Anti‐tuberculosis effect of isoniazid scales accurately from zebrafish to humans

Background and Purpose There is a clear need for innovation in anti‐tuberculosis drug development. The zebrafish larva is an attractive disease model in tuberculosis research. To translate pharmacological findings to higher vertebrates, including humans, the internal exposure of drugs needs to be quantified and linked to observed response. Experimental Approach In zebrafish studies, drugs are usually dissolved in the external water, posing a challenge to quantify internal exposure. We developed experimental methods to quantify internal exposure, including nanoscale blood sampling, and to quantify the bacterial burden, using automated fluorescence imaging analysis, with isoniazid as the test compound. We used pharmacokinetic–pharmacodynamic modelling to quantify the exposure–response relationship responsible for the antibiotic response. To translate isoniazid response to humans, quantitative exposure–response relationships in zebrafish were linked to simulated concentration–time profiles in humans, and two quantitative translational factors on sensitivity to isoniazid and stage of infection were included. Key Results Blood concentration was only 20% of the external drug concentration. The bacterial burden increased exponentially, and an isoniazid dose corresponding to 15 mg·L−1 internal concentration (minimum inhibitory concentration) leads to bacteriostasis of the mycobacterial infection in the zebrafish. The concentration–effect relationship was quantified, and based on that relationship and the translational factors, the isoniazid response was translated to humans, which correlated well with observed data. Conclusions and Implications This proof of concept study confirmed the potential of zebrafish larvae as tuberculosis disease models in translational pharmacology and contributes to innovative anti‐tuberculosis drug development, which is very clearly needed.


| INTRODUCTION
Tuberculosis (TB) is the leading cause of death from infectious diseases in adults, and Mycobacterium tuberculosis is becoming the deadliest pathogen on the planet (Furin, Cox, & Pai, 2019). The United Nations Sustainable Development Goals aim to eradicate the TB epidemic before 2030, but progress is stalling due to ineffectiveness of currently available treatments (United Nations, 2019). Drug development is a challenging, lengthy and costly process, generally requiring a decade for drugs to reach the market with estimated costs of $1-2.5 billion per approved new drug (DiMasi, Grabowski, & Hansen, 2016).
Development of anti-TB drugs is especially difficult, because of the laboratory biosafety issues (World Health Organization, 2012), the slow replication rate of M. tuberculosis, and the long duration of treatment and patient follow-up (Ginsberg & Spigelman, 2007). As a result, there is a clear and urgent need for innovations in the development of new TB treatments (Ginsberg & Spigelman, 2007;Van Wijk, Ayoun Alsoud, Lennernäs, & Simonsson, 2020). the zebrafish larva as model organism (Schulthess et al., 2018).
The zebrafish (Danio rerio) is increasingly used in biomedical research, because of its many advantages, which include high fecundity, rapid development, transparency throughout the first period of life, easy genetic modification, and limited ethical constraints (Rennekamp & Peterson, 2015;Schulthess et al., 2018;Van Wijk, Krekels, Hankemeier, Spaink, & Van der Graaf, 2016). Zebrafish larvae infected by Mycobacterium marinum, a close relative of M. tuberculosis, are an established TB disease model to study host-pathogen interaction (Meijer, 2016;Meijer & Spaink, 2011) and to screen for novel drugs (Carvalho et al., 2011;Ordas et al., 2015), with faster replication times and less biosafety risks than experiments with M. tuberculosis (Tobin & Ramakrishnan, 2008). Experiments in zebrafish larvae are commonly performed before the moment they start independent feeding (Rennekamp & Peterson, 2015), which is ethically preferable (Strähle et al., 2012). Other important pathophysiological aspects of TB such as granuloma formation and the drug effect thereon can be studied in later stages of zebrafish development (Van Wijk, Ayoun Alsoud, et al., 2020), but these aspects will not be considered further in this paper.
Imaging of zebrafish larvae infected with fluorescent mycobacteria (Stoop et al., 2011) allows for repeated longitudinal measurements from individual larvae, which reduces overall noise in data as biological and experimental variability can be quantified separately. This is in contrast to ex vivo bacterial burden organ count by colony-forming units (CFU) or mycobacteria growth indicator tube (MGIT) liquid media systems currently utilized in preclinical TB research (Kolibab, Yang, Parra, Derrick, & Morris, 2014;Kumar et al., 2014).
Translation of drug response between species is challenging and considerably limits drug development (Bartelink et al., 2017). Currently, pharmacological treatment of zebrafish larvae is performed by dissolving drugs into the water in which the larvae swim, without taking into account how much drug is actually taken up by the larvae.
Translating pharmacological response between species however requires quantification of the drug exposure at the site of action as a basis for the quantification of the exposure-response relationship (Morgan et al., 2012;Van Wijk, Ayoun Alsoud, et al., 2020). Although such quantification is a challenge because of the small size of the larvae, we have developed new experimental methods to quantify internal drug exposure based on ultra-sensitive analytical techniques and a novel method for nanoscale blood sampling. Pharmacological modelbased approaches can then be used to quantitatively link the internal What is already known • The zebrafish is a well-known tuberculosis disease model for qualitative efficacy screens in drug development.
• The information content and translational potential of qualitative assessments are low.

What this study adds
• Isoniazid exposure-response relationship was quantified in zebrafish larvae using integrated experimental and computational innovation.
• With pharmacokinetic-pharmacodynamic modelling, quantitative translation of pharmacological findings from zebrafish larvae to humans is confirmed.

What is the clinical significance
• Adding zebrafish larvae to the translational drug development pipeline could expedite development of antituberculosis drugs. van WIJK ET AL. exposure over time (pharmacokinetics) of anti-TB drugs to bactericidal response in the zebrafish larvae observed by fluorescence microscopy (pharmacodynamics). The exposure-response relationship that is thus obtained is the basis for translational pharmacology to higher vertebrates, including humans.
Here, for the first time, we present an integration of experimental and computational approaches in preclinical TB research, using zebrafish larvae infected with M. marinum and treated with increasing doses of waterborne isoniazid, from 0.25 to 10 times the minimum inhibitor concentration (MIC) (3.75-150 mgÁL −1 ). The internal exposure is quantified in homogenates and blood samples of the larvae, and the bacterial burden is quantified by automated fluorescence image analysis, according to the design in Figure 1. Pharmacokineticpharmacodynamic modelling was performed to quantify the exposure over time and the exposure-response relationship. Isoniazid was chosen because it is known to have the largest early bactericidal activity for single drug treatments among the current standard of care drugs against TB (Jindani, Aber, Edwards, & Mitchison, 1980). The quantified exposure-response relationship in the zebrafish larvae together with simulated concentration-time profiles in TB patients was utilized to translate the findings on isoniazid response in the zebrafish larvae to humans. A quantitative comparison with reported observations from patients was made as a proof of concept, to assess translational value of this new disease model in anti-TB drug development.

| Study design
Zebrafish embryos of the VUmc wild-type strain were dechorionated and infected with M. marinum strain E11 at 28 hours post fertilization (hpf) and kept at 28 C throughout the experiment. At 2 days post infection (dpi), the bacterial burden in zebrafish larvae was quantified by fluorescence microscopy, and waterborne treatment with isoniazid was commenced at external concentrations of 0, 3.75, 7.5, 15, 30, and 75 mgÁL −1 corresponding to 0, 0.25, 0.5, 1, 2, and 5× MIC. Quantification of the bacterial burden was repeated within individual larva at 3 and 4 dpi to assess individual early bactericidal response. Internal isoniazid exposure was quantified by LC-MS/MS of whole zebrafish larval homogenate samples as well as in larval blood samples in a parallel experiment with uninfected larvae treated with external isoniazid concentrations of 7.5, 15, 30, 75, and 150 mgÁL −1 corresponding to 0.5, 1, 2, 5, and 10× MIC. All replicates are biological replicates; technical replicates were not obtained. Figure 1 shows a schematic overview of the study design.

| Zebrafish husbandry
Zebrafish were maintained and handled following international consensus protocols (Westerfield, 2000 Adult zebrafish were set-up for breeding overnight, and in the morning at lights on, separators between males and females were removed, and fertilized eggs were collected within 30 min of fertilization. Eggs, embryos, and larvae were kept in embryo medium (60 μgÁml −1 Instant Ocean Sea Salts [Sera, Heinsberg, Germany] in demineralized water, daily refreshed) at 28 C. For the duration of treatment, larvae were kept in embryo medium with isoniazid at 28 C until imaging or sampling.

| Internal exposure of isoniazid in zebrafish larvae
Internal exposure of isoniazid in zebrafish larvae was quantified as drug amounts in larval homogenates or as drug concentrations in blood samples after waterborne treatment with isoniazid concentrations at external concentrations of 0.5, 1, 2, 5, and 10× MIC (7.5, 15, 30, 75, and 150 mgÁL −1 ). Experiments were performed on different days per dose, and the experimentalist was unblinded to the external concentration when blood samples or homogenates were obtained. As no outcome measurements were acquired at this stage, this was not expected to result in measurement bias.
Homogenate samples of pooled samples with five larvae at time points 0.25, 0.5, 1, 2, 3, 6, 8,9,18,20,22,24,26,32,42,44,46, 48, F I G U R E 1 Experimental study design. Fertilized eggs at 0 days post fertilization (dpf) were harvested and injected with 200 CFU Mycobacterium marinum at 1 dpf. After 2 days of establishing the infection, fluorescence imaging was performed (orange arrow; n ≥ 20 larvae per group), and the treatment with isoniazid dissolved in the external treatment medium (0.25-10-fold MIC, MIC = 15 mgÁL −1 , and control) was started. Fluorescence imaging was repeated daily (orange solid arrows). In satellite larvae groups, destructive homogenate and blood samples for internal isoniazid exposure quantification were taken from 0 to 50 h of treatment (blue dashed arrows) and 50 h after start of treatment were obtained at n = 3, except for external concentration of 1× MIC (15 mgÁL −1 ), which was repeated to check for inter-day variability, thus resulting in n = 6 replicates.
Zebrafish larvae were washed 4 times with 20/80 methanol/water (v/v) using Netwell inserts (Corning Life Sciences B.V., Amsterdam, The Netherlands) and transferred to Safe-Lock tubes (Eppendorf Nederland B.V., Nijmegen, The Netherlands). Excess volume was removed, and 50 μl of 200 ngÁml −1 isoniazid-D4 internal standard was added after which the larvae were snap frozen in liquid nitrogen and stored at −80 C until quantification.
Blood samples (n = 4) were taken after 48 h of treatment at the highest external concentration of 150 mgÁL −1 to ascertain quantifiable levels, using a previously published method (Van Wijk, Krekels, Kantae, Ordas, et al., 2019). In short, zebrafish larvae of 5 dpf were washed 4 times as described above while being unanaesthetized to prevent bradycardia and decreased sample yield. The larvae were superficially dried and transferred to an agarose microscopy slide, An image of each blood sample was captured to calculate the obtained blood volume before the sample was injected into a 2-μl heparin droplet (5 IUÁml −1 ) and pooled into a 0.5-ml tube (Eppendorf Nederland B.V.). Blood samples were kept at −80 C until quantification. Table S1 reports the number of larvae sampled per pooled sample, as well as the blood volume and isoniazid concentration per time point.

| Quantification of isoniazid in homogenate and blood samples
Isoniazid was quantified by LC-MS/MS. Derivatization with cinnamaldehyde was performed to achieve adequate retention for LC separation. The operator was unblinded, and to reduce measurement bias, all samples were randomized prior to injection into the LC-MS/MS, and peak processing was automated (LCQuan software v. 2.7, Thermo Fisher Scientific, Breda, The Netherlands).
Samples of whole zebrafish larva were thawed, and 100 μl methanol and 100 μl of 0.5 mm zirconium oxide bullets (NextAdvance, New York, USA) were added. The samples were homogenized using a Bullet Blender (NextAdvance) for at least two rounds of 5 min at The Netherlands). The temperature of the column was maintained at 40 C. Gradient elution was performed with two ultra HPLC pumps using methanol/water mixtures with 0.01% formic acid. The gradient started at the time of injection and increased from a 49/51 methanol/water v/v ratio to a 72/28 methanol/water v/v ratio within 4 min. The column was flushed with 95/5 methanol/water v/v ratio starting at 4.1 min for 2.4 min, after which the system was equilibrated to initial conditions. Within the MS system, the vaporizer temperature was set at 300 C and capillary temperature at 250 C. Sheath gas pressure was 40 psi, and multiple reaction monitoring was used to quantify the isoniazid derivative (MH + = 252.1 m/z) and the isoniazid-D4-adduct (MH + = 256.1 m/z). For the isoniazid derivative, the fragments were 79.01 and 121.01 m/z, and for the internal standard derivative, the fragments were 83.10 and 124.96 m/z. The sheath gas pressure was 40 psi, auxiliary gas pressure was 15 psi, S-lens RF amplitude was 63 V, and capillary pressure was 1.190 mTorr. Detection limit was 0.5 ngÁml −1 , and lower limit of quantification (LLOQ) was 1.75 ngÁml −1 . The LC-MS/MS method was validated according to the US Food and Drug Administration guidelines (US Food and Drug Administration, 2018). LCQuan software was used for data acquisition where isoniazid peak area was corrected by internal standard peak area, and calibration was performed with weighted linear regression using 1/y as weighting factor. Of the pharmacokinetic data points, 3% were below the LLOQ, and 1% were above the highest calibration standard; these samples were excluded from the analysis (Beal, 2001). van WIJK ET AL.
The experimenter was unblinded to the dose, but the potential of measurement bias was limited as image analysis was automated.
Automated image analysis was performed as reported before (Nezhinsky & Verbeek, 2010;Stoop et al., 2011), where count of pixels with fluorescence is assumed to correlate directly with the bacterial burden. Possible differences in the bacterial burden between individual larvae at the start of treatment were tested by nonparametric Kruskal-Wallis test; 12.7% (16/126) of the larvae were removed from the data set due to improper injection, developmental defects (e.g., cardiac oedema), or death due to mechanical damage from handling or otherwise.
2.7 | Quantification of the exposure-response relationship for isoniazid in zebrafish larvae Data were analysed by non-linear mixed effects modelling, which was used to develop a pharmacokinetic-pharmacodynamic model that In the pharmacokinetic component of the model, the homogenate and blood sample data were fitted simultaneously. The treatment medium was represented as depot compartment from which a firstorder absorption rate constant into a one-compartment model with distribution volume and linear or non-linear (Michaelis-Menten) elimination was estimated. Concentration in the treatment medium was assumed to be constant, as supported by measurements of external drug concentrations ( Figure S1).
We expected age to be a covariate (predictor) on the absorption and elimination rate constants, but because the effects of age on absorption and elimination are indistinguishable at steady state, we have estimated a net effect of age as covariate for a net increase on absorption only, which reflects the relative increase (or decrease in case of a negative value) in absorption rate constant, compared to the increase in elimination rate constant. For this, we tested a linear, power, or exponential relationship. In addition, a discrete increase in in which Bac represented the bacterial burden, t time (h), k g the growth rate (h −1 ), and B max the maximum capacity of the system (log 10 fluorescence). Because no ceiling of bacterial growth was observed in the data, the maximum capacity of the system could not be estimated for the Gompertz or logistic growth function. Growth and decay could not be estimated separately as this was mathematically not identifiable, given the data. Therefore, a net effect was estimated for growth, which means a negative growth represents a net decay or kill of bacteria.
Based on the homogenate and blood sample data at 5 dpf, the distribution volume at this age was estimated. This distribution volume was subsequently scaled to 3 and 4 dpf based on total larval volume quantified previously (Guo, Veneman, Spaink, & Verbeek, 2017).
The pharmacokinetic component of the model converted total isoniazid amounts from homogenates to concentrations using these distribution volumes (Equation 4).
The exposure-response relationship was linked to the bacterial burden as either inhibition of growth or as kill term as shown for the exponential growth function as example in Equations 8 and 9, respectively.
Biological (inter-individual) variability was tested on the estimate of the inoculum, as well as on the parameters of the exposureresponse relationship and reported as coefficient of variation. A proportional error model, parameterized as an additive error on the log 10 transformed data, was used to describe the experimental (residual) variability in the bacterial burden.
Non-linear mixed effects modelling does not assess statistical differences in outcome measures between treatment groups, rather it assesses whether inclusion of model features is statistically supported.
In this framework, model selection was performed based on the likelihood ratio test between nested models, in which a drop in objective function value of 3.84 corresponds to P < 0.05 between models with a single degree of freedom difference, assuming a χ 2 distribution. Precision of the estimates of structural parameters was considered to be acceptable when relative standard errors of these estimates remained below 50%, which is a second means of assessing that the model is supported by the data. The physiological plausibility of parameter estimates and visual assessment of goodness-of-fit plots (Nguyen et al., 2017) were also used for model selection.

| Translation of isoniazid response to humans
To quantitatively compare the findings for isoniazid response in zebrafish infected with M. marinum to findings in humans infected with M. tuberculosis, the exposure-response relationship obtained in the zebrafish larvae was translated to humans as a proof of concept. zebrafish, which is in logarithmic phase, and the patients, which are assumed to be in stationary phase (150 days of infection) , was taken into account by scaling the isoniazid drug response as quantified previously (Clewe, Wicha, de Vogel, de Steenwinkel, & Simonsson, 2018). The ratio of the maximum isoniazid-induced kill rates for the logarithmic and the stationary phase was utilized as translational factor to account for the difference in infection stage . Inoculum of the simulation was set at the mean of the reported inoculi in humans (Hafner et al., 1997;Johnson et al., 2006;Li et al., 2010). Third, the obtained isoniazid response was quantitatively compared to published bacterial burden data in humans (Hafner et al., 1997;Johnson et al., 2006;Li et al., 2010).

| Materials
Isoniazid, cinnamaldehyde, =tricaine and PVP40 were acquired from Sigma-Aldrich and isoniazid-D4 internal standard from Santa Cruz Biotechnology (Santa Cruz, USA). Nanopure water was used from a

| Mycobacterium marinum bacterial burden in zebrafish larvae upon isoniazid treatment
The bacterial burden of M. marinum was quantified through fluorescence imaging with repeated measurements of each individual larva.
Doses ranging from 0.25 to 5× MIC were chosen based on a feasibility study and on the fluorescence detection limit. Figure  Internal exposure linearly increases with dose, and steady state amounts increase with age, suggesting increased net absorption. Open symbols show observations below LLOQ highest dose of 5× MIC, which corresponded to 1× MIC blood concentration, showed a decline in the bacterial burden over time.

| Quantification of the exposure-response relationship for isoniazid in zebrafish larvae
To quantify the exposure-response relationship, a sequential modelling approach was performed. First, the internal exposure over time was quantified in the pharmacokinetic component of the model, after which this was linked to the isoniazid response in the final pharmacokinetic-pharmacodynamic model (Figure 4).
A one-compartment model with first-order absorption and first-order elimination best described the data on internal exposure. Because the larvae are still developing, both absorption and elimination are expected to increase with age (Van Wijk, Krekels, Kantae, Harms, et al., 2019). The data show an increase in the steady state amounts with increasing age, suggesting that absorption rates increase faster than elimination rates. A net increase of absorption with age was found, which reflects a faster increase in absorption compared to the increase in elimination. Age was included as predictor (covariate) on the absorption rate constant (k a ) in two ways: first, in an exponential relationship per hpf, and second, based on knowledge on the physiology of the GI tract, which opens between 3 and 4 dpf (Van Wijk, Krekels, Kantae, Harms, et al., 2019), as discrete increase at 4 dpf (Equation 10).
in which k a,0 is the absorption rate constant at the median age of 101 hpf, k a,hpf is the constant in the exponential covariate relationship, and k a,GI is the discrete factor with which the absorption rate constant increased at 4 dpf. A linear covariate relationship was statistically significantly worse (P < 0.001) compared to an exponential relationship, and a power relationship was statistically similar (P > 0.1) but resulted in worse precision of the parameter estimates.
Precision of all pharmacokinetic parameters in the final model was acceptable (relative standard errors <36%), with only the relative standard error of k a,GI being slightly higher than the cut-off of 50%.
Removing the effect of the opening of the GI tract worsened the fit significantly (P < 0.05), and as the opening of the GI tract was physiologically expected to affect absorption, the relationship was retained despite the slight imprecision of the obtained estimate. The precision of the remaining model parameters confirmed that the obtained model and parameter values are supported by the data. Goodness-offit plots further confirm an unbiased fit of the data by the model (Figure S2). A visual predictive check is provided in Figure 5, which showed good prediction of the typical trends and a slight overprediction of the variability of the observed data by the model.
The data of bacterial burden did not support separate estimation of growth and decay; therefore, only the net effect of these two processes was estimated in the model, which was found to be best described with an exponential growth model. In case of bacterial kill exceeding bacterial growth with sufficient isoniazid response, this net growth rate would become negative. A linear exposure-response relationship for inhibition of bacterial growth fitted the data best. The biological variability between larvae was quantified by inclusion of inter- individual variability on the inoculum (coefficient of variation 204%) and slope of the drug response (coefficient of variation 50.5%).
The experimental variability was very reasonable with a coefficient of variation of 36.3%. The predicted isoniazid response in zebrafish larvae is shown in Figure 6 (individual predictions: Figure S3), with a clear increase in antibacterial response with increasing external concentrations, which is in line with the observations. Goodnessof-fit plots show unbiased model fit ( Figure S4). Parameter estimates of the final pharmacokinetic-pharmacodynamic model are given in Table 1.
As a proof of concept for the translation of the quantified exposureresponse relationship in zebrafish larvae to humans, the relationship for isoniazid in zebrafish larvae was translated to humans, assuming the isoniazid response of M. tuberculosis in humans to be similar to the isoniazid response of its close relative M. marinum in the larvae. Two translational factors were taken into account . The first corrected for the difference in sensitivity to isoniazid between M. marinum and M. tuberculosis as reported by a difference in MIC.
The second corrected for the difference in stage of infection between the fresh experimental infection and more chronic clinical infection, where it was assumed that patients start treatment 150 days after initial infection .
The exposure-response relationship quantified in the current study was linked to simulated isoniazid concentration-time profiles using a previously published pharmacokinetic model for isoniazid in humans (Wilkins et al., 2011). Simulations with three human doses were performed: a subtherapeutic and a super therapeutic dose of 150 and 450 mg, in addition to the recommended dose of 300 mg.
Reported observations in patients of M. tuberculosis bacterial burden quantified in sputum after isoniazid monotherapy of daily doses of 300 mg (Hafner et al., 1997;Johnson et al., 2006;Li et al., 2010) served as quantitative comparison for our simulated isoniazid effect.

| DISCUSSION
There is a clear need for innovation in anti-TB drug development.
With its high-throughput potential, its possibility for repeated fluorescence imaging to quantify infection, and its fast, cheap, and relatively safe experimentation, the zebrafish larva tuberculosis disease model combines the advantages of in vitro experiments within a wholeorganism vertebrate with its translational potential. Additionally, zebrafish embryos and larvae prior to independent feeding are not considered experimental animals, and no approval of an ethical committee is required for these experiments (EU, 2010;Strähle et al., 2012). In this work, we developed an experimental approach to acquire data on the internal drug exposure and on the bacterial bur-   Kantae, Ordas, et al., 2019). This methodology proved sensitive enough to quantify isoniazid in larval homogenates and blood samples of only nanolitres in volume. The latter is especially of importance because blood concentrations are required to scale total amounts in homogenates to blood concentrations, which is essential for establishing a drug concentration-effect relationship, which is the basis for inter-species translation. The stateof-the-art blood sampling method cannot yet be performed in high throughput mode, but automation based on previously developed automated injection systems is being developed (Spaink et al., 2013).
Utilizing fluorescence imaging in transparent larvae to establish the response of mycobacterial burden to isoniazid has clear advantages over CFU plating, as the latter has been reported to show large sample-to-sample variability (Gillespie, Gosling, & Charalambous, 2002). An important advantage of fluorescence imaging in zebrafish larvae is the possibility of repeated longitudinal measurements of the bacterial burden within a single individual, which is uncommon in preclinical TB research. These repeated measurements not only suppress noise by distinguishing biological from experimental variability but also reduce the number of subjects needed in an experiment, which is ethically preferable. Additionally, fluorescence imaging of bacteria does not require these bacteria to grow on solid or in liquid media and will include both multiplying and non-multiplying (i.e., dormant) bacteria. Currently, our imaging set-up is restricted by its fluorescence detection limit to clearly quantify mycobacterial kill, which is why the F I G U R E 7 Translation of isoniazid response in zebrafish larvae to humans, using a model-based pharmacokinetic-pharmacodynamic approach. (a) isoniazid concentration-time profile (median: solid line, 80% prediction interval: shaded area) after 7 days of daily isoniazid doses of 150, 300, and 450 mg as simulated from a previously published pharmacokinetic model (Wilkins et al., 2011). (b) Simulated median (solid line) and 80% prediction interval (shaded area, including biological and experimental variability) bacterial burden in CFUÁml −1 sputum based on the human isoniazid concentration-time profile for 1,000 individuals per dose group. Concentration-time profiles were linked to the exposure-response relationship quantified in zebrafish larvae together with the translational factors on isoniazid sensitivity (MIC) and stage of infection (logarithmic vs. stationary). Translated response corresponds well to the observed bacterial burden in sputum, from Hafner et al., (1997;triangles), Johnson et al., (2006;squares) and Li et al., (2010;circles). Orange part of the prediction is extrapolated in time from the 48 h of treatment studied in the zebrafish, shown in black highest dose of 10× MIC was not tested in the bacterial burden study.
Imaging systems are continuously being improved, pushing the detection limit to the individual mycobacterium (Greenwood et al., 2019).
\Mycobacterium marinum and M. tuberculosis show different sensitivities to isoniazid. Isoniazid MIC against M. marinum ranges between 1.6 and 32 mgÁL −1 (Aubry et al., 2000;Boot et al., 2016;Boot et al., 2017;Weerakhun, Hatai, Murase, & Hirae, 2008), in concordance with the value obtained in our analysis. The MIC against M. tuberculosis is lower and ranges from 0.016 to 0.2 mgÁL −1 (Budha, Lee, Hurdle, Lee, & Meibohm, 2009;Gumbo et al., 2007;Hemanth Kumar et al., 2016;Jayaram et al., 2004;Schön et al., 2009). This difference in sensitivity was taken into account when predicting isoniazid response in humans by scaling the slope with the ratio of MICs as the scaling term, as shown earlier as a method for handling differences in sensitivities to drugs between strains . The difference in stage of infection between the logarithmic growth upon a fresh experimental infection in the zebrafish here and the chronic, stationary infection of patients starting treatment was taken into account as well. This translational factor has been presented before and is important to consider when translating results from a well-controlled experimental context to the real world .
The pharmacokinetic-pharmacodynamic model developed here was data driven. The internal exposure over time could be described  Van der Sar, et al., 2020).
The reported bacterial burden in humans after daily 300mg isoniazid monotherapy fell within the 80% prediction interval (including biological and experimental variability) obtained from simulated concentration-time profiles in humans and the exposure-response relationship quantified in zebrafish larvae here. The median simulated decline in the bacterial burden of 0.7-0.9 log 10 CFUÁml −1 Áday −1 after 2 days of treatment, has been reported in humans before (Jindani, Doré, & Mitchison, 2003). The large variability in the prediction interval from this work was largely due to the high biological variability in the inoculum, resulting from the establishment of infection during the first 2 days and the slope of the linear drug response quantified in zebrafish larva. Care must also be taken when extrapolating a linear exposure-response relationship to exposures outside the studied range, as well as extrapolating the response of treatment after the 48 h of treatment in zebrafish larvae. The treatment duration in the zebrafish larvae was not extended beyond 48 h, to remain within the ethical larval age limit (Figure 1) (Strähle et al., 2012).
In conclusion, we have developed a new experimental and computational approach to translate the pharmacokineticpharmacodynamic relationship of the early bactericidal response to an antibiotic in a zebrafish model of TB to humans. We propose that this approach is used in the search for novel TB treatments.