Design of Host-Directed Therapies using Quantitative Systems Pharmacology Modelling

The overall outcome of Mtb disease and treatment is reliant on the integrated results of the molecular and cellular events, and their reflection at tissue, organ, and host level dynamics occurring at different time scales. As such, it can be challenging to predict patient responses to different HDT strategies. Species differences in immune response characteristics make it more challenging to translate the results from preclinical studies to clinical scenarios. In addition, determination of the effects of treatments and disease progression in specific patient populations, can be challenging, i.e., in patients with weakened immune response and/or other conditions, patients with specific genotype known to affect certain pharmacology. QSP modelling can address these hurdles through quantitative integration of Mtb host-pathogen interaction mechanisms with PK and PD aspects of HDTs, making it a relevant tool to guide drug discovery and development of HDTs for Mtb. Here we discuss three main components of the QSP framework to evaluate HDTs for Mtb infection, (1) drug PK models, (2) host immune response models, and (3) pathogen dynamic models. The considerations for identification of drug- and system specific parameters to facilitate scaling, and the incorporation of variability are also discussed. Lastly, we discuss applications of the QSP models to evaluate HDTs. An overview of the QSP framework components and applications is provided inFigure 2 and Figure 3 , respectively.

Pharmacokinetics

Pharmacokinetics describes the concentration-time profile of drugs and is determined by absorption, distribution, metabolism. and elimination processes, which may differ between organisms. Consideration of concentration-effect relationships, and therefore the PK, is of essential value for design of HDT strategies. Mathematical PK models quantitatively characterize PK based on parameters accounting for the underlying processes.
Physiologically based PK (PBPK) models describe the concentration profiles in specific tissues of interest and are informed by both drug- and system-specific parameters. PBPK models are of relevance to scale PK between preclinical species and towards humans in a mechanistic-fashion. For Mtb infection, PBPK models describing lung exposure are of specific relevance. In addition, their mechanism-based approach allows for incorporation of drug-drug interactions, which often occur for Mtb combination therapies[92]. In the clinical phase, quantifying inter-patient variability in PK is important. Here, population PK (PopPK) models are of relevance, which capture inter-individual variation in underlying PK parameters that can be explained by specific patient-specific covariates[93]. It is furthermore helpful that because many HDTs involve approved drugs, often PK models are available already to characterize their PK[94,95].

Immune Responses

Models describing the key immune response components, such as dynamics of macrophage counts, cytokines, and CD4+ and CD8+ T lymphocytes are essential for QSP models to study HDTs. Systems biology models describing the host-Mtb interactions within the site of infection (lungs)[56] have been previously developed, and later linked with lymphatics[50] and blood circulations[96]. The states included in these models were resting-, activated-, and critically infected-macrophages, cytokines, such as IFN-γ, IL-10, and IL-12, immature- and mature- dendritic cells, CD4+ lymphocytes, and intra- and extra-cellular Mtb populations. The key feature of this model was contributions of various immune components on intra- and extra-cellular Mtb. The above-developed model was later expanded to include CD8+ cells dynamics in lungs and lymph[49,97]. The parameters in these models were identified from published human-derived or non-human primate (NHP) experimental results or model fitting to in vitro or in vivo (mice) data. These models can be expanded to include key drug targets involved in Mtb HDTs and their downstream effects on functional immune response changes and the quantitative interaction with Mtb bacteria.
To the best of our knowledge, there are currently no mathematical models available in literature describing HDT-relevant pathways, such as autophagy in Mtb infections; however, components and parameter estimates from single cell systems biology models[98–103] can be adapted and extended using experimental in vitro and in vivo data. For example, a HDT model containing key biological features of autophagy[98] including HDAC1-related components may be developed. The model parameters can be informed using prior knowledge available in literature[98] and data from in vitro experiments[40]. The model may describe dynamics of the phagocytic cells and zebrafish infection with Mm bacterial load overtime in HDAC1 inhibitors exposed macrophage cell cultures as compared to controls, and this would allow estimation of parameters relevant to HDAC1 effect. The simulations from the models may be compared with the experimental outcomes, preferably from different experimental conditions than the original experiments used for parameter estimation. This allows validation of the model structure and parameter estimates. In the above example, the simulations from the QSP model including autophagy components may be validated against data from zebrafish exposed to HDAC1 inhibitors (at various HDAC1 levels) experiments[40].

Pathogen Dynamics

Models for the population dynamics of pathogens include the effect of antimicrobial drug on the growth and inhibition-dynamics of Mtb bacteria and emergence of treatment resistance. In vitro and in vivo kill dynamic studies have enabled our understanding of parameters of Mtb growth rates[18], bactericidal and bacteriostatic effects of conventional anti-TB drugs[76], and resistance development rates of bacteria[104,105]. Through the use of PK/PD modelling, dosing strategies can be designed that optimize dosing schedules for maximal bacterial control and reduced risk of resistance development. The incorporation of immune cell effects on pathogen killing is a key required step to study the effects of HDTs on Mtb treatment. Published host-Mtb interaction models[50] can be updated to include contributions of key HDT components on pathogen killing, as well as pathogen evasion mechanisms. For example, an autophagy model may contain quantitative relationship between bacterial load, mTOR, and autophagy. This will allow evaluations and predictions of various mTOR inhibitors on Mtb clearance by autophagy.

Implementation and Applications of the QSP Modelling Framework

QSP modelling have successfully influenced various decision making processes at different stages starting from discovery to late phase development in various therapeutic areas[16] and offer potential for the challenges faced in translation and design of HDT (combination) treatments in Mtb infections. A QSP framework to translate and optimize optimal HDTs should contain a combination of aforementioned model components for PK of one or more (investigational) drugs, immune/host response and pathogen dynamics, including their interactions. Depending on the type of HDT drug studied, QSP models may be parametrized and/or adapted in specific ways, e.g., to capture the drug-specific parameters for PK, pathogen kill and immune system effects, and induction of specific immune system effects. Various considerations and applications of the HDT QSP modelling framework are discussed below.

Target Identification and Drug Discovery

QSP models integrate various host-pathogen interactions and drug PK/PD components; therefore, they can readily provide assessment of target engagement upon stimulation or inhibition of certain target molecules at various doses and affinities and its impact on overall treatment outcome. This allows evaluations of the iterative process of hypotheses generation, designing new experiments, hypotheses validation and/or generation of new hypotheses. This approach can be applied to evaluate known HDT targets and HDT candidate molecules, to discover new HDT targets, or to discover and evaluate new HDT molecules. With advances in technologies, applications of combining QSP modelling and machine learning approaches to screen virtual drug compounds to enable discovery of drugs with optimal PK/PD characteristics are being evaluated[106].

Translational Predictions

With increased complexity and innovation in design of new drugs within the last two decades, mechanistic QSP models are increasingly being applied to inform translation of the results across different experimental conditions and species[107,108]. The systematic incorporation of system-specific parameters not only for various species, such as zebrafish, mice and humans, but also incorporation of differences between in vitro systems and in vivo models, is crucial to enable translation towards clinical HDT treatment designs[77,86]. In some cases, i.e. for scaling from in vitro HFIM to humans, such scaling is already well studied[76], whilst further studies are needed for the host’s immune response components[109]. Consolidating immune-relevant differences between preclinical models and humans[109] may be challenging and resource intensive, as there are varying strains of models used across different experiments depending on the objectives of the experiments. On the other hand, the shown evolutionary conservation of the metabolic responses to mycobacterial infection in human patients and mice and zebrafish animal models show that basic disease symptoms such as wasting syndrome are not depending on species or varying strains[110]. Gene expression analysis data across species may be used to inform parameters of expressions of genes responsible for certain immune functions[111]. Such expression data studies can be used to predict metabolism in a whole-genome metabolic network theoretical modelling approach in various model organisms such as zebrafish[112]. Factors such as state or severity of infection, intensity of resistance, and sensitivity of drugs to bacterial strains (for example between Mtb and Mm) may also be applied within the QSP framework.

Variability and Precision Medicine

The presentation and severity of TB is variable amongst patients, and thus treatment responses, especially to HDTs, are variable. Many factors such as age, sex, genotypes, co-morbid conditions (HIV, type 2 diabetes) play role in determining the outcome of the disease and treatment. Thus, considering these factors in the QSP framework is very important. For example, known differences in PK and immune-response components for HIV co-infected TB patients may be incorporated in the framework, and extrapolate results from studies in TB patients to HIV-TB co-infected patients[113]. Many PopPK models have evaluated these factors’ impact on variability in PK of conventional anti-TB drugs[114], and thus can be included in QSP simulations framework. In addition to external factors, considering immune-response relevant endotypes is also important[115,116]. Technological advances within the last century enabled generation of large-scale data, including omics data. The large-scale omics data may enable us to better understand the inter-individual variations associated with the parameters of the QSP models[117,118]. For example, parameters, together with inter-individual variations in them, describing the expression of baseline state of immune response components within lymph nodes and blood were estimated using data from a flow cytometry analysis of blood leukocytes and genome-wide DNA genotyping from 1000 healthy humans[117]. In addition, parameters, together with inter-individual variations in them, describing fractions of various lymphocytes within tumour microenvironment were informed using transcriptomics data from cancer patients.[117] Gene expression analysis of omics datasets from total of 443 TB patients enabled stratification of the patients into two groups. One of the two groups was characterized by increased gene activity score for inflammatory response and decreased gene activity score for metabolism-relevant pathways, and patients in this group showed slower time to negative TB culture conversion and poor clinical outcome[115,116]. Similarly, gene expression data can be used to include variability in the QSP models and inform outcome of certain HDT treatment.

Selection of Optimal Dosing Regimens and Combination Therapies

QSP models are also suitable to evaluate various combination therapies with optimal dosing regimen efficiently and can be especially valuable for difficult to treat diseases, such as TB. A QSP model enabled simulations of multiple combination therapies and identified the most effective dual-drug combination for the treatment of advanced castration-resistant prostate cancer where effectiveness of immunotherapy was previously insufficient [119]. In the TB disease space, QSP modelling has recently been applied to predict patient outcome with intensive dosing regimen and to explore shorter treatment duration scenarios for conventional anti-TB drug therapy[18]. Overall, the use of QSP modelling can serve as a valuable tool to efficiently design and develop HDTs for treatment of TB.