Mediodorsal thalamus and ventral pallidum contribute to subcortical regulation of the default mode network

Humans and other animals readily transition from externally to internally focused attention, and these transitions are accompanied by coactivation of a group of brain regions collectively known as the default mode network (DMN). While the DMN was considered a cortical network, recent evidence suggests subcortical structures are part of the DMN. Here we investigated the role of ventral pallidum (VP) and mediodorsal thalamus (MD) in DMN regulation in the tree shrew, a close relative of primates. We combine electrophysiology and deep learning-based motion tracking to perform unsupervised classification of behavioral states. We found gamma oscillations in VP and MD coordinated with gamma in the anterior cingulate (AC) cortex specifically during DMN states. Similar enhancements were found for high gamma, but only at subcortical sites. Cross-frequency coupling between gamma and delta oscillations were higher during DMN than other behaviors, underscoring the engagement of MD, VP, and AC circuits. Our findings highlight the importance of VP in DMN regulation in the tree shrew, consistent with rodent studies, and demonstrate a role for MD thalamus in DMN regulation. Our results extend homologies in DMN regulation among mammals, and underline the importance of thalamus and basal forebrain to the regulation of DMN brain states.

Methods
Materials and methods

The local ethical committee on animal experimentation (canton of Fribourg), approved all experimental procedures.

Animals. Three adult tree shrews, T. belangeri, of either sex were housed under a 13/11 LD cycle in a 3 m3 cage with branches, some enrichment elements, and ad libitum access of food and water. The cage was connected to a nest box with a tube.

Surgical Procedures. Animals first received i.m injections of Alfaxan (40 mg/kg) to induce anaesthesia and Atropine (0.08 mg/kg) to prevent secretions. Animals were then intubated using a modified otoscope (Bebird, Alhambra, CA),  ventilated at 100 bpm (Small Animal Ventilator, Harvard Apparatus, Cambridge, MA), and placed in a stereotactic frame (David Kopf Instruments Tujunga, CA). Anaesthesia was maintained with isoflurane (1-3%) in pure oxygen, and end tidal CO2 was monitored (Physiosuite, Kent Scientific Torrington CT) and maintained at ~4%. Lidocaine (0.5 ml 1%) was injected near the incision site, a midline incision was made and the skull was exposed. Three 1.5 mm stainless-steel bone screws (WPI Hertfordshire, UK) were implanted with two located above the cerebellum as a reference and ground. Burr holes were drilled, and epoxy coated tungsten electrodes (FHC Bowdoin ME)  with a tip resistance ~ 150 kW were lowered to the recording sites: Ventral pallidum (AP 7.6 mm, ML 3.0 mm, DV -7.8 mm), anterior cingulate cortex ( AP 11.1 mm, ML 0.8 mm, DV -1.5 mm) mediodorsal thalamus (AP 4.9 mm, ML 1.0 mm, DV -5.4 mm) and primary visual cortex (AP 2.0 mm, ML 1.4 mm, DV -1.0 mm). All coordinates are from the interaural line. Electrodes were fixed to the skull with super glue (LOCTITE, Westlake OH) and Paladur dental cement (Kulzur Inc. Hanau Germany). Electrodes were wired to a socket connector, and the connector was attached to the skull with dental acrylic. The incision was closed about the connector with sutures, and the animal was allowed to recover for at least one week prior to testing.

Data Acquisition. LFP and accelerometer data were collected using a wireless battery-powered data logger (Neurologger 2A, Zürich Switzerland), additionally, an infrared receiver on the neurologger was used for aligning the LFP and accelerometer data with video recordings. All channels of neural signals and accelerometer data were digitized at 400 Hz and no further filtering was performed on the LFP data.

Home cage Recording. Video recordings of the animals in their home cage were made using a wide field, 103⁰x58⁰, CMOS camera (DS-2CD2143G0-IS, HIKVISION, Hangzhou China) mounted on top of the cage. After connecting the Neurologger, tree shrews were initially kept in their nest box for 10 min in order to acclimate to the Neurologger device. Home cage recordings typically lasted for 5-6 hours, between 9:00-18:00 during the animals’ perspective daytime.

DeepLabCut Tracking. DeepLabCut (DLC) (Mathis et al., 2018; Nath et al., 2019) was used to track the animal’s location in the home cage. First, videos were pre-processed by cropping appropriately and down sampling to 2 fps, and then cut into 20 minute segments. The model was trained for 500,000 iterations after manually labelling 50 frames from each video segment. For each frame, we labelled the nose and neck. The output consisted of body part coordinates in x,y coordinates and the corresponding likelihood estimate, using the same model to analyse all the video segments from the same animal. In most circumstances, DLC can accurately track the tree shrews in most of the frames. For those videos with obvious missing frames, an additional 50 frames were hand labelled, and the network was retrained with additional 200,000 iterations.

Preprocessing and Spectral Analysis. We partitioned the LFP data into 0.5-s epochs for further analysis. Power spectra were calculated by fast Fourier transforms (FFT). We calculated the band power by calculating the mean value of the power spectrum between 40 - 60 Hz (gamma band) and 60-150 Hz (high gamma band).

Hidden Markov Model.

We designed an HMM with three states to capture the three groups of behavioural states detailed in the results, and four output symbols corresponding to High/High, High/Low, Low/High, and Low/Low combinations of ACL and DLC signal values based on preliminary observations of our data. The thresholds between High and Low sensor values were determined based on the median value of the signal across the recording session. We used maximum likelihood estimation to find HMM state transition and output symbol emission probabilities and the Viterbi algorithm to compute the most probable state sequence given the estimated parameters. 

Granger Causality.

To test the information transfer between VP, AC, MD, and V1 brain regions, we used LFPs and a multivariate linear vector autoregressive (VAR) model from Matlab Multivariate Granger Causality (MVGC) toolbox (Barnett and Seth, 2014) for granger causality analyses. The maximum model order for model order estimation was 20 ms, and Akaike information criteria (AIC) was used. The model parameter for the VAR model estimation was the locally weighted linear regression (LWR). We used F-testing with a false discovery rate (Q < 0.05) for the pairwise conditional Granger causality estimation. Kruskal-Wallis test was applied to determine the significance between the information transfer directions. T-test was applied to compare the significance of granger causality between the DMN state and the exploration state.

Cross-frequency Coupling

We first filtered our raw data with butterworth band pass filter into delta (0.5-4 Hz) and gamma (40-60 Hz) bands. Then we applied the Hilbert transform to extract the instantaneous phase from the delta band and the instantaneous amplitude from the gamma band. We display the cross-frequency coupling in polar format (Fig. 7B).

Statistical analyses reproducibility.

Experiment details are provided in the text and Materials and Methods section. All the statistical analyses were performed in Matlab. In comparison of likelihood in Fig.1, we applied the non-parametric Wilcoxon signed rank test as the data did not follow a normal distribution. For the same reason, to compare granger causality values within VP, AC, and MD, we used Kruskal-Wallis test (Fig.4G). For normally distributed data with equal variance, t-test or ANOVA tests were applied according to group number. In Fig.6E, we applied the Warren-Sarle test in order to assess the bimodality of the delta band amplitude distribution for sleep posture epochs. In Fig.7G, we used the Circular Median test (Matlab CircStat Toolbox) for a non-parametric multi-sample test of equal medians for circular data.