Main

CTCs are pioneers of the metastatic cascade in several cancer types, including breast cancer1. The factors that regulate spontaneous CTC intravasation in physiological settings are poorly understood, and the general assumption is that CTCs are constantly generated from invasive cancerous tissues2, or generated following mechanical cues such as surgery3 or physical activity4. In patients and in mouse cancer models, the exact timing of the events that characterize metastatic cancer progression, and the principles that dictate intravasation of CTCs and their proclivity to metastasize, are unclear. A better understanding of these processes may result in new approaches for cancer investigation and treatment.

Circadian rhythm and CTC intravasation

We first sought to determine CTC abundance and composition in hospitalized women with progressive breast cancer who had no treatment or were temporarily off-treatment and who consented to donate blood during the active (10:00 am) and rest (4:00 am) phases of the same day, including a total of 30 patients (Fig. 1a). Of these, 21 patients were diagnosed with early breast cancer (no metastasis) and 9 were diagnosed with stage IV metastatic disease at the time of blood sampling (Supplementary Table 1). Strikingly, following antigen-agnostic microfluidic capture of CTCs and confirmation through immunofluorescence staining5, we found most CTCs (78.3%) in samples obtained at night time during the rest phase, including single CTCs, CTC clusters and CTC–white blood cell (WBC) clusters (Fig 1a, Extended Data Fig. 1a and Supplementary Table 1). To test the generality of these findings and to finely characterize the precise timing of the events, we used four different mouse models of breast cancer, including xenografts derived from human breast CTCs (NSG-CDX-BR16), xenografts with established human breast cancer cells (NSG-LM2) or mouse breast cancer cells (NSG-4T1), and an immunocompetent syngeneic breast cancer model (BALB/c-4T1). Following transplantation of breast cancer cells in the mammary fat pad and tumour growth, we examined spontaneous CTC generation over time by means of terminal blood sampling and microfluidic CTC capture. Consistently with patient data, we found most CTC events (99.2% in NSG-LM2, 97% in NSG-CDX-BR16, 93.8% in NSG-4T1, 87% in BALB/c-4T1) to be present in samples obtained through cardiac puncture during the mouse rest phase (corresponding to daylight time, given the inverted circadian rhythm of rodents compared to humans6; Fig. 1b,c and Extended Data Fig. 1a). More precisely, when carrying out a time kinetic analysis of a 24-h time period at intervals of 4 h, we observed a very prominent oscillatory pattern of CTC release, peaking between 4 and 12 h zeitgeber time (ZT; with ZT0 defined as 06:00 am when the lights turn on, and ZT12 defined as 06:00 pm when the lights turn off) corresponding to their rest phase (Fig. 1b and Extended Data Fig. 1b) in mice with analogous tumour burden (Extended Data Fig. 1c,d). When focusing on the two most representative time points for the rest (ZT4) and active (ZT16) phases of the mouse circadian rhythm, we observe marked differences in absolute and normalized CTC counts in all tested models, with a 6- to 88-fold increase for single CTCs, a 12- to 278-fold increase for CTC clusters and an 8- to 34-fold increase for CTC–WBC clusters during ZT4 (Fig. 1c and Extended Data Fig. 1e–g), whereas no changes are observed in the ratio of the various CTCtypes (Extended Data Fig. 1h). Further, we identified highly similar (yet, even more pronounced) oscillatory patterns in CTC abundance when blood samples were obtained directly from the tumour draining vessel (TDV; Fig. 1b and Extended Data Fig. 2a–e). Given these results, the extremely short circulation half-life of CTCs7,8, and their similar clearance rate during different time points of the rest and active phases (Extended Data Fig. 2f,g), we conclude that the main differences in CTC abundance observed during the rest versus the active phase are to be ascribed to differences in intravasation rates. We then attempted to perturb the physiological rhythm of tumour-bearing mice by different means. First,on the one hand, we used well-established methods to shift the normal light/dark (LD) cycle, provoking a jet-lag effect9, and on the other hand, we treated control and jet-lagged mice with melatonin, a key hormone that regulates the sleep cycle (Extended Data Fig. 3a). Strikingly, when analysing blood in all mice at ZT4, we find that jet-lag induction leads to a 38- to 282-fold decrease in single CTCs, a 63- to 484-fold decrease in CTC clusters and a 28- to 219-fold decrease in CTC–WBC clusters compared to those in control mice and with no changes in primary tumour size (Extended Data Fig. 3b–d). Further, when treating tumour-bearing mice with melatonin (daily, 2 h before the start of the rest phase), and exposing them to a jet-lagged or normal LD cycle for a total duration of 22 days, we find a marked melatonin-induced increase in the production of single CTCs, CTC clusters and CTC–WBC clusters in all cases, rescued by the melatonin receptor antagonist luzindole (Extended Data Fig. 3e). Along with the effects of melatonin in increasing CTC production and luzindole in decreasing it without affecting primary tumour size, we observe an augmented or reduced metastatic burden, respectively (Extended Data Fig. 3f,g). We then exposed tumour-bearing mice to altered LD cycles. We used a long-day photoperiod (14:10 LD), as well as two different T-cycles with LD cycles that differ from 24 h (20-t, 10:10 LD and 28-t, 14:14 LD), and tested their effect on CTC generation. We observe a consistent increase in CTC counts during the rest phase in each of these light conditions (Extended Data Fig. 4a–d), suggesting a key role for light exposure, and its consequences, in CTC intravasation. Last, given the oscillatory pattern of CTC intravasation and its relation to the circadian rhythm, we tested whether this pattern was abolished in the context of a syngeneic transplantation of E0771.lmb mouse cancer cells in either wild-type BL/6 mice (BL/6-E0771.lmb) or in Bmal1-homozygous-knockout mice (BL/6-Bmal1−/−-E0771.lmb), the only arrhythmic single-gene-knockout model10 (Fig. 1d). We find that, whereas CTC counts from both cardiac puncture and the TDV follow a typical oscillatory pattern in control BL/6-E0771.lmb mice, oscillation in CTC counts is lost in BL/6-Bmal1−/−-E0771.lmb mice (Fig. 1d and Extended Data Fig. 4e). Of note, BL/6-Bmal1−/−-E0771.lmb mice generally fail to generate CTCs despite identical tumour size and timing of sample collection compared to those for BL/6-E0771.lmb control mice (Extended Data Fig. 4f,g), highlighting that a disruption of the master regulator of the circadian clock results in abolished CTC intravasation. Taken together, these results demonstrate that CTCs are not shed continuously during tumour progression, but the greatest release of single and clustered CTCs is achieved during sleep in both patients with breast cancer and mouse models.

Fig. 1: CTCs intravasate during the rest phase of the circadian rhythm.
figure 1

a, Left: graphical representation of the human circadian rhythm. The white and black bars represent environmental light (active period) and dark (rest period) conditions, respectively. Right: radial histograms showing the percentage of single CTCs, CTC clusters and CTC–WBC clusters isolated during the rest and active phases in patients with early- or late-stage breast cancer. n = 21 patients with early-stage cancer and n = 9 patients with late-stage cancer. Pat, patient. b, Top: graphical representation of the mouse circadian rhythm. The white and black bars represent environmental light (rest period) and dark (active period) conditions, respectively. Bottom: time kinetic analysis showing CTC counts in the NSG-CDX-BR16 mouse model of breast cancer, from blood collected through cardiac puncture or TDV over a 24-h time period (n = 4 for all time points, except ZT12 (cardiac puncture and TDV) and ZT20 (TDV), for which n = 3). c, Box plots showing the distribution of the number of CTCs collected at ZT4 and ZT16 in the immunocompromised NSG-LM2 (n = 6; single CTCs P = 0.0152; CTC clusters and CTC–WBC clusters P = 0.0087) and NSG-4T1 (n = 4; P = 0.0286 for all) and immunocompetent BALB/c-4T1 (n = 5; single CTCs P = 0.0159; CTC clusters P = 0.0079; CTC–WBC clusters P = 0.0317) mouse models of breast cancer. d, Left: graphical representation of physiological (BL/6-EO771.lmb mice) versus impaired circadian rhythm (BL/6-Bmal1−/−-EO771.lmb mice). Right: graphs showing time kinetic analysis of CTC counts (single CTCs, CTC clusters and CTC–WBC clusters) in BL/6-EO771.lmb (ZT4, ZT12, ZT16 n = 4; ZT0 n = 3) and BL/6-Bmal1−/−-EO771.lmb (n = 3) mice, from blood collected through cardiac puncture over a 24-h time period. The data in b,d are presented as mean ± s.e.m.; for c, the centre lines in the box represent the median, the box limits represent the first and third quartiles, and the extremes of the whisker lines represent the minimum and maximum observed values. *P < 0.05, **P < 0.01 by two-sided Mann–Whitney test. n represents the number of biologically independent mice. The human figure, sun and moon were created with BioRender.com.

Source data

Full size image

Metastatic ability of CTCs during sleep

We next investigated whether, in addition to their increased production during the rest phase, CTCs that are generated during different phases of the circadian rhythm also harbour a different potential to successfully metastasize. To this end, we used the NSG-LM2 xenograft model, exclusively labelled with either GFP or RFP, and following tumour development, we isolated spontaneously shed CTCs at ZT4 (GFP-labelled) and ZT16 (RFP-labelled) by microfluidics-based capture. With a robotic micromanipulator, we isolated 150 ZT4-generated GFP-labelled CTCs and 150 ZT16-generated RFP-labelled CTCs, simultaneously, each of which composed of 110 single CTCs, 35 CTC clusters and 5 CTC–WBC clusters (representing typical CTC ratios in the NSG-LM2 model), and co-injected them through the tail vein of tumour-free recipient mice at different time points of the circadian rhythm (ZT0, ZT4, ZT12, ZT16) to measure their direct metastatic ability (Fig. 2a). Through in vivo bioluminescence imaging, we find the highest metastatic burden during the rest phase, and in particular at ZT4 (Fig. 2b). To dissect whether these metastases were derived from ZT4 or ZT16 CTCs, we conducted immunohistochemical analysis of the lungs with anti-GFP and anti-RFP antibodies. Remarkably, we find that most metastases derive from ZT4-generated GFP-labelled CTCs (Fig. 2c). These results highlight a substantial contribution of ZT4 CTCs to metastasis formation, as well as a higher proclivity of ZT4 CTCs to form metastases when injected into mice during their rest phase. Next we sought to extend our findings to further models and to precisely quantify the metastatic ability of rest-phase versus active-phase single CTCs, CTC clusters and CTC–WBC clusters, individually. We used both the NSG-CDX-BR16 and NSG-LM2 xenograft models, and following tumour development, we isolated spontaneously shed CTCs by microfluidics-based capture. With a robotic micromanipulator, we then isolated 100 single CTCs, 100 CTCs from CTC clusters and 100 CTCs from CTC–WBC clusters of mice during their rest (ZT4) and active (ZT16) phases, respectively, and injected them through the tail vein of tumour-free recipient mice at ZT12 to measure their direct metastatic ability (Fig. 2d). By means of bioluminescence imaging, we confirm that CTCs obtained during ZT4 exhibit an extraordinary metastasis-forming capacity compared to CTCs that are obtained during ZT16 (Fig. 2e and Extended Data Fig. 5a–f). Of note, when isolated during the rest phase and compared to those in the active phase, CTC clusters and CTC–WBC clusters seem to be endowed with greater metastasis-forming properties than those of single CTCs (Fig. 2e and Extended Data Fig. 5a–f), suggesting that most of the rest-phase-dependent metastatic spread of breast cancer could be ascribed to both homotypic and heterotypic CTC clusters. Together, these results indicate not only that CTC intravasation rates are increased, but also that their metastatic ability is augmented during the rest phase.

Fig. 2: The metastatic potential of CTCs is highest during the rest phase.
figure 2

a, Schematic illustration of the experimental design for b,c, Equal numbers of spontaneously shed ZT4-generated GFP-labelled CTCs and ZT16-generated RFP-labelled CTCs from NSG-LM2 mice were co-injected through the tail vein into tumour-free recipient mice at different time points of the circadian rhythm (ZT0, ZT4, ZT12, ZT16) to measure their direct metastatic ability. b, Top: representative bioluminescence images of lungs from mice being co-injected simultaneously with ZT4-generated GFP-labelled CTCs and ZT16-generated RFP-labelled CTCs from NSG-LM2 mice. Bottom, plot showing normalized bioluminescence signal obtained from lungs of mice used in the same panel (n = 3). c, Left: representative immunofluorescence images of GFP (green) and RFP (red) in lungs of mice shown in b. Nuclei are stained with 4′,6-diamidino-2-phenylindole (DAPI; blue). Right: plot showing GFP and RFP levels in lungs of mice used in the same panel (n = 3; ZT4 P = 0.0406; ZT12 P < 0.0001). a.u., arbitrary units. Scale bars, 100 μm. d, Schematic illustration of the experimental design for e. Single CTCs, CTC clusters and CTC–WBC clusters are collected at ZT4 and ZT16 and separately injected into the tail vein of recipient tumour-free mice to measure their direct metastatic potential. e, Top: representative bioluminescence images of lungs from mice injected with single CTCs, CTC clusters or CTC–WBC clusters collected at ZT4 and ZT16 from NSG-LM2 mice. Bottom: plot showing normalized bioluminescence signal obtained from lungs of mice used in the same panel (n = 3 for ZT4 single CTCs and CTC clusters; n = 4 for ZT16 single CTCs and CTC clusters; n = 2 for CTC–WBC clusters, owing to their rarity; P = 0.0272 for CTC clusters at ZT4 versus ZT16). For all panels, the data are presented as mean ± s.e.m.; *P < 0.05, ***P < 0.001 by unpaired two-sided t-test. n represents the number of biologically independent mice. The sun and moon were created with BioRender.com. 

Source data

Full size image

Time-dependent gene expression in CTCs

Next we sought to investigate molecular features determining the differential ability of CTCs to seed metastasis during the rest and active phases, respectively. Following mammary fat pad engraftment of BR16 and LM2 breast cancer cells, tumour growth and spontaneous CTC generation, we isolated single CTCs, CTC clusters and CTC–WBC clusters during the rest (ZT4) and active (ZT16) phases of the mouse circadian rhythm, and subjected them individually to single-cell RNA sequencing11,12 (scRNA-seq; Fig. 3a). In total, after filtering for high-quality samples (that is, taking into account the number of expressed genes per cell, the total number of reads per cell and the proportion of reads aligning to mitochondrial genes; see Methods), we obtained a total of 138 CTCs from the NSG-CDX-BR16 model and 108 CTCs from the NSG-LM2 model, representing all types of CTC at ZT4 and ZT16. Using principal component analysis, we find that time point (ZT4 versus ZT16) is a key feature driving variance of gene expression in CTCs (Fig. 3b and Extended Data Fig. 6a), suggesting time-point-driven gene expression changes. Differential gene expression of samples isolated during the rest (ZT4) versus the active (ZT16) phase reveals a set of 121 upregulated genes in ZT4 CTCs (log2[fold change] ≥ 0.5 and false discovery rate ≤ 0.05) and a set of 156 upregulated genes in ZT16 CTCs (log2[fold change] ≤ −0.5 and false discovery rate ≤ 0.05; Fig. 3c and Supplementary Table 2). Of note, we observe that most of the genes defining ZT4 and ZT16 expression signatures are consistently found highly upregulated (that is, fold change) in all types of CTC, yet statistical significance is highest in CTC clusters and CTC–WBC clusters (Extended Data Fig. 6b,c). This isconsistent with a higher variability and higher dropout rate expected in single-cell samples. Gene set enrichment analysis (GSEA) of genes upregulated during ZT4 and ZT16 highlights a highly consistent activity of pathways that support mitosis and cell division during ZT4 (adjusted P value ≤ 0.0001), mirrored by pathways that support ribosomal biogenesis and translation of genes during ZT16 (adjusted P value ≤ 0.0001; Fig. 3d,e, Extended Data Fig. 6d,e and Supplementary Table 3). These findings are consistent with prototypical gene expression timing in eukaryotic cells (that is, comprising recurring ribosome biogenesis and gene translation phases followed by the expression of cell cycle progression genes and the execution of cellular division within a 24-h time frame)13. Of note, in human CTCs isolated from patients with breast cancer during the active (10:00 am) and rest (4:00 am) phases of the same day, we confirm the same pattern of gene expression and pathway activity as observed in mouse models (Fig. 3f and Extended Data Fig. 6e). Gene expression changes ascribed to cell division and translation and inferred at time points ZT4 and ZT16, respectively, are also consistently observed across different time points during the rest and active phases (Fig. 3g and Extended Data Fig. 6f). Given the short half-life of CTCs, we reasoned that oscillatory changes in cellular proliferation could also be visible at the level of the primary tumour when analysed at different times. Accordingly, when staining for the proliferation marker Ki67 in tumours from the NSG-CDX-BR16 and the NSG-LM2 models, along with their CTCs, we find a marked upregulation of Ki67 during the rest phase (Fig. 3h and Extended Data Fig. 7) and consistent with the timing of highest CTC intravasation and expression of mitosis-related genes. Together, molecular gene expression analysis of CTCs from patients and mouse models, isolated during the rest and active phases, highlights very distinct gene expression patterns. During sleep, gene expression is dominated by cell division and mitosis genes, whereas during the active phase, we observe high ribosome biogenesis activity. This oscillatory proliferation timing is observed not only in CTCs but also in the primary tumour, suggesting this as a general phenomenon occurring in breast cancer cells during disease progression.

Fig. 3: Rest-phase CTCs are highly proliferative.
figure 3

a, Illustration of the experimental design. CTCs are collected at ZT4 or ZT16, and then directly processed for scRNA-seq. b, Plot showing the principal components PC4 and PC5 of gene expression in CTCs from NSG-CDX-BR16 mice. The upper and right panels show the density of values for the active (blue) and rest (red) phases. c, Heatmap showing row-normalized abundance of differentially expressed genes between the rest and active phases in CTCs from NSG-BR16-CDX mice. N represents the number of upregulated genes. d, Heatmap showing the pairwise similarity of enriched gene sets in CTCs of the rest and active phases from NSG-CDX-BR16 mice. The heatmaps on the right represent the adjusted GSEA P value and normalized enrichment score (NES). e, Plots comparing the GSEA results (NES and P value) in the NSG-CDX-BR16 (left) and NSG-LM2 (right) models for gene sets shown in d. f, Left: illustration of the experimental design. CTCs were collected from patients with breast cancer during the rest (04:00 am) and active (10:00 am) phases, and then directly processed for scRNA-seq. The bar on the top represents environmental light (white) and dark (black) phases. Right: GSEA results in patient CTCs as described in e. g, Average gene set variation analysis (GSVA) score for translation (yellow, n = 5) and cell division (blue, n = 17) gene sets in CTCs from the NSG-LM2 time kinetic experiment (ZT0 n = 3, ZT4 n = 3, ZT12 n = 3, ZT16 n = 3, ZT20 n = 3). The background represents environmental light (white) and dark (grey) conditions. h, Left: representative immunofluorescence images of Ki67 (green) and Pan-CK (red) in primary tumours from NSG-LM2 mice dissected at ZT0, ZT4, ZT12, ZT16, ZT20 (n = 3; P = 0.002). Nuclei are stained with DAPI (blue). Scale bars, 100 μm. Right: plot showing the intensity of Ki67 in tumours of NSG-LM2 mice during different time points. The data are presented as mean ± s.e.m.; **P < 0.01 by unpaired two-sided t-test. n represents the number of biologically independent mice. The human figure, sun and moon were created with BioRender.com.

Source data

Full size image

Regulators of CTC intravasation

Mechanistically, to identify the master regulators of circadian-rhythm-driven CTC generation and proliferation, we took several approaches. We first investigated whether the expression of prototypical circadian clock genes in cancer cells changed between the rest (ZT4) and active (ZT16) phase. Similarly to previous reports highlighting disruption of rhythmicity in the expression of circadian clock genes in cancer14,15, we could not detect differential expression in CTCs or primary tumour cells (Extended Data Fig. 8a–c), whereas we could confirm rhythmicity in non-neoplastic tissues (Extended Data Fig. 8c). We next investigated whether oscillations in CTC counts could be explained by changes in interstitial fluid pressure, differential interplay with immune cells and damage due to different haemolysis rates during different phases of the circadian rhythm. Notably, we found no difference in YAP and TAZ expression levels or localization (as sensors of interstitial pressure), in the abundance of circulating or tumour-infiltrated immune cells, or in apoptotic levels of CTCs during the rest (ZT4) versus active (ZT16) phase (Extended Data Fig. 9a–f). Last, we interrogated our RNA-seq data from CTCs of xenografts and patients to determine the expression levels of receptors for well-known circadian-rhythm-regulated hormones, growth factors or molecules, reasoning that the daily oscillation of their systemic levels could affect cancer cells in a time-dependent fashion. We evaluated the expression of 63 receptors for circadian-rhythm-regulated candidates (Supplementary Table 4), looking for those with a high level of expression in most CTCs and independently of a specific time point (that is, stable expression over time and activity proportional to the levels of their ligand). With these criteria, we find that expression of the glucocorticoid receptor, androgen receptor and insulin receptor is highly represented among single CTCs, CTC clusters and CTC–WBC clusters (Extended Data Fig. 10a), suggesting the involvement of their ligands in time-point-driven CTC generation and proliferation (Fig. 4a). To test this hypothesis, we first treated tumour-bearing mice with either dexamethasone (specific glucocorticoid receptor ligand) or testosterone (the main androgen receptor ligand), both found at high levels in physiological conditions at the onset of the active phase16,17 (that is, when CTC numbers are low). Accordingly, both a single treatment with dexamethasone at 4 mg kg−1 during the rest phase (ZT2) and implantation of a testosterone pellet (slow, continuous release) resulted in a marked reduction in single CTCs, CTC clusters and CTC–WBC clusters when sampled at the peak time during the rest phase (ZT4; Fig. 4b,c and Extended Data Fig. 10b–f). Of note, whereas treatment with dexamethasone or testosterone did not affect primary tumour size (Extended Data Fig. 10c,d), we observed a reduction in the metastatic burden of testosterone-treated mice (Extended Data Fig. 10e), consistent with a prolonged suppression of CTC generation alongside the continuous testosterone release from the pellet. Further, given the well-established link between insulin stimulation and subsequent cell growth and division13,18, we investigated whether insulin oscillations (with insulin being higher during the active phase following glucose intake in physiological conditions19) could also influence the proliferation timing and intravasation of breast cancer cells (that is, whether insulin stimulation during the rest phase could invert the dynamics of CTC release and proliferation). To address this, following tumour development, we treated tumour-bearing mice daily (during the rest phase, at ZT3) for 1 week with 0.7 U kg−1 of insulin and 1 g kg−1 glucose, and quantified CTC abundance during the rest and active phases, respectively. Consistently, we find that insulin treatment during the rest phase decreases CTC intravasation at ZT4, and increases it at ZT16 (Fig. 4d), with no significant changes in primary tumour volume (Extended Data Fig. 10g,h). Of note, treatment with insulin during the rest phase also inverts the proliferation cycle of breast cancer cells (that is, it decreases proliferation during the rest phase and it increases it during the active phase; Fig. 4e). Together, our findings indicate that proliferation and intravasation of breast cancer cells are dictated by daily oscillations in key circadian-rhythm-regulated hormones, whose action influences breast cancer metastasis dynamics.

Fig. 4: Dexamethasone, testosterone and insulin regulate CTC intravasation.
figure 4

a, Schematic illustration of the expression of three key receptors (insulin receptor, glucocorticoid receptor and androgen receptor) on breast cancer cells, activated by their circadian-rhythm-regulated ligands (insulin, glucocorticoid and testosterone, respectively) during the active phase. b, Box plots showing the distribution of the number of single CTCs, CTC clusters and CTC–WBC clusters isolated from mice treated with dexamethasone (4 mg kg−1) or vehicle (0.03% dimethylsulfoxide; control) at ZT4 (n = 4; P = 0.0286 for all). c, Box plots showing the distribution of the number of single CTCs (P = 0.0159), CTC clusters (P = 0.0317) and CTC–WBC clusters (P = 0.0159) isolated from testosterone-treated (n = 5) or untreated (control) mice (n = 4) at ZT4. d, Box plots showing the distribution of the number of single CTCs (ZT4 P = 0.0002; ZT16 P = 0.0007), CTC clusters (ZT4 P = 0.0002; ZT16 P = 0.0153) and CTC–WBC clusters (ZT4 P = 0.0011; ZT16 P = 0.0053) isolated from mice treated with insulin (0.7 U kg−1) or vehicle (phosphate-buffered saline; control) at ZT4 or ZT16 (n = 8, except insulin-treated mice at ZT16 with n = 6). e, Representative immunofluorescence images of Ki67 (green) and Pan-CK (red) in primary tumours from control or insulin-treated mice, dissected at ZT4 or ZT16 (n = 4 except control mice at ZT4 with n = 3). Nuclei are stained with DAPI (blue). Scale bars, 100 μm. f, Pie charts showing the mean percentage of Ki67 intensity in tumours shown in e. The NSG-LM2 model was used for all treatments. For b–d, the white and black bars on the horizontal axis represent environmental light and dark conditions, respectively; the grey arrows indicate treatment timing; the centre lines in the box represent the median; the box limits represent the first and third quartiles; the extremes of the whisker lines represent the minimum and maximum observed values. *P < 0.05, **P < 0.01, ***P < 0.001 by two-sided Mann–Whitney test. n represents the number of biologically independent mice.

Source data

Full size image

Discussion

Our data provide new insights into the processes that dictate the generation of metastasis-competent CTCs. Previous reports have suggested a role of the circadian rhythm in tumorigenesis, with most involving epidemiological studies and linking disruption of the circadian clock to accelerated onset of cancer15,20,21. Yet, dynamics that prospectively govern metastatic disease progression in this context remained poorly characterized. More recent studies using in vivo imaging technologies and interrogation of physiological models have highlighted various mechanisms adopted by cancer cells during the intravasation process22,23,24; however, a detailed understanding of the specific timing of CTC intravasation has been lacking. We find that, in both patients with breast cancer and mouse models, generation of CTCs is highly restricted to the rest phase, and that rest-phase CTCs are endowed with a much greater metastatic proclivity compared to active-phase CTCs. This augmented metastatic ability is conferred by high proliferation rates that occur in a time-dependent manner, and it is influenced by the action of circadian-rhythm-regulated hormones, suggesting the need for time-controlled approaches for the characterization and treatment of breast cancer. These could include the interrogation of clinical samples at highly controlled time points to minimize variability, as well as cancer treatment approaches that are tuned to be maximally effective during sleep.

Methods

Patient samples

All patients gave their informed written consent to participate in the study that took place at the University Hospital Basel under the Clinical Research Protocol (number 2020-00014) approved by the Swiss authorities (EKNZ, Ethics Committee northwest/central Switzerland) and in compliance with the Declaration of Helsinki. All patients were hospitalized and were either temporarily off-treatment (patients with stage IV disease) or before operation (patients with stage I–III disease) at the time of blood sampling. A 7.5-ml sample of peripheral blood was collected from patients with breast cancer during the rest (04:00 am) and active (10:00 am) phases of the same day in EDTA vacutainers. The time point for each sample collection was strictly followed.

Cell culture

Human CTC-derived BR16 cells were generated as previously described25 from a patient with hormone-receptor-positive breast cancer at the University Hospital Basel and propagated as suspension cultures in a humidified incubator at 37 °C with 5% O2 and 5% CO2. MDA-MB-231 LM2 human breast cancer cells (obtained from J. Massagué, Memorial Sloan Kettering Cancer Center), E0771.lmb mouse breast cancer cells (obtained from R. Anderson, Olivia Newton-John Cancer Research Institute) and 4T1 mouse breast cancer cells (ATCC) were grown in Dulbecco's modified Eagle medium (Gibco, 11330-057) supplemented with 10% fetal bovine serum (Gibco, 10500064) in a humidified incubator at 37 °C with 20% O2 and 5% CO2. LM2, BR16 and 4T1 cells were transduced with lentiviruses carrying either GFP–luciferase or mCherry–luciferase. Cell lines did not belong to the list of commonly misidentified cell lines (International Cell Line Authentication Committee) and were confirmed negative for mycoplasma contamination. Authentication is not applicable for the human CTC-derived BR16 cell line, the MDA-MB-231 LM2 human breast cancer cell line, and the E0771.lmb mouse breast cancer cell line. 4T1 mouse breast cancer cells were authenticated by Multiplexion GmbH. Finally, for the in vivo mouse immunocompetent experiments, 4T1 and E0771.lmb cells were transduced with lentiviruses carrying CD90.1.

Mouse experiments

All mouse experiments were carried out according to institutional and cantonal guidelines (approved mouse protocol number 3053, cantonal veterinary office of Basel-City; and approved mouse protocol number 33688, cantonal veterinary office of Zurich). Experimental endpoints that were allowed in our approved licence included tumour-related factors such as a maximum tumour size of 2,800 mm3 or severe ulceration, as well as appearance and behaviour features such as hunching, piloerection or decreased activity. These limits were not exceeded in any of the experiments. Sample size calculations were not predetermined, but the number of animals was chosen to comply with the 3R principles. All mice were randomized before the start of each experiment, but blinding was not carried out. NSG (NOD-scid-Il2rgnull; The Jackson Laboratory), BALB/c (Janvier Labs) and C57BL/6J (The Jackson Laboratory) female mice were kept in pathogen-free conditions, according to institutional guidelines. Bmal1-knockout mice (C57BL/6J background) were purchased and genotyped from the The Jackson Laboratory. Animals were kept in a standard light-cycle photoperiod (12 h light/12 h dark; 12:12 LD) with ZT0 defined as lights on (06.00 am) and ZT12 defined as lights off (06.00 pm). For the 20-t and 28-t cycle studies, animals were kept in 10:10 LD or 14:14 LD cycle conditions, respectively. Orthotopic breast cancer lesions were generated in 8-week-old NSG females following injection with either 1 × 106 LM2-mCherry–luciferase cells (NSG-LM2 model), 1 × 106 BR16-GFP–luciferase cells (NSG-CDX-BR16 model) or 0.5 × 106 4T1-GFP–luciferase cells (NSG-4T1 model) into the mammary fat pad. Similarly, 0.5 × 106 4T1-CD90.1 cells were orthotopically injected into the mammary fat pad of 8-week-old BALB/c female mice (BALB/c-4T1 model). Finally, 1 × 106 E0771.lmb-CD90.1 cells were orthotopically injected into the mammary fat pad of 8-week-old wild-type (BL/6-E0771.lmb model) or Bmal1-knockout (BL/6-Bmal1−/−-E0771.lmb model) mice. In all cases, breast cancer cells were inoculated in 100 μl of 50% Cultrex PathClear Reduced Growth Factor Basement Membrane Extract (R&D Biosystems, 3533-010-02) in phosphate-buffered saline (PBS). Blood draw for CTC analysis, organ dissection and IVIS bioluminescence imaging were carried out during the rest and active phases after 4.5 weeks for NSG-LM2, 4 weeks for NSG-4T1 and BALB/c-4T1, 3 weeks for BL/6-E0771.lmb and BL/6-Bmal1−/−-E0771.lmb and 5–6 months for NSG-CDX-BR16 mice. The time point for each sample collection was strictly followed. All mice were randomized before mouse experiments and blindly selected before injection. Maximal approved tumour volume was never exceeded.

Mouse treatments

A total of 1 × 106 LM2-mCherry–luciferase cells were orthotopically injected into the mammary fat pad of 8-week-old NSG female mice. Following tumour development, mice were treated with different circadian-regulated hormones on the basis of their pharmacokinetic profiles and the possibility of developing negative regulatory loops following prolonged treatment. For melatonin, treatments started 10 days after the tumour injection, when tumours started growing exponentially and CTCs were not yet detectable in peripheral blood. Mice were treated daily with melatonin (20 mg kg−1; Sigma-Aldrich, M5250-1G) alone or in combination with luzindole (5 mg kg−1; Sigma-Aldrich, L2407). Luzindole treatments were carried out 30 min before melatonin treatment, which was administrated 1.5 h before the onset of the rest phase (ZT0). Blood collection and CTC analysis were carried out at ZT0. For dexamethasone, mice were treated with dexamethasone (4 mg kg−1; Sigma-Aldrich, D1159-500MG) once 2 h before the blood collection (ZT4) to avoid the activation of the negative regulatory loop of the hypothalamus–pituitary–adrenal axis26. For testosterone, mice were injected with testosterone implants (Belma Technologies, T-M/60) 4 days before the tumour injection. Implants were kept until the day of the blood collection (ZT4). For insulin, treatments started 25 days after the tumour cell injection to avoid an effect of insulin on tumour growth. Mice were treated daily with insulin (0.7 U kg−1; Humalog) in parallel with glucose (1 g kg−1; Sigma-Aldrich, G7021) at ZT3. Blood collection and CTC analysis were carried out at ZT4 and ZT16. All treatments were administered as intraperitoneal injections in a final volume of 100 µl.

Jet-lag experiment

A total of 1 × 106 LM2-mCherry–luciferase or 0.5 × 106 4T1-GFP–luciferase cells were orthotopically injected into the mammary fat pad of 8-week-old NSG female mice. Jet lag was initiated 1 week after the tumour injection by placing the animals in altered light-cycle conditions with an 8-h light advance every 2–3 days9. Melatonin treatments in jet-lagged mice were administered daily, 1.5 h before the onset of each jet-lagged rest phase. Blood collection and CTC analysis were carried out at the onset of the rest phase.

CTC capture

For patient samples, 7.5 ml of peripheral blood was processed for microfluidic-based CTC capture within 1 h from blood draw. Using the Parsortix Cell Separation System (ANGLE), CTCs were captured in Cell Separation cassettes (GEN3D6.5) and then stained with an antibody cocktail containing EpCAM–AF488 (Cell Signaling Technology, CST5198), HER2–AF488 (BioLegend, 324410), EGFR–FITC (GeneTex, GTX11400) and CD45–BV605 (BioLegend, 304042). For mouse experiments, 0.8 ml of blood was collected through cardiac puncture and processed immediately. For the immunocompromised models, samples were stained only for CD45, as cancer cells were identified on the basis of mCherry or GFP expression. For the immunocompetent models, anti-CD45 staining was carried out in parallel with staining for CD90.1 (OX-7 clone, BioLegend, 202508) to identify WBCs and cancer cells, respectively. The number of captured CTCs, including single CTCs, CTC clusters and CTC–WBC clusters, was determined while cells were still in the cassette. CTCs were then released from the cassette in Dulbecco's PBS (Gibco,14190169) onto ultra-low-attachment plates (Corning, 3471-COR) for further downstream analysis.

Direct metastatic potential assay

A total of 1 × 106 LM2-mCherry–luciferase, LM2-GFP–luciferase or BR16-GFP–luciferase cells were orthotopically injected into the mammary fat pad of 8-week-old NSG female mice. Following tumour development, blood was collected through heart puncture at ZT4 or at ZT16, and run through the Parsortix system, and captured CTCs were released onto ultra-low-attachment plates. Using the CellCelector, an automated single-cell picking system (ALS), single CTCs, CTC clusters and CTC–WBC clusters were individually micromanipulated and then each category was injected into the tail vein of recipient NSG mice. Metastasis onset and growth rate in lungs were non-invasively monitored on a weekly schedule with the IVIS bioluminescence system. The experiment was terminated 4 months post injection of LM2-mCherry–luciferase and LM2-GFP–luciferase cells or 5 months post injection of BR16-GFP–luciferase CTCs.

Immunofluorescence staining and confocal analysis

Dissected organs and primary tumours were fixed in 4% paraformaldehyde at 4 °C overnight. After paraffin embedding, the Thermo Scientific Rotary Microtome Microm HM 355S was used to cut slices of 7 µm in thickness. Following a standard deparaffinization/antigen retrieval protocol, samples were stained for pan-CK (1:65; GeneTex, GTX27753), Ki67 (1:250; Abcam, ab15580), GFP (D5.1; 1:200; Cell Signaling Technology, 2956), mCherry (16D7; 1:300: Invitrogen, M11217) and DAPI (1 µg ml−1; Sigma-Aldrich, D9542-1MG). CTCs were cytocentrifuged onto microscope slides and fixed in 4% paraformaldehyde. Post-fixation, cells were washed with PBS, permeabilized for 5 min in 0.5% TritonX-100/PBS and blocked with 5% BSA in 0.1% Triton/PBS for 1 h before Ki67 (1:250; Abcam, ab15580), TAZ (1:100; BD Biosciences, 560235), YAP (D8H1X; 1:1000; Cell Signaling Technology, 14074) or cleaved caspase 3 (5A1E; 1:100; Cell Signaling Technology, 9664) antibody was added. Immunofluorescence imaging was carried out on a Leica SP5 confocal microscope, and images were taken using the 60× oil lens. All images were analysed by the Fiji image processing software (2.1.0/1.53c).

Flow cytometry

Mouse tumours were minced into fragments and enzymatically digested for 15 min with 2 mg ml−1 type IV collagenase plus 50 U ml−1 bovine DNase. The digested tumours were mechanically dissociated in C tubes using a GentleMACS device (Miltenyi), and then subjected to red blood cell lysis using ACK Buffer (Lonza) and immediately stained. Whole mouse blood was pelleted and red blood cells were lysed using ACK Buffer and immediately stained. For immunostaining, cells were blocked in a 2% FCS solution containing 2 mM EDTA and FcR blocking reagent (Miltenyi). Mouse peripheral blood cells were stained with the following antibodies: Alexa Fluor 594 anti-mouse Ly-6G (1:500; 1A8 clone, BioLegend, 127602), Alexa Fluor 594 anti-mouse CSF-1R/CD115 (1:200; AFS98 clone, BioLegend, 135520), APC/Cyanine7 anti-CD11b antibody (1:500; M1/70 clone, BioLegend, 101226), Brilliant Violet 421 anti-mouse CD3 (1:200; 17A2 clone, BioLegend, 100228), PE anti-mouse NKp46/CD335 (1:100; 29A1.4 clone, BioLegend, 137647), Alexa Fluor 647 anti-mouse CD49b (1:200; DX5 clone, BioLegend, 103511), Alexa Fluor 488 anti-mouse CD8a (1:200; 53-6.7 clone, BioLegend, 100723), Brilliant Violet 510 anti-mouse CD4 (1:500; GK1.5 clone, BioLegend, 100449), PE anti-mouse CD223/LAG-3 (1:100; C9B7W clone, BioLegend, 125224). Dissociated tumour cells were stained with the above-mentioned antibodies plus CD90.1 (1:500; OX-7 clone, BioLegend, 202508) to distinguish CD90.1-labelled cancer cells from the infiltrating stroma. All samples were processed on an LSR Fortessa device (BD) and further analysed with FlowJo (Tree Star).

scRNA-seq

Using the CellCelector, single CTCs, CTC clusters and CTC–WBC clusters were collected and immediately transferred into individual tubes (Axygen, 321-032-501) containing 2.5 µl RLT Plus lysis buffer and 1 U SUPERase IN RNase inhibitor (Invitrogen, AM2694). Samples were immediately frozen and kept at −80 °C until further processing. Amplified cDNA was prepared according to the Smart-seq2 protocol. Libraries were prepared using Nextera XT (Illumina) and sequenced on an Illumina NextSeq500 instrument in 75-base-pair single-read mode. This yielded a median raw sequencing depth of 1.64 million reads per sample.

RNA-seq analysis

Sequencing reads were quality trimmed with Trim Galore! (v0.6.5, https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/; parameters: –q 20 –length 20) and Cutadapt (v3.4). Quality assessment of RNA-seq data was carried out using FastQC (v0.11.4, https://www.bioinformatics.babraham.ac.uk/projects/fastqc) and FastQ Screen (v0.11.4, https://www.bioinformatics.babraham.ac.uk/projects/fastq_screen/) and visualized with MultiQC (v1.7). Trimmed reads were aligned to human (GRCh38) genome reference using STAR (v.2.7.3a; parameters: --twopassMode Basic --outSAMmapqUnique 60 --sjdbGTFfile) with splice junctions from the human GENCODE annotation (release 35). To eliminate residual contamination from mouse RNA, reads derived from xenograft models were also aligned to the mouse (GRCm38) genome reference using STAR (v.2.7.3a; parameters: --twopassMode Basic --outSAMmapqUnique 60 --sjdbGTFfile), with splice junctions from the mouse GENCODE annotation (release M25) and assigned to either human or mouse using Disambiguate (v1.0.0). Resulting BAM files were sorted by Samtools (v1.10), and the alignment quality was evaluated using RSeQC (v.2.6.4). The gene-level expression counts were computed with featureCounts (v.2.0.1; parameters: -t exon -g gene_id --minOverlap 10 -Q 10) using the human gene annotations from GENCODE (release 35). Genes present with at least 3 reads in 50% of the samples were kept for the analysis. Single-cell samples were retained for further analyses if they had at least 50,000 reads, at least 5,000 genes with non-zero expression and less than 50% of reads mapping to mitochondrial genes. For samples containing more than 1 cell (CTC clusters and CTC–WBC clusters), the minimum number of genes was set to 8,000. Read counts were normalized using the trimmed mean of M-values method implemented in the R/Bioconductor package edgeR (v3.34.1). Quality control and visualization of processed data were carried out with the help of the R/Bioconductor scater package (v1.20.1). After normalization, principal component analysis was conducted using gene expression (log2-normalized counts) of the top 500 genes with the largest biological components according to the getTopHVGs function from the R/Bioconductor package scran (v1.20.1). Selected principal components were associated with technical and biological variables using Pearson correlation. The number of PCs selected was defined by the elbow method.

Differential expression and GSEA

Differential expression was computed with the quasi-likelihood approach from the edgeR R/Bioconductor package (v3.34.1) using robust dispersion estimates. Before differential expression analysis, genes detected in less than 50% of the samples, considering the size of the smallest group of replicates, were removed from the analysis (threshold 5 counts per million). P values were adjusted for multiple comparisons using the Benjamini–Hochberg method. GSEA was conducted with the fast GSEA method implemented in the R/Bioconductor package clusterProfiler (v4.0.5). As input for GSEA, we used a list of genes ranked by fold-change and two gene set collections from the Molecular Signatures Database (MsigDB, v7.4): C2 canonical pathways and C5 GO biological process. Fast GSEA carries out a preliminary estimation of enrichment P values using a permutation test (1,000 permutations) and a secondary estimation of low P values using the multilevel algorithm with a 1 × 10−10 boundary. An adjusted P-value cutoff of 0.0001 was applied to define enriched gene sets. Only gene sets with a size between 10 and 500 genes were included in the analysis. The Jaccard coefficient was computed to measure the similarity between the enriched terms using the genes included in the GSEA leading-edge subset within each gene set. In NSG-LM2 and patients, the GSEA analysis was carried out using only the enriched terms from the NSG-CDX-BR16 model. GSVA was conducted with the R/Bioconductor GSVA package to obtain sample level enrichment scores for the same MSigDB collections evaluated in the GSEA analysis. Differences in enrichment score across the multiple time points were estimated using the moderated F statistic obtained through the empirical Bayes approach implemented in the R/Bioconductor package limma v3.48.3. Time points with fewer than three replicates were removed from this analysis.

Data analysis

Data analysis, statistical testing and visualization were conducted in Graphpad Prism (v.9.1.1), R (version 4.1.0; R Foundation for Statistical Computing) and Bioconductor (v.3.13). Illustrations were created with BioRender. The figure legends describe the statistical approach used for each analysis.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.