Mapping of brain activity by automated volume analysis of immediate early genes

Imaging and registration of IEGs in the mouse brain

The original iDISCO protocol for immunolabeling and brain clearing permits detection of immunolabeled structures throughout the intact adult mouse brain and is compatible with a wide variety of antibodies to diverse epitopes (Renier et al., 2014). In particular, it is compatible with antibodies that detect a variety of IEGs (see below). Specific signals are detected using secondary antibodies coupled to Alexa fluorophores that emit in the far-red spectrum (647 nm laser excitation), to which the cleared brain is essentially transparent (Renier et al., 2014). In the blue-green spectrum (480 nm laser excitation), intrinsic tissue fluorescence highlights the major structures in the brain, such as myelinated tracts, cortical regions and various deep nuclei (Figure 1B and S1A), thus providing a counterstain without the need for counterstaining (Movie S1).

Building on iDISCO, we developed ClearMap, a cell detection and registration pipeline outlined in Figure 1A and Movie S2, which relies on key improvements in iDISCO sample preparation and imaging procedures, as follows.

First, we optimized imaging conditions to achieve the minimal x,y,z pixel size of 4.06 × 4.06 × 3μm/pixel necessary to image a large field of view at cellular resolution while still preserving acquisition speed and a relatively low data footprint (See Methods and Figure S1). We found that imaging at 1.6X with a 2X 0.5NA objective and continuous horizontal scanning of the light sheet provided a resolution sufficient to image single c-Fos positive cells within the sharp center of the lens, which covers the length of the entire mouse forebrain in sagittal orientation (Figure S1A). At this magnification, the optical maximal resolution is almost isotropic at the center of the imaging lens (Figure 1E and S5), so that images acquired in one plane can be combined to give an undistorted representation in the other orthogonal planes. The measured z-resolution (thickness) of the light sheet at the numerical aperture used (0.1) is about 5μm at its thinnest point (manufacturer's information). To maximize the z-resolution throughout the dorso-ventral length of the brain, the thinnest focal point of the light sheet was horizontally scanned and a composite image was created for each plane by using a contrast projection, thus circumventing the Gaussian shape of the beam. While the camera sensor field of view (10.4mm in length) and the tiling of 3 parallel light sheets covers the mouse brain in sagittal orientation from the cerebellum to the olfactory bulb, the sharp center of the imaging lens covers a smaller region (7.2mm in length), which limits the region covered by one tile to the forebrain excluding the olfactory bulbs (Figure S4A). However, the consistency of the sharpness and signal intensity for detected cells throughout this region suggested that it could be possible to build a robust pipeline to automatically analyze the distribution of IEGs within the brain (Figure S1B,C).

Second, several efficient 3D image registration software packages have recently been developed. As mentioned, the intrinsic tissue fluorescence highlighting major structures (Figure 1B and Figure S1A) offers enough detail and contrast to allow in principle for automated registration of scans using signal harvested with the 480nm laser excitation. However, tissue distortion due to shrinkage during 3DISCO clearing (when used after the iDISCO staining) precluded automated image registration (Menegas et al., 2015) (Figure 1B). We traced the major source of shrinkage to TetraHydroFuran used in the 3DISCO protocol, and were able to modify the protocol to dispense with this chemical, replacing it with a combination of Methanol and DiChloroMethane (Figure 1B). With this novel iDISCO protocol, which we term iDISCO+, only minor tissue deformation occurred (see below). The clearing performance of iDISCO+ enables imaging of c-Fos+ cells with a consistent signal throughout the working distance of the objective, well over 6mm deep in the tissue (Figure S1A and B). There was no drop in the number of c-Fos+ cells detected throughout the imaging depth, as the density of c-Fos+ cells was identical on both sides of the midline (figure S1B). We verified that the new clearing protocol remains compatible with all tissues and has an identical clearing performance to that seen with iDISCO/3DISCO (Figure S2G and data not shown).

The ClearMap pipeline includes interfaces to the Elastix open-source software (Klein et al., 2010; Shamonin et al., 2014) to automatically register the iDISCO+ light sheet scans, as previously developed for STP imaged mouse brains (Kim et al., 2015). We quantified the extent of deformation by registering lightsheet scans of cleared half-brains to matched MRI scans of the same samples (Figure 1C and Figure S4A-B). The optimized affine transformation coefficients determined by Elastix indicate a modest linear reduction in size of 7% and 9% in the medio-lateral and rostro-caudal axis respectively and a small linear expansion of 5% along the dorso-ventral axis (n=5). The overall volume shrinkage as determined by the determinant of the transformation matrix was 11%, down from the reported 50% shrinkage observed with the original 3DISCO clearing protocol (Richardson and Lichtman, 2015). For whole brains, which require a little distortion to fit in the MRI tube, we observed similar but smaller coefficients when registering whole brains to the STP reference atlas (5% volume difference, n=6, Figure S4B), suggesting that whole brain morphology may be better preserved during clearing. The average displacement after applying the optimized non-linear (b-spline) transformation from the sample to the reference was 4 voxels (Figure 1C). Most of the non-linear transformations corrected for a small expansion of the ventricular volume and adjustment of the cortical curvature. The new clearing protocol therefore enabled image registration of the cleared samples datasets to the STP reference (STPR) mouse brain aligned to the Allen Brain 25μm reference Atlas, whereas the original clearing protocol would have required custom reference Atlases (Kim et al., 2015) (Figure 1D and Figure S4D-F).

We next developed a complete suite of open-source image processing tools within our modular ClearMap software to detect cells in 3D and register their coordinates onto the reference atlas, which has a similar workflow to previously used analysis pipelines (Kim et al., 2015; Menegas et al., 2015), and runs under 90 min per brain in fully automated fashion on a desktop workstation, facilitated by our optimized 3D cell detection algorithm and compact dataset. To validate the detected accuracy, the cell detection was run on a dataset for which positive cell voxels were manually annotated by two users. The filters and threshold were tuned until the disagreement ratio between the manual and automated annotation was similar to the ratio between the two users (Figure 1F, S3C and see Extended Methods).

As iDISCO+ relies on immunostaining, it offers flexibility when choosing the IEGs used to map brain activity. To test the ability of the method to detect the induction of known IEGs, we relied on activity evoked in the barrel cortex by whisker stimulation as a tool to validate antibodies for c-Fos, Arc, Egr-1, Phospho-S6 and Npas4. Cortical Layer 4 neurons receiving inputs from the large mystacial whiskers on the snout are topographically organized in the barrel cortex. Their map can be readily visualized using intrinsic tissue fluorescence, which is preserved in the blue-green spectrum in the iDISCO+ protocol. We trimmed all whiskers except select rows or arcs; exposed animals for 1 hr to an enriched environment in which the remaining whiskers were stimulated as the animals explored; processed brains using the iDISCO+ protocol with each of the antibodies, using secondary antibodies conjugated to a fluorophore that emits in the far-red spectrum; and manually segmented cortical layers to visualize the activity map of the barrel cortex in a virtual projection of layer 4, or as a virtual section along the remaining row or arc (Figure 2A and B). The IEGs tested all showed increased staining in the stimulated rows or arcs (B row shown in Figure 2). Of those tested, in this system, c-Fos and Phospho-S6 showed the highest induction in labeled cell density between stimulated and deprived barrels (4.1 and 6 respectively, ratio of the number of cells at the center of B2 over C2). Egr-1 had the highest baseline activity (ratio of 1.3). Arc showed a banded pattern by cortical layers (high expression in layers 6, 4, 3 and 2) and an induction ratio of 1.7 in this paradigm (Figure 2C). Npas4 showed modest up-regulation of expression 30 min and 1 hr after stimulation (2.2 and 2.1 respectively) (Figure 2A and B), well within the induction window reported in vivo for the cortex (Lin et al., 2008; Spiegel et al., 2014).

We then used ClearMap to analyze the data obtained with whisker trimming and stimulation for all IEGs and different combinations of rows and arcs, and were able to detect the increased cell density in the corresponding stimulated barrels for various whisker stimulations (Figure 3A-D). To check the accuracy of the registration, we verified that we could use the Atlas annotations to automatically isolate activity maps in the barrel cortex and in its thalamic relay station, the Ventro Postero Medial (VPM) nucleus, from the complete 3D dataset (Figure 3D-E).

Mapping of pharmacologically induced brain activity

Because of their broad spectrum of action, many classical psychoactive drugs affect the function of multiple brain regions. The systematic mapping of brain activity provides the opportunity for an unbiased investigation of regions up or down regulated by such drugs. As an example, we selected the antipsychotic Haloperidol, a well-studied high affinity antagonist of D2 dopamine receptors and low affinity agonist of serotonin receptors.

Mice were treated with an acute saline or Haloperidol injection, and their brains were processed through our analysis pipeline (n=4 for each group). ClearMap detected statistically significant increases in activity in hallmark target regions of Haloperidol, including the internal segment of the Globus Pallidus (GPi), the Caudo-Putamen formation (CPu) and the Nucleus Accumbens (NAc), showing that the deepest structures in the samples are effectively labeled by the antibodies to c-Fos (Figure 4 and table S1). To visually parse the data and find activity hot spots, we generated density maps by summing spheres of 375μm (15 pixels) diameter and uniform intensity centered on each cell. Results obtained from automated segmentation of the brain regions were confirmed by inspection of both voxelized density maps and raw data (Figure 4A-C). A useful way to depict differences in cell counts in various regions between experimentally-treated animals versus controls is with the use of P-value maps, in which regions are highlighted only when the P value is below a certain level (typically 0.05 or 0.01, indicated in each case below), thus indicating statistical significance; one color (typically green) is used when the number is higher in treated than controls, and a different color (typically red) when the number is lower (Figure 4). This enables a rapid and unbiased discovery method for identifying responsive regions for further analysis. The density maps, P-value maps and Result Tables can be presented either as total cell counts or by summing the peak intensity of each cell (Figure S6A and B). Using total cell counts, peak intensity counts, or cell size weighted counts yielded very similar results (see Methods), although weighted counts typically yielded greater statistically significant differences of tests versus controls (Figure S6A).

These analysis tools again highlighted known areas of c-Fos induction, and also identified many others of interest (Figure 4 and table S1). Interestingly, a few nuclei in table S1, such as the Ventral Tegmental Area (VTA), that trended towards significance but failed to achieve it had in fact a significantly higher activity restricted to a sub-region of the nucleus (here, the lateral part of the VTA) when manually inspected on the density map (Figure 4C and D); the failure to achieve significance across the entire subregion presumably reflects a localized increase in c-Fos activation within the larger, otherwise unchanged subregion. The single cell resolution of our method thus enabled us to detect a change in the detailed activation pattern of a nucleus.

Mapping of cortical activity during an exploration task

Live recording of whole mammalian brain activity is not possible for behaviors involving active exploration, and IEGs can provide a simple solution to map brain activity in freely moving mice. To demonstrate this, we chose to look for brain regions differentially regulated when mice explore a new environment with or without their whiskers. Mice had all whiskers either trimmed or spared, were transferred to an empty cage in the dark for 24 hr, after which they were transferred to a new cage containing an enriched environment for 1 hr in the dark, and then sacrificed.

Automated isolation of cortical regions was performed using ClearMap, (Figure 5A) to display averaged density maps as well as their variability shown as standard deviation maps for both groups. While the mapped values for standard deviation are naturally higher in active regions, they provide a visual way to assess the variability and significance of activity differences between groups (see Figure 5C). As expected, in the shaved group, c-Fos positive neuron density was significantly reduced in the caudomedial region of the barrel cortex, the region of the brain that receives input from the whiskers, and also in a patch at the center of the primary motor cortex that controls whisker movements (Figure 5A). Moreover, activity in the dorso-medial part of the VPM of the thalamus (relaying whisker input from the brainstem to the cortex) was also reduced in the shaved group, while other parts of the VPM and PO nucleus (receiving inputs from other parts of the body) had higher densities of active neurons (Figure 5B). There is a strong cross-regulation in the cortex between different sensory modalities (He et al., 2012), and we indeed recorded significant differences in the primary auditory and piriform cortex between shaved and non-shaved groups, but not in the visual cortex (as expected, since the experiment was performed in the dark) (Figure 5A). Interestingly, we noticed a significant depression of the activity in the anterior part of the Agranular Insular Cortex in the shaved animals (Figure 5C), a region not traditionally associated with the sensory integration of whisker-evoked signals. We also observed increased activity in shaved animals in the regions of the somatosensory cortex associated with the nose and mouth (Figure 5A), which suggests that, lacking whiskers, the shaved animals instead actively explored the enriched environment by touching objects with those regions of their faces.

Activity mapping with axonal tracing to investigate parental behavior

While IEG maps obtained from electrical, optogenetic or pharmacological stimulation of specific brain nuclei can provide a good indication that the activity detected arises from connected regions (Bepari et al., 2012), maps derived from natural behaviors cannot provide a definitive indication about the functional relationships between the active regions. Therefore, it is useful to combine axon tracing with IEG mapping, to determine whether regions connected by specific axonal projections show correlation (or anti-correlation) in IEG expression, indicative of potentially functional links.

Parental care is a critical behavior hard-wired in the mammalian brain, showing sexual dimorphism in both activity patterns and connectivity (Dulac et al., 2014). Previous work has shown that in laboratory mice, virgin males are aggressive towards pups and switch to parenting behavior after mating (Saal, 1985; Tachikawa et al., 2013). The activity of a group of galanin-expressing neurons located in the Medial Pre-Optic nucleus (MPO) in the hypothalamus has been shown to drive this behavior (Wu et al., 2014). We examined the 3D c-Fos expression patterns in parenting females (n=3) and infanticidal males (n=6), compared to their respective controls (n=6 for each group). The ventral side of the brain showed a clear dichotomy in the distribution of activity between females and males in the hypothalamus (Figure 6A). Females showed a higher density of activated neurons in the anterior region of the hypothalamus, while the posterior hypothalamic region was more active in males. Indeed, as previously described (Wu et al., 2014), the density of active neurons was lower in the MPO in virgin males and control females than in parenting females (Figure 6B), while the opposite was observed in the DMH and PH Nuclei (Figure S7A). The number of active neurons was also higher in the MPO of aggressive males compared to control males, although not as high as in parenting females (Figure 6B). It is possible that the active neurons of the MPO in infanticidal males are a different population than the Gal+ neurons active in parenting females. Indeed the activity patterns in the MPO appear complementary between parenting females and infanticidal males (Figure 6B). The Bed Nucleus of the Stria Terminalis also showed an interesting dichotomy, with more active neurons found ventral to the anterior commissure in parenting females while the dorsal side was more active in infanticidal males.

To test whether we could correlate the activity of MPO neurons during parenting tasks to a modulation of activity in other regions of the brain, we injected a floxed eYFP expressing virus into the MPO of Gal::Cre virgin females and virgin males. Using iDISCO+, we double-immunolabeled the samples for c-Fos and YFP expression. Volume imaging showed that the Galanin neurons in the rostral part of the hypothalamus had extensive projections to the thalamus, ipsilateral olfactory bulb, retro-splenial cortex, amygdalar cortex, pons and peri-acqueductal grey (Movie S3). Very extensive projections from the MPO^Gal neurons were seen to the Cortical Amygdalar Area (CoA) in both males and females (Figure 7B). The c-Fos density in the CoA was found to be higher in infanticidal males than in control males and both parenting and control females, specifically in the region receiving the bulk of the projections from the hypothalamus (Figure 7C). MPO neurons are inhibitory as evidenced by Gad1 expression (Wu et al., 2014), so it is possible that the increased activity in the MPOA neurons could drive a tonic inhibition of the CoA, which could act as a brake when the global activity in the MPO is high in parenting females. These results show that our method can be used to automatically map the position of c-Fos positive neurons in the brain, and correlate active and inactive regions to virally traced axonal projections, thereby generating hypotheses for further testing.

Animals

Animals were maintained on 12 h:12 h light/dark cycle with food and water available ad libitum. All surgical and behavioral procedures in mice were performed in compliance with NIH guidelines and protocols approved by the IACUC of the Rockefeller University and of the Harvard University. The Gal::Cre BAC transgenic line (STOCK Tg(Galcre)KI87Gsat/Mmucd, 031060-UCD) was imported from the Mutant Mouse Regional Resource Center and is described (Wu et al., 2014).

iDISCO+ sample processing

Modifications and continuous updates to the protocol can be found at http://www.idisco.info.

iDISCO+ protocol

Adult mice were euthanized with a rising gradient of CO2 and fixed with an intracardiac perfusion of 4% PFA in PBS. All harvested samples were post-fixed overnight at 4°C in 4% PFA in PBS. Samples were processed with the iDISCO immunolabeling protocol, as detailed in the supplementary methods and in the bench protocol.

Light sheet imaging

Cleared samples were imaged in sagittal orientation (right lateral side up) on a light-sheet microscope (Ultramicroscope II, LaVision Biotec) equipped with a sCMOS camera (Andor Neo) and a 2X/0.5 objective lens (MVPLAPO 2x) equipped with a 6mm working distance dipping cap. Version v144 of the Imspector Microscope controller software was used. The samples were scanned with a step-size of 3μm using the continuous light sheet scanning method with the included contrast blending algorithm for the 640nm and 595nm channels (20 acquisitions per plane), and without horizontal scanning for the 480nm channel.

Image processing and analysis

For display purposes in the figures, a gamma correction of 1.47 was applied on the raw data obtained from the light-sheet microscope (but not on the density maps, which are linearly scaled). Maximum projections were performed using ImageJ (NIH, http://imagej.nih.gov/ij/). Imaris (Bitplane, http://www.bitplane.com/imaris/imaris) was used for the manual 3D annotations in figure 2, 3 and movies. Imaris was also used to generate the orthogonal projections of the unprocessed data in all figures.

ClearMap analysis

All analysis and quantifications where performed with the newly developed open source ClearMap software as shown in the example scripts (Supplementary data 2), and the associated documentation. The latest version of ClearMap can be downloaded from www.idisco.info.

The cell detection is tailored for cell nuclei and uses a background subtraction via morphological opening, followed by a sequence of filters, morphological operations and a 3D peak detection. We chose a threshold of 20 voxels for the cell size. The background was removed by subtraction of the morphological opened image with a disk shape structure element with main axis of 7 pixels of diameter. Cells were detected from peaks and subsequent watershedding, removing background pixels below an intensity cutoff of 700. Density maps were generated by summing spheres of 375μm diameter and uniform intensity centered on each cell. Cell density maps and statistics are presented in the figures. Samples were registered using the average autofluorescence STPR brain (Kim et al., 2015) registered to the Allen Brain Institute 25μm map, and companion annotation map (http://alleninstitute.org/).

Statistics

Cell counts or intensities of each sample in considered regions or annotated brain areas between different groups were compared using the independent two sample student t-test assuming unequal variances. Statistical tests were performed numerically using the SciPy statistics library (http://www.scipy.org/). Using ClearMap as a discovery tool p-values were corrected for multiple comparison to q-values to control for false-discovery rate (Benjamini and Hochberg, 1995).

MRI scanning

Wild type adult littermates were transcardially perfused with 0.9% saline and 4% PFA. The brains were dissected, post-fixed, cut sagitally in half (to fit in the MRM scanner tube without deformation) and stored in PB until imaging.

All MRM scans were performed on a 9.4T Bruker vertical bore magnet. Fomblin was used to seal the specimen to prevent dehydration and as the same time to prevent susceptibility effects.

Haloperidol injections

Adult (10 weeks old) male littermate mice were individually housed for 5 days and on the 6-th day were injected i.p. either 1 mg/kg Haloperidol or vehicle (0.3%-TWEEN-80 in 0.9% saline). The mice were euthanized 3 hours after injection, transcardially perfused with 0.9% saline and 4% PFA. The brains were dissected out, post-fixed overnight in 4% PFA at 4°C and washed in PB until entering the iDISCO+ pipeline.

New environment exploration

Mice were anesthetized with a mixture of Ketamine / Xylazine (Ketamine at 0.1mg/g, Xylazine at 0.01mg/g) and their whiskers were trimmed or left intact (sham operation for the intact-whiskers group). Mice were then left to recover single-housed in an empty cage with access to food and water for 24h in the dark. Mice where then transferred to a new cage containing an enriched environment crowded with cardboard tubes and wire tube racks for 1h to explore. Mice were then promptly anesthetized, perfused and processed according to the iDISCO+ protocol.

Viral injections

Viral tracing experiments were performed in Gal::Cre mice at ∼8-12 weeks of age. Mice were anesthetized with 100 mg/kg ketamine and 10 mg/kg xylazine via intra-peritoneal injection. 500 nl of AAV1-EF1a-DIO-hChR2(H134R)eYFP (UPenn Vector Core) were injected into the right medial preoptic area. Two weeks later, mice underwent behavioral testing.

Parental behavior assay

Parental behaviour assays were performed in sexually naive Gal-cre females at ∼8-12 weeks of age, 2 weeks after viral injection. Animals were individually housed for one week before testing. On testing day, animals were habituated to the testing arena for 30min. Two C57BL/6J pups (age P1-3) were then introduced into the cage and placed at the farthest corners from the resident's nest. A timer was started at the first retrieval of a pup to the nest. 90 min after retrieval, mice were deeply anesthetized with isoflurane and transcardially perfused with ice-cold PBS followed by 4% paraformaldehyde (PFA, in PBS).

Infanticide behavior assay

Infanticide behavior assays were performed in sexually naive C57BL/6J males at ∼8-12 weeks of age that had not been exposed to pups. Animals were individually housed for one week before testing. On testing day, animals were habituated to the testing arena for 2h with hydrogel and food in the cage. Two C57BL/6J pups (age P1-3) were then introduced into the cage and placed at the farthest corners from the resident's nest. The first olfactory investigation marked the beginning of the assay, which was continued if the resident attacked at least one pup. When a pup was attacked, the assay was ended immediately and the wounded pup was euthanized. 90 min after first olfactory investigation, infanticidal males were euthanized, perfused and their brains collected and post-fixed for iDISCO treatment.