Circadian rhythm disruption by PARP inhibitors correlates with treatment toxicity in patients with ovarian cancer and is a predictor of side effects

Abstract

Background: Ovarian cancer is among the most lethal malignancies in women. The advent of PARP inhibitors (PARPi) has improved outcomes. However, treatment-related toxicity remains a critical challenge, impacting patient quality of life and treatment adherence.

Methods: In a circadian sub-study of the MAMOC trial-a double-blind, phase III study-42 patients (FIGO stage IIIA-IV) were randomised in a 2:1 ratio to receive rucaparib or placebo. In a subset of these patients, we performed differential gene expression and rhythmicity analysis on up to 800 genes, including clock and clock-controlled genes. Machine learning algorithms and mathematical modelling were employed to simulate patient-specific toxicity profiles and to explore correlations between gene expression patterns and treatment-related side effects.

Findings: Our analysis revealed significant disruptions in circadian rhythms, specifically in the expression of the core clock genes BMAL1 and PER2, following treatment. These disruptions strongly correlated with the severity and frequency of side effects, including nausea and fatigue, displaying opposite trends between the placebo and rucaparib-treated groups. K-means clustering successfully distinguished rucaparib-treated patients from those receiving placebo based on BMAL1 phase and gene expression profiles. In addition, rucaparib therapy also altered the expression of several clock-controlled genes, including SIRT1, BRCA1, BRCA2, and TP53. Notably, our data suggest that individual differences in circadian rhythms may lead to distinct 24-h toxicity profiles among patients.

Interpretation: These findings suggest that circadian rhythm dysregulation may contribute to the toxicity of PARPi therapy. Aligning treatment timing with circadian rhythms could mitigate these adverse effects, and improve patient outcomes.

Funding: This study was funded by the Dr. Rolf Schwiete Stiftung and the MSH Medical School Hamburg, Germany. The MAMOC trial (ClinicalTrials.gov: NCT04227522) was funded by Clovis Oncology, United States.

Keywords: Adverse events; Chronotherapy; Circadian profiles; Circadian rhythms; Mathematical modelling; Ovarian cancer; PARP inhibitors; Rucaparib.

Fig. 1

Overview of the circadian sub-study within a Phase III double-blind, placebo-controlled trial. A total of 42 patients with high-grade ovarian cancer were randomised to receive either rucaparib (N_P = 28) or placebo (N_P = 14). Saliva samples from 15 of those patients enabled molecular profiling of circadian rhythms during therapy. Gene expression analysis revealed treatment-specific circadian clock alterations in the rucaparib group, which correlated with side effects of the therapy. Mathematical modelling of toxicity profiles underscored the potential of circadian-based treatment timing to reduce side effects and improve outcomes (created in https://BioRender.com).

Fig. 2

Study design, patient recruitment timeline, and clinical characteristics of the cohort. a. Schematic representation of the overall study design. 42 patients with high-grade ovarian cancer (14 placebo, 28 rucaparib) were recruited. Saliva samples from 15 patients were collected at multiple time points throughout treatment (N_P = number of patients; N_S = number of samples). All the patients completed various questionnaires covering topics such as quality of life and adverse event grading. Additional data on dosage changes, treatment interruptions, and activity tracker metrics were collected. b. Overview of patient recruitment timeline and collection of PRO for those who provided saliva samples. The time points for saliva sample collection (for circadian rhythm analysis) and PRO (EORTC-QLQ-C30, FSI, etc.) are shown. Green bars indicate the end of previous bevacizumab therapy, grey bars mark saliva sample collection for molecular circadian analysis, and yellow bars indicate questionnaire completion during treatment. Patient IDs (P01–P15) are colour-coded: placebo (brown) and rucaparib (orange). c-e. Descriptive analysis of cancer characteristics. The bar chart displays the clinical classification of 42 patients based on c. ovarian cancer type, d. cancer severity according to FIGO staging, and e. reasons for treatment discontinuation (EOT). Statistical analysis was performed using Fisher's exact test.

Fig. 3

Descriptive analysis of patientreported outcomes(PRO)and correlations with activity tracker data. a-c. PRO. Patients in the placebo and rucaparib groups completed questionnaires evaluating quality of life and adverse events. “+" indicates fewer negative effects, and "–" represents more negative effects. The density plot shows the distribution of the full MAMOC cohort (14 placebo, 28 rucaparib), while the jitter plots and boxplots represent the Circadian cohort (5 placebo, 10 rucaparib). a-b. The EORTC-QLQ-C30 and EORTC-QLQ-OV28 showed worsened physical functioning and increased severe symptoms in the rucaparib group. c. The FSI revealed more severe fatigue in the rucaparib group. Statistical analysis was performed using Wilcoxon rank-sum test with BH correction; error bars are 95% confidence intervals. d-f. Correlations between activity and PRO. Step count recorded over one month, was plotted against PRO data for the same period. d. Step count correlated positively with the Global health scale. e. Step count decreased with higher PRO-CTCAE scores. f. Step count decreased with higher insomnia scores. OLS regression was used for the fitting, and p-values were obtained for each correlation. g-i. Global health scale correlates negatively with fatigue. OLS regression was used in panel g, Wilcoxon rank-sum test was used in panels h and i.

Fig. 4

Rucaparib therapy resulted in alterations of core-clock gene expression and its circadian properties. a. Graphical representation of the interaction between the core-clock network, especially BMAL1 and PARP1. b.BMAL1 showed significantly lower expression in Placebo vs. Baseline, and higher expression in Rucaparib_A_EOT vs. Placebo. Whereas, PER2 showed significantly higher expression in Rucaparib_A_EOT vs. other groups. PARP1 expression was higher in Rucaparib and Rucaparib_A_EOT groups, while no significant differences were observed. Outliers were defined as data points falling outside 1.5 times the interquartile range (IQR) from the first (Q1) and third (Q3) quartiles, following standard statistical practices. Individual data points are overlaid as jittered dots. Statistical test used: Wilcoxon rank-sum test (5 placebo, 10 rucaparib). c. Lollipop plot represents the changes in the expression of the NCRG genes due to Rucaparib therapy or no therapy (Placebo) vs. the Baseline (N_S = 84). d. Acrophase bin plot represents the changes in phase (first peak expression) in BMAL1, PER2, and PARP1 genes due to Rucaparib therapy vs the Baseline (5 placebo, 10 rucaparib).

Fig. 5

Alterations of clock properties correlate withPROand adverse events in the rucaparib group. To investigate potential correlations between BMAL1, PER2, and PARP1 gene expression, their circadian properties, and PRO, we performed a Spearman correlation analysis at the patient level (5 placebo, 10 rucaparib). Spearman rank correlation and permutation tests were used to assess the significance of the correlation pairs. a. At baseline, BMAL1 amplitude showed a weak negative correlation with insomnia. b. In the placebo group, the correlation between BMAL1 amplitude and insomnia was slightly positive. c. In the rucaparib treatment group, this correlation was strongly negative. d. After the end of treatment, the correlation remained negative but was weaker than during treatment.

Fig. 6

Circadian dysregulation among patients with rucaparib therapy correlates with reported adverse events. a-f The rucaparib group showed an increase in adverse events such as fatigue and insomnia a-c in the MAMOC cohort (14 placebo, 28 rucaparib), as well as d-f in the Circadian cohort (5 placebo, 10 rucaparib) as assessed by a questionnaire. The questionnaire outcomes are visualised across these stages to assess changes over the course of treatment (C: cycle; D: day; SFU: safety follow-up; FU: follow-up; only one patient filled out the questionnaire during C1D1 and C14D1). Error bars are defined as Mean ± SD. g The line graphs depict the circadian profile of core clock genes (BMAL1 and PER2) over the treatment timeline for each patient. Adverse events reported by the patients are indicated at relevant time points, highlighting potential associations between circadian profile alterations and side effect occurrences.

Fig. 7

Circadian dysregulation among patients may reflect the connection strength within the core-clock network and result in the differential regulation and differential rhythmicity of cancer markers. a-b. Correlation heatmap represents the discrepancies in the strength of the core-clock network due to rucaparib therapy in P07 and P14. c-d. Volcano plot represents the differential gene regulation among P07 and P14 patients during rucaparib therapy. e-f. Acrophase plot depicts the differential rhythmicity of core-clock and cancer markers due to rucaparib therapy in P07 and P14.

Fig. 8

In silico simulations show an effect of different core clock phenotypes in daily drug toxicity across treatment cycles. Toxicity profiles have been generated using a previously established model fitted to another drug with a similar mechanism of action also interfering with DNA repair machinery, for different dosing schedules.a. Graphical representation of PARP1's mode of action. b. A previously published mathematical model for predicting drug toxicity targeting the DNA repair pathway and cell cycle duplication. c. Circadian profile of BMAL1 and PER2 genes among P07, P09, and P14 based on the gene expression normalised to the model's scale (dots) and the corresponding model fit (line). d. Based on the fitted expression profiles, the model predicted the level of rucaparib drug toxicity. Different curves represent specific drug toxicity profiles which result from alterations in circadian rhythms. Marked with dashed lines are the predicted treatment times with minima in toxicity.

Fig. 9

BMAL1 and PER2 circadian properties influencePROand can be used to stratify patients into treated and untreated groups. a-f OLS with bootstrapping analyses demonstrates the relationships between BMAL1 and PER2 circadian properties and PRO such as adverse events and quality of life. The error bar represents 95% CI (5 placebo, 10 rucaparib). g-h Unsupervised k-means clustering based on BMAL1 gene expression and PER2 MESOR identifies distinct patient groupings between the placebo and rucaparib groups (5 placebo, 10 rucaparib).