Abstract

   Regulatory T cells (T[regs]) in non-lymphoid organs provide critical
   brakes on inflammation and regulate tissue homeostasis. While so-called
   “tissue-T[regs]” are phenotypically and functionally diverse, serving
   to optimize their performance and survival, upregulation of pathways
   related to circadian rhythms is a feature they share. Yet, the diurnal
   regulation of T[regs] and its consequences are controversial and poorly
   understood. Here, we profiled diurnal variations in visceral-adipose
   tissue (VAT) and splenic T[regs] in the presence and absence of
   core-clock genes. VAT, but not splenic, T[regs] upregulated their
   cell-intrinsic circadian program, and exhibited diurnal variations in
   their activation and metabolic state. BMAL1 deficiency specifically in
   T[regs] led to constitutive activation and poor oxidative metabolism in
   VAT, but not splenic, T[regs]. Disruption of core-clock components
   resulted in loss of fitness: BMAL1-deficient VAT T[regs] were
   preferentially lost during competitive transfers and in heterozygous
   BMAL1-deficient females. After 16 weeks of high-fat-diet feeding, VAT
   inflammation was increased in mice harboring BMAL1-deficient T[regs],
   and the remaining cells lost the transcriptomic signature of bona fide
   VAT T[regs]. Unexpectedly, VAT T[regs] suppressed adipocyte lipolysis,
   and BMAL1 deficiency specifically in T[regs] abrogated the
   characteristic diurnal variation in adipose-tissue lipolysis, resulting
   in enhanced suppression of lipolysis throughout the day. These findings
   argue for the importance of the cell-intrinsic clock program in
   optimizing VAT T[reg] function and fitness.

ONE-SENTENCE SUMMARY:

   Visceraladipose tissue Tregs turn up cell-intrinsic circadian rhythms
   to promote cellular fitness and regulate diurnal lipolysis

INTRODUCTION

   Regulatory T cells (T[regs]), particularly those expressing the
   transcription factor (TF) Foxp3, play an essential role in restraining
   overexuberant immune responses. Although early studies focused on
   T[regs] circulating through lymphoid organs, it is by now well
   established that T[regs] are located in non-lymphoid tissues as well.
   These so-called “tissue T[regs]” have unique transcriptomes, T cell
   receptor (TCR) repertoires and growth/survival-factor dependencies that
   enable them to thrive in particular tissue environments, where they
   help maintain homeostasis ([36]1). Even though tissue T[regs] exhibit a
   diversity of phenotypes and functions, they also share transcriptomic
   modules that set them apart from their lymphoid-organ counterparts
   ([37]1, [38]2). Intriguingly, pathways related to circadian rhythms are
   amongst the upregulated pathways shared most prominently, raising the
   question of whether tissue T[regs] are subject to distinct diurnal
   regulation.

   Mammalian circadian rhythms evolved to anticipate and adapt to diurnal
   variations in the environment. A set of “core-clock” genes encodes the
   molecular drivers of this process, the expression, activity and
   degradation of which form an autoregulatory feedback loop lasting
   approximately 24 hours ([39]3). Heterodimeric complexes of BMAL1:CLOCK
   drive a core component of circadian transcription, including genes
   encoding its inhibitors, PER and CRY, and genes specifying its
   transcriptional regulators, the REV-ERBs and ROR [reviewed in ([40]3)].
   These molecular oscillators are stratified into a hierarchy of central
   and peripheral pacemakers according to whether they are localized in
   cells of the suprachiasmic nucleus of the brain or in cells of
   peripheral tissues. Peripheral clocks interpret synchronization signals
   from the central pacemaker, which is primarily synchronized by light,
   and from other environmental cues, like feeding and temperature, to
   generate cell-intrinsic rhythmicity ([41]4–[42]6).

   Our understanding of circadian regulation of T cells and of the
   functions of their core-clock genes is just beginning to emerge.
   Rhythmic T cell migration between the circulation and lymphoid organs
   is the best-described circadian behavior: in mice, peak numbers of
   lymphocytes are found in circulation around zeitgeber time (ZT) 5 (ZT0
   signalling onset of the light period), and in lymphoid organs around
   ZT13 ([43]7–[44]10). Additionally, naïve murine T cells isolated during
   the day are more likely to differentiate into T helper 17 (Th17) cells
   than those taken at night ([45]11). However, core-clock genes were
   reported to be dispensable for T cell differentiation and for their
   responses to bacterial and viral infection ([46]12), and expression of
   core-clock genes in splenic T[regs] seemed to lack rhythmicity
   ([47]13). Thus, the contexts wherein T[regs] are subject to circadian
   regulation, as well as the consequences of such regulation, remain in
   question.

   We decided to investigate the importance and function of circadian
   rhythms in the paradigmatic tissue-T[reg] population found in
   epididymal visceral-adipose tissue (VAT) of lean mice ([48]14). Thymic
   T[regs] seed VAT in the first weeks of life, where they proliferate
   indolently until they dominate the CD4^+ T cell pool around 20–30 weeks
   of age ([49]15). Their accumulation and survival depend on a unique
   transcriptome largely driven by the critical transcriptional regulator
   of adipocyte differentiation, PPARγ, as well as on their TCR and
   cytokines such as IL-33 ([50]15–[51]18). Importantly, VAT T[regs]
   regulate local and systemic metabolic tenor by controlling both
   immunologic and non-immunologic cells ([52]14, [53]16, [54]17).

   Combining population-level and single-cell transcriptomics, adoptive
   transfer experiments, genetic ablation, obesity-inducing challenges,
   lipolysis assays, and cytofluorimetry-based single-cell metabolic
   assays, we uncovered a role for the cell-intrinsic clock in promoting
   the fitness of VAT, but not splenic, T[regs], regulating, for example,
   adipose-tissue lipolysis. The heightened circadian regulation in
   T[regs]operating in non-lymphoid-tissue environments re-emphasizes
   their adaptability and striking divergence from their lymphoid-organ
   counterparts.

RESULTS

Elevated expression of core-clock genes in VAT T[regs]

   Since pathways related to circadian rhythms appeared to be upregulated
   in tissue T[regs] ([55]1, [56]2), we employed published microarray data
   to compare expression of the core-clock genes in VAT ([57]16), muscle
   ([58]19) and colonic lamina propria ([59]20) T[regs] with that of their
   lymphoid-organ counterparts. Core-clock genes, across all major arms of
   the cell-intrinsic clock machinery ([60]Fig. 1A), were expressed at
   higher levels in T[regs] from non-lymphoid than lymphoid tissues
   ([61]Fig. 1B and [62]fig. S1A). Focusing on VAT, we found much less
   evident up-regulation of the core-clock genes in conventional CD4^+ T
   cells (T[convs]) than in T[regs] ([63]Fig. 1C and [64]fig. S1B). Data
   from a T[reg] adoptive-transfer system confirmed that the VAT
   microenvironment up-regulated core-clock gene expression: transfer of
   T[regs] from a VAT-T[reg] TCR-transgenic (tg) mouse line ([65]18) into
   non-tg recipients resulted in accumulation of donor T[regs] with
   elevated expression of core-clock genes in VAT in comparison with
   spleen ([66]Fig. 1D and [67]fig. S1C). Thus, expression of core-clock
   genes was elevated in tissue Tregs, VAT T[regs] in particular, in
   comparison with their splenic counterparts, an increase induced by the
   tissular microenvironment.

Fig. 1: Upregulation of core-clock gene expression in VAT T[regs].

   Fig. 1:
   [68]Open in a new tab

   (A) Schematic of the cell-intrinsic circadian rhythms executed by
   core-clock gene products. (B) Core-clock gene expression by tissue
   T[regs] from published microarray data: VAT ([69]16), skeletal muscle 4
   days after cardiotoxin injury ([70]19), and colonic lamina propria
   ([71]20) T[regs] in comparison with lymphoid-organ T[regs] from the
   same mice. (C) Core-clock gene expressions in VAT-T[reg] or -T[conv]
   cells, from published microarray data on retired male breeders ([72]28)
   in comparison with lymphoid-organ T[regs] or T[convs] from the same
   mice. (B, C) p: * ≤ 0.05, ** ≤ 0.01, *** ≤ 0.001 according to Student’s
   t-test comparing expression in tissue versus paired lymphoid-organ
   T[regs]. (D) Core-clock gene expression at zeitgeber 0 (onset of light
   phase) in donor T[regs] 9 weeks after transfer in recipients’ VAT and
   spleen. T[regs] from Foxp3-Cre.YFP^+CD45.1^+TCR-tg^+ donors were
   transferred into Foxp3-Cre.YFP^+CD45.1/2^+ recipients. FDR: * ≤ 0.05,
   ** ≤ 0.01, *** ≤ 0.001 according to the quasi-likelihood F-test by
   EdgeR. n ≥ 3 per condition. VAT, visceral adipose tissue; Spl, spleen;
   AU, arbitrary units; TCR, T cell receptor; tg, transgenic.

Rhythmic expression of VAT-T[reg] core-clock genes

   We next assessed diurnal expression profiles of the core-clock genes by
   population-level RNA sequencing of cytofluorimetrically sorted VAT
   T[regs]. Plotting gene expression versus time, where zeitgeber (ZT) 0
   denotes onset of the light phase and ZT12 the dark phase, revealed that
   expression of most of the core-clock genes (such as Bmal1, Per1-3, and
   Nr1d1,2) exhibited significant rhythmicity in VAT T[regs] ([73]Fig.
   2A). The times of peak core-clock gene expression in VAT T[regs] were
   similar to those previously reported for liver cells ([74]21, [75]22),
   reinforcing the notion that peripheral tissues are usually
   synchronized. Expression of a few core-clock genes, specifically Cry1-2
   and Rorα, lacked rhythmicity in VAT T[regs], which is not surprising
   given the report that not all core-clock genes are rhythmically
   expressed in peripheral tissues ([76]23). In stark contrast,
   rhythmicity of core-clock genes in splenic T[regs] appeared muted and
   was not statistically significant, with the exception of Nr1d2
   ([77]Fig. 2A).

Fig. 2: Rhythmic expression of core-clock genes and other loci in VAT
T[regs].

   Fig. 2:
   [78]Open in a new tab

   Retired B6 male breeders (>35 weeks old) were sacrificed every 4 hours
   during a 12:12 light:dark cycle. ZT0 signals onset of the light period
   (see bars below relevant panels). T[regs] were sorted as CD4^+CD25^+.
   (A) Expression of core-clock genes over time. (B) Heatmap of
   rhythmically expressed genes in VAT T[regs]. (C) Distribution of the
   time of peak expression for rhythmically expressed genes in VAT
   T[regs]. (D) Fold-change of maximum/minimum gene expression over time
   in rhythmically expressed genes. (E) Ingenuity Pathway Analysis of
   rhythmic genes. (F, G) Expression profile of selected rhythmicaly
   expressed genes in the TCR signaling pathway (F) and cholesterol
   biosynthesis pathways (G) from panel E. Circadian rhythmicity
   determined by JTK_Cycle. p: * ≤ 0.05, ** ≤ 0.01, *** ≤ 0.001. n = 3 for
   each time point from ZT0 to ZT16, n = 2 for ZT20. ZT, Zeitgeiber time.
   TCA, tricarboxylic acid; VAT, visceral adipose tissue; Spl, spleen; AU,
   arbitrary units; TCR, T cell receptor. Flow-cytometric gating
   strategies for all quantified or isolated cell-types can be found in
   [79]fig. S7.

   Looking beyond the core-clock genes to loci they might regulate, we
   identified over 600 genes with rhythmic expression in VAT T[regs] after
   filtering out genes with low expression or a period of rhythmicity less
   than 16 hours ([80]Fig. 2B, [81]Table S1). Peak expression of most of
   these genes was near either ZT0 or ZT12 ([82]Fig. 2C), with about a
   1.5-to 2-fold change in expression between their highest and lowest
   expression ([83]Fig. 2D). Pathway analysis on the rhythmically
   expressed genes in VAT T[regs] revealed an enrichment in pathways
   related to protein ubiquitination, sirtuin signalling, cholesterol
   biosynthesis, T cell activation and mitochondrial metabolism ([84]Fig.
   2E). Transcripts encoding molecules involved in TCR signaling and
   cholesterol biosynthesis peaked mostly around ZT0/24 ([85]Fig. 2F,
   [86]G). Thus, clock genes were rhythmically expressed in VAT Tregs but
   generally not in splenic Tregs.

Core-clock-gene products enhance the fitnessof VAT T[regs]

   To assess the functional importance of the T[reg]-intrinsic clock, we
   abrogated expression of BMAL1 (which drives the forward loop of the
   clock and thereby promotes rhythmicity) specifically in T[regs] by
   crossing Foxp3-Cre.YFP mice with a Bmal1^flox/flox strain ([87]24). The
   mutant offspring will hereafter be referred to as T[reg]^Bmal1Δ mice
   (Foxp3-Cre.YFP^+Bmal1^flox/flox) as opposed to the T[reg]^Bmal1WT
   controls (Foxp3-Cre.YFP^+Bmal1^wt/wt). We confirmed that the mutant
   T[regs] lacked the floxed exon 8 ([88]fig. S2A) by inspecting pile-ups
   generated from population-level RNA sequencing (RNA-seq). Furthermore,
   we verified the specificity of deletion under the Foxp3-Cre driver by
   quantitative RT-qPCR, and observed that Bmal1 expression was
   substantially decreased in T[regs] but remained unaffected in CD4^+
   T[convs] or CD8^+ T cells from the spleen ([89]fig. S2B). As expected,
   loss of BMAL1 led to significant alterations in the expression of other
   core-clock genes at ZT0 in sorted VAT T[regs]. Specifically, there was
   elevated expression of some genes, such as Cry1 and Npas2, and reduced
   expression of others, such as Nr1d1 and Nr1d2 ([90]Fig. 3A).

Fig. 3: Loss of competitive fitness in VAT T[regs] in the absence of BMAL1 or
REV-ERBα.

   Fig. 3:
   [91]Open in a new tab

   (A, B): VAT T[regs] were isolated at ZT0 from 18-to-30-week-old male
   T[reg]^Bmal1WT (WT) or Treg^Bmal1Δ (Δ) mice. (A) Expression of
   core-clock genes at 18 weeks of age determined by RNA-seq analysis. (B)
   Fraction of VAT T[regs] across age groups at ZT0 (left) and at ZT0 vs
   ZT12 (right). Left: pooled data from 4 independent experiments; right:
   representative data of 3 independent experiments. n ≥ 2 per group per
   experiment. (C-G) Competitive transfer of a 1:1 mix of T[regs] from
   6-to-8-week-old male mice of the indicated genotype into
   8-to-10-week-old recipients, as schematized in (C). (D, E) Fractions of
   Bmal1 WT and Bmal1 Δ donor cells expressed as a percentage of total
   T[regs] (D) or of total donor T[regs] (E) in the recipients’ VAT. (F)
   similar to panel E but from the recipients’ spleen 12 weeks after
   transfer. (G) Fraction of donor cells in cycle (Ki67^+). For 9 and 12
   weeks after transfer, representative data of at least 3 experiments, n
   ≥ 3 per experiment. (H) T[regs] in gonadal fat of
   Foxp3-Cre^+/−.YFP^+/−Bmal1^flox/flox female heterozygotes. Fraction of
   Cre^− versus Cre^+ T[regs]; left: representative flow plot; middle:
   frequency; right: cell number. (I-K) Competitive transfer of a 1:1 mix
   of T[regs] from the indicated genotypes into Foxp3-Cre.YFP^+CD45.1/2^+
   recipients and assessed 12 weeks after transfer. Data pooled from 2
   independent experiments. (I,J) Fractions of Nr1d1 WT and Nr1d1 Δ donors
   expressed as a percentage of total T[regs] (I) or of total donor
   T[regs](J) in the recipient’s VAT. (K) Same as panel I, but from the
   recipients’ spleen. (p: * ≤0.05, ** ≤0.01, *** ≤ 0.001 according to
   Student’s t-test for the transfer experiments; quasi-likelihood F-test
   by EdgeR for RNA-seq data. wks, weeks; VAT, visceral adipose tissue;
   Spl, spleen; AU, arbitrary units; TCR, T cell receptor; tg, transgenic.
   Flow-cytometric gating strategies for all quantified or isolated
   cell-types can be found in [92]fig. S7.

   Since the VAT-T[reg] compartment in adult mice receives little
   contribution from circulating T[regs] ([93]15), we expected diurnal
   variation in the circulation of lymphoid-organ T cells ([94]8) to have
   little impact on the VAT-T[reg] pool size. Accordingly, the number and
   fraction of VAT T[regs] remained similar at ZT0 and ZT12, and between
   T[reg]^Bmal1WT and T[reg]^Bmal1Δ mice at ZT0 ([95]Fig. 3B and [96]fig.
   S2C). The frequency of T[regs] in the spleen also lacked diurnal
   variation and appeared normal in the absence of BMAL1 ([97]fig. S2D).

   We suspected that the T[reg]-intrinsic clock might promote fitness
   without impacting accumulation at homeostasis. Thus, we assessed the
   fitness of BMAL1-deficient T[regs] in a competitive adoptive-transfer
   system. When polyclonal lymphoid-organ T[regs] are transferred under
   this protocol, they do not accrue in VAT; however, Tregs from the
   VAT-T[reg] TCR-tg mouse line mentioned above do accumulate after
   transfer, reflecting efficient recognition of their TCR’s ligand
   ([98]18). We generated Foxp3-Cre.YFP^+.CD45.1^+.TCR-tg^+Bmal1 WT and
   Foxp3-Cre.YFP^+.CD45.2^+.TCR-tg^+Bmal1 Δ donor strains. Bmal1 WT and
   Bmal1 Δ T[regs] were transferred at a 1:1 ratio into
   Foxp3-Cre.YFP^+.CD45.1/2^+ recipients, and donor-T[reg] accumulation
   was quantified in the VAT and spleen at various timepoints thereafter
   ([99]Fig. 3C). Donor T[regs] were first detected in VAT 6 weeks after
   transfer, when the Bmal1 WT and Bmal1 Δ cells were present at an equal
   ratio. By 7.5 to 9 weeks, the Bmal1 Δ population had undergone a
   preferential expansion, representing over 70% of the total donor-T[reg]
   pool. However, this transient dominance was followed by a crash of the
   Bmal1 Δ T[reg] population, such that mutant cells represented less than
   10% of donor T[regs] 12 weeks after transfer ([100]Fig. 3D–[101]E).
   Similar dynamics were observed for donor-T[reg] numbers ([102]fig.
   S2E). In contrast, Bmal1 WT and Bmal1 Δ T[regs] persisted equally well
   in the spleen at 12 weeks after transfer ([103]Fig. 3F), suggesting a
   tissue-specific role for the T[reg]-intrinsic clock in promoting
   fitness. The reduced number of Bmal1 Δ cells was not due to a lack of
   proliferation as the fractions of Ki67^+ donor cells were similar in
   the presence and absence of Bmal1 ([104]Fig. 3G); nor did it appear to
   reflect migration to other tissues ([105]fig. S2F). We also verified
   that the reduction of Bmal1 Δ T[regs] was not due to the CD45 allele
   they expressed, as competitive transfer of CD45.1^+ and CD45.2^+ WT
   cells led to similar persistence in the recipients’ VAT ([106]fig.
   S2G). Therefore, loss of Bmal1 Δ donor cells most likely issued from a
   difference in survival.

   Finally, loss of fitness was not restricted to adoptive transfer
   settings nor to TCR-tg T[regs]. Because the Foxp3 gene maps to the X
   chromosome, random X-inactivation in female mice randomly silences one
   of the Foxp3 alleles in each cell. Thus, in
   Foxp3-Cre.YFP^+/−Bmal1^flox/flox heterozygotes, half of the T[regs]
   express BMAL1 (Cre^-) while half do not (Cre^+). The number and
   fraction of Cre^+ T[regs] in the gonadal fat of female heterozygotes
   were substantially lower than those of co-resident Cre^− T[regs]
   ([107]Fig. 3H). Only 20% of total T[regs] were Cre^+ and their numbers
   were significantly less than Cre^− Tregs ([108]Fig. 3H). As expected,
   control Foxp3-Cre.YFP^+/− females without Bmal1 deletion did not show
   these differences: Cre^+ cells represented 40% of total T[regs] in the
   gonadal fat, and the numbers of Cre^+ and Cre^− T[regs] were similar
   ([109]fig. S2H, [110]I).

   To rule out transcriptional effects of Bmal1 unrelated to its role in
   circadian rhythms, we performed analogous experiments on another
   core-clock gene, Nr1d1, encoding REV-ERBα. Like the T[reg]^Bmal1Δ
   strain, Foxp3-Cre.YFP^+.Nr1d1^flox/flox (hereafter referred to as
   T[reg]^Nr1d1Δ) mice showed normal accumulation of VAT T[regs]
   ([111]fig. S2J). However, donor Nr1d1 Δ T[regs] did not persist like
   Nr1d1 WT T[regs] did in the recipients’ VAT 12 weeks after transfer
   ([112]Fig. 3I, [113]J), whereas competitively transferred T[regs] of
   the two genotypes were maintained equally well in the spleen ([114]Fig.
   3K). Since Nr1d1 Δ T[regs] showed no reduction in Bmal1 expression at
   ZT12 ([115]fig. S2K), consistent with the role of REV-ERBα in
   repression of Bmal1 expression, the loss of competitive fitness of both
   Bmal1 Δ and Nr1d1 Δ donor cells indicated that circadian rhythms,
   rather than expression of Bmal1 per se, promoted VAT-T[reg] fitness.

Diurnal variation of the activation state of ST2^+ T[reg] subtypes

   As a preliminary step in understanding the mechanisms underlying
   diurnal control of VAT-T[reg] fitness, we performed single-cell
   (sc)RNA-seq on VAT T[regs] at ZT0 and ZT12 from T[reg]^Bmal1WT and
   T[reg]^Bmal1Δ animals. To permit simultaneous isolation of cell
   populations at ZT0 and ZT12, we housed half of the experimental animals
   in a room with the opposite light cycle for at least 14 days ([116]25),
   and verified that the animals therein indeed entrained to the new light
   regimen ([117]fig. S3A). Cells from the four conditions were separately
   hashtagged and encapsulated together. After demultiplexing, we obtained
   robust scRNA-seq data for over 6,500 VAT CD4^+ T cells, with an average
   of 1,300 genes detected per cell.

   Overlay of Foxp3 expression on a Uniform Manifold Approximation and
   Projection (UMAP) of the pooled datasets allowed us to distill the
   T[reg] compartment ([118]fig. S3B), consisting of over 2,000 cells. We
   then reclustered the T[regs], revealing five distinct subtypes
   ([119]Fig. 4A). The p1 ST2^+, p2 ST2^+, Tbet^+, IL18R^+ and resting
   clusters were distinguished by divergent and diagnostic gene expression
   profiles ([120]Fig. 4B). For instance, the resting cluster expressed
   elevated levels of Sell and Ccr7; the Tbet^+ cluster higher levels of
   Cxcr3 and Tbx21; and the IL18R^+ cluster elevated levels of Il18r,
   Cxcr3 and S1pr1. Two clusters expressed Il1rl1, encoding ST2: p1 with
   higher levels of Areg, Nfkbia and Junb transcripts; p2 with elevated
   Ly6c expression ([121]Fig. 4B). A differential density map revealed
   enrichment of the p1 over the p2 ST2^+ subtype in ZT0 VAT T[regs], as
   well an increased representation of the Il18r^+ subtype ([122]Fig. 4C).

Fig. 4: Diurnal variation of activation state within the ST2^+ T[reg]
population.

   Fig. 4:
   [123]Open in a new tab

   (A-D, G-J) scRNA-seq analysis of 21-to-23-week-old Foxp3-Cre.YFP^+ (WT)
   and Foxp3-Cre.YFP^+Bmal1^flox/flox (Δ) VAT T[regs] at ZT0 and ZT12. (A)
   Heterogeneity of VAT T[regs] visualized by UMAP. Left: a composite of
   all four conditions; right: disentangled plots for each condition. (B)
   Heatmap of top 30 differentially expressed genes of the clusters
   distinguished in panel A. (C) Cell-density differential between the
   UMAPs of ZT0 and ZT12 Bmal1 WT VAT T[regs]; frequencies of p1 and p2
   subtypes of ST2^+ cells are indicated. (D) Differential RNA expression
   of cell-surface markers for flow cytometric distinction of p1 ST2^+ and
   p2 ST2^+ T[regs]. (E) Fraction of CD69^hi VAT T[regs] at ZT0 or ZT12 in
   B6 mice. Left: flow cytometric dot plot; right: summary data. (F)
   Fraction of CD69^hi splenic T[regs] at ZT0 and ZT12. (E,F) Pooled data
   from two independent experiments. (G) KEGG pathway analysis by EnrichR
   of differentially expressed genes in cells of the p1 and p2 ST2^+
   clusters. (H-J) Examples of differentially expressed genes encoding
   activation (H), effector (I), and survival (J) molecules. ** p ≤ 0.01
   by Student’s t-test. n ≥ 3 per group per experiment. UMAP, Uniform
   Manifold Approximation and Projection; VAT, visceral adipose tissue;
   Spl, spleen. Flow-cytometric gating strategies for all quantified or
   isolated cell-types can be found in [124]fig. S7.

   To validate these findings at the protein level, we searched for
   differentially expressed genes whose products could be detected by flow
   cytometry. Antibodies recognizing CD62L, ST2, IL18R, CD39, CD69 and
   CD25 distinguished the five T[reg] clusters defined in [125]Fig. 4A
   ([126]fig. S3C). For the p1 and p2 ST2^+ clusters, levels of Cd69, Cd44
   and Il2ra transcripts were all higher in the p1 cluster, whereas Ly6c
   transcript levels were augmented in the p2 cluster ([127]Fig. 4D).
   However, by flow cytometry, cell-surface expression of only CD69
   clearly distinguished the two subtypes of ST2^+ cells. CD69^high ST2^+
   T[regs] showed a higher mean fluorescence intensity (MFI) of CD44 and
   CD25 staining, and a lower MFI of Ly6c staining compared with those of
   CD69^low ST2^+ T[regs], mirroring the scRNA-seq data ([128]fig. S3D).
   Consistent with the enrichment of p1 ST2^+ T[regs] at ZT0 ([129]Fig.
   4C), the fraction of CD69^high cells was significantly higher within
   the ST2^+ T[reg] population at ZT0 compared with ZT12 ([130]Fig. 4E).
   In contrast, splenic T[regs] showed a constant component of CD69^high
   cells ([131]Fig. 4F).

   We then used the scRNA-seq data to explore differences between the p1
   and p2 ST2^+ subtypes as well as to examine their interrelationships.
   Consistent with the higher CD44, CD25 and CD69 expressions, which are
   canonical T cell activation markers, T[regs] of the p1 ST2^+ subtype
   appeared to be more activated. Pathway analysis identified heightened
   NF-κB and TCR signaling pathways ([132]Fig. 4G), and transcripts
   encoding activation markers such as IκBα, NR4A1, EGR1 and JUNB were
   expressed at an elevated level ([133]Fig. 4H). Furthermore, p1 ST2^+
   cells expressed higher levels of transcripts encoding T[reg] effector
   molecules such as Areg, Ctla4 and Il10 ([134]Fig. 4I), as well as
   transcripts encoding pro-survival molecules such as Tnfaip3, Bcl2a1b
   and Il10ra ([135]Fig. 4J). On the other hand, p2 ST2^+ cells expressed
   more Ki67 transcripts, indicative of more cells in cycle ([136]fig.
   S3E); thus the two subtypes might reflect a balance of activation
   versus proliferation. RNA-velocity analysis, which predicts the likely
   future state of a cell based on the ratio of spliced to unspliced
   transcripts ([137]26, [138]27), revealed distinct trajectories at the
   two time-points. At ZT12, a trajectory from the p2 toward the p1 ST2^+
   subtype was predicted. That trajectory was absent at ZT0; instead, a
   subset of p1 ST2^+ cells exhibited a trajectory toward the p2 subtype
   ([139]fig. S3F). In brief, then, the VAT-Treg compartment includes
   multiple Treg subtypes. Clear diurnal variation was seen in the
   relative contributions of the p1 and p2 subtypes of ST2^+ cells.

Heightened activation state of Bmal1 Δ VAT T[regs]

   Diurnal variations within the ST2^+ T[reg] population raised the
   question of whether they depended on the T[reg-]intrinsic circadian
   clock. Analysis of cell density differences on the UMAP reproduced from
   [140]Fig. 4A ([141]Fig. 5A, left) revealed an enrichment in ST2^+
   T[regs], particularly the p1 ST2^+ subtype, at both ZT0 ([142]Fig. 5A,
   center) and ZT12 ([143]Fig. 5A, right) in Bmal1 Δ compared with Bmal1
   WT T[regs]. Furthermore, Bmal1 Δ T[regs] showed an enrichment of the p1
   ST2^+ subtype at both ZT0 and ZT12 ([144]Fig. 5A), in contrast to WT
   cells, for which we observed a diurnal switch in the dominant ST2^+
   subtype ([145]Fig. 4C), suggesting that Bmal1 Δ VAT T[regs] might be
   constitutively activated. Indeed, population-level RNA-seq revealed
   heightened activation terms such as “Th1 and Th2 activation”, “TNFR2
   signaling” and “IL-2 signaling”, as well as elevated expression of
   genes encoding cytokine common-γ-chain receptors (Il2ra, Il7r) and
   members of the NF-κB signaling pathway (Nfkbia, Nfkb1, Traf1) in Bmal1
   Δ cells ([146]Fig. 5B and [147]fig. S4A). Consistent with the scRNA-seq
   data, T[reg]^Bmal1Δ mice had an elevated fractions of ST2^+ T[regs]
   ([148]Fig. 5C), which translated into an enrichment of the canonical
   VAT-T[reg] signature ([149]28) ([150]Fig. 5D). Transcripts upregulated
   in the p1 vs p2 ST2^+ clusters were also enriched in the Bmal1 Δ T[reg]
   transcriptome ([151]Fig. 5E). According to flow cytometric analysis,
   the fraction of CD69^high cells within the ST2^+ VAT-T[reg] population
   was higher in T[reg]^Bmal1Δ than T[reg]^Bmal1WT mice ([152]Fig. 5F),
   consistent with an enrichment of the p1 ST2^+ subtype. This enrichment
   was accompanied by a non-significant relative decrease in the Tbet^+
   (Il18r^−) subtype in Bmal1 Δ VAT T[regs] ([153]fig. S4B). In contrast,
   the fraction of CD69^high T[regs] was the same in T[reg]^Bmal1WT and
   T[reg]^Bmal1Δ spleens. ([154]Fig. 5F). Consistent with the
   transcriptomic data, the MFI of CD44 staining was significantly higher
   and the MFI of CD25 staining trended higher in Bmal1 Δ than in Bmal1 WT
   VAT T[regs] ([155]Fig. 5G). These variations were not present in
   splenic T[regs] ([156]fig. S4C).

Fig. 5: Elevated activation state of Bmal1 Δ T[regs].

   Fig. 5:
   [157]Open in a new tab

   (A) Cell-density differential comparing Bmal1 WT and Bmal1 Δ VAT
   T[regs]at ZT0 (middle) and ZT12 (right). The left panel re-presents
   [158]figure 4A for convenient and accurate comparison. (B-D)
   Population-level RNA-seq of Bmal1 WT and Bmal1 Δ VAT T[regs] from
   18-week-old male mice at ZT0. (B) Ingenuity Pathway Analysis. (C)
   Fraction of ST2^+ VAT T[regs] from 18–30-week-old male mice at ZT0. (D)
   Volcano plot overlaid with up- (red) and down- (blue) regulated
   VAT-T[reg] signature genes ([159]28). (E) Volcano plot overlaid with
   genes upregulated in the p1 ST2^+ cluster (orange) or p2 ST2^+ cluster
   (pink). (F) Left: flow plot. Right: summary data showing fraction of
   CD69^high T[regs] in VAT and spleen at ZT0 or ZT12. (G) Geometric MFI
   (gMFI) of CD44 (left, middle) and CD25 (right) expression at ZT0. For
   flow cytometric data, p: * ≤0.05, ** ≤0.01, *** ≤ 0.001 according to
   Student’s t-test. Th, T helper cell; TNFR2, tumor necrosis factor
   recptor 2; ZT, Zeitgeiber time; VAT, visceral adipose tissue; Spl,
   spleen. Flow-cytometric gating strategies for all quantified or
   isolated cell-types can be found in [160]fig. S7.

   Male and female mice have distinct gonadal VAT T[reg] compartments
   ([161]18, [162]29, [163]30). We identified five T[reg] subtypes in
   T[regs] isolated from the gonadal fat (gVAT) of 8-to-12-week-old female
   mice using the gating strategy adopted from our single-cell analysis
   ([164]fig. S4D). The fraction of ST2^+ cells was lower in female than
   male mice, consistent with prior reports ([165]29, [166]30). Similar to
   males, female Bmal1 Δ mice exhibited an increased fraction of gVAT
   ST2^+ T[regs] and a decrease in the Tbet^+ (IL18r^−) T[reg] subtype
   ([167]fig. S4E). However, representation of the CD69^hi p1 subtype
   within the ST2^+ population was unchanged in mutant T[regs] ([168]fig.
   S4E).

   A heightened activation state in the absence of Bmal1 likely also
   explained the preferential expansion of the Bmal1 Δ donor-derived Treg
   population at 9 weeks after competitive transfer. Indeed,
   population-level RNA-seq on sorted Bmal1 WT and mutant donor-derived
   cells at this time-point revealed an enrichment in activation pathways
   such as “TNF-alpha signaling via NF-kb” and “IL-2/Stat5 Signaling”
   ([169]fig. S4F). Expression of Egr1, Nfkb1, Cd44, Il2ra and Il7r were
   increased in Bmal1 Δ donor-derived cells ([170]fig. S4G), as they were
   in the non-competitive setting ([171]fig. S4A). Flow cytometric
   analysis confirmed the expected increase in CD44 MFI in Bmal1 Δ
   donor-derived cells 12 weeks after transfer ([172]fig. S4H). Thus,
   Bmal1 Δ T[regs] showed an enrichment of the ST2^+ subtype in both male
   and female mice at steady-state, and were more activated than their WT
   counterparts in male mice and after adoptive transfer.

VAT T[regs] enforce diurnal lipolysis

   We next sought to identify the physiological function(s) of the
   cell-intrinsic clock in VAT T[regs]. When fed on lean chow,
   T[reg]^Bmal1Δ mice exhibited a small increase in adiposity compared
   with their WT littermates ([173]Fig. 6A), without significant changes
   in their glucose tolerance or insulin sensitivity ([174]fig. S5A).
   Because increased adiposity can result from reduced lipolysis, and
   given that the extent of adipose-tissue lipolysis is known to oscillate
   diurnally ([175]31), we interrogated the role of T[regs] in this
   process. Systemic depletion of T[regs], achieved through diphteria
   toxin (DT) administration to >16-week-old Foxp3-DTR^+ mice ([176]Fig.
   6B), provoked a rapid decrease in epididymal VAT (eVAT) weight 3 days
   after DT injection ([177]Fig. 6C), and the epididymal fat pads were
   visibly smaller ([178]Fig. 6D), suggestive of increased lipolysis. To
   assess the lipolytic rate of adipose tissue, we explanted eVAT and
   stimulated it with norepinephrine (NE) in vitro, as described
   previously ([179]32, [180]33). This approach circumvents the challenge
   of interpreting serum glycerol levels, which depend on both lipolytic
   output and peripheral-organ uptake ([181]34). Explants from
   T[reg]-depleted animals were less sensitive to NE stimulation and
   resisted further lipolysis, expressed as micromolar release of glycerol
   per milligram per hour, while basal glycerol release were similar
   ([182]Fig. 6E). This diminished lipolysis in vitro likely resulted from
   an elevated lipolysis in vivo. For instance, injection of the
   β3-adrenergic agonist CL316,243 (CL) to B6 mice, which is well known to
   induce extensive lipolysis in vivo, also resulted in diminished
   lipolysis in vitro ([183]fig. S5B).

Fig. 6. VAT T[regs] enforce diurnal lipolysis.

   Fig. 6
   [184]Open in a new tab

   (A) Body composition of 18-to-24-week-old male mice. (B-E, G-H)
   lipolysis experiments in 22-to-32-week-old male Foxp3-DTR^+ mice. (B)
   schematic of DT injection and subsequent analyses. (C-E) Body and
   epididymal VAT weight (C); photo of epididymal fat pad (D); in vitro
   glycerol release rate of VAT explants (E) 3 days after DT injection at
   ZT12 (F) in vitro glycerol release rate of VAT explants isolated from
   T[reg]^Bmal1WT male mice stimulated with 0.1μM norepinephrine. (G-H)
   Fraction of p1 ST2^+ (CD69^hi) T[regs] (G); in vitro glycerol release
   rate of VAT explants 4 weeks after DT injection stimulated with 0.1μM
   norepinephrine (H). (I) in vitro glycerol release rate of VAT explants
   of 12-to-23-week-old T[reg]^Bmal1WT and T[reg]^Bmal1Δ male mice
   stimulated with 0.1μM norepinephrine pooled from two independent
   experiments. For in vitro lipolysis assays, each dot is one explant,
   two explants were taken per mouse per condition. Paired Student’s
   t-test between littermate pairs was performed for panel A. For other
   panels, p: * ≤0.05, ** ≤0.01, *** ≤ 0.001 according to Student’s
   t-test. DTR, diphteria toxin receptor; μM, micro-molar; mg, milligram;
   h, hour; NE, norepinephrine; ZT, zeitgeiber time. Flow-cytometric
   gating strategies for all quantified or isolated cell-types can be
   found in [185]fig. S7.

   eVAT explants taken at ZT12 underwent significantly more in vitro
   lipolysis in response to NE stimulation than did those taken at ZT0
   ([186]Fig. 6F), indicative of a diurnal rhythm in lipolysis. However,
   explants from mice deficient in the ST2^+ T[reg] population, in
   particular the p1 subtype, did not exhibit this rhythmicity. Four weeks
   after DT injection, eVAT weight and VAT T[reg] numbers had normalized
   in DT-injected mice ([187]fig. S5C), but the fraction of p1 ST2^+
   T[regs] remained low ([188]Fig. 6G). Explants from these animals
   exhibited constitutively low lipolysis in vitro ([189]Fig. 6H),
   suggesting that p1 ST2^+ T[regs] de-sensitized adipocytes to NE-induced
   lipolysis in vivo and was required for diurnal rhythmicity. Finally, we
   assessed whether adipose-tissue lipolysis was controlled by BMAL1 in
   T[regs]. Diurnal rhythms of lipolysis in eVAT from T[reg]^Bmal1Δ mice
   appeared blunted: in vitro lipolysis was elevated at both ZT0 and ZT12
   after NE stimulation ([190]Fig. 6I), indicative of reduced lipolysis in
   vivo and consistent with the increased adiposity of T[reg]^Bmal1Δ mice
   ([191]Fig. 6A). Thus, BMAL1 expression enables VAT T[regs] to help
   enforce a diurnal rhythm in adipocyte lipolysis.

Accelerated activation but loss of fitness during high-fat-diet challenge

   The heightened activation state of Bmal1 Δ T[regs] in lean mice
   prompted us to characterize these mutant cells in the context of
   chronic activation such as high-fat-diet (HFD)-induced obesity. Upon
   HFD feeding, the VAT-T[reg] compartment responds in two distinct stages
   ([192]35). VAT T[regs] initially undergo a period of proliferation to
   counter mounting levels of inflammation ([193]35) and their
   transcriptome shows an enrichment of the VAT-T[reg] signature
   ([194]28). With chronic HFD feeding, bona fide (ST2^+, KLRG1^+,
   GATA3^+) VAT T[regs] are greatly reduced and the remnant cells lack the
   canonical VAT-T[reg] signature ([195]14, [196]28, [197]35). We fed
   T[reg]^Bmal1WT and T[reg]^Bmal1Δ mice on a lean diet for 14 weeks to
   establish a normal pool of VAT T[regs], followed by HFD feeding for 4,
   8 or 16 weeks ([198]Fig. 7A). Weight gain was similar between the two
   genotypes ([199]fig. S6A), as was the fraction and number of VAT
   T[regs]([200]Fig. 7B and [201]fig. S6B). But after 4 weeks of HFD,
   transcriptional profiling of VAT T[regs] revealed an enrichment of the
   VAT-T[reg] signature in the mutants’ T[reg] transcriptome ([202]Fig.
   7C), associated with an elevated fraction of ST2^+ T[regs] ([203]fig.
   S6C), and higher cell-surface expression of CD44 and CD25 ([204]Fig.
   7D). In contrast, after 16 weeks of HFD, the VAT-T[reg] signature was
   decreased in Bmal1 Δ compared with Bmal1 WT T[regs] ([205]Fig. 7E),
   suggestive of poor adaptation of Bmal1 Δ T[regs]. As a result,
   T[reg]^Bmal1Δ mice had a higher fraction of CD11c^+ inflammatory
   macrophages ([206]Fig. 7F), and a trend toward a greater accumulation
   of CD8^+ T cells at this time-point ([207]fig. S6D), idicative of
   increased VAT inflammation. In line with this notion, inflammatory
   pathways distinguished the eVAT depots from T[reg]^Bmal1WT and
   T[reg]^Bmal1Δ mice after 16 weeks of HFD, as assessed by whole-tissue
   RNA-seq. Pathways related to interferon, IL6 and TNFα signaling were
   upregulated in the VAT of T[reg]^Bmal1Δ mice, whereas metabolic
   pathways such as fatty acid metabolism and oxidative phosphorylation
   were downregulated ([208]Fig. 7G, [209]H). Glucose and insulin
   tolerance, assessed at 7 and 9 weeks after HFD feeding, were similar
   between the two genotypes ([210]fig. S6E). Thus, disruption of the
   cell-intrinsic clock impaired VAT T[regs]’ ability to adapt to the
   dysregulated VAT environment induced by HFD feeding.

Fig. 7. Accelerated activation but loss of fitness in Bmal1 Δ T[regs] during
HFD challenge.

   Fig. 7
   [211]Open in a new tab

   T[reg]^Bmal1WT and T[reg]^Bmal1Δ mice were fed a LFD for 14 weeks
   followed by a HFD for indicated durations; VAT T[regs] were isolated at
   ZT0, schematized in (A). (B) Percent of VAT T[regs]. (C) Volcano plot
   of VAT-T[reg] transcriptome after 4 weeks of HFD, overlaid with
   VAT-T[reg] signature genes similar to [212]figure 6D. (D) gMFI of CD44
   (left) and CD25 (right) staining after 4 weeks of HFD. (E) Volcano plot
   of VAT-T[reg] transcriptome after 16 weeks of HFD, overlaid with
   VAT-T[reg] signature genes. (F) Fraction of CD11c^+ macrophages. (G)
   Volcano plot of whole VAT transcriptome; up- or downregulated genes
   with p ≤ 0.05 and ≥1.5 fold change are highlighted; genes of interest
   are boxed. (H) Gene Set Enrichment Analysis (GSEA) of whole VAT
   transcriptome. For flow cytometric data, p: * ≤0.05, ** ≤0.01, *** ≤
   0.001 according to Student’s t-test. Data representative or pooled of
   at least two experiments. NES, normalized enrichment score; HFD,
   high-fat diet; wks, weeks. Flow-cytometric gating strategies for all
   quantified or isolated cell-types can be found in [213]fig. S7.

Diurnal variation in mitochondrial electron transport chain activity

   To explore potential mechanisms of how BMAL1 promoted T[reg] fitness,
   we returned to the rhythmic pathways identified in VAT T[regs]
   ([214]Fig. 2E). Besides loci associated with TCR activation, we noted
   that the rhythmically expressed genes in VAT T[regs] encoded molecules
   related to several mitochondrial oxidative metabolism pathways
   ([215]Fig. 2E). Genes encoding components of the electron transport
   chain (ETC) and the tricarboxylic acid (TCA) cycle peaked between ZT20
   and ZT0/24 according to JTK_CYCLE analysis ([216]Fig. 8A). We analyzed
   the metabolic states of VAT T[regs] at ZT0 and ZT12 to be consistent
   with other experiments, although analysis at ZT20 might have revealed
   even greater differences. Due to the very low number of VAT T[regs] in
   individual mice, we could not implement conventional methods such as
   Seahorse assays to evaluate mitochondrial metabolism. Instead, we
   turned to flow cytometry to assess mitochondrial oxidative metabolism
   by measuring the mitochondrial membrane potential (ΔΨ[m]) at
   steady-state and after inhibition of ATP synthase (Complex V).
   Catabolism of molecules such as glucose, fatty acids and amino acids
   ultimately feeds into the TCA cycle, where acetyl-CoA is sequentially
   oxidized and electrons are passed to the ETC via reduction and
   oxidation of NAD+/NADH. Electron transport is coupled to translocation
   of protons from the mitochondrial matrix into the intermembrane space.
   Such proton translocation generates an electrochemical gradient across
   the inner mitochondrial membrane, the ΔΨ[m], which is primarily used to
   power ATP synthesis at Complex V. While the basal ΔΨ[m] is a function
   of proton influx and efflux, inhibition of Complex V by oligomycin
   impairs proton influx to the matrix, and the resulting elevation of
   ΔΨ[m] is indicative of the extent of mitochondrial oxidative metabolism
   ([217]Fig. 8B). In other words, higher TCA cycle and ETC activity would
   result in a larger increase in mitochondrial membrane potential after
   oligomycin treatment.

Fig. 8: Circadian regulation of mitochondrial electron transport chain
activity.

   Fig. 8:
   [218]Open in a new tab

   (A) Circadian expression profiles of rhythmically expressed genes in
   the oxidative phosphorylation and TCA cycle pathways. RNA-seq data from
   [219]figure 2. (B) Schematic of mitochondrial membrane potential at
   baseline (left) and after inhibition of Complex V by oligomycin
   (right). Red color emphasizes changes after oligomycin inhibition. (C,
   D) Basal mitochondrial membrane potential measured by TMRE at ZT0 and
   ZT12 in VAT (C) and splenic (D) T[regs]. (E) VAT and splenic T[reg]
   responses to oligomycin. Ratio calculated as post-oligomycin gMFI/basal
   gMFI. (F) Basal TMRE in p1 ST2^+ (CD69^hi ST2^+) and p2 ST2^+ (CD69^lo
   ST2^+) VAT T[regs] at ZT0 and ZT12; (G) their response to oligomycin.
   (H, I) Basal mitochondrial membrane potential of Bmal1 WT and Bmal1 Δ
   VAT (H) and splenic (I) T[regs] at ZT0. (J) VAT and splenic T[reg]
   responses to oligomycin in T[reg]^Bmal1WT and T[reg]^Bmal1Δ mice at
   ZT0. p: * ≤0.05, ** ≤0.01 by Student’s t-test. Basal TMRE plots:
   representative of at least two independent experiments.
   Oligomycin/basal FC: pooled from at least two independent experiments.
   n ≥ 2 per condition per experiment. IMS, intermembrane space; oxphos,
   oxidative phosphorylation; TMRE, tetramethylrhodamine ethyl ester; FC,
   fold change. Flow-cytometric gating strategies for all quantified or
   isolated cell-types can be found in [220]fig. S7.

   At baseline, VAT T[regs] had a lower ΔΨ[m] at ZT0 than at ZT12, as
   measured by tetramethylrhodamine ethyl ester (TMRE) labeling ([221]Fig.
   8C), unlike splenic T[regs], wherein the ΔΨ[m] remained constant
   ([222]Fig. 8D). The lower ΔΨ[m] at ZT0 reflected greater oxidative
   phosphorylation, since inhibition by 1 μM oligomycin induced a greater
   increase in membrane potential in VAT T[regs] at ZT0 than ZT12
   ([223]Fig. 8E); in contrast, splenic T[regs] responded similarly to
   oligomycin at the two timepoints ([224]Fig. 8E). Within ST2^+ T[regs],
   the p1 subtype exhibited the most prominent diurnal changes ([225]Fig.
   8F, [226]G).

   VAT T[regs] from T[reg]^Bmal1Δ mice exhibited a higher baseline ΔΨ[m]
   at ZT0 than their counterparts from Bmal1 WT littermates ([227]Fig.
   8H), whereas such differences were not observed in the spleen
   ([228]Fig. 8I). These ΔΨ[m] differences translated into a greater
   response to oligomycin in Bmal1 WT than Bmal1 Δ VAT T[regs], but not in
   the corresponding splenic T[regs] ([229]Fig. 8J). Overall, these
   findings illustrated VAT-specific diurnal variation in the extent of
   mitochondrial oxidative metabolism and suggested a role for BMAL1 in
   promoting ETC activity in VAT T[regs].

DISCUSSION

   It has been reported that the adaptive immune response is unaffected by
   the cell-intrinsic clock ([230]12). However, that study focused on
   lymphocytes within lymphoid organs, and we found that the
   transcriptomes of T[regs] within non-lymphoid, compared with lymphoid,
   tissues were enriched in pathways associated with circadian rhythms
   ([231]1, [232]2). In this report, we showed that genes encoding core
   components of the cell-intrinsic clock were rhythmically expressed in
   VAT, but not splenic, T[regs]. T[reg]-specific mutations of core-clock
   genes impacted VAT, but not splenic, T[reg] activation, metabolism and,
   ultimately, fitness. These differences had organismal consequences: for
   example, the cell-intrinsic clock enabled VAT T[regs] to enforce a
   diurnal rhythm in epididymal VAT lipolysis.

   As T[regs] refine their phenotype in various tissues, they become more
   activated and adapt their cellular metabolism to their specific locale
   ([233]1). There is past and current evidence that both of these
   processes fall under circadian control. T cells isolated at different
   ZTs are differentially sensitive to TCR stimulation ([234]36, [235]37).
   Murine CD8^+ T cells isolated during the day are more robustly
   activated after recognition of cognate antigen in vitro, and exhibit a
   diurnal variation in their activation after DC-mediated vaccination in
   vivo ([236]37). BMAL1 restrained VAT, but not splenic, T[reg]
   activation, similar to its role in other immunocyte populations.
   Inflammatory cytokine production in macrophages or IL-22 and IL-17
   production in ILC3s are heightened after Bmal1 or Nr1d1 deletion
   ([237]38–[238]44). The brakes on cell activation imposed by BMAL1 in
   multiple immunocyte lineages that lack TCRs also suggest that the NF-κB
   signaling pathway might be a primary target of circadian regulation.
   Indeed, Bmal1 deletion heightens NF-κB signaling in macrophages
   ([239]38) and intestinal epithelial cells ([240]45).

   Many of the rhythmically expressed genes in VAT T[regs] belonged to
   metabolic pathways. Our results evidenced a clear diurnal variation in
   the TCA cycle and in ETC activity. Rhythms in mitochondrial ETC
   activity can originate from several processes. For instance,
   mitochondria undergo diurnal cycles of fusion and fission
   ([241]46–[242]48), which modulate the efficiency of ATP production.
   Fused mitochondria, which are more efficient at producing ATP, occur
   early in the day ([243]46–[244]49) and are associated with a heightened
   mitochondrial oxygen consumption rate and ATP production ([245]47,
   [246]48). In VAT T[regs], the timing of increased oxidative metabolism
   coincided with the period of heightened activation state to potentially
   meet an elevated energy demand. These variations were not present in
   splenic T[regs], highlighting the different metabolic needs of their
   VAT-T[reg] counterparts.

   In the absence of these and presumably other temporal modulations, VAT
   T[regs] lost fitness in competitive transfers, in female Bmal1
   heterozygotes, and in response to HFD challenge. The reduced fitness
   resulted from disruption of the core-clock program rather than loss of
   BMAL1 expression per se, since adoptive transfer of either Bmal1 or
   Nr1d1 Δ T[regs] led to loss of fitness. Some of the phenotypes in
   T[reg]^Bmal1Δ mice might be attributable to other core-clock proteins,
   such as REV-ERBα, whose expression was significantly altered by loss of
   BMAL1. The constitutively heightened activation state in
   Bmal1-deficient T[regs] suggests dysregulation of the NF-κB signaling
   pathway, which is intimately related to apoptosis and necrosis. Indeed,
   the upstream regulators of NF-κB signaling serve a second, less
   recognized, function as cell death checkpoints, and NF-κB target genes
   are known to regulate cell-death (reviewed in ([247]50, [248]51)).
   However, the in vivo signals that result in cell death instead of
   pro-survival signaling by NF-κB in VAT T[regs] after competitive
   transfer remain undefined. Furthermore, ETC activity appeared dampened
   in Bmal1 Δ VAT T[regs] and might contribute to the loss of fitness.
   BMAL1 promotes mitochondrial oxidative metabolism ([249]42, [250]46,
   [251]52). An inefficient mitochondrial respiration in the absence of
   BMAL1 increases the production of reactive oxygen species (ROS)
   ([252]53) and ROS exhibits circadian rhythmicity ([253]41, [254]42,
   [255]47). Poorly optimizied mitochondrial function can impair T[reg]
   fitness and function, as has been shown in other reports
   ([256]54–[257]58).

   We uncovered a role for VAT T[regs] in regulating diurnal lipolysis of
   the adipose tissue, which requires BMAL1 expression in T[regs].
   Norepinephrine-induced lipolysis is an important mechanism for
   providing nutrients during states of energy deprivation ([258]59), but
   excessive lipolysis recruits inflammatory infiltrates to the adipose
   tissue ([259]60, [260]61). Thus, adipocyte lipolysis is highly
   regulated. In mice, lipolysis peaks near ZT0 ([261]31), and we
   recapitulated these findings in vitro as VAT explants isolated at ZT0
   resisted further norepinephrine stimulation, compared with those taken
   at ZT12. We found that ST2^+ T[regs], in particular the p1 subtype,
   suppressed lipolysis in lean animals, and rhythmic lipolysis was
   disrupted in both DT-treated Foxp3-DTR^+ mice and in T[reg]^Bmal1Δ
   animals. It seems likely that VAT T[regs] regulate lipolysis by
   controlling the inflammatory tenor of the adipose tissue, since
   cytokines such as TNFα induce lipolysis ([262]62, [263]63).
   Alternatively, T[regs] might control other VAT immunocytes that
   modulate sympathetic nervous signaling, such as macrophages ([264]32,
   [265]64). Lastly, VAT T[regs] might control the sympathetic tone
   directly through an unknown mechanism. That VAT T[regs] enforce diurnal
   lipolysis is consistent with their role in promoting metabolic
   homeostasis in VAT ([266]14, [267]16, [268]17).

   It is unclear how VAT T[regs] are entrained by circadian rhythms.
   Identifying specific signals that entrain them, indeed non-neuronal
   cells in general, remains an important goal in the field. Hormonal,
   neuronal and metabolic signals are drivers of some of the mechanisms
   (reviewed in ([269]6)). TCR activation upregulates the expression of
   core-clock genes such as Bmal1 and Nfil3 ([270]65, [271]66), and thus
   might kick-start the cell-intrinsic clock program. Metabolic cues such
   as insulin signaling ([272]67) might also entrain resident immunocytes.
   Given the diurnal lipolysis of adipose tissue, whether lipid species
   also provide entrainment cues to VAT T[regs] is an intriguing
   possibility.

   While we did not find most core-clock genes to be significantly
   rhythmic in splenic T[regs], it remains possible that a more frequent
   and prolonged sampling, for instance over 48 hours with a 2-hour
   sampling interval, might detect weak rhythmicity in the expression of
   other core-clock genes. Nonetheless, other studies have noted a
   similarly weak rhythmicity in splenocytes ([273]12, [274]13).
   Rhythmicity in VAT T[regs] entailed changes on the order of 2-fold,
   which is less than what has been observed with some cell-types, such as
   hepatocytes, but is similar to observations on other immunocytes
   ([275]68, [276]69). Lastly, due to the very low number of VAT T[regs]
   per mouse, we could not perform metabolomics or metabolite-tracing
   experiments to pinpoint the specific metabolic reactions under diurnal
   control.

   Tissue microenvironments undergo significant changes throughout the
   day. The elevated rhythmicity of core-clock gene expression provides a
   molecular mechanism for VAT T[regs] to synchronize and adapt to the
   changing tissue environment. An improved understanding of the diurnal
   behaviors of tissue T[regs] might help maximize the efficacies of
   T[reg]-based immunotherapeutic strategies.

MATERIALS AND METHODS

Study design

   This study aimed to establish to what extent circadian rhythms occurs
   in T[regs] in VAT versus the spleen, and how important it is. To that
   end, we performed transcriptomic profiling by population-level and
   single-cell RNA sequencing over circadian time, as well as
   cytofluorimetry-based metabolic assays in C57BL/6 and mutant
   littermates lacking core-clock components.

Mice

   C57BL/6 (B6), B6.CD45.1 (stock number 002014, B6.SJL-Ptprca Pepcb/BoyJ)
   and Bmal1^flox/flox (stock number 007668, B6.129S4(Cg)-Arntltm1Weit/J)
   mice were obtained from the Jackson Laboratory. Nr1d1^flox/flox mice
   have been reported (University of Pennsylvania) ([277]70). Foxp3-Ires
   GFP.hDTR mice were obtained from A. Rudensky (Memorial Sloan Kettering
   Cancer Center). The B6.Foxp3-Ires GFP ([278]71), B6.Foxp3-Cre.YFP
   ([279]72), and VAT TCR-tg lines ([280]18) have been described. Mice
   lacking Bmal1 or Nr1d1 specifically in T[regs] were generated by
   crossing B6.Foxp3-Cre.YFP mice with the respective flox strains, and
   are typically referred to in the text as T[reg]^Bmal1Δ or Treg^Nr1d1Δ
   For transfer experiments, knock-out donor strains were generated by
   crossing B6.Foxp3-Cre.YFP.CD45.2^+ mice with VAT TCR-tg mice and either
   the Bmal1^flox/flox or Nr1d1^flox/flox strain to obtain
   Foxp3-Cre.YFP.CD45.2^+[.]TCR-tg^+ Bmal1^flox/flox or Nr1d1^flox/flox
   offspring. Female heterozygous Cre strains were generated by crossing
   Foxp3-Cre.YFP^+/+Bmal1^flox/flox with Bmal1^flox/flox mice. Wild-type
   donor strains were obtained by crossing B6.Foxp3-Cre.YFP mice with
   B6.CD45.1 and VAT TCR-tg mice to obtain Foxp3-CreYFP.CD45.1^+.TCR-tg^+
   offspring. Recipients were generated by crossing B6.Foxp3-Cre.YFP with
   B6.CD45.1 to obtain Foxp3-Cre.YFP.CD45.1/2^+ offspring. Appropriate
   littermate controls were always used except for transfer experiments,
   where co-housed donors of similar age were used. All animals were
   housed in specific pathogen-free facilities at Harvard Medical School’s
   Center for Animal Resources and Comparative Medicine. All experiments
   were performed under protocols approved by the HMS Institutional Animal
   Care and Use Committee (protocol# IS00001257).

   Mice were housed under a strict 12h:12h light: dark cycle. In some
   experiments, they were entrained in a reverse light-cycle room for a
   minimum of 14 days with a 12h:12h dark: light cycle, which enabled
   simultaneous tissue collection from mice at different circadian
   time-points. To assess adaptation to reverse light cycle, mice were
   housed individually with unrestricted access to running wheels equipped
   with activity tracking (Columbs Instruments). Wheel revolutions were
   recorded in each cage every 10 seconds and cumulative wheel turns from
   all mice are plotted.

   Unless otherwise noted, weaned mice were fed a low-fat-diet chow
   (no.5053, Picolab Rodent Diet 20; 13% kcal fat) ad libitum. For
   high-fat-diet experiments, mice were fed a low-fat-diet chow (no.5053)
   ad libitum until 14 weeks of age, then fed a high-fat diet (Open Source
   Diet D12492, 60% kcal fat) ad libitum for 4, 8 or 16 weeks.

Tissue preparations for flow cytometry and cell sorting

   Tissue preparation has been previously described with slight
   modifications ([281]29) Briefly, male mice were euthanized by CO[2].
   Epididymal VAT and spleen were excised and placed in Dulbecco’s
   modified Eagle’s medium (DMEM) containing 2% fetal bovine serum (FBS)
   and 10mM HEPES (referred to as FACs buffer). VAT depots were minced
   with scissors and incubated with FACs buffer containing 1.5mg/mL
   collagenase type II (Sigma) at 37°C for 20 minutes. The adipocyte layer
   was removed after centrifugation at 420 g. Red blood cells were lysed
   by ACK lysing buffer (Gibco). Spleen and lymph nodes were mechanically
   dissociated by mashing against a 40 μM filter, followed by red blood
   cell removal by ACK lysis. For liver, lung and kidney preparations,
   excised tissues were minced with scissors and incubated in 1% FCS DMEM
   containing 0.5mg/mL collagenase IV (Gibco), 150 μg/mL DNase I (Sigma)
   at 37°C for 30–45 minutes. Digested tissues were filtered and washed.
   For liver and kidney, leukocytes were enriched by centrifugation (10
   min at 800 × g) with 36% Percoll (GE Healthcare).

   For surface-antigen staining, single-cell suspensions were incubated in
   phosphate-buffered saline (PBS) containing live/dead stain and surface
   antibodies for 10 minutes on ice, followed by washes in FACs buffer.
   For intracellular staining, single-cell suspensions were fixed in
   eBioscience Fix/Perm buffer for 20–30 minutes at room temperature,
   followed by permeabilization in eBioscience permeabilization buffer.
   Cells were stained with intracellular antibodies for 30 minutes at room
   temperature followed by washes with permeabilization buffer.

   The following antibodies were used: anti-mouse -CD4 (GK1.5), -CD8α
   (53–6.7), -CD11b (M1/70), -CD11c (N418), -CD19 (6D5), -CD39 (Duha59),
   -CD45 (30-F11), CD45.1 (A20), -CD45.2 (104), CD62L (MEL-14), -CD69
   (H1.2F3), -Il18r/CD218α (BG/IL18RA), -CD44 (IM7), -Il17r/CD127 (A7R34),
   -CD206 (C068C2), -Ly6c (HK1.4), -F4/80 (BM8) and -I-A/I-E (M5/114.15.2)
   all from BioLegend; -CD25 (PC61), -TCRβ (H57–597) and -Ki67 (B56) from
   BD bioscience; -Foxp3 (FJK-16s), -Il33R/ST2 (RMST2–2) and -Klrg1 (2F1)
   from Thermofisher. Live/Dead fixable viability dye yellow, near-IR or
   UV (Thermofisher) were used to distinguish live and dead cells.
   Typically T[regs] were defined as CD45^+ TCRβ^+ CD4^+ Foxp3^+. For cell
   sorting, T[regs] were defined as CD45^+ Tcrb^+ CD4^+ YFP^+ CD8α^−
   CD11b^− CD11c^− CD19^− DAPI^−, unless otherwise specified.

   For 5-ethynyl-2’-deoxyuridine (EdU) experiments, 40μg/g of body weight
   EdU (Thermofisher) was intraperitoneally injected 12 hours before
   sacrifice. EdU was detected using Click-iT plus Pacific Blue Flow
   Cytometry Assay Kit (Thermofisher) according to the manufacturer’s
   instructions.

RNA Sequencing

   RNA sequencing was performed as described at [282]www.immgen.org. 1000
   T[regs] were double-sorted directly into 5 μl TCL-buffer (Qiagen)
   containing 1% β-mercaptoethanol (Sigma). For adoptive transfer
   experiments, 500 cells were double-sorted. Smart-seq2 libraries were
   prepared and sequenced as previously described ([283]73) using the
   Broad Genomics Platform. Briefly, RNA was captured and purified using
   RNAClean XP beads (Beckman Coulter). Polyadenylated mRNA was selected
   using an anchored oligo(dT) primer (50
   –AAGCAGTGGTATCAACGCAGAGTACT30VN-30) and converted to cDNA by reverse
   transcription. Limited PCR amplification was performed on the
   first-strand cDNA followed by Tn5 transposon-based fragmentation using
   the Nextera XT DNA Library Preparation Kit (Illumina). Samples were PCR
   amplified for 12 cycles using barcoded primers such that each sample
   carried a specific combination of eight base Illumina P5 and P7
   barcodes for subsequent pooling and sequencing. Paired-end sequencing
   was performed on an Illumina NextSeq 500 using 2 × 38bp reads with no
   further trimming. Reads were aligned to the mouse genome (GENCODE
   GRCm38/mm10 primary assembly and gene annotations vM16;
   [284]https://www.gencodegenes.org/mouse_releases/16.html) with STAR
   2.5.4a ([285]https://github.com/alexdobin/STAR/releases). Ribosomal RNA
   gene annotations were removed from GTF (General Transfer Format) file.
   Gene-level quantification was calculated by featureCounts
   ([286]http://subread.sourceforge.net/). Raw read counts tables were
   normalized by the median of ratios method with DESeq2 package from
   Bioconductor
   ([287]https://bioconductor.org/packages/release/bioc/html/DESeq2.html)
   and then converted to GCT and CLS format. Samples with less than 1
   million uniquely mapped reads or with fewer than 8,000 genes over ten
   reads were considered low quality and removed, followed by further
   screening for contamination by using known cell-type-specific
   transcripts. Biological replicates with poor Pearson correlation (<0.9)
   were removed. Genes with less than 10 reads across samples or a high
   coefficient of variation (typically >0.6) were removed. Differential
   gene-expression analysis was performed using EdgeR ([288]74);
   differentially expressed genes were defined as p ≤ 0.05 according to
   quasi-likelihood F-test. For pathway enrichment analysis,
   differentially expressed genes were analyzed by Ingenuity Pathway
   Analysis (Qiagen,
   [289]https://www.qiagenbioinformatics.com/products/ingenuitypathway-ana
   lysis) ([290]75), Gene Set Enrichment Analysis (Subramanian et al 2005)
   or EnrichR ([291]76, [292]77). Enriched pathways were defined as FDR
   ≤0.05. In some cases, previously established signatures were tested for
   enrichment in our dataset, where significant enrichment was defined as
   p ≤0.05 according to Fisher’s exact test.

   For whole-tissue RNA-sequencing, RNA was extracted from VAT using the
   RNeasy lipid tissue mini kit following the manufacturer’s protocol
   (Qiagen). 2ng of RNA was added into 5 μl TCL-buffer (Qiagen) containing
   1% β-mercaptoethanol (Sigma) and sequenced as above.

RT-PCR

   RNA was extracted from 100,000 sorted splenocytes using Trizol
   (Invitrogen) following the manufacturer’s protocol. Complementary DNA
   synthesis was performed using a Maxima H Minus Reverse Transcriptase
   kit (Thermo Fisher Scientific). Quantitative PCR (qPCR) was run on a
   QuantStudio 5 Real-Time PCR System (Applied Biosystems). The expression
   of Bmal1 was calculated by the 2^−ΔΔCt method relative to the
   expression of the ribosomal subunit 36b4 (Rplp0). Primer sequences:
   Bmal1 AGGATCAAGAATGCAAGGGAGG (F), TGAAACTGTTCATTTTGTCCCGA (R); Rplp0
   GGAGTGACATCGTCTTTAAACCCC (F), TCTGCTCCCACAATGAAGCA (R).

Adoptive transfers

   Cells from 6-to-8-week-old male donors were pooled from spleen and
   peripheral lymph nodes, followed by enrichment for CD4^+ cells using
   Dynabeads untouched Mouse CD4 (Thermofisher). Cells were single sorted
   for YFP^+ T[regs]. For the competitive transfer experiments, a 1:1 mix
   of 250,000 donor T[regs] from wild-type or knock-out donors (500,000
   total T[regs]) were resuspended in PBS and intravenously injected via
   the tail vein into 8-to-10-week-old male recipients. Donor and
   recipient mice were fed a low-fat-diet chow with 21% kcal fat (no.
   5058, Picolab Rodent Diet 20)

scRNA-seq

   34,000 VAT and 32,000 splenic CD4^+ T cells were sorted and multiplexed
   using TotalSeq hashing antibodies (BioLegend). Cell encapsulation, RNA
   isolation, and hash library construction and sequencing were performed
   by the Broad Institute Genomic Services using the 10X Genomics
   platform. Samples were processed using 3’ v3.1 chemistry and hashed
   cells were loaded onto one lane. cDNA and gene expression libraries
   were constructed following the manufacturer’s instructions (10X
   Genomics). Gene expression libraries were sequenced on HiSeq X. Samples
   were demultiplexed using cellranger v6.0 and hash libraries were
   processed using cite-seq-count
   ([293]https://hoohm.github.io/CITE-seq-Count/)

   Data analysis was performed following the Seurat v3 pipeline ([294]78).
   Low-quality cells were removed and were defined as those with less than
   700 or more than 3000 unique genes, as well as those with >10%
   mitochondrial gene reads. Clusters of proliferating cells were also
   removed. Non-CD4^+ T cell contaminants were removed using well-known
   cell-identity markers. Clustering, dimensionality reduction onto UMAP,
   differential gene expression, and visualization were performed using
   the built-in functions of Seurat v3. Density distributions were
   calculated using two-dimensional kernel density estimation.

   RNA velocity was performed as described ([295]27). Briefly, the spliced
   and unspliced counts were calculated from the original sequencing data
   using the RNA velocity package ([296]26). The spliced and unspliced
   expression matrices were then integrated into the single-cell data
   matrix, and the scvelo package was used to calculate the velocity
   vectors using the dynamical model of transcriptional dynamics and
   splicing kinetics. The calculated velocity vectors were then projected
   onto the UMAP embedding calculated during the scRNA-seq analysis.

Lipolysis assays

   The in vitro lipolysis assay was adapted from previous studies
   ([297]32, [298]33, [299]79). Briefly, approximately 20mg of epididymal
   VAT was excised into 200μl of 1g/L glucose DMEM (Gibco) supplemented
   with 10 mM HEPES, 2% fatty-acid free bovine serum albumin (Sigma), and
   5 μM Triacsin C (Sigma) (collectively referred to as “lipolysis
   buffer”), with or without norepinephrine (A0937 Sigma) for 1 hour at
   37°C. Explants were then transferred into fresh lipolysis buffer with
   or without NE for an additional hour, and the supernatant was collected
   for glycerol quantification following the manufacturer’s protocol
   (SGA1, Zenbio). To quantify protein content, the we placed the fat pad
   into a 1 mL of chloroform:methanol (2:1 v/v) with 1% glacial acetic
   acid (extraction solution) for 1 hour at 37°C while rotating, and then
   incubated the material in 500μl of 0.3N NaOH with 0.1% SDS in dH[2]O
   (lysis buffer) overnight at 55 °C. The lysis buffer supernatant was
   taken for protein quantification using the Pierce BCA Assay following
   the manufacturer’s protocol (Thermofisher). Glycerol release rate was
   expressed as μM glycerol per milligram protein per hour.

   For induction of lipolysis in vivo, 1mg/kg CL316,243 (Sigma) was
   i.p-injected. Serum was collected via the tail vein using
   non-heparinized capillary tubes (Fisher) and was allowed to clot at
   room temperature for 15 minutes. After centrifugation at 2000 × g for
   15 minutes, the supernatant was assayed for free fatty acid or glycerol
   levels following the manufacturer’s protocol (GFA-1 Zenbio).

   Foxp3-DTR^+ mice was i.p-injected with 20ng per g body weight of DT
   once per day, at ZT10-ZT12, for three consecutive days.

Organismal metabolism

   Organismal metabolism was assessed by the glucose tolerance test (GTT)
   and insulin tolerance test (ITT). For the GTT, mice were fasted for 5
   hours and acclimatized to the experiment room for 45 minutes, followed
   by intraperitoneal injection of 1g/kg body weight of glucose. Glucose
   levels were then measured from tail vein blood at 0, 20, 40, 60, 90 and
   120 minutes after the glucose injection. For ITT, mice were fasted for
   4 hours, followed by 45 minutes of acclimatization to the experiment
   room. They were intraperitoneally injected with 0.6 unit/kg (Humulin R
   U-100 from Eli Lilly) body weight, followed by glucose level
   measurements at 0, 20, 40, 60, 90 and 120 minutes after injection.
   Blood glucose levels were measured using a Contour glucose meter.

Cellular metabolism

   Mitochondrial membrane potential was measured using
   tetramethylrhodamine ethyl ester perchlorate (TMRE) from Thermofisher.
   Single-cell suspensions were stained with surface antibodies as
   described above and washed in FACs buffer. Cells were then incubated in
   round-bottom 96-well tissue-culture plate (Corning) in 100μl of DMEM +
   2% FBS + 10 mM HEPES + 2mM glutamine at 37°C for 30 minutes. Where
   applicable, cells were treated with 1 μM oligomycin (delivered at 3X
   concentration in 50 μL) at 37°C for 10 minutes. Finally, cells were
   stained with 100 μM TMRE (delivered at 4X in 50 μL) at 37°C for 20
   minutes. Cells were spun down and resuspended in cold FACs buffer
   without any further washes before immediate acquisition on the flow
   cytometer.

Statistical analyses

   Data are routinely shown as mean ± SD. Unless stated otherwise,
   statistical significance was determined by two-tailed Student’s t-test
   using GraphPad Prism 9.0. * p <0.05; ** p <0.01; *** p < 0001. A value
   of more than three standard deviations from the mean was adopted as
   criteria to exclude outliers. Independent replicate experiments were
   typically performed with at least two biological replicates per group.
   Circadian rhythmicity of gene expression was assessed using JTK_CYCLE
   ([300]80). Rhythmic genes were defined as p ≤ 0.05, gene expression ≥
   20, and period ≥ 16 hours. Time of peak expression was determined by
   JTK_CYCLE.

Supplementary Material

   XiaoEtAl-FigsS1-S6

   fig. S1: Expression of core-clock genes.

   fig. S2: Clock genes confer fitness to VAT T[regs].

   fig. S3: T[reg] heterogeneity.

   fig. S4: Characterization of Bmal1Δ Tregs.

   fig. S5: VAT Tregs regulate adipocyte lipolysis.

   fig. S6: T[reg]^Bmal1WT and T[reg]^Bmal1Δ mice during HFD challenge

   fig. S7: Gating strategies
   [301]NIHMS1854639-supplement-XiaoEtAl-FigsS1-S6.pdf^ (736KB, pdf)
   XiaoEtAl_RawDataFile
   [302]NIHMS1854639-supplement-XiaoEtAl_RawDataFile.xlsx^ (4.7MB, xlsx)
   XiaoEtAl_Table_S1

   Table S1: JTK analysis of VAT and splenic T[reg] transcriptome over
   circadian time
   [303]NIHMS1854639-supplement-XiaoEtAl_Table_S1.xlsx^ (4.6MB, xlsx)

ACKNOWLEDGEMENTS