Model‐Informed Approaches to Optimizing Therapeutics in the African Patient Populations

We explore model-informed approaches to address pharmacokinetics variability in African populations due to genetic differences in the expression and activity of drug metabolizing enzymes and potential impact of Mycobacterium tuberculosis-mediated drug metabolism. Application of physiologically based pharmacokinetic models that factor in genetic variants predicted from artificial intelligence or built using in vitro drug metabolism data from African liver subcellular fractions with integrated effects of Mycobacterium tuberculosis can help design doses tailored to African populations. Pharmacokinetics variability arises from individual differences in how drugs are absorbed, distributed, metabolized, and excreted by the body. Pharmacokinetic variations can be influenced by several intrinsic and extrinsic factors, including genetic polymorphisms, weight, sex, age, disease state, or polypharmacy.1 Genetic factors such as polymorphisms in drug metabolizing enzymes can account for 20 to 95 percent patient variability in response to individual pharmacotherapy.1 Understanding the genetic diversity in genes encoding for drug-metabolizing enzymes is essential in facilitating the development of more effective therapeutics tailored to diverse populations. Africa exhibits higher genetic diversity than other populations worldwide, a characteristic that extends to variability within genes encoding the cytochrome P450 enzyme system and drug transporter proteins.2 It is noteworthy that genetic variability in African populations is not representative of that in the African American population and therefore cannot be assumed to be similar. Variations in gene sequence ultimately translate into decreased, normal, increased, or no enzyme function. This gives rise to metabolic phenotypes such as poor metabolizers, normal metabolizers, and fast metabolizers, which are associated with adverse drug reactions, suboptimal treatment efficacy, and poor treatment outcomes, respectively. This genetic variability has contributed to the high incidences of adverse drug reactions and poor therapeutic outcomes, especially in tuberculosis (TB) and human immunodeficiency virus (HIV) cases reported in Africa.2 Hence, variations in pharmacokinetics pose challenges and create disparities in treatment outcomes among the genetically diverse African populations. These challenges are compounded by lack of sufficiently powered clinical pharmacokinetics or pharmacogenetics studies to accurately quantify the influence of the prevalent pharmacogenetic variants on drug exposure in most African populations.3 In addition, despite Africa accounting for around 15% of the global population, it remains understudied and largely absent from biomedical research. Therefore, it is indispensable to adopt model-informed strategies to address disparities in therapeutic outcomes among African patient populations. These approaches involve the application of data-driven models to guide decision-making and optimize therapeutic interventions for African patient populations. In the absence of African-derived biobanks, this requires the building or refining of existing physiologically based pharmacokinetic (PBPK) models that factor in African genetic variants predicted from artificial intelligence (AI) as illustrated in Figure 1. Here, we propose in vitro and in silico methods that may help explain pharmacokinetic variability and inform dose regimen optimization. By incorporating information such as genetic factors, comorbidities such as TB infection, and treatment response patterns, healthcare providers can tailor therapies to the specific needs of African patients. Tuberculosis (TB) is a serious infectious disease that is caused by Mycobacterium tuberculosis (Mtb) and mostly affects not only the lungs but also other organs. Treatment typically involves a combination of antibiotics taken over a long period (4–6 months for drug sensitive TB and 6–20 months for drug-resistant TB) to eliminate the bacteria and prevent the development of drug-resistant strains. The burden of Mtb is high, particularly in low- and middle-income countries of Africa where co-morbidities are prevalent. The burden of noncommunicable diseases (NCDs) in Africa such as diabetes and cardiovascular diseases are a growing concern, accounting for over 30% of the continent's mortality. These NCDs also account for the most common comorbidities in tuberculosis patients. The impact of comorbid conditions on the PK of drugs used to treat Mtb is a crucial area of study, as these interactions can potentially affect treatment outcomes.4 Understanding how comorbidities such as hypertension, diabetes and HIV infection can influence the PK is essential for optimizing therapeutic regimens. Given the high burden of Mtb coinfections worldwide, it is important to explore the complex interplay between the Mtb infection and other comorbidities to develop more effective treatment strategies. Metabolism of drugs is mostly attributed to liver enzymes particularly the cytochrome P450s. Mtb possesses its own set of enzymes (Table 1) capable of catalyzing biotransformation reactions on xenobiotics, including some antituberculosis drugs. These enzymes enable Mtb to modulate the activity of drugs through various mechanisms, such as deactivation or activation of drugs, ultimately influencing the effectiveness of TB treatment.5, 6 The extent of Mtb drug metabolism is dependent on the bacterial load. It was shown in a large cohort of African patients receiving first-line antituberculosis drugs (isoniazid, rifampicin, pyrazinamide, and ethambutol) that Mtb bacterial load persisted to an average amount of 2 log10 CFU/mL after 12 weeks of treatment from an average baseline of 6 log10 CFU/mL.7 In the case of a drug that is metabolized by Mtb, it is expected for drug clearance to decrease or normalize after a longer duration of treatment. Hence, Mtb-mediated drug metabolism may potentially impact the efficacy and treatment outcomes of tuberculosis or other comorbidities whose drug treatment may be prone to Mtb-mediated metabolism. PBPK models are sophisticated mathematical tools that simulate the PK of drugs within the body. These models rely on accurate data to represent physiological processes, and the availability of such data from diverse populations is essential to ensure the models' reliability across different demographic groups.8 Hence, the models constructed using data derived from African liver subcellular fractions (microsomes, cytosol, and S9 fraction) can potentially hold significant promise in advancing our understanding of drug metabolism and pharmacokinetics within African populations. These models offer a nuanced depiction of how drugs interact with the body at a molecular level, considering factors such as enzyme activity, protein binding, and tissue distribution specific to individuals from the African populations. It is imperative to stress that in vitro data generated from African liver subcellular fractions is not generalizable across all African populations due to genetic diversity. Likewise, the PBPK model can be extended to model the in vivo effect of Mtb by integrating Mtb-mediated in vitro kinetics data. This makes it possible to estimate the contribution of Mtb to total drug clearance. Hence, PBPK models that utilize data from African liver subcellular fractions and in vitro Mtb kinetics can offer a means to address this disparity by providing a framework to simulate drug behavior in populations typically underrepresented in clinical studies. This approach has the potential to enhance stratified medicine initiatives and facilitate the development of therapeutics tailored to the specific needs of African populations, ultimately improving therapeutic outcomes, and reducing disparities in clinical outcomes. Harnessing the power of AI to predict prevalent drug-metabolizing gene variants and transporter proteins within African populations holds immense potential for enhancing the accuracy and applicability of PBPK models. However, there is paucity of pharmacogenetic data in African populations, which accounts for less than 5% of the pharmaco-genomics data in the PharmGKB database.9 This challenge can be addressed through the implementation of the AI transfer learning approach. Transfer learning involves re-using the knowledge gained from a dataset to improve performance on another related dataset. In the context of pharmacogenetics, AI transfer learning can be used to predict drug-metabolizing gene variants and/or drug transporter proteins prevalent in African populations.9 These predictive models can help elucidate the genetic factors influencing pharmacokinetics, thereby facilitating the integration of relevant gene variants into PBPK models. This integration of AI-predicted African prevalent drug-metabolizing gene variants would enable PBPK models to simulate pharmacokinetics within African populations. These models can offer valuable information on pharmacogenetic diversity and its impact on drug exposure and safety. We hold the perspective that the synergy between AI-predictive models and PBPK modeling represents a significant advancement toward personalized medicine approaches that consider the genetic diversity and the influence of Mtb-mediated drug metabolism on pharmacokinetics. Designing optimum treatment regimens using nonlinear-mixed effects (NLME) modeling and simulation represents an advanced approach to treatment optimization. The NLME modeling approach allows for the incorporation and modeling of Mtb-mediated metabolism or genetic factors as covariates into these mathematical models, enabling the exploration of complex pharmacokinetic scenarios. By leveraging pharmacokinetic data predicted from PBPK models, NLME modeling can characterize the variability in drug exposure and identify factors such as genetic factors, demographics, and disease status such as TB as significant covariates that influence treatment outcomes. Simulation techniques based on NLME models enable the exploration of different dosing strategies, treatment durations, and combination therapies in silico,10 providing valuable insights into the optimal design of treatment regimens for specific African patient populations. Improving therapeutic outcomes entails administration of the right dose or dose regimen. Model-informed approaches offer a promising avenue for optimizing therapeutics in the African patient population. By building PBPK models using African liver subcellular fractions or AI-predicted prevalent African gene variants (drug metabolizing enzymes) and integrating Mtb-mediated drug metabolism into PBPK models we can predict and capture pharmacokinetic variability. The NLME modeling of PBPK simulated pharmacokinetic profiles enhance the understanding of pharmacogenetic variability, which enables the development of more tailored and precise treatment regimens, minimizing the risk of adverse drug reactions and optimizing therapeutic outcomes among African populations. K.C. is the Neville Isdell Chair in African-centric Drug Discovery and Development and thanks Neville Isdell for generously funding the Chair. The authors declared no competing interests for this work.