Pharmacokinetics and pharmacodynamics of imatinib for optimal drug repurposing from cancer to COVID-19

Abstract

Introduction

In the randomized double-blind placebo-controlled CounterCOVID study, oral imatinib treatment conferred a positive clinical outcome and a signal for reduced mortality in COVID-19 patients. High concentrations of alpha-1 acid glycoprotein (AAG) were observed in these patients and were associated with increased total imatinib concentrations.

Aims

This post-hoc study aimed to compare the difference in exposure following oral imatinib administration in COVID-19 patients to cancer patients and assess assocations between pharmacokinetic (PK) parameters and pharmacodynamic (PD) outcomes of imatinib in COVID-19 patients. We hypothesize that a relatively higher drug exposure of imatinib in severe COVID-19 patients leads to improved pharmacodynamic outcome parameters.

Methods

648 total concentration plasma samples obtained from 168 COVID-19 patients were compared to 475 samples of 105 cancer patients, using an AAG-binding model. Total trough concentration at steady state (Ct_trough) and total average area under the concentration-time curve (AUCt_ave) were associated with ratio between partial oxygen pressure and fraction of inspired oxygen (P/F), WHO ordinal scale (WHO-score) and liberation of oxygen supplementation (O₂lib). Linear regression, linear mixed effects models and time-to-event analysis were adjusted for possible confounders.

Results

AUCt_ave and Ct_trough were respectively 2.21-fold (95%CI 2.07–2.37) and 1.53-fold (95%CI 1.44–1.63) lower for cancer compared to COVID-19 patients. Ct_trough, not AUCt_ave, associated significantly with P/F (β=-19,64; p-value=0.014) and O₂lib (HR 0.78; p-value= 0.032), after adjusting for sex, age, neutrophil-lymphocyte ratio, dexamethasone concomitant treatment, AAG and baseline P/F-and WHO-score. Ct_trough, but not AUCt_ave associated significantly with WHO-score. These results suggest an inverse relationship between PK-parameters, Ct_trough and AUCt_ave, and PD outcomes.

Conclusion

COVID-19 patients exhibit higher total imatinib exposure compared to cancer patients, attributed to differences in plasma protein concentrations. Higher imatinib exposure in COVID-19 patients did not associate with improved clinical outcomes. Ct_trough and AUCt_ave inversely associated with some PD-outcomes, which may be biased by disease course, variability in metabolic rate and protein binding. Therefore, additional PKPD analyses into unbound imatinib and its main metabolite may better explain exposure-response.

Graphical abstract

1. Introduction

Coronavirus disease 2019 (COVID-19) is caused by infection with severe acute respiratory syndrome-coronavirus 2 (SARS-CoV-2) and, in severe cases, is associated with diffuse alveolar damage, endothelial injury and vascular leakage into the alveoli (Gibson et al., 2020; Diamond, 2021). These pathological features have many similarities with those observed in acute respiratory distress syndrome (ARDS) (Teuwen et al., 2020; Smadja et al., 2021). A combination of accumulation of fluid, loss of hypoxic pulmonary vasoregulation due to endothelial dysfunction and over inflammatory processes contribute to a high mortality rate in severe COVID-19 (Nitsure et al., 2020). The long-term pulmonary and extra-pulmonary consequences for severe COVID-19 survivors are often disabling, and include cognitive decline and persistent skeletal-muscle weakness (Thompson et al., 2017). The function of the pulmonary vascular endothelium is to form a barrier between circulating fluid and inflammatory cells from the pulmonary interstitium and alveolar space. In COVID-19 ARDS, disruption of cell-cell adherence junctions in the endothelium leads to gaps and consequently vascular leakage (Aman et al., 2012; Rizzo et al., 2015). Key mediators in the regulation of vascular permeability as well as oxidant-induced endothelial cell injury are the Abl family nonreceptor tyrosine kinases c-Abl and Abl related gene (Arg, Abl2) (Aman et al., 2012; Rizzo et al., 2015). There are currently no effective treatments targeting endothelial barrier dysfunction.

Imatinib mesylate is a phenylaminopyrimidine derived compound that targets constitutive ATPase activity of tyrosine kinases and is approved by the Food and Drug Administration and European Medicine Agency for the treatment of chronic myeloid leukemia (CML) and gastrointestinal stromal tumors (GIST) (Peng et al., 2005). It has been demonstrated in vitro and in vivo that imatinib has protective effects on the endothelial vascular barrier function by c-Abl and Arg/Abl2 kinase inhibitory effects (Aman et al., 2012; Chislock and Pendergast, 2013). This evidence has raised interest in the repurposing of imatinib for improvement of vascular barrier dysfunction in COVID-19 patients. The recent CounterCOVID study assessed the efficacy of oral imatinib on clinical recovery from COVID-19. Hypoxemic COVID-19 patients were given 400 mg of oral imatinib for 9 days following a loading dose of 800 mg (Aman et al., 2021). Treatment with imatinib did not reduce the time to liberation of oxygen and ventilatory support in hospitalized COVID-19 patients. Patients in the imatinib group, however, had a reduced mortality, shorter duration of invasive ventilation and faster improvement of oxygenation status in invasively ventilated patients (Duijvelaar et al., 2022). These findings imply that imatinib may confer clinical benefit, particularly in critically ill patients (Aman et al., 2021; Duijvelaar et al., 2022).

The dosing schedule in the CounterCOVID study was based on the dosage used for the indications CML and GIST. The target inhibitory effects on Arg/Abl2 kinases in COVID-19, resemble the targets in these indications. Total imatinib target trough concentrations (Ct_trough) of 1000 μg/L for efficacy has been determined for CML/GIST (Takahashi and Miura, 2011; Farag et al., 2017). One study showed that only 50% reached a Ct_trough above this target, after receiving standard 400 mg daily dosing (Larson et al., 2008). Our preclinical studies demonstrated that the target concentration of imatinib to prevent endothelial leakage also resembles the target concentration in CML/GIST (Aman et al., 2012). For tyrosine kinase inhibitors (TKIs), including imatinib, a correlation between drug concentrations and effect (oncolytic response and toxicity) have been observed, making most TKIs good candidates for therapeutic drug monitoring (Larson et al., 2008; Cortes et al., 2009; Demetri et al., 2009; Picard et al., 2007; Kantarjian et al., 2010; Awidi et al., 2010; Li-Wan-Po et al., 2010; Singh et al., 2009). Most TKIs, including imatinib, show a wide inter-patient variability in pharmacokinetics (PK) and have a relatively small therapeutic window in cancer patients which further demonstrates the relevance of therapeutic drug monitoring during imatinib treatment (Cortes et al., 2009; Schmidli et al., 2005; Judson et al., 2005). High PK variability of imatinib was also observed in COVID-19 patients (Bartelink et al., 2021).

Therefore, we hypothesize that in COVID-19 patients similar associations between PK and clinical responses, survival and adverse events exist, compared to the observations in CML/GIST patients. Furthermore, imatinib, like most TKIs is a highly plasma protein-bound drug, approximately 95%, and mainly binds to albumin and alpha-1-acid glycoprotein (AAG) (Peng et al., 2005). Previously we observed high total imatinib concentrations in COVID-19 patients, compared to a cohort of CML/GIST patients (Bartelink et al., 2021). In these PK analyses, we showed that variability in the acute phase protein concentrations only partially explained the interpatient variability (Bartelink et al., 2021). Most studies on exposure-response relationships of TKIs were performed based on total plasma exposure estimates. However, for imatinib, the association improved when unbound instead of total (bound and unbound) concentration-time profiles were correlated with efficacy and toxicity in cancer (Widmer et al., 2008). In COVID-19 it remains to be elucidated how potential exposure-response associations can be best described and what the influence of higher total drug concentrations is on the effectivity of imatinib in this new indication (Benet, 2002). This study aims to find potential PKPD relationships following oral imatinib 400 mg administration in COVID-19 patients. Possible PKPD associations could be of relevance for repurposing of imatinib for endothelial barrier dysfunction in ARDS caused by multiple risk factors including COVID-19. We hypothesize that a relatively higher drug exposure of imatinib in severe COVID-19 patients leads to improved pharmacodynamic outcome parameters.

2. Methods

2.1. Study design

This imatinib PKPD study is a post-hoc analysis of the CounterCOVID study. Hospitalized patients, >18 years old, enrolled in the multi-center randomized double-blind placebo-controlled clinical trial, CounterCOVID, had a confirmed SARS-CoV-2 infection and had a peripheral oxygen saturation <94% or arterial partial oxygen pressure <9 kPa (Rizzo et al., 2015). The most important exclusion criteria were lympho- or thrombocytopenia, active liver disease, pre-existing heart failure or pulmonary disease, active treatment of hematologic or non-hematologic malignancies within 12 months before enrollment and concomitant treatment with medication that can cause strong drug-drug interactions (Rizzo et al., 2015). Patients were randomized 1:1 to oral imatinib treatment or placebo for 10 days. Imatinib-treated patients (N = 179) were given a loading dose of 800 mg on the first day of treatment, day 0, followed by 400 mg once daily on day 1–9.

2.2. Data and sample collection

Demographic and clinical data were collected during the study period. Clinical data were collected up to day 28 and then again on day 90 after randomization. Blood samples were collected on day 0 prior to the loading dose and again ∼ 4 and 8 h post-dose. From day 1 onward, samples were withdrawn once daily on days 1, 2, 3, 5, 7, and 10, or until hospital discharge or death. Plasma was collected after centrifugation of whole blood samples and stored at −80 °C until analysis. For the current study, imatinib-treated patients were included if they had ≥1 blood plasma sample available for PK-modeling. The total imatinib plasma concentrations and free drug concentration were determined using validated methods (Bartelink et al., 2021; Haouala et al., 2009; Bouchet et al., 2013). AAG concentrations were measured with enzyme-linked immunosorbent assay (Bartelink et al., 2021).

2.1.1. Pharmacodynamic outcomes

Outcomes of interest were clinical parameters of which data was collected during imatinib treatment and followed up on day 28 and 90 after randomization (Aman et al., 2021). Due to a low number of deaths (N =15) in the imatinib treatment group, mortality was not considered to be a relevant pharmacodynamic endpoint. Clinical parameters that were included for analysis included:

• •Being alive and >48-hour liberation from oxygen supplementation and ventilation (O₂lib), monitored untill 90 days after randomization.
• •The ratio between the oxygen saturation (SpO₂) and the fraction of inspired oxygen was used as estimate for the P/F, monitored during imatinib treatment. The fraction of inspired oxygen differs among methods of oxygen supplementation methods. Therefore, the P/F was corrected for the method of oxygen support. Ventilation and oxygen support methods differ in fraction of inspired oxygen. Therefore, the P/F was stratified for ventilated and non-ventilated patients.
• •The World Health Organization modified ordinal score for clinical improvement (WHO-score) during imatinib treatment, monitored untill 90 days after randomization (Marshall et al., 2020). Category 1 indicates that the patient was not hospitalized, and received no oxygen supplementation; 2. was not hospitalized, but received supplemental oxygen; 3. was hospitalized, without the use of supplemental oxygen; 4. was hospitalized and received supplemental oxygen using a nasal cannula or mask; 5. was hospitalized and received oxygen through non-invasive ventilation or high-flow devices; 6. was hospitalized and received invasive ventilation with no extra organ support; 7. was hospitalized and received invasive ventilation plus additional organ support: vasopressors, renal replacement therapy (RRT), or extra corporal membrane oxygenation (ECMO); and 8. died.

2.3. PK-estimation

Bartelink et al. recently published a study in which a large fraction of collected imatinib plasma PK-samples from the CounterCOVID study were used to predict PK-profiles of imatinib in COVID-19 patients (Bartelink et al., 2021). The PK-parameters and covariate associations of this previously published final covariate model were re-estimated and re-validated. The original model was based on 74 CounterCOVID patients and validated on an additional sample set of 60 CounterCOVID patients. In the current study, we analyzed the full PK-sample dataset of 168 patients. A mean prediction error below 20% was considered adequate. The applied AAG-PK-model is a first order absorption, one-compartment model in which imatinib binds nonlinearly to AAG and forms a complex (Drug_unbound + AAG $\rightleftarrows$ Drug_AAG), with an in vitro/in vivo estimated dissociation constant Kd and linear elimination of the unbound fraction (Bartelink et al., 2021). An exponential scale was used for the population parameters and a natural log transformation was applied to the covariates before including them into the final model. The previously observed effect of intensive care unit (ICU) admission in the 28-day study period on PK was re-evaluated. Dosing was continued during and after mechanical ventilation and duration of nasogastric tube feeding was included in the analysis. During mechanical ventilation in the ICU, imatinib tablets were solved in water and administered via a nasogastric tube. The effect of a nasogastric tube on bioavailability was estimated. The relevant covariates in the full dataset were included using a full covariate model, followed by a backward elimination procedure that was described previously (Bartelink et al., 2021). Visual predictive checks (VPCs) and forest plots were used to assess the predictions of the re-estimated AAG-PK-model and to define the clinically significant covariates on imatinib exposure.

Individual and independent post-hoc PK profiles were derived as estimated imatinib exposure in this study:

• •Trough concentration on day 3 of imatinib treatment (C_trough)
• •Average area under the concentration-time curve per day during imatinib treatment (AUC_ave)

The correlation of the total and unbound PK-parameters was assessed, and PK-parameters were included only if the correlation (R) was smaller than 0.70.

2.4. Potential confounders

Based on biological plausibility and from available literature, potential confounders for clinical outcomes were selected (Clift et al., 2022; Prozan et al., 2021; Horby et al., 2021; Williamson et al., 2020; van Erp et al., 2009). These include sex, age, the number of administered dosages, neutrophil-lymphocyte ratio, dexamethasone concomitant treatment, AAG, lactate hydrogenase, comorbid hypertension, cardiovascular disease and diabetes, CRP, smoking-status and baseline P/F and WHO-score. Potential confounders were considered as relevant if they influenced one of the PKPD associations by ≥10%. Relevant confounders were included into the final adjusted PKPD analyses.

2.5. PKPD analysis

For statistical analyses, patients were separated into 3 equal exposure category groups (tertiles), based on C_trough and AUC_ave. Statistical analyses were performed on the total imatinib treatment group as well as on the exposure categories. Regression analyses were then performed using exposure as a continuous variable. Linear regression was used to analyze the change in WHO-score between the start and end imatinib treatment. For the P/F during imatinib treatment, a linear mixed effects model was used for data analysis since P/F contains time series data of included patients. Survival analyses (Kaplan Meijer and COX regression analyses) were performed for O₂lib. All analyses were performed crude and adjusted.

2.6. Software

Data collected from electronic patient files were stored in Castor EDC https://data.castoredc.com. Rstudio (version.4.0.3; R Foundation for Statistical Computing, Vienna, Austria) was used to perform analyses and visualize data. Nonlinear mixed-effects modeling software NONMEM (version 7.3, Globomaxx LLC, Hanover, MD, USA) with Piraña Software (version 3.0, Certara) and Perl-speaks-NONMEM (PsN) were used for PK-modeling. Visual predictive check graphs (VPCs) were created using the VPC package (version 1.0.1; R).

3. Results

3.1. Patient characteristics

From 197 enrolled patients in the imatinib-treated group, 168 patients had ≥1 PK-sample available for analysis. Compared to the previous published PK-analysis, the PK-samples of 34 additional patients were available for this study (Bartelink et al., 2021). These samples included 648 total and 48 unbound plasma imatinib samples. The observed treatment period ranged between 1 and 10 days, with a median of 9 days. However, 33% of all patients had a treatment-observation period of less than 5 days. Most patients were male (76.8%). The most common comorbidities were hypertension (32.1%) and diabetes (22.0%). Out of 168 patients, 43 patients (25.6%) had a BMI above 30. Patient characteristics are presented in Table 1 .

Table 1.

Baseline patient characteristics (N = 168).

Demographic characteristics
Patients – no.	168
Age – median [IQR]	65 [57–73]
Bodyweight (kg) - median [IQR]	84.0 [75.0–95.3]
Height (cm) - median [IQR]	175 [169–180]
BMI (kg/m2) - median [IQR]	27.2 (24.8–30.5)
Males - no. (%)	129 (77)
Comorbidities - no. (%)
Current/former smoker	59 (35)
Obesity (BMI >30 kg/m²)	43 (26)
Diabetes	37 (22)
COPD/asthma	28(17)
Chronic kidney disease	6 (4)
Rheumatic disease	8 (5)
Neurological disease	14 (8)
Hypertension	54 (32)
Heart failure	8 (5)
Coronary artery disease	17(10)
Heart arrhythmia (Atrial fibrillation or flutter)	10 (6)
Clinical presentation – median [IQR]
SpO2/FiO2 ratio	321 (268–373)
Time from symptom onset to randomization (days)	10 [6–14]
Laboratory findings – median [IQR]
eGFR (ml/min/1.73 m2)	87 [72–90]
Albumin (g/L)	34.0 [30.0–38.0]
AAG (g/L)	1.88 [1.62–2.18]
CRP (mg/L)	95 [33–153]
ALT (U/L)	36.0 [26.0–55.0]
AST (U/L)	46.0 [35.0–61.8]
Bilirubin (μmol/L)	8.0 [6.0–10.0]
GGT (U/L)	60.5 [37.8–104.3]
Hb (mmol/L)	8.4 [7.8–9.1]
Leukocyte count (*109/L)	7.65 [5.58–10.50]
Neutrophil to lymphocyte ratio	6.44 [3.62–9.82]
LDH (U/L)	348 [274–441]
Medication initiated at admission - no. (%)
Dexamethasone	119 (71)
Remdesivir	25 (15)
(Hydroxy)chloroquine	32 (19)
Use of medical resources
Drug administration through nasogastric tube at any time– no. (%)	33 (20)
Admission to the intensive care unit – no. (%)	33 (20)
Need for invasive ventilation – no. (%)	16 (10)
Duration of hospital admission (days) median [IQR]	7 [4–11]

BMI = body mass index, AAG = alpha-1-acid glycoprotein, CRP = C-reactive protein, ALT = alanine transaminase, AST = aspartate transferase, GGT = gamma glutamyl transferase, Hb = hemoglobin, LDH = lactate dehydrogenase, NLR = neutrophil to lymphocyte ratio, PF0 = peripheral oxygen saturation and fraction of inspired oxygen ratio (P/F ratio) at baseline, COPD = chronic obstructive pulmonary disease, ICU = intensive care unit, eGFR = estimated glomerular filtration rate, The eGFR values were calculated with the Chronic Kidney Disease Epidemiology Collaboration equation (CKD-EPI), SpO2=oxygen saturation. FiO2=fractional concentration of oxygen in inspired air.

3.2. PK-parameters

For an optimal PK description of all patients in the CounterCOVID study and to further evaluate the difference in PK between CML/GIST and COVID, the PK parameters were re-estimated based on the combined dataset and the AAG-PK model, total C_trough (Ct_trough^,median 1.37 mg/L, IQR 0.035–2.71) and total AUCt_aver (median 3.12 mg*h/L, IQR 0.48–5.75).(table 2 ). Interindividual variability (IIV) was included on unbound apparent clearance and distribution volume. IIV was tested on affinity for the AAG-binding pocket (Kd AAG). The model fit improved (ΔOFV −14.1) but shrinkage of 62% was estimated and IIV on Kd was therefore removed. The AAG, age, bodyweight, and albumin were identified as significant covariates, similar to the prior PK-analyses results (Appendix Fig. 1 ). The forest plot demonstrates that covariates bodyweight, albumin and age had a statistically, but not a clinically significant effect on apparent clearance, whereas AAG predicted most of the variability in PK (Appendix Fig. 1 and 3A). Providing imatinib via a nasogastric tube explained variability in relative bioavailability (F) between the ICU and non-ICU patients as a F of only 66% (CV 9%) in patients with a nasogastric tube was found (∆OFV −10, P = 0.0016, compared to the model with no variability in F) (appendix Fig. 2 B). In the final PK-model, ICU and CRP no longer significantly affected the PK (∆OFV −0.28 and −6.265 compared to the model with CRP or ICU, respectively).

$${\mathrm{C}}_{\text{tot}}=\frac{{\mathrm{C}}_{\mathrm{u}}*\mathrm{L}*\text{AA}{\mathrm{G}}_{\text{tot}}}{{\mathrm{K}}_{\mathrm{d}}+{\mathrm{C}}_{\mathrm{u}}}+\frac{{\mathrm{K}}_{\mathrm{d}}}{{\mathrm{C}}_{\mathrm{u}}};\mathrm{L}=11700$$

Table 2.

The PK parameter estimates final PK model.

Fixed effect parameters*	Final AAG-PK model
KA (L/h)	0.682 (1.7%)
CL/Fu (L/h)	307.4 (0.3%)
KD (µg/L)	290 (1%)
Bodyweight (kg)	1.38 (3.7%)
Age (years)	0.594 (1.9%)
Albumin	1.174 (3.2%)
Vd/Fu (L)	7785.3 (0.3%)
Fnasogastric tube	0.656 (8.7%)
IIV on CL (CV%)	29.5% (5.6%, 28%)
IIV on V (CV%)	35.5% (6.9%, 38%)
corr. IIV Cl-V	0.549 (15.9%)
IIV on Ka (CV%)	129.5% (5.6%, 58%)
Box dis	0.1755 (56.8%)
RUV on Ctotal	16.2% (4.1%, 7%)
RUV on Cunbound	14.4% (1.8%, 15%)
Corr. Error	0.0218 (0%)

K_A = absorption rate, CL/F_u = apparent clearance unbound fraction, Vd/F_u = apparent unbound volume of distribution, F_{naso gastric tube} = bioavailability with the use of a nasogastric tube, IIV = inter-individual variability.

*Unbound CL/F_u and Vd/F_u were calculated. Total concentrations were derived from unbound concentrations using the equation:.

Fig. 1.

Prediction corrected, simulated imatinib concentration-time profiles in COVID-19 patients who were admitted to the Intensive Care Unit (ICU) during the 28 days after treatment onset, stratified for the administration of the drug via a nasogastric tube at the ICU. The dots = observed data, dashed black lines = 5th percentile and 95th percentile of observed concentrations, solid black line = median of observed concentrations, semitransparent dark blue field= simulation-based 95% confidence interval, SONDEANY = use of nasogastric tube during mechanical ventilation in the treatment period.

Fig. 3.

Mean WHO-score±standard deviation (SD) on day 0, 9, 28 and 90 of imatinib treatment and individual observations in relation to total trough concentration at steady state and (Ct_trough) and total average area under the concentration-time curve (AUCt_ave).

Fig. 2.

Liberation of oxygen supplementation (O₂lib) during 90 days after randomization in relation to total trough concentration at steady state (Ct_trough⁾ and total average area under the concentration-time curve (AUCt_ave) of imatinib concentrations divided in low, medium and high exposure categories.

The goodness-of-fit plots showed an adequate model fit, with some overprediction of the population at high AAG values, but adequate individual predictions (Appendix Fig. 3 ). Ct_trough of total imatinib concentrations in COVID-19 was adequately predicted, with a mean prediction error of 15.1% (± SD 67.4%). Fig. 1 shows a VPC of patients who were admitted to the ICU during any of the 28 days after treatment onset, stratified for nasogastric tube during dosing and PK-sampling. The results show that the observed total concentrations of imatinib in ICU patients were adequately predicted in all critically ill patients, also when the drug was administered via a nasogastric tube. All observed vs predicted PK-data stratified for AAG level are presented in appendix Fig. 4 . The correlation between the total and unbound day 3 C_trough and AUC_ave PK-parameters, were 0.880 and 0.884 respectively (appendix figure 5). As most PK data were derived from the observed total concentration-time profiles, in the current analysis the total day 3 Ct_trough and AUCt_ave PK-parameters were further explored.

Fig. 4.

Total trough concentration at steady state (Ct_trough) and average area under the concentration-time curve (AUCt_ave) on each day of imatinib treatment with corresponding partial oxygen pressure and fraction of inspired oxygen (P/F) including patient groups with and without need for invasive ventilation.

3.3. PKPD associations

Exploratory plots and univariate analyses showed that the magnitude of the total imatinib exposure (Ct_trough and AUCt_ave) did not positively associate with any of the outcomes: P/F, WHO-score change during imatinib treatment and O₂lib during 90 days after randomization (Table 3 ).

Table 3.

Associations between total imatinib exposure and O₂lib during 90 days after randomization (A) as well as WHO-score change (B) and P/F (C) during imatinib treatment adjusted for sex, age, AAG, P/F and WHO baseline, NLR, number of administered dosages and dexamethasone concomitant treatment.

A
O2lib
Total treatment group (N = 168)
Parameter	HR	95%CI	p-value	HR **	95%CI	p-value **
Cttrough (mg/L)	0.73	0.60 – 0.88	0.0014	0.78	0.61 – 0.99	0.032
AUCtave(mg*h/L)	0.93	0.86 – 1.01	0.087	0.93	0.81 – 1.1	0.36
Exposure categories (reference category=low)
Cttrough (mg/L) - High	0.51	0.34–0.78	0.001	0.61	0.36–1.00	0.062
Cttrough (mg/L) - Medium	0.76	0.51–1.12	0.17	1.15	0.72–1.84	0.56
AUCtave(mg*h/L) - High	0.44	0.29–0.66	<0.001	0.58	0.33–1.0	0.066
AUCtave(mg*h/L) - Medium	0.56	0.38–9.84	0.0046	0.50	0.31–0.81	0.0044
** Adjusted for confounders

B
WHO-score
Total treatment group (N = 168)
Parameter	β	95%CI	p-value	β **	95%CI	p-value **
Cttrough (mg/L)	0.38	0.08–0.69	0.014	0.33	−0.004–0.69	0.073
AUCtave(mg*h/L)	0.11	−0.015–0.23	0.085	−0.051	−0.25–0.15	0.62
Exposure categories (reference category=low)
Cttrough (mg/L) - High	0.83	0.15–1.5	0.17	0.63	−0.17–1.4	0.12
Cttrough (mg/L) - Medium	0.28	−0.40–0.96	0.42	−0.049	−0.76–0.67	0.89
AUCtave(mg*h/L) - High	0.62	−0.07–1.3	0.08	0.46	−0.35–1.28	0.26
AUCtave(mg*h/L) - Medium	0.30	−0.39–0.98	0.40	0.55	−0.14–1.24	0.12
*WHO-score corrected for baseline WHO-score, ** adjusted for confounders, Total trough concentration at steady state (Cttrough) and average area under the concentration-time curve (AUCtave).

C
P/F
Total treatment group (N = 168)
Parameter	β	95%CI	p-value	β **	95%CI	p-value **
Cttrough (mg/L)	−16.15	−28.6- −3.67	0.011	−19.64	−35.4- −3.90	0.014
AUCtave(mg*h/L)	−2.95	−8.14–2.24	0.27	−1.76	−10.6–7.12	0.70
Exposure categories (reference category=low)
Cttrough (mg/L) - High	−32.00	−60.1- −3.95	0.025	−30.28	−65.9–5.29	0.10
Cttrough (mg/L) - Medium	−5.94	−34.4–22.5	0.68	1.96	−30.0–33.9	0.90
AUCtave(mg*h/L) - High	−26.98	−55.8–1.88	0.067	−36.25	−72.5–0.020	0.050
AUCtave(mg*h/L) - Medium	−15.68	−44.4–13.03	0.29	−33.03	−63.7- −2.38	0.035

*partial oxygen pressure and fraction of inspired oxygen (P/F) corrected for baseline P/F, ** adjusted for confounders, total trough concentration at steady state (Ct_trough) and average area under the concentration-time curve (AUCt_ave).

3.4. O₂lib

Kaplan-Meier curves of total Ct_trough and AUCt_ave, and O₂lib showed that patients who survived and experienced >48-hour liberation from oxygen supplementation and ventilation had lower imatinib exposure (Fig. 2). The number of observed events at day 90 per low, medium and high exposure categories were 89%, 86% and 79% for Ct_trough (Logrank p = 0.005) and 88%, 86% and 80% for AUCt_ave (Logrank p=<0.001). In total 142 O₂lib events were observed during a follow up period of 90 days. For Ct_trough the unadjusted HR was 0.73, p = 0.0014, and the adjusted HR was 0.78, p = 0.032, suggesting that low total Ct_trough imatinib exposure was associated with a higher likelihood of experiencing an O₂lib event, with a similar trend for observed AUCt_ave (Table 3A). These results remain similar when looking at exposure categories low, medium, high (Table 3A).

3.5. WHO-score

In Fig. 3, WHO-scores on day 0, 9, 28 and 90 after the first imatinib dose are displayed per low, medium and high exposure categories for Ct_trough and AUCt_ave. For both PK-parameters, visually, a trend was observed, in which patients with higher exposure showed higher average improvement in WHO-score. For Ct_trough this trend is visible over time, whereas for AUCt_ave this trend disappears at day 90. Consequently, the change in WHO-score during treatment, so between day 9 and 0, was analyzed. No associations were found between change in WHO-score and Ct_trough (adjusted β=0.33, p-value=0.073) and AUCt_ave (adjusted β=−0.051, p-value=0.62), table 3B, appendix figure 6. Analysis of exposure categories showed similar trends (Table 3B).

3.5. P/F

Ct_trough and AUCt_ave did not associate with the P/F on each specific day of treatment and the PK-parameters on the corresponding day, when stratified for patients with and without invasive ventilation (Fig. 4). Plots of P/F over time against total Ct_trough on day 3 and AUCt_ave, divided into tertiles, show that up until day 4 the lowest exposure categories demonstrated the highest PF-ratios, whereas after day 4, the medium exposure categories lead to the best P/Fs, and for the low exposure categories the P/F declines (appendix figure 7). After day 6 of treatment, the percentage of missing values for the P/F is ≥50%. An increase in Ct_trough associated with an improvement in P/F during imatinib treatment, (unadjusted, only adjusting for baseline P/F, β=−16.15, p-value=0.011 and adjusted β=−19.64, p-value=0.014), whereas AUCt_ave, did not associate significantly with P/F when looking at the total treatment group (Table 3C), although significant associations were found for the AUCr_ave exposure categories. Based on the current data, no associations were found between P/F or WHO-score at baseline and pharmacokinetics of imatinib (appendix figure 8).

3.6. Confounders

All potential continuous confounders were included in a correlation matrix as shown in Appendix figure 9. Based on ≥10% influence on the investigated associations in this study, sex, age, the number of administered dosages, dexamethasone concomitant treatment, AAG, NLR and PF-ratio/WHO-score at baseline were considered confounders.

4. Discussion

The aim of this study was to find potential PKPD relationships following oral administration of 400 mg imatinib in COVID-19 patients. Our hypothesis that high total imatinib exposure leads to improved pharmacodynamic outcomes was not confirmed in this study. This is the first PKPD study concerning imatinib treatment in COVID-19 patients, and is enabling the opportunity to compare with prior PKPD studies of imatinib in CML/GIST. In CML/GIST some studies observed a relationship between higher drug exposure and improved clinical outcomes (Larson et al., 2008; Picard et al., 2007; Singh et al., 2009; Widmer et al., 2008; Takahashi et al., 2010; Widmer et al., 2010). Our study suggests that in contrast to patients with cancer, in COVID-19 dose individualization based on total concentrations may not lead to improved clinical outcomes. This PKPD analysis found inverse associations between total imatinib Ct_trough and AUCt_ave and clinical outcomes, liberation of oxygen, WHO clinical status and P/F.

Although it remains to be elucidated, this inverse association may be explained by the severity of the infectious disease course. The severity of the disease course may influence pharmacokinetic processes of imatinib including distribution through protein-binding and metabolism (Delbaldo et al., 2006). There are studies that question the clinical relevance of protein-binding (Benet, 2002; Smith et al., 2010). However, this clinical relevance may be influenced by the affinity of a certain drugs for plasma proteins, because some plasma proteins are more influenced by disease-state than others (Gabrielsson et al., 2009). Inflammatory cytokines including tumor necrosis factor (TNF)-α, interleukin(IL)−1, IL-6 and C-reactive protein can regulate AAG expression (Ruiz, 2021). AAG, also known as orosomucoid (ORM), belongs to the protein family of lipocalins. Lipocalins transport small hydrophobic molecules and play a role in pro-and anti-inflammatory immune mechanisms affecting viral infections, but also in other pathologies including cancer (Fournier et al., 2000; Hochepied et al., 2003). Imatinib binds to AAG with high affinity (Schmidt et al., 2010). According to the free drug hypothesis, only the free drug concentration in plasma can be of relevance for its pharmacological actions (Schmidt et al., 2010). In healthy humans, AAG is present in the systemic circulation in levels from 0.5 to 1.0 mg/ml, whereas in inflammatory diseases and cancer these levels can increase two-to six-fold (Kremer et al., 1988). In the current study AAG levels ranged between 1.4–3.6 mg/ml. Concentrations of AAG typically rise within the first 24 hours of infection or injury and slowly decline during the days thereafter. Because AAG concentrations are influenced by a patient's disease course, this could affect the total drug concentration of imatinib. Furthermore, isoforms of AAG can change throughout the course of an infection by the heterogeneity in glycosylation patterns (Almquist and Lausing, 1957) resulting in different binding capacities, further influencing imatinib concentrations. Disease course may not only influence the total drug concentration by AAG-imatinib binding, but may also affect drug metabolism. For example, immune responses to viral infections may alter the activity of cytochrome P450(CYP)-enzymes that are involved in the metabolism of imatinib (van Erp et al., 2009; Gréen et al., 2010). It is, for instance, reported that IL-6 may suppress CYP3A4 activity (Jover et al., 2002). Suppression of CYP3A4 may in turn increase drug exposure of imatinib, since IL-6 is upregulated during COVID-19 (Chen et al., 2020).

In the current study an established AAG-PK model was used to predict interpatient-variability in the total drug exposure PK parameters, Ct_trough and AUCt_ave in the CounterCOVID study (Bartelink et al., 2021; Haouala et al., 2013). The final covariate model showed accurate individual predictions for imatinib total concentration-time profiles in all 168 patients. The bioavailability of imatinib decreased during mechanical ventilation in ICU patients compared to non-ventilated patients, which could be explained by either or both of altered gastro-intestinal adsorption or use of the nasogastric tube  (Smith et al., 2012). The effect of nasogastric tubing on bioavailability should be studied further. Intravenous injection of imatinib is suggested as possible alternative route of administration in order to circumvent the potential influence upon bioavailability of imatinib of bowel motility and nasogastric tube adsorption. Moreover, the underlying mechanism of the direct effect of albumin on unbound clearance is a topic of subsequent study: low albumin may decrease protein binding in the interaction with the AAG-imatinib-complex and low albumin may reflect disease severity, which in turn can affect the metabolic rate of imatinib. Total concentrations of imatinib in COVID-19 patients at early time points (0–4 hours post-dose) were adequately predicted. However, the lack of early PK sampling limited the exploration of C_max in the PKPD analyses. Also, Ct_trough of imatinib in COVID-19 patients were adequately predicted, with a mean prediction error of 15.1% (± SD 67.4%). The current study primarily focused on total imatinib concentrations, by a lack in the availability of unbound imatinib samples. To further explore the influence of protein-binding on imatinib PKPD in COVID-19 patients, future studies should include a larger number of unbound imatinib plasma samples. Multiple imatinib PKPD studies related to CML/GIST patients recommended to focus on the unbound concentrations of imatinib, because of the complexity of factors affecting the total drug concentration of imatinib including AAG (Bartelink et al., 2021; Delbaldo et al., 2006; Gandia et al., 2013) potentially providing a more direct link with response. Other than the unbound fraction, the main metabolite of imatinib, GCP74588, may also be considered for the accuracy of additional PKPD analyses. This metabolite is active against BCR/ABL in GIST/CML (Menon-Andersen et al., 2009). Although, it remains unclear whether this metabolite has similar activity against tyrosine kinases involved in vascular leakage, including c-Abl and Arg.

Of potential covariates that were identified prior to performing the study, the variables sex, age, the number of administered dosages, NLR, dexamethasone concomitant treatment, AAG and baseline P/F and WHO-score were included into the final PKPD model. Some of these variables such as the sex and age, the NLR, P/F-and WHO-score at baseline may reflect COVID-19 severity of disease at baseline. The disease course of COVID-19 during treatment can moreover influence AAG that can in turn influence exposure parameters. Age and sex are furthermore known to influence the pharmacokinetics of drugs. Therefore, the inclusion of these covariates seems biologically plausible.

This study did not investigate the association between imatinib concentrations and toxicity. However, dose/concentration-toxicity studies suggest that doses above 600 mg and high unbound, rather than total concentrations may increase toxicity (Singh et al., 2009; Widmer et al., 2006). In the CounterCOVID study, more imatinib treated patients stopped treatment prematurely due to gastrointestinal (GI) adverse reactions, but cardiotoxic effects such as QT prolongation in the imatinib group were not observed (Aman et al., 2021; Duijvelaar et al., 2022). Further analyses of exposure and GI toxicity may help to improve adherence and thereby treatment outcome. Another study limitation is that data points were missing not at random. Patients who died or were discharged from the hospital during imatinib treatment could not be followed up in a similar way to patients that were still admitted. The majority of missing data was attributed to either discharge or death. Over time, the data were more enriched with patients who were hospitalized and alive, because clinical blood collection and vital parameters assessments halted after discharge. Although clinical outcome was available for every patient throughout the study, these competing factors could have potentially effected the collected data. It therefore remains to be elucidated whether this inverse exposure-response trend can be explained by missing data or whether it is related to other factors.

In conclusion, the current study found inverse PKPD relationships of imatinib in COVID-19 patients, after COVID-19 patients were treated with 400 mg imatinib daily for 10 days. A study with multiple dose levels would answer the question whether this dose and corresponding exposure provides optimal response COVID-19. In this study, lower Ct_trough and AUCt_ave associated with improved clinical outcomes including liberation from oxygen, WHO clinical status and P/F. The underlying causes of these observations remain to be elucidated. Disease course can affect pharmacokinetic processes, including distribution by AAG binding and metabolism, that may explain inverse exposure-response relationships of imatinib in COVID-19 patients. The current applied AAG-PK model accurately predicted total imatinib concentrations. However, for further PKPD research purposes of imatinib in COVID-19 or other conditions that can lead to ARDS vascular leakage, it is recommended to consider the unbound drug concentration as well as the active metabolite of imatinib.

Credit author statement

IB, NB, HJB, JA, AM, NS, ED, PB contributed to conceptualization and data curation. IB, MS performed the PK analysis and visualization. NB, KM performed the PKPD analysis and visualization. JT, CL performed and provided input for the statistical analyses. NB, IB, MS, ED were involved in writing of the original draft. All authors were involved in writing – review and editing. IB was involved in investigation, supervision. HJB was involved in funding acquisition and project administration.

Data availability statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declaration of Competing Interest

J.A. is an inventor on a patent (WO2012150857A1, 2011) covering protection against endothelial barrier dysfunction through inhibition of the tyrosine kinase abl-related gene (arg). J.A. served as a non-compensated scientific advisor for Exvastat Ltd. S.S. is an employee of Exvastat Ltd. and has no other conflicts of interests. All other authors declare no competing interests.

Data availability

• Data will be made available on request.