Predictions of Systemic, Intracellular, and Lung Concentrations of Azithromycin With Different Dosing Regimens Used in COVID-19 Clinical Trials.

Azithromycin (AZ), a broad-spectrum macrolide antibiotic, is being investigated in patients with coronavirus disease 2019 (COVID-19). A population pharmacokinetic model was implemented to predict lung, intracellular poly/mononuclear cell (peripheral blood monocyte (PBM)/polymorphonuclear leukocyte (PML)), and alveolar macrophage (AM) concentrations using published data and compared against preclinical effective concentration 90% (EC90 ) for severe acute respiratory syndrome-coronavirus 2 (SARS-CoV-2). The final model described the data reported in eight publications adequately. Consistent with its known properties, concentrations were higher in AM and PBM/PML, followed by lung tissue, and lowest systemically. Simulated PBM/PML concentrations exceeded EC90 following the first dose and for ~ 14 days following 500 mg q.d. for 3 days or 500 mg q.d. for 1 day/250 mg q.d. on days 2-5, 10 days following a single 1,000 mg dose, and for > 20 days with 500 mg q.d. for 10 days. AM concentrations exceeded the 90% inhibitory concentration for > 20 days for all regimens. These data will better inform optimization of dosing regimens for AZ clinical trials.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

☑ Azithromycin (AZ) is currently being used in clinical trials of patients with coronavirus disease 2019 (COVID‐19), although its optimal dose is unknown.

WHAT QUESTION DID THIS STUDY ADDRESS?

☑ The study was able to predict AZ concentrations in relevant tissues for antiviral activity including lung, polymorphonuclear and mononuclear cells, and alveolar macrophages using different dosing regimens and compare against in vitro effective concentration 90% (EC90) for severe acute respiratory syndrome‐coronavirus 2 (SARS‐CoV‐2).

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

☑ Azithromycin predicted exposure exceeded target EC90 in relevant tissues. The analysis provides a rationale to support dosing of AZ in clinical trials using az alone or in combination with other agents.

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

☑ Results will better inform optimization of dosing regimens for clinical trials of AZ.

Azithromycin (AZ), a broad‐spectrum macrolide antibiotic with a long half‐life and extensive tissue distribution, is being investigated in multiple clinical trials in patients with coronavirus disease 2019 (COVID‐19; clinicaltrials.gov). An early uncontrolled clinical study in a small number of patients showed that AZ combined with hydroxychloroquine contributed to the reduction in viral load in patients with COVID‐19. The study was expanded to a pilot study in a total of 80 mildly infected subjects who showed clinical improvement with administration of the combined medications. Controlled clinical studies in a larger number of patients are required to understand the clinical effect of AZ in this patient population. To support the planning and safe administration of AZ alone and in combination with other agents, the clinical pharmacology, activity against different viral agents, and safety of AZ were recently reviewed by Damle et al.

AZ is not approved for antiviral therapy, including infections with severe acute respiratory syndrome‐coronavirus 2 (SARS‐CoV‐2), the virus causing COVID‐19. However, several mechanisms have been postulated to support its activity against viral pathogens. AZ is a weak base ionized at cellular pH and, thus, accumulates in intracellular compartments, more specifically in lysosomes, and remains in these cells for an extended period of time. The accumulation of ionized AZ in these cells would increase pH, altering the acidic environment required for uncoating of enveloped viruses, and potentially impairing viral replication. In a recent preclinical study, the in vitro 50% effective concentration and 90% effective concentration (EC90) of AZ against SARS‐CoV‐2 was 2.12 µM (~ 1,600 ng/mL) and 8.65 µM (~ 6,500 ng/mL), respectively, as determined in a viral assay model using VeroE6 cells.

To better understand its efficacy against bacterial infections in the lungs, investigators have measured AZ in different tissues, including epithelial lining fluid, lung tissue homogenate, lung alveolar macrophages (AMs), and intracellularly in peripheral blood mononuclear cells (monocyte or lymphocyte (PBM)) or polymorphonuclear leukocyte (PML; Table ). , , , , , , , , A population pharmacokinetic (PopPK) model describing tissue distribution of unbound drug in muscle and subcutis and in PML separately for AZ un‐ionized in cytosol and ionized in lysosome compartments has been developed by Zheng et al., based on data originally published in Matzneller et al. This paper summarizes the extension of the PopPK model to predict tissue distribution in the lungs and AMs, as intracellular concentrations are most relevant to antiviral activity. The model can be utilized to assess the effectiveness of different dosing regimens based on comparison of the predictions in tissues to the in vitro antiviral EC90 value. Assessed regimens included those approved by the US Food and Drug Administration (FDA) for different mild to moderate bacterial infections, including 500 mg as a single dose on day 1 followed by 250 mg q.d. on days 2 through 5; 500 mg once daily for 3 days; and 1,000 or 2,000 mg as a single dose in adults.

Table 1

Summary of published systemic and tissue pharmacokinetics of azithromycin

	Zheng et al., 2014 5 Matzneller et al., 2013 6	Sampson et al., 2014 7	Lucchi et al., 2008 8	Liu et al., 2007 9	Danesi et al., 2003 10	DiPaolo et al., 2002 11	Ballow et al., 1998 12	Olsen et al., 1996 13
Dose	500 mg D1–D3	250 and 1,000 mg SD	500 mg SD	500 mg D1–D3 c	500 and 1,000 mg D1–D3	500 and 1,000 mg D1–D3	500 mg D1, 250 mg D2–D5	500 mg D1, 250 mg D2–D5
No. enrolled	6	10/dose	32 a	12	24/dose a	28/dose a	14	25 a
Plasma/serum	Plasma		Serum	Serum	Plasma	Plasma	Serum	Serum
Sampling b	0–7 days		0–72 hours	0–5 days	0–204 hrs	0–204 hours	0–264 hours	0–508 hours
PML (poly)	X	X (PMN)		X (PMNL)			X
PBM (mono)		X (PBMC)		X (MNL)				X
Lung			X		X	X
AM			X					X
Other	ISF in muscle and subcutis	Blood	ELF		Bronchial washing	Bronchial washing	Urine, RBC, blister fluid	ELF

AM, alveolar macrophage; D, days; ELF, epithelial lining fluid; ISF, interstitial fluid; MNL, mononuclear leukocytes; PBM, peripheral blood monocyte; PBMC, peripheral blood mononuclear cells; PML, polymorphonuclear leukocytes; PMN, polymorphonuclear cells; PMNL, polymorphonuclear leukocytes; RBC, red blood cell.

Number represents total number of subjects; N for tissue samples ranged from 3 to 6 per timepoint.

Sampling time is reported relative to first dose.

Another cohort of subjects received an extended release formulation.

Zheng et al., 2014

Matzneller et al., 2013

METHODS

Model development

For the current analysis, model development and simulations were undertaken using the following steps, which are further described below.

Internalize model developed by Zheng et al., to simulate concentrations in plasma, muscle/subcutis, and PML;

Perform external visual predictive checks (VPCs) for plasma/serum and white blood cell (WBC) concentration‐time profiles against mean/median and available individual data and adjust parameters, as required;

Using the same model structure as Zheng et al. for tissue and intracellular concentrations, add parameters to describe published lung and AM concentration‐time data;

Conduct sensitivity analysis to verify assumptions around final parameter values;

Use the model to simulate relevant dosing regimens of AZ.

First, the AZ specific PopPK model developed by Zheng et al. was utilized as the initial base model. This model consisted of a three‐compartment model with first order absorption and elimination, with an absorption lag describing the disposition of unbound AZ in plasma, and both fast and slow equilibrating tissues (Figure ). The model also included two tissue compartments one each for muscle and subcutis, and another compartment for PMLs, each with an associated deep‐tissue compartment. Distribution to tissue compartments was driven by unbound drug from plasma, whereas distribution to PML cytosol was driven by unbound un‐ionized drug from plasma. Tissue and cell concentrations were scaled according to estimated distribution factors representing the ratio between tissue/cell concentrations and plasma concentrations.

Figure 1

Structural model of azithromycin pharmacokinetics derived from Zheng et al. and Zhao et al. . AM, alveolar macrophage; fu, fraction unbound; funion, fraction un‐ionized; ISF, interstitial fluid; WBC, white blood cell.

As reported by Zheng et al., the base model was fitted to free plasma concentrations, with the model data derived from total plasma concentrations (C) using a concentration‐dependent fraction unbound (f) (Eq. 1):with f equal to 0.4984 at very low AZ concentration and up to 0.8 at a concentration of 300 ng/mL. A full description of the model can be found in the original publication.

Second, an evaluation of the typical plasma predictions and parameter estimates of the initial base model was conducted against other AZ PopPK models in the literature (Supplementary Table ) , , , , , and published summary PBM and PML data, , , , as defined in the section below (Data for External Validation). Changes to the initial three‐compartment structure (systemic disposition) parameters were made based on comparison to other published AZ PopPK models, after adjusting for the difference in free and total concentrations. AZ PopPK models fitted to total plasma concentrations required the f to be incorporated when drug was distributed to the muscle, subcutis, and PML compartments of the base model (Eq. 2):where A tis = amount in tissue, A plas = amount in plasma, A tis,deep = amount in deep compartment connected to tissue, k in = rate constant for uptake in tissue/PML, k out = rate constant for distribution out of tissue/PML, and k on/k off = on and off rate constants in tissue/PML compartments.

In addition, the model was validated using a VPC against external individual plasma, PML, and PBM data. PML and PBM concentrations were assumed to be similar based on data from Sampson et al., 2014.

Third, once the improvement of plasma and PML/PBM predictions was completed, development then focused on the extension of the model to predict concentrations in lung tissue and AMs (Figure ). The lung compartment was parameterized assuming unit density (1 g/cm3) when comparing lung predictions with digitized lung concentrations. , , Extension of the model to AMs assumed a similar description regarding intracellular concentrations as that described in PMLs after adjusting for distribution in lung tissue. These compartments shared the parameters used for distribution into tissue and cell compartments from the base model and are shown in Eq. 3 for change in drug in lung over time (dA lung/dt) and Eq. 4 for change in AM over time (dA AM/dt): where the parameter f unionized is calculated based on AZ pKa (pKa1: 8.1, pKa2: 8.8) and physiologic pH as reported in Zheng et al., using the same assumption that plasma and tissue compartment pH are similar, and k in, k out, k off, k on, and A plas have been defined above. The rate constants of distribution into and out of the lung and AMs (k in/k out ratios) were assumed to be the same as muscle/subcutis and PMLs, respectively. It should be noted that this assumption does not cover all distribution parameters, as these rate parameters were allowed to scale with the estimation of a distribution factor.

A local sensitivity analysis was performed to assess the sensitivity of area under the concentration‐time curve (AUC) in AM and lung tissue to changes in model parameters. Parameters were perturbed by up to ± 20% with 2% intervals (−20%, −18%, −16%, etc.) to simulate AUC and were compared against the reference AUC simulated from unperturbed parameters.

Model simulations were conducted using the R statistical and programming language (version 3.6.1) with mrgsolve and tidyverse packages. , , Models implemented in mrgsolve were compared against NONMEM (version 7.4.3) to validate model specifications. Visual evaluation was used to assess models during development, consisting of: (i) comparison of simulated typical model predictions against digitized mean and SD data in plasma, lungs, muscle, subcutis, AMs, PBMs, and PMLs; (ii) VPC using individual clinical plasma concentration data; and (iii) VPC using individual clinical WBC concentration data. VPCs were generated by simulating from the observed data 1,000 times, with the median 50th, 5th, and 95th percentiles of the individual predictions from each simulation and 95% prediction intervals presented with an overlay of the observed data in those tissues.

Data for external validation

Data used for model evaluation included digitized aggregate tissue data from literature and individual clinical trial data. Digitized tissue data were collected from eight different studies as reported in the literature, , , , , , , , , which consisted of mean and SD values over time for six unique dosing regimens. Table presents tissue/cell concentrations and a high‐level summary of the study designs that were available from each literature source.

Individual clinical trial data were available from two published phase I studies. , The first study used an open‐label randomized single dose design (n = 40), to estimate the bioavailability of a fixed dose combination of AZ‐chloroquine tablets relative to co‐administration of separate AZ and chloroquine tablets. These formulations were considered bioequivalent based on the conclusions of the study.

The second study, already mentioned above, was an open‐label randomized parallel‐group study (n = 24) comparing the plasma, PML, and MNL pharmacokinetics (PKs) of AZ single‐dose sustained release AZ (2 g) and a 3‐day regimen of immediate release AZ (3 × 500 mg). Only the immediate release data (n = 12) were used in the present analysis.

Model simulations and companion app

The final model was used for simulations of drug concentrations based on planned AZ dosing regimens in COVID‐19 clinical trials, as reported on registry clinicaltrials.gov (accessed on April 8, 2020; Supplementary Table ). Simulations consisted of 1,000 individuals, with median predictions and 90% prediction intervals in plasma, PMLs/PBMs, lung tissue, and AMs compared with the EC90 (8.65 µM) of AZ against SARS‐CoV‐2.

A companion app was also developed for the simulation of AZ concentrations in relevant tissues using a variety of different dosing regimens (https://github.com/jhhughes256/azithro_pk). The application was designed to allow for individual and population simulations, user‐defined dosing regimens, comparison of predictions between different dosing regimens, adjustment of EC90 values, and assessment of time above the EC90 in lungs and AMs.

RESULTS

Implementation and evaluation of the initial base model developed by Zheng et al. highlighted differences in the modeled population, with parameter estimates and typical predictions differing substantially from AZ plasma/serum profiles reported by Zhao et al. (Supplementary Figure .1) and by Liu et al. (Supplementary Figure .2), and PML and MNL concentrations by Liu et al. (Supplementary Figure .3). The estimate of total apparent clearance (CL/F) was reported as 258 L/hour (bootstrap 95% confidence interval (CI) 71–517) by Zheng et al. and was derived based on a small number of subjects (N = 6). Although not directly comparable to the CL/F estimates for PopPK models of total plasma concentrations, when accounting for f (~ 0.5–0.9) the CL/F estimate (129–232 L/hour) was equal or greater than other reported PopPK values (range 100–158 L/hour; Supplementary Table ). To adjust for the misspecification against external central tendencies, the parameters describing the systemic disposition of AZ including absorption were updated using data reported by Zhao et al. The Zhao et al. model was selected as the ideal candidate due to its three‐compartment structure, number of subjects included in the analysis, robustness of the analysis, and inclusion of weight as a covariate enabling predictions in other populations. Mean parameter estimates were similar to the other reported values (Supplementary Table ).

When implementing the parameters from the Zhao et al. three‐compartment model, the apparent volume of distribution in the fast distributing compartment was corrected (2890 L to 2490 L) to account for the added volume of 400 L attributed to the addition of both the muscle and subcutis compartments. The value was derived based on K in and K out values reported by Zheng et al. The combination of parameters in the hybrid model (Table ) provided adequate prediction of both the digitized data from Zheng et al. (Supplementary Figure ) and the observed data used in development of the Zhao et al. model used as internal validation.

Model predictions from the hybrid model were also validated against mean reported values of both PMLs and PBMs. , , The pooled white blood cell (WBC; either PML or PBM) predictions used total concentrations in WBCs, which included both ionized (major component in cells) drug in lysosomes and un‐ionized drug in cytosol. As the model predictions for the pooled WBC compartment tended to underestimate the concentrations of different PBM/PML data sources (not shown), the distribution factor for the pooled WBC cytosol was increased from 52 to 77. This final estimate was contained within the reported bootstrap CIs of the distribution factor estimate for PML cytosol (95% CI 39–423) and this adjustment visually improved model predictions.

As described in Methods, the hybrid model was extended to represent the lung tissue and AM cells (Figure ). Volume displacement caused by adding the lung compartment was accounted for by reducing the apparent fast‐distributing peripheral compartment, as described previously. The initial values for distribution factors to the new lung and AM compartments were determined by calculating the ratio between tissue and plasma from Lucchi et al. and Danesi et al. and further adjusted based on visual evaluation. The final model parameters are presented in Table and code provided in the Supplementary Material. VPC of the individual serum and WBC (observed PML and MNL) concentrations are shown in Figure and supported the validity of the final model.

Table 2

Pharmacokinetic parameters of the base, hybrid, and final models

	Base model 5	Hybrid model 5 , 14	Final model
Fixed effect parameters
T lag, hour	1.45	‐	‐
k a, hour−1	0.88	0.259	0.259
CL/F, L/hour	258	100	100
Normalization for body weight	NA	0.75	0.75
Vc/F, L	160	186	186
Normalization for body weight	NA	1.00	1.00
Q P1/F, L/hour	207	180	180
Normalization for body weight	NA	0.75	0.75
VP1/F, L	1190	2490	2290
Normalization for body weight	NA	1.00	1.00
QP2/F, L/hour	101	10.6	10.6
VP2/F, L	9721	2610	2610
k in, hour−1	0.16	0.16	0.16
k out, hour−1	0.15	0.15	0.15
k on, hour−1	0.56	0.56	0.56
k off, hour−1	0.05	0.05	0.05
DFmuscle	0.55	0.55	0.55
DFsubcutis	0.25	0.25	0.25
DFPML/WBC(cytosol)	52	52	77
DFlung	‐	‐	140
DFAM(cytosol)	‐	‐	730
Interindividual variability (%CV)
T lag	17.6	‐	‐
CL/F	29.3	31.3	31.3
Vc/F	168.3	113	113
k in	0.22	0.22	0.22
DFmuscle	26.9	26.9	26.9
DFsubcutis	31.5	31.5	31.5
DFPML/WBC(cytosol)	0.22	0.22	0.22

CL/F, apparent clearance; CV, coefficient of variation; DF, distribution factor; k a, first‐order absorption rate constant; kin, rate constant for uptake into tissue compartments; k on and k off, rate constants for nonspecific tissue binding; k out, rate constant for distribution out of tissue; NA, not applicable; QP1/F, apparent fast peripheral intercompartment clearance; QP2/F, apparent slow peripheral intercompartment clearance; T lag, absorption lag; Vc/F, apparent central volume of distribution; VP1/F, apparent fast peripheral volume of distribution; VP2/F, apparent slow peripheral volume of distribution.

Figure 2

Visual predictive checks of the final model for azithromycin concentrations in plasma (a) and white blood cell (b) using external data from Liu et al. The observed azithromycin concentrations are represented by blue circles, whereas the median, 5th, and 95th percentiles of the observed data are represented by the solid, lower‐dashed, and upper‐dashed red lines. The median, 5th, and 95th percentiles of predictions from 1,000 simulations of the observed data are represented by the solid, lower‐dashed, and upper‐dashed black lines. The 95% prediction intervals for the median, 5th, and 95th percentiles of the simulated data are represented by the red, lower‐blue, and upper‐blue shaded areas.

Results of the local sensitivity analysis in AM and lung tissue (not shown) demonstrated that AUC was highly sensitive to the value of CL/F and the distribution factor for the respective tissues. AUC in AMs was also sensitive to changes in k in and k out. These parameters were shared with the PML compartment, assuming a similar ratio between distribution into and out of both cell types, with the scalar distribution factor controlling the magnitude of the rate of distribution. Although the sensitivity of AUC to changes in k in and k out supported estimation of tissue‐specific values for these parameters, there was insufficient data to support determination of all three sensitive parameters (k in, k out, and distribution factorAM). As the main difference between PML and AM concentration profiles were the magnitude of the concentrations, no changes were made to the underlying assumptions of the model based on the sensitivity analysis.

As shown in Figure , the final model adequately described the mean systemic (plasma/serum), pooled WBC, tissue (muscle and subcutis), lung, and AM concentrations across different studies using different dosing regimens. The model was sufficiently robust to describe the AZ PKs following single dose administration (doses of 250, 500, and 1,000 mg), 3‐day regimen (500 and 1,000 mg), and the commonly used regimen of 500 mg on day 1 followed by 250 mg days 2 to 5.

Figure 3

Comparison of final model predictions with digitized literature values. The digitized mean and SDs from literature data are represented by the shapes and error bars respectively. The solid lines represent the typical simulated predictions in different tissues when simulating patients (body weight 79 kg) treated with azithromycin: (a) 500 mg initial dose, followed by 250 mg daily for 4 days; (b) 500 mg daily for 3 days; (c) 1,000 mg daily for 3 days; (d) 500 mg single dose; (e) 1,000 mg single dose; and (f) 250 mg single dose. AM, alveolar macrophage; MNL, morphonuclear leukocyte; PBM, peripheral blood monocyte; PML, polymorphonuclear leukocyte; WBC, white blood cell.

Model simulations

Based on the most commonly reported dosing regimens for AZ investigated in clinical trials of COVID‐19 (Supplementary Table ), systemic concentrations, total concentrations in lung tissue, and intracellular concentrations in PML/PBM and AM, were predicted and compared against the reported 90% inhibitory concentration (IC90) for antiviral activity in vitro (Figure ). Consistent with known properties of AZ, drug concentrations were higher intracellularly as shown in AM and WBC (PBM/PML), followed by lung tissue and plasma. Concentrations in WBC exceeded the in vitro IC90 starting following first dose and for ~ 14 days with either of the currently approved regimens for bacterial pulmonary infections (500 mg q.d. days 1–3 or 500 mg on day 1, followed by 250 mg q.d. days 2–5), 10 days following a single 1,000 mg dose, and > 20 days with administration of 500 mg q.d. for 10 days. Trough concentrations in AM were ~ 4‐fold greater than those in WBC after first dose and exceeded the IC90 for > 20 days based on the four regimens tested. Concentrations in total lung tissue were generally below IC90 except following administration of 500 mg q.d. for 10 days.

Figure 4

Final model simulation of azithromycin concentrations for different treatment regimens. Treatment regimens were: (a) 500 mg daily for 3 days; (b) 500 mg initial dose, followed by 250 mg daily for 4 days; (c) 1,000 mg single dose; and (d) 500 mg daily for 10 days. Solid lines represent the median concentration from 1,000 simulated individuals (body weight 79 kg), while shaded areas represent the 90% prediction intervals. Black dashed line represents in vitro 90% inhibitory concentration.

DISCUSSION

As the world faces the COVID‐19 pandemic with a rapidly growing number of individuals infected, the demand for therapeutics against SARS‐COV‐2 is most urgently met by repurposing approved compounds and evaluating their activity against the virus. The use of hydroxychloroquine in this population has been extensively covered by the press, and different quantitative approaches have been implemented to support its use, including a physiologically‐based pharmacokinetic model, a PK/virologic/corrected QT model, and a web application quantifying exposure (http://www.covid19pkpd.eu/simulator/). In vitro screening of a chemical library of compounds identified activity of a selected number of drugs against SARS‐COV‐2 replication, including AZ with a potent 50% effective concentration of 2.12 µM. To further support the rationale for evaluation of AZ in these patients, a PK model was developed based on an existing semi‐physiologic model and existing tissue data to predict intracellular concentrations.

Modeling of the viral dynamics of SARS‐CoV‐2 described in a recent preprint by Gonçalves et al. identified that treatment either initiated at the time of infection, symptom onset, or 3 days post‐symptom onset would require at least a 60%, 90%, or 99% efficacy in reducing viral replication, respectively, to reduce the peak viral load by > 2 log units. According to the work presented herein, predicted AZ in AMs and WBCs exceeded the in vitro EC90 for AZ against SARS‐CoV‐2. Providing that these cells represent the site of action against SARS‐CoV‐2, AZ may be effective even after starting treatment up to 3 days after symptom onset. However, if the lung tissue better represents effective concentration at the site of action, then treatment would need to be initiated prior to or during symptom onset.

Simulations of the final model show that AZ accumulates in intracellular compartments (AM and PBM/PML), with slow distribution out of these cells over time. As a result, the extent of exposure in these cells is dependent on the total dose administered regardless of regimen. The 3‐day and 5‐day regimens with a total dose of 1.5 g provided similar profiles in these compartments, whereas lower exposure was noted with 1 g administered as a single dose and higher exposure noted with 5 g administered over 10 days. Considering the large number of possible dosing regimens and the assumptions that can be made regarding clinical effectiveness, the model, analysis, and web application code have been provided for researchers to analyze different dosing scenarios other than those included herein (https://github.com/jhhughes256/azithro_pk).

One limitation of the approach used for implementation of the present model was the use of data collected in healthy volunteers, whereas lung data was obtained in patients without infection undergoing lung resection. In the presence of viral infection, distribution in lung tissue may be increased, and, thus, the model may underestimate lung concentrations in infected individuals. Despite this limitation, the present analysis and web application enables the evaluation of alternate scenarios, and will better inform optimization of dosing regimens for ongoing and future AZ clinical trials.

Funding

This analysis was funded by Pfizer.

Conflict of Interest

All authors are employees of Pfizer Inc. and own stocks.

Author Contributions

J.H., K.S., S.A., and D.O. wrote the manuscript. J.H., K.S., S.A., and D.O. designed the research. J.H., K.S., S.A., and D.O. performed the research. J.H. analyzed the data.

Supplementary Material

Click here for additional data file.

Table S2

Click here for additional data file.