Cold-Inducible RNA-binding protein (CIRBP) adjusts clock gene expression and REM sleep recovery following sleep deprivation

Sleep depriving mice affects clock gene expression, suggesting that these genes partake in sleep homeostasis. The mechanisms linking wakefulness to clock gene expression are, however, not well understood. We propose CIRBP because its rhythmic expression is i) sleep-wake driven and ii) necessary for high-amplitude clock gene expression in vitro. We therefore expect Cirbp knock-out (KO) mice to exhibit attenuated sleep-deprivation (SD) induced changes in clock gene expression, and consequently to differ in their sleep homeostatic regulation. Lack of CIRBP indeed blunted the SD-incurred changes in cortical expression of the clock gene Rev-erbα whereas it amplified the changes in Per2 and Clock. Concerning sleep homeostasis, KO mice accrued only half the extra REM sleep wild-type (WT) littermates obtained during recovery. Unexpectedly, KO mice were more active during lights-off which was accompanied by an acceleration of theta oscillations. Thus, CIRBP adjusts cortical clock gene expression after SD and expedites REM sleep recovery.


34
The sleep-wake distribution is coordinated by the interaction of a circadian and a sleep 35 homeostatic process (Daan et al., 1984). The molecular basis of the circadian process consists of 36 clock genes that interact through transcriptional/translational feedback loops. 37 CLOCK/NPAS2:BMAL1 heterodimers drive the transcription of many target genes, among which 38 Period (Per1-2), Cryptochome (Cry1, -2), and Rev-Erb (Nr1d1, -2). Subsequently, PER:CRY complexes 39 inhibit CLOCK/NPAS2:BMAL1 transcriptional activity and thus prevent their own transcription. The sleep homeostatic process keeps track of time spent awake and time spent asleep, during 44 which sleep pressure is increasing and decreasing, respectively. The mechanisms underlying this 45 process are to date unknown. However, accumulating evidence implicates clock genes in sleep 46 homeostasis [reviewed in (Franken, 2013)]. This is supported by studies in several species (i.e. mice, 47 Two independent studies showed that the temperature-driven changes in CIRBP are required 66 for high amplitude clock gene expression in temperature synchronized cells (Morf et al., 2012, Liu 67 et al., 2013. Therefore, we and others (Archer et al., 2014) hypothesized that changes in clock gene 68 expression during SD are a consequence of the sleep-wake driven changes in CIRBP. We used mice 69 Figure 1 The sleep-wake distribution drives daily changes of central Cirbp expression in the mouse. After the onset of the baseline rest phase (ZT0), when mice spend more time asleep and thus Tcx decreases, Cirbp expression increases (blue symbols and lines), whereas between ZT12-18, when mice spent most of their time awake and Tcx is high, Cirbp decreases. When controlling for these diurnal changes in sleep-wake distribution by performing four 6h SD [sleep deprivation] starting at either ZT0, -6, -12, or -18, the diurnal amplitude of Cirbp is greatly reduced (red symbols represent levels of expression reached at the end of the SD). Nine biological replicates per time point and condition from three different inbred strains of mice were used (one data point missing at ZT18), and RNA was extracted from whole brain tissue (see Maret et al. 2007 for details). Data were taken from GEO GSE9442 and accessible in source file 1. 7

Daily cycles in Tcx are determined by sleep-wake state 121
After having established that the rapid changes in Tcx are indeed evoked by changes in sleep-122 wake state, we next wondered if the large daily change in Tcx is also due to the daily rhythms in 123 sleep-wake state and LMA, and therefore inspected these variables per hour. The LMA data were 124 log2 transformed to allow for parametric assessment. Tcx, waking and LMA oscillated over the course 125

THE INFLUENCE OF SD AND CIRBP ON TRANSCRIPTS IN CORTEX AND LIVER 186
After establishing that also in the mouse the sleep-wake distribution is the major determinant 187 of Tcx, we assessed whether the SD-incurred changes in CIRBP participate in linking the effect of 188 SD to clock gene expression. To this end, we quantified 11 transcripts from liver and 15 from cortex 189 an emphasis on clock genes. Mice were sacrificed before SD at ZT0, or 6 hours later after SD (ZT6-192 SD) together with non-sleep deprived control mice that could sleep ad lib (ZT6-NSD). Statistics on 193 ZT0 (t-test) and ZT6 (2-way ANOVA) can be found in Table 1.  RBM3 is another cold-inducible RNA Binding Protein and, like CIRBP, conveys temperature 203 cycles into high-amplitude clock gene expression in vitro (Liu et al., 2013). A long and a short 204 isoform of Rbm3 (Rbm3-long and -short, resp.) that differ in their 3'UTR length were discovered in 205 the mouse cortex. Although both isoforms are referred to as 'cold-induced', they exhibit opposite 206 responses to SD (Wang et al., 2010), with a decrease in the short isoform and an increase in the long 207 isoform. We found that overall, the short isoform was more common than the long isoform in the 208 cortex (PCR cycle detection number for all samples pooled: cortex: Rbm3-short: 25.6±0.2, Rbm3-209 long: 29.7±0.1, amplification efficiency Rbm3-short: 2.11 and Rbm3-long 2.07). In the liver, only the 210 Figure 4 Cortical expression of several genes is affected by SD and the lack of CIRBP. NRQ: Normalized Relative Quantity, SD: sleep deprivation, GT: Genotype. n=5 for each group, each symbol represents an observation in one mouse. Mice were sacrificed at ZT0, at ZT6 after sleep deprivation (ZT6-SD) or after sleeping ad lib (ZT6-NSD). Statistics are performed separately for ZT0 (factor GT, t-test), and ZT6 (factor GT and SD; 2-W ANOVA). Significant (p<0.05) GT differences are indicated by a black line and *, the effect of SD in WT mice with a grey line and *, and in KO mice with a green line and *. Interaction effects (GTxSD) at ZT6 are indicated by a red *. See Table 1 for statistics.
short isoform was detected (liver: Rbm3-short: 28.2±0.2, Rbm3-long: >32; i.e., beyond reliable 211 detection limit). We confirmed that after SD, Rbm3-short was decreased in the cortex (Figure 4 Dusp4, Hspa5/BiP, Hsp90b, and Hsf1 was increased by SD (Figure 4, supplement 2). Post-hoc tests 217 revealed that the latter two were significantly increased only in Cirbp KO mice. Furthermore, the 218 effect of SD on the transcripts Hsp90b and Hspa5 was significantly amplified in Cirbp KO mice 219 compared to WT mice. Unexpectedly, no changes in the expression of heat shock transcripts 220 incurred by SD or genotype were detected in the liver (Figure 4, supplement 1). 221 Figure 4, supplement 1 Changes in transcripts incurred by the absence of CIRBP and/or SD in the liver. Legend same as in Figure 4. See Table 1 for statistics.

13
In vitro studies have shown that the presence of CIRBP is associated with longer 3'UTRs of its 222 target genes, such as the transcript splice-factor proline Q (Sfpq), resulting in a higher prevalence 223 of long isoforms (extended or ext) over all isoforms (common or com), and thus an increased  Compared to the cortex, the clock gene expression in the liver appeared more resilient to the 249 effects of SD, as only Dbp and Rev-erbα were significantly affected and not Per2 (Figure 4, 250 supplement 1). The lack of CIRBP did not interfere with this response, nor did it contribute to 251 genotype dependent changes of other (clock) gene transcripts in the liver. 252 Taken together, the absence of CIRBP modulated the SD induced changes in the cortical 253 expression of the clock genes Rev-erbα, Clock and Per2. Furthermore, the expression of transcripts 254 in the heat shock pathway were also affected in a genotypic manner by SD. 255 256 257 Figure 4, supplement 2 Changes in transcripts incurred by the absence of CIRBP and/or SD in the cortex. Legend same as in Figure 4. See Table 1 for statistics.

Baseline characteristics of sleep-wake behavior do not differ between Cirbp KO and WT mice 282
During the two baseline days, no significant differences in waking, NREM or REM sleep were 283 observed. This was neither the case for time spent in these three behavioral states per light and 284 dark phase (Table 2), nor for the distribution of sleep and waking across the day (see   Differences in delta power can be attributed to changes in the dynamics of the underlying 298 homeostatic process, Process S, and/or to changes in the sleep-wake distribution. Evidence 299 supporting the latter possibility was observed because Cirbp KO mice tended to spend less time in 300 NREM sleep (and more time awake) during the early dark phase compared to WT mice, reaching 301 significance during the recovery (Figure 5-A; 3 rd graph from top). To test if these changes in the 302 sleep-wake distribution were indeed sufficient to raise NREM delta power above WT levels, we 303 estimated the increase (τi) and decrease (τd) rate of delta power by a simulation of Process S based 304 on the sleep-wake distribution. We assumed Process S to increase exponentially during waking and 305  Table 3). Hence, the reduction in 311 NREM sleep in Cirbp KO mice in the beginning of the dark period caused the higher NREM EEG 312  The amount of REM sleep is also homeostatically defended (Franken, 2002). At the end of REC1, 321 both WT and KO mice spent more time in REM sleep compared to corresponding baseline hours.  While quantifying sleep-wake states, we observed that Cirbp KO mice were more active than 333 their WT littermates during the dark phase (t(31)=-2.56, p=0.015, see also Table 2). More specifically, 334 Cirbp KO mice were almost twice as active in the first 6hrs of the dark phase (movements: WT: 335 128  Figure 6-A). Interestingly, this pronounced 337 increase was not associated with a significant increase in time spent awake during BL (per 12 hrs: 338 t(35)=1.2, p=0.24, and see Table 2 p=0.01). Note that also Tcx was not significantly increased in Cirbp KO mice during the dark phase 341 Because Cirbp KO mice are not more awake (Table 2 and Figure 6), we wondered if their 344 increased LMA is associated to the prevalence of sub-states of waking. Theta-dominated waking 345 (TDW) is a sub-state of waking that correlates with activity, prevails during the dark phase and SD, 346 and is characterized by the presence of EEG theta-activity (Buzsáki, 2006, Vassalli and Franken, 347 2017). Despite their increased LMA, Cirbp KO mice did not spend more time in TDW during the 348 dark phase of the BL (see Table 1  In contrast, the spectral composition of the EEG during 'quiet' waking (i.e. all waking that is not 358 TDW) was remarkable similar between the two genotypes ( Figure 6, figure supplement 1), 359 demonstrating that the changes in spectral composition of TDW EEG are not the result of a general 360 effect of CIRBP on the waking EEG. 361 Moreover, we observed a decrease in slow and a non-significant increase in faster theta 362 activity in the TDW EEG of Cirbp KO mice, together hinting at an acceleration of theta peak 363 frequency (TPF; lower panel in Figure 6   estimates of TPF during TDW are less precise, we found more non-significant associations in both 390 genotypes (KO: 3/ 16; WT: 3/ 17 mice). Altogether, these results provide further evidence that LMA 391 contributes to TPF and suggests that CIRBP, through its effects on LMA, reduces TPF. 392 The SD altered the distribution of waking during the recovery relative to BL (3-way RM ANOVA, 393 factor Time, GT and BL/REC, factor BL/REC: REC1: F(1,558)=42.7, p<0.0001; REC2: F(1,1514)=441.8, 394 p<0.0001; see triangles in REC1 and REC2, Figure 6-A). Surprisingly, while time spent awake was 395 overall decreased compared to baseline, we observed several intervals during the recovery in which The EEG spectra during TDW in REC1 and REC2 showed similar profiles as during BL (see Figure  405 6, figure supplement 2), although there were some changes that in recovery reached significance 406  In this study, we showed that, like in other rodents, the sleep-wake distribution is the major 419 determinant of Tcx in the mouse. Because of the well-established link between temperature and 420 CIRBP levels, it is likely that the equally well-known sleep-wake driven changes in Cirbp expression 421 in the brain are conveyed through the sleep-wake driven changes in brain temperature. As 422 predicted, the SD-incurred changes in the expression of clock genes was modulated by the presence 423 of CIRBP. However, only for Rev-Erbα did we observe the anticipated attenuated response to SD in 424 Cirbp KO mice, whereas the changes in the expression of Per2 and Clock were amplified compared 425 to WT mice. Moreover, we did discover evidence of altered dynamics of the process regulating time 426 spent in REM sleep. Unexpectedly, Cirbp KO mice are more active during the dark phase, and have 427 during TDW reduced power in the gamma band and increased TPF. 428

CHANGES IN CORTICAL TEMPERATURE ARE SLEEP-WAKE DRIVEN 430
When sleep and waking occur at their appropriate circadian times, the changes in both brain 431 and body temperature have a clear 24-hour rhythm and therefore appear as being controlled 432 directly by the circadian clock. However, sleep-wake cycles contribute significantly to both the daily 433 changes in brain and body temperature. In humans, this involvement is powerfully illustrated by 434 spontaneous desynchrony, where body temperature follows both a circadian and an activity-rest 435 (and presumably, sleep-wake) dependent rhythm (Wever, 1979). The contribution of sleep-wake 436 state to the daily dynamics in body temperature is further supported by forced desynchrony studies, 437 such as (Dijk and Czeisler, 1995), estimating that 'masking' effects of rest-activity and sleep-wake 438 cycles contributed between 30% and 50% to the amplitude of the circadian body temperature In contrast to body temperature, brain temperature in rodents is much more determined by 444 sleep-wake state: 80% of its variance can be explained by the sleep-wake distribution ( (Franken et 445 al., 1992) and this study). Likewise, the sleep-wake driven changes in brain temperature are still 446 present in arrhythmic animals (Edgar et al., 1993, Baker et al., 2005, pointing to a more important 447 sleep-wake dependency of brain temperature compared to body temperature. In our study, we also 448 estimated the contribution of LMA to changes in Tcx and found that waking with higher LMA is 449 associated with higher Tcx. Although significant, the contribution of LMA to the daily changes in 450 Tcx was modest and explained only 2% more of the variance compared to waking alone. Can we 451 optimize the prediction of Tcx? A non-linear relationship between sleep-wake state and Tcx was 452 assumed previously (Franken et al., 1992) and could have improved the prediction of our model 453 further. This is supported by the residuals from the complete model (see Figure 3, supplement 2), 454 that exhibit under baseline conditions a circadian distribution, whereas during the SD, they remain 455 increased as during the dark phase. Thus, the model overestimates Tcx during periods with little 456 waking (light phase) and underestimates Tcx during periods that are dominated by waking (dark 457 phase and SD), suggesting a non-linear relationship between these two variables. 458 Important to consider is that the influence of LMA on Tcx is likely affected by the type of activity; 459 for example, rats on a running wheel activity can increase their brain temperature by 2°C within 30 460 minutes (Fuller et al., 1998). Also exercise in men leads to an increase in (proxies) of brain power. However, there is a clear increase in power of the high gamma band specifically during the 478 SD (see Figure 6, supplement 2), as noted previously (Vassalli and Franken, 2017). This increase was 479 present in both genotypes suggesting that while KO mice seem to have a reduced capacity to 480 produce fast gamma activity, SD is still able to activate their fast-gamma circuitry. These results, 481 together with the observation that during the light phase the decreased power in the gamma bands 482 was still present at a time of day when LMA did not significantly differ, argue against an association 483 between the decreased power in the gamma bands of Cirbp KO mice and their increased LMA. Several aspects of waking that appeared to differ between Cirbp KO and WT mice under baseline 498 dark conditions but were non-significant, reached significance during the recovery dark phase. For 499 example, during baseline Cirbp KO mice were 4% more awake and 13% more in TDW compared to 500 their WT littermates, which was amplified to 8% and 20%, respectively, during recovery. Also, TPF 501 and the genotype-dependent association between overall TPF and LMA reached significance during 502 the recovery. This suggest that SD amplified the genotypic differences. Other sleep deprivation 503 studies found evidence for similar phenomena, where sleep disturbance can amplify molecular and 504 behavioral phenotypes of Alzheimers' mouse models (for review, see (Musiek and Holtzman, 2016)) 505 and sensitivity to pain (Sutton and Opp, 2014). Our data indicates that a similar phenomenon 506 occurs in Cirbp KO mice, where a single 6-hr SD reveals the suggestive baseline genotypic 507 differences. It would be interesting to understand the dynamics of this change; e.g. if they are 508 reversible or if a second SD could augment genotypic differences further. The other aspects of the homeostatic regulation of sleep that we inspected, NREM EEG delta 531 power and time spent in NREM sleep after sleep deprivation, were unaffected in Cirbp KO mice. 532 Thus, CIRBP participates specifically in REM sleep homeostasis, whereas we do not find evidence 533 for its participation in NREM sleep homeostatic mechanisms. 534

OTHER MECHANISMS LINKING SLEEP-WAKE STATE TO CLOCK GENE EXPRESSION 536
Our results show that other pathways besides CIRBP must contribute to the sleep-wake driven 537 changes in clock gene expression. Some suggestions for such pathways are shortly discussed below, 538 as well considerations that could potentially account for the absence of a more widespread CIRBP 539 dependent change in clock gene expression that we expected based on in vitro results. when necessary; for details see (Mang and Franken, 2012). Briefly, six gold-plated screws (diameter 633 1.1 mm) were screwed bilaterally into the skull over the frontal and parietal cortices. Two screws 634 served as EEG electrodes and the remaining four anchored the electrode connector assembly to the 635 skull. As EMG electrodes, two gold wires were inserted into the neck musculature. Of all EEG/EMG 636 implanted mice, 8 KO and 9 WT mice were additionally implanted with a thermistor (serie 637 P20AAA102M, General Electrics (currently Thermometrics), Northridge, California, USA) which 638 was placed on top of the right cortex (2.5 mm lateral to the midline, 2.5 mm posterior to bregma). 639 The EEG and EMG electrodes and thermistor were soldered to a connector and cemented to the 640 skull. Mice recovered from surgery during 5-7 days before they were connected to the recording 641 cables in their home cage for habituation, which was at least 6 days prior to the experiment. In total 642 no less than 11 days were scheduled between surgery and start of experiment. Tcx reached in each of these 3 mice during the SD to the average Tcx reached over the same recording 715 period in the remaining 9 mice. Of note, most of our Tcx analysis focuses on its relative sleep-wake 716 dependent changes, which are not affected by differences in absolute Tcx values. Finally, the baseline 717 Tcx data that was used to construct Figure 2B was based on 7 WT and 7 KO mice. During the 718 recording, one KO mouse and one WT mouse had random fluctuations of Tcx beyond physiological 719 reach and were therefore excluded from analysis involving the daily dynamics of Tcx (Figure 3). In 720 the recovery, a KO mouse was excluded due to aberrant high Tcx that could not be accounted for 721 by the sleep-wake distribution, leaving 6 WT and 5 KO mice for analyses involving REC1 and REC2. 722 We inspected Tcx 1.5 min before and after sleep-wake transitions (i.e., transitioning from wake 729 to NREM sleep, NREM to REM sleep, REM sleep to wake and NREM sleep to wake). A sleep-wake 730 transition was selected when the state before and after the transition lasted at least 8 epochs (i.e. 731 >32 sec). With this criterion, an average of 38 wake to NREM sleep, 101 NREM sleep to REM sleep, 732 28 REM sleep to wake and 32 NREM sleep to wake transitions per mouse during the two baseline 733 days was detected. The temperature profile of Tcx before and after the transition was constructed 734 by depicting Tcx relative to the mean Tcx at a given sleep-wake transition (i.e., the average of Tcx in 735 the epoch before and after the sleep-wake transition). We subsequently constructed an individuals' 736 average change in Tcx for each sleep-wake transition. For this average, at least 10 traces were 737 contributing at a given point in time to prevent skewing of the average individual temperature 738 profile by few Tcx traces. Thus, the further from the sleep-wake state transition, the less epochs 739 contributed to the average individual Tcx profile. One WT mouse exhibited an extreme drop in Tcx 740 (-0.2°C in a 4-second epoch) after the transition from NREM sleep to wake in its average Tcx trace, 741 but not in other sleep-wake transitions. We attributed this observation to a technical artefact and 742 therefore this mouse was excluded from the NREM sleep to wake transitions. 743 The residuals of the correlation between waking and Tcx exhibited a circadian pattern under 744 baseline conditions. We visualized the properties of this pattern further by fitting a sinewave 745 through the data (Prism, non-linear regression; sine-wave with non-zero baseline; least squares fit). 746

ANALYSIS OF EEG BASED ON BEHAVIORAL STATE 748
Unless otherwise stated, all mice (20 WT and 17 KO) were included in the analyses based on the 749 EEG data. Spectral content of the EEG within sleep-wake states was calculated as follows. To 750 account for inter-individual differences in overall EEG power, EEG spectra were expressed as a 751 percentage of an individual reference value calculated as the total EEG power across 0.75-45 Hz and 752 all sleep-wake states in the 48h baseline. This reference value was weighted so that for all mice the 753 relative contribution of the three sleep-wake states (wake, NREM and REM sleep) to this reference 754 value was equal. 755 Theta peak frequency (TPF) was calculated by determining the frequency at which power 756 density peaks per 4-s epoch and subsequently averaged per individual. Power density peaks were 757 quantified from 6.5 to 12.0 Hz band and from 5.5 to 12.0 Hz band for TDW and REM sleep, 758

respectively. 759
Time course analysis of EEG delta power (i.e., the mean EEG power density in the 0.75-4.0 Hz 760 range in NREM sleep) during baseline and after SD was performed as described previously (Franken 761 et al., 1999), and similar to the analysis of LMA per unit of waking. The light periods of BL1, BL2, 762 and REC2 were divided into 12 percentiles, the REC1 light period (ZT6-12) into 8 sections, and all 763 dark periods into 6 sections. The timing of these percentiles was based on the prevalence of NREM We applied a computational method to predict the change in delta power during NREM sleep 778 based on the sleep-wake distribution as described before (Franken et al., 2001). Process S is 779 exponentially increasing with time constant τi during waking and REM sleep, and exponentially 780 decreasing by τd during NREM sleep (eq. (4) and (5), respectively). 781 In these simulations, UA represents the upper asymptote, LA the lower asymptote and dt the 784 time step of the iteration (4 seconds). Both asymptotes were estimated for each individual mouse. 785 The upper asymptote was based on the 99% level of the relative frequency distribution of delta 786 power reached in all 4s epochs scored as NREM sleep in the 4-day recording. As an estimate of the 787 lower asymptote, the intersection of the distribution of delta power values in NREM sleep with 788 REM sleep was taken. At the start of the simulation, an iteration through the first 24-hr (BL1) was 789 performed with S0=150 at t=0. The value reached after 24-hrs is independent of S0 at t=0 and, 790 assuming a steady state during baseline, reflects Process S at the start of the baseline for a given 791 and Td giving the best fit was used to assess differences in process S between genotypes. 796 We noted a subtle but consistent linear discrepancy in the alignment of the simulated Process 797