidic measured pH value in ELF of bacterially infected pneumonia patients is 5.6. Another study indicated that the pH of exhaled breath condensate (EBC), which reflects the pH on of the lining fluid of the lung, was lower in mechanically ventilated patients compared to spontaneously breathing individuals (mean±standard deviation values of 5.86±0.32 vs. 7.47±0.48) (24). In the absence of human data describing the pH value for lung mass, a value of 6.69 based on experimental data obtained in dog ((25);range of 6.46 -6.97) is used in the model. Based on these data, the range for sensitivity analysis on lung mass pH in the current study was set to 6.0 to 6.7.

Serum albumin is commonly measured as part of the initial evaluation of critically ill patients with infectious disease. Several studies show that the frequency of hypoalbuminemia at admission is high in hospitalized patients with community acquired pneumonia (26). The leading causes of hypoalbuminaemia are malnutrition, reduced hepatic synthesis, renal losses, and inflammation; however, the predominant mechanism by which serum albumin decreases is secondary to increased capillary permeability and redistribution from plasma to the interstitium in critically ill patients (27). A summary of

This article is protected by copyright. All rights reserved serum albumin levels in differing types of viral pneumonia in the literature is presented in Table S3 . The weighted mean albumin levels in severe cases of viral pneumonia is 29.3 g/L. Concentrations of an inflammation marker, alpha-1-acid-glycoprotein (AAG), are generally believed to increase during an infection, however, quantitative information is limited in patients with viral pneumonia. A 1.7-fold to 3-fold increase in AAG levels in CAP (28) or SARS-CoV patients (29) has been reported. As the above parameters appear to be reasonably consistent with those previously cited for patients with severe renal impairment (9), they were accounted for in simulations of virtual patients with GFR<30 ml/min.

The predicted Kp values for lung after a single oral dose of 300 mg CQ base are shown in Figure 1 . Observed data in humans are not available for comparison and rat data from several different studies are presented. The predicted Kp value after 10 days (Kp=166) appears to be reasonably consistent with the rat Kp values (approximately 220) that have been previously published (McChesney et al., 1967) . Of particular importance, is the finding that the predicted Kp, and hence the exposure in the lung, is very sensitive to small changes in pH. Indeed, a reduction in pH from 6.7 to 6.5 or down to 6.0, results in an increase in predicted Kp from 166 to 412 and 5208, respectively.

The simulated concentration-time profiles for CQ in plasma and lung (lung tissue and ELF) based on various different dosage regimens are shown in Figure 2 . When comparing the unbound lung exposure of CQ against the minimum cited EC 50 value for inhibition of SARS-CoV-2, all dosage regimens appear to generate high enough exposures from the first day of dosing. However, when compared against the EC 90 value (6.9 µM), the only dosage regimen that meets this target on day 1 is 600 mg BID for 10 days ( Table 2 ). For 2 of the dosage regimens, the effect of reducing lung pH to 6 (lung tissue and ELF) and renal impairment are shown in Figure 3 . Although the increase in CQ lung exposure is more sensitive to changes in pH than the added complication of renal impairment (protein binding and GFR), the combined effect resulted in 50-to 100fold increases relative to the control scenario (lung pH=6.7 and normal GFR) ( Table 2) .

This article is protected by copyright. All rights reserved 

After a single oral dose of 500 mg AZ, predicted plasma C max and AUC values were within 80-120% of observed values ((32); data not shown). Predicted C max values in the ELF and lung were 1.39 and 7.36 µg/mL, respectively, which were consistent with observed values of 1.20 and 8.30 µg/mL, respectively. Based on C max values, predicted tissue to plasma ratios were 3.73 and 19.8 for ELF and lung tissue, respectively, which were again similar to observed values of 3.08 and 21.3. Concentration-time profiles for AZ in plasma and lung (lung tissue and ELF) after a single oral dose of 500 mg followed by 250 mg daily for 4 days are shown in Figure 5 . A reduction in lung pH does not lead to a change in plasma exposure but a 2.7-fold increase in C max is expected in the lung (12.4 versus 33.5 µM). A 20% increase in plasma exposure is predicted in subjects with renal impairment and the C max in lung is further increased to 42.5 µM. Unbound lung concentrations of AZ do not attain the cited EC 50 or EC 90 values ( Figure 5) .

In the current study, we sought to demonstrate that the accumulation of CQ, HCQ and AZ in the lungs is sensitive to lung pH, a parameter, which could change significantly in patients as a consequence of having COVID-19. The degree of accumulation is largely

This article is protected by copyright. All rights reserved dependent on drug related parameters (log D, pKa, fu plasma and fu mass ) and the pH gradients maintained across the blood vessels and lung. This complex interplay of parameters is captured by the mechanistic multi-compartment permeability limited lung model (12) and allowed us to investigate the effect of COVID-19 and associated comorbidities on the predicted concentrations of CQ, HCQ and AZ in the lung and plasma. A change in pH from 6.7 to 6 in the lung, which is observed in respiratory disease, led to 20-, 4.0-and 2.7-fold increases in lung exposure of CQ, HCQ and AZ, respectively, due to an increase in ionisation of the drugs. Furthermore, the high concentrations of CQ and HCQ in lung tissue were sustained long after administration of the drugs had stopped. A significant number of patients with COVID-19 present with kidney failure (7). Our simulations indicate that renal impairment (plus lung pH reduction) One of the limitations of the study is that predicted baseline concentrations of the drugs in lung could only be verified for AZ. Clinical data for AZ were characterised in plasma, ELF, alveolar macrophages (AM) and lung tissue itself (32). For CQ and HCQ, Kp values for lung derived from rat data provided some level of verification. The lung model has been applied previously, with reasonable success, to predict the lung exposure of at least 7 drugs used in the treatment of tuberculosis (12) . However, future studies that characterise CQ and HCQ concentrations in ELF, plasma, AM and/or bronchial tissue are warranted. In addition, robust in vitro data relating to the permeability of the drugs across the lung are required to enhance the PBPK models for these drugs.

It is assumed that drug efficacy at the site of action is determined by the unbound drug in the tissue. The unbound fraction in lung was not available for these drugs and therefore, had to be predicted based on physicochemical properties (19) . 

This article is protected by copyright. All rights reserved dosage regimens that were simulated for CQ, HCQ and AZ appeared to attain concentrations that were comparable with the in vitro derived EC 50 values, despite the EC 50 values being highly variable for each compound. When compared against the EC 90 values, it appears that for HCQ, the dosage regimen investigated in the French study (155 mg base three times a day for 10 days) (3) was not able to attain lung exposures in this target range, a finding which was in agreement with Arnold & Buckner(11). Yao et al. (2) reported that based on their PBPK modelling results, HCQ (a loading dose of 310 mg base given twice on day 1 followed by 155 mg twice daily for 4 days) should be recommended for treatment of COVID-19 as this dosage regimen was likely to attain high enough concentrations in the lung to inhibit the SARS-CoV-2 virus.

For CQ, the potential 20-fold increase in lung exposure in COVID-19 patients relative to healthy subjects, led to predicted unbound concentrations in the lung that were higher than the EC 90 from day 1 even for one of the more conservative dosage regimens (300 mg for days 1 to 4, followed by 150 mg QD). Given the uncertainty associated with the predicted fu mass values and their impact on dose selection, appropriate in vitro studies to measure tissue binding in lung homogenate should be conducted to lend more weight to the predictions. It may also be pertinent to use EC 90 values derived from measured intracellular concentrations rather than those added to the cell culture media in vitro (33).

As uptake of both CQ and HCQ is sensitive to pH, it cannot simply be assumed that uptake into cells in vitro reflects accumulation of the drugs in the lung (33).

Last, but certainly not least, concerns relating to CQ and HCQ toxicity should also help inform dose selection. There are an increasing number of reports of associated QT prolongation and it is possible that patients with COVID-19 may be more susceptible to CQ and HCQ toxicity (34). Going forward, efforts should be made to link the exposure of CQ and HCQ in the heart with markers of QT prolongation, such as the EC 50 data derived for inhibition of cardiac ion channels (35) to better understand margins of cardiac safety.

Whilst it is important to acknowledge the limitations described and collect more clinical data to verify drug exposures in the lung, as highlighted by Arnold & Buckner, (11) , our study does show that there is a place for PBPK modelling to inform the advancement of

This article is protected by copyright. All rights reserved HQ and CQ and AZ as potential candidates for treatment of COVID-19. Furthermore, the strategy taken here may be useful to assess and prioritise other drugs that are potential candidates for repurposing.

Several publications have recently appeared relating to the application of PBPK models for prediction of the exposure of CQ and HCQ in lung and plasma. Both drugs have been identified as potential candidates for treatment of COVID-19.

We sought to demonstrate that a modelling approach can be used to assess the effects of physiological changes, potentially evoked by COVID-19, on the plasma and lung exposure of CQ, HCQ and AZ.

We confirm that the accumulation of CQ, HCQ and AZ in the lung is very sensitive to changes in pH, a parameter that could be affected by COVID-19. We also demonstrate that renal impairment, a comorbid condition of COVID-19, may lead to increased concentrations of all 3 drugs in the lung.

The approach and mechanistic lung model described here could be applied to other drugs being investigated as COVID-19 therapeutics. 

This article is protected by copyright. All rights reserved mass unbound concentration to EC 90 value (3.6 µM) against SARS-CoV-2 virus in Vero E6 cells. Table S1 3. Table S2 4. Table 23 5. Figure S1 6. Figure S2 7. Figure S3 Accepted Article This article is protected by copyright. All rights reserved : Ratio of the simulated lung mass unbound concentration to EC 90 value (6.9 µM) against SARS-CoV-2 virus in Vero E6 cells .

Breakthrough: Chloroquine phosphate has shown apparent efficacy in treatment of COVID-19 associated pneumonia in clinical studies

In vitro antiviral activity and projection of optimized dosing design of hydroxychloroquine for the treatment of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)

Hydroxychloroquine and azithromycin as a treatment of COVID

Hydroxychloroquine for treatment of SARS-CoV-2 infection? Improving our confidence in a model-based approach to dose selection

Development of a multicompartment permeability-limited lung PBPK model and its application in predicting pulmonary pharmacokinetics of antituberculosis drugs

Hydroxychloroquine, a less toxic derivative of chloroquine, is effective in inhibiting SARS-CoV-2 infection in vitro

In vitro screening of a FDA approved chemical library reveals potential inhibitors of SARS-CoV-2 replication

Remdesivir and chloroquine effectively inhibit the recently emerged novel coronavirus (2019-nCoV) in vitro

How does in vivo biliary elimination of drugs change with age? Evidence from in vitro and clinical data using a systems pharmacology approach

Interpretation of antibiotic concentration ratios measured in epithelial lining fluid

A mechanistic framework for in vitro-in vivo extrapolation of liver membrane transporters: prediction of drug-drug interaction between rosuvastatin and cyclosporine

Physiologically based pharmacokinetic modeling 1: predicting the tissue distribution of moderate-to-strong bases

Using human plasma as an assay medium in Caco-2 studies improves mass balance for lipophilic compounds

Kinetics and thermodynamics of chloroquine and hydroxychloroquine transport across the human erythrocyte membrane

Assessment of macrolide transport using PAMPA, Caco-2 and MDCKII-hMDR1 assays

We would like to acknowledge The Bill & Melinda Gates Foundation for funding the development of the original chloroquine model. We would like to thank Dr Ping Zhao for his review of the manuscript and Eleanor Savill for her assistance with manuscript preparation.

This article is protected by copyright. All rights reserved

This article is protected by copyright. All rights reserved