Identification of Circadian Determinants of Cancer Chronotherapy through In Vitro Chronopharmacology and Mathematical Modeling

Most biologic functions in experimental rodents and humans are rhythmically moderated by the Circadian Timing System (CTS) over 24 hours (1). The CTS is constituted of a network of genetic cellular circadian clocks which are coordinated by the suprachiasmatic nuclei, a hypothalamic pacemaker, through the generation of rhythmic physiology (2). The CTS further controls drug metabolism and transport, as well as cell-cycle progression, DNA repair, and apoptosis in experimental rodents and in humans. As a result, drug pharmacokinetics (PK) and/or pharmacodynamics (PD) can vary as a function of dosing time in whole mammalian organisms, as shown for several hundreds of medications (3).

Circadian timing is especially relevant for anticancer drugs whose optimal dose and delivery schedule are most critical for safely achieving best antitumor efficacy (2). Indeed, up-to-several-fold changes in treatment tolerability and/or efficacy were found according to dosing time for 40 anticancer drugs in rodents (2). Moreover, the administration of anticancer agents at the circadian time when they were the safest also achieved best efficacy both in rodents (4) and in cancer patients (5–8). Large intersubject differences in CTS further need to be taken into account for the personalization of circadian delivery of cancer treatments (9, 10). Thus, a systematic mapping of the critical pathways of anticancer drug chronopharmacology and their molecular clock control is required for optimizing treatment effects through tailoring circadian drug delivery to individual CTS.

Drug chronopharmacology is modulated at the cellular level by molecular clocks which are constituted by interconnected transcription/translation loops involving 15 clock genes, where PER2, REV-ERBα, and BMAL1 play major roles (3). These genes display circadian rhythms in their expression and in turn generate oscillations in several membrane transporters and enzymes involved in drug metabolism, cell cycle, DNA repair, or apoptosis (2, 11, 12). Thus, the CTS could rhythmically control nearly half of the liver metabolome (13), including many DNA synthesis (14) and detoxification pathways (11).

Here, we use a novel systems chronopharmacology approach that combines in vitro and in silico models, in order to characterize the chronopharmacology of the anticancer drug irinotecan in human colon carcinoma cells. The drug is taken as a model of an anticancer drug with a complex metabolism (Fig. 1; ref. 15). It is indicated against colorectal cancer, despite severe and dose-limiting intestinal and hematologic toxicities (16). Irinotecan efficacy and toxicity displayed circadian rhythms in groups of mice and patients (2, 17). Several genes and proteins involved in irinotecan PK and/or PD display circadian rhythms in mouse liver and/or intestine, including the drug target Top1 (18), the bioactivation enzymes Ces1 and Ces2 (19), the detoxification enzyme Ugt1a1 (19), and the efflux transporters Abcb1a, Abcb1b, and Abcc2 (17, 18, 20–22). Studies in 8 mouse categories, based on differences in sex and genetic background, revealed large differences in irinotecan chronopharmacology. Statistical modeling of 27 circadian gene expression data in liver and colon highlighted the robust prediction of optimal irinotecan timing with a mathematical model of the Rev-erbα and Bmal1 transcription regulatory loop (23).

In contrast with chronopharmacology studies in whole organisms, investigations in circadian synchronized cell culture models could enable systematic and quantitative information to be gathered, so as to precisely determine and model molecular chronopharmacology (24). Here, we conduct such an in vitro–in silico circadian investigation of irinotecan PK–PD in synchronized colorectal cancer cells for the first time. Our aim is the identification of the most critical clock-controlled determinants of drug cytotoxicity. Such knowledge would help optimizing circadian drug delivery patterns according to the host and tumor clocks (25, 26).

The human colon tumor cell line Caco-2 was obtained from the ATCC in 2007. Cells were grown in DMEM:Ham F12 medium [DMEM:F12 (1:1)] supplemented with an antibiotic cocktail (penicillin 1000 U/L, streptomycin 100 μg/mL), 2 mmol/L glutamine (Fischer Scientific) and FCS (Dutscher; 10%). They were maintained in a humidified atmosphere containing 5% CO₂ at 37°C. For experiments, cells were seeded on Petri dishes at 2,500 cells/cm². Two days after confluence, cells were transfected with 10 nmol/L of siRNA of BMAL1 (NM_001030272, mix of S1616 and S1618 siRNA sequence from Ambion) or of siRNA control using Lipofectamine RNAiMAX transfection reagent (Invitrogen) as recommended by supplier. Six hours after transfection, cells were replaced in normal culture medium for 2 days. Next, circadian synchronization of cells was performed by a serum shock that consisted in a 2-hour exposure to serum-rich medium (DMEM:F12 containing 50% FBS). The beginning of the serum shock defined Time 0 (T0). Then cells were replaced in normal culture medium.

At indicated circadian times (Ts), siRNA-treated synchronized cells were scrapped in guanosine isothiocyanate and frozen at −80°C until RNA extraction. Total RNA was then extracted as described in (27). Reverse transcription was realized with SuperscriptII RT- (Invitrogen). Quantitative PCR was performed with LightCycler 480 using LightCycler 480 SYBR Green I master kit (Roche). Primers used for gene amplification were previously described (24, 28). Hybridization temperature for all the primers was 60°C. Relative quantification of target RNA using 36B4 as reference were realized with Relquant software (Roche).

Circadian gene expressions in synchronized Caco-2 cells were modeled as a damped cosine (24):

The common period T and gene parameters λ, M, A, and ϕ were simultaneously estimated for all genes, under control or BMAL1 siRNA conditions, using a previously described bootstrap approach (ref. 24; Supplementary Table S1). We removed the first 2 data points at 0 and 4 hours from the parameter estimation procedure as purely circadian variations may be perturbed by the seric shock during the first 4 hours. Indeed, this allowed us to achieve a better fit to equation (1) compared with taking into account all data points.

Irinotecan was purchased from Pfizer. For chronoPK experiments, Caco-2 cells were first exposed to control or BMAL1 siRNA before being synchronized. Cells were then exposed to 75 μmol/L of CPT11 for 6 hours starting at T2, 14, and 20. Irinotecan was added to the culture medium 6 hours before incubation with cells in order to reach the lactone–carboxylate equilibrium. CPT11 and SN38 extra- and intracellular concentrations were measured by the method of high-performance liquid chromatography (HPLC) described (24).

For circadian assessment of TOP1 complexes, siRNA-treated synchronized cells (1 × 10⁶ to 10 × 10⁶) were exposed to SN38 (0.1 μmol/L) at 37°C during 30 minutes at indicated CTS. DNA–topoisomerase I complexes were then measured by a slot-blot method adapted from Subramanian and colleagues (26) using the Topo I Link Kit (TopoGen) as previously described (24).

Cell viability and apoptosis measurements were performed using respectively the CellTiter-Glo Kit and caspase 3/7-Glo Kit (Promega). Cells were seeded in 96-well plates treated with control and BMAL1 siRNA and synchronized and treated 3 days after confluence with CPT11 (85 μg/mL) during 2 hours starting at T2, 14, 20, and 28. Twenty-four hours after treatment, reaction buffer was added in each well and bioluminescence was measured using Luminometer (Berthold Technologies). Apoptosis percentages compared with untreated cells were then normalized to cell viability percentages.

Statistically significant differences according to circadian time and experimental condition were tested with multiple-way ANOVA, including Scheffe contrast test, using the SPSS software (v.16.0) for Windows software. Circadian rhythmicity of mRNA levels was validated by statistical tests of null amplitude (See Supplementary Information). The Pearson χ² test was used to measure the goodness of the fit of the irinotecan PK–PD model.

CPT11 molecular PK–PD was represented by a mathematical model on the basis of ordinary differential equations that were solved using the ode15s MATLAB function. The final model is an extended version of a previously published one (24). The equations describing the dynamics of CPT11, SN38, and SN38G extra- and intracellular concentrations together with those of DNA/TOP1, reversible and irreversible SN38/DNA/TOP1 complexes were kept unchanged. To account for the fact that irreversible DNA damage may trigger the apoptotic machinery, we supplemented the existing model with the following equation that phenomenologically computes the percentage of apoptotic cells Apop:

The state variable I_compl represents the intracellular concentration of irreversible SN38/DNA/TOP1 complexes and the parameter k_apop accounts for DNA repair mechanisms and apoptosis pathways which both display daily variations (29). Therefore, we allow k_apop to have circadian rhythms:

Circadian variations are also assumed for the enzymatic activity of irinotecan bioactivation (CES), SN38 detoxification (UGT1As), and of the cellular efflux of irinotecan (ABC_CPT) and SN38 (ABC_SN) in agreement with experimental results of Fig. 1 and with literature (23, 24, 30). Protein activities were modeled by a cosine function of common period T as determined in the mRNA studies:

Parameters M^activity, A^activity, and ϕ^activity were computed for each protein activity in the parameter estimation procedure. No circadian variations were considered for TOP1 as suggested by constant nucleic protein amount (Supplementary Fig. S1).

In total, 21 PK parameters and 12 circadian parameters had to be estimated. TOP1 total protein concentration was set to 70 nmol/L as observed in mouse cells (31). Model parameters were estimated to optimally reproduce the chronoPK results (extra- and intracellular CPT11 and SN38 concentrations, Fig. 2), the percentage of DNA-bound Top1 in presence of SN38 (Fig. 3) and CPT11-induced apoptosis (Fig. 4), in the presence of control or BMAL1 siRNA treatment, for the 3 studied Ts of exposure (See Supplementary Tables S2 and S3 for final parameter values). Initial search values were set to parameter estimates of the previous version of the model (24). The parameter estimation consisted in a bootstrap approach based on a weighted least-square method, as previously described (24).

We performed a Sobol variance–based global sensitivity analysis on the model as follows. First, parameters' lower and upper bounds were set to 100-fold lesser and greater than their estimated values and 50,000 parameter sets were generated from cross-sampling by Saltelli's extension of Sobol's method using the MOEA framework (version 2.0, http://www.moeaframework.org/ ref. 32). For each parameter set, the considered model output was computed using MATLAB. Finally, the Sobol analysis function of the MOEA framework was used to compute parameters' total-order sensitivity indices—which represent the contribution to the output variance of the studied parameter including all variances caused by its interactions with other parameters—together with their confidence intervals. Only parameters with a sensitivity of >5% are shown.

We studied the sensitivity of all model parameters—except K_m's of enzymatic reactions—together with that of irinotecan initial concentration and exposure duration (T_expo). Irinotecan cytotoxicity was assessed through DNA damage—defined as the sum of reversible and irreversible SN38/DNA/TOP1 complexes formed during irinotecan exposure. Drug-induced apoptosis was not used as an endpoint in sensitivity analysis because it was not mechanistically modeled.

The circadian mRNA expression patterns of 3 core clock genes (PER2, REV-ERBα, BMAL1) and 4 critical activation, transport, detoxification, and target genes for irinotecan metabolism (CES2, ABCB1, UGT1A1, TOP1) were first determined in synchronized human colon cancer Caco-2 cells in the presence of control or BMAL1 siRNA. In cells treated with control siRNA, the 3 clock genes and the 4 pharmacologic genes displayed circadian rhythms in their mRNA expression with amplitude values A ranging from 40% to 80% of mean values M (Figs. 2 and Supplementary Table S1). The common period was estimated to τ = 28 h 06 min (SD, 1 h 41 min). Differences between phases of BMAL1 and PER2 and between BMAL1 and REV-ERBα were respectively equal to 13 h 33 min (=0.94 π rad) and 15 h 41 min (=1.1 π rad). Thus, the circadian expression pattern of BMAL1 was in antiphase to that of REV-ERBα and PER2 in this colon cancer cell line. This finding was consistent with the reciprocal regulation of these 3 clock genes within the molecular circadian clock in healthy cells or tissues (12). We further noticed a mild dampening of the mRNA oscillations in the range of λ ≈ 10⁻² per hour, which may account for a slight desynchronization of cells over time. Times of maximal mRNA expressions (acrophases) were located at 11 h 28 min for CES2, at 14 h 26 min for UGT1A1, at 16 h 51 min for ABCB1, and at 14 h 30 min for TOP1 modulo the period of 28 h 06 (Supplementary Table S1).

As expected, the exposure of Caco-2 cells to BMAL1 siRNA resulted in a 56% decrease in BMAL1 mean mRNA expression as compared with controls (Fig. 2). si-BMAL1 exposure completely turned off the circadian oscillations of the mRNA expression of the 3 clock genes and those of the 4 metabolism genes in Caco-2 cell populations (Supplementary Table S1). BMAL1 silencing also resulted in a 27% to 48% decrease in the mean mRNA level of the other 6 genes of interest—as compared with their respective controls—and reduced mRNA expressions by 48% to 72% at the time of their respective peak expressions (Supplementary Table S1).

The above mRNA circadian expression data in clock-proficient Caco-2 cells were in line with those obtained earlier (24). They led to model and predict that irinotecan toxicity would be least following exposure onset at T2 and highest after treatment initiation at T14 or T20 (24). These 3 circadian timings were used here for testing the relevance of the molecular clock for irinotecan metabolism and cellular PK following irinotecan exposure (75 μg/mL) over 6 hours. The extracellular concentration pattern of irinotecan displayed minor yet statistically significant differences according to treatment timing in the clock-proficient Caco-2 cells (Fig. 3). Thus, the mean area under the curve over 6 h (AUC_{0–6 h}) of extracellular irinotecan concentration varied according to treatment time (ANOVA, P = 0.001). Mean AUC_{0–6 h} (±SEM) was lowest following treatment at T2 (392.3 ± 2.3 μmol/L·h), intermediate after treatment at T20 (433.0 ± 5.6 μmol/L ·h), and highest after drug exposure at T14 (458.1 ± 5.6 μmol/L·h). However, irinotecan timing did not moderate intracellular irinotecan concentration dynamics or AUC_{0–6 h} (Fig. 3C and D; ANOVA, P = 0.117). In contrast, irinotecan timing profoundly modified both intracellular and extracellular concentration dynamics of SN38, the bioactive metabolite of irinotecan. These differences translated into large and statistically significant changes in both intra- and extracellular AUC_{0–6 h} of SN38 according to the T of drug exposure (P < 0.0001). Mean AUC_{0–6 h} values of intracellular SN38 (± SEM) ranged from 265.13 ± 4.5 nmol/L·h for cells exposed at T2, to 489.7 ± 2.59 nmol/L·h for those exposed at T20, and 341 ± 3.3 nmol/L·h following treatment at T14. Corresponding values for extracellular SN38 AUC_{0–6 h} were 0.15 ± 0.004, 0.3 ± 0.003, and 0.26 ± 0.001 nmol/L ·h, respectively. These findings highlighted the key role of the molecular clock for irinotecan and SN38 transport and metabolism.

In sharp contrast, the disruption of the circadian clock with BMAL1 siRNA completely erased the large dosing time dependencies found in clock-proficient cells (Fig. 3). Indeed, no dosing time dependency was apparent for the extracellular and intracellular concentrations of irinotecan and SN38, as well as for their respective AUC_{0–6 h} (ANOVA; P = 0.05).

Irinotecan dosing at T2 resulted in similar values for intracellular or extracellular SN38 AUC_{0–6 h} irrespective of clock silencing, with corresponding means close to 250 and 0.07 nmol/L·h, respectively. However, clock silencing markedly decreased irinotecan bioactivation into SN38 after drug exposure at T14 or T20 down to its level found at T2 in clock-proficient cells (Fig. 3E–H).

We then questioned the relevance of the molecular circadian clock for SN38 molecular PK. The amount of DNA-bound TOP1 protein varied according to SN38 timing with statistical significance in clock-proficient Caco-2 cells (ANOVA, P < 0.0001; Fig. 4A). Indeed, mean DNA-bound TOP1 amount nearly doubled between treatment at T2 and T14 (P = 0.011) or T20 (P = 0.02), whereas differences between T14 and T20 were minor and not statistically significant (ANOVA, P = 0.86). In contrast, DNA-bound TOP1 protein amount did not differ according to SN38 timing in clock-defective cells, with amounts similar to those found in clock-proficient cells treated at T2.

The implications of the circadian clock control of molecular chronoPK and chronoPD for drug toxicity was further determined, through the quantification of caspase-3 activation as a marker of drug-induced apoptosis. Clock-proficient cells exposed to control siRNA displayed statistically significant time-dependent apoptosis (ANOVA, P < 0.0001; Fig. 4B). Apoptosis induction was increased by 14%, 52%, and 63% following irinotecan treatment at T2, T14, and T20, respectively, as compared with untreated controls. We thus observed a 4.5-fold difference in irinotecan-induced apoptosis as a function of whether irinotecan treatment was initiated at T2 or T20.

Modest time-dependent changes were found according to irinotecan timing in clock-defective cells. No statistically significant difference characterized treatment initiated at T2 or T14 (P = 0.38), yet apoptosis induction was largest at T20 as compared with T14 (P = 0.002) or T2 (P = 0.025). Clock silencing with BMAL1 siRNA dramatically decreased irinotecan-induced apoptosis by 72% in cells treated at T14 and by 65% in cells treated at T20. In contrast, apoptosis increased by 25% in the siBMAL1 Caco-2 cells treated at T2 as compared with control conditions.

A comprehensive mathematical analysis of all the above experimental results was implemented. We first fitted the mathematical model of irinotecan molecular chronoPK–PD to the current multidimensional datasets (see Materials and Methods and Supplementary Data). In the presence of BMAL1 siRNA treatment, all circadian rhythms were assumed to be disrupted—that is all circadian amplitudes were set to zero—and mean protein activities were allowed to be different from control conditions as suggested by differences in the mean mRNA levels of metabolism genes.

The calibrated mathematical model displayed good qualitative and quantitative agreement with all datasets, thus theoretically confirming the circadian disruption through BMAL1 siRNA (Figs. 3 and 4). The Pearson χ² test validated, with a probability P > 0.975, that the irinotecan PK–PD model was correctly representing the 3 datasets of irinotecan PK, TOP1 activity, and drug-induced apoptosis measured at T2, T14, and T20 in both control and siBMAL1-exposed cells. The model predicted circadian activities for the proteins involved in irinotecan bioactivation (CES), SN38 detoxification (UGT1As), as well as for those responsible for irinotecan and SN38 efflux (ABC_CPT, ABC_SN) in control conditions. Predicted amplitude values (A^activity) ranged from 12.7% to 80% of the corresponding mean values (M^activity) and predicted acrophases (ϕ^activity) were shifted by 2 h 08 to 20 h 30 as compared with those of the corresponding mRNA expression rhythms (Fig. 5A and Supplementary Table S2). Furthermore, the fitting procedure predicted a circadian rhythm in k_apop, the parameter that links drug-induced DNA damage to apoptosis in the model. This finding suggested an important impact of the circadian control of DNA repair and apoptosis genes and proteins for irinotecan cytotoxicity. The best-fit model predicted highest cytotoxicity following drug exposure onset at 17 h 48 min and least if drug exposure timing started at 5 h 24 min modulo the period of 28 h 06 min (Fig. 4C).

In cells exposed to BMAL1 siRNA, mean activities of irinotecan efflux transporters and UGT1As enzymes were unchanged compared with clock-proficient cells. In contrast, SN38 efflux transporters and irinotecan bioactivation activities were decreased by 17% and 37%, respectively. The k_apop model parameter value was increased by 2.85-fold compared with control, a result supporting a greater susceptibility of clock-deficient cells to irinotecan-induced DNA damage (Supplementary Table S2). The comprehensive mathematical model was consistent with all experimental data except the time-dependent changes in irinotecan extracellular concentrations in the clock-proficient cells, which were not well reproduced by the best-fit simulations. An additional in vitro chronoPK experiment was thus performed (See Supplementary Fig. S2). Overall, interstudy variability was considered as accounting for the minor discrepancy in irinotecan extracellular dynamics between both experiments.

The influence of model parameters on irinotecan toxicity pattern was assessed through global sensitivity analyses using DNA damage as the main toxicity endpoint (see Materials and Methods). Parameter sensitivity analyses were performed for (i) the circadian timing of irinotecan exposure onset associated to minimal DNA damage and (ii) the circadian amplitude of irinotecan-induced DNA damage, defined as the difference between the mean and the minimum values of the rhythm.

The circadian timing corresponding to least toxicity was mostly determined by the mean value and the circadian amplitude of both SN38 detoxification (UGT1A) and irinotecan bioactivation (CES), whose respective circadian acrophases ranked as first and second most sensitive parameters (Figs. 5B and 6). Irinotecan exposure duration ranked third, and the detoxification and bioactivation circadian amplitudes ranked fourth and fifth, respectively.

The amplitude of the toxicity rhythm was largely determined by the 3 circadian parameters of bioactivation, whose amplitude, mean value, and acrophase ranked as first, second, and fifth most sensitive parameters, respectively (Figs. 5C and 6). DNA complex formation and dissociation parameters also contributed to drug toxicity amplitude as k_f1 (DNA/TOP1 complex formation), k_f2 (SN38/DNA/TOP1 complex formation), k_d1 (DNA/TOP1 complex dissociation), k_DSB (irreversible complex formation), and k_d2 (SN38/DNA/TOP1 complex dissociation), respectively, ranked as third, sixth, seventh, eighth, and ninth parameters.

Improving our understanding of the circadian rhythms that govern anticancer drug toxicities and efficacy requires identifying the critical molecular determinants of the circadian control of drug metabolism. To this end, we here investigated the in vitro chronopharmacology of irinotecan as a first proof of concept of chronopharmacology in synchronized cancer cell populations. This novel in vitro–in silico approach to developmental chronotherapeutics should be viewed as aiming to help dissect the relative contribution of the cellular molecular clock and that of systemic factors for the chronopharmacology of irinotecan. Here, circadian rhythms in irinotecan pharmacology and toxicity were found for each studied parameter in synchronized Caco-2 cell cultures and were all ablated following exposure to BMAL1 siRNA. Thus, irinotecan chronopharmacology resulted from the control of its metabolism by the molecular circadian clock. Moreover, irinotecan cytotoxicity appeared to be positively correlated to the expression level of clock gene BMAL1, both in clock-proficient and in clock-deficient cells. In contrast, the cytotoxicity of oxaliplatin, another effective anticancer drug against colorectal cancer, reportedly displayed a negative correlation with BMAL1 expression level (33). Taken together, the findings suggest that BMAL1 tumor expression could help select the most effective drug for a given patient.

The physiologically based model of irinotecan chronoPK–PD closely agreed with experimental results. Model parameter sensitivity analyses highlighted the overall predominant influence of bioactivating CES and detoxifying UGT1As circadian rhythms on irinotecan chronotoxicity pattern. The mRNA expression circadian rhythms were in good agreement with previously published data (24). The circadian acrophases of BMAL1, TOP1, UGT1A1. and ABCB1 were indeed similar to those earlier found in a separate study, whereas those of PER2, REV-ERBα and CES2 differed by less than |$\frac{\pi}{2}$| rad (Supplementary Fig. S3). The mathematical model of irinotecan chronoPK–PD, which was initially designed for unsynchronized Caco-2 cells, successfully fitted the current datasets in synchronized cells, a step that further validated the model structure. New parameter estimates were obtained here, which highlighted kinetics differences between synchronized and unsynchronized cells, thus providing an opportunity for differentiating chronopharmacology in healthy and malignant tissues (Supplementary Table S3).

An interesting added value of the model involved its ability to accurately predict the circadian patterns in the main metabolism proteins, based upon gene and pharmacology data. The estimated protein acrophases were shifted by 1 h 24 min to 20 h 30 min relative to those found for the corresponding genes. Intervals ranging from 0 to 20 hours (modulo, 24 hours) were previously reported between several mRNA and protein expressions in mouse tissues (30). The model predictions based on synchronized Caco-2 data were rather consistent with mouse liver data, following the normalization of the circadian period to 24 hours for the Caco-2 cells. Thus, the model predicted a time lag of 14 hours for UGT1As mRNA and protein expression rhythms, whereas it was equal to 15 hours for Ugt1a1 in mouse liver (30). Similarly, the model predicted a 22-hour time lag between CES mRNA and protein, which differed from mouse liver Ces2 by only 4 hours (30).

The slight shift between the circadian rhythm in irinotecan-induced DNA-bound TOP1 and that in extent of apoptosis suggested the occurrence of an additional circadian control of DNA repair and apoptosis processes. The best-fit mathematical model indeed included a non-zero circadian amplitude for the parameter k_apop, which represents the DNA damage response phenotype, including the P53 network, DNA repair and, eventually, apoptosis. Besides its regulatory effect on apoptosis, P53 also regulates the molecular clock so that P53 mutation could modify the circadian rhythms of the protein activities involved in irinotecan chronopharmacology and therefore alter the circadian pattern in irinotecan cytotoxicity (29, 34).

Altogether, the current model-driven chronopharmacology study has established the molecular bases of irinotecan chronopharmacology, through setting up a comprehensive mechanistic circadian PK–PD model for this drug. Several levels of prediction have been confirmed, regarding the circadian control of gene transcription and proteins, based on a recent literature survey. Moreover, the temporal relation between low BMAL1 expression and low irinotecan toxicity found in vitro was in good agreement with experimental studies in mice regarding the circadian toxicity pattern of this drug as well (23). The results advocate thus for the model validity and accuracy, which now deserves further prospective adjustments in selected experimental models, before clinical testing of personalized chronopharmacology delivery in cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: S. Dulong, A. Ballesta, A. Okyar, F. Lévi

Development of methodology: S. Dulong, A. Ballesta, F. Lévi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Dulong, F. Lévi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Dulong, A. Ballesta, A. Okyar, F. Lévi

Writing, review, and/or revision of the manuscript: S. Dulong, A. Ballesta, A. Okyar, F. Lévi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Dulong, A. Ballesta

Study supervision: S. Dulong, A. Ballesta, F. Lévi

The authors thank Dr. Jean Clairambault for his role in the initiation of this project and subsequent fruitful discussions.

This work was co-funded by the ARC (CR109/8003) to F. Lévi, the European Commission through the specific targeted research project TEMPO (LSHG-CT-2006-037543) to all authors, ERASsysBio+ (C5Sys, LHSB-2010-005137) to F. Lévi and S. Dulong, CaSyM (HEALTH-F4-2012-305033) to A. Ballesta and F. Lévi, and the research fund of Istanbul University (N-7762/2010) to A. Okyar.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.