Pharmacokinetics-Pharmacodynamics of a Sordarin Derivative (GM 237354) in a Murine Model of Lethal Candidiasis

Abstract

Sordarins are a new class of antifungal agents which selectively inhibit fungal protein synthesis (FPS) by impairing the function of elongation factor 2. The present study investigates possible correlations between sordarin pharmacokinetic (PK) properties and therapeutic efficacy, based on a murine model of invasive systemic candidiasis, and provides a rationale for dose selection in the first study of efficacy in humans. A significant correlation between PK parameters and the in vivo activity of GM 237354, taken as a representative FPS inhibitor, was demonstrated in a murine model of lethal systemic candidiasis. Area under the concentration-time curve (AUC) and maximum concentration of drug in serum (C_max) over 24 h were determined after a single GM 237354 subcutaneous (s.c.) dose (50 mg/kg of body weight) in healthy animals (no significant PK changes with infection were observed for other sordarin derivatives). These results have been used to simulate PK profiles obtained after several doses and/or schedules in animal therapy. A PK-pharmacodynamic (PD) parameter such as the time that serum drug concentrations remain above the MIC (t > MIC) was also determined. Treatment efficacies were evaluated in terms of the area under the survival time curve (AUSTC), using Kaplan-Meier survival analysis and in terms of kidney fungal burden (log CFU/gram) after s.c. doses of 2.5, 5, 10, 20, and 40 mg/kg every 4, 8, or 12 h (corresponding to total daily doses of 5 to 240 mg/kg). The results show all treatments to significantly prolong survival versus that of infected and nontreated controls (P < 0.05). Relationships between simulated PK and PK-PD parameters and efficacy were explored. A good correlation independent of the dosing interval was observed with AUC (but not C_max or t > MIC) and both AUSTC and kidney burden. Following repeated dosing every 8 h, AUC₅₀ (AUC at which 50% of the maximum therapeutic efficacy is obtained) was estimated as 21.7 and 37.1 μg · h/ml (total concentrations) for AUSTC and kidney burden using a sigmoid E_max and an inhibitory sigmoid E_max PK-PD model, respectively. For an efficacy target of 90% survival, AUC was predicted as 67 μg · h/ml. We conclude that the PK-PD approach is useful for evaluating relationships between PK parameters and efficacy in antifungal research. Moreover, the results obtained with this approach could be successfully applied to clinical studies.

Infections caused by opportunistic fungal pathogens remain an important clinical problem. Candida albicans is the major fungal pathogen. Deep-seated infections due to this organism are an important cause of nosocomial infections, and the morbidity and mortality associated with C. albicans infections remain significant. Although in recent years there has been an expansion in the number of antifungal drugs available, in many cases the treatment of fungal infections is unsatisfactory.

This situation has led to an ongoing search for new antifungal agents. The determination of ultimate outcome is a function of multiple variables, though much of it can be explained on the basis of intrinsic microbiological activity (in vitro) and the serum concentration-time profile (in vivo) (11). Correlations are available for β-lactam antibiotics, aminoglycosides, and quinolones (12–15, 19, 21, 24, 25, 27, 28). In the case of β-lactams, improved outcome is associated with time that serum drug concentrations remain above the MIC (t > MIC) for the pathogenic agent (22, 23, 28). In contrast, for aminoglycosides and quinolones, the maximum antimicrobial effect is associated with higher ratios of maximum concentration of drug in serum (C_max) or area under the concentration-time curve (AUC) to the MIC (C_max/MIC or AUC/MIC ratios, respectively) (13, 15, 26).

However, little is known about antifungal agents (7, 29, 30). Anaissie et al. (2) showed some correlation between in vitro parameters and in vivo efficacy, but no clear reference to pharmacokinetic (PK) parameters was described. Graybill et al. (16) reported results obtained with fluconazole-susceptible or -resistant isolates in an experimental murine candidiasis model. The correlation found in this study was not very high. In vitro susceptibility tests could predict in vivo response to fluconazole: susceptible C. albicans strains (MIC ≤ 0.25 μg/ml) required lower daily doses than did resistant C. albicans strains (MICs from 8 to 64 μg/ml). However, in most cases, therapeutic decisions are governed more by the clinical experience of the physician than by the preclinical results even when amphotericin B is considered (18).

Recently, Louie et al. (20) defined the pharmacodynamic (PD) parameter that optimizes outcome in deep-seated C. albicans infections treated with fluconazole intraperitoneally, based on a murine model of systemic candidiasis, and taking into account the fungal reduction in the kidneys. Dose fractionation studies showed that the AUC/MIC ratio (20) best predicted the outcome with fluconazole (3).

Fungal protein synthesis (FPS) inhibitors are a new family of antifungal drugs, with a novel mechanism of action (9, 10, 17) and no former related therapeutic experience. The aim of the present study was thus to define a possible correlation between the PK properties of FPS inhibitors and therapeutic efficacy, using a murine model of invasive systemic candidiasis, and to provide a rationale for dose selection in the first study of efficacy in humans.

(Part of this work was presented at the 38th Interscience Conference on Antimicrobial Agents and Chemotherapy, San Diego, Calif., 24 to 27 September 1998 [P. Aviles, C. Falcoz, C. Efthymiopoulos, R. San Roman, A. Martinez, E. Jimenez, M. S. Marriott, A. Bye, F. Gomez De Las Heras, and D. Gargallo-Viola, Abstr. 38th Intersci. Conf. Antimicrob. Agents Chemother., abstr. J-074, 1998].)

MATERIALS AND METHODS

C. albicans isolate.

C. albicans 4711E (GlaxoWellcome culture collection, Greenford, United Kingdom) was used throughout the study. The strain was maintained in Sabouraud dextrose (SAB) agar (Difco, Detroit, Mich.) with 15% glycerol at −70°C until required. For inoculum preparation, C. albicans was cultured on SAB agar (Difco) plates. The resulting growth was collected from the plates in sterile 0.9% NaCl, and infecting inocula were adjusted by the spectrophotometric method to the appropriate concentrations (colony counts were verified on SAB plates).

The MIC of GM 237354 was determined on five separate occasions by a method described elsewhere (17). Briefly, a Microlab AT Plus robot (Hamilton Bonaduz, Bonaduz, Switzerland) was used to prepare microdilution panels containing twofold dilutions of the drugs in 0.1 ml of medium. Starting inocula were adjusted by the spectrophotometric method to 10⁶ CFU/ml. Then, the adjusted yeast suspensions were diluted 1:10 with medium, and microtiter plates were inoculated with this dilution (by using the Hamilton system to dispense 10 ml into each well) to obtain a final inoculum of approximately 10⁴ yeast cells per ml. The inoculated plates were incubated at 35°C without agitation for 24 h. Following incubation and after agitation with a microtiter plate shaker for 5 min, the plates were read visually with the aid of a reading mirror and spectrophotometrically with an automatic plate reader (IEMS; Lab Systems, Helsinki, Finland) set at 620 nm. The MIC was defined as the lowest concentration of antifungal agent which prevented any visible growth or which inhibited growth by 95% compared with that in drug-free control wells. The median MIC after the incubation period was 0.001 μg/ml (range, 0.001 to 0.004 μg/ml).

Antifungal agent.

GM 237354 (as potassium salt) was synthesized at Glaxo Wellcome S.A. (Tres Cantos, Madrid, Spain). For animal treatments, the compound was dissolved in sterile deionized water to reach desired concentrations. The drug was used immediately. The antifungal doses were expressed as milligrams of base (active compound) per kilogram of body weight.

Mice.

CD-1 male nonimmunosuppressed mice (weight, 25 to 30 g; Charles River France Inc., Lyon, France) were used. The animals were housed in cages of 10 individuals per group and allowed to acclimatize for at least 7 days prior to infection. The mice had free access to food and water throughout the experimental period. The research complied with Spanish national legislation and with company policy on the care and use of animals and related codes of practice.

Serum protein binding of GM 237354.

Binding was determined in mouse serum by equilibrium dialysis using [³H]GM 237354. Radiochemical purity was assessed by high-pressure liquid chromatography (HPLC)–UV analysis. [³H]GM 237354 at concentrations ranging from 8 to 80 μg/ml in 1.0 ml of blank serum was incubated at 37°C for 2 h before starting the experiment. The dialysis chambers (Cellu-Sep; Spectrum) have a volume of 250 μl and are separated by a membrane measuring 1 cm². Visking tubes were cut into 1- by 1-cm squares and soaked in red blood cell buffer (150 mM sodium chloride, 0.8 mM magnesium chloride, 0.2 mM calcium chloride in deionized water) for 1 h. Once the membrane was fixed in place, the first compartment was filled with 200 μl of serum sample. Serum samples were dialyzed against the same volume of red blood cell buffer placed into the second compartment. The chamber was placed in a rotator (Dianorm, Munich, Germany), and dialysis was carried out at 16 rpm at 37°C for 24 h. Following incubation, aliquots of both compartments were counted and the free fraction was calculated.

Single-dose PKs of GM 237354.

Noninfected mice were injected with a 50-mg/kg single dose in 500 μl of vehicle subcutaneously (s.c.). The dose proportionality of the kinetics had been verified in a pilot study where healthy animals were treated with 40 and 5 mg of GM 237354 per kg. At 0, 0.25, 0.5, 0.75, 1.5, 2, 2.5, and 3 h, four animals were sacrificed, blood samples were collected by cardiac puncture from each animal and centrifuged, and then serum samples were stored at −20°C. The concentration of GM 237354 in each serum sample was determined by HPLC as described below, and the mean of four values obtained for each sampling time was used for PK analysis. The HPLC analysis used a previously described method (P. Aviles, A. Pateman, R. San Roman, and D. Gargallo-Viola. Abstr. 37th Intersci. Conf. Antimicrob. Agents Chemother., abstr. F-066, 1997). Briefly, GM 237354 was resolved by using an HP 1090 HPLC equipped with a diode array UV detector (Hewlett-Packard, Palo Alto, Calif.), a 4.5- by 15-cm Novapack RP-C₁₈ column with a guard column of the same material (Waters Corporation, Milford, Mass.) maintained at 50°C, and a Vectra 486/66U computer with HP Chemstation software (Hewlett-Packard). The mobile phase was shaped by acetonitrile (Panreac Quimica S.A., Barcelona, Spain) and phosphate-octane sulfonic acid (Reactivos Scharlau S.L., Barcelona, Spain) solution buffered at pH 5. The flow rate was 1 ml/min, and UV detection of the compound was performed at 215 nm. Chromatography was isocratic at 78% acetonitrile. Standard curves (linearity from 100 to 2 μg/ml) were generated by adding known amounts of GM 237354 to pooled mouse serum (Charles River France Inc.). Before HPLC analysis, standard and unknown samples were deproteinized by mixing (1:1 [vol/vol]) with acetonitrile. Mixtures were shaken for 2 min and then centrifuged at 1,000 × g for 15 min at 5°C. Twenty microliters of supernatant was injected for HPLC analysis. Areas of peaks were measured, and concentrations of unknown samples were extrapolated from the regression line calculated with standard sample results.

PK analysis.

The concentration of GM 237354 in each serum sample was determined by HPLC as described above, and the mean of four values obtained for each sampling time was used for PK analysis. PK parameters were derived from the serum concentration-time data on the basis of a one-compartment open model with first-order absorption-elimination kinetics. AUC and C_max were calculated with WinNonlin software (Scientific Consulting, Inc., Apex, N.C.).

PK simulations.

PK profiles for 40, 20, 10, 5, and 2.5 mg/kg were simulated considering the experimental profile obtained after a single s.c. dose of 50 mg/kg. The AUC over 24 h (AUC) and C_max were simulated after a multiple-dose regimen with the above doses administered every 4, 8, and 12 h. Simulations were performed using the WinNonlin software package.

Systemic infection.

A pilot experiment to determine the inoculum sizes of C. albicans that would result in a survival time for infected-nontreated control animals of at least 7 days was performed (data not shown). Mice were challenged intravenously with 200 μl of the appropriate inoculum (10⁵ CFU) into the lateral tail vein. Thirteen infected animals were left untreated, and the rest were randomly assigned to treatment groups of 10 to 13 animals each.

Antifungal treatment.

Therapy was initiated 1 h after inoculation and was continued for 7 days. Deaths were recorded daily up to 28 days postinoculation (1). GM 237354 was administered s.c. at 40, 20, 10, 5, and 2.5 mg/kg every 4, 8, and 12 h for total daily doses of 5 to 240 mg/kg.

Fungal burdens of kidneys.

Twelve hours after the end of treatment, three randomly selected animals from groups dosed every 8 h were sacrificed. The kidneys were removed, weighed, and homogenized with 5 ml of cold sterile saline in a blender (Stomacher 400; Seward Medical, London, United Kingdom). Samples from each specimen were diluted and spread onto SAB plates. After 24 h of incubation at 35°C, the log CFU per gram of kidney were calculated. Drug carryover was avoided by using a washout procedure described elsewhere (25).

Efficacy parameters.

The efficacy parameters used to assess treatment success were the following: (i) percentage of survivors and median survival day obtained from Kaplan-Meier analysis; (ii) survival time expressed as a net effect, ES = K_t − K_k, where K_t is the area under the survival time curve (AUSTC) obtained with infected animals receiving treatment and K_k is the AUSTC obtained with infected-nontreated animals; and (iii) kidney burden expressed as the measured effect (log CFU per gram) in the absence and presence of treatment.

Correlation between survival and PK parameters (AUC and C_max).

Most of the interdependence among PK-pharmacodynamic parameters can be reduced by comparing the results of dosage regimens that are based on different dosing intervals (8). The relationship between PK parameters (total serum concentrations) and effect on survival time curve (AUSTC) was first evaluated graphically, for each dosing interval. Only the PK parameter(s) which could describe efficacy independently of the dosing regimen with a similar trend for all three dosing intervals was selected for the PK-PD modelling analysis. The relationship between AUC and the net effect on AUSTC was then modeled with the Hill equation using WinNonlin software and the Nelder-Mead algorithm. No weighting was used (21). The general equation used is given as

where AUC is the steady-state 24-h AUC obtained after PK simulations, ES is the net effect observed, E_max is the maximum net effect, AUC₅₀ is the AUC at which 50% of the maximum efficacy is obtained, and γ is the Hill factor determining the slope of the curve. After a preliminary analysis which showed a large uncertainty for the 12-h-dosing-interval PK-PD parameter estimates, E_max was fixed at 1,825% · day, which is equal to the theoretical maximal value for K_t which would be observed in noninfected untreated control animals (i.e., 100% survival during the experimental period = 2,800% · day = 28 days × 100%) − the effect K_k measured in control, infected-nontreated animals (975% · day). The model could not be fitted to individual data due to the nature of the endpoint. The sigmoid E_max model was evaluated in three separate ways: (i) for each set of dosing interval data separately, (ii) with common γ values but a different AUC₅₀ value for each of the three dosing intervals, and (iii) by pooling all the data for the three dosing intervals. The simple E_max model (γ = 1) was also explored with pooled data for the three dosing intervals.

The AUC which would provide a desired effect, E_ther (i.e., 90% of the maximum therapeutic efficacy), can be calculated as follows:

Thus, for 90% efficacy,

Correlation between kidney burden and PK parameters (AUC and C_max).

The relationship between i.e. AUC (total serum concentrations) and kidney burden was evaluated with an inhibitory sigmoid E_max model, including the baseline E₀, which is the kidney burden in infected-nontreated animals at the end of the 7-day treatment period. The equation used is given as:

where AUC is the steady-state 24-h AUC obtained after PK simulations, E is the effect measured, E_max is the maximum net effect, AUC₅₀ is the AUC at which 50% of the maximum therapeutic efficacy is obtained, and γ is the Hill coefficient. Individual data (approximately three animals per dose) were used for the analysis. The same modeling conditions as described above were used.

The AUC which would provide a desired therapeutic effect, E_ther (i.e., 90% of the maximum therapeutic efficacy), can be calculated as follows:

Thus, for 90% of maximal efficacy:

Correlation between efficacy and PK-PD parameters (AUC/MIC and t > MIC).

AUC/MIC, defined as the ratio between AUC and the corresponding MIC (27), would be another parameter to be assessed in terms of free serum concentrations for different compounds of the same class. AUC₅₀/MIC values could be calculated in terms of free concentrations by correcting for the binding of a specific compound (unbound fraction = 0.05 for GM 237354 in mice), the AUC₅₀ values being estimated for total concentrations as described in the previous section. However, PK-PD data are available only for one FPS inhibitor, GM 237354, and consequently AUC/MIC values cannot be compared for different compounds of the same class and will not be presented in Results.

On the other hand, t > MIC has been successfully used to describe PK-PD relationships with different antibacterials (14, 27, 28). t > MIC was determined for each dosing regimen based on free compound concentration. Its relationship with effect was first evaluated graphically and could be described by a sigmoid E_max model for the 8- and 12-h dosing intervals (data for the 4-h dosing interval could not be analyzed, as t > MIC was 100% whatever the dose and the corresponding ES value [see Fig. 6]) as

FIG. 6.

Graphic representation of the relationship between observed net effect on survival time curve (ES), survival rate, kidney burden, and t > MIC based on free serum concentrations (Tau = dosing interval).

RESULTS

PKs.

Figure 1 displays the GM 237354 serum concentration-time curve observed after s.c. administration of 50 mg/kg (total concentration) or simulated after i.e. a 40-mg/kg s.c. dose (free and total concentrations). Table 1 displays the main PK parameters, including those obtained in a pilot study performed to explore dose proportionality.

FIG. 1.

Simulated PK profiles for total (solid line) and free (dashed line) GM 237354 serum concentrations following a single s.c. 40-mg/kg dose. The inset displays mean (± standard deviation) total serum concentrations observed following a single 50-mg/kg dose.

TABLE 1.

PK parameters of GM 237354 in healthy mice for total concentration

Parametera	s.c. single dose (mg/kg)
	50	40	5
AUC0–∞ (μg · h/ml)	46.04	30.7	2.33
Cmax (μg/ml)	23.04	21.8	3.16
t1/2 (h)	0.52	0.40	0.36
Tmax (μg/ml)	0.60	0.14	0.52

^aAUC_0–∞, AUC from 0 h to infinity; t_1/2, half-life; T_max, time after administration at which C_max is reached. 

PK simulations for AUC and C_max were based on total serum concentrations and were carried out for repeated dosing using the PK parameters estimated above for the 50-mg/kg dose. Single-dose PK parameters were used to predict repeated-dosing PKs (Table 2), as GM 237354 did not show accumulation, even when using the shortest dosing interval (every 4 h) and the highest dose simulated (40 mg/kg). t > MIC values were derived from free concentration profiles, derived from simulated total serum concentrations corrected for binding.

TABLE 2.

PK parameters (AUC and C_max) predicted after s.c. repeated dosing

Simulated dose (mg/kg)	Predicted value
	AUC over 24 h (μg · h/ml) for dosing interval:			Cmax (μg/ml) for all dosing intervalsa
	4 h	8 h	12 h
40	220.8	110.4	73.6	18.72
20	110.4	55.3	36.8	9.36
10	55.3	27.6	18.4	4.68
5	27.6	13.8	9.2	2.34
2.5	13.8	6.9	4.6	1.17

^aEqual value for all dosing intervals as there is no evidence of compound accumulation upon repeated dosing. 

Efficacy parameters.

All untreated animals died within 8 and 11 days postinfection. Survival accumulative distributions are displayed in Fig. 2, with derived AUSTC and ES values shown in Table 3. Kaplan-Meier survival analysis showed statistical differences between the treated animals and the untreated controls, even at the lowest dose, i.e., 2.5 mg/kg every 12 h (P ≤ 0.0001). As shown in Fig. 3, a good consistency was observed between short-term (C. albicans in kidneys 7 days postinfection) and long-term (survival rate at day 28 or AUSTC) measures (data available for all endpoints only for the 8-h dosing interval).

FIG. 2.

Cumulative survival of mice with systemic candidiasis treated with GM 237354 administered at doses of 2.5 (⧫), 5 (▾), 10 (▴), 20 (■), or 40 (●) mg/kg every 4 (A), 8 (B), and 12 (C) h or left untreated (○).

TABLE 3.

Long-term efficacy obtained with different dosing intervals throughout the treatment period

Dose (mg/kg)	Value (% · day) for dosing intervala
	AUSTC			ES = Kt − Kk
	4 h	8 h	12 h	4 h	8 h	12 h
40	2,800	2,755	2,330	1,825	1,780	1,355
20	2,775	2,620	1,510	1,800	1,645	535
10	2,350	2,065	1,665	1,375	1,090	690
5	1,535	1,390	1,345	560	415	370
2.5	1,330	1,349	1,318	355	374	342
Untreated		975			0

^aAUSTC values are obtained from plots of cumulative survival versus days (Fig. 2). 

FIG. 3.

Influence of dose (milligrams per kilogram every 8 h) on observed efficacy parameter values.

Correlation between survival and PK parameters (AUC and C_max).

Using total plasma concentrations, a graphic visual evaluation showed efficacy to be related independently of the dosing interval only to AUC, not to C_max (Fig. 4 and 5). Results from Table 3 were modeled to obtain the parameters shown in Table 4. Several models were tested, as summarized in Table 4 and explained as follows.

FIG. 4.

Relationship between observed net effect on survival time curve (ES) and PK parameters (AUC and C_max). Data were pooled for the three dosing intervals.

FIG. 5.

Graphic representation of the relationship between observed net effect on survival time curve (ES) and PK parameters (AUC and C_max). Data are shown for each dosing interval (Tau).

TABLE 4.

PK-PD parameter estimates for the relationship between the net effect ES on survival time curve (AUSTC) and kidney burden and GM 237354 AUC (total serum concentrations)^a

Parameter	Model	Dosing interval (h)	γ (CV %)	Emax (CV %)	Baseline E0 (CV %)	AUC50 (CV %)	Correlation
Survival time	Sigmoid Emax model for each dosing interval	12	0.9 (40.7)		NA	41.2 (37.5)	0.862
		8	1.9 (20.3)		NA	21.7 (11.4)	0.985
		4	2.3 (19.7)		NA	34.7 (9.4)	0.989
Survival time	One sigmoid Emax model for each dosing interval, common γ	12				39.8 (19)	0.882
		8	1.7 (17.7)		NA	21.4 (18.4)	0.982
		4				33.5 (18.6)	0.985
Survival time	One sigmoid Emax model for all dosing intervals (pooled data)	12, 8, 4	1.5 (19.4)		NA	29.3 (12.7)	0.927
Survival time	One simple Emax model for all dosing intervals (pooled data)	12, 8, 4	NA		NA	27.8 (18.7)	0.920
Kidney burden	Inhibitory sigmoid Emax model	8	1.5 (29.7)	4.4 (21.5)	6.03 (3.3)	37.1 (31.9)	0.970

^aAUC₅₀ values are shown as microgram-hours per milliliter; E₀ and E_max values are shown as log CFU per gram. The net effect was defined as ES = (K_t − K_k) (see Materials and Methods). E_max values for survival were fixed at 1,825% survival · day. Correlation is that between observed and predicted values provided by WinNonlin software. NA, not applicable. 

Using a sigmoid E_max model for each dosing interval separately, the data could be well fitted for the 8- and 4-h dosing intervals; however, the fit was reasonably good only for the 12-h dosing data (larger coefficients of variation [CVs] for the estimated parameters). AUC₅₀ was estimated at 21.7 and 34.7 μg · h/ml and γ was estimated at 1.95 and 2.34 for dosing intervals of 8 and 4 h, respectively. For an efficacy target of 90% of maximum effect, AUC₉₀ was predicted as 67 μg · h/ml (8-h dosing interval).

A sigmoid E_max model with a common γ value but a different AUC₅₀ value for each dosing interval did not provide any substantial improvement in the fit. No real change in AUC₅₀ estimates was observed.

A sigmoid E_max model pooling data from all three dosing intervals provided a reasonable fit, with AUC₅₀ estimated at 29.3 μg · h/ml and γ estimated at 1.5 (Table 4). The simple E_max model pooling data from all three dosing intervals gave a poorer fit.

Correlation between survival and t > MIC.

The PK-PD relationship with t > MIC at steady state was derived for ES (net effect on AUSTC) using t > MIC calculated for unbound concentrations (5% of the corresponding total concentrations) as depicted in Fig. 1. Figure 6 shows the very steep relationship between ES and t > MIC, with 50% of E_max being achieved with a t > MIC of 77.1 and 59.2% for the 8- and 12-h dosing intervals, respectively. γ was estimated as 15.9 and 6.2 for the 8- and 12-h intervals, respectively (Table 5). For the 4-h dosing interval, t > MIC was 100% for all doses, although efficacy was poor for certain doses (Fig. 2 and 6). This shows that t > MIC is not a valid PK-PD predictor.

TABLE 5.

PK-PD parameter estimates for the relationship between the net effect ES on survival time curve (AUSTC) and t > MIC (free serum concentrations)^a

Dosing interval	Estimated value
	t > MIC (CV %)	γ (CV %)	Correlation
4 h	NA	NA
8 h	77.1 (2.1)	15.9 (35.6)	0.979
12 h	59.2 (5.9)	6.2 (46.5)	0.824

^aNA, not applicable, as t > MIC values are 100% for all doses. t > MIC values are shown as percentages of the dosing intervals. Correlation is that between observed and predicted values provided by WinNonlin software. 

Correlation between kidney burden and PK parameters (AUC and C_max).

Figure 7 shows an excellent correlation between AUC (total serum concentrations) and kidney burden (log CFU per gram) (observed and fitted data; data available only for the 8-h dosing interval). PK-PD parameters were estimated at 37.1 μg · h/ml for AUC₅₀, 4.40 log CFU/g for E_max, and 1.51 for γ. The baseline E₀ was estimated at 6.03 log CFU/g. For an efficacy target of 90% of maximum effect, AUC₉₀ was predicted as 159 μg · h/ml.

FIG. 7.

Relationship between effect on kidney burden (individual values) and AUC (8-h dosing interval). Fitted and observed data are shown.

A satisfying consistency in the PK-PD parameter estimates was observed between the two different efficacy endpoints, i.e., AUSTC and kidney burden (Table 4 and Fig. 3; 8-h interval).

No modeling was performed with C_max, as (i) C_max was not selected as a satisfying PK-PD predictor for the PK-PD analysis of AUSTC (see above) and (ii) the kidney burden and AUSTC endpoints are consistent regarding the dose relationship (Fig. 3, similar dose providing 50% of net effect).

DISCUSSION

Other authors have already published some correlations between in vitro and in vivo antifungal activities in animal models (L. Appenzeller, E. Lim, P. Wong, M. Fadem, P. Motchnik, M. Bakalinsky, and R. Little, Abstr. 36th Intersci. Conf. Antimicrob. Agents Chemother., abstr. F187, 1996; 16, 20, 29). However, much less experience exists with a rational approach for the design of the first clinical trial involving antifungals. One difficulty is the extrapolation of the experimental animal model results to humans, due to PK dissimilarities between humans and laboratory animals (4, 5).

The present study was designed to establish correlations between PK parameters and efficacy. The ultimate aim is to identify a PK parameter (e.g., AUC or C_max) or a PK-PD parameter (e.g., t > MIC or AUC/MIC ratio) which could be used as a common predictor of in vivo antifungal activity. The PK characteristics are different in each species and depend more critically on specific metabolic paths (6). However, the same PK or PK-PD parameter values (based on free concentrations) predictive of i.e. 90% efficacy in an animal species are likely to provide a similar efficacy and thus could be used to more accurately extrapolate results within different species.

In order to identify the predictor of efficacy, it is critical to explore several daily doses fractionated using different dosing intervals. Most of the interdependence among PK-PD parameters can be reduced by comparing the results of dosage regimens that are based on different dosing intervals (8).

The E_max model has been successfully used to describe PK-PD relationships (12, 19, 21, 25) in the antibacterial field. In these reports, the most frequently used efficacy parameter was the difference between the CFU in the absence and that in the presence of the antibacterial compound. In our case, we have also used survival-related parameters such as AUSTC. This approach has been validated because we found a good PK-PD relationship for both survival-related parameters and the presence of C. albicans in kidneys.

Regarding survival measures, AUSTC was selected for PK-PD modeling as being more sensitive than the percentage of survivors at the end of the study or the mean survival day. The relationship between the net effect on AUSTC and the 24-h GM 237354 AUC at steady state could be well described by a sigmoid E_max model independently of the dosing interval. The Hill coefficient γ estimated values were very close: 1.9 and 2.3 for 8- and 4-h intervals, respectively. The AUC (total concentrations) at which 50% of the maximum effect was reached (AUC₅₀) were 21.7 and 34.7 μg · h/ml for the 8- and 4-h intervals, respectively. A poorer fit was obtained for the 12-h dosing interval, as a result of greater variability in the data. The AUC₅₀ estimates for the three dosing intervals were reasonably close (approximately twofold range); however, they did not provide exactly the same values. The reason for this is not known, though this may be due to experimental variability. A sigmoid E_max model pooling data from the three dosing intervals could be used as an approximation (a simple sigmoid E_max model could not provide a good representation of the data), and γ was estimated at 1.5 and AUC₅₀ was estimated at 29.3 μg · h/ml.

This analysis can be the first step to test prospectively more-targeted doses in larger species (i.e., humans) because, at least theoretically, the AUC value (in terms of free concentrations to account for any differences in binding between species) producing favorable outcomes (i.e., 90% of the maximal effect) in any species can be predicted using the above model.

Another PK-PD parameter, t > MIC (calculated in terms of free serum concentrations), showed a very steep relationship with efficacy. Considering for example the 8-h dosing interval, for 50 and 90% efficacy in terms of AUSTC, t > MIC was 77 and 89%, respectively. This implies that, when near-complete efficacy was observed, unbound concentrations were essentially above the MIC over the whole dosing interval. However, t > MIC was 100% in the 4-h dosing interval group though good efficacy was not observed at low doses. This demonstrated that t > MIC is not a PK-PD predictor of efficacy.

Regarding tissue burden, the AUC₅₀ value needed to diminish the C. albicans burden in its target organ (kidney in this infection model) was found to be at 37.1 μg · h/ml and the Hill coefficient γ was found to be at 1.5 following repeated dosing every 8 h (no data for the other dosing intervals). These values are similar to the estimates for the net effect on AUSTC (AUC₅₀ = 21.7 μg · h/ml; γ = 1.9) for the same dosing interval.

We can conclude the following. (i) A good agreement was found between the PK and therapeutic efficacy of GM 237354 at different dosing regimens using an experimental systemic C. albicans infection model in mice. The 7-day kidney colonization represents infection of the target organ, and a good consistency was found between Candida burden and mortality (survival time and percentage of survivors). (ii) PK-PD relationships between efficacy measures (survival time curve and kidney burden) and C_max or 24-h AUC at steady state were evaluated using PK parameters obtained with total plasma concentrations. PK-PD relationships using t > MIC at steady state were assessed using unbound serum concentrations. (iii) The effect was well predicted independently of the dosing interval only by AUC. The various efficacy endpoints used for modeling (net effect on survival time curve, percentage of survivors, and kidney burden) provided similar PK-PD trends. (iv) For 50% efficacy, AUC₅₀ was estimated at 21.7 and 37.1 μg · h/ml for the survival time curve and the reduction in kidney burden (8-h dosing interval data), respectively. These values corresponded to a daily dose of ∼60 mg/kg in mice. (v) At near-maximal efficacy for the effect on survival, t > MIC₉₀ was close to 90% in the 8- and 12-h dosing interval groups. However, t > MIC is not predictive of efficacy: in the 4-h dosing interval group, efficacy was dose dependent, although t > MIC was 100% at all doses. (vi) PK-PD relationships have to be validated using new FPS inhibitor compounds and Candida spp. with different susceptibility patterns. However, these relationships could be already useful for the more accurate design of studies involving large animal species and humans, after correction for any plasma protein binding between species. Future work destined to increase knowledge in this field is warranted.