Personalized chronotherapy in glioblastoma: integrating circadian profiling and PK–PD modelling to optimize temozolomide timing

All three cell lines exhibited robust circadian oscillations at the promoter level. Notably, each showed a significant phase difference between BMAL1 and PER2 activity: 0.4 rad in U87-MG (p = 0.0004), 1.5 rad in T98G (p = 0.0028), and 0.7 rad in U251-MG (p = 0.0005). U87-MG and U251-MG did not differ significantly in circadian period for either promoter (BMAL1: p = 0.21; PER2: p = 0.18). U87-MG displayed an extended period (26.9 ± 0.9 h), exceeding the canonical 24-hour cycle, while U251-MG oscillated within the normal range (24.0 ± 0.8 h). T98G cells exhibited discordant periods between promoters: BMAL1 oscillated normally (24.8 ± 1.9 h), but PER2 had a significantly shortened period of 19.9 ± 2.1 h (p = 0.0023).

Amplitude and phase comparisons further highlighted cell line-specific differences. At the BMAL1 promoter level, T98G showed significantly lower amplitude than U87-MG (p = 0.0053) and U251-MG (p = 0.030). U87-MG had a significantly longer period than T98G (p = 0.012) and U251-MG (p = 0.0067). Phase was significantly delayed in U251-MG compared to T98G (p = 0.0019) and U87-MG (p < 0.0001), while U87-MG and T98G had comparable phases (p = 0.79). At the PER2 promoter level, amplitudes did not differ significantly, but both period and phase were significantly different across all three cell lines (period: all comparisons p < 0.01; phase: all comparisons p < 0.001).

To extend these observations to the transcript level, we measured endogenous expression of six core-clock genes (BMAL1, PER2, NR1D1, NR1D2, CRY1, and CRY2) by qRT-PCR over a 54-hour period in 3-hour intervals (Fig. 1b). Expression was normalized to GAPDH and mean expression. The best-fit period used for Cosinor analysis was 24 h in T98G, 35 h in U87-MG, and 20 h in U251-MG.

Cosinor analysis revealed significant differences in both phase and amplitude between cell lines. T98G displayed the most robust rhythmicity, with all six core-clock genes showing significant 24-hour oscillations (p < 0.05) and stable amplitude across the time course. In contrast, U87-MG and U251-MG lacked rhythmicity in some genes and showed either lengthened (U87-MG) or shortened (U251-MG) periods, consistent with promoter-level observations. These disrupted rhythms likely reflect desynchronization due to imbalances in the clock network, which may underlie the observed rapid damping of oscillations in these two cell lines.

Interestingly, while BMAL1 and PER2 phase relationships were largely preserved between promoter activity and mRNA expression, discrepancies in rhythm robustness suggest post-transcriptional regulation plays a role in shaping the observed RNA profiles. In U87-MG and U251-MG, but less so in T98G, oscillations appeared less stable at the mRNA level than at the promoter level, possibly due to cell line-specific differences in mRNA processing, stability, or feedback regulation.

These findings underscore substantial variability in circadian clock function across GBM cell lines, supporting the hypothesis that inter-patient circadian differences may influence response to chronotherapy.

Having established distinct circadian profiles across our GBM cell line models, we next investigated whether these differences influence responses to chronotherapy. We focused primarily on TMZ, which has been reported to exhibit time-dependent efficacy in both preclinical and clinical GBM studies^8–14. TMZ is a well-established component of standard GBM therapy and has a short half-life (~1.8 h), making it amenable to precise timing of administration²².

As BMAL1 expression levels have previously been correlated with TMZ sensitivity, we selected treatment timepoints based on the circadian phase of BMAL1 expression: 0 h (trough), 5 h (ascending), 10 h (peak), 15 h (descending), and 20 h (trough) (Fig. 2b). Cells were treated at these timepoints with TMZ, KL001, or SR9011 alone or in combination, at their respective IC₅₀ concentrations (Supplementary Fig. 1). Proliferation was monitored in real time (Fig. 2c, d; Supplementary Figs. 2–4), and treatment efficacy was quantified as the normalized area under the curve (AUC) relative to DMSO control. The treatment effect was defined as 1 - AUC, where positive values indicate growth inhibition and negative values indicate enhanced proliferation (Supplementary Table 1). Mean treatment effects across all timepoints were also calculated for each condition (Fig. 2e).

In T98G cells, TMZ-based treatments were clearly more effective than those with clock modulators alone. KL001 exhibited a consistent negative treatment effect across all timepoints, while SR9011 alone showed limited efficacy. Interestingly, the combination of KL001 + SR9011 yielded a comparable effect to TMZ alone (p = 0.62) and was significantly more effective at 20 h (p = 0.0395). The most effective treatments were combinations of TMZ with one or both clock modulators. TMZ + SR9011 and TMZ + KL001 + SR9011 significantly outperformed TMZ monotherapy independent of treatment timing. These findings support the potential of clock-targeting agents to enhance TMZ efficacy in T98G cells (Fig. 2c–e; Supplementary Fig. 2, Supplementary Table 1).

U87-MG cells exhibited a different profile. KL001 consistently induced negative treatment effects, and in combination with TMZ, appeared to antagonize its efficacy. In contrast, SR9011 showed the highest overall efficacy among monotherapies, significantly outperforming TMZ alone (p = 0.01).

Notably, the efficacy of SR9011 and its combinations was highly time-dependent. The most effective treatments included SR9011 at 0 h and TMZ + SR9011 at 10 h, 15 h, and 20 h. These results highlight that both the choice of agent and treatment timing are critical for maximizing efficacy in U87-MG cells (Fig. 2c–e; Supplementary Fig. 3, Supplementary Table 1).

U251-MG cells demonstrated a strong time-of-treatment effect, with the poorest responses generally occurring at 15 h. In contrast, 20 h emerged as the optimal treatment time (p = 0.005 vs. 15 h). The most effective single agent was TMZ at 20 h, followed by KL001 at 15 h, which was surprising given its poor performance in the other two cell lines.

Indeed, KL001 exhibited a notably different response in U251-MG cells, with a positive mean treatment effect, suggesting a potentially unique CRY-dependent regulatory mechanism in this line. This differential response may reflect underlying differences in the CRY network, given KL001’s activation of CRY^23,25 (Fig. 2c–e; Supplementary Fig. 4, Supplementary Table 1).

Furthermore, T98G cells displayed an overall lower amplitude in treatment effect—particularly for the clock modulators—which may correlate with their lower circadian amplitude observed at the promoter level (Fig. 2e). This suggests that intrinsic clock robustness may influence responsiveness to clock-targeting chronotherapeutics.

Together, these results demonstrate that the circadian clock modulates the efficacy of TMZ and clock-targeting agents in a time- and cell line-dependent manner. Most treatments showed significant time-of-day variation, and pharmacological manipulation of the CCN altered treatment responses. Importantly, each GBM cell line, representing a distinct circadian phenotype, responded differently to both the timing and type of treatment. These findings underscore the importance of personalized chronotherapy strategies based on individual circadian profiles in GBM.

We found several significant correlations between KL001 sensitivity and components of the negative limb of the circadian clock. In T98G cells, KL001 treatment effect negatively correlated with NR1D1 expression (p = 0.009, r = –0.96). In U87-MG, KL001 effect showed strong negative correlations with both PER2 (p = 0.022, r = –0.93) and CRY2 (p = 0.026, r = –0.922). These results suggest that KL001 efficacy may be modulated by the expression of genes involved in circadian repression.

Previous studies have reported a positive association between BMAL1 expression and sensitivity to TMZ^8–10. While we did not observe a statistically significant correlation between BMAL1 and TMZ response in our data, we noted a moderately strong positive correlation in T98G cells (p = 0.116, r = 0.780), in line with earlier findings (Supplementary Fig. 5).

No further significant correlations were found regarding the other drugs used in this study (Supplementary Figs.–, Supplementary Table). 5 7 2

Many treatments led to pronounced alterations in circadian rhythms, including a complete or partial loss of oscillatory behavior. We quantified the remaining circadian strength by calculating the amplitude relative to DMSO-treated controls. For partially arrhythmic conditions, amplitude was estimated from the mean-to-maximum luminescence values.

In T98G cells, we observed a significant negative correlation between PER2 amplitude and mean treatment effect (p = 0.03, r = –0.802), suggesting that higher treatment efficacy was associated with greater disruption of PER2-driven rhythms. A similar, though not statistically significant, trend was observed for BMAL1 (p = 0.11, r = –0.657). In contrast, no consistent correlations between amplitude and treatment effect were observed in U87-MG or U251-MG cells (Fig. 4c).

It has been reported that Luminescence reduction can be a result of reduced cell viability rather than alterations of biological processes²⁶. To ensure that the correlation we found was not the result of reduced cell viability we assessed cell viability in parallel and did not find any significant correlation to Lumicycle amplitude (Supplementary Fig. 8). This suggests that the changes in amplitude are due to alterations in the circadian clock network rather than viability.

These findings imply that the extent to which treatment disrupts circadian amplitude may serve as a predictive marker for treatment response in certain GBM phenotypes. However, this predictive value appears to be cell line–specific and may depend on intrinsic clock robustness or additional, yet unidentified, genetic or molecular factors.

To further probe the role of the circadian clock in treatment response, we generated CRISPR-Cas9 knockouts (KOs) of three key clock components—BMAL1, PER2, and NR1D1—in the T98G glioblastoma cell line. We selected T98G cells primarily due to its robust circadian rhythmicity, especially on the RNA level, which was not observed in the two other cell lines. We therefore hypothesized that KO of clock components would have more profound effects in a circadian healthy cell line like T98G, and thus a circadian healthy cell line would represent a good choice for the initial model development. The focus on a single cell line for genetic manipulation shows the validity of our approach, still it introduces limitations, and we aim to expand genetic clock manipulation to less circadian robust cell lines in future studies.

Drug response analysis revealed that core-clock gene disruption significantly impacted treatment sensitivity in a time- and drug-dependent manner (Fig. 5c, d; Supplementary Figs. 9–11). Specifically: (1) TMZ Resistance: All three KO lines displayed increased resistance to TMZ, most notably in NR1D1 KO cells, which were resistant across all timepoints (p = 0.004 vs WT). PER2 KO also exhibited significantly reduced sensitivity (p = 0.004 vs WT), except at 15 h (p = 0.41 vs WT). In BMAL1 KO cells, TMZ resistance was only significant at 15 h and 20 h (p = 0.02 for both vs WT). (2) KL001 Sensitivity: While KL001 had a negative or no effect in WT cells, both BMAL1 and PER2 KOs showed enhanced sensitivity across multiple timepoints. BMAL1 KO displayed significant treatment effects at 0, 5, and 10 h (p = 0.02 vs DMSO). PER2 KO cells responded significantly at all timepoints (p = 0.0001 vs DMSO). In contrast, NR1D1 KO abolished the negative treatment effect observed in WT cells at 15 h but did not show a significantly improved mean response (p = 0.90 vs WT). (3) SR9011 Sensitivity: WT T98G cells did not respond to SR9011 alone. In contrast, both BMAL1 and PER2 KO cells exhibited strong and consistent sensitivity across all timepoints (BMAL1 KO: p = 0.0001; PER2 KO: p < 0.0001 vs DMSO). NR1D1 KO cells showed increased sensitivity only at 15 h (p = 0.001 vs DMSO). (4) KL001 + SR9011 Combination: WT cells had a moderate response to the combination. This was markedly enhanced in BMAL1 (p = 0.0003 vs WT) and PER2 KO cells (p < 0.0001 vs WT), but not in NR1D1 KO (p = 0.70 vs WT), where the effect was significant only at 15 h and reduced at 20 h (p = 0.031 vs WT).

To assess reproducibility, we generated PER2 KO in U87-MG cells, which mirrored the T98G results (Supplementary Fig. 12). TMZ resistance was increased in U87-MG PER2 KO, with significant reductions at 15 h (p = 0.004 vs DMSO, p < 0.0001 vs WT) and 20 h (p = 0.006, p = 0.0004). In line with T98G, PER2 KO U87-MG cells showed significantly enhanced sensitivity to SR9011 (p = 0.01 vs WT) and KL001 + SR9011 (p = 0.003 vs WT).

Clock gene knockouts disrupted circadian rhythms under treatment (Fig. 5e), with most conditions resulting in a loss of rhythmicity. Notably, the previously observed correlation between BMAL1 amplitude and mean treatment effect in WT cells was weakened in the KO lines (Fig. 5f).

PER2 KO cells exhibited a near-complete loss of time-dependency in treatment response, except for the TMZ + KL001 and TMZ + SR9011 conditions. In contrast, BMAL1 and NR1D1 KO cells retained time-dependency, though the optimal timepoints shifted. For example, SR9011 became time-independent in BMAL1 KO cells, while KL001 + SR9011 was time-independent in NR1D1 KO cells.

Given the observed correlation between PER2 amplitude and treatment effect in WT cells—but not for BMAL1—these findings suggest that PER2 plays a particularly strong role in mediating chronotherapy responsiveness. Its loss most profoundly altered both treatment sensitivity and timing effects.

Altogether, these results support a strong and consistent relationship between circadian clock integrity and treatment efficacy in GBM cells. Genetic disruption of key clock components alters drug sensitivity, modifies circadian rhythms, and reshapes the temporal dynamics of treatment response. These effects were observed across multiple cell lines, underscoring the potential of circadian phenotyping as a predictive framework for optimizing personalized chronotherapy.

Our approach is the first to implement a detailed, biologically validated model of the core-clock machinery in the form of a coupled ODE system previously published by our group¹⁸. This ODE system is connected to a TMZ efficacy model consisting of three sub-components: (1) a pharmacokinetic (PK) component describing cellular uptake of TMZ, (2) a pharmacodynamic (PD) component describing molecular activation of and DNA alkylation induced by TMZ, and (3) a cell population model describing how alkylated DNA affects cellular proliferation. The efficacy model (i.e., PK-PD-cell population) is linked to the core-clock machinery (i.e., CCN) via the detoxifying enzyme O6-methylguanine DNA methyltransferase (MGMT) reversing TMZ induced DNA alkylation. This MGMT enzyme in turn has been described to be co-expressed with CRY2 and to follow BMAL1²⁷ expression in an anti-phasic manner, offering two connection points to the core-clock machinery.

To address this discrepancy, we performed additional simulations assuming partial, rather than complete, BMAL1 knockdown. Restoring as little as 30% of BMAL1 activity was insufficient to recapitulate observed rhythmicity. At least ~70% residual BMAL1 activity was required to reintroduce time-of-day–dependent fluctuations in treatment response (Fig. 8b), suggesting the presence of compensatory mechanisms maintaining residual circadian function.

With our model we lay the foundation for a tool to generate patient individual TMZ chronotherapeutic recommendations based on gene expression data.

The human glioblastoma cell line T98G, originally derived from a primary tumor of a 61-year-old Caucasian male, and widely used in GBM research was used as an in vitro cellular model. T98G cells harbor deletions in CDKN2A, PART3, TP53, and PTEN, and are IDH wild-type (IDH-WT). For assessment of phenotypic variation in drug response, U87-MG cells, derived from a Caucasian male of unknown age, which exhibit mutations in NF1, PTEN, TP53, and TERT (IDH-WT), and U251-MG cells, derived from a 75-year-old Caucasian male, with mutations in PTEN, TERT, and TP53 (IDH-WT), have been included. The MGMT expression and promotor status was reported as follows: T98G MGMT promotor methylation negative, high MGMT expression⁶⁴; U87-MG MGMT promotor methylation positive, low MGMT expression^64,104; U251-MG MGMT promotor methylation positive, low MGMT expression¹⁰⁵.

All cell lines were authenticated by the Eurofins Cell Line Authentication Service. T98G and U87-MG cells were cultured in Minimum Essential Medium (MEM; Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 2 mM GlutaMAX™, 1 mM sodium pyruvate, 0.1 mM non-essential amino acids (NEAA), 1% penicillin-streptomycin, and 10% fetal calf serum (FCS). U251-MG cells were cultured in RPMI 1640 medium (Gibco) supplemented with 2 mM GlutaMAX™, 1% penicillin-streptomycin, and 10% FCS. CRISPR-Cas9-mediated knockouts of key core-clock genes (BMAL1, NR1D1, and PER2) were generated as previously described¹⁰⁶.

Circadian rhythms in cell lines were characterized using the bioluminescence reporters PER2::LUC-Blasticidine (PLB) and BMAL1::LUC-Puromycin (BLP) encoding for the BMAL1 or PER2 promotor fused to luciferase as described previously^106–108. Briefly, lentiviral particles were produced in HEK-293 cells and glioblastoma cell lines were transduced and selected with the appropriate antibiotic (1 µg/ml puromycin or 10 µg/ml blasticidine) until the control cells died. Cells were seeded in phenol-red-free medium, synchronized by medium change and supplemented with 250 µM D-Luciferin (Bio-Rad laboratories, Hercules, CA, USA) and optionally indicated treatments. Bioluminescence was recorded for at least 4 days using the Lumicycle Device (Actimetrics, Wilmette, IL, USA). Circadian parameters were evaluated from 24 h average detrended curves using the Chronostar 3 software and visualized in GraphPad Prism 10.4.1.

Time-course gene expression analysis of selected core-clock genes was performed by qRT-PCR as described previously¹⁰⁶. Briefly, 3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}×10⁴ cells/well were seeded in 24-well plates, allowed to adhere for 24 h and synchronized by medium change. Cells were harvested at indicated timepoints (every 3 h for 54 h) and RNA was extracted using the RNeasy Plus Mini kit (Qiagen, Hilden, Germany) according to the manufacturer. One microgram of RNA was reverse transcribed to cDNA and qRT-PCR was performed using QuantiTect Primers (Qiagen, Hilden, Germany): GAPDH: QT00079247; BMAL1: QT00011844; PER2: QT00011207; NR1D1: QT00000413; CRY1: QT00025067, CRY2: QT00094920 and NR1D2: QT00008897. CT values were normalized to GAPDH and mean gene expression to obtain fold-change values. Best-fit period values were determined using the R package DiscoRhythm based on PER2 and BMAL1 curves. Cosinor analysis was performed to obtain p-values of rhythmicity and circadian parameters (amplitude, acrophase). Data were visualized in GraphPad Prism 10.4.1. All measurements were performed in 2–3 independent biological replicates, including 3 technical replicates each. One-way ANOVA was used for comparisons between groups.

Cell population behavior under TMZ treatment was simulated using a modified growth model with Allee effect. Classical model equation was expanded by a first-order decay term to account for toxic effects of alkylated DNA (DNA_A) adducts:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{dN}}{{dt}}={C}_{{KO}}{k}_{p}N\left(1-\frac{N}{{K}_{c}}\right)\left(\frac{N}{{N}_{c}}-1\right)-{k}_{t}\left[{{\rm{DNA}}}_{{\rm{A}}}\right]$$\end{document}dNdt=CKOkpN1−NKcNNc−1−ktDNAAwhere N is cell population size, k_p is the populations’ proliferation rate constant, K_c is the environmental carrying capacity, N_c is the critical cell population size, k_t is the toxicity induced population decay rate constant, and [DNA_A] is the concentration of alkylated DNA adduct. The factor C_KO was introduced to account for minor losses in cell fitness due to the introduction of gene knock-outs (wild-type cell lines C_KO = 1; knock-out cell lines C_KO = 0.85).

Cellular uptake, intracellular modifications, accumulation of methyl-diazonium ion (CH₃N₂⁺), and DNA alkylation were simulated using the model presented by Stéphanou et al.¹⁰⁹. Constant intra- and extra-cellular pH of 7.4 was assumed over the simulation time course. Extra-cellular volume was set to 1 mL to account for in vitro experimental settings. The detoxifying reaction catalyzed by O6-methylguanin-DNA-methyltransferase (MGMT) was simulated using the model presented by Zang et al.¹¹⁰.

To model circadian fluctuations in MGMT¹¹¹ expression, we implemented a core-clock network model previously developed by our group¹⁸. Connection between TMZ effect- and core-clock network model was established via the CRY component of the latter by adjusting the term for MGMT:DNA_A complex formation for the MGMT concentration to follow the current-to-mean CRY level-ratio in the model equation describing DNA_A formation. The inverse of this term was introduced for BMAL to model its anti-phasic relation to MGMT expression:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d\left[{{\rm{DNA}}}_{{\rm{A}}}\right]}{{dt}}={k}_{a}\left[{{\rm{CH}}}_{3}{{\rm{N}}}_{2}^{+}\right]+{k}_{d}\left[{\rm{MGMT}}:{{\rm{DNA}}}_{{\rm{A}}}\right]-{k}_{f}\frac{{\left[{\rm{CRY}}\right]}_{t}}{{\left[{\rm{CRY}}\right]}_{{mean}}}\frac{{\left[{\rm{BMAL}}\right]}_{{mean}}}{{\left[{\rm{BMAL}}\right]}_{t}}\left[{\rm{MGMT}}\right]\left[{{\rm{DNA}}}_{{\rm{A}}}\right]$$\end{document}dDNAAdt=kaCH3N2++kdMGMT:DNAA−kfCRYtCRYmeanBMALmeanBMALtMGMTDNAAHere, k_a is the alkylation reaction rate constant, k_d is the MGMT:DNA_A complex dissociation rate constant, k_f is the MGMT:DNA_A complex formation rate constant, [CH₃N₂⁺] is the methyl-diazonium ion concentration, [MGMT:DNA_A] is the MGMT:DNA_A complex concentration, [CRY]_t is the current CRY level, [CRY]_mean is the mean CRY level over simulation time (equivalent for BMAL terms), [MGMT] is MGMT concentration, and [DNA_A] is the concentration of alkylated DNA adduct. The complete set of model equations, in addition to the ones in the core-clock network model previously developed by our group¹⁸ is listed below. Variables and parameters are as here specified and as previously defined in the published used models by our group¹⁸ (a model from our group with the core-clock can be found at biomodels https://www.ebi.ac.uk/biomodels/MODEL2109140001↗), and by Zang et al.¹¹⁰ for MGMT, and also Stéphanou et al.¹⁰⁹ for DNA alkylation.

Pharmacokinetic and dynamic model

Extracellular TMZ concentration: 3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{T}_{o}}{{dt}}=-\frac{{p}_{T}}{{V}_{o}}{T}_{o}$$\end{document} d T o dt p T V o = − T o

With p_T being the inwards cellular transport rate, Vo being the extracellular volume, and To being the extracellular TMZ concentration.

Intracellular TMZ concentration: 4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{dT}}_{i}}{{dt}}=-\left(\frac{{p}_{T2}}{{V}_{i}}+{k}_{T}\right){T}_{i}+\frac{{p}_{T}}{{V}_{i}}{T}_{o}$$\end{document} dT i dt p T V i = − + p T 2 V i + k T T i T o

With p_T2 being the outward cellular transport rate, V_i being the intracellular volume, k_T being the MTIC formation rate, and T_i being the intracellular TMZ concentration. Note that - deviating from Stéphanou et al. - rate constants were not considered as a function of pH assuming a constant pH of 7.4 over the course of simulation.

MTIC formation: 5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{dM}}_{i}}{{dt}}={k}_{T}{T}_{i}-{k}_{M}{M}_{i}$$\end{document} dM i dt = − k T T i k M M i

With k_M being the methyl-diazonium ion formation rate, and M_i being the intracellular MTIC concentration.

Methyl-diazonium ion formation: 6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{dC}}_{i}}{{dt}}={k}_{M}{M}_{i}-\left({k}_{{cat}}+{k}_{{add}}\right){C}_{i}$$\end{document} dC i dt = − k M M i C i k cat k add +

With k_cat being the methyl-diazonium ion, k_add being the DNA adduct formation rate, and C_i being the intracellular methyl-diazonium ion concentration.

DNA adduct formation: 7 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d\left[{{\rm{DNA}}}_{{\rm{A}}}\right]}{{dt}}={k}_{a}\left[{{\rm{CH}}}_{3}{{\rm{N}}}_{2}^{+}\right]+{k}_{d}\left[{\rm{MGMT}}:{{\rm{DNA}}}_{{\rm{A}}}\right]-{k}_{f1}\frac{{\left[{\rm{CRY}}\right]}_{t}}{{\left[{\rm{CRY}}\right]}_{{mean}}}\frac{{\left[{\rm{BMAL}}\right]}_{{mean}}}{{\left[{\rm{BMAL}}\right]}_{t}}\left[{\rm{MGMT}}\right]\left[{{\rm{DNA}}}_{{\rm{A}}}\right]$$\end{document} d DNA A dt CRY t CRY mean BMAL mean BMAL t = + − k a k d k f 1 CH 3 N 2 + MGMT : DNA A MGMT DNA A

MGMT catalyzed detoxification

DNA adduct-MGMT associate formation: 8 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{{MGMT}}_{1}}{{dt}}={k}_{f1}\frac{{\left[{\rm{CRY}}\right]}_{t}}{{\left[{\rm{CRY}}\right]}_{{mean}}}\frac{{\left[{\rm{BMAL}}\right]}_{{mean}}}{{\left[{\rm{BMAL}}\right]}_{t}}\left[{\rm{MGMT}}\right]\left[{{\rm{DNA}}}_{{\rm{A}}}\right]-{k}_{r1}{{MGMT}}_{1}-{k}_{f2}{{MGMT}}_{1}+{k}_{r2}{{MGMT}}_{2}$$\end{document} d MGMT 1 dt CRY t CRY mean BMAL mean BMAL t = − − + k f 1 k r 1 MGMT 1 k f 2 MGMT 1 k r 2 MGMT 2 MGMT DNA A

With k_f1 being the DNA adduct-MGMT associate formation rate, k_r1 being the DNA adduct-MGMT associate dissociation rate, k_f2 being the DNA adduct-MGMT complex formation rate, k_r2 being the DNA adduct-MGMT complex dissociation rate, MGMT₁ being the DNA adduct-MGMT associate, and MGMT₂ being DNA adduct-MGMT complex.

DNA adduct-MGMT complex formation: 9 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{dMGMT}}_{2}}{{dt}}={k}_{f2}{{MGMT}}_{1}-{k}_{r2}{{MGMT}}_{2}-{k}_{f3}{{MGMT}}_{2}$$\end{document} dMGMT 2 dt = − − k f 2 MGMT 1 k r 2 MGMT 2 k f 3 MGMT 2

With k_f3 being the DNA diacylation rate constant.

DNA de-acylation: 10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{dMGMT}}_{3}}{{dt}}={k}_{f3}{{MGMT}}_{2}$$\end{document} dMGMT 3 dt = k f 3 MGMT 2

With k_f3 being the DNA diacylation rate constant.

Cell population model

Cell-proliferation: 11 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{dN}}{{dt}}={C}_{{KO}}{k}_{p}N\left(1-\frac{N}{{K}_{c}}\right)\left(\frac{N}{{N}_{c}}-1\right)-{k}_{t}\left[{{\rm{DNA}}}_{{\rm{A}}}\right]$$\end{document} dN dt = − C KO k p k t N 1 − N K c N N c − 1 DNA A

Cells were seeded at a concentration of 2.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document}×10³/well and allowed to adhere for 24 h. Cells were synchronized by medium change for 0, 5, 10, 15 or 20 h and treated with the indicated treatments or treatment combinations of TMZ, KL001 or SR9011 at a pre-determined IC₅₀ concentration. Cell confluence was measured every 2 h using the IncuCyte S5 (Sartorius, Göttingen, Germany). Cell confluency was normalized to cell confluency at t = 0 and AUC values were determined using GraphPad Prism 10.4.1. and normalized to the respective AUC values of DMSO-treated cells for each plate. All measurements were performed in 2–3 independent biological replicates, including 3 technical replicates each. One-way ANOVA was used for comparisons between groups.