Effects of a patient-derived de novo coding alteration of CACNA1I in mice connect a schizophrenia risk gene with sleep spindle deficits

CACNA1I, a schizophrenia risk gene, encodes a subtype of voltage-gated T-type calcium channel CaV3.3. We previously reported that a patient-derived missense de novo mutation (R1346H) of CACNA1I impaired CaV3.3 channel function. Here, we generated CaV3.3-RH knock-in animals, along with mice lacking CaV3.3, to investigate the biological impact of R1346H (RH) variation. We found that RH mutation altered cellular excitability in the thalamic reticular nucleus (TRN), where CaV3.3 is abundantly expressed. Moreover, RH mutation produced marked deficits in sleep spindle occurrence and morphology throughout non-rapid eye movement (NREM) sleep, while CaV3.3 haploinsufficiency gave rise to largely normal spindles. Therefore, mice harboring the RH mutation provide a patient derived genetic model not only to dissect the spindle biology but also to evaluate the effects of pharmacological reagents in normalizing sleep spindle deficits. Importantly, our analyses highlighted the significance of characterizing individual spindles and strengthen the inferences we can make across species over sleep spindles. In conclusion, this study established a translational link between a genetic allele and spindle deficits during NREM observed in schizophrenia patients, representing a key step toward testing the hypothesis that normalizing spindles may be beneficial for schizophrenia patients.

Instruments) and electrophysiological properties of TRN neurons were recorded under the whole cell configuration using the pClamp software (Molecular Devices). Under current clamp configuration, hyperpolarization induced rebound burst firing and depolarization induced tonic firing were observed using different current injection protocol ranging from -200 pA to +200 pA (∆ 50 mV) using a K-methanesulfonate based internal solution (in mM: 140 KMeSO4; 10 KCl; 10 HEPES; 0.1 EGTA; 4 Mg-ATP; 0.2 Na-GTP; 10 phosphocreatine). Rebound burst firing was observed from different holding membrane potentials ranging from -80 mV to -55 mV. Tonic firing was observed primarily from a holding potential of around -60 mV. Sampled data was analyzed using Clampfit (Molecular Devices). For rebound burst firing properties, hyperpolarization induced bursts were visually identified and counted for each neuron for the different holding membrane potentials and plotted. Bursts were defined as events containing 2 or more action potentials with a maximum of 70 ms interspike interval. The maximum number of bursts ever observed in a neuron irrespective of holding membrane potential and hyperpolarization amplitude was also plotted as the max number of bursts observed. Other properties of the first of such bursts, such as no. of action potentials, threshold of action potentials in the burst, frequency, duration, latency from the end of hyperpolarization pulse and after-hyperpolarization (AHP) observed at the end of the burst were quantified and plotted. In addition to this, frequency and no. of action potentials observed in the second subsequent burst and the inter-burst interval (IBI) were also calculated for neurons displaying more than one burst. For tonic firing, frequency of action potentials was calculated for each depolarizing current amplitude and plotted. The maximum frequency of tonic firing for each neuron and the fast-AHP (fAHP) observed after each single action potential during tonic firing was also calculated and plotted. When a depolarizing pulse displayed burst firing in neurons, the frequency of tonic firing was calculated following the burst, without including the bursting spikes.
For recording T-type Ca 2+ currents from TRN neurons, slices were transferred to a submersion recording chamber perfused with a modified aCSF (in mM: 120 NaCl, 3 KCl, 20 TEA-Cl; 5 CsCl, 10 HEPES, 2 MgCl2, 2.5 CaCl2, 10 HEPES) at room temperature containing 0.5 uM tetrodotoxin (TTX). 2-4 MOhm resistance borosilicate pipettes containing a Cs-methanesulfonate based internal solution (in mM: 130 CsMeSO4; 10 TEA-Cl; 5 MgCl2; 10 HEPES; 10 EGTA; 5 Na-ATP) were used to record Ca 2+ currents under whole cell voltage clamp configuration. To isolate T-type Ca 2+ currents from the TRN neurons, we utilized a subtraction technique. All neurons were initially held at -60 mV and then a current-voltage relationship (I-V; -90 mV to +20 mV; ∆ 5 mV steps) of the neurons were measured with (-100 mV I-V protocol) and without (-60 mV I-V protocol) a 1s hyperpolarization to -100 mV. The traces obtained from the -60 mV I-V protocol was then subtracted from the -100 mV I-V protocol to isolate the T-type Ca 2+ currents using Clampfit and the I-V relationship was calculated from the subtracted traces. Peak current densities were calculated by dividing the peak amplitude by the cell capacitance value.

In vivo Electrophysiology:
12-to 19-week-old mice were deeply anesthetized with isoflurane. One intracranial frontal EEG electrode screw (AP Bregma ~+1.5mm, ML Bregma ~+ 1.5mm) and one intracranial parietal EEG electrode screw (AP Bregma ~-1.3mm, ML Bregma ~+ 2.3mm) and a common ground/reference electrode screw above the cerebellum (AP Lambda ~-1, ML Lambda ~0.0) were chronically implanted using a stereotaxic device (David Kopf Instruments). The electromyogram (EMG) electrodes were placed in the nuchal muscle of mice. Electrodes were soldered to EEG/EMG headmount (Pinnacle Technology Inc., part # 8402-SS, KS, USA). Dental acrylic was used to encase the connections. Mice were tethered to a Pinnacle® recording system for >48 hours habituation in recording chambers following one full week of post-operative recovery.
EEG/EMG signals were recorded for 60 hours from the onset of the light phase (7am; ZT0) when animals are mostly inactive and spend more time sleeping. All signals were digitized at a sampling rate of 1,000 Hz, filtered (1-100 Hz bandpass for EEG; 10-1 kHz bandpass for EMG), and acquired using Sirenia Acquisition program (Pinnacle Technology). Sleep scoring was performed manually using 10 second epochs as previously described 46 with Sirenia Sleep software (Pinnacle Technology Inc.).

Sleep Spindle Analysis:
For the 61 mice, we extracted 6 hours of EEG, recorded on two channels from ZT0-6 hrs, and combined these data with manual stage assignments of NREM, REM, wake and unscoredper each 10-second epoch. Differences between genotype groups in distributions of stage durations and the number and type of stage transitions were tested using linear regression models. The sleep spindle analysis pipeline is described below and illustrated in supplementary For the EEG signal data, we performed a series of channel-level and epoch-level checks for likely artifact, by calculating the per-epoch root mean square (RMS), three Hjorth parameters 4 , and the proportion of clipped points (signal at its maximum or minimum value). Within each mouse and conditional on manually-assigned sleep stage, we then iteratively masked epochs for which any of these five metrics were outliers (+/-3 standard deviation units) for either channel, performing this procedure twice. We additionally estimated per-epoch mutual information as well as cross-spectra and magnitude-squared coherence between the two EEG channels and the electromyography (EMG) channel, to detect recordings with unusually high cross-talk between the EEG and/or EMG signals. For each mouse, we generated a series of figures for visual review: a) epoch-level sequences of manually assigned stages, masked epochs, signal RMS and sigma power, b) stage-specific power spectra for each channel, c) the distribution of absolute log delta power, and RMS of the EMG channel, to determine the broad validity of manual staging for that animal, d) a scatter plot of both absolute and relative sigma power between the two EEG channels, during manually-scored NR epochs that passed the initial round of filtering, e) epochlevel sequences of mutual information and coherence measures between channels, f) spectrograms of epoch-level power over the whole night, g) raw EEG signals from exemplar epochs based on affinity propagation clustering 5 of within-individual epoch-by-epoch distance matrices based on permutation distribution clustering of the EEG 6 , to point to channels with unusually high rates of artifact otherwise missed by our primary approach. Based on visual review of these figures, we eliminated 5 mice for which neither EEG channel yielded sufficient signal data (3 Cacna1i +/+ , 1 Cacna1i RH/RH and 1 Cacna1i +/RH ). In the remaining mice, genotype groups had broadly similar numbers of epochs removed by the above procedure: there were no statistically significant differences for NR epochs, although there were small, nominally significant differences in REM (Cacna1i +/RH and Cacna1i RH/RH had 7.4% and 8.5% of epochs removed compared to 8.9% in Cacna1i +/+ (p = 0.02 and 0.04 respectively), whereas in wakefulness, Cacna1i +/had 7.5% of epochs removed compared to 9.7% in Cacna1i +/+ (p=0.05).
We detected spindles using Morlet wavelet analysis 7 , targeting center frequencies of FC = 9, 11, 13 and 15 Hz (supplementary figure 10), in analyses performed 1) across all epochs, as well as 2) separately within NR, R and W, to allow for stage-specific differences in background sigma activity.
The differences between these two approaches is that different thresholds for spindle detection will be set, as spindles are detected above a multiplicative factor of that individual's median baseline wavelet power value, either based on all epochs (universal baseline), or all epochs of a particular stage (stage-specific). The primary comparisons described here between genotypes, the comparisons focused on differences between stages, or around NREM-REM transitions, were based on spindles detected across all epochs (i.e. using the universal baseline).
Nonetheless, all primary results described in the paper are the same, no matter which approach is used. We used four target frequencies (FC=9, 11, 13, 15 Hz), with each analysis detecting spindles with a peak frequency approximately +/-1 Hz (see Supp Figure 5 and 6), although strong spindles may be detected at multiple FC values. For each individual spindle, the actual frequency can be calculated afterwards. That is, "FC=11 Hz spindles", the group of spindles detected when targeting a center frequency of 11 Hz, is likely to encompass a range of spindle frequencies, mostly between 10 and 12 Hz.
Spindles were defined as intervals for which the wavelet power exceeded a threshold of a) 6 times the (stage-specific) median for that mouse for at least 0.3 seconds (the spindle 'core'), and b) 3 times the (stage-specific) median for at least 0.5 seconds (the 'flanking' waxing/waning region), but not more than 3.0 seconds in total (supplementary figure 10). The threshold of 6 was determined by applying Otsu's method for selecting a threshold that maximizes the between-class variance in wavelet power between putative "spindle" and "non-spindle" intervals, with the flanking threshold being 50% of the core threshold (i.e. 3). Spindles within 0.5 seconds of each other were merged, unless the abovementioned 3.0 second criterion would be violated, in which case they were both excluded. We found that defining thresholds as a multiplicative function of the individual's median wavelet power was more robust compared to using the mean, as wavelet power has a very skewed distribution and outliers (i.e. in large part reflecting true spindles) disproportionately influence the mean but not the median.
After detecting spindles, we recorded the count for each epoch, as well as the following summary statistics: density (spindles per minute), count, mean duration, mean amplitude (peak- For the channels retained in the primary analyses, epochs masked by the initial outlier procedure were excluded from analyses, as were those with an uncertain manual stage assignment. Considering the within-individual distributions of both absolute and relative spectral band power (including slow and gamma bands) for each retained channel, we additionally masked any epoch that was a statistical outlier (+/-3 SDs) for any measure. On average, ~9% of epochs were masked per individual by this procedure. This approach is likely conservative, but feasible because of the relatively long recordings for each mouse, which means sufficient data are retained for analysis.
Of the 56 remaining mice, there were 29 for which both EEG channels passed the above QC procedure, with the remaining 27 having only a single channel passing (11 parietal and 16 frontal channels). In order to maximize the effective sample size and thus statistical power, we combined derived estimates of key measures (e.g. of spindle density, etc) across channels for the 29 mice with both channels retained, otherwise selecting the value for the remaining QCpositive channel, such that all analyses are based on an initial N of 56 (prior to any subsequent measure-specific outlier removal).
Although not a focus of the current work, there are likely detectable topographic differences between frontal and parietal channels. Importantly, in our sample which channels were retained was not associated with genotype (p=0.24), based on Fisher's exact test of the table of genotype (5 levels) by channel (4 levels: neither, frontal, parietal, or both). Furthermore, analyzing the 29 mice with two channels passing QC, we did not observe marked differences in the sample averages for key measures between the two EEG channels: e.g. for relative sigma power (p=0.4) or FC=11 Hz spindle density (p=0.42). In these same mice, key metrics showed moderate to strong intra-individual, cross-channel correlations, e.g. r=0.55 (p=0.00007) for relative sigma power, and r=0.58 (p=0.000001) for FC=11 Hz spindle density, supporting the decision to base our primary analyses on a set of composite measures, to increase power.
Nonetheless, secondary analyses performed on only one of the two channels yielded substantively similar results.
Initially mice were assigned to two Cacna1i +/+ groups, intended for comparison with KO and RH mice respectively. As we did not observe differences between Cacna1i +/+ groups for the primary outcome measures, we pooled all Cacna1i +/+ mice into a single group to increase power.
Based on manual staging, we annotated "stable" epochs as those flanked by similarly scored epochs. We also identified all NREM-REM transitions, grouping those epochs into 20second bins (i.e. pairs of epochs) labeled E-5 to E+1 as follows: E0 marks the transition period (the two epochs flanking the transition), E-1 marks from 30 to 10 seconds prior to the transition, E-2 from 50 to 30 seconds, and so on, and E+1 marks from 10 to 30 seconds after the NR/R transition.
Primary analyses between genotype groups controlled for age, sex (2 Cacna1i +/+ mice were female), and the day (relative to the start of data collection for that animal) as well as the date on which EEG data for that animal were extracted for analysis. For all primary comparisons, statistical outliers (+/-3 SDs) in the dependent variable were removed from analysis. For analyses of epoch-level data, e.g. around NREM/REM transitions, we first created individual-level means and then performed standard linear regression on these measures.

Statistics, Reagents and animal models:
Sample sizes were chosen as per literature standard for biochemistry, slice electrophysiology and in vivo EEG experiments. It has been shown previously in the literature that such sample sizes are sufficient for confident statistical conclusions. A small number of mice in the in vivo EEG experiments were excluded based on poor quality recording, as detailed in the manuscript. For the remaining mice, we removed epochs that were statistical outliers as detailed in the sleep spindle analysis section. For slice physiology and in vivo EEG, the data analysis was performed in a blinded fashion. In particular, rebound bursting and calcium current analysis and the EEG manual sleep scoring for the different genotypes were de-identified before analysis. All Statistical tests used are listed in the appropriate results section. Non-parametric statistical tests were utilized whenever necessary. There is no estimate of variation and the variance between groups were not statistically compared.
We used a custom-made CaV3.3 antibody that can be available from us when requested.
We validated the CaV3.3 antibody in the Cacna1i -/animals. The data is provided in Figure 1.
All of the experiments with animals were approved by the Broad Institute IACUC (Institutional Animal Care and Use Committee). All animals used were male C57BL6 mice. We