Timing gone awry: distinct tumour suppressive and oncogenic roles of the circadian clock and crosstalk with hypoxia signalling in diverse malignancies

Abstract

Background: The circadian clock governs a large variety of fundamentally important physiological processes in all three domains of life. Consequently, asynchrony in timekeeping mechanisms could give rise to cellular dysfunction underpinning many disease pathologies including human neoplasms. Yet, detailed pan-cancer evidence supporting this notion has been limited.

Methods: In an integrated approach uniting genomic, transcriptomic and clinical data of 21 cancer types (n = 18,484), we interrogated copy number and transcript profiles of 32 circadian clock genes to identify putative loss-of-function (Clock^Loss) and gain-of-function (Clock^Gain) players. Kaplan-Meier, Cox regression and receiver operating characteristic analyses were employed to evaluate the prognostic significance of both gene sets.

Results: Clock^Loss and Clock^Gain were associated with tumour-suppressing and tumour-promoting roles respectively. Downregulation of Clock^Loss genes resulted in significantly higher mortality rates in five cancer cohorts (n = 2914): bladder (P = 0.027), glioma (P < 0.0001), pan-kidney (P = 0.011), clear cell renal cell (P < 0.0001) and stomach (P = 0.0007). In contrast, patients with high expression of oncogenic Clock^Gain genes had poorer survival outcomes (n = 2784): glioma (P < 0.0001), pan-kidney (P = 0.0034), clear cell renal cell (P = 0.014), lung (P = 0.046) and pancreas (P = 0.0059). Both gene sets were independent of other clinicopathological features to permit further delineation of tumours within the same stage. Circadian reprogramming of tumour genomes resulted in activation of numerous oncogenic pathways including those associated with cancer stem cells, suggesting that the circadian clock may influence self-renewal mechanisms. Within the hypoxic tumour microenvironment, circadian dysregulation is exacerbated by tumour hypoxia in glioma, renal, lung and pancreatic cancers, resulting in additional death risks. Tumour suppressive Clock^Loss genes were negatively correlated with hypoxia inducible factor-1A targets in glioma patients, providing a novel framework for investigating the hypoxia-clock signalling axis.

Conclusions: Loss of timekeeping fidelity promotes tumour progression and influences clinical outcomes. Clock^Loss and Clock^Gain may offer novel druggable targets for improving patient prognosis. Both gene sets can be used for patient stratification in adjuvant chronotherapy treatment. Emerging interactions between the circadian clock and hypoxia may be harnessed to achieve therapeutic advantage using hypoxia-modifying compounds in combination with first-line treatments.

Keywords: Circadian clock; Gain-of-function; Glioma; Hypoxia; Loss-of-function; Oncogene; Pan-cancer; Renal cancer; Tumour suppressor.

Fig. 1

Circadian reprogramming in diverse cancer types. a Schematic diagram depicting the project design and the identification of putative loss-of-function and gain-of-function clock genes. Somatic copy number alteration (SCNA) and transcript expression of 32 clock genes are investigated in 21 cancer types. A total of 19 or 12 genes are recurrently lost or gained respectively. Of these SCNA events, 11 or two genes are also downregulated or upregulated in tumours, representing Clock^Loss and Clock^Gain gene sets respectively. Both gene sets are prognostic in seven cancer cohorts. Pie slices indicate the number of patients within each cancer type. Crosstalk between circadian genes and tumour hypoxia is investigated. b The proportion of samples with deep and shallow somatic alterations are represented using stacked bar graphs. The number of samples within each cancer type is represented by the width of the stacked bars. c Somatic losses and differential expression profiles of 19 clock genes that are recurrently deleted in at least seven cancer types. d Somatic gains and differential expression profiles of 12 clock genes that are recurrently amplified in at least seven cancer types. Bar charts on the far right represent the number of cancers with at least 20% of samples affected by copy number alteration. Heatmaps on the far left depict the cohort fraction in which a given gene is deleted or amplified. Cancer types are ordered using Euclidean distance metric. Heatmaps in the centre represent differential expression values between tumour and non-tumour samples. Clock^Loss and Clock^Gain genes are highlighted in red. Cancer abbreviations are listed in Additional file 2

Fig. 2

Prognostic significance of Clock^Loss and Clock^Gain. Kaplan–Meier plots are generated using a Clock^Loss and b Clock^Gain. Patients are quartile stratified based on their clock gene scores. P values are obtained from log-rank tests. c Ordination plots of multidimensional scaling analyses using Clock^Loss genes reveal significant differences between tumour and non-tumour samples. P values are obtained from PERMANOVA tests. d Expression distribution of Clock^Gain scores in tumour and non-tumour samples with statistical analyses performed using Mann–Whitney–Wilcoxon tests. P values are represented by ****< 0.00001. ns non-significant

Fig. 3

Clock^Loss and Clock^Gain are independent of tumour stage. Kaplan–Meier plots are generated from patients stratified according to TNM stage and a Clock^Loss and b Clock^Gain. TNM staging is first used to stratify patients, followed by median stratification into high- and low-score groups using Clock^Loss or Clock^Gain. b Glioma histological subtypes, astrocytoma and oligodendroglioma, are quartile stratified using Clock^Gain. P values are obtained from log-rank tests. ROC analyses on c Clock^Loss and d Clock^Gain to determine the specificity and sensitivity of both gene sets in predicting 5-year overall survival rates. ROC curves generated from clock gene sets are compared to those generated from TNM staging. AUCs for TNM stage are in accordance with previous work utilising TCGA datasets [–71]

Fig. 4

Circadian dysregulation drives malignant progression. Differential expression analyses are performed between 4th and 1st quartile patients determined using Clock^Loss or Clock^Gain. a Venn diagrams illustrate the number of differentially expressed genes (DEGs) and their overlapping patterns in five cohorts. Numbers in parentheses represent DEGs. Table inset depicts the number of genes that are found to be in common in five, > four and > three cancer types. Overlap between common Clock^Loss and Clock^Gain genes are also depicted. Enriched b GO terms, c KEGG ontologies and d transcription factor binding associated with DEGs

Fig. 5

Prognostic relevance of the hypoxia and clock crosstalk. Scatter plots depict a significant negative correlations and b significant positive correlations between hypoxia and Clock^Loss or Clock^Gain scores respectively. Patients are grouped into four categories based on median clock and hypoxia scores. At the x- and y-axes, density plots depict the distribution of clock and hypoxia scores. Kaplan–Meier analyses are performed on the four patient groups to determine the effects of crosstalk between hypoxia and c Clock^Loss and d Clock^Gain on overall survival in multiple cancers including glioma histological subtypes

Fig. 6

Model of the hypoxia-clock signalling axis in glioma. The circadian clock exerts tumour promoting or tumour suppressing qualities that are dependent on cellular types. Tumour suppressive Clock^Loss genes are negatively correlated with HIF-1A target genes (CA9, VEGFA and LDHA) in glioma. Clock^Loss scores are plotted such that each spoke of the circular heatmap represents individual patients that are sorted in descending order. Circular heatmaps for HIF-1A target genes are plotted with patients sorted in descending order of Clock^Loss scores. Spearman’s correlation coefficients between Clock^Loss and individual HIF-1A genes are depicted in the centre of the heatmap