Proteomic insights into circadian transcription regulation: novel E-box interactors revealed by proximity labeling Authors: Manon Torres, Marieluise Kirchner, Caroline G. Marks, Philipp Mertins, Achim Kramer Published: 2024 ABSTRACT In this study, Torres et al. use novel genomic locus-specific, proximity labeling proteomics to quantitate circadian-dependent protein binding at the E-box of the clock-controlled Dbp gene. This work delineates known and novel protein regulators that play a role in orchestrating circadian transcriptional regulation. Circadian clocks are molecular oscillators that are present in most mammalian cells and operate on an ∼24 h cycle. They coordinate the daily rhythms of molecular, physiological, and behavioral processes. These clocks play a central role in maintaining health, and disruptions caused by factors such as shift work and jet lag have been linked to the onset of various diseases such as cancer, diabetes, obesity, and cardiovascular and neurodegenerative diseases. Therefore, deciphering the basic mechanisms that control circadian rhythms is essential for the development of effective health strategies. At the heart of molecular oscillators is the regulation of circadian transcription. Several cis-regulatory proteins and enhancer elements, including the E-box, Rev–ErbA/ROR response element (RRE), and D-box, have been identified as necessary elements of transcriptional–translational feedback loops (TTFLs) that drive cell-autonomous oscillations. The transcription factors BMAL1 and CLOCK form a protein complex that rhythmically binds to E-box enhancer elements typically located near the transcription start site of clock or clock-controlled genes (Gekakis et al. 1998; Ripperger et al. 2000; Ripperger and Schibler 2006). This complex promotes the transcription of its own inhibitors, especially PERs and CRYs, which in turn form a repressor complex. This inhibitory complex attenuates CLOCK:BMAL1 transcriptional activity with a delay of several hours (Aryal et al. 2017; Gabriel et al. 2021). The oscillation is strengthened by REV–ERB and ROR proteins that bind to the RREs in the Bmal1 promoter. The network of clock-controlled genes directly regulated by CLOCK:BMAL1 includes several transcription factors and coregulators. For example, the CLOCK:BMAL1 complex orchestrates the rhythmic expression of the D-box binding PAR BZIP transcription factor (Dbp) gene (Ripperger et al. 2000), which in turn controls a large number of genes containing D-box sequences (Bozek et al. 2007; Yoshitane et al. 2019). This system allows for broad temporal control of downstream factors via rhythmic transcriptional regulation. Mechanistically, the transcriptional activity of BMAL1:CLOCK is regulated at multiple levels, including its abundance in the nucleus (Kondratov et al. 2003; Kwon et al. 2006), formation of the BMAL1:CLOCK heterodimer, DNA binding, and histone modifications (Etchegaray et al. 2003; Curtis et al. 2004; Hirayama et al. 2007; Asher et al. 2008; Nakahata et al. 2008; Koike et al. 2012; Shi et al. 2016; Beytebiere et al. 2019), as well as protein turnover. Heterodimerization and turnover of BMAL1 and CLOCK are closely related to post-translational modification processes, including phosphorylation (Spengler et al. 2009; Tamaru et al. 2009; Sahar et al. 2010; Kwak et al. 2013; Aryal et al. 2017; Yoshitane et al. 2019; Klemz et al. 2021), ubiquitylation (Kwon et al. 2006; Stratmann et al. 2012), or SUMOylation (Cardone et al. 2005; Lee et al. 2016). In addition, characterization of the BMAL1 and CLOCK cistromes in mouse livers has shown that DNA binding by the BMAL1:CLOCK complex does not determine the transcription phase of most target genes (Menet et al. 2012; Trott and Menet 2018), highlighting the influence of additional transcriptional modulators on clock-regulated gene expression. This suggests that BMAL1:CLOCK regulation of clock-regulated gene expression relies on cooperation between CLOCK:BMAL1 and other transcription factors, allowing for interplay with other signaling pathways. The study of proteins involved in the circadian regulation of transcription has so far relied on antibody-based chromatin immunoprecipitation (ChIP), protein–protein interaction (PPI), or cistrome analysis. However, these methods are only partially suitable for the de novo identification of proteins involved in the circadian transcription machinery. Recently, proximity labeling in combination with mass spectrometry proteomics has emerged as a complementary strategy (Niinae et al. 2021). Techniques based on proximity biotinylation have been used to profile various protein interactions, including insoluble protein complexes and temporal protein interaction profiling. Engineered peroxidases (e.g., APEX, APEX2, and HRP) oxidize biotin-phenols to biotin-phenoxyl radicals in the presence of H2O2, and the resulting radicals biotinylate electron-rich amino acids (e.g., Tyr, Trp, Cys, and His). Engineered peroxidases have been used to develop new tools to identify regulatory proteins involved in the control of gene expression. In their studies, Gao et al. (2018) and Myers et al. (2018) used APEX or APEX2 fused to a catalytically dead Cas9 mutant (dCas9) for the biotinylation of DNA-binding proteins near specific DNA elements. Here, we adapted a DNA–protein proximity labeling method based on APEX2 biotinylation and subsequent liquid chromatography mass spectrometry (LC-MS/MS) proteomics to identify proteins that bind to the E-box region of the classical clock-controlled gene Dbp in mouse NIH3T3 fibroblasts. We identified 69 proteins that are present at the E-box of the first intron of Dbp at the time of maximal BMAL1 binding, including proteins already known to be involved in circadian transcription via E-boxes such as BMAL1, HDAC3, and H2AZ, as well as previously unknown E-box proteins such as XPO7 and MINK1. The latter play a critical role in normal E-box-mediated transcription, as their depletion in human U-2 OS cells leads to disturbed circadian rhythms. This demonstrates that our proximity labeling approach is suitable for the unbiased identification of novel circadian transcription modulators. Results Protein–DNA proximity labeling to study circadian E-box regulators To identify proteins localized at the E-box of Dbp at a given time point, we used a novel proximity labeling method in which a deactivated CAS9 nuclease is fused to an engineered ascorbic acid peroxidase (APEX2) enzyme (dCAS9–APEX2). In this approach, CRISPR–CAS9 technology is used to guide the dCAS9–APEX2 fusion protein (CASPEX complex) to a desired location using specially designed sgRNAs. APEX2 is a peroxidase that catalyzes the biotinylation of surrounding proteins when biotin-phenol and H2O2 are added to the culture medium (Fig. 1A). The biotinylated proteins can then be enriched due to their strong affinity for streptavidin and identified by proteomic methods such as immunodetection or mass spectrometry. This rapid labeling reaction (1 min) is particularly advantageous to achieve the temporal resolution required in circadian studies. Proximity labeling of E-box proteins using the CASPEX method. (A) Schematic description of the method. The sgRNAs target the dCAS9–APEX2 fusion protein downstream from the E-box of the first intron of the Dbp gene. After induction, dCAS9–APEX2 binds the Dbp gene, and upon addition of biotin-phenol and H2O2, the APEX2 peroxidase induces biotinylation of proteins in close proximity. This reaction occurs simultaneously at the 200 repeats of the Dbp gene expressed in this Dbp(luc) array NIH3T3 mouse cell line. The biotinylated proteins are then purified using their affinity for streptavidin and subsequently identified by LC-MS/MS. (B) Immunodetection of dCAS9–APEX2 and validation of its enzymatic activity. dCAS9–APEX2 was immunodetected in nuclear lysates of Dbp(luc) array NIH3T3 mouse cells using an anti-CAS9 antibody. In the same samples, biotinylated proteins were detected using horseradish peroxidase (HRP)-conjugated streptavidin followed by chemiluminescence imaging. All cells were treated with biotin-phenol, and H2O2 was added where indicated. (C) Circadian period of different cell lines. Period values were extracted using Chronostar software from 3 day bioluminescence time series recorded 24 h after synchronization and dCAS9–APEX2 induction. No significant differences were detected. One-way ANOVA, n = 5, P > 0.05; i.e., not significant (NS). (D) Example of confocal microscopy fluorescence images of dCAS9–APEX2 (anti-CAS9 antibody) and biotinylated proteins (streptavidin, Alexa fluor 514 conjugate) in Dbp(luc) array NIH3T3 mouse cells expressing both dCAS9–APEX2 and guide sgRNAs 24 h after dCAS9–APEX2 induction and fixed after addition of APEX2 substrates (biotin-phenol and H2O2). High-intensity aggregates were observed in the nucleus, similar to the BMAL1-YFP aggregates on the Dbp(luc) array in the NIH3T3 cell model (Stratmann et al. 2012). (E,F) Quantification of aggregates of dCAS9–APEX2 (E) and biotinylated proteins (F) from confocal microscopy images (as shown in D). For each field of view, the percentage of nuclei containing an aggregate of dCAS9–APEX2 or biotinylated proteins was quantified and is expressed as log2 fold change compared with the control cell type (dCAS9–APEX2 noGuide). n = 3–4 biological replicates, each with five to 12 fields of view per cell line. (G) Chromatin immunoprecipitation experiment on NIH3T3 Dbp(luc) array cells 24 h after treatment with doxycycline (purple) or solvent (gray). Enrichment of Dbp's first intron E-box was analyzed by qPCR (expressed as fold enrichment compared with the respective solvent-treated sample) for immunoprecipitation with CAS9 antibody. n = 3 biological replicates. To improve the enrichment of E-box-binding proteins, we used a Dbp(luc) array NIH3T3 mouse cell line, a creation of the Schibler laboratory (also characterized in our laboratory), which contains ∼200 copies of the Dbp gene, a known E-box-regulated gene (Stratmann et al. 2012; Klemz et al. 2021). In previous studies, such tandem arrays were used to monitor the binding of fluorescently labeled transcription factors to DNA (McNally et al. 2000; Bosisio et al. 2006; Karpova et al. 2008; Stratmann et al. 2012). We selected this cell line to overcome potential sensitivity limitations associated with proximity labeling technology. Dbp(luc) array NIH3T3 cells were stably transfected with an inducible dCAS9–APEX2 construct, which allowed the induction of dCAS9–APEX2 expression by addition of doxycycline to the culture medium. We first determined the concentration of doxycycline required for robust expression of dCAS9–APEX2 (0.25 µg/mL) in the nuclear fraction (Supplemental Fig. S1). We then validated the enzymatic activity or biotinylation activity of dCAS9–APEX2. Using Western blot analysis, we assessed the amount of biotinylated proteins in the nuclear lysate as a function of the expression of the CASPEX complex and/or the addition of its substrates (biotin-phenol and H2O2). The presence of both dCAS9–APEX2 and the substrates resulted in a significant increase in the total intensity of biotinylated proteins (Fig. 1B; Supplemental Fig. S2). To specifically direct the dCAS9–APEX2 fusion protein to the E-box in the first intron of the Dbp gene, we designed sgRNA guides targeting positions at different distances from the E-box. Based on the APEX2 biotinylation range estimated by Myers et al. (2018), the sgRNA sequences were positioned within 500 bp of the targeted E-box to ensure effective biotinylation by the APEX2 peroxidase (Fig. 1A). Next, we validated the ability of our sgRNA to guide the dCAS9–APEX2 complex to the Dbp(luc) array by quantifying the accumulation of both dCAS9–APEX2 and biotinylated proteins on the Dbp(luc) array. Biotinylation was induced in the cells 24 h after CASPEX induction by doxycycline, followed by immediate fixation of the cells. The dCAS9–APEX2 complex was immunodetected with a CAS9-specific antibody, and biotinylated proteins were detected with streptavidin-conjugated fluorophores. Using confocal microscopy, we counted fluorescent aggregates in Dbp(luc) array NIH3T3 cells expressing the dCAS9–APEX2 complex (Fig. 1D). Cells expressing guide1 at +40 bp or guide2 at +440 bp showed an increased amount of CASPEX aggregates (1.9-fold and 1.3-fold change, respectively) (Fig. 1E) and a higher number of aggregates of biotinylated proteins (3.7-fold and 2.7-fold change, respectively) (Fig. 1F) compared with cells expressing a noGuide construct. Using ChIP, we further confirmed the accumulation of the dCAS9–APEX2 complex on Dbp’s first intron region. We compared enrichment of Dbp’s first intron region after CAS9 immunoprecipitation in noGuide cells with cells expressing guide1 or guide2, treated with doxycyclin or solvent. dCAS9–APEX2 induction by doxycyclin increased its accumulation on Dbp’s first intron only in the presence of the guides designed to target this region (Fig. 1G), indicating that these guides effectively bring dCAS9–APEX2 to the E-boxes of the Dbp(luc) array to induce biotinylation of the proteins localized on it. To ensure that the presence of CASPEX near the E-box of Dbp does not interfere with the circadian machinery of rhythmic Dbp expression, we compared the rhythmicity of cells expressing guide1 and guide2 with cells expressing the noGuide construct. Because the Dbp(luc) array NIH3T3 cells were equipped with a luciferase reporter in the sequence of each Dbp gene repeat, we could monitor Dbp E-box-dependent transcription via live-cell bioluminescence recording. Cells expressing guide1, guide2, or noGuide were synchronized with dexamethasone, and bioluminescence signals were monitored for ∼4 days. Circadian rhythmicity in cells expressing sgRNA guides directing CASPEX near the E-box of Dbp did not differ from the control; i.e., there were no significant changes in period length (Fig. 1C) or other circadian parameters (Supplemental Fig. S3). Identification of proteins located near the E-box of the Dbp gene Our next goal was to identify the proteins located at the E-box of the Dbp gene. To this end, we synchronized the circadian rhythms of our cell lines with dexamethasone, and 24 h later, at the time of maximal BMAL1 binding to E-boxes (Stratmann et al. 2012), we induced biotinylation of proteins in the vicinity of dCAS9–APEX2 targeted close to the E-box of the Dbp gene (Supplemental Fig. S4). The biotinylated proteins were subsequently enriched from the nuclear lysates with streptavidin-coated magnetic beads and identified by label-free quantitative LC-MS/MS proteomics (Fig. 2A). The experiment included cells expressing two different sgRNA guides (guide1 and guide2) and a control cell type transduced with an empty vector guide (noGuide). A total of 2687 proteins was detected in at least three replicates of at least one cell type (with a standard deviation of <1 between replicates). In guide1, guide2, and noGuide cell types, we identified 2074, 2281, and 2105 proteins, respectively, which were detected in at least three replicates (Fig. 2B; Supplemental Table S1), suggesting a substantial enrichment of nonspecific proteins, possibly due to APEX2-independent biotinylation, high overall expression of dCAS9–APEX2 in the nucleus, or contamination by nonbiotinylated proteins. Identification of proteins located at the E-box of the Dbp gene using mass spectrometry. (A) Experimental pipeline. (B) Number of identified proteins in total, per cell type, and overlapping between cell types. Proteins were considered reliably identified if they were detected in three or more replicates of a cell type and the standard deviation between replicates of the same cell type was <1. (C) Fold change of the detection intensity of all samples compared with the mean intensity of noGuide for the 1685 proteins that were detected in all three cell types. Warm colors indicate a positive fold change compared with noGuide cells, and cold colors indicate a negative fold change (three or more replicates per cell type). (D) Volcano plot of the log2 fold change of proteins versus the −log10 Q-value in guide1 and guide2 compared with noGuide cells. The proteins with significant enrichment (i.e., fold change of >2 and Q-value < 0.1) are highlighted (three or more replicates per cell type). (E) Abundance (in log2 LFQ intensity) of the 153 proteins that were detected in both guide1 and guide2 in three or more replicates but had detection in less than three replicates of noGuide cells (three or more replicates per cell type). To specifically identify proteins in the vicinity of the E-box of Dbp, we required that a protein must be detected in both guide1 and guide2 cell types and either statistically lower or absent in the noGuide control. For the 1685 proteins detected in all three cell types, we analyzed the fold change in intensity of each replicate compared with the mean intensity of the noGuide control and found that ∼60% of these proteins had lower levels in guide1 and guide2 compared with noGuide cells, whereas ∼40% showed enrichment in guide1 and guide2 compared with noGuide cells (Fig. 2C). To identify additional candidates located near the E-box, we used both their fold change (≥2) and their statistical enrichment in the guide compared with the noGuide cells using the false discovery rate (FDR) approach (Q-value < 0.1) as further criteria. This analysis identified 33 proteins enriched around Dbp’s first intron E-box (Fig. 2D; Supplemental Table S1). In addition, 153 proteins were detected in three or more replicates in both guide1 and guide2 cells but in less than three replicates in noGuide cells (Fig. 2E; Supplemental Table S1). No statistical analysis was possible for these proteins, but the complete lack of detection in noGuide samples for 36 proteins from this list suggests an enrichment of proteins specific to the presence of guides targeting Dbp’s first intron E-box. Thus, we considered these proteins to be specific for CASPEX localization to the E-box of Dbp. Taken together, we identified 69 proteins that are enriched in both guide1 and guide2 cells and are absent or significantly lower in the noGuide control cells (Supplemental Table S1). These proteins are presumably located near the E-box of Dbp, and their role in the circadian transcriptional machinery can be further investigated. Proteins enriched at the E-box of the Dbp gene Among the enriched proteins, we found the central transcription protein BMAL1, as expected for this time point (Fig. 3A), demonstrating the specificity of the approach. In addition to BMAL1, our protein list included six other transcription factors or coactivators (YAP1, RUNX2, SMAD1, CUX1, SRCAP, and ARID5B) (Fig. 3A) and three chromatin modifiers (HDAC3, HISTONE H2A.V/Z, and SMARCAL1) (Fig. 3B). Furthermore, we highlighted three proteins involved in post-translational modifications (i.e., the kinase MINK1, the ubiquitin-conjugating enzyme UBE2L3, and the ubiquitin–protein ligase DTX2) (Fig. 3C), as well as three proteins associated with protein transport (XPO7, APPL1, and MVP) (Fig. 3D). (A–D) Identified proteins and their functional roles. Log2 LFQ intensity values of the proteins identified in guide1 and guide2 compared with noGuide control cells. (ND) Not detected; statistical significance is defined as fold change of ≥2 and a Q-value < 0.1. Interestingly, the detection of the histone variants H2A.Z and HDAC3 is consistent with their known roles in the circadian transcriptional machinery in liver cells (Shi et al. 2016; Tartour et al. 2022), suggesting a mechanism that goes beyond liver specificity. In addition, both RUNX2 and YAP1 have been identified as transcription factors that influence circadian rhythmicity. In the mouse suprachiasmatic nucleus, knockdown of RUNX2 resulted in altered mPER2:luc rhythms in embryonic cultures, whereas Runx2+/− mice exhibited a prolonged period of free-running activity (Reale et al. 2013). YAP1 is shown to repress clock gene expression in undifferentiated pleomorphic sarcomas, thereby promoting cell proliferation (Rivera-Reyes et al. 2018), and available ChIP-seq data sets from cancer cell lines suggest localization in several E-box-regulated core clock genes (Stein et al. 2015; Zanconato et al. 2015). It remains to be determined whether any of these effects on circadian rhythmicity are mediated by involvement in the circadian transcriptional machinery in healthy cells. In addition, for several of the identified proteins, we found reports of protein–protein interactions with other candidates from our list, such as RUNX2 with YAP1 and SRCAP (Guo et al. 2021; Thomsen et al. 2021), SMAD1 with YAP1 and MINK1 (Alarcón et al. 2009; Kaneko et al. 2011; Chen et al. 2023), HISTONE 2AZ with SRCAP (Liang et al. 2016; Tartour et al. 2022), and HDAC3 with APPL1 (Fan et al. 2019). Overall, for many proteins identified using this approach, we found data in the literature that indeed suggest their localization at the E-box of Dbp and for their interaction with each other. To experimentally test whether an identified protein associates with a specific genomic region, we used ChIP as an orthogonal method. Some of our identified proteins, such as BMAL1, H2A.Z, and HDAC3, have previously been shown by ChIP to localize to E-boxes. Here, we investigated two additional candidates (RUNX2 and SMAD1) for which high-affinity ChIP-grade antibodies were available, which is often limited to known transcription factors or coactivators. Using NIH3T3 Dbp-luciferase (Dbp-luc) cells, we performed ChIP 24 h after synchronization and quantified the binding of RUNX2 and SMAD1 to the first intron of the Dbp gene (encompassing the E-box) using qPCR. As a negative control, we used a same-species nonspecific IgG, and as a positive control, we immunoprecipitated BMAL1. As expected, BMAL1 immunoprecipitation resulted in a significantly (P < 0.05) higher enrichment of the Dbp E-box sequence compared with the nonspecific IgG (Fig. 4A). Similar significant enrichment was also observed with antibodies against RUNX2 (Fig. 4B) and SMAD1 (Fig. 4C), indicating a localization of these proteins to the E-box of the Dbp gene. Protein location on the E-box of the Dbp gene. Chromatin immunoprecipitation experiment on NIH3T3 Dbp(luc) array cells 24 h after synchronization. Enrichment of Dbp’s first intron E-box was analyzed by qPCR, expressed as fold enrichment compared with the corresponding IgG, for immunoprecipitation with BMAL1 antibody (A), RUNX2 antibody (B), and SMAD1 antibody (C). n = 3 biological replicates; t-test, P < 0.05. New regulators of the circadian oscillator We were interested in proteins that localize to E-boxes because they may be novel regulators of circadian transcription and thus modulators of circadian dynamics. To test this, we performed loss-of-function experiments (knockdown via RNA interference) for some of the identified candidates in human U-2 OS cells, an established and widely used cell model for circadian clocks (Maier et al. 2021). These cells express a Bmal1-luciferase reporter construct, allowing real-time monitoring of luminescence and quantification of rhythmicity. To minimize potential off-target effects that could lead to false positives or negatives, we used multiple shRNAs to knock down each target protein. From the identified proteins highlighted in Figure 3, we selected candidates for knockdown based on prescreening of knockdown efficiency and normal cell growth after knockdown (data not shown). Knockdown of RUNX2 using shRNAs resulted in knockdown efficiencies ranging from 50% to 90% (Supplemental Fig. S5A) and showed a correlation between RUNX2 transcript abundance and BMAL1 luciferase signal intensity (Fig. 5A), suggesting a direct or indirect role for RUNX2 for BMAL1 transcription. Furthermore, a lower RUNX2 transcript level correlated with a higher sinusoid fitting error; i.e., at low RUNX2 transcript levels, we could no longer detect robust rhythmicity (Fig. 5A). Similarly, downregulation of the APPL1 transporter resulted in knockdown efficiencies ranging from 50% to 95% (Supplemental Fig. S5B). No specific effect on the Bmal1 luciferase signal intensity or sinusoid fit was observed; however, four out of five shRNA constructs resulted in a shortening of the circadian period by ∼1 h (Fig. 5B), also suggesting a role for APPL1 in controlling circadian dynamics. For the nuclear transporter XPO7 and the kinase MINK1, knockdowns with four different shRNAs led to a reduction of the respective transcript levels by ∼90%. XPO7 knockdown led to a period shortening of >0.5 h in three out of four constructs (Fig. 5C; Supplemental Fig. S5C), and MINK1 knockdown led to a period shortening of >0.7 h in three out of four constructs (Fig. 5D; Supplemental Fig. S5D). This suggests that normal transcript levels are critical for normal circadian period length also for these two E-box-associated candidates. Depleting identified E-box-enriched proteins affects circadian dynamics. Real-time bioluminescence monitoring of U-2 OS Bmal1 luciferase cells after synchronization with dexamethasone. (A) Knockdown of Runx2. (Left) Representative time series. (Middle) Change in Bmal1 luciferase signal intensity compared with nonsilencing control as a function of the percentage of RUNX2 transcript. (Right) Error of fit to a sinosoidal curve as a function of the percentage of RUNX2 transcript. The red line represents a fit of the data to a monophase decay curve. (B–D) Knockdown of APPL1 (B), XPO7 (C), and MINK1 (D). (Left) Representative time series. (Right) Change in period length compared with the nonsilencing controls as a function of the shRNA construct. The dotted red line represents the mean change in period length for all shRNA combined. Dots represent the results of independent experiments. Discussion The identification and characterization of proteins involved in the circadian transcriptional machinery on key cis-regulatory elements depends on the ability to selectively enrich proteins that are in close proximity to a particular genomic sequence at a given time point. Previous methods aiming at this task relied on chemical cross-linking to immobilize proteins in their spatial and temporal context. However, chemical cross-linking leads to changes in the protein structure that can interfere with the identification by mass spectrometry (MS) (Toews et al. 2008). Here, we used the biotinylating enzyme APEX2 targeted close to the E-box of the Dbp gene by a catalytically inactive CAS9 and specifically designed sgRNAs. In contrast to previous approaches, this method does not require chemical cross-linking or the availability of specific antibodies. In addition, the reaction time of APEX2 of only 1 min allows labeling of proximal proteins with a high time resolution. This unbiased characterization of the interactome at a specific chromosomal locus complements the chromatin immunoprecipitation (ChIP)-seq method, which delineates the multiple binding regions of a particular protein of interest. In this first study, we focused on the E-box of the Dbp gene, a known E-box-regulated gene whose binding to core clock proteins has previously been characterized by ChIP-seq strategies. To increase the enrichment efficiency, we used a Dbp(luc) array in the mouse fibroblast NIH3T3 cell line, which expresses 200 copies of the Dbp gene as tandem repeats. After establishing stable expression of dCAS9–APEX2 and the sgRNAs in this cell model, we observed, as expected, substantial accumulation of the fusion protein on the Dbp(luc) array but also in other cell compartments (especially in the cytoplasm) (data not shown), leading to nonspecific biotinylation of proteins. To distinguish E-box proteins from nonspecific biotinylated proteins, we used two cell lines with different sgRNAs for chromosomal targeting, resulting in overlapping labeling radii. Using our criteria, we identified a total of 69 proteins that are most likely enriched at the E-box of the Dbp gene. Using the extensive literature on E-box binding proteins, we found evidence for localization at the E-box of Dbp for several of the proteins identified here. The key proteins expected at the E-box of Dbp at this specific time point are BMAL1 and CLOCK. Original work on the Dbp(luc) array NIH3T3 cell line showed that overexpressed BMAL1-YFP reaches its highest binding to the Dbp gene array ∼24 h after cell synchronization by dexamethasone (Stratmann et al. 2012). Accordingly, using the same cell line and inducing proximity labeling 24 h after cell synchronization, we enriched endogenous BMAL1 specifically in both cell types where the labeling enzyme is brought into proximity to the E-box of Dbp. In ChIP-seq time series experiments using mouse livers, both BMAL1 and CLOCK were detected at the E-box of the first intron of Dbp at the time of highest BMAL1 binding (Koike et al. 2012). However, in our data set, CLOCK was not enriched in any sample of the cell lines analyzed, including in the noGuide control, indicating that the amount of biotinylated endogenous CLOCK was below the detection limit. In addition to the core clock protein CLOCK and BMAL1, it has been described that p300 interacts with CLOCK–BMAL1 at E-boxes during this phase of circadian transcription (Koike et al. 2012). In our data set, p300 is robustly detected in all cell types, including the noGuide control, but was not significantly enriched in cells expressing E-box-specific sgRNA (Supplemental Table S1). This could mean that p300 is not specifically localized to the E-box of the first intron of Dbp, which is consistent with the p300 ChIP-seq results from mouse livers (Koike et al. 2012). However, it could also be an artifact resulting from the inverse relationship between protein abundance and the ability to determine when it is enriched over background. Similarly, six of the RNA polymerase II subunits were detected with high intensity in all cell types and replicates, with no statistical enrichment in the guide cell lines, likely reflecting their overall abundance. The H2A.Z variant histone 1 and the isoform H2A.Z variant histone 2 were found to be significantly enriched in our sgRNA cells compared with noGuide cells. This is consistent with H2A.Z ChIP-seq data showing that in mouse livers, H2A.Z not only flanks the core clock binding sites of E-box-regulated genes such as Per2 or Cry1 but is also required for the binding of BMAL1 to E-box-controlled genes, including Per1, Per2, and Dbp (Tartour et al. 2022). Furthermore, the same study suggests that H2A.Z nucleosome incorporation may be regulated by SRCAP and/or the p400-TIP60 chromatin remodeler complexes, which agrees with the specific enrichment of SRCAP at the E-box of Dbp in our data set. HDAC3, a histone deacetylase enzyme, was identified exclusively in both sgRNA cell types but not in the noGuide control. In liver and muscle tissue, HDAC3 has been detected by ChIP-seq at many RORE sites together with REV–ERBα but also at many E-box-regulated genes (Feng et al. 2011; Hong et al. 2017). Although HDAC3 has been shown to be recruited to the repressive REV–ERBα complex, it has also been described as associating with CLOCK and BMAL1 in the nucleus (Hosoda et al. 2009; Shi et al. 2016) and as promoting the binding of BMAL1 to E-boxes in the main binding phase, independent of its enzymatic deacetylase activity (Shi et al. 2016). We used ChIP to validate the localization of several newly identified proteins for which we were able to obtain specific high-affinity ChIP-grade antibodies. We validated the location of the transcription factor RUNX2 and the transcriptional coactivator SMAD1. Interestingly, we found that RUNX2 is important for maintaining a normal circadian oscillation in human U-2 OS cells, consistent with findings in Runx2+/− mice (Reale et al. 2013). Taken together, these results suggest that RUNX2's contribution to the circadian oscillator is mediated through an action on E-boxes. Many of the proteins that we have identified have not yet been described as linked to the transcriptional machinery. Therefore, for most of them, high-affinity ChIP-grade antibodies do not exist, which prevents the use of the ChIP method to validate their binding to chromatin and specifically to the E-box region. We therefore used alternative methods (i.e., genetic loss-of-function experiments) and identified three proteins from our list that are required for normal circadian dynamics: the transporter proteins XPO7 and APPL1 and the kinase MINK1. Interestingly, it has already been shown by others that XPO7 enhances the activation of E-boxes by the bHLH transcription factor TCF3 (Lee et al. 2010), supporting data showing that transporters can also influence transcriptional activity (Tamanini et al. 2005), albeit indirectly. APPL1 has been described as enriched in the nucleus of brown adipose tissue cells to directly interact with HDAC3 (a known E-box regulator also detected in our study) and as regulating its activity (Fan et al. 2019). Because downregulation of APPL1 led to a significant shortening of the circadian period, this result may indicate a role of APPL1 in E-box transcription via interaction with HDAC3. Our results also support a role of MINK1 in the circadian transcriptional machinery, possibly through its modulation of SMAD1's phosphorylation state (Kaneko et al. 2011). Overall, our study represents a novel strategy to characterize the landscape of the circadian cistrome and to identify novel regulators of the complex machinery of circadian transcription in an unbiased manner. This approach not only identifies known E-box regulators but also allows the discovery of novel players in the circadian transcriptional machinery, including proteins with no described role in transcriptional mechanisms. Our study is therefore an excellent starting point not only to discover new circadian modulators but also to characterize the chromatin binding dynamics of these regulators and thus obtain a complete time-resolved picture of circadian transcriptional regulation. Materials and methods Cell lines and culture Mouse fibroblast NIH3T3 cells stably expressing a tandem array of ∼250 Dbp copies [Dbp(luc) array NIH3T3] were a kind gift from Ueli Schibler (University of Geneva) (Stratmann et al. 2012). Exon 4 of the Dbp gene contains an open reading frame for luciferase (luc) preceded by an internal ribosome entry site (IRES) to allow monitoring of transgene expression on the array by luminescence recording. Human osteosarcoma U-2 OS Bmal1-luciferase cells (ATCC HTB-96) stably expressing firefly luciferase from a 0.9 kb mouse Bmal1 promoter fragment were a kind gift from Steven Brown (University of Zurich). All cell types were grown in DMEM with 10% fetal bovine serum, 25 mM HEPES, and antibiotics (100 U/mL penicillin, 100 μg/mL streptomycin). Generation of the Dbp(luc) array NIH3T3 dCAS9–APEX2 cell line The inducible CASPEX expression plasmid (Addgene 97421) was stably transfected into Dbp(luc) array NIH3T3 cells using Lipofectamine 3000 from Invitrogen. Nontransduced cells were eliminated by both antibiotic selection with 10 μg/mL puromycin and fluorescence-activated cell sorting (FACS). The inducible expression of dCAS9–APEX2 after doxycycline treatment was confirmed by SDS-PAGE and Western blot. The same cell line was used for all subsequent experiments. Design and cloning of guide RNAs Guide RNA sequences were generated using the CRISPOR web tool (Concordet and Haeussler 2018) and designed to target the E-box of the first intron of the Dbp gene with a maximum distance of 500 bp. Sequences with a high on-target score were cloned into lentiGuide-Hygro-dTomato (Addgene 99376), following the guidelines of Shalem et al. (2014). Production and transduction of lentiviruses Lentiviruses were produced in HEK293T cells as described by Maier et al. (2021). The virus-containing supernatants were filtered and the target cells were transduced with the virus filtrate. Dbp(luc) array NIH3T3 dCAS9–APEX2 cells were stably transduced with guide RNA-containing lentiviruses. Nontransduced cells were eliminated by antibiotic selection (0.25 mg/mL hygromycin) as well as by FACS. U-2 OS Bmal1-luciferase cells were transduced with a lentivirus containing shRNA and selected for 3 days with 10 μg/mL puromycin. APEX2-mediated labeling Prior to protein labeling, circadian rhythms were synchronized by incubating cells with 1 μM dexamethasone (Sigma-Aldrich D4902) for 20 min. The cells were then rinsed twice with warm PBS, and normal culture medium containing 0.25 µg/mL doxycycline was added to induce dCAS9–APEX2 expression (Clontech 631311). Twenty-four hours later, biotin-tyramide-phenol (Iris Biotech GNBH 41994-02-9) resuspended in DMSO was added to the culture medium at a final concentration of 500 μM. After incubation for 30 min at 37°C, biotinylation of the proteins was induced by adding 1 mM H2O2 for exactly 60 sec. After 60 sec, the medium was aspirated and the cells were washed three times with ice-cold quenching buffer that inactivated APEX2 activity (TBS with 1 mM CaCl2, 10 mM sodium ascorbate, 1 mM sodium azide, 1 mM Trolox, 1 mM DTT). Cell lysis and enrichment of nuclear proteins Nuclear proteins were isolated according to the Abcam nuclear fractionation protocol. On ice, cell medium was replaced with fractionation buffer (10 mM HEPES at pH 7.4, 1.5 mM MgCl2, 10 mM KCl, 0.5 mM DTT, 0.05% NP40, 1:100 protease inhibitor cocktail), and cells were scraped, collected, and incubated for 10 min on ice. After centrifugation at 720g for 10 min at 4°C, the supernatant containing cytoplasm, membranes, and mitochondria was aspired and discarded. The pellet containing the nuclei was suspended in a lysis buffer (5 mM HEPES at pH 7.4, 1.5 mM MgCl2, 0.2 mM EDTA, 0.5 mM DTT, 26% glycerol, 1:100 protease inhibitor cocktail) containing 300 mM NaCl. Samples were homogenized with a 27 gauge needle and incubated for 30 min on ice. After centrifugation at 24,000g for 20 min at 4°C, the supernatants containing nuclear proteins were separated. The protein concentration was determined using a BCA assay. SDS-PAGE and Western blot Equal amounts of proteins (50–100 µg) were denatured by heating for 5 min at 95°C in 1:4 SDS loading buffer. Protein separation was performed by SDS-PAGE. Proteins were transferred to a nitrocellulose membrane overnight in 5% methanol blotting buffer. Protein smears were visualized by incubation in Ponceau for 10 sec. To block unspecific binding sites, the membranes were incubated for 1 h in TBS-T containing 5% bovine serum albumin (BSA). The membranes were then incubated overnight with primary antibodies in TBS-T (αCAS9 at 1:1000 and αHDAC1 at 1:1000), washed, and then probed with the corresponding HRP-conjugated secondary antibodies (1:1000). Biotinylated proteins were detected by incubating the membranes with HRP-conjugated streptavidin (1:10,000). A chemiluminescence reaction was performed for protein visualization. Chromatin immunoprecipitation (ChIP) NIH3T3 cells were grown in 15 cm dishes to 95% confluence. Twenty-four hours before collection, cells were synchronized and dCAS9–APEX2 was induced, as described previously. Cross-linking was induced by incubation in 1% paraformaldehyde for 10 min and then stopped by the addition of glycine at 125 mM. Cells were scraped and lysed in a series of cell lysis buffers with a composition favorable to chromatin solubilization (see the Supplemental Material for a complete list of buffers). DNA was sheared using a Bioruptor Plus sonication system (Diagenode) for 20 cycles of 15 sec on and 15 sec off at the high setting. Cell debris was removed by centrifugation at 13,000 rpm for 10 min. Supernatants were diluted 1:8, precleared with agarose A or G beads (sc-2001 and sc-2002) for 1 h at room temperature, incubated with 2 μg of antibody (IgG mouse, IgG rabbit, RUNX2 [12556], BMAL1 [sc-365645], or CRISPR/Cas9 [C15310258-100]) for 1 h, and then incubated with the addition of agarose beads for another 1 h at room temperature with rotation. Precipitates were washed sequentially in a series of TSE buffers (see the buffer list in the Supplemental Material) before cross-linking was reversed overnight at 65°C in cross-linking reversal buffer. DNA fragments were purified using a QIAquick spin kit (Qiagen). qPCR was performed with SYBR Green (Fermentas) using a CFX384 real-time PCR instrument (Bio-Rad) with a primer pair targeting Dbp’s first intron E-box primers 5′-GAGGAATGTGCCTGAGCAGT-3′ (forward) and 5′-GGGTTTAGTGACAGGGTGGG-3′ (reverse). Enrichment of biotinylated proteins for LC-MS/MS analysis Twenty 20 cm dishes with confluent cells were used for each replicate of the proteomic experiments to ensure an input of 5 mg of nuclear proteins for each sample. To each sample, 500 µL of Dynabeads MyOne streptavidin T1 (Invitrogen 65601) was added, followed by overnight incubation at 4°C on a rotating wheel. Using a magnetic support, the streptavidin beads (and attached biotinylated proteins) were washed twice with cold lysis buffer, once with 1 M cold KCl, once with 100 mM Na3CO2, and twice in 2 M cold urea in 50 mM ammonium bicarbonate (ABC) (as described by Myers et al. 2018). The washing solution was removed and the proteins on the beads were resuspended in 20 μL of urea buffer (2 M urea, 50 mM ammonium bicarbonate at pH 8.0), reduced in 12 mM dithiothreitol for 30 min at 25°C, and then alkylated with 40 mM chloroacetamide for 20 min at 25°C. The samples were digested with 0.5 μg of trypsin/LysC overnight at 30°C. The peptide-containing supernatant was collected and the digestion was stopped with 1% formic acid. The peptides were desalted and purified according to the stage tip protocol (Rappsilber et al. 2003). Samples were eluted with 80% acetonitrile/0.1% formic acid, dried with a Speedvac, resuspended in 3% acetonitrile/0.1% formic acid, and analyzed by LC-MS/MS. Liquid chromatography–mass spectrometry Peptides were analyzed on a reversed-phase column, a 20 cm long fritless silica microcolumn with an inner diameter of 75 μm, packed with ReproSil-Pur C18-AQ 3 μm resin (Dr. Maisch GmbH) using a 90 min gradient at a flow rate of 250 nL/min with increasing concentrations (2%–60%) of buffer B (90% acetonitrile/0.1% formic acid) on a high-performance liquid chromatography (HPLC) system (Thermo Fisher Scientific), ionized with an electrospray ionization (ESI) source (Thermo Fisher Scientific) and analyzed on a Thermo Q Exactive Plus instrument. The instrument was operated in data-dependent mode, selecting the 10 most intense ions in the MS full scans (ranging from 350 to 2000 m/z) with a resolution of 70,000, an ion count of 3 × 106, and an injection time of 50 msec. Tandem MS was performed with a resolution of 17,500. The MS2 ion counting target was set to 5 × 104 with a maximum injection time of 250 msec. Only precursors with charge state of 2–6 were selected for MS2. The dynamic exclusion time was set to 30 sec with a 10 ppm tolerance for the selected precursor and its isotopes. Raw data were analyzed with the MaxQuant software package (v1.6.3.4) (Tyanova et al. 2016) using the UniProt mouse database (MOUSE.2019-07) and the APEX2 construct sequence database with forward and reverse sequences. The search included variable modifications such as methionine oxidation, N-terminal acetylation, and asparagine and glutamine deamidation. Carbamidomethylation of cysteine was defined as a fixed modification. The minimum peptide length was set at seven amino acids and a maximum of three missed cleavages was allowed. The FDR (false discovery rate) was set at 1% for peptide and protein identification. Unique and Razor peptides were considered for quantification. Retention times were recalibrated based on the integrated nonlinear time-scaling algorithm. MS2 identifications were transferred between runs using the “Match between runs” option, with the maximum retention time window set to 0.7 min. The LFQ algorithm (label-free quantification) was activated. Immunofluorescence microscopy Prior to APEX2-mediated labeling, cells were seeded to 90% confluence on glass-bottomed chambers (Ibidi 80826) coated with fibronectin. At the end of the labeling protocol, the medium was removed and the cells were fixed by incubation in 4% paraformaldehyde in PBS (pH 7.4) for 10 min. After three washes with cold PBS, cells were permeabilized by incubating three times for 5 min in 0.1% Triton X-100 in PBS alternating with 5 min washes with cold PBS. dCAS9–APEX2 proteins were detected with a first antibody against CAS9 (1:1000) and a secondary antibody combined with 10 µg/mL fluorophores. Biotinylated proteins were detected by incubation with 5 µg/mL streptavidin Alexa fluor 514 conjugate. Cell nuclei were labeled by staining with 300 nM DAPI. Images were captured with a Nikon scanning confocal A1Rsi+ at 60×. The number of nuclei that had dCAS9–APEX2 or biotinylated protein dots was manually determined using Fiji (Schindelin et al. 2012) and is expressed as a percentage of the total number of nuclei on the image. Bioluminescence imaging Cells were seeded either on a white 96 well plate or in 35 mm NUNC dishes at the density required for confluence 1–3 days later. Circadian rhythms were then synchronized with 1 μM dexamethasone for 30 min, and cells washed with warm PBS and placed in phenolred-free DMEM culture medium supplemented with 10% fetal bovine serum, antibiotics (100 U/mL penicillin, 100 μg/mL streptomycin), and 250 μM D-luciferin (Biothema). Bioluminescence recordings were performed at 35°C–37°C in a 96 well plate luminometer (TopCount, PerkinElmer) or LumiCycle (Actimetrics). The data were analyzed with ChronoStar software (Maier et al. 2021). Briefly, the first 12 h of recording were excluded, time series data were trend-eliminated by dividing by a 24 h average, and rhythm parameters were estimated by fitting a cosine wave function with an exponential decay term. Determination of knockdown efficiency Total RNA was purified using the Pure Link RNA mini kit (Thermo Fisher Scientific 12183025) according to the manufacturer's instructions, with an additional on-column deoxyribonuclease I digestion. For quantitative reverse transcription PCR (RT-qPCR), RNA was reverse-transcribed using the Moloney murine leukemia virus reverse transcriptase kit (Thermo Fisher Scientific 28025013) with random hexamers (Thermo Fisher Scientific N8080127) according to the manufacturer's instructions. RT-qPCR was performed using the SYBR Green master mix (Thermo Fisher Scientific A46112) and a CFX96 PCR system with CFX Manager software (Bio-Rad). Proteomics data analysis We used the Perseus software (Tyanova et al. 2016) to reduce our matrix of label-free quantification (LFQ) values by filtering out reversed proteins, proteins identified only by site, and contaminants. The values were converted to log2(x) values. We filtered out proteins that were not detected in at least three replicates of any of the cell types and had a standard deviation between replicates >1. For proteins that were detected in all three cell types, we determined the statistical enrichment in guide1 or guide2 cells compared with the noGuide control using a multiple t-test with the false discovery rate approach (Benjamini–Hochberg method). Significant enrichment of a protein at the E-box position was defined by a Q-value of <0.1 in both guide cell types compared with the control (defined by an unpaired t-test) and a fold change of >2 in both cell types compared with the control. Supplemental Material