Technical Comments

Comment on “Circadian rhythms in the absence of the clock gene Bmal1

See allHide authors and affiliations

Science  16 Apr 2021:
Vol. 372, Issue 6539, eabf0922
DOI: 10.1126/science.abf0922


Ray et al. (Reports, 14 February 2020, p. 800) recently claimed temperature-compensated, free-running mRNA oscillations in Bmal1–/– liver slices and skin fibroblasts. We reanalyzed these data and found far fewer reproducible mRNA oscillations in this genotype. We also note errors and potentially inappropriate analyses.

Note: Although this Comment originally addressed the version published on 14 February 2020, our key points also apply to the erratum and revision published on 29 January 2021.

Circadian (~24-hour) clocks endow organisms with the fundamental capability to tell time and regulate behavior and physiology accordingly (1). BMAL1 is one of a set of transcription regulators that drive the mammalian clock and is necessary for rhythmic locomotor activity or programs of rhythmic gene expression (2). However, Ray et al. (3) reported that Bmal1–/– mice liver slices and skin fibroblasts exhibit bona fide free-running 24-hour circadian-like oscillations of the transcriptome, which they interpreted as the signature of a novel Bmal1-independent, cell-autonomous molecular oscillator.

Reexamination of the original data (3) revealed that wild-type and Bmal1–/– liver datasets were switched. Although this error has now been corrected, several key Bmal1–/– results therefore originally described wild-type slices. Despite this correction, true Bmal1–/– slice oscillations are overall of low amplitudes, most likely of artifactual origin. The mouse skin fibroblast (MSF) data show a similar trend. The data taken together therefore still do not support a Bmal1-independent, cell-endogenous molecular oscillator.

We originally plotted core clock transcripts from the provided FPKM table on GEO (Fig. 1A); the levels of direct and indirect BMAL1 targets were incompatible with canonical and published results from Bmal1–/– (4). For example, Nr1d1 levels were high rather than low, whereas Cry1 levels were low rather than high (Fig. 1A). Remapping of reads showed that the GEO raw data genotype labels were correct: The Bmal1–/– samples included reads mapping to the neomycin resistance gene (Fig. 1B), and the deleted exon was evident. It is thus unambiguous that reversed genotypes were used in the 2020 published paper [figure 3C, figure S2, C and D, and table S2 of (3)]; claims about Bmal1–/– liver slices pertain to the wild type and vice versa.

Fig. 1 Reanalysis of raw data in wild-type and Bmal1–/– liver slices reveals inverted genotypes and absence of robust high-amplitude rhythms.

(A) Transcript levels (log2 FPKM) of Nr1d1 and Cry1 are shown from the original FPKM table in wild-type (blue) and Bmal1–/– (red) liver slices. (B) Only Bmal1–/– (KO) GEO samples show raw sequencing reads mapping to the neomycin cassette. However, the original FPKM table contained inverted genotype labels. (C) Transcript levels (log2 CPM) of Nr1d1 and Cry1 after remapping and quantification of the GEO raw data. Genotypes are inverted relative to (A). (D) Transcript levels of Per1 and Arntl genes are shown for the wild type (blue) and Bmal1–/– (red) after reassigning correct genotypes. (E) With correct genotypes, 24-hour rhythmicity as detected by JTK is higher in the KO than in the wild type (WT). No genes pass q < 0.05 in WT. (F) The oscillatory amplitudes of the detected rhythms (P < 0.05) are a factor of <2 [unlike figure S3A of (3)]. (G) The corresponding gene patterns are strongly correlated, showing two anticorrelated groups of genes (k-means clustering with two centroids). Data were filtered before analysis to remove genes with a mean expression of <3 FPKM, or with more than six (24-sample datasets) or four (16-sample datasets) zero values. Data were log2-transformed. For simplicity, because this was sufficient to make our points, we used one of the rhythmicity methods used in (3), JTK-cycle, with the following parameters: 24-hour period, P < 0.05. Genome-wide q-value correction was done with the BH method (9). Amplitude was calculated using JTK-cycle.

Although mean levels of core clock transcripts now appear correct in the properly assigned Bmal1–/– (Fig. 1C), their temporal patterns are strange. For example, Bmal1 should be rhythmic only in the wild type but shows a very similar and weakly rhythmic pattern in both genotypes (Fig. 1D). Equally surprising, most other clock transcripts showed no rhythms in the wild type but exhibited 24-hour–like fluctuations in Bmal1–/– (Fig. 1D). Moreover, these clock transcript fluctuations showed peaks or spikes at similar times in this deletion genotype, which suggests that these fluctuations are not driven by residual core clock circuit interactions in Bmal1–/–. The lack of rhythms in wild-type samples also suggests something technically flawed in the liver experiments.

In addition, genome-wide 24-hour periodicity was now strikingly higher in the (real) Bmal1–/– samples than in wild-type samples (e.g., no wild-type genes passed the q < 0.05 JTK filter) (Fig. 1E). With a more relaxed filter (P < 0.05), however, all oscillatory amplitudes in both genotypes were low (factor of <2; Fig. 1F) and unlike the large amplitudes in figure S3A of (3). Strikingly, these weakly rhythmic fluctuations were all related and show only two anticorrelated patterns in both genotypes [Fig. 1G; comparison to figure S2, C and D, of (3) clearly illustrates the genotype inversion]. The lack of sole dependence on an internal oscillator is also buttressed by the lack of progressive amplitude damping after the single initiating pulse of dexamethasone. Wild-type amplitudes even increase substantially on the third day, nearly 5 days after the initiating pulse (Fig. 1G).

These considerations suggest that the liver slice fluctuations could have been influenced, or even caused, externally—for example, by medium changes or other unintended daily interventions. The identification of ETS factors as possible drivers of wild-type and Bmal1–/– rhythms could therefore reflect such external causes even though these factors have been previously linked to diurnal transcript rhythms (5, 6).

Ray et al. also presented 12 datasets for MSFs, six for the wild type and six for Bmal1–/–: an MSF dataset [37°C; figure 2, E and F, of (3)], three temperature datasets [one dataset each at 27°, 32°, and 37°C; figure 2, A to D, of (3)], an AM dataset [37°C; figure 2, E and F, of (3)], and a PM dataset [37°C; figure 2, E and F, of (3)]. All cells were synchronized with dexamethasone, with the AM and PM synchronization offset by 12 hours.

The absence of transcript oscillation criteria in the methods section of (3) hampered our attempts to illustrate cycling properties of those datasets. So we chose to set our criteria such that the number of cycling transcripts was similar but somewhat fewer than reported [figure S1C of (3)]; thus, our JTK criteria are presumably similar or slightly more stringent than used in (3). Nonetheless, the number of rhythmic transcripts varied widely in each dataset (Fig. 2A). Moreover, they almost all (98.4%) had low amplitudes, less than 0.5 (log2), and less than 5% met the stricter and often used q < 0.05 criterion (genome-wide Benjamini-Hochberg).

Fig. 2 In Bmal1–/– mouse fibroblasts, very few genes are consistently rhythmic and exhibit free-running and temperature-compensated rhythms.

(A) Rhythmic transcripts were identified in four Bmal1–/– MSF experiments performed at 37°C. Only 38 genes were rhythmic in three of the four datasets. (B) The 38 genes from (A) were examined in the 27°, 32°, and 37°C datasets. Only 16 of these 38 genes were rhythmic at all temperatures and were therefore likely temperature-compensated. (C) The gene expression patterns of the 16 temperature-compensated, consistently rhythmic genes from (B) were examined in Bmal1–/– AM (blue) or PM (yellow) datasets. Seven genes were rhythmic in both datasets; however, only one shows antiphase AM/PM rhythms and this is limited to day 1. Gene expression is represented as FPKM (log2). Time (CT) is shown on the x axis. (D) Most of these temperature-compensated, Bmal1–/– rhythmic genes are also rhythmic in at least one wild-type dataset. Data preprocessing and rhythmicity analysis was as in described in Fig. 1. Intersectional analyses were performed with UpsetR (10). In both Figs. 1 and 2, we did not analyze non–24-hour periodicities or other temporal patterns.

We carefully examined cycling transcripts from the four Bmal1–/– datasets collected under similar conditions, namely MSFs cultured at 37°C: MSF, AM, PM, 37. These rhythmic transcripts had minimal overlap (Fig. 2A), as most were in only one of the four datasets, whereas only 38 genes (~1%) were in at least three of four datasets. Notably, only 1 of the 11 genes highlighted in (3) was among these 38 consistently rhythmic genes [figure 2, E and F, of (3)]. This low overlap cannot be explained by a particular cycling algorithm, in this case JTK, because a simple plotting of gene expression also revealed a lack of consistency in the samples. There is a comparable lack of reproducibility in the wild-type data; only 59 genes (1.1%) are in three of the four wild-type datasets.

We tested whether the 38 consistently rhythmic Bmal1–/– transcripts at 37°C were also identified at 27°C and 32°C. This criterion was met by 16 of the 38 transcripts, or 42% (Fig. 2B), indicating that these 16 oscillating transcripts may indeed be temperature-compensated.

We examined in more detail these 16 transcripts in the AM/PM datasets. Only seven were rhythmic in both (Fig. 2C), of which only one showed the expected antiphasic pattern from oscillator synchronization 12 hours apart by dexamethasone; moreover, this pattern was only visible on the first day. Nearly all of the other six rhythmic transcripts showed a similar mid-morning phase in both AM/PM datasets, which suggests that these rhythms may be externally driven rather than free-running.

Although it was implied that most Bmal1–/– rhythmic transcripts are specific for this genotype (3), we identified 14 of the 16 candidate temperature-compensated transcripts in the wild-type cycling data (Fig. 2D). Even the six prime examples of Bmal1-independent temperature compensated cyclers [figure 2D of (3)] were also rhythmic in at least one wild-type dataset.

In summary, most Bmal1–/– MSF rhythmic transcripts are of low amplitude, low significance, and little consistency across datasets; notably, MSF rhythmic transcripts are unchanged in the revised version. Nonetheless, there may be a small number of Bmal1–/– transcripts that meet rhythmic criteria. They are not, however, convincingly free-running. It is also notable that there are no core clock transcript oscillations in the wild-type liver slice data even after reassigning the genotypes, whereas some of them showed periodic-like but atypically phased patterns in Bmal1–/–. Moreover, and unlike previous liver slice oscillations (7, 8), no damping was observed. Our analysis therefore argues against the claims of a Bmal1-independent and cell-autonomous oscillator.

Data source: Note that the processed data files on GEO (GSE111696) were updated on 11 September 2020 with the genotype assignments corrected, after the initial version of this manuscript was sent to the authors on 27 August 2020.


Acknowledgments: We thank J. Hogenesch and R. Braun for sharing their independent analyses of the liver slice and fibroblast datasets. Funding: M.R. and K.C.A. were funded by the Howard Hughes Medical Institute. Research in the Naef lab is supported by Swiss National Science Foundation grant 310030_173079 and by EPFL. Author contributions: Conceptualization and writing, K.C.A., C.G., F.N., and M.R.; formal analysis and visualization, K.C.A. and C.G.; funding acquisition, F.N. and M.R. Competing interests: None. Data and materials availability: Original and updated datasets from (3) as well as supplementary analysis and scripts are available at

Stay Connected to Science

Navigate This Article