Abstract
In vivo pharmacokinetic simulations and virtual bioequivalence (BE) evaluation of cilostazol have not yet been described for humans. Here, we successfully developed a physiologically based absorption model to simulate plasma concentrations of cilostazol. In addition, virtual population simulations integrating dissolution of 0.3% sodium dodecyl sulfate water media were executed to evaluate the BE of test and reference formulations. Simulation results show that test and reference formulations were bioequivalent among 28 subjects, but not nine subjects, consistent with clinical studies. The model proved to be an important tool to show potential BE for cilostazol. This finding may facilitate understanding of the potential risks during the development of generic products.
Study Highlights.
WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?
Physiologically based absorption modeling that integrates the clinical and non‐clinical results can be used to predict bioequivalence (BE) study outcomes for various Biopharmaceutics Classification System (BCS) II compounds.
WHAT QUESTION DID THIS STUDY ADDRESS?
This study explored the BE of two cilostazol formulations using a physiologically based absorption model and clinical studies.
WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?
Virtual BE simulations based on an established physiologically based absorption model indicate that the two cilostazol formulations were bioequivalent among 28 subjects, but not nine subjects, which is consistent with clinical studies.
HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?
These results support physiologically based absorption modeling as an important tool to predict pharmacokinetic and BE of BCS II drugs, which facilitates understanding of potential risks and guides design of BE studies.
INTRODUCTION
In vivo dissolution and release of orally administered drug products is one of the rate‐limiting steps of drug absorption from the gastrointestinal tract. Thus, it can be an indicator to predict pharmacokinetic (PK) behavior in humans. 1 , 2 However, oral absorption of a drug involves a highly complex process that is influenced by various factors, such as physicochemical properties of the drug, characteristics of the formulation, and physiological factors of the gastrointestinal tract. 3 , 4 Therefore, it is necessary to develop a predictable evaluation method to link in vivo PK profiles with in vitro dissolution data. Computer simulation approaches integrating key properties of the drug and formulation, and the gastrointestinal tract physiological parameters have been used to predict oral absorption, 5 support formulation development, 6 and preclinical to clinical translation. 7 In addition, these models are now routinely used for regulatory review and decision making. 8
In vivo bioequivalence (BE) studies are pivotal to show similarity between a generic product and reference product, and are routinely applied during the development of various generic products. 9 , 10 , 11 However, these studies often require costly clinical investigations. Early assessment of potential BE risk is critical for an effective and streamlined clinical pharmacology strategy. Therefore, application of modeling and simulation approaches to accurately predict BE outcome can help guide formulation development and risk assessments.
Cilostazol, a class II drug in the biopharmaceutical classification system (BCS), 12 is widely used for the treatment of peripheral ischemia, such as intermittent claudication. 13 The PK profile of cilostazol has been reported. 10 , 14 The maximum plasma concentration of cilostazol occurs ~3 h after administration. 15 Hepatic metabolism is the major route of elimination for cilostazol. Two of its metabolites, 3,4‐dehydro‐cilostazol (OPC‐13015) and 4′‐trans‐hydroxy‐cilostazol (OPC‐13213), are identifiable and considered pharmacologically active. Cilostazol has a half‐life of ~10 h. 14 An in vivo BE study is required for the development of a generic cilostazol drug. To reduce clinical failure, it is necessary to develop an oral absorption model to predict BE outcomes for cilostazol formulations.
This study examines the accuracy of a human model for predicting the PK parameters of cilostazol tablets, including the effects of in vitro dissolution behavior on the in vivo absorption outcome, and BE of cilostazol using GastroPlus software.
METHODS
Materials
Cilostazol 50 mg tablets were obtained from Zhejiang Yongning Pharmaceutical, Ltd. (test formulation), and Teva Pharmaceuticals USA, (reference formulation).
In vitro dissolution study
A United States Pharmacopeia type II dissolution apparatus (paddle) was used for in vitro dissolution testing. For testing, dissolution media consisted of water with 0.3% sodium dodecyl sulfate. Each dissolution vessel was filled with 900 mL of dissolution medium. The medium in the vessel was kept at 37°C. The paddle rotational speed was 75 rpm. Samples were withdrawn at 5, 10, 15, 20, 30, 60, 90, and 120 min. Sample concentrations were determined using ultraviolet–visible spectrophotometry. To evaluate the discriminatory power, similarity factor (f2) analysis for in vitro dissolution profiles was conducted. When the f2 value was greater than or equal to 50, the dissolution profiles were considered to be similar.
Construction of in silico model
In vitro and in silico input parameters
Modeling was conducted using GastroPlus (Simulations Plus). On the basis of the cilostazol structure, we used the ADMET predictor module in GastroPlus to predict physicochemical and biopharmaceutic parameters, including Log P, pKa, intestinal effective permeability (P eff), blood to plasma ratio (Rbp), unbound fraction in plasma (F up). Certain parameters were replaced by measured values in the literature. Parameter sensitivity analysis (PSA) was used to assess the importance of input parameters in predicting the speed and degree of drug absorption. Physicochemical and pharmaceutics parameters used for model construction are shown in Table S1. The corresponding dissolution profile of the reference formulation was also used as an input parameter. Percentages of in vivo drug release were calculated with the “single Weibull function” in GastroPlus software. For modeling, the dosage form option “CR: Integral tablet” was selected in GastroPlus software.
Physiology
To accurately reflect the physiological absorption process of the drug, gastrointestinal tract simulation based on the advanced compartmental absorption and transit (ACAT) model of GastroPlus software was used. This absorption model is a physiologically based transit model, comprising different compartments of the gastrointestinal tract (stomach, duodenum, jejunum, ileum, cecum, and colon), that describes the release, dissolution, and absorption of the compound in each compartment. The default human fasted physiological model of the ACAT module in GastroPlus (Opt logD SA/v6.1) was used for simulations.
Compartment model
PK parameters were estimated using the PKPlus module of GastroPlus. The mean plasma concentration profile of clinical study I was entered into the PKPlus module to evaluate the optimal compartmental model and related disposition parameters. One‐, two‐, and three‐compartmental model parameters are summarized in Table S2. Compartmental models were compared by evaluating the coefficient of determination (R 2) and Akaike Information Criterion (AIC). The one‐compartmental model produced the lowest AIC value, whereas the two‐compartmental model yielded a higher R 2. Ultimately, we selected the two‐compartmental model to generate disposition parameters for simulation.
Prediction accuracy
To evaluate the prediction accuracy of the model, the absolute percent prediction error (%PE) was calculated using the following equation:
The acceptance criterion for differences between observed and predicted values for maximum concentration (C max), area under the concentration‐time profiles (AUC0‐t ), and area under the concentration‐time curve from zero to infinity (AUC0‐∞) was no more than 15% of %PE.
Virtual BE simulation
A population simulation model integrating test and reference product dissolution data was used to predict the outcome of pivotal BE studies. Based on clinical study I, virtual BE simulations were conducted with nine subjects and 28 subjects. To evaluate the reproducibility of this simulation, the virtual BE simulation for 28 subjects was repeated 10 times. In the simulation process, each simulated subject has a random set of physiological and PK parameters to imitate variances of absorption, distribution, metabolism, and excretion. Coefficients of variation for C max and AUC0‐t in BE simulations was 20% and 10%, respectively. The geomean ratio and 90% confidence interval (CI) were automatically calculated by GastroPlus. The test formulation was considered bioequivalent to the reference formulation if 90% CIs for C max, AUC0‐t , and AUC0‐∞ were within the range 80.00%–125.00%. Observed PK results for clinical studies I and II were used to verify the model.
In vivo clinical study
Subjects
Human PK data were generated from clinical studies I and II. The two studies were completed before and after modeling, respectively; inclusion and exclusion criteria were identical. Subjects aged 18 to 45 years with a body mass index between 19 and 26 kg/m2 were eligible for the studies. Subjects were considered healthy based on history and physical examination, clinical laboratory tests, echocardiogram, and chest X‐ray examination. Subject exclusion criteria were as follows: history of any disease; abnormal bleeding; positive blood screen for human immunodeficiency virus, Treponema pallidum antibodies, or hepatitis B surface antigen; history of alcoholism; positive urine screen for drugs and nicotine; blood donation within 3 months before administration (≥400 mL); taking any prescription, over‐the‐counter medications, or herbal medicine within 14 days; allergic to any component of the tablet. Female subjects needed to be non‐breastfeeding and have a negative pregnancy test. Informed consent was obtained from all subjects before the study.
Study design
The studies were performed at the Center of Clinical Pharmacology, the Second Affiliated Hospital, School of Medicine, Zhejiang University. The protocol and informed consent form were reviewed and approved by the Human Subject Research Ethics Committee of the Second Affiliated Hospital School of Medicine, Zhejiang University (2019), LSYD No. (557), and (2022) LSYD No. (500). Both studies were conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines.
Both randomized, open‐label, two‐period crossover studies were executed under fasting conditions. A total of nine and 28 healthy adult subjects were enrolled in clinical studies I and II, respectively. Subjects in both studies were randomized at a 1:1 ratio to receive a single oral dose of 50 mg cilostazol tablet test formulation or reference formulation. After a 7‐day washout period, the same procedure was repeated with the crossover drug. For study I, blood samples were collected before drug administration (0 h) and 0.5, 1, 2, 2.5, 3, 3.5, 4, 4.5, 5, 6, 8, 10, 12, 16, and 24 h after drug administration. For study II, blood samples were collected before drug administration (0 h) and 0.5, 1, 2, 2.5, 3, 3.5, 4, 4.5, 5, 6, 8, 10, 12, 16, 24, 36, 48, and 72 h after drug administration. Within 60 min of collection, plasma was obtained via centrifugation (2000×g, 10 min, 2–8°C) and stored at an ultra‐low temperature refrigerator. Samples were analyzed using the liquid chromatography–tandem mass spectrometry method.
Statistical analyses
All clinical PK parameters of cilostazol from in vivo clinical study were calculated using Phoenix WinNonlin software (version 7.0; Pharsight Corporation). Ratios of geometric least‐squares mean of C max, AUC0‐t , and AUC0‐∞, and their 90% CIs were calculated. The two formulations were judged as bioequivalent if the 90% CIs were within the equivalent range of 80%–125%.
RESULTS
In vitro dissolution behavior
Dissolution profiles for the two cilostazol formulations in water media with 0.3% SDS are shown in Figure 1. For in vitro dissolution results, f2 was calculated to evaluate similarity between the two formulations. The results showed that the dissolution profiles of reference and test formulations were similar (f2 > 50), although the two formulations did not dissolve rapidly (<85% of the formulations dissolved within 30 min).
FIGURE 1.
In vitro dissolution profiles of cilostazol reference formulation in aqueous media with 0.3% SDS (n = 12).
Construction of an oral absorption model
In this study, an oral absorption model was initially constructed based on the physicochemical, pharmaceutic, disposition parameters, and gastrointestinal tract physiological parameters using the GastroPlus single‐simulation mode. Simulated and observed plasma concentration–time profiles (Figure 2a ) indicate that the simulation profile failed to accurately capture the observed results. To better fit the observed results, we optimized the P eff (from 3.89 to 1.2) and Rbp (from 0.7 to 0.2) values. The optimized simulated and observed plasma concentration–time profiles are shown in Figure 2b. Simulations with the optimized model show an improved match to plasma concentrations. %PE values for C max, AUC0‐t , and AUC0‐∞ for the two formulations met our acceptance criteria for prediction accuracy (≤15%; Table 1). These results confirm that the oral absorption model of cilostazol was successfully built. In fact, various PSA was performed before model optimization, including evaluation of P eff, Rbp, F up, log P and effective particle radius. PSA results indicate that C max and AUC were sensitive to the variation of some input parameters, especially Peff and F up (Figure 2c,d). In addition, we investigated the in vivo regional absorption of cilostazol in healthy humans based on the model. As shown in Figure S1, high fractions of cilostazol are predicted to be absorbed in the small intestine (jejunum and ileum1).
FIGURE 2.
Model simulation and pharmacokinetic assessment of cilostazol reference formulation: (a) original model, (b) model with optimized parameters, (c) PSA on C max, (d) PSA on AUC0‐t . AUC0‐t , area under the concentration‐time profiles; C max, maximum concentration; PSA, parameter sensitivity analysis.
TABLE 1.
Simulated and observed PK parameters after orally administrated 50 mg cilostazol reference formulation.
| Parameters | Observed | Simulated | %PE |
|---|---|---|---|
| C max (ng/mL) | 318.14 | 343.84 | −8.1 |
| AUC0‐t (ng·h/mL) | 4183.2 | 4220.8 | −0.9 |
| AUC0‐∞ (ng·h/mL) | 5701.1 | 6274.0 | −10.0 |
| T max (h) | 5 | 4.56 | / |
Abbreviations: %PE, percent prediction error; AUC0‐∞, area under the concentration‐time curve from zero to infinity; AUC0‐t , area under the concentration‐time profiles; C max, maximum concentration; PK, pharmacokinetic; T max, time to C max.
In silico BE simulation
We first conducted a clinical BE study with the nine subjects before virtual BE prediction. The demographic characteristics of subjects are summarized in Table S3. As shown in Table S4, the two formulations were not equivalent because 90% CIs for the geometric least‐squares mean ratio of C max were 100.05%–128.18%. To demonstrate whether the model can reproduce the PK parameters of clinical study I, an in silico BE simulation was conducted using the developed oral absorption model. Virtual BE results for study I are shown in Table 2. In this simulation, the test formulation was judged as non‐bioequivalent because 90% CIs for the geometric mean ratio of C max were outside the range 100%–125%. This result is consistent with the observed clinical study results. Therefore, the in silico BE model can accurately reproduce clinically relevant behavior.
TABLE 2.
Virtual BE results in nine subjects of clinical study I.
| Parameters | Geometric least‐squares mean | T/R % | 90% CI | |
|---|---|---|---|---|
| Test (n = 9) | Reference (n = 9) | |||
| C max (ng/mL) | 237 | 234 | 101.6 | 81.36–126.93 |
| AUC0‐t (ng·h/mL) | 3273 | 3170 | 103.2 | 88.47–120.47 |
| AUC0–∞ (ng·h/mL) | 4754 | 4609 | 103.1 | 87.63–121.41 |
Abbreviations: AUC0‐∞, area under the concentration‐time curve from zero to infinity; AUC0‐t , area under the concentration‐time profiles; BE, bioequivalence; CI, confidence interval; C max, maximum concentration; T/R, the ratio of test formulation to reference formulation.
In silico BE simulation was conducted again, with a sample size expanded to 28. To evaluate the reproducibility of this simulation, virtual BE simulation was repeated 10 times. Representative population simulation profiles of the two formulations are shown in Figure 3. The simulation BE results for 10 trials are shown in Table 3. The test formulation is judged as bioequivalent in comparison with the reference formulation, as the 90% CI ln‐transformed of the ratio of C max and AUC was within the regulatory acceptance limit for BE (80%–125%) in all 10 simulation trials.
FIGURE 3.
Representative virtual BE profile of 28 subjects after oral administration of 50 mg cilostazol reference and test formulations. BE, bioequivalence; CI, confidence interval.
TABLE 3.
In silico BE simulation for 10 trials in 28 healthy subjects.
| Parameters | In silico BE judgment | 90% CI |
|---|---|---|
| C max (ng/mL) | 10/10 | 87.05 to 92.71~104.15 to 118.89 |
| AUC0‐t (ng·h/mL) | 10/10 | 89.11 to 96.11~104.06 to 113.68 |
| AUC0‐∞ (ng·h/mL) | 10/10 | 86.47 to 93.25~106.43 to 116.86 |
Abbreviations: AUC0‐∞, area under the concentration‐time curve from zero to infinity; AUC0‐t , area under the concentration‐time profiles; BE, bioequivalence; CI, confidence interval; C max, maximum concentration.
The clinical BE results of study II
On the basis of simulation results from the in silico BE model, we performed clinical study II, in which the sample size was increased to 28. The demographic characteristics of subjects are summarized in Table S5. Mean plasma concentration‐time curves for test and reference formulations are shown in Figure 4. Mean PK parameters and BE evaluation results are summarized in Tables S6 and S7, respectively. The results suggest that PK properties of the two formulations were similar. Moreover, the test formulation was bioequivalent to the reference formulation, consistent with in silico BE simulation results.
FIGURE 4.
Mean plasma concentration‐time curves for test and reference formulations in clinical study II. Data present mean ± SD. For test formulation, n = 27. For reference formulation, n = 28.
DISCUSSION
In the present study, we first predicted human PK characteristics of cilostazol using physiologically based absorption modeling. Specifically, an in vitro dissolution profile was entered into the model and simulated values were in accordance with observed in vivo results. In addition, we report the use of absorption modeling for BE prediction of test and reference formulations. The model was used to assist risk assessment for a pivotal BE study.
The relationship between in vitro dissolution behavior and in vivo performance of BCS II drugs has been widely investigated. 1 , 6 , 16 Cilostazol, a BCS II neutral drug, has poor solubility and high permeability. 12 Hence, dissolution rate‐limited absorption was expected, making it imperative that in vitro dissolution values be representative of in vivo dissolution. During drug development, several formulations with varying dissolution profiles were investigated. We selected the test formulation in our study on the basis of the outcome. We found that in vitro dissolution profiles of the two formulations were very similar. When in vitro dissolution data were integrated into the model, it accurately captured drug in vivo performance, indicating that the dissolution has clinical relevance. However, the dissolution boundary for maintaining BE between the test and reference needs further evaluation.
In our study, we first conducted a clinical BE study enrolling nine subjects, in which PK parameters were found to be very similar between the two formulations, indicating that failure of this BE study likely occurred due to a small sample size. However, it was uncertain whether the two formulations would meet the BE standard with an increased sample size. Virtual BE modeling can simulate distinct individuals with different physiological parameters, and has been widely applied to screening of formulations, dissolution specification evaluation, design of BE studied, and even risk accessment. 1 , 6 , 17 Therefore, we used our model to simulate and verify the study with nine subjects, and we plan to evaluate the reason for BE failure and further assess the risk of BE trials with expanded sample sizes. The simulation results of our model are consistent with the observed results with nine subjects, indicating the reliability of model prediction. Thus, we further conducted 10 consecutive virtual BE trials with 28 subjects. The results of all 10 trials show that the test formulation was bioequivalent to the reference formulation. These findings give us more confidence to subsequently conduct a formal BE trial with sufficient sample size.
In conclusion, we developed physiologically based absorption modeling by integrating clinical and non‐clinical results. Through the established model, the non‐bioequivalent batch was discriminated from the bioequivalent batch in clinical studies. These approaches are a potentially useful tool for robust development of drug products and risk assessment of clinical BE studies.
AUTHOR CONTRIBUTIONS
L.W. and B.J. wrote the manuscript. L.W., P.Z., J.C., and B.J. designed the research. L.W., P.Z., T.L., D.Y., Q.J., and B.J. performed the research. H.L., Z.R., and B.J. analyzed the data.
FUNDING INFORMATION
This work was supported by the National Major Science and Technology projects of China (2020ZX09201022).
CONFLICT OF INTEREST STATEMENT
All the authors declare that they have no conflict of interest.
Supporting information
Figure S1
Data S1
Tables S1–S7
ACKNOWLEDGMENTS
The authors thank all participating subjects and clinical center personnel who contributed to conducting this study. Moreover, we thank Liwen Bianji (Edanz; www.liwenbianji.cn/) for basic language editing of a draft of this manuscript.
Wang Lu, Zhao P, Luo T, et al. Physiologically based absorption modeling to predict the bioequivalence of two cilostazol formulations. Clin Transl Sci. 2023;16:2323‐2330. doi: 10.1111/cts.13633
REFERENCES
- 1. Kato T, Nakagawa H, Mikkaichi T, Miyano T, Matsumoto Y, Ando S. Establishment of a clinically relevant specification for dissolution testing using physiologically based pharmacokinetic (PBPK) modeling approaches. Eur J Pharm Biopharm. 2020;151:45‐52. [DOI] [PubMed] [Google Scholar]
- 2. Jambhekar SS, Breen PJ. Drug dissolution: significance of physicochemical properties and physiological conditions. Drug Discov Today. 2013;18:1173‐1184. [DOI] [PubMed] [Google Scholar]
- 3. Abuhelwa AY, Williams DB, Upton RN, Foster DJ. Food, gastrointestinal pH, and models of oral drug absorption. Eur J Pharm Biopharm. 2017;112:234‐248. [DOI] [PubMed] [Google Scholar]
- 4. Hurst S, Loi CM, Brodfuehrer J, El‐Kattan A. Impact of physiological, physicochemical and biopharmaceutical factors in absorption and metabolism mechanisms on the drug oral bioavailability of rats and humans. Expert Opin Drug Metab Toxicol. 2007;3:469‐489. [DOI] [PubMed] [Google Scholar]
- 5. Ojala K, Schilderink R, Nykanen P, et al. Predicting the effect of prandial stage and particle size on absorption of ODM‐204. Eur J Pharm Biopharm. 2020;156:75‐83. [DOI] [PubMed] [Google Scholar]
- 6. Mitra A, Petek B, Bajc A, Velagapudi R, Legen I. Physiologically based absorption modeling to predict bioequivalence of controlled release and immediate release oral products. Eur J Pharm Biopharm. 2019;134:117‐125. [DOI] [PubMed] [Google Scholar]
- 7. Gao ZW, Zhu YT, Yu MM, et al. Preclinical pharmacokinetics of TPN729MA, a novel PDE5 inhibitor, and prediction of its human pharmacokinetics using a PBPK model. Acta Pharmacol Sin. 2015;36:1528‐1536. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Zhao P, Zhang L, Grillo JA, et al. Applications of physiologically based pharmacokinetic (PBPK) modeling and simulation during regulatory review. Clin Pharmacol Ther. 2011;89:259‐267. [DOI] [PubMed] [Google Scholar]
- 9. Wang L, Ruan Z, Yang D, et al. Pharmacokinetics and bioequivalence evaluation of erlotinib hydrochloride tablets: randomized, open‐label, 2‐period crossover study in healthy Chinese subjects. Clin Pharmacol Drug Dev. 2021;10:166‐172. [DOI] [PubMed] [Google Scholar]
- 10. Lee D, Lim LA, Jang SB, et al. Pharmacokinetic comparison of sustained‐ and immediate‐release oral formulations of cilostazol in healthy Korean subjects: a randomized, open‐label, 3‐part, sequential, 2‐period, crossover, single‐dose, food‐effect, and multiple‐dose study. Clin Ther. 2011;33:2038‐2053. [DOI] [PubMed] [Google Scholar]
- 11. Shao R, Yang DD, Ruan ZR, et al. Pharmacokinetic and bioequivalence evaluation of 2 tadalafil tablets in healthy male Chinese subjects under fasting and fed conditions. Clin Pharmacol Drug Dev. 2022;11:165‐172. [DOI] [PubMed] [Google Scholar]
- 12. Miyake M, Oka Y, Mukai T. Food effect on meal administration time of pharmacokinetic profile of cilostazol, a BCS class II drug. Xenobiotica. 2020;50:145‐149. [DOI] [PubMed] [Google Scholar]
- 13. Chapman TM, Goa KL. Cilostazol: a review of its use in intermittent claudication. Am J Cardiovasc Drugs. 2003;3:117‐138. [DOI] [PubMed] [Google Scholar]
- 14. Schror K. The pharmacology of cilostazol. Diabetes Obes Metab. 2002;4(Suppl 2):S14‐S19. [DOI] [PubMed] [Google Scholar]
- 15. Chatsiricharoenkul S, Nanchaipruek Y, Manopinives P, Atakulreka S, Niyomnaitham S. Bioequivalence study of 100‐mg cilostazol tablets in healthy Thai adult volunteers. Curr Ther Res Clin Exp. 2019;91:11‐16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Jereb R, Opara J, Legen I, Petek B, Grabnar‐Peklar D. In vitro‐in vivo relationship and bioequivalence prediction for modified‐release capsules based on a PBPK absorption model. AAPS PharmSciTech. 2019;21:18. [DOI] [PubMed] [Google Scholar]
- 17. Wang J, Chen J, Wang L, et al. Evaluating the bioequivalence of two pitavastatin calcium formulations based on IVIVC modeling and clinical study. Clin Transl Sci. 2023;16:85‐91. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1
Data S1
Tables S1–S7