Physiologically‐Based Pharmacokinetic Modelling of Creatinine‐Drug Interactions in the Chronic Kidney Disease Population

Abstract

Elevated serum creatinine (S_Cr) caused by the inhibition of renal transporter(s) may be misinterpreted as kidney injury. The interpretation is more complicated in patients with chronic kidney disease (CKD) due to altered disposition of creatinine and renal transporter inhibitors. A clinical study was conducted in 17 patients with CKD (estimated glomerular filtration rate 15–59 mL/min/1.73 m²); changes in S_Cr were monitored during trimethoprim treatment (100–200 mg/day), administered to prevent recurrent urinary infection, relative to the baseline level. Additional S_Cr‐interaction data with trimethoprim, cimetidine, and famotidine in patients with CKD were collated from the literature. Our published physiologically‐based creatinine model was extended to predict the effect of the CKD on S_Cr and creatinine‐drug interaction. The creatinine‐CKD model incorporated age/sex‐related differences in creatinine synthesis, CKD‐related glomerular filtration deterioration; change in transporter activity either proportional or disproportional to glomerular filtration rate (GFR) decline were explored. Optimized models successfully recovered baseline S_Cr from 64 patients with CKD (geometric mean fold‐error of 1.1). Combined with pharmacokinetic models of inhibitors, the creatinine model was used to simulate transporter‐mediated creatinine‐drug interactions. Use of inhibitor unbound plasma concentrations resulted in 66% of simulated S_Cr interaction data within the prediction limits, with cimetidine interaction significantly underestimated. Assuming that transporter activity deteriorates disproportional to GFR decline resulted in higher predicted sensitivity to transporter inhibition in patients with CKD relative to healthy patients, consistent with sparse clinical data. For the first time, this novel modelling approach enables quantitative prediction of S_Cr in CKD and delineation of the effect of disease and renal transporter inhibition in this patient population.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

☑ Serum creatinine (S_Cr), the key endogenous biomarker of kidney function, increases due to inhibition of renal transporters even in the absence of kidney injury. Disposition of both creatinine and renal transporter inhibitors differs between healthy subjects and patients with chronic kidney disease (CKD). Physiologically‐based pharmacokinetic (PBPK) models to simulate creatinine‐drug interactions have previously only been reported for healthy populations.

• WHAT QUESTION DID THIS STUDY ADDRESS?

☑ Can PBPK modelling of creatinine predict the disease effect on S_Cr and creatinine‐drug interactions in patients with CKD?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

☑ A PBPK model that can account for disposition of creatinine and effects of renal transporter inhibitors in patients with CKD has been developed. The model can successfully simulate creatinine‐drug interactions in this population.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

☑ The developed model enables quantitative translation of renal transporter in vitro inhibition data together with disease‐related changes to predict the extent of changes in S_Cr in patients with CKD.

Estimated glomerular filtration rate (eGFR) based on serum creatinine (S_Cr) is widely used clinically as an index of renal function. ¹ However, inhibition of renal transporters leads to transient increase in S_Cr (creatinine‐drug interaction) even in the absence of kidney injury, ² because a certain proportion of creatinine is eliminated by active secretion via renal transporters. ³ , ⁴ Therefore, a method that can identify the cause of increased S_Cr would be useful in clinical practice.

Chronic kidney disease (CKD) is often associated with progressive renal dysfunction, characterized by increased S_Cr. ¹ In addition to gradual decline in glomerular filtration rate (GFR), patients with CKD show several physiological changes in the kidneys and other organs that can affect elimination of drugs/endogenous substances. These include accumulation of uremic solutes, ⁵ , ⁶ reduced serum albumin concentration, ⁷ metabolic acidosis, ⁸ , ⁹ and reduced expression/activity of metabolizing enzymes and transporters in the liver. ¹⁰ These physiological changes in CKD can affect the disposition of both creatinine and drugs that inhibit renal transporters.

A number of clinical analyses assumed that tubular secretion of solutes decreases in CKD in proportion to GFR (“intact nephron hypothesis" (INH) ¹¹ ). ⁵ , ⁶ In contrast, other studies reported changes in tubular secretion (e.g., via organic anion transporters (OATs)) that were inconsistent with GFR. ⁵ , ⁶ , ¹² In the case of creatinine, smaller extent of decline in tubular secretion relative to GFR was implied because of reported increase in ratio of creatinine clearance to GFR (C_Cr/GFR) in patients with CKD. ³ , ⁴ Moreover, the C_Cr/GFR in patients with CKD approached the level of healthy subjects after administration of cimetidine (renal transporter inhibitor). ¹³ Therefore, a higher degree of creatinine‐drug interaction may occur in the CKD population than in subjects with normal renal function (assuming equal dosing) due to combination of (i) higher exposure of inhibitor drug due to lower impaired hepatic and/or renal elimination, and (ii) decline in transporter activity disproportionate to GFR (higher C_Cr/GFR ratio in patients with CKD relative to healthy).

Physiologically‐based pharmacokinetic modelling has been applied to predict the effect of CKD on drug exposure. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ In the case of creatinine, several models have been reported and applied to simulate creatinine‐drug interactions in healthy subjects. ¹⁸ , ¹⁹ , ²⁰ , ²¹ However, there is currently no model in place to capture CKD‐related changes in creatinine renal disposition. Our recently published physiologically‐based creatinine model, developed for healthy subjects, accounted for multiple transporters involved in renal creatinine elimination, assuming either unidirectional or bidirectional transport via organic cation transporter (OCT) 2 (uptake‐OCT2 or bidirectional‐OCT2 model), driven by an electrochemical gradient (Figure  1a ). ²⁰ , ²¹ In addition, the models incorporated endogenous creatinine synthesis, glomerular filtration, and passive diffusion across proximal tubule cells. The models, initially based on proteomics‐informed in vitro‐in vivo extrapolation of transporter kinetics, were optimized by creatinine‐trimethoprim interaction data and successfully simulated the percent change in S_Cr (%ΔS_Cr) postdosing of 11 further inhibitors.

Figure 1.

Model optimization for creatinine‐drug interaction in patients with chronic kidney disease (CKD). (a) A reprinted model structure from the previous study showing the creatinine models for healthy subjects. ²⁰ , ²¹ Permeation mechanisms at proximal tubule cell in two creatinine models with different description of organic cation transporter (OCT) 2 were presented in purple shaded area: uptake‐OCT2 model (green arrow only) and bidirectional‐OCT2 model (both green and yellow arrows). See Scotcher et al. ²⁰ , ²¹ regarding details of the models and system parameters. Parameters optimized for patients with CKD in this study were enclosed by red dashed squares. (b) Strategy of model optimization. The simulation of creatinine‐drug interaction in patients with CKD was implemented in three steps; (1) optimization of creatinine model for patients with CKD, (2) development of inhibitors’ pharmacokinetic (PK) models for patients with CKD, and (3) simulation of creatinine‐drug interaction in patients with CKD. In step 1, both uptake‐OCT2 and bidirectional‐OCT2 models optimized for healthy subjects (step 1‐0) were extended for patients with CKD in four sub‐steps. In step 1‐1, estimated glomerular filtration rate of CKD patient i (eGFR_i) was calculated using Chronic Kidney Disease Epidemiology Collaboration (CKD‐EPI) equation ¹ (Eq.  S1 ), where S_Cr,i and Age_i represent serum creatinine (S_Cr) and age of CKD patient i, a = −0.329 and k = 0.7 for women, and a = −0.411 and k = 0.9 for men, min and max indicate the minimum of S_Cr/k or 1, or maximum of S_Cr/k or 1, respectively. In step 1‐2, creatinine synthesis rate of CKD patient i (R_syn,i) was calculated using the reported regression equation ²² (Eq. S5), with correction using the calculated body surface area. WT_i represents body weight of CKD patient i, C0 = 27 and C1 = 0.173 for men, and C0 = 25 and C1 = 0.175 for women. In step 1‐3, a value of parameter j in the proximal or distal tubule of CKD patient i (SysPara(j)_CKD.i) was altered in proportion to glomerular filtration rate (GFR; Eq. S6, intact nephron hypothesis (INH)), where Sys Para(j)_healthy represents a representative value of system parameter j in healthy subjects,GFR_CKD,i and GFR_healthy are GFR in CKD patient i and a healthy subject (125 mL/min), respectively. Abbreviations of optimized parameters are listed in Table  2 . In step 1‐4, clearances of renal transporters were altered disproportional to GFR (non‐INH scenario). Relative change inintrinsic clearance (CLint) of organic anion transporter (OAT) 2 in CKD patient i (CLint(OAT2)_CKD,i/CLint(OAT2)_healthy) was calculated as a function of relative change in GFR (GFR_CKD,i/ GFR_healthy) and additional deterioration of OAT2 clearance beyond INH (Fx_OAT2,i; Eqs.  S9,S10 ). Relative change in clearances of OCT2 and multidrug and toxin extrusion protein (MATE) transporters in CKD patient i (expressed as CLint(j)_CKD,i/CLint(j)_healthy where CLint(j)_CKD,i and CLint(j)_healthy represent CLint of transporter j in representative populations) were estimated as a linear function of GFR, where Coeff_CKD,TP represents the slope of the linear function (Eq. S11). Combined with PK models for inhibitors developed in step 2, the creatinine models were used to simulate creatinine‐drug interaction in CKD population in step 3.

This study aimed to extend the existing creatinine model to the CKD population by accounting for physiological changes associated with the disease and to predict creatinine‐drug interactions in these patients for inhibitors of OCT2 and multidrug and toxin extrusion protein (MATE) transporters. Literature data were collated based on availability of both clinical pharmacokinetics (PKs) for the inhibitor, and interaction data in patients with moderate (G3; eGFR 15–29 mL/min/1.73 m²) to severe (G4, eGFR 30–59 mL/min/1.73 m²) CKD. In addition, a new clinical study was conducted in 17 patients with moderate‐to‐severe CKD and their S_Cr was monitored during prophylactic trimethoprim treatment (100–200 mg/day). The creatinine‐CKD models were developed and evaluated in a stepwise manner (Figure  1b ):

1. Modification of the creatinine models and corresponding system parameters to account for physiological changes in CKD and model verification against independent clinical dataset.

2. Development of PK models for different inhibitors using reported plasma concentration‐time profiles in patients with CKD.

3. Simulation of creatinine‐drug interactions in patients with CKD and evaluation of predicted %ΔS_Cr against clinical observations.

METHODS

Collation of clinical data in patients with CKD

Clinical creatinine‐drug interaction data were obtained from a new clinical study and the existing literature. The Salford Kidney Study is a large longitudinal CKD cohort investigation conducted with > 3,000 patients with non‐dialysis CKD, recruited during the period 2002–2015 in Salford Royal NHS Service Foundation Trust (Supplementary Material Section  S1 ). From the entire cohort, data from 17 patients with CKD (6 men and 11 women, age 22–88 years, stage G3–4) were included for evaluation of the creatinine‐trimethoprim interaction. Subjects on any additional comedications known to cause creatinine‐drug interaction in healthy subjects ²¹ were excluded. Trimethoprim treatment was usually for prophylaxis against recurrent urinary infection (100–200 mg/day), lasting on average 91 days (ranging from 10–420 days). Patients’ S_Cr at the baseline (S_Cr,baseline) were defined as the mean of measurements in the period up to 1,000 days prior to initiation of trimethoprim; “day 1” was the first S_Cr measurement after initiation of trimethoprim treatment. The %ΔS_Cr was calculated as percent change from S_Cr,baseline to the S_Cr at day 1. S_Cr values after day 1 were excluded due to potential confounding factors (e.g., deterioration of CKD or adaptation to trimethoprim treatment).

Literature clinical creatinine‐drug interaction data were collated for 15 renal transporter inhibitors that were evaluated with the existing creatinine model for healthy subjects in our previous study (Table  S3 ). ²¹ Inclusion criteria for clinical studies are detailed in the Supplementary Material Section  S2 . Reported C_Cr/GFR ratio data in healthy and CKD populations were collated (see Supplementary Material Section  S3 ). Mean values and standard deviations (SD) of C_Cr/GFR were calculated for each CKD group accounting for the number of subjects in each study.

Simulation of creatinine‐drug interaction in patients with CKD

Simulations of creatinine‐drug interaction were implemented in three steps: (i) optimization of creatinine CKD models, (ii) development of PK models for each inhibitor, and (iii) simulation of the creatinine‐drug interaction in patients with CKD (Figure  1b ). Uptake‐OCT2 or bidirectional‐OCT2 models were optimized independently.

Optimization of creatinine models for CKD

Creatinine CKD models were developed in a stepwise manner. GFR parameter was informed by either (i) gold‐standard exogenous markers (e.g., inulin and iothalamate), or, when such measurements were not available, (ii) eGFR based on measured S_Cr and the Chronic Kidney Disease Epidemiology Collaboration (CKD‐EPI) equation ¹ , which was validated against S_Cr and iothalamate renal clearance in patients with CKD (Eq.  S1 ). Endogenous creatinine synthesis rate (R_SYN) was calculated using a published regression equation ²² using demographic data (age, sex, and weight (WT); Eq. S5). Renal blood flow was decreased by 27% and 42% in CKD G3 and G4 relative to healthy subjects, respectively, based on magnetic resonance imaging. ²³ The volume of distribution in the central compartment was assumed to be the same as in healthy subjects, as CKD has minimal effect on total body water. ²⁴ Values of pH and membrane potential in the proximal tubule, were assumed the same as in healthy populations due to the scarcity of information in the CKD population.

Parameters relevant to proximal and distal tubule, such as passive membrane permeability (CL_PD), volumes of proximal tubule compartments, and filtrate flow rate (Q_U‐filt), were decreased proportionally to GFR (Eq. S6), in line with INH. Modification of Q_U‐filt was based on the assumption that flow out of the proximal tubular filtrate changed in proportion to GFR; changes in CL_PD were attributed to decreased membrane surface area of proximal and distal tubule in CKD. CKD‐dependent change in filtrate pH was assumed to have minimal effect on creatinine CL_PD, supported by our previous study. ⁹ These assumptions for CL_PD and Q_U‐filt resulted in the same creatinine fraction reabsorbed in the distal tubule in healthy patients and patients with CKD (Eqs.  S7,S8 ). Absolute amount of water reabsorbed in distal tubule was reduced as per INH.

Two scenarios were investigated in creatinine CKD model with respect to transporter clearance parameters: (i) activity of all transporters decreased proportionally to GFR (Eq. S6, “INH scenario”) or (ii) change in transporter activity was disproportionate to GFR, supported by higher C_Cr/GFR ratio in CKD relative to healthy (“non‐INH scenario”). Deterioration of OAT2 activity in CKD was based on the analysis of clinical data for OAT2 substrates ¹² (Eqs.  S9,S10 ). Relative change in clearances of other transporters in CKD were expressed as a linear function of GFR (Eq. S11), and slope of the function (Coeff_CKD,TP) were estimated for uptake‐OCT2 or bidirectional‐OCT2 models independently by fitting the models to overall means of C_Cr/GFR in CKD and healthy populations using the lsqnonlin function in Matlab (R2017a; MathsWorks). The same Coeff_CKD,TP value was assumed for OCT2 and MATE transporters. Sensitivity of the creatinine CKD models to the uncertainty in system parameters or degree of transporter inhibition is shown in Supplementary Material Section  S5 .

Development of PK models for the inhibitors in patients with CKD

One‐compartment or two‐compartment PK models were developed for trimethoprim, cimetidine, and famotidine (details in Supplementary Material Section  S6 ). As there was no clear relationship between exposure of the inhibitors and CKD stage in our scarce dataset, PK models were developed by simultaneous fitting of all available plasma concentration‐time profiles in patients with CKD G3–4. One‐compartment model for trimethoprim was developed in NONMEM version 7.42 using WT as a covariate. Models for cimetidine and famotidine were based on mean PK profiles using the naïve pooled method.

Simulation of creatinine‐drug interaction in patients with CKD

Simulation of creatinine‐drug interactions was performed by combining the creatinine and inhibitor models for patients with CKD. Inhibitory effect on intrinsic clearances of individual transporters was simulated as described previously ²¹ , ²⁵ (Supplementary Material Section  S7 ).

Verification of creatinine CKD model

The predictability of the creatinine models was evaluated with S_Cr,baseline and %ΔS_Cr as end points. The predictability of S_Cr,baseline was assessed by comparing the observed and predicted values, quantified by geometric mean fold‐error (gmfe; Eq. 1), ²⁶ and percentage of simulated data within 1.2‐fold of observed values. Acceptable creatinine models were selected based on gmfe < 1.15 and recovery of reported C_Cr/GFR. The predictability of S_Cr,baseline was also evaluated using independent external dataset (Supplementary Material Section  S8 ).

$$\text{gmfe}={10}^{\frac{1}{n}\sum \left|{\log }_{10}\left(\frac{S_{\text{Cr,baseline,predicted}}}{S_{\text{Cr,baseline,observed}}}\right)\right|}$$ (1)

The predictability of %ΔS_Cr was evaluated by comparing means of observed and predicted %ΔS_Cr for each study using mean absolute‐error (Eq. 2). In addition, prediction performance of %ΔS_Cr was evaluated using novel prediction limits that accounted for intra‐individual variability in S_Cr,baseline in patients with CKD; these boundaries are much stricter than conventional twofold (details in Supplementary Material Section  S9 ). ²¹ , ²⁷ When data on timing of blood sampling for S_Cr was missing, the maximum predicted %ΔS_Cr during the potential sampling period was used.

$$\text{MAE}=\frac{1}{n}\sum \left|\%\mathrm{\Delta }{S}_{\text{Cr,predicted}}‐\%\mathrm{\Delta }{S}_{\text{Cr,observed}}\right|$$ (2)

RESULTS

Analysis of creatinine data in patients with CKD

In the Salford Kidney Study, S_Cr at baseline and post‐trimethoprim were evaluated in 17 patients with CKD G3 (n = 12) and G4 (n = 5) (Table  S1 ). Trimethoprim caused a statistically significant increase in S_Cr from the mean value of 1.7 mg/dL at baseline (1.1–3.2 mg/dL) to 2.0 mg/dL (1.2–3.1 mg/dL) 91 days post‐trimethoprim (mixed effects model, P < 0.01; Figure  2a , Table  S2 ). Mean %ΔS_Cr post‐trimethoprim was 20% higher relative to the baseline (ranged from −12 to 86%), with no direct correlation between eGFR and %ΔS_Cr (Figure  2b ). The intra‐individual coefficient of variability of S_Cr,baseline was 8.9 ± 4.9%, which was higher than the reported value in healthy subjects (4.7%). ²⁸

Figure 2.

Evaluation of creatinine‐drug interaction in Salford Kidney Study. (a) Serum creatinine (S_Cr) at baseline and post trimethoprim of 17 patients with chronic kidney disease (CKD) in Salford Kidney Study. Filled symbols and error bars represent means and standard deviations of S_Crat the baseline (−1,000 day to last blood test prior to trimethoprim) in each patient; Open symbols represent S_Cr at the first test post trimethoprim (day 1); Circles: CKD G3 (estimate glomerular filtration rate (eGFR) 30–59 mL/min/1.73 m²); Triangles: CKD G4 (eGFR 15–29 mL/min/1.73 m²). (b) Percent change in S_Cr post trimethoprim plotted against eGFR in Salford kidney study. Each symbol represents an individual patient with CKD; symbols for CKD stage, as described in a). Solid line and dashed lines represent mean and mean ± SD of percent change in S_Cr of all patients with CKD, respectively.

A literature search identified 15 clinical studies evaluating either S_Cr,baseline or %ΔS_Cr in patients with CKD for three inhibitors of renal transporters (trimethoprim, cimetidine, and famotidine) that met inclusion criteria for the current analysis (Table  1 ). Data on %ΔS_Cr in healthy patients and patients with CKD from the literature and the Salford Kidney Study were analyzed with respect to the daily dose of inhibitors (Table  S4 , Figure  S2 ). Despite the lower daily doses in patients with CKD relative to healthy subjects, maximum mean %ΔS_Cr across the studies was ~ 30% in both populations following administration of trimethoprim (13–31% in healthy patients and 7–33% in patients with CKD) and cimetidine (14–26% in healthy patients and 10–31% in patients with CKD). Although mean %ΔS_Cr in patients with CKD tended to be higher than in healthy subjects when comparing the effects of trimethoprim of < 400 mg/day, the trend was inconclusive due to sparse and variable data.

Table 1.

Summary of clinical studies evaluating creatinine‐drug interaction in patients with CKD

Inhibitor	Clinical study	Subject information (M; male, F; Female)	GFR, mL/min/1.73 m2	Study design	Blood sampling for SCr post inhibitor	% change in SCr, mean ± SD	Individual’s baseline SCr
Trimethoprim	Salford study group 1	n = 13, M3 F10, 25–79 years	26–56	100 mg q.d. oral	>Day 10, 12 hours after last dose	23 ± 25	Yes
	Salford study group 2	n = 4, M3 F1, 22–88 years	26–43	200 mg q.d. oral	>Day 10, 12 hours after last dose	9 ± 10	Yes
	Myre et al. (1987) 38	n = 5, M2 F3, 37–57 years	15–23	100 mg b.i.d. oral	Day 10, 2–4 hours after morning dose	33 ± 26	Yes
	Tasker et al. (1975) 45 group 1 a	n = 6, M5 F1, 34–80 years	27–47	160 mg b.i.d. oral	Days 6–10	28 ± 34	Yes
	Tasker et al. (1975) 45 group 2 b	n = 4, M0 F4, 47–78 years	22–23	160 mg b.i.d. oral (days 1–3), q.d. oral (day 4~)	Days 6–10	7 ± 20	Yes
	Rieder et al. (1974) 30	n = 9, M4 F5, 25–69 years	17–56	‐	‐	‐	Yes
Cimetidine	Larsson et al. (1980) 46 group 1 c	n = 8, 29–77 years	21–40	200 mg q.d. oral	Day 7	25	No e
	Larsson et al. (1980) 46 group 2 e	n = 9, 29–77 years	36–55	(200 mg × 3 + 400 mg) per day oral	Day 7	31	No e
	Ishigami et al. (1989) 39	n = 8, M7 F1, 28–67 years	16–29	400 mg b.i.d. oral	Day 7	12	No e
	Hilbrands et al. (1991) 47	n = 5, 25–66 years	20–40	(400 mg × 2 + 600 mg) per day oral	Steady‐state	26 ± 14	No e
	Ma et al. (1978) 31	n = 8, M8 F0, 34–66 years	18–57	300 mg SD intravenous	24–48 hours after dose	10 ± 10	No e
	Larsson et al. (1981) 32	n = 17, M12 F5, 31–68 years	19–60	‐	‐	‐	Yes
	Bjaeldager et al. (1980) 48	n = 6, M2 F4, 39–65 years	23–41	‐	‐	‐	Yes
Famotidine	Ishigami et al. (1989) 39	n = 8, M7 F1, 28–67 years	16–29	20 mg b.i.d. oral	Day 7	7	No e
	Abraham et al. (1987) 34	n = 12, M10 F2, 28–54 years	10–41	10 mg SD intravenous	0–4 hours after dose	0	No e

CKD, chronic kidney disease; GFR, glomerular filtration rate; S_Cr, serum creatinine; SD, single dosing.

^aGroup with creatinine clearance above 25 mL/min.

^bGroup with creatinine clearance 15–25 mL/min.

^cGroup with creatinine clearance of 30–50 mL/min.

^dGroup with creatinine clearance of 50–75 mL/min.

^eExcluded from the evaluation of baseline S_Cr due to lack of individual’s S_Cr. Ma et al. (1978) ³¹ was not used for the evaluation of baseline S_Cr because individual’s age was missing.

Optimization of creatinine models for patients with CKD

The creatinine models were optimized to capture reported S_Cr,baseline (1.1 to 3.9 mg/dL) in 64 patients with CKD (35 with G3 and 29 with G4, 31 men and 33 women, ages 22–88 years) from 8 clinical studies (Table  1 ). Initial application of the creatinine models ²⁰ , ²¹ based on population average values for physiological model parameters for healthy subjects failed to capture increased S_Cr,baseline in patients with CKD (gmfe = 2.33–2.34; Figure  S5 , Table  2 ). In particular, more pronounced underprediction of higher S_Cr,baseline values was apparent. Refinement of the model to capture CKD‐related decrease in glomerular filtration (Eqs.  S1,S2 ) substantially reduced the gmfe with both creatinine models, but resulted in overestimation (gmfe = 1.37–1.38). Creatinine is produced in the muscle and R_SYN depends on WT, age, and sex. ²² R_SYN in the creatinine models for healthy subjects was based on young adult men, ²⁹ whereas demographics of patients with CKD in this study were variable (e.g., WT range 44–96 kg). Consideration of both changes in GFR and R_SYN (Eq. S5) resolved the S_Cr,baseline overprediction (gmfe = 1.14–1.15). However, certain underprediction was still evident, implying the necessity to consider additional factors contributing to decreased creatinine elimination in CKD.

Table 2.

System parameters in creatinine models optimized for patients with CKD

Parameter	Unit	Description	Optimization for CKD	Value in healthy a , c (GFR = 125 mL/min)	Value in CKD (GFR = 15 mL/min) b , c
					INH scenario d		Non‐INH scenario d
GFR	L/h	Glomerular filtration rate	eGFR calculated using CKD‐EPI equation 1 (Eqs.  S1,S2 )	7.5	0.9
QPT,blood	L/h	Blood flow rate to proximal tubule	Percent reduction in renal blood flow in CKD relative to healthy 23	58	34
RSYN	mg/h	Endogenous creatinine synthesis rate	Calculated using the equation from Bjornsson et al. (Eq. S5) 22	71	46
VPT,bi	L	Volume of water in blood and interstitial space of cortex	Reduction in proportion to GFR (Eq. S6)	0.082	0.0098
VPT,cell	L	Volume of water in proximal tubule cells		0.066	0.0079
VPT,filt	L	Volume of proximal tubule filtrate		0.054	0.0064
QPT‐U,filt	L/h	Filtrate flow rate out of proximal tubule		2.7	0.3
CLPD,trans	L/h	Passive permeability by the transcellular routes		(Uptake) 0.89 (Bidirectional) 0.43	(Uptake) 0.11 (Bidirectional) 0.052
CLPD,para	L/h	Passive permeability by the paracellular routes		(Uptake) 5.9 (Bidirectional) 2.9	(Uptake) 0.71 (Bidirectional) 0.34
CLint,OAT2	L/h	CLint of OAT2 transporter	INH scenario d ; Reduction in proportion to GFR (Eq. S6) Non‐INH scenario d ; Additional deterioration in OAT2 (Eqs.  S9,S10 ) + clearances of other transporters optimized (Eq. S11)	(Uptake) 20.8 (Bidirectional) 21.5	(Uptake) 2.49 (Bidirectional) 2.58	(Uptake) 1.40 (Bidirectional) 1.45
CLint,OCT2	L/h	CLint of OCT2 transporter		(Uptake) 23.9 (Bidirectional) 8.75	(Uptake) 2.87 (Bidirectional) 1.05	(Uptake) 5.47 (Bidirectional) 3.33
CLint,MATE1	L/h	CLint of MATE1 transporter		(Uptake) 0.16 (Bidirectional) 0.16	(Uptake) 0.019 (Bidirectional) 0.019	(Uptake) 0.036 (Bidirectional) 0.062
CLint,MATE2‐K	L/h	CLint of MATE2‐K transporter		(Uptake) 0.51 (Bidirectional) 0.53	(Uptake) 0.061 (Bidirectional) 0.063	(Uptake) 0.117 (Bidirectional) 0.200

CKD, chronic kidney disease; CKD‐EPI, Chronic Kidney Disease Epidemiology Collaboration; CL_int, intrinsic clearance; eGFR, estimated glomerular filtration rate; INH, intact nephron hypothesis; MATE, multidrug and toxin extrusion protein transporter; OAT2, organic anion transporter 2; OCT2, organic cation transporter 2.

^aValues optimized for healthy patients, see Scotcher et al. (2019). ²⁰

^bAssuming a man at age 65 years old with serum creatinine of 4.5 mg/dL, body weight of 70 kg, and height of 170 cm.

^c(Uptake); parameter values in uptake‐OCT2 model, (Bidirectional); parameter values in bidirectional‐OCT2 model.

^dINH scenario ‐ decline in transporter activity proportional to GFR, non‐INH scenario ‐ changes in transporter activity disproportionate to GFR (details in Method section).

Assuming INH decline in transporter activity, all transporter clearances were decreased 53% in G3 (GFR 59 mL/min/1.73 m²) and 88% in G4 (GFR 15 mL/min/1.73 m²; Table  2 ). This approach recovered observed S_Cr,baseline (gmfe = 1.11–1.12; Figure  3a,c ), but underestimation of C_Cr/GFR was evident (Figure  3e,f ). Assuming decline in transporter activity disproportionate to GFR, deterioration of OAT2 activity was 65% or 93% in patients with GFR 59 and 15 mL/min/1.73 m², respectively (Figure  S4 ). Estimated Coeff_CKD,TP for uptake‐OCT2 and bidirectional‐OCT2 models were 0.88 and 0.70, respectively, resulting in less pronounced decrease in OCT2 and MATE‐mediated clearances relative to GFR (e.g., when GFR decreases 88% (15 mL/min/1.73 m²) relative to healthy, transporter clearances decrease 77% or 62% in uptake‐OCT2 and bidirectional‐OCT2 models, respectively). The non‐INH scenario successfully recovered both S_Cr,baseline (gmfe = 1.13; Figure  3b,d ) and overall means of C_Cr/GFR in each CKD group (Figure  3e,f ). Decrease in renal blood flow in CKD had marginal effect on S_Cr,baseline (data not shown).

Figure 3.

Development of creatinine chronic kidney disease (CKD) models. Predictability of serum creatinine (S_Cr,baseline, a–d) and ratios of creatinine clearance to glomerular filtration rate (GFR; C_Cr/GFR, e and f). Both creatinine uptake‐organic cation transporter (OCT)2 model (a, b, e) and bidirectional‐OCT2 model (c, d, and f) were optimized for patients with CKD based on two scenarios for transporter clearances: (a, c, and blue lines in e and f) decline in transporter activity proportional to GFR (intact nephron hypothesis (INH) scenario), (b,d, and red lines in e and f) changes in transporter activity disproportionate to GFR decline (non‐INH scenario, details in Methods section). a–d Circles represent patients with CKD from 8 clinical studies (Table  1 ), and solid and dashed lines represent a line of unity and 1.2‐fold error lines, respectively. gmfe, geometric mean fold‐error. In figure e and f, open circles represent mean C_Cr/GFR in individual clinical studies and filled circles represent overall means for each CKD stage (Figure  S3 ): blue = G1 (GFR > 90 mL/min/1.73 m²), cyan = G2 (GFR 60–89mL/min/1.73 m²), orange = G3 (GFR 30–59 mL/min/1.73 m²), pink = G4 (GFR 15–29 mL/min/1.73 m²), and red = G5 (GFR < 15mL/min/1.73 m²). Blue and red solid lines are simulated C_Cr/GFR based on INH scenario and non‐INH scenario for changes in transporter activity, respectively. Black dashed lines represent C_Cr/GFR = 1.

Verification of the developed creatinine CKD models was performed against independent datasets, including 42 patients with CKD (24 with G3 and 18 with G4, 22 men and 20 women, aged 22–68 years; Table  S8 ); S_Cr,baseline gmfe were < 1.32 for both INH and non‐INH scenarios (Figure  S16 ). Sensitivity analysis showed no sensitivity of %ΔS_Cr to changes in GFR in INH scenario (i.e., models predicted comparable extent of interaction between heathy and CKD). In contrast, in the non‐INH scenario, simulated %ΔS_Cr in patients with CKD were higher relative to healthy patients in case of OCT2 or MATE inhibition, whereas the opposite trend was seen for OAT2 (Figure  S9 ).

Pharmacokinetic models for renal transporter inhibitors in patients with CKD

A literature search identified one, four, and two clinical studies evaluating plasma concentration‐time profiles of trimethoprim, ³⁰ cimetidine, ³¹ , ³² , ³³ and famotidine, ³⁴ , ³⁵ respectively, in patients with CKD (Table  3 ). Fraction of unbound inhibitors in plasma in patients with CKD was 0.51, 0.84, and 0.72 for trimethoprim, cimetidine, and famotidine, respectively (Table  S6 ). These clinical data were used to develop operational PK models for each inhibitor (Supplementary Material Section  S6 ).

Table 3.

Pharmacokinetic studies of renal transporter inhibitors in patients with CKD

Inhibitor	Subject information (M; male, F; Female)	GFR, mL/min/1.73 m2	Study design	Blood sampling points, time after last dose	Reference
Trimethoprim	n = 9, M4 F5, 25–69 years	17–56	160 mg oral SD	1–48 hours	Rieder et al. (1974) 30 , a
Cimetidine	n = 5, 26–76 years	30–52b	200 mg oral SD	0.75–9 hours	Larsson et al. (1979) 33 , c
	n = 6, M4 F2, 43–66 years	23–47	Day 1–6; (200 mg × 4) per day oral Day7; 200 mg SD oral	0–9 hours on day 7	Larsson et al. (1981) 32 , d
	n = 8, M6 F2, 31–68 years	36–69	Day1‐6; (200 mg × 3 + 400 mg) per day oral Day7; 200 mg SD oral	0–9 hours on day 7	Larsson et al. (1981) 32 , e
	n = 8, M8 F0, 34–66 years	23–65	300 mg intravenous SD	0.25–16 hours	Ma et al. (1978) 31 , f
Famotidine	n = 5, M2 F3, 60–71 years	6–38b	20 mg oral SD	1–24 hours	Inotsume et al. (1989) 35
	n = 12, M10 F2, 28–54 years	10–41	10 mg intravenous SD	2.5 minutes–4 hours	Abraham et al. (1987) 34

CKD, chronic kidney disease; GFR, glomerular filtration rate; SD, single dosing.

^aSubjects in G3‐4 group (eGFR 15–59 mL/min/1.73 m²) was extracted based on individuals’ eGFR

^bCreatinine clearance (mL/min)

^cGroup with creatinine clearance of 30–52 mL/min

^dGroup with creatinine clearance of 30–50 mL/min

^eGroup with creatinine clearance of 50–75 mL/min

^fGroup with creatinine clearance of 49–87 mL/min (mild renal failure).

Prediction of creatinine‐drug interaction in patients with CKD

In total, 12 clinical studies (90 patients in CKD G3–4, age 22–88 years) were collated for the evaluation of the ability of creatinine CKD model to predict %ΔS_Cr (Table  1 ). The effect of renal transporter inhibitors was initially simulated using unbound plasma concentration (C_p,u) as an inhibitory concentration against all transporters. Assuming that transporter activity changes disproportionately to disease‐related changes in GFR resulted in higher predicted %ΔS_Cr than the model with INH assumptions; this difference was more evident in the bidirectional‐OCT2 model (Figure 4). Non‐INH model assumptions resulted in 66% of predicted %ΔS_Cr within prediction limits relative to 58% for the INH scenario; trends were consistent regardless of OCT2 directionality assumption (Table  S9 ). Relatively higher predictability was seen for trimethoprim and famotidine (60 or 100% of studies within prediction limits, respectively), whereas underestimation of %ΔS_Cr was seen for 40–60% of cimetidine studies regardless of the model. One potential contributor to this underprediction is the accumulation of inhibitors within the proximal tubule that was not accounted for when C_p,u was applied as inhibitory concentration. Use of inhibitor concentrations in proximal tubular filtrate as a pragmatic/worst‐case scenario for MATE transporters ²¹ improved overall predictability (75–83% within the prediction limits), except for the bidirectional‐OCT2 model in the non‐INH scenario (58% within the prediction limits; Figure  S17 and Table  S10 ). Predictability of cimetidine interactions was overall improved regardless of the model (80–100% within the prediction limits).

Figure 4.

Predictability of percent change in serum creatinine after administration of renal transporter inhibitors. Predicted percent change in serum creatinine (S_Cr) post administration of inhibitors using (a, b) uptake‐organic cation transporter (OCT)2 model and (c,d) bidirectional‐OCT2 model based on two scenarios for transporter clearances: a, c decline in transporter activity proportional to glomerular filtration rate (GFR; intact nephron hypothesis (INH) scenario), b, d changes in transporter activity disproportionate to GFR (non‐INH scenario, details in Method section). Filled symbols and error bars represent means and standard deviations of percent change in S_Cr in each clinical study with three inhibitors; red circles = trimethoprim, green triangles = cimetidine, and blue squares = famotidine. Simulations were performed based on unbound concentrations of inhibitors in plasma as inhibitory concentration for all transporters. Solid and dashed lines represent line of unity and prediction error limits considering intra‐individual variability in baseline S_Cr in the CKD population (8.9%), respectively. MAE, mean absolute error.

In addition to prediction of the mean inhibitory effect per study, the predictability of individual %ΔS_Cr was evaluated using the clinical data from 32 patients with CKD (G3; 18 patients, G4; 14 patients) that received trimethoprim (Table  1 ). The individual %ΔS_Cr were highly variable (ranging from −20% to > 50%) in both CKD G3 and G4 (Figure  S18 ). Simulations based on C_p,u as an inhibitory concentration resulted in 34–47% of predicted individual data within assigned limits (Table  S11 ). There was a tendency for higher prediction accuracy in CKD G3 (33–67% vs. 21–36% for patients with CKD G4), but this trend was based on a limited number of subjects.

DISCUSSION

Increased S_Cr post drug dosing requires careful interpretation because it can be caused by inhibition of renal transporters even in the absence of kidney injury, leading to the inappropriate discontinuation of medical treatments or misinformation in clinical trials in drug development. ² , ³⁶ Further consideration may be necessary for patients with CKD due to altered disposition of both creatinine and inhibitors as a result of the disease. Regulatory agencies have alerted about the possibility of altered drug‐drug interactions in patients with impaired renal function. ³⁷ Therefore, a tool elucidating the true cause of increased S_Cr in this patient cohort would be useful to improve decision making in clinical practice. Several studies have reported creatinine models that can simulate creatinine‐drug interaction risk in healthy subjects, ¹⁸ , ¹⁹ , ²⁰ , ²¹ but to the best of our knowledge, so far these efforts have not been extended to patients with CKD. This study showed a novel approach to simulate creatinine‐drug interaction in patients with CKD using mechanistic physiologically‐based pharmacokinetic models of creatinine combined with conventional PK models for inhibitors of renal transporters.

Patients with CKD in the Salford Kidney Study showed higher intra‐individual variability in S_Cr,baseline (8.9%) than healthy subjects (4.7%). ²⁸ In addition, large interindividual variability in %ΔS_Cr was evident, consistent with previous clinical studies in the CKD population. The deterioration of renal function over time (not considered in our model), could contribute to these variabilities in patients with CKD. A continuous increase in S_Cr due to the progression of CKD can result in a large change in S_Cr during the observation period, which could lead to the underestimation of true S_Cr,baseline and potential overestimation of %ΔS_Cr (patient ID8 and 12; Figure  S1 ). Higher interindividual variability in C_Cr/GFR in CKD (G3; 34% and G4; 42%) relative to healthy subjects (G1; 18%) may also contribute to large interindividual variability in %ΔS_Cr (Table  S5 )_.

Degree of creatinine‐drug interaction in healthy subjects and patients with CKD

Only a few clinical studies compared the %ΔS_Cr with the same dosage regimen between healthy and CKD populations in a single clinical study. ³⁸ , ³⁹ Our comprehensive literature analysis showed the tendency for higher %ΔS_Cr in patients with CKD relative to healthy subjects at a daily dose of < 400 mg/day of trimethoprim (Figure  S2 ). The overall comparison between two populations was based upon insufficient data to be conclusive on whether CKD leads to more pronounced %ΔS_Cr. Nevertheless, higher interaction in the CKD population remains a possibility because dosage regimens of inhibitors had already been adjusted for reduced renal function in some studies reported in patients with CKD, possibly masking the difference between populations for trimethoprim (CKD = 33% vs. healthy = 15%) and famotidine (CKD = 7% vs. healthy = 1%), ³⁸ , ³⁹ albeit with larger variability in CKD.

Optimization of creatinine models for patients with CKD

In order to capture disease‐related physiological changes, the creatinine CKD models included decreased glomerular filtration and modification of multiple physiological parameters based on several assumptions. For example, R_SYN regression equation accounted for differences in WT and age, based upon three independent clinical studies with subjects who did not show severe CKD (mean S_Cr < 1.8 mg/dL) ⁴⁰ , ⁴¹ , ⁴² ; interindividual variability in R_SYN was < 35% regardless of age or sex (Figure  S6 ). R_SYN was assumed to be unaffected by the progression of CKD because marginal changes in synthesis were reported in individuals with S_Cr ranging from 1.5 to 5 mg/dL. ⁴⁰ Application of INH assumptions to volumes of proximal tubule compartments, CL_PD, and Q_U‐filt was based on the principle that the number of proximal tubular cells, tubular surface area, and filtrate flow out of the proximal tubule are likely to decrease in proportion to the number of intact nephrons, respectively. Despite CKD‐dependent changes occurring in filtrate pH and flow rate, fraction of creatinine reabsorbed in distal tubule was not affected, supported also by a previous study reporting no sensitivity of creatinine renal clearance to these parameters. ⁹

In addition to INH assumptions, where transporter activity declines proportionally to GFR, an alternative scenario was explored in the creatinine CKD model, assuming changes in transporter clearances that are not consistent with the GFR decline. Deterioration of OAT2 activity implemented in this non‐INH scenario (65–93%) was comparable to those reported for OAT1/3 (66–95%). ⁶ In the case of OCT2 and MATEs, relative decline in transporter activity was smaller compared with proportional changes assumed under the INH. Further investigations are necessary to elucidate fully changes in the functional activity of OCT2 and MATEs in patients with CKD.

Predictability of creatinine CKD models

Following non‐INH assumptions for transporter clearances, creatinine CKD models showed higher sensitivity to inhibition of OCT2/MATEs relative to models for healthy populations; opposite trend was seen for OAT2 (Figure  S9 ). These differences are attributed to changes in fraction transported and change in overall contribution of secretion compared with filtration and reabsorption in the non‐INH scenario. In contrast, CKD models assuming decline in transporter activity proportional to GFR (INH scenario) showed similar sensitivity to transporter inhibition to healthy subjects, because fraction of creatinine transported by renal transporters were minimally affected under these assumptions.

Higher sensitivity of the models with non‐INH transporter assumptions to creatinine‐drug interactions was also reflected in the predictive performance of %ΔS_Cr (Figure  4 ). Simulations of the %ΔS_Cr based on C_p,u as inhibitory concentration for all transporters resulted in 66% of clinical studies within the proposed prediction limits (Table  S9 ) and improved predictive performance to healthy population (59% and 51% in uptake‐OCT2 and bidirectional‐OCT2 model, respectively). ²¹ Underestimation of %ΔS_Cr for cimetidine and improved predictability with C_PT,filt were consistent between creatinine models for CKD and healthy populations. ²¹ Use of C_PT,filt tended to exacerbate overestimation of trimethoprim‐creatinine interactions in CKD. The original creatinine models ²⁰ , ²¹ were optimized with trimethoprim interaction in healthy subjects and with C_p,u as inhibitory concentration for all transporters. This approach may have resulted in bias by compensating for the difference in the inhibitor concentration in plasma and the proximal tubular filtrate, leading to the overestimation of trimethoprim interaction when C_PT,filt was applied. The application of C_PT,filt was a pragmatic approach to explore the worst‐case scenario. Improved predictability for cimetidine and overestimation for trimethoprim with C_PT,filt highlight potential limitations of empirical PK models ignoring the intracellular concentration of inhibitors in proximal tubular cells. Mechanistic modelling of inhibitors, ⁴³ , ⁴⁴ which was beyond the scope of current work, would enable us to address these limitations.

Despite reasonable recovery of the mean observed %ΔS_Cr per study, interindividual variability of %ΔS_Cr was not captured by the proposed creatinine CKD models. Multiple factors could contribute to the underestimation of the extent of this interindividual variability. Empirical compartment PK models of inhibitors could not consider interindividual variability in the inhibitors’ plasma exposure due to limited data. PK data from patients with CKD G3 and G4 were not differentiated in the development of these PK models, potentially resulting in underestimation of the impact of CKD severity on the PK of these drugs. In addition, lack of description of disease progression and longitudinal changes in GFR and other physiological parameters or interindividual variability in C_Cr/GFR in the model may contribute to underestimation of interindividual variability in %ΔS_Cr.

In conclusion, elevation of S_Cr is likely to be interpreted as acute kidney injury and can result in the discontinuation of new drug development or clinical treatment. Inhibition of renal transporters also causes elevated S_Cr, as observed in our creatinine‐trimethoprim interaction study in patients with moderate‐to‐severe CKD. The developed creatinine CKD model enabled quantitative prediction of the increase in S_Cr resulting from deteriorated renal function and identified challenges in quantitative translation to patients. In addition, modelling allowed differentiation of the effect of disease from inhibition of renal transporters with the ultimate goal to provide a valuable tool for prospective evaluation of drug interaction risk via renal transporters in this patient population.

Funding

H.T. was financially supported by a fellowship grant from Asahi Kasei Pharma Corporation.

Conflict of Interest

The authors declared no competing interests for this work.

Author Contributions

H.T., D.S., R.C., P.K., and A.G. wrote the manuscript. H.T., D.S., R.C., P.K., and A.G. designed the research. H.T. and R.C. performed the research. H.T., D.S., and R.C. analyzed the data.

Supporting information

Supplementary Material

Supplementary Material