Telling time in tumor samples reveals diversity of clock disruption

Epidemiological studies have revealed increased cancer incidence in people exposed to circadian disruption, such as that caused by night shift work, with the strongest evidence of such an association for breast cancer. However, mechanisms underlying enhanced cancer risk associated with circadian disruption are unknown and are likely pleiotropic. Studies in mouse models indicate that circadian disruption may enhance tumor burden by suppressing immune surveillance (1–3), altering proteostasis signaling pathways (4), suppressing tumor-intrinsic circadian clocks (5, 6), or disrupting circadian regulation of growth-regulatory pathways (7–10). A major challenge for studying circadian rhythms in humans is the inability to collect biopsies from individual people at multiple times of day. Despite the availability of enormous publicly available databases that have allowed major research progress in cancer (11), such projects do not record the time of day at which samples are collected, making it difficult to investigate the implications of circadian disruption in such data. Informatic approaches can address these challenges by using defined patterns of expression for genes that exhibit high-amplitude daily oscillations to estimate collection time (i.e., circadian phase) for individual samples (12–16) and to evaluate the robustness of circadian rhythmicity (12, 17, 18) in groups of samples segregated by tissue type or disease state.

Algorithms for estimating circadian phase or evaluating circadian robustness in samples with unknown collection times rely on analyzing the expression of a core set of genes that exhibit high-amplitude oscillations of expression across all tissue types and can therefore be applied to any type of sample. These programs have been applied to large sequencing datasets to demonstrate that many tumors have less robust circadian rhythmicity than healthy samples of the same tissue of origin (18). However, increasing evidence suggests that the connection between circadian disruption and exacerbated cancer is not universal. For example, deletion of core clock components suppresses tumor growth in glioblastoma (19), hepatocellular carcinoma (20), and leukemia (21), and chronic disruption of circadian light exposures does not affect MYC-driven lymphoma in a mouse model (22). While only a small number of genes are ubiquitously rhythmic, many more transcripts exhibit consistent rhythmicity within specific anatomic locations (23). In this issue of PNAS, Li et al. (24) use sequencing data from breast tumor and matched normal breast tissue biopsies with recorded collection times to refine informatic approaches for estimating collection time and circadian robustness in breast biopsies (Fig. 1). With this machine learning approach, they generated an improved breast-specific algorithm and used it to demonstrate that circadian disruption is most severe in aggressive breast cancers—but unexpectedly correlates with better survival outcomes in the least aggressive (luminal A) subtype.

Fig. 1.

Sample collection time is rarely recorded when biopsies are performed and is not included in clinical data associated with large sequencing projects. Several informatic approaches have been developed to estimate circadian phase (a proxy for sample collection time) from gene expression data. These methods can be refined by training them on data generated from relatively small sets of time-stamped samples, thus enabling investigation of circadian properties in existing large datasets. This figure was created with Biorender.com.

Breast cancers are typically grouped into four major subtypes, which are associated with the presence or absence of receptors for the hormones estrogen (ER) and progesterone (PR) and the epidermal growth factor receptor modulator HER2 and defined by additional histological and clinical features. Luminal A tumors (ER+ and/or PR+; HER2−) are the most common and have the best prognosis. Triple-negative breast cancers (TNBCs, ER−; PR−, HER2−) are the most aggressive and have poorer outcomes. Consistent with some prior studies, Li, Hammarlund, Wu, and colleagues show that circadian disruption is more pronounced in the more aggressive subtypes luminal B and TNBC than it is in luminal A breast cancers. However, within the group of patients with luminal A–type breast tumors, they find that patients for whom biopsies exhibit greater circadian rhythmicity have worse 5-year survival outcomes.

Notably, circadian rhythmicity can be disrupted in myriad ways and the state in which the clock is “broken” may have important implications for cancer biology. Reduced robustness of circadian rhythmicity in a biopsy could be caused by decreased amplitude of circadian expression for some or all genes included in the algorithm, altered phase relationships between the genes, or altered cellular composition of the collected sample. The current study demonstrated that reduced rhythmicity in TNBC compared to healthy breast tissue and less aggressive tumors is likely intrinsic to the tumor cells by measuring circadian rhythms in organoid cultures. Organoids formed from luminal A tumor cells exhibit less robust rhythmicity than those from normal breast cells but greater rhythmicity than those formed from TNBC. It is unclear whether enhanced rhythmicity indicated by gene expression patterns in biopsies from patients with luminal A tumors that had decreased survival can be attributed to tumor-intrinsic mechanisms.

In PNAS, Li et al. use sequencing data from breast tumor and matched normal breast tissue biopsies with recorded collection times to refine informatic approaches for estimating collection time and circadian robustness in breast biopsies.

Over the past decades, public investment has enabled the generation of large datasets that provide opportunities to define genetic and epigenetic perturbations associated with the occurrence, clinical features, and prognosis of diverse tumor types. Development of informatic approaches like those in the highlighted study that enable robust characterization of circadian properties in these valuable datasets will be essential for translating research findings regarding mechanisms by which circadian disruption can influence tumor development from model organisms and to enable rational application of the principles of circadian medicine to cancer treatment.