Evidence for Weakened Intercellular Coupling in the Mammalian Circadian Clock under Long Photoperiod

The experiments in this study were performed in accordance to the Dutch law on animal welfare and approved by the Dutch government. Permit (DEC 11010) was issued by the animal ethical committee of the Leiden University Medical Center. Homozygous PERIOD2::LUCIFERASE (PER2::LUC) knock-in mice [17] were obtained from G. Lundkvist (Karolinska Institute, Stockholm, Sweden) and rederived at the animal facility of the Leiden University Medical Center using female wild type C57BL/6J (Charles River). This homozygous line was generated by backcrossing heterozygous knock-in mice to wild type C57BL/6J for more than 10 generations and then intercrossed. Male mice—average age 6.3 ± 0.6 months—were maintained in climate controlled cabinets equipped with full-spectrum diffused lighting of 50–100 lux intensity (Osram truelight TL). Mice had access to food and water ad-libitum, were kept in groups of 2–5 animals and had access to nesting material. Mice were subjected to either a long photoperiod (LP; 16 h light, 8 h dark) or a short photoperiod (SP; 8 h light; 16 h dark) schedule for at least 28 days. Animals were sacrificed within two hours before lights off, since dissection during that period is found to least affect the SCN rhythm [18, 19]. Experimental time was expressed as the projected external time (ExT), with ExT 0 representing the midnight, and ExT 12 the midday of the previous light regime.

Slice cultures of the SCN were prepared as described previously [20]. Briefly, mice were decapitated and the brain dissected and placed in ice cold, low Ca²⁺ and high Mg²⁺ artificial cerebrospinal fluid (ACSF), containing (in mM): NaCl (116.4), KCl (5.4), NaH₂PO₄ (1.0), MgSO₄ (0.8), CaCl₂ (1.0), MgCl₂ (4.0), NaHCO₃ (23.8), D-glucose (16.7) and 5 mg/l gentamicin (Sigma Aldrich) saturated with 95% O2−5% CO₂ (pH 7.4). From each animal, two consecutive coronal slices (200 μm thick) of the hypothalamus containing the SCN were made with a VT 1000S vibrating microtome (Leica), the slice containing the anterior SCN was optically identified by using the third ventricle and the optic chiasm as landmarks (S1A Fig). The next slice was considered to contain the posterior SCN. The SCN was isolated and slice cultures were maintained on a Millicell membrane insert (PICMORG50, Millipore). In a subset of experiments the SCN tissue of two animals (four slices) were cultured on a single membrane insert. Membrane inserts were placed in a sealed 35 mm dish containing 1.2 ml of Dulbecco’s Modified Eagles Medium (D7777, Sigma-Aldrich) supplemented with 10 mM HEPES-buffer (Sigma-Aldrich), 2% B-27 (Gibco), 5 U/ml penicillin and 5 μg/ml streptomycin (0.1% penicillin-streptomycin, Sigma-Aldrich) and 0.2 mM D-luciferin sodium salt (Promega), adjusted to pH 7.2 with NaOH.

SCN slices were immediately transferred to an upright microscope (BX51WIF, Olympus) housed in a light-tight and temperature-controlled chamber, at 37°C (Life Imaging Services, Reinach, Switzerland). The microscope was equipped with a long-working distance objective (HN10X/22, Olympus), a cooled CCD camera (ORCA UU-BT-1024, Hamamatsu) and a motorized stage (XY-shifting table 240, Luigs & Neumann Ratingen, Germany) as well as focus control (MA-42Z, Märzhäuser, Wetzlar, Germany) both connected to an OASIS-4i Four Axis Controller. Bioluminescence images were obtained from either four or two SCN cultures using exposure times of 14.5 and 29 minutes respectively. Stage and focus position as well as image acquisition was controlled by Image Pro Plus software (MediaCybernetics, Warrendale PA USA; StagePro plug-in, Objective Imaging, Cambridge, UK).

A MATLAB-based (Mathworks, Natick, MA) custom-made program (source code available on request) was used to analyze the time series of images. First, cell-like regions of interest (ROIs) were defined by automated detection of groups of pixels with luminescence intensity above the noise level. The location of these cell-like ROIs had to be consistent throughout the recordings to include them for further analysis. For convenience we will refer to these cell-like ROIs as “single cells”. In order to reduce noise and increase the efficiency for subsequent analyses, a previously described algorithm [21] was used to smooth and resample the data to one data point per minute. This provides a sinus-like wave for each single cell. Finally, the smoothed data were detrended and normalized using a fourth order polynomial from the TSA toolbox for MATLAB.

The processed intensity traces from the single-cell rhythms were evaluated on sustained PER2::LUC signal and circadian rhythmicity. Single cells were only used if the time series contained at least three cycles, with peaks above and troughs below the trend line, and with an averaged interpeak interval in the circadian range of 20–28 hours. The selection procedure yielded on average 134 cells per slice, with a minimum of 56, and a maximum of 254 cells. Although the average number of cells in the anterior SCN was slightly higher than in the posterior SCN, there was no significant difference between any of the groups (). S1E Fig

To assess peak time and peak time variability, the first peak after the first trough of the PER2::LUC expression rhythm was used as a phase marker. Peak time variability was defined as the standard deviation (SD) of the peak times from all cells per slice.

We determined the cycle-to-cycle interval between the half-maximum of the rising edge of the PER2::LUC expression rhythm and calculated the day-to-day variability in period in accordance to previous studies using this parameter as an indication for period stability and precision [22, 23]. We defined the period variability as the standard deviation (SD) of the cycle-to-cycle interval of individual cells [23]. We calculated the period variability for the first three cycles for each cell in a slice, and averaged the single-cell period variability of all cells per slice.

Data were further analyzed using GraphPad Prism software. For comparison between groups, a two-way ANOVA was used, followed by a Sidak post hoc analysis for significance. The correlation between peak time SD and period SD was analyzed with a linear regression.

A spatial analysis of the variability in period of single cells in the SCN was performed. From a binary representation matrix of both the left and right SCN, the center of mass—which is the weighted average position of mass in a body or a system of particles—was calculated using the following mathematical formula: xcm=∑ximi/M;ycm=∑yimi/M where x_cm, y_cm is the position of centre of mass, x_i, y_i is the position of the i-th particle with mass m_i and M is the total mass of the body. Based on these centers, each SCN was divided into six regions and all ROIs within each region analyzed. For each slice, the average single-cell period SD was calculated per region and weighted for the corresponding number of RIOs. We then calculated the weighted average for all slices of each experimental group (LP anterior and posterior, SP anterior and posterior). This resulted in average regional values for period variability for every experimental group for six regions in both the left and the right SCN.

To resolve functional modules in the SCN neuronal network, a recently developed advanced method for the identification of community structure was employed [15, 24]. First, a cross-correlation matrix from the multiple time series of the bioluminescence intensity traces was constructed. Next, using random matrix theory, both the local (neuron-specific) noise and global (SCN-wide) dependencies from the correlation matrix were filtered out [15]. In contrast to alternative community detecting methods, there is no need to introduce arbitrary thresholds. The method leads to the identification of functional modules (communities) that are guaranteed to have positive overall correlation within communities and negative overall correlation between communities, relative to the global SCN activity.

To assess the phase relationship between the anterior and posterior SCN in both LP and SP conditions, PER2::LUC peak time was determined from smoothed bioluminescence intensity traces of the single cells, and averaged per slice (Fig 1A and 1B). The average peak time–expressed in ExT–was about two hours later in the anterior, compared to the posterior slice, for both LP and SP (Fig 1C; LP, anterior: ExT 16.7 ± 0.5, n = 5, posterior: ExT 14.2 ± 0.9, n = 5; p < 0.05; SP, anterior: ExT 19.8 ± 0.3, n = 4, posterior ExT 17.8 ± 0.3, n = 4; p < 0.05; cf. S2A Fig). In LP, all average peak times per slice preceded the projected time of lights-off, while in SP all average peak times followed after the projected time of lights-off (Fig 1C). The data on peak time show that the temporal relationship of anterior to posterior SCN is maintained in different photoperiods.

To test whether photoperiod affects the phase relationship among cells, we used the standard deviation (SD) of peak times as a measure of phase dispersion within slices (Fig 2B). Photoperiod had a clear effect on the distribution of peak times in cellular PER2::LUC rhythms (Fig 2A). The SD of peak times was significantly higher in LP than in SP, but only in the anterior SCN (Fig 2B; anterior, LP: 3.42 ± 0.19, n = 5, SP 1.42 ± 0.04, n = 4; p < 0.001; posterior, LP: 1.81 ± 0.15, n = 5, SP: 1.40 ± 0.13, n = 4; p > 0.05; cf. S2B Fig). This photoperiod-induced difference in phase dispersion was not dependent on the time in culture (S3 Fig). The results emphasize the spatial heterogeneity in photoperiodic response within the SCN.

We next sought to determine whether an increase in period variability of single cells could contribute to an increase in peak time dispersion. The fluctuation of the cycle-to-cycle period of individual cells was analyzed (Fig 3A). The standard deviation (SD) of the cycle intervals of the first three cycles of PER2::LUC rhythms was determined for individual cells, and averaged per slice (Fig 3B). The average variability in single-cell period was increased in LP compared to SP in the anterior SCN (p < 0.05), and in anterior vs. posterior in both long and short photoperiod (p < 0.001 and p < 0.01 respectively; anterior, LP: 1.78 ± 0.30, n = 5, SP: 1.28 ± 0.17, n = 4; posterior, LP: 0.97 ± 0.18, n = 5, SP: 0.71 ± 0.10, n = 4; Fig 3B), meaning that the cycle interval of individual cells varied on average 0.5 hours per cycle more in LP than in SP. It should be noted that the average period of the cellular PER2::LUC rhythm over 3 days was not affected by photoperiod (S4B Fig). Increasing period variability of single cells could be a mechanism to achieve peak phase dispersion. Indeed, assessment of the relation between phase dispersion and single-cell period variability showed that peak-time SD and the single-cell period SD was strongly correlated (R² = 0.60, p < 0.001; Fig 3C; cf. S2D Fig).

To map the spatial heterogeneity of period variability within slices, we arbitrarily divided the SCN in six regions for which the single-cell period variability was averaged (Fig 4). We found only minor differences between the regions in period variability in SP, which could be expected from the overall lower variability in phases in this photoperiod. Interestingly, the differences between the regions were more substantial in LP. The average variability of single cells was found to be lower in the medial than in the lateral region of the SCN. Furthermore, the differences were consistently larger in the anterior SCN compared to the posterior SCN.

To extend the analysis of regional differences in the coronal plane and identify clusters of cells expressing similar dynamic behavior, we employed a recently developed community detection approach designed specifically for correlation matrices [15, 16]. This method consistently yielded two main groups of cells, which displayed a robust spatial distribution within both the anterior and posterior slice. This correlation-based differentiation yielded a similar spatial distribution as described above. More specifically, in the anterior slice the communities divided the SCN in a ventromedial (VM) region and a dorsolateral (DL) region (Fig 5A). Note that these regions do not correspond with the division of the SCN in dorsomedial (shell) and ventrolateral (core) SCN defined originally in rats by the distribution of neuropeptide content [25], but later revised for mice [26]. In the posterior slice the two communities were found in the medial (M) and lateral (L) regions of the SCN. We characterized these clusters in peak time, phase dispersion and single-cell period variability. The VM in the anterior, and the M region in the posterior SCN peaked earlier compared to the DL and L region (Fig 5B; anterior LP: DL: ExT 18.1 ± 0.5; VM: ExT 15.2 ± 0.5, n = 5, p < 0.01; SP: DL: ExT 20.5 ± 0.3; VM: ExT 19.0 ± 0.4, n = 4, p < 0.01; posterior LP: L: ExT 15.1 ± 0.9; M: ExT 13.6 ± 0.8, n = 5, p < 0.01; SP: L: ExT 18.8 ± 0.3 M: ExT 17.1 ± 0.5, n = 4, p < 0.01). With respect to phase dispersion, there were no differences between the regions (Fig 5C). Interestingly, in the anterior SCN in LP, the single-cell period variability was higher in the DL region, compared to the VM region (Fig 5D; LP, anterior, DL: SD 2.06 ± 0.31, VM: SD 1.47 ± 0.29, n = 5, p < 0.001). The spatial distribution of the two groups of cells identified with the impartial method of community detection is similar to the pattern defined by physiological parameters (cf. Fig 4).

In the present study we observed a difference in mean peak time of PER2::LUC rhythms between the anterior and posterior SCN, with the posterior SCN peaking earlier than the anterior SCN (Fig 1C). This is consistent with previous studies on gene expression rhythms in the anterior vs. posterior SCN [3, 7, 28], however see [29].

Exposure to LP, but not to SP, led to considerably larger phase differences in PER2::LUC expression rhythms between the single cells of the anterior, compared to posterior SCN (Figs 1B and 2), which is in accordance with previous studies on single-cell recordings in Per1 expression [4]. In addition, both PER2::LUC and Bmal1-ELuc recordings from mice exposed to LP also revealed an increased regional phase dispersion in the (anterior) SCN, suggesting a broader influence of environmental photoperiod on the transcription-translation feedback loop [6, 30]. To our knowledge, this is the first study showing that the photoperiod-induced PER2 peak time distribution of single cells is restricted to the anterior SCN (Fig 2B). Notably, in the anterior SCN the difference in average peak time standard deviation between LP and SP is about two hours, suggesting it may strongly contribute to the photoperiod-induced broadening of the electrical activity rhythm in the SCN [5].

The increase in peak phase dispersion in LP can be established in three ways: (1) by increased phase dispersion between cells achieving a new stable phase relationship, but maintaining the same period; (2) by variability in period between individual cells, but with little instability in the period of the single cell; or (3) by variability in the cycle-to-cycle period of single cells (see insets in Figs 2B, 3B and S5). We hypothesize that coupling within the network is a potential mechanism that will both affect the accuracy of the period in single cells (option 3) as well as the variability of the period between the cells (option 2). Although there is increasing evidence for a stable phase relationship of gene expression rhythms between regions in the SCN (option 2) in different photoperiods [6, 28, 30, 31], we were interested to know if single-cell period variability (option 3) may contribute to peak phase dispersion. We therefore analyzed the period standard deviation, and found that it was increased in LP in the anterior SCN. Interestingly, the variability in period of the first three cycles in vitro was positively correlated with the peak time distribution in the first cycle (Fig 3C). Thus, a larger peak time distribution in the first cycle seems to be predictive of a large variability in cycle-to-cycle period. The increased single-cell period variability from cycle to cycle may indicate weak coupling among the cells. In this context it should be noted that period stability and the resulting high precision of the overt rhythms are a hallmark of the circadian clock [22, 32]. However, this seems to be primarily a result of a coupled network consisting of less accurate single cell clocks [23]. We therefore propose that photoperiod-induced variability in cycle-to-cycle period of the cells in the network is a consequence of a change in coupling strength, which contributes to the alteration in phase dispersal.

We next sought to determine whether the cycle-to-cycle period variability was spatially distributed in the SCN. Indeed, the mapping of regional averages of period variability revealed that the largest variability was found in the lateral part of the SCN (Fig 4). Notably, this was the case throughout the rostrocaudal plane, with the most pronounced differences in LP, in the anterior SCN. This spatial distribution coincides with the phase distribution found by Evans and colleagues [28], who reported a ~1.5 hour phase difference between the ventrolateral and dorsomedial part of the anterior SCN.

Subsequently we used an impartial community detection approach, which could identify functional clusters of cells, without knowledge on spatial locations [15]. Surprisingly, this method persistently identified two populations of neurons, of which the locations coincided largely with the regions described above (Fig 5A). While the cluster of cells in the ventromedial (VM) region of the anterior SCN, as well as the clusters located in the posterior SCN are more rigid in their oscillatory behavior, the clusters of cells found in the dorsolateral (DL) region of the anterior SCN show more oscillatory variability, raising the possibility that neurons in the DL and VM region of the anterior SCN are differentially affected by long photoperiod. The location of clusters defined by our community detection method overlap only partially with previously defined regions of the SCN based on neuropeptide content. Nevertheless, they are comparable to the regions described by Abel and colleagues [33], who found hubs in a small-world network located in the central SCN, corresponding to our DL cluster, and only sparsely connected cells in a region similar our VM cluster. Together with the results from the current study, this could mean that the DL network functions differently than the VL network, and responds differently to photoperiodic input.

Several previous studies have used clustering methods to identify functional groups of neurons within the SCN. Most methods require the specification of an arbitrary threshold value for inter-neuron correlation in order to project correlation matrices to networks and subsequently identify clusters within these networks. The resulting clusters are therefore unavoidably threshold-dependent. As an additional limitation, such thresholding methods tend to detect only the single cluster with the largest internal correlation, thereby failing to identify additional structures. The advantage of the method used in the current study is that no definition of an arbitrary threshold is needed. Two previous papers described clustering of “super-pixels” from imaging of PER2::LUC in the SCN, in which a k-medoid cluster analysis with the gap statistic with a cosine distance function was used to produce a pre-defined number (5) of clusters [34, 35]. Since these studies use super-pixels and we use cell-like ROIs it is difficult to directly compare the results, although the locations of the clusters that were found did not contradict the spatial organization of our clusters. Myung and co-workers [31] also used cluster analysis to identify regions with different temporal relationships; the spatial distribution of clusters found using two independent methods was only similar to the present study in the anterior SCN. The discrepancies may result from differences in rhythm dynamics between PER2 and BMAL1 expression as already suggested by the authors [31].

This study shows for the first time that exposure to long photoperiod affects rhythm stability of single cells. This effect of photoperiod on single-cell period variability can be caused by different mechanisms and on different levels. On the single-cell level, period precision can be reduced by increased molecular noise [36]. At the network level, changes in the intercellular connectivity could result in period variability [23, 37]; changes in intercellular signaling could actively induce variability in the period of single cells [30]. Alternatively, reduced connectivity, meaning reduced intercellular signaling could cause larger “drifts” in period, before the signal is strong enough to correct a drifting cell.

Studies using computational modeling to evaluate the effect of molecular noise on circadian rhythmicity aimed at explaining the robustness of circadian rhythms in the SCN. This work shows that it takes considerably low gene expression of the core clock genes to affect period variability [36, 38, 39]. Although we cannot exclude a role of molecular noise in the period variability found in this study, our results do not suggest an effect of photoperiod on the amplitude in single-cell gene expression of PER2::LUC as would be predicted by the models (S1B and S1D Fig).

Previous experimental and modeling work suggests that alterations in the balance between attractive and repulsive coupling can induce a phase gap in BMAL expression rhythms between the dorsal and ventral region of the SCN [30, 40]. While the underlying neuronal mechanisms of photoperiod-induced phase adjustments within the SCN network is still not fully understood, there is mounting evidence for a role of GABA in dorsal-ventral communication, and seasonal encoding in the SCN [6, 30, 41]. However, the increased phase dispersion we found in the current study is present within, not between those regions, making it conceivable that other mechanisms are at play. The model of Myung and colleagues [30] is able to explain the emergence of a phase gap between different regions in the SCN, based on differences in period. However, it does not explain the period variability of single cells we see in this study. For that, a daily variability in attractive/repulsive coupling should take place between single SCN neurons and not only regions, which would be mechanistically challenging to explain.

Several studies have shown that reduced intercellular contact results in increased single-cell period variability [23, 37]. These studies have found that low-density plating of SCN neurons results in higher period variability in both electrical activity [37] and in Per1 gene expression [23], compared to cultured slices or high-density plating, respectively. These studies indicate that the increased period variability found after exposure to long photoperiod could result from reduced intercellular connectivity. These findings make it likely that the increased single-cell period variability documented in the present study is the result of reduced coupling within the SCN neuronal network.