High-Throughput Chemical Screen Identifies a Novel Potent Modulator of Cellular Circadian Rhythms and Reveals CKIα as a Clock Regulatory Kinase

By applying a high-throughput circadian assay system using human U2OS cells with Bmal1-dLuc reporter [29], we analyzed approximately 120,000 uncharacterized compounds corresponding to diverse chemical scaffolds [43]–[44] at a final concentration of 7 µM. We identified a number of compounds with different scaffolds that lengthened the circadian period of cellular luminescence rhythms. Among them, we selected one purine derivative compound 1 (Figure 1A) for follow-up studies, because it strongly lengthened the period in a dose-dependent manner and showed less effect on the amplitude of Bmal1-dLuc rhythms (Figure S1). A preliminary structure-activity relationship study helped to identify a derivative of compound 1 that is 3 times more potent and able to generate >10 h period change at a concentration of 10 µM (Figure 1B and Table 1). We termed this derivative “longdaysin” (Figure 1A), based on its prominent period lengthening effect. We further investigated the effect of longdaysin (Figure 1C–E) by using primary cells and tissues isolated from mPer2^Luc knockin mice harboring a mPer2^Luc reporter [45]–[46] as an additional clock-controlled reporter different from Bmal1-dLuc used in the screen. Longdaysin consistently caused dose-dependent period lengthening in adult tail fibroblasts (Figure 1C) and lung explants (Figure 1D), which represent peripheral clocks, and in SCN explants (Figure 1E), which represent the central clock. The effect of longdaysin was reversible, as the period length returned to normal after washout of the compound (Figure S2). Taken together, these results demonstrate that longdaysin potently lengthens the circadian period in multiple mammalian cells including SCN neurons.

In order to identify potential biological targets of longdaysin by affinity-based proteomic approaches [38], we synthesized longdaysin analogs with an aminohexyl linker, based on the preliminary structure-activity relationship analysis. Among them, compound 2 with a linker at the C2 position (Figure 2A) retained the period lengthening effect in the cell-based circadian assay (Figure 2B). We then prepared agarose-conjugated compound 3 (Figure 2A) and incubated it with U2OS cell lysate in the presence or absence of 100 µM longdaysin as a soluble competitor (Figure 2C). Proteins that bound to the affinity resin, and could be competed off by free longdaysin, were separated by SDS-PAGE and analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS). This analysis yielded 10 proteins (Figure 2D) including the protein kinases (highlighted in blue) CKIδ (CSNK1D), CKIα (CSNK1A1), ERK2 (MAPK1), CDK7, and p38α (MAPK14). Independent affinity chromatography followed by Western blotting with specific antibodies confirmed both the binding of the protein kinases to the affinity resin as well as decreased binding in the presence of free longdaysin (Figure S3). Furthermore, in vitro kinase assays revealed that longdaysin inhibited CKIδ, CKIα, ERK2, and CDK7 activities (IC₅₀ = 8.8, 5.6, 52, and 29 µM, respectively; Figure 2E and Table 1), while it had much less effect on p38α (unpublished data). In contrast, compound 1 inhibited CKIδ, CKIα, and ERK2 with ∼3 times less potency than longdaysin and inhibited CDK7 similarly to longdaysin (Figure 2E and Table 1). The difference in potency between longdaysin and compound 1 against CKIδ, CKIα, and ERK2 was consistent with their cellular period effects (Table 1), suggesting an involvement of these three kinases in the period regulation.

To identify the protein(s) mediating longdaysin effect on period length, we first tested the contribution of CKIδ, a well-characterized kinase in period regulation [12],[18]–[19],[47], by using embryonic fibroblasts prepared from CKIδ deficient mice harboring the mPer2^Luc knockin reporter [18]. In a 384-well plate format, the period of CKIδ deficient (CKIδ^Δ2/Δ2) cells was 1.1 h longer than that of wild type (CKIδ^+/+) cells (CKIδ^Δ2/Δ2, 25.9±0.5 h; CKIδ^+/+, 24.8±0.9 h; n = 48), consistent with a previous report [18]. We found that longdaysin lengthened the period in a dose-dependent manner in CKIδ deficient cells as well as in wild type cells (Figure 3A). This result indicates the presence of additional longdaysin-target(s) that regulate period length besides CKIδ.

To investigate the effects of RNAi-mediated inhibition of potential longdaysin-targeted kinases on the circadian period, we conducted knockdown experiments by applying four independent siRNAs against each gene. At least two siRNAs for CSNK1D (encoding CKIδ), CSNK1A1 (CKIα), and MAPK1 (ERK2) caused period lengthening (Figure 3B, red box), while those for CDK7 and MAPK14 (p38α) did not, thus proposing CKIδ, CKIα, and ERK2 as the potential clock-acting targets of longdaysin. We also tested siRNAs against the close homologs of these three kinases (CSNK1E for CSNK1D, CSNK1A1L for CSNK1A1, and MAPK3 for MAPK1) and found that the homologs had little or no effect on the period (Figure 3B). The minor period effects for MAPK14, CSNK1E, and CSNK1A1L are in line with previous reports [23],[42]. We further looked at the primary screening data from our genome-wide RNAi study [33] in which we used four siRNAs different from this study by combining two siRNAs as a pool. Among the 10 longdaysin-interacting proteins identified by the affinity chromatography (Figure 2D), only CSNK1D, CSNK1A1, and MAPK1 showed period lengthening of the reporter with both siRNA pairs (Figure S4, red box), supporting important roles for CKIδ, CKIα, and ERK2 in period regulation as longdaysin targets. We then characterized the period lengthening effects of siRNAs against CSNK1D, CSNK1A1, and MAPK1 by using an 8-point dilution series of the effective siRNAs. All siRNAs tested gave dose-dependent changes of the period (Figure 3Ci) and reduction of the target gene mRNA levels without affecting the levels of closely homologous genes (Figure 3Cii). The only exception was CSNK1A1 si1, which has sequence similarity against CSNK1A1L mRNA (96% identical on a 23 bp stretch) and reduced its level at higher dose (Figure 3Cii). The correlation between period effect and mRNA knockdown effect matched well for two siRNAs against CSNK1A1 or MAPK1 (Figure S5). The proportional changes of circadian function by dose-dependent knockdown of CSNK1D, CSNK1A1, and MAPK1 (Figure 3C) are common characteristics among the core clock components and clock modifiers [23],[33]. Taken together, these results illustrate the involvement of CKIα and ERK2 in the period regulation, as well as confirming the importance of CKIδ.

We further tested knockdown of all three kinases CSNK1D, CSNK1A1, and MAPK1 in combination, in order to determine if their concomitant reduction could explain the strong effect of longdaysin as an inhibitor of all three kinases. Combinatorial knockdown of the three genes caused strong and dose-dependent lengthening of the period to >10 h (Figure 3Di). The multiple gene knockdown effect (Figure 3Dii, red line) matched well with the theoretical sum of the effect of single gene knockdown (black line). These results suggest that the knockdown of these three kinases works in an additive manner to cause prominent period lengthening similar to that generated by longdaysin.

In contrast to the well-characterized roles of CKIδ/ε-mediated phosphorylation of PER proteins [9]–[21], the functions of CKIα and ERK2 in period regulation have yet to be characterized. To examine the interaction of CKIα and ERK2 with the core clock proteins, we co-expressed HA-tagged kinases with Flag-tagged clock proteins in HEK293T cells. Immunoprecipitation assay revealed interactions of both CKIα and ERK2 with PER1 and PER2, and to a lesser extent, CRY1 and CRY2 (Figures 4A, S6A, and S6B). Co-expression of PER1 with CKIα generated lower electrophoretic mobility forms of PER1 (Figure 4B), which disappeared upon phosphatase treatment (Figure S7A), suggesting CKIα-dependent phosphorylation of PER1. This modification relied on the kinase activity of CKIα, because the kinase-dead K46R mutant of CKIα [CKIα (KR)] [48] did not cause the mobility-shift of PER1 (Figure 4B). In contrast, ERK2 showed no detectable effect on the PER1 mobility (unpublished data). Treatment of the cells with longdaysin reduced the CKIα- and CKIδ-dependent mobility-shift of PER1 (Figures 4C, S7B, and S7C), consistent with a potential mode of longdaysin action through CKIα and CKIδ.

As phosphorylation of PER1 modulates its stability [16]–[17],[21], we further tested the effect of CKIα on the stability of PER1 by expressing luciferase-fused PER1 protein (PER1-LUC) in HEK293T cells. Luminescence changes were monitored following inhibition of de novo protein synthesis by cycloheximide treatment. Co-expression of CKIα but not CKIα (KR) accelerated PER1-LUC degradation relative to that of LUC (Figure 4D). The CKIα- and CKIδ-dependent degradation of PER1 was inhibited by longdaysin treatment in a dose-dependent manner (Figure 4E). Similar results were obtained by using PER1 without the LUC fusion (Figure S8A). In contrast, we found that CKIα had no effect on the stability of a PER2-LUC fusion protein while CKIδ accelerated its degradation (Figure S8B). These results demonstrated the selectivity of CKIα against PER1 degradation over PER2. Longdaysin inhibited the CKIδ-mediated PER2-LUC degradation in a dose-dependent manner (Figure S8C), suggesting its role in regulating both PER1 and PER2 stabilities through CKIδ and/or CKIα.

We then investigated the effect of longdaysin on the protein level of endogenous PER1 in Bmal1-dLuc U2OS cells during circadian cycles. The cells were synchronized with medium change and collected every 6 h from 28 h to 58 h after the medium change (Figure 4Fi). Consistent with the longdaysin-dependent shift of the second trough of Bmal1-dLuc luminescence rhythm (34, 38, and 46 h for 0, 3, and 9 µM longdaysin, respectively; Figure 4Fi), the second peak of PER1 protein rhythm shifted in parallel (40, 40–46, and 52 h for 0, 3, and 9 µM longdaysin, respectively; Figure 4Fii). Furthermore, 3 or 9 µM longdaysin treatment strongly up-regulated overall protein amount of PER1 compared with 0 µM control (Figure 4Fii) without affecting its mRNA level (Figure 4Fiii), demonstrating post-transcriptional increase of endogenous PER1 by longdaysin. The progressive phosphorylation of PER1 was still observed in the presence of longdaysin (Figure 4Fii), possibly because of the phosphorylation by kinase(s) that was not affected by longdaysin. Collectively, these results provide a possible mechanism of longdaysin action for period regulation through the CKIα- and CKIδ-mediated control of PER1 stability.

Lastly, we investigated if longdaysin had any in vivo efficacy by using zebrafish, which provide a useful model system for studies on circadian rhythms at the level of the whole organism [49], and have conserved CKI and ERK family genes [50]–[51]. By using transgenic zebrafish harboring a per3-luc reporter [52], we first established an in vivo circadian assay to investigate the effects of compounds. Larval per3-luc fish were entrained in 12 h light/12 h dark cycles from day 3 to 6 postfertilization and then placed in an individual well of a 96-well plate to monitor luminescence rhythms under constant darkness. By using this assay, we found that longdaysin treatment caused >10 h period lengthening in a dose-dependent manner in per3-luc reporter fish (Figure 5A and 5B), without affecting body size (Figures 5C and S9). The in vivo period changes were similar to those observed in mammalian tissues and cells (Figure 1), showing the prominent characteristics of longdaysin as a period lengthening compound.

All animal studies were approved by the University of California, San Diego, Institutional Animal Care and Use Committee and performed in accordance with the guidelines.

Synthesis of compound 1, longdaysin, compound 2, and compound 3 is described in Text S1. The dilution series of the compounds was made on 384-well plates by using a robotic liquid handling system (MiniTrak, Perkin-Elmer).

The compound screen was done with the high-throughput circadian assay system as described previously [29]. In brief, Bmal1-dLuc U2OS cells were suspended in the culture medium (DMEM supplemented with 10% fetal bovine serum, 0.29 mg/ml L-glutamine, 100 units/ml penicillin, and 100 µg/ml streptomycin) and plated onto 384-well white solid-bottom plates at 20 µl (2,000 cells) per well. After 2 d, 50 µl of the explant medium (DMEM supplemented with 2% B27, 10 mM HEPES, 0.38 mg/ml sodium bicarbonate, 0.29 mg/ml L-glutamine, 100 units/ml penicillin, 100 µg/ml streptomycin, 0.1 mg/ml gentamicin, and 1 mM luciferin, pH 7.2) was dispensed to each well, followed by the application of 500 nl of compounds (dissolved in DMSO; final 0.7% DMSO). The plate was covered with an optically clear film and set to luminescence monitoring system equipped with a CCD imager (ViewLux, Perkin Elmer). The luminescence was recorded every 2 h for 3–4 days. In follow-up studies, the luminescence was recorded every 100 min by using a microplate reader (Infinite M200, Tecan). The period parameter was obtained from the luminescence rhythm by curve fitting program CellulaRhythm [29] or MultiCycle (Actimetrics), both of which gave similar results.

Luminescence rhythms of adult tail fibroblasts [46] and embryonic fibroblasts [18] from mPer2^Luc knockin mice were analyzed similarly to U2OS cells, except that 1,800 cells were plated per well. Because of the low luminescence intensity of the fibroblasts, the higher sensitivity ViewLux imager was used for rhythm recording.

Explants of lung and SCN were dissected from mPer2^Luc knockin mice [45] and cultured in explant medium as described previously [46]. The medium was changed every week with increasing concentration of longdaysin each time (from 0 to 9 µM, final 0.7% DMSO). The luminescence was recorded every 10 min with LumiCycle luminometer (Actimetrics), and the period parameter was obtained by using LumiCycle Analysis software (Actimetrics).

U2OS cells kept in confluence (2×10⁸ cells) were collected with ice-cold PBS and homogenized by using Dounce homogenizer in 5 ml of lysis buffer (25 mM MOPS, 15 mM EGTA, 15 mM MgCl₂, 1 mM DTT, 60 mM β-glycerophosphate, 15 mM p-nitrophenyl phosphate, 1 mM Na₃VO₄, 1 mM NaF, 1 mM phenyl phosphate, 10 µg/ml leupeptin, 10 µg/ml aprotinin, 10 µg/ml soybean trypsin inhibitor, 100 µM benzamidine, pH7.2). The homogenate was sonicated and centrifuged (16,000×g) at 4°C for 20 min. The resulting supernatant was split into two, and each portion was incubated with or without 100 µM longdaysin (final 0.1% DMSO) at 4°C for 10 min (Figure 2C). Then, 120 µl of compound 3 [50% slurry in bead buffer (50 mM Tris, 250 mM NaCl, 5 mM EDTA, 5 mM EGTA, 0.1% NP-40, 5 mM NaF, 10 µg/ml leupeptin, 10 µg/ml aprotinin, 10 µg/ml soybean trypsin inhibitor, 100 µM benzamidine, pH 7.4)] was added to the mixture and incubated at 4°C for 1 h with rotation. The agarose beads were washed 6 times with 2 ml of the bead buffer. The bound proteins were eluted with SDS sample buffer and separated by SDS-PAGE (4%–12% gradient gel, Invitrogen). The gel was CBB stained, and the gel lane for each condition was cut horizontally into 24 pieces.

All gel bands were subjected to LC-MS/MS analysis as described previously [66]. Tandem MS data were analyzed using Sequest (ThermoFinnigan, San Jose, CA; Version 3.0). Sequest was set up to search a Homo sapiens subset of the EBI-IPI database (Version 3.32) to which a reversed copy of the protein database was appended, assuming the digestion enzyme trypsin. Sequest was searched with a fragment ion mass tolerance of 0 Da and a parent ion tolerance of 3.0 Da. Iodoacetamide derivative of cysteine was specified in Sequest as a fixed modification. Oxidation of methionine was specified in Sequest as a variable modification.

Scaffold (version Scaffold_2_05_00, Proteome Software Inc., Portland, OR) was used to validate MS/MS based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95.0% probability as specified by the Peptide Prophet algorithm. Protein identifications were accepted if they could be established at greater than 99.0% probability and contained at least three unique peptides. Protein probabilities were assigned by the Protein Prophet algorithm. Crude differential quantitation of proteins identified in both pulldown experiments was performed by comparing the number of assigned peptides.

The CKIδ, CKIα, CDK7, and ERK2 kinase assays were performed on 384-well plates (10 µl volume). The reaction mixture was as follows: for CKIδ, 2 ng/µl CKIδ (Millipore, 14-520), 50 µM peptide substrate RKKKAEpSVASLTSQCSYSS corresponding to human PER2 Lys659-Ser674 [47], and CKI buffer (40 mM Tris, 10 mM MgCl₂, 0.5 mM DTT, 0.1 mg/ml BSA, pH 7.5); for CKIα, 1 ng/µl CKIα (Invitrogen, PV3850), 50 µM CKI peptide substrate (Anaspec, 60547-1), and CKI buffer; for CDK7, 5 ng/µl CDK7 (Millipore, 14-476), 100 µM Cdk7/9 peptide substrate (Millipore, 12-526), and CKI buffer; for ERK2, 1.5 ng/µl ERK2 (Millipore, 14-550), 0.8 µg/µl MBP (Millipore, 13-104), and ERK buffer (50 mM Tris, 10 mM MgCl₂, 0.5 mM DTT, 1 mM EGTA, pH 7.5). Five hundred nl of compound was added to the mixture (final 5% DMSO), and the reaction was started by adding ATP (final 5 µM). After incubation at 30°C for 3h, 10 µl of Kinase-Glo Luminescent Kinase Assay reagent (Promega) was added, and the luminescence was detected to determine remaining ATP amount. All of the tested compounds did not inhibit luciferase activity directly. IC₅₀ value was obtained by using Prism software (GraphPad Software).

siRNAs against protein kinase genes (obtained from Human Protein Kinome Set, Integrated DNA Technologies) were tested on 384-well plates in Figure 3B, and resynthesized siRNAs (Table S1, Integrated DNA Technologies) were tested on 96-well plates in Figure 3C,D by using Bmal1-dLuc U2OS cells as described previously [29],[33]. In brief, for 96-well plates, the siRNA was spotted onto white solid-bottom plates, and 60 µl of Opti-MEM (Invitrogen) containing 0.4 µl of Lipofectamine 2000 (Invitrogen) was dispensed onto each well. After incubation at room temperature for 20 min, 60 µl of the cells in DMEM supplemented with 20% fetal bovine serum was dispensed (6,000 cells/well). The cells were cultured overnight, and the medium was changed to 180 µl of the culture medium. After 2 d, the medium was changed to 180 µl of the explant medium, and the plate was covered with optically clear film. The luminescence was recorded every 36 min by using the Tecan luminometer. The period parameter was obtained from the luminescence rhythm by using MultiCycle software.

Bmal1-dLuc U2OS cells were transfected with siRNAs as described above and harvested just before the change to the explant medium (i.e., the cells were unsynchronized at the time of harvest). Total RNA preparation and RT-qPCR were performed as described previously [29],[33]. The primers for qPCR are listed in Table S2.

HEK293T cells (1.25×10⁶ cells) were reverse transfected on 6-well plates by Lipofectamine 2000 with 1 µg each of expression vectors for C-terminally 3XFlag-tagged clock protein (in p3XFLAG-CMV-14, Sigma) and N-terminally HA-tagged kinase (in p3XFLAG-CMV-14). For ERK2, 0.05 µg of expression vector with 0.95 µg of empty vector was used because of its efficient expression. After 24 h, the cells were collected with ice-cold PBS and suspended in 100 µl of incubation buffer [50 mM Tris, 50 mM NaCl, 2 mM EDTA, 10% glycerol, 1 mM DTT, Complete Protease Inhibitor Cocktail (Roche), Phosphatase Inhibitor Cocktail 1 and 2 (Sigma), pH 8.0]. The mixture was supplemented with NP-40 (final 1%) and incubated on ice for 15 min, followed by centrifugation (16,000×g) at 4°C for 10 min. A part of the resulting supernatant (40 µl) was incubated with 0.4 µg of anti-HA antibody (Roche, 11867423001) cross-linked with Dynabeads Protein G (Invitrogen) at 4°C for 2 h with rotation. The beads were washed twice with the incubation buffer supplemented with 1% NP-40. The bound proteins were eluted with SDS sample buffer, separated by SDS-PAGE (4%–12% gradient gel), and analyzed by Western blotting with anti-Flag antibody (Sigma, F1804) or anti-HA antibody conjugated with HRP (Roche, 12013819001). For the analysis of PER1 electrophoretic mobility-shift, the cell extracts were separated by SDS-PAGE (3%–8% gradient gel, Invitrogen) and analyzed by Western blotting with anti-Flag antibody or anti-α-tubulin antibody (Santa Cruz Biotechnology, sc-32293). Protein concentration of each sample was measured by the Lowry method using DC protein assay (BioRad).

HEK293T cells (6.0×10⁴ cells) were reverse transfected on 96-well white solid-bottom plates by Lipofectamine 2000 with 40 ng each of expression vectors for C-terminally luciferase-fused PER1 (in p3XFLAG-CMV-14) and N-terminally HA-tagged kinase (in p3XFLAG-CMV-14). For luciferase (in p3XFLAG-CMV-14), 2 ng of expression vector with 38 ng of empty vector was used because of its efficient expression. After 48 h, the medium was supplemented with luciferin (final 1 mM) and HEPES-NaOH (pH 7.2; final 10 mM). After 1 h, cycloheximide (final 20 µg/ml) was added to the medium, and the plate was covered with optically clear film. The luminescence was recorded every 10 min by using the Tecan luminometer. Half-life was obtained by using Prism software (GraphPad Software).

Bmal1-dLuc U2OS cells were plated onto 6-well-plates (2.0×10⁵ cells/well). After 2 d, the medium was replaced with 2 ml explant medium containing 0, 3, or 9 µM longdaysin. The plate was covered with film and kept at 36°C. At indicated time points, the cells were collected with ice-cold PBS and stored at −80°C. Then the cell pellets were homogenized in SDS sample buffer and analyzed by Western blotting with anti-PER1 antibody (Cosmo Bio, KAL-KI044) or anti-α-tubulin antibody. In parallel, luminescence rhythms of the cells plated on 35 mm dishes were recorded with LumiCycle luminometer at 36°C.

The per3-luc transgenic line [52] was obtained from Zebrafish International Resource Center. Hemizygote larval fish were entrained in 12 h light/12 h dark cycles from day 3 to 6 postfertilization. They were then placed in an individual well of a 96-well white solid-bottom plate with 180 µl of E3 solution (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, and 0.33 mM MgSO₄, pH 7.0) containing 0.5 mM luciferin, 0.013% Amquel Plus Instant Water Detoxifier (Kordon brand; Novalek, Hayward, California, United States), and various concentrations of longdaysin (final 0.1% DMSO). The plate was covered with optically clear film, and the luminescence was recorded every 36 min by using the Tecan luminometer at 25°C. The period parameter was obtained from the luminescence rhythm by using MultiCycle software.