Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

An Optimized Mouse Brain Atlas for Automated Mapping and Quantification of Neuronal Activity Using iDISCO+ and Light Sheet Fluorescence Microscopy

An Optimized Mouse Brain Atlas for Automated Mapping and Quantification of Neuronal Activity... In recent years, the combination of whole-brain immunolabelling, light sheet fluorescence microscopy (LSFM) and subsequent registration of data with a common reference atlas, has enabled 3D visualization and quantification of fluorescent markers or tracers in the adult mouse brain. Today, the common coordinate framework version 3 developed by the Allen’s Institute of Brain Science (AIBS CCFv3), is widely used as the standard brain atlas for registration of LSFM data. However, the AIBS CCFv3 is based on histological processing and imaging modalities different from those used for LSFM imaging and consequently, the data differ in both tissue contrast and morphology. To improve the accuracy and speed by which LSFM-imaged whole-brain data can be registered and quantified, we have created an optimized digital mouse brain atlas based on immunolabelled and solvent-cleared brains. Compared to the AIBS CCFv3 atlas, our atlas resulted in faster and more accurate mapping of neuronal activity as measured by c-Fos expression, especially in the hindbrain. We further demonstrated utility of the LSFM atlas by comparing whole-brain quantitative changes in c-Fos expression following acute administration of semaglutide in lean and diet-induced obese mice. In combination with an improved algorithm for c-Fos detection, the LSFM atlas enables unbiased and computationally efficient characterization of drug effects on whole- brain neuronal activity patterns. In conclusion, we established an optimized reference atlas for more precise mapping of fluorescent markers, including c-Fos, in mouse brains processed for LSFM. . . . . . Keywords Light sheet fluorescence microscopy iDISCO Tissue clearing Brain atlas C-Fos Whole brain imaging Introduction expression patterns of c-Fos, a proxy for neuronal acti- vation (Dragunow and Faull 1989). Rodent models are important tools in preclinical drug devel- Recent advances in immunohistochemical methods and opment for central nervous system (CNS) disorders (Bobela optical clearing techniques have, together with ex vivo imag- et al. 2014; Esquerda-Canals et al. 2017; Leung and Jia ing technologies such as light sheet fluorescence microscopy 2016). A common method for characterizing central ef- (LSFM), enabled whole-organ imaging (Chung et al. 2013; fects of potential novel therapies is to quantify Ertürk et al. 2012;Jensen et al. 2015; Kjaergaard et al. 2019; Renier et al. 2014; Rocha et al. 2019; Secher et al. 2014). As a Electronic supplementary material The online version of this article result, it is now possible to visualize c-Fos expression at the (https://doi.org/10.1007/s12021-020-09490-8) contains supplementary single cell level in the intact adult mouse brain (Kjaergaard material, which is available to authorized users. et al. 2019;Nectowetal. 2017;Renier et al. 2016). In recent years, automated image analysis algorithms have * Jacob Hecksher-Sørensen been developed enabling 3D quantification of activated neu- jhs@gubra.dk rons and their signal intensities in the adult mouse brain (Detrez et al. 2019;Jensenetal. 2015;Liebmannetal. Gubra ApS, 2970 Hørsholm, Denmark 2016;Nectow et al. 2017; Salinas et al. 2018;Schneeberger Department of Applied Mathematics and Computer Science, et al. 2019). The first step of the analysis process is to register Technical University Denmark, 2800 Kongens Lyngby, Denmark LSFM imaging data onto a common reference brain which Danish Research Centre for Magnetic Resonance, Centre for contains annotated brain regions. Today, the most widely used Functional and Diagnostic Imaging and Research, Copenhagen University Hospital Hvidovre, 2650 Hvidovre, Denmark mouse brain atlas is the common coordinate framework 434 Neuroinform (2021) 19:433–446 version 3 (CCFv3), developed by the Allen Institute for Brain Materials and Methods Science (AIBS) (Allen Institute for Brain Science 2011, 2015, 2017;Kuan et al. 2015;Wang et al. 2020). For quantification Animals of fluorescent signals, registration is followed by cell detec- tion, e.g. ClearMap, to segment and count c-Fos positive cells Male C57Bl/6 J mice were obtained from Janvier Labs (Le (Nectow et al. 2017; Renier et al. 2016) or extract voxel in- Genest-Saint-Isle, France), and were maintained in standard tensities (Salinas et al. 2018). Finally, the results can be housing conditions (12 h light/dark cycle and controlled tem- assigned to specific regions using the anatomical reference perature of 21–23 °C). Mice had ad libitum access to tap water atlases such as those provided by AIBS. and regular chow (Altromin 1324, Brogaarden, Hørsholm, LSFM image processing pipelines have improved quanti- Denmark) or high fat diet (60% fat, 21% carbohydrates, tative whole-brain 3D imaging. However, the quality of the 19% protein; Ssniff Spezialdiäten GmbH, Soest, Germany). LSFM results is highly dependent on sample processing and The LSFM atlas was established based on analysis of 139 the imaging methods applied. Whole-organ immunolabelling brains from 8 to 10 weeks old male chow-fed mice. The requires lipid extraction to make the tissue permeable to anti- pharmacology-induced neuronal activity study involved two bodies (Kim et al. 2018) and enable deep tissue imaging groups of lean mice and two groups of DIO mice. All groups (Vigouroux et al. 2017). In particular, myelin fibers which were aged matched (38 weeks) and consisted of n = 6. Lean are lipid-rich (Villares et al. 2015), are more likely to be af- and DIO control group animals received phosphate buffered fected by lipid extraction, leading to non-uniform morpholog- saline with BSA, lean and DIO treatment group animals re- ical changes within the brain. Also, various clearing medias ceived semaglutide (Ozempic®, Novo Nordisk A/S, have different chemical properties which will result in either Bagsværd, Denmark) dose of 0.04 mg/kg. Both groups were shrinkage or expansion of brain structures (Wan et al. 2018). administered subcutaneously 5 ml/kg and the animals were In contrast, the AIBS CCFv3 is based on vibratome-sectioned sacrificed 4 h post-dose. All animal procedures were conduct- and two-photon microscopy imaged brains. Consequently, ed in compliance with internationally accepted principles for brains imaged with LSFM differ from the AIBS CCFv3 atlas the care and use of laboratory animals and were approved by with respect to morphology and signal intensity. This affects the Danish Animal Experiments Inspectorate (license #2013- the registration accuracy and because the morphological 15-2934-00784). changes introduced by the sample processing are tissue-de- pendent, some brain regions are more prone to erroneous Sample Preparation for Immunohistochemistry alignment than others. As result, subsequent data analysis re- quires time-consuming validation and manual correction to Animals were transcardially perfused with heparinized PBS ensure accurate quantification. This is particularly relevant in and 40 ml of 10% neutral buffered formalin (CellPath, pre-clinical research where group sizes are often relatively Newtown, UK) under Hypnorm-Dormicum (fentanyl large in order to provide better statistical power. 788 μg/kg, fluanisone 25 mg/kg and midazolam 12.5 mg/kg, In our experience, the hindbrain is particularly sensitive to subcutaneous injection) anesthesia. Brains were carefully dis- erroneous registration when cleared samples are mapped di- sected and immersion-fixed in 10% neutral buffered formalin rectly onto the AIBS CCFv3. High quality registration can be overnight at room temperature on a horizontal shaker. Whole- achieved using a multi-regional approach where each larger brain samples were washed 3 × 30 min in PBS with shaking part of the brain, e.g. the hindbrain, is registered separately. and dehydrated at room temperature in methanol/H Ogradi- However, this procedure reduces analysis speed as initial seg- ent to 100% methanol (20%, 40%, 60%, 80%, 100% metha- mentation of the larger brain structures is required. We aimed nol, each step 1 h). The brains were stored in 100% methanol to preserve both data flow and quality by generating a refer- (VWR International A/S, Søborg, Denmark) at 4 °C until ence template based on iDISCO+ processed and LSFM- further processing. imaged mouse brains and aligning the AIBS CCFv3 with the template through multi-regional registration. Whole-Brain Immunohistochemistry for Labeling of c- The LSFM atlas enables fast brain-wide inter-modality regis- Fos Positive Cells and Clearing tration of other LSFM samples. To confirm accuracy and demon- strate the utility of the LSFM-based reference brain atlas, we de- The iDISCO+ (immunolabeling-enabled three-dimensional termined the c-Fos expression signature of semaglutide, a long- imaging of solvent-cleared organs) protocol was used for acting glucagon-like peptide-1 (GLP-1) receptor agonist. The whole brain immunolabelling (Renier et al. 2014, 2016). LSFM atlas enabled precise mapping of semaglutide-induced c- Samples were washed with 100% methanol for 1 h and incu- Fos expression in the mouse whole-brain. In addition to c- bated overnight in 66% dichloromethane/33% methanol Fos imaging, application of the atlas includes also map- (VWR International A/S, Søborg, Denmark) at room temper- ping other fluorescent markers imaged by LSFM. ature. Then, samples were washed twice in 100% methanol Neuroinform (2021) 19:433–446 435 for 30 min and bleached in chilled fresh 5% H O (Acros Images were acquired at 0.63 x magnification (1.2 × total 2 2 Organics, Fisher Scientific Biotech Line A/S, Slangerup, magnification) with an exposure time of 254.47 ms in a z- Denmark) in methanol overnight at 4 °C. Subsequently, the stack at 10 μm intervals. Acquired volumes (16-bit tiff) had samples were rehydrated in methanol/PBS series (80%, 60%, an in-plane resolution of 4.8 μm and z-resolution of 3.78 μm 40%, 20% methanol with 0.2% Triton X-100 (Merck, (NA = 0.156). Horizontal focusing was captured in 9 planes Darmstadt, Germany), each step 1 h) at room temperature, with blending mode set to the centre of the image to merge the washed in PBS with 0.2% Triton X-100 twice for 1 h at room individual raw images. Data was acquired in two channels, temperature and in permeabilization solution (PBS with 0.2% autofluorescence and antibody-specific channel, because the Triton X-100, supplemented with 20% volume of DMSO former provides information on tissue structure and the latter (Merck, Darmstadt, Germany) and 2.3% weight/volume gly- on neuronal activity. Autofluorescence volumes were ac- cine (Merck, Darmstact, Germany)) for 3 days at 37 °C. For c- quired at excitation wavelength of 560 ± 20 nm and emission Fos labeling, unspecific antibody binding was blocked in wavelength of 650 ± 25 nm, laser power was set to 80%. blocking solution (PBS, 0.2% TritonX-100, 10% DMSO/6% Fluorescently labelled c-Fos positive cells were captured donkey serum (Jackson ImmunoResearch, Cambridgeshire, in a specific channel at excitation wavelength of 630 ± UK)) for 2 days at 37 °C before incubated in the primary 15 nm and emission wavelength of 680 ± 15 nm, laser antibody buffer (PTwH, 5% DMSO, 3% donkey serum, power was set to 100%. 0.2% of 10% NaN (Merck, Darmstadt, Germany)) for 7 days at 37 °C. For visualization of c-Fos expression, rabbit anti-c- Fos antibody (1:5000, Cell Signaling Technology, Image Processing for Creating the Mouse Brain Atlas Massachusetts, US, cat number #2250) was used. Following incubation with primary antibody, the brains were washed in An average LSFM mouse brain volume was created from PTwH (PBS, 0.2% Tween 20 (Merck, Darmstadt, Germany), 139 individual mouse brain autofluorescence datasets by an 0.1% of 10 mg/ml heparin solution) for 1 × 10 min, 1 × iterative multi-resolution image registration algorithm 20 min, 1 × 30 min, 1 × 1 h, 1× 2 h and 1× 2 days. (Kovačević et al. 2005;Kuanetal. 2015;Umadevi Subsequently, the brains were incubated in secondary anti- Venkataraju et al. 2019). Pre-processing was initiated by body solution (PTwH, 3% donkey serum, 0.2% of 10% down-sampling to 20 μm isotropic resolution. N3 method NaN ) for 7 days at 37 °C with donkey anti rabbit Cy-5 anti- (Larsen et al. 2014; Sled et al. 1998; Van Leemput et al. body (1:1000, Jackson ImmunoResearch, Cambridgeshire, 1999) was applied to correct intensity inhomogeneity. UK, cat no #711–175-152) and washed in PTwH for 1 × Subsequently, the intensity histograms of the individual 10 min, 1 × 20 min, 1 × 30 min, 1 × 1 h, 1× 2 h and 1× 3 days. volumes were normalized and, contrast adaptive histogram For clearing, the brains were dehydrated in a methanol/H2O equalization was performed (Fig. 1a, left). For generating an series (20%, 40%, 60%, 80% and 100% methanol, each step average mouse brain template, a reference volume was ran- 1 h) at room temperature, incubated in 66% dichloromethane/ domly selected as a starting point. Six iterative multi- 33% methanol for 3 h at room temperature with shaking and in resolution registration steps – one affine and five B-spline 100% dichloromethane twice for 15 min with shaking to re- transformations were performed for the remaining samples move traces of methanol. Finally, the samples were trans- (Fig. 1a, middle). In the first step the brains were registered ferred to dibenzyl ether (Merck, Darmstadt, Germany) and to the chosen reference brain and in subsequent steps stored in closed glass vials until imaged with light sheet fluo- aligned to the average of all brains from the previous step. rescence microscope. Due to the limit in scanning depth in the Z-dimension, which is about 6 mm for our LSFM setup, about half a Light Sheet Fluorescence Microscopy of Labeled and millimetre of the dorsal cortex was not imaged. To produce Cleared Mouse Brains a template with full cortex, 15 additional image stacks of cor- tices were acquired, pre-processed and aligned to the average All whole-brain samples were imaged in an axial orientation mouse brain volume. Subsequently, both volumes were on a LaVision ultramicroscope II setup (Miltenyi Biotec, merged. Satisfactory axial symmetry was achieved by divid- Bergisch Gladbach, Germany) equipped with a Zyla 4.2P- ing the template brain volume into three coronal slabs with CL10 sCMOS camera (Andor Technology, Belfast, UK), equal thickness and manually rotating them into correct posi- SuperK EXTREME supercontinuum white-light laser EXR- tion. The final template was created by mirroring one hemi- 15 (NKT photonics, Birkerød, Denmark) and MV PLAPO sphere to the opposite side and merging the hemispheres with 2XC (Olympus, Tokyo, Japan) objective lens. The samples a sigmoidal blending function for receiving a symmetric tem- were attached to the sample holder with neutral silicone and plate brain (Fig. 1a, right) Additionally, a tissue mask and a imaged in a chamber filled with dibenzyl ether. Version 7 of ventricular mask were added to the LSFM template from the the Imspector microscope controller software was used. AIBS CCFv3 and manually adapted to fit the template. 436 Neuroinform (2021) 19:433–446 3D autofluorescence images Iterative normalization and averaging Final template I A A 1 1 2 6 + post-processed 1, pre-processed I A 6 + post-processed AIBS CCFv3 Final LSFM atlas AIBS CCFv3 A + segmentations 6 + post-processed Multi-regional registration 6 + post-processed Fig. 1 LSFM-based mouse brain atlas. a) Generation of a brain where y stands for the iteration step. B) Transfer of brain region template based on the LSFM autofluorescence volumes of 139 mice segmentations from the AIBS CCFv3 to the LSFM mouse brain brains using an iterative registration and averaging algorithm. Raw light template. Brain regions of the AIBS CCFv3 were mapped to the LSFM sheet scans are annotated with I where x stands for the animal number, template in six parts, e.g. cortex to cortex, hindbrain to hindbrain etc. and the intermediate average mouse brain volumes are annotated with A Brain regional annotations were transferred to the LSFM Segmentation refinements were performed with micros- template from the AIBS CCFv3 (Fig. 1b) (Allen Institute for copy image analysis software Imaris™ version 2 Brain Science 2011, 2015, 2017;Kuan etal. 2015;Wang etal. (Oxford instruments, Abington, UK). Image processing 2020). First, the mouse brain template of AIBS was registered was performed in Python and Elastix toolbox (Klein onto the LSFM template using multi-resolution affine and B- et al. 2010; Shamonin et al. 2014) was used to imple- spline registration. Subsequently, the registered AIBS CCFv3 ment the registrations. Detailed description of the atlas template and its segmentations were divided into six parental creation procedure and full sets of parameters can be brain regions – cerebral cortex, cerebral nuclei, hindbrain, found in the Online Resource 1. cerebellum, septal regions and interbrain together with midbrain. The parental regions were then separately reg- Quantification of c-Fos Positive Cells istered to the corresponding areas of the LSFM tem- plate. Manual corrections were performed for regions Neuronal activity was quantified by detecting and counting c- near to ventricular system, such as AP and SFO. Fos positive cells using an adapted ClearMap routine (Renier Neuroinform (2021) 19:433–446 437 et al. 2016). In brief, the volume pairs collected from the generalized linear model provided a suitable fit to our c-Fos cell autofluorescence and c-Fos specific channel were aligned count data. For each generalized linear model, a Dunnett’stest slice-by-slice using affine registration in 2D with mattes mu- was performed. Statistical analysis for determining differences in tual information as a similarity measure and background c-Fos expression between semaglutide and vehicle treated mice subtracted through morphological opening using a disk ele- involved p value adjustment using a multiple comparison meth- ment. For removing false positive c-Fos signal originating od called false discovery rate. Statistical analysis of the data was from increased tissue autofluorescence, a signal appearing performed using R statistics library. both in the autofluorescence and the c-Fos specific channel Further, all significantly regulated brain regions underwent was removed from the specific channel. For identifying c-Fos a two-step manual validation procedure for checking if the positive cells, local intensity peaks were monitored by moving used statistical model fits the data points, the significance of a filter cube over the specific channel volume followed by the brain regions is not achieved due to outliers and the raw seeded watershed for segmenting the c-Fos positive cells. signal is truly originating from the region. First, the fit of cell The initial parameters were taken from the original counts to the generalized linear model was evaluated. This ClearMap implementation (Renier et al. 2016) but optimized was done by investigating deviance residuals and checking to fit our data, being acquired under different conditions, in- if the residuals aligned with the assumptions of normality cluding image resolution. The size of the filter cube was set to and homoscedasticity. Furthermore, Cook’s distance was cal- 5x5x3 pixels for effectively detecting all possible c-Fos posi- culated for each cell count data point in the model as a mea- tive cell candidates. The third dimension of the filter cube was sure of model influence. Regions where the generalized linear chosen to be smaller than the first and second dimension of the model showed severe violations of the assumptions, or the cube since z-resolution of the LSFM volumes was lower than model contained overly influential data points, were the in-plane resolution. The coordinates of the detected local discarded. Secondly, the remaining brain regions were visual- intensity peaks were used as seeds in watershed segmentation ly studied for possible spillover signal from neighboring re- with a background intensity cut-off of 800 and the resulting gions. If the c-Fos response in a region seemed to originate segmentations were filtered by removing cell segmentation from the neighboring region, e.g. very few c-Fos positive cells regions smaller than 8 voxels and bigger than 194 voxels. were observed only in the border areas of the region while the Following c-Fos positive cell detection in the specific channel, neighboring areas were exhibiting very high signal, it was the corresponding autofluorescence volumes underwent bias declared as not significant. field correction and contrast limited adaptive histogram equal- ization (similar procedure as for the LSFM mouse brain tem- plate creation). For quantifying the number of c-Fos positive Results cells in individual brain regions, the LSFM atlas was aligned to c-Fos specific channel volumes of individual mice over pre- LSFM Reference Atlas of the Adult Mouse Brains processed autofluorescence volumes and the number of c-Fos positive was counted in every brain region. Heatmaps visual- The standard way of aligning a LSFM scanned mouse brain izing the density of the c-Fos positive cells were created by with the AIBS CCFv3 is to perform a single cross-modality mapping the specific channel volumes to the LSFM atlas registration of the full brains by computing a global affine and using the inverse transform, generating and summing the local B-spline transformation in a one-to-one manner spheres of uniform value and 20 μm radius around the centers (Fig. 2a). However, an alternative strategy is to perform mul- of the c-Fos positive cells (Renier et al. 2016). Image tiple registrations, where each of the major brain structures is processing and analysis was performed in Python. 3D aligned individually (Fig. 2b). By comparing the two ap- visualizations of heatmaps were created with microscopy proaches we observed that multiple registrations yield higher image analysis software Imaris™ version 2 (Oxford in- quality registrations in some parts of the brain, e.g. the area struments, Abington, UK). postrema (Fig. 2c). However, aligning LSFM-imaged brains using multiple registrations is time-consuming and require Statistics both initial segmentation of the larger brain structures and manual validation for each brain which is not compatible with For simplicity, 666 individual brain region segmentations of the high-throughput analysis. Our solution to this dilemma was to LSFM atlas were collapsed to their parental regions using the build an LSFM-based reference atlas by aligning the AIBS hierarchy tree of the atlas ontology (Online Resource 2) resulting CCFv3 to the LSFM-based mouse brain template through in 284 regions in which the statistical analysis was performed. multi-regional registrations. The present LSFM-based mouse For determining the difference in the c-Fos positive cell counts, a brain reference atlas can be used to analyze individual LSFM- generalized linear model was fitted to the cell counts observed in imaged samples directly by fast one-to-one registrations or for each brain region in every animal group. A negative binomial improved alignment to the AIBS CCFv3 space if needed (Fig. 438 Neuroinform (2021) 19:433–446 Whole-brain one-to-one Multi-regional Example of one-to-one and ab c registration registration multi-regional registration Cleared LSFM-imaged brain (raw data) Cleared LSFM- Cleared LSFM- AIBS CCFv3 imaged brain imaged brain template AP Cerebral cortex Septal AIBS CCFv3 registered to the regions cleared LSFM-imaged brain Hindbrain Cerebral nuclei AP AP Mid- and interbrain Whole-brain Multi- AIBS CCFv3 one-to-one regional Cerebellum template registration registration Fast Slow Fast inaccurate accurate accurate Whole-brain one- Multi-regional Whole-brain one- to-one registration registration to-one registration (performed once) Cleared LSFM- Cleared LSFM- LSFM AIBS CCFv3 imaged brain imaged brain template Fig. 2 Techniques for registering LSFM-imaged samples with the aligning cleared LSFM-imaged samples with the AIBS CCFv3 template AIBS CCFv3. a) Illustration of one-to-one registration between a cleared provides better accuracy but is relative slow compared to the one-to-one LSFM-imaged sample and the AIBS CCFv3 template. b) Illustration of registration. By generating a template from cleared LSFM-imaged brains multi-regional registration between a cleared LSFM-imaged sample and and registering the AIBS CCFv3 with it once using multi-regional regis- the AIBS CCFv3 template, where the brain volumes have been divided tration approach we ensure good alignment accuracy between the two into six larger brain areas that are mapped individually. c) Example of the templates. Subsequent registrations of cleared LSFM-imaged brains with registration quality in area postrema (AP) using either one-to-one or the LSFM template can then be done directly using fast one-to-one reg- multi-regional registration. d) Illustration of the registration flow de- istrations. This way it is possible to achieve both fast as well as accurate scribed in this manuscript. Using one-to-one registration for aligning registration of cleared LSFM-imaged brains. Regardless of computer per- cleared LSFM-imaged samples with the AIBS CCFv3 is fast but inaccu- formance the speed of analysis improved by a factor of six compared to rate in some brain regions like the AP. Multi-regional registration for the multiregional registration 2d). Regardless of computer performance we found that direct viewed from the coronal and horizontal orientation. The axial alignment to the LSFM atlas improved the registration speed resolution of the mouse brain template is 20 μm. Brain region for each brain sample volume by a factor of six. annotations for the LSFM template were imported from the An LSFM-based mouse brain reference atlas containing an AIBS CCFv3 by image registration (Fig. 1b). The annotations average anatomy template with corresponding brain region were imported as six separate pieces with manual corrections annotations was created. The mouse brain template was gen- to mitigate the challenge of cross-modality registration. The erated from 139 3D autofluorescence-scanned brain volumes final atlas contains 666 brain region segmentations with by an iterative multi-resolution image registration algorithm anatomical nomenclature corresponding to the AIBS (Fig. 1a). Post-processing of the template involved refinement CCFv3 (hierarchy tree of the atlas ontology in of the axial symmetry to obtain a midline symmetric atlas Online Resource 2) (Dong 2008). Neuroinform (2021) 19:433–446 439 Improved Registration of LSFM-Imaged Mouse Brains the AIBS CCFv3 templates, as well as in the same ten indi- vidual brain volumes which were previously used for registra- To validate that the LSFM reference atlas improved alignment tion evaluation (Fig. 3c; an atlas template containing the 27 of LSFM-acquired brain volumes, we tested alignment of ten landmarks together with the intensity variance map is raw LSFM-imaged mouse brain volumes and compared the available at GitHub and the atlas coordinates for each results to alignment with the AIBS CCFv3 using identical landmark can be found in the Online Resource 5). For the registration procedures. By computing the amount of defor- placement of each landmark several factors were considered. mation needed to register each brain into the two atlases, we The landmarks should be: 1) easily recognizable in both the evaluated the voxel-wise magnitude of displacement neces- AIBS CCFv3 and LSFM templates; 2) distributed brain-wide sary to convert the individual brain volumes to either of the such that several landmarks were located in cerebral cortex, atlas template (Fig. 3a). As expected, the LSFM-imaged brain cerebral nuclei, interbrain, midbrain, hindbrain and cerebel- volumes are less deformed when aligned with the generated lum; 3) distributed along the midline as well as in more lateral LSFM atlas compared to alignment with the AIBS CCFv3. parts of the brain; 4) placed in regions with increased local We found deformations ranging up to 13 voxels with the intensity variance, if possible (Fig. 3b). Following the regis- AIBS CCFv3 compared to deformations ranging up to 8 tration of the individual brains to the LSFM atlas and voxels with the LSFM atlas. Furthermore, the volume of the the AIBS CCFv3, the Euclidean distance between the regis- area affected by the deformation is smaller for the brains tered and atlas landmarks was calculated. Although this ap- aligned to the LSFM atlas compared to the brains aligned to proach also reflects the inherent variation that occurs when the AIBS CCFv3. The results show that deformations are placing landmarks, it consistently showed more accurate reg- most pronounced in the midbrain and hindbrain (Fig. 3a) istration when the LSFM atlas was used as a template. and most likely the reflect they morphological changes inflicted by tissue processing and clearing. Accurate c-Fos Quantification in LSFM-Imaged Mouse As the magnitude of the deformation is only an indicative Brains measure by which the registration quality cannot be fully assessed, we further investigated the alignment quality using For evaluating the performance of the LSFM atlas to assign c- a standardized metric called intensity variance developed by Fos positive cells to anatomical brain regions, we conducted a the Non-Rigid Image Registration Evaluation Project separate experiment where we mapped the brains from (NIREP) (Christensen et al. 2006). Intensity variance quan- semaglutide-dosed lean mice onto the LSFM and AIBS tifies how much the signal intensity differs per voxel between CCFv3 atlas, respectively, and compared the distribution the set of registered brain volumes and hence, estimates the and number of c-Fos positive cells counted using each atlas amount of noise in the data set. We therefore computed the (Fig. 4a). The two atlases showed highly overlapping results intensity variance for all brain regions using ten LSFM- in the majority of brain regions. However, 11 regions showed imaged mouse brains registered to both the LSFM atlas and significant differences in the number of c-Fos positive cells the AIBS CCFv3 (Fig. 3b). The mean intensity variance de- when comparing data analyzed with the two atlases (Fig. 4b- termined for the six major brain volumes registered to the c). Hence, to determine how registration accuracy impacts the AIBS CCFv3 was 17.42 for cerebral cortex, 19.09 for cerebral localization of c-Fos positive cells, we compared c-Fos signa- nuclei, 21.77 for interbrain, 32.38 midbrain, 49.34 for cere- tures in the hindbrain regions, i.e. the nucleus of the solitary bellum and 53.90 for hindbrain. In contrast, the mean intensity tract (NTS) and the dorsal motor nucleus of the vagus nerve variance computed for volumes registered to the LSFM atlas (DMX). According to the LSFM atlas, most c-Fos positive was 19.00 for cerebral cortex, 16.01 for cerebral nuclei, 18.64 cells were localized to the NTS (234 ± 38 cells) compared to for interbrain, 24.13 for midbrain, 44.19 for cerebellum, 30.31 the DMX (144 ± 14 cells) (Fig. 4d). In contrast, the AIBS for hindbrain. To analyse these findings in more detail, the CCFv3 revealed an opposite pattern (NTS, 95 ± 16 cells; intensity variance for all sub-regions within the six major DMX, 205 ± 25 cells) (Fig. 4e). To clarify which atlas is more brain regions, were plotted in scatter plots with AIBS accurate in the signal localization, we compared the raw mi- CCFv3 values on the y-axis and LSFM atlas values on the croscope images to heatmaps representing c-Fos signal densi- x-axis. As for the deformation (Fig. 3a), the most substantial ty using either atlas (Fig. 4f). The autofluorescence intensity differences in intensity variance were observed in the mid- of NTS is brighter than the intensity of surrounding tissue brain and hindbrain. The improvement of the registration ac- making it easy to delineate and shows that the raw c-Fos signal curacy using the LSFM atlas was particularly notable for hind- is indeed localized in the NTS, thus validating the LSFM atlas brain due to significantly lower intensity variance for majority mapping. Signal localization accuracy of the LSFM atlas was of the sub-regions when LSFM atlas was used for registration. also assessed for the other nine brain regions with conflicting To further compare the registration quality between the two c-Fos data (data not shown). While improved c-Fos signal atlases, 27 landmarks were identified in both the LSFM and localization by the LSFM atlas was confirmed for additional 440 Neuroinform (2021) 19:433–446 Neuroinform (2021) 19:433–446 441 Fig. 3 Improved registration of LSFM-acquired brain volumes using autofluorescence channel was applied for correction in the LSFM atlas. a) Heatmaps illustrate the average magnitude of the whole-brain mounts in both lean and DIO mice deformation resulting from the registration of ten random raw LSFM (Supplementary Fig. 1), resulting in significantly improved brain volumes to the AIBS CCFv3 and to the LSFM atlas. b) Registration using the LSFM mouse brain atlas enables improved signal-to-noise ratio specifically in DIO mice alignment between individual brains. Intensity variance, a measure for (Supplementary Fig. 2). To identify the differences between registration performance, was calculated per brain region for the ten the semaglutide and the vehicle dosed mice, average signal random brain volumes aligned to the LSFM atlas and for the same ten heatmaps in semaglutide-treated lean and obese mice were brain volumes aligned to the AIBS CCFv3. Highest intensity variance was detected in both cases in ventricular and hindbrain regions (example subtracted voxel-wise from the corresponding vehicle control sections, left). Statistical analysis of the intensity variance was performed group (Fig. 5a, c) with statistical analyses on the raw c-Fos using two-tailed Welch’s t-test and the resulting significant regions are positive cell counts (Fig. 5b, d). Compared to vehicle controls, visualized in the scatter plot (right) along with the mean intensity variance 9 brain regions were significantly regulated by semaglutide per major brain region for both atlases (denoted as mean IV). The results indicate that the difference in intensity variance values was small for treatment in both lean and DIO mice. Semaglutide treated lean cortical areas. However, majority of brain regions in cerebral nuclei, and DIO mice showed similar increased c-Fos expression in interbrain, midbrain, cerebellum and hindbrain exhibited higher intensity the bed nuclei of the stria terminalis (BST), paraventricular variance when the AIBS CCFv3 was used for registration compared to nucleus of the thalamus (PVT), xiphoid thalamic nucleus when the LSFM atlas was used for registration. c) Registration of the ten brain volumes was further evaluated using 27 landmarks distributed over (Xi), central amygdalar nucleus (CEA), parabrachial nucleus the whole brain (overview of the landmark positions, left). The landmarks (PB), nucleus of the solitary tract (NTS) and dorsal motor were divided between the six major brain areas in both atlases as well as nucleus of the vagus nerve (DMX) compared to the vehicle in the ten brain volumes. Distances between the corresponding landmarks treated controls. Additionally, semaglutide treated DIO mice in the individual brains and the atlas templates were calculated after registering the ten brain volumes to the LSFM atlas and the AIBS exhibited increased c-Fos expression in the parataenial nucle- CCFv3 (bar plot, right). For most landmarks, the calculated distances us (PT) and parasubthalamic nucleus (PSTN), whereas are lower when the LSFM atlas is used as template. Significant differ- semaglutide treated lean mice showed increased c-Fos ex- ences in distances between the two atlases was consistently observed in pression in the pedunculopontine nucleus (PPN) and cerebral cortex and hindbrain. Two-tailed Welch’st-test was appliedfor determining statistical significance in landmark distances between the mediodorsal nucleus of thalamus (MD) compared to the re- atlases: ∗ for 0.01 ≤ p <0.05, ∗∗ for 0.001 ≤ p <0.01 and ∗∗∗ for spective vehicle treated controls. p <0.001 Discussion five regions (hypoglossal nucleus (XII), presubiculum (PRE), nodulus (NOD), nucleus of the optic tract (NOT) and We present here the generation of an LSFM-based mouse postsubiculum (POST)). The AIBS CCFv3 performed better brain atlas. Compared to the AIBS CCFv3 (Allen Institute in one region, flocculus (FL), while three regions (lateral part for Brain Science 2011, 2015, 2017;Kuan et al. 2015), the of the central amygdalar nucleus (CEAl), parabrachial nucleus LSFM reference mouse brain atlas provides more accurate (PB) and pedunculopontine nucleus (PPN)), could not be anatomical segmentation and quantitative detection of properly evaluated because the ground truth could not be iden- immunolabelled markers in iDISCO+ processed and LSFM- tified due to dispersed c-Fos signal. imaged mouse brains, exemplified by characterization of whole-brain c-Fos responses to semaglutide treatment in both C-Fos Detection in a Pharmacological Study lean and DIO mice. To create an atlas template that is fully representative for To exemplify the use of the LSFM atlas we performed a study average brain anatomy we developed the LSFM atlas based on with the aim of quantifying c-Fos expression in mice dosed the variational atlas algorithm previously described with the GLP-1 receptor agonist semaglutide. Semaglutide (Kovačević et al. 2005;Kuanetal. 2015; Umadevi and vehicle was administered peripherally to lean and DIO Venkataraju et al. 2019). This algorithm avoids bias of the mice, and the c-Fos expression was evaluated 4 h post- template towards the shape of a single chosen reference brain dosing (Fig. 5). When examining the raw LSFM volumes of and accounts for morphological differences between the indi- DIO mice we observed increased autofluorescence in both the vidual brains. specific and the autofluorescence channel, which could poten- In comparison with the AIBS CCFv3, the created LSFM tially lead to false positive c-Fos signals (Supplementary brain template resulted in registrations with lower amount of Figs. 1 and 2). Increased autofluorescence in DIO mice was deformation, and the intensity variance as well as landmark present throughout the brain, but strongest in the cerebellum distances showed improved alignment for LSFM-imaged (Supplementary Fig. 2). Since the increased tissue fluores- samples. This is particularly relevant for tissue samples im- cence was apparent in both channels, but true positive c-Fos aged with LSFM since the samples have been cleared and/or signal was only present in the specific channel, the immunolabelled prior to scanning which affect brain 442 Neuroinform (2021) 19:433–446 b c Average regional cell count Top 20 regions Regions with significant differences Higher in AIBS CCFv3 AIBS CCFv3 LSFM atlas Higher in LSFM atlas Significant regions d e LSFM atlas AIBS CCFv3 NTS NTS DMX DMX 200 200 NTS DMX NTS DMX Raw data LSFM atlas AIBS CCFv3 NTS NTS NTS c-Fos c-Fos c-Fos signal cells signal No. c-Fos positive cells No. c-Fos positive cells Neuroinform (2021) 19:433–446 443 Fig. 4 Choice of brain atlas influences the number of c-Fos positive Our results demonstrate that c-Fos signal distribution in cells per brain region. Comparison of number of c-Fos positive cells in hindbrain regions is less accurately mapped using the AIBS response to semaglutide treatment using the LSFM atlas and the AIBS CCFv3 compared to delineation of signals using our LSFM CCFv3. a) Average number of detected c-Fos expressing cells in every brain region after registration to either the AIBS CCFv3 or the LSFM reference brain atlas. The large difference may be explained atlas. Regions in which the c-Fos positive cells are differentially quanti- by the highamountoflipid-richmyelinfibersinthispartof fied are highlighted by a circle surrounding the data points. An average the brain (Smith 1973). As solvent-based tissue clearing cell count per group below ten is considered too low to judge. b) The bar removes lipids, this could explain the difficulty of mapping chart lists the brain regions and the corresponding mean log fold changes of quantified c-Fos positive cells in these regions according to the p value. certain brain volumes to the AIBS CCFv3 which is based on Blue = higher with LSFM atlas, red = higher with CCFv3. NS stands for non-cleared tissue samples. In addition to the NTS and DMX, not significant, ∗ for 0.01 ≤ p < 0.05, ∗∗ for 0.001 ≤ p < 0.01 and ∗∗∗ for we found improved signal localization using the LSFM atlas p < 0.001. c) Horizontally and sagittally depicted brain volumes highlight in five other brain areas. In four of these areas the improve- the regions in which the c-Fos cells were differentially quantified while using the LSFM atlas and the AIBS CCFv3 for the analysis (same colour ment could be assigned to the detailed ventricular mask creat- code as in b and c). See Online Resource 2 for full names of the brain ed for the LSFM atlas. Because the AIBS CCFv3 template regions. d-e) Comparison of total number of c-Fos positive cells quanti- depicts a narrower ventricular system compared to the LSFM fied in 3D-volumes of the nucleus of the solitary tract (NTS) and the atlas template, this may have resulted in incorrectly assigned dorsal motor nucleus of the vagus nerve (DMX) using the LSFM atlas and the AIBS CCFv3. DMX (blue) and NTS (grey) volumes of both c-Fos signal from the choroid plexus to nearby brain regions. atlases in which the signal (glow colormap) was quantified is visualized In the FL, the AIBS CCFv3 performed better than the LSFM in 3D renderings. d) Quantification of c-Fos positive cells following reg- atlas. However, as the FL is often damaged or dislocated dur- istration of the LSFM atlas to the LSFM-acquired brain volumes showed ing dissection of the brain this may impact the subsequent that in average 234 ± 38 c-Fos positive cells were found in the NTS and 144 ± 14 in the DMX. e) Quantification of c-Fos positive cells following mapping. In three brain regions we detected a significant dif- registration of the AIBS CCFv3 to the LSFM-acquired brain volumes. ference in the mapping, but we were unable to determine Here the majority of the signal is found in the DMX. Quantification which of the two atlases performed best because the c-Fos revealed that on average 95 ± 16 c-Fos positive cells are counted in the signal was too scattered. NTS and 205 ± 25 in the DMX. f) Comparing the raw data to the data in alignment with atlases. DMX has a dense dark appearance compared to In terms of c-Fos detection, DIO mice exhibited relatively NTS high unspecific background signals as compared to lean con- trols, most likely attributed to lipid-associated autofluores- cence. Lipid-containing residues of lysosomal digestion, lipofuscins, have also been reported to increase during aging morphology by shrinkage/expansion and de-lipidation (Kim and oxidative stress (Boellaard et al. 2004) leading to in- et al. 2018; Wan et al. 2018). Furthermore, the contrast within creased autofluorescence (Cho and Hwang 2011;Di Guardo anatomical structures in the brain that are important for sub- 2015; Schnell et al. 1999). When comparing the c-Fos activity sequent image registration differs between the AIBS CCFv3 maps between lean and DIO mice we found that the response template and brain processed for LSFM. These issues have to semaglutide looked overall similar in both phenotypes with also been recognized by other researchers and a need for a significant c-Fos activation in BST, PVT, Xi, CEA, PB, NTS dedicated atlas for cleared LSFM-imaged brains has previous- and DMX. Semaglutide is a glucagon-like peptide-1 (GLP-1) ly been highlighted (Umadevi Venkataraju et al. 2019). analogue which has been shown to activate GLP-1 receptors In this study annotations from the AIBS CCFv3 were in the hypothalamus and brainstem (Secher et al. 2014)and mapped to the LSFM atlas template (Wang et al. 2020). markedly stimulates c-Fos expression in mice (Kjaergaard However, as annotation volumes are continuously refined et al. 2019; Salinas et al. 2018). The observed c-Fos expres- (Chon et al. 2019), these can also be aligned to the LSFM sion pattern observed in this study fits well with these previous template. The process of mapping annotations from an reports. Only slight differences were seen between lean and existing atlas to the LSFM-template depends on cross- DIO as exemplified by only DIO mice showed significantly subject cross-modality registration (i.e. different brain, differ- upregulated c-Fos expression in the PT and PSTN. It should ent microscope) which is difficult and often requires manual be noted that lean mice demonstrated a similar c-Fos expres- corrections. With the respect to mapping annotations from the sion pattern in these regions which, however, did not attain AIBS CCFv3 to the LSFM-template, the main difficulty was statistical significance. related to morphology differences in the hindbrain and ven- In this study a c-Fos was detected using a Cy5 labelled tricular system. This was solved by stepwise mapping of the secondary antibody. Consequently we used the 560 nm to annotations for larger parts of the brain such as the hindbrain, record the autofluorescence which is different from the map- together with manual corrections around the ventricular sys- ping reported in the original ClearMap protocol (Renier et al. tem. Now complete, the LSFM atlas provides the benefit of 2016). However, since the choice of fluorophores might vary improved registration of other LSFM-samples together with from study to study, we tested how the choice of autofluores- cence impacts the subsequent mapping (atlas-registered detailed brain region annotations. 444 Neuroinform (2021) 19:433–446 a b Vehicle vs semaglutide, lean mice BST CEA PVT PB NTS DMX c d Vehicle vs semaglutide, DIO mice BST CEA PVT PSTN PB NTS DMX Fig. 5 Differentially regulated c-Fos expression in response to vehicle treatment and corresponding mean log fold changes of c-Fos semaglutide administration. Up (red) and down (blue) regulation of c- positive cells in these regions in b) lean and d) DIO mice. ∗ stands for Fos expression in a) semaglutide treated lean mice in comparison to 0.01 ≤ p < 0.05, ∗∗ for 0.001 ≤ p<0.01 and ∗∗∗ for p < 0.001. P-values vehicle treated lean mice and c) in semaglutide treated DIO mice in were adjusted for multiple comparisons using the false discovery rate. See comparison to vehicle treated DIO mice. Differentially regulated brain Online Resource 2 for full names of the brain regions regions in response to semaglutide administration in comparison to autofluorescence volumes can be found in Github). Although, accumulation. In this case it did not impact on the registration, we obtained the best registration using the 560 nm channel to but it will always be important to consider the possibil- record the autofluorescence, channels below 700 nm ity that the choice of model may influence registration worked as well. When reaching the NIR spectrum the and quantification. endogenous fluorescence become so weak it can no lon- In conclusion, we developed a dedicated reference atlas ger be used for registration. allowing faster and more accurate mapping of iDISCO+ proc- The average brain generated in this study was created from essed and LSFM-imaged whole mouse brains. In combination 8 to 10 week old C57Bl/6 J male mice. Since factors such as with an improved c-Fos detection algorithm, our pipeline en- age, sex and strain are known to affect brain size and anatomy, ables for unbiased, automated and computationally efficient it is possible deviations from the average parameters may have quantitative analysis of drug-induced c-Fos expression in the a slight impact on the overall quality of registration and quan- intact mouse brain. The LSFM atlas is highly applicable for tification. Indeed, we observed that obesity led to an unexpect- fast and precise mapping of fluorescent markers in both the ed increase in autofluorescence, presumably due to lipofuscin normal mouse brain and mouse models of CNS diseases as Neuroinform (2021) 19:433–446 445 Boellaard, J. W., Schlote, W., & Hofer, W. (2004). Ultrastructural pathol- well for improved delineation of compound distribution in the ogy species-specific ultrastructure of neuronal Lipofuscin in CNS imaged by LSFM (Liebmann et al. 2016; Roostalu et al. Hippocampus and Neocortex of subhuman mammals and humans. 2019; Salinas et al. 2018;Secher etal. 2014). Ultrastruct Pathol, 28(5–6), 341–351. https://doi.org/10.1080/ Cho, S., & Hwang, E. S. (2011). Fluorescence-based detection and quan- tification of features of cellular senescence. Methods Cell Biol, 103, Information Sharing Statement 149–188. https://doi.org/10.1016/B978-0-12-385493-3.00007-3. Chon, U., Vanselow, D. J., Cheng, K. C., & Kim, Y. (2019). Enhanced and unified anatomical labeling for a common LSFM reference atlas files are freely accessible at https:// mouse brain atlas. Nat Commun, 10, 5067. https://doi.org/ github.com/Gubra-ApS. Quantitative c-Fos data for all brain 10.1038/s41467-019-13057-w. regions is available as Online Resources 3 and 4. Source code Christensen, G. E., Geng, X., Kuhl, J. G., Bruss, J., Grabowski, T. J., Pirwani, I. A., et al. (2006). Introduction to the non-rigid image used for generating the LSFM reference atlas along with the registration evaluation project (NIREP). Lecture Notes in code for detecting and quantifying the number of c-Fos posi- Computer Science (including subseries Lecture Notes in Artificial tive cells in LSFM mouse brain volumes is accessible at Intelligence and Lecture Notes in Bioinformatics), 4057,128–135. https://github.com/Gubra-ApS. https://doi.org/10.1007/11784012_16. Chung, K., Wallace, J., Kim, S.-Y., Kalyanasundaram, S., Andalman, A. Acknowledgements The authors would like to acknowledge Lotte S., Davidson, T. J., Mirzabekov, J. J., Zalocusky, K. A., Mattis, J., Denisin, A. K., Pak, S., Bernstein, H., Ramakrishnan, C., Grosenick, Ankjær Frederiksen and Hanne Duus Laustsen for skillful technical L., Gradinaru, V., & Deisseroth, K. (2013). Structural and molecular assistance. interrogation of intact biological systems. Nature, 497(7449), 332– 337. https://doi.org/10.1038/nature12107. Funding The work was funded by Gubra ApS and Innovation Fund Detrez, J. R., Maurin, H., Van Kolen, K., Willems, R., Colombelli, J., Denmark grant number (8053-00121B). Lechat, B., et al. (2019). Regional vulnerability and spreading of hyperphosphorylated tau in seeded mouse brain. Neurobiol Dis, Compliance with Ethical Standards 127,398–409. https://doi.org/10.1016/j.nbd.2019.03.010. Di Guardo, G. (2015). Lipofuscin, lipofuscin-like pigments and autoflu- orescence. Eur J Histochem, 59(1), 1–2. https://doi.org/10.4081/ejh. Conflict of Interest Johanna Perens, Casper Gravesen Salinas, Jacob 2015.2485. Lercke Skytte, Urmas Roostalu, Franziska Wichern, Pernille Barkholt, Dong, H. . (2008). The Allen reference atlas: A digital color brain atlas of and Jacob Hecksher-Sørensen are currently employed by Gubra ApS. Niels Vrang and Jacob Jelsing are the owners of Gubra ApS. the C57BL/6J male mouse. Wiley. Dragunow, M., & Faull, R. (1989). The use of c-fos as a metabolic marker in neuronal pathway tracing. J Neurosci Methods, 29(3), 261–265. Ethics Approval All animal procedures were conducted in compliance https://doi.org/10.1016/0165-0270(89)90150-7. with internationally accepted principles for the care and use of laboratory Ertürk, A., Becker, K., Jährling, N., Mauch, C. P., Hojer, C. D., Egen, J. animals and were approved by the Danish Animal Experiments G., Hellal, F., Bradke, F., Sheng, M., & Dodt, H. U. (2012). Three- Inspectorate (license #2013-15-2934-00784). dimensional imaging of solvent-cleared organs using 3DISCO. Nat Protoc, 7(11), 1983–1995. https://doi.org/10.1038/nprot.2012.119. Open Access This article is licensed under a Creative Commons Esquerda-Canals, G., Montoliu-Gaya, L., Güell-Bosch, J., & Villegas, S. Attribution 4.0 International License, which permits use, sharing, (2017). Mouse models of Alzheimer’s disease. J Alzheimers Dis, 57, adaptation, distribution and reproduction in any medium or format, as 1171–1183. https://doi.org/10.3233/JAD-170045. long as you give appropriate credit to the original author(s) and the Jensen, C. B., Secher, A., Hecksher-Sørensen, J., Conradsen, K., & source, provide a link to the Creative Commons licence, and indicate if Larsen, R. (2015). Quantification of brain access of exendin-4 in changes were made. The images or other third party material in this article the C57BL mouse model by SPIM fluorescence imaging and the are included in the article's Creative Commons licence, unless indicated Allen mouse brain reference model. In Lecture Notes in Computer otherwise in a credit line to the material. If material is not included in the Science (including subseries Lecture Notes in Artificial Intelligence article's Creative Commons licence and your intended use is not and Lecture Notes in Bioinformatics) https://doi.org/10.1007/978-3- permitted by statutory regulation or exceeds the permitted use, you will 319-19665-7_38. need to obtain permission directly from the copyright holder. To view a Kim, J. H., Jang, M. J., Choi, J., Lee, E., Song, K. D., Cho, J., et al. copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. (2018). Optimizing tissue-clearing conditions based on analysis of the critical factors affecting tissue-clearing procedures. Sci Rep, 8, 12815. https://doi.org/10.1038/s41598-018-31153-7. Kjaergaard, M., Salinas, C. B. G., Rehfeld, J. F., Secher, A., Raun, K., & References Wulff, B. S. (2019). PYY(3-36) and exendin-4 reduce food intake and activate neuronal circuits in a synergistic manner in mice. Allen Institute for Brain Science. (2011). Allen Mouse Brain Atlas. Neuropeptides, 73,89–95. https://doi.org/10.1016/j.npep.2018.11. Available from: http://mouse.brain-map.org. Allen Institute for Brain Science. (2015). Allen mouse common coordi- Klein, S., Staring, M., Murphy, K., Viergever, M. A., & Pluim, J. P. W. nate framework. Technical White Paper. (2010). Elastix: A toolbox for intensity-based medical image regis- Allen Institute for Brain Science. (2017). Allen mouse common coordi- tration. IEEE Trans Med Imaging, 29(1), 196–205. https://doi.org/ nate framework and reference atlas. Technical White Paper. 10.1109/TMI.2009.2035616. Bobela, W., Zheng, L., & Schneider, B. L. (2014). Overview of mouse Kovačević, N., Henderson, J. T., Chan, E., Lifshitz, N., Bishop, J., Evans, models of Parkinson’sdisease. Current protocols in mouse biology, A. C., Henkelman, R. M., & Chen, X. J. (2005). A three- 4(3), 121–139. https://doi.org/10.1002/9780470942390.mo140092. dimensional MRI atlas of the mouse brain with estimates of the 446 Neuroinform (2021) 19:433–446 average and variability. Cereb Cortex, 15(5), 639–645. https://doi. Schnell, S. A., Staines, W. A., & Wessendorf, M. W. (1999). Reduction org/10.1093/cercor/bhh165. of Lipofuscin-like autofluorescence in fluorescently labeled tissue. J Kuan, L., Li, Y., Lau, C., Feng, D., Bernard, A., Sunkin, S. M., Zeng, H., Histochem Cytochem, 47(6), 719–730. https://doi.org/10.1177/ Dang, C., Hawrylycz, M., & Ng, L. (2015). Neuroinformatics of the 002215549904700601. Allen mouse brain connectivity atlas. Methods, 73,4–17. https://doi. Secher, A., Jelsing, J., Baquero, A. F., Hecksher-Sørensen, J., Cowley, org/10.1016/j.ymeth.2014.12.013. M. A., Dalbøge, L. S., Hansen, G., Grove, K. L., Pyke, C., Raun, K., Larsen, C. T., Iglesias, J. E., & Van Leemput, K. (2014). N3 Bias Field Schäffer, L., Tang-Christensen, M., Verma, S., Witgen, B. M., correction explained as a Bayesian modeling method. Bayesian and Vrang, N., & Bjerre Knudsen, L. (2014). The arcuate nucleus me- grAphical Models for Biomedical Imaging. Lect Notes Comput Sci, diates GLP-1 receptor agonist liraglutide-dependent weight loss. J 8677,1–12. https://doi.org/10.1007/978-3-319-12289-2. Clin Investig, 124,4473–4488. https://doi.org/10.1172/JCI75276. Leung, C., & Jia, Z. (2016). Mouse genetic models of human brain dis- Shamonin, D. P., Bron, E. E., Lelieveldt, B. P. F., Smits, M., Klein, S., & orders. Front Genet, 7, 40. https://doi.org/10.3389/fgene.2016. Staring, M. (2014). Fast parallel image registration on CPU and GPU for diagnostic classification of Alzheimer’sdisease. Liebmann, T., Renier, N., Bettayeb, K., Greengard, P., Tessier-Lavigne, Frontiers in Neuroinformatics, 7,50. https://doi.org/10.3389/fninf. M., & Flajolet, M. (2016). Three-dimensional study of Alzheimer’s 2013.00050. disease hallmarks using the iDISCO clearing method. Cell Rep, Sled, J. G., Zijdenbos, A. P., & Evans, A. C. (1998). A nonparametric 16(4), 1138–1152. https://doi.org/10.1016/j.celrep.2016.06.060. method for automatic correction of intensity nonuniformity in mri Nectow, A. R., Schneeberger, M., Zhang, H., Field, B. C., Renier, N., data. IEEE Trans Med Imaging, 17(1), 87–97. https://doi.org/10. Azevedo, E., Patel, B., Liang, Y., Mitra, S., Tessier-Lavigne, M., 1109/42.668698. Han, M. H., & Friedman, J. M. (2017). Identification of a brainstem Smith, M. E. (1973). A regional survey of myelin development: Some circuit controlling feeding. Cell., 170,429–442.e11. https://doi.org/ compositional and metabolic aspects. J Lipid Res, 14(5), 541–551. 10.1016/j.cell.2017.06.045. Umadevi Venkataraju, K. U., Gornet, J., Murugaiyan, G., Wu, Z., & Renier, N., Adams, E. L., Kirst, C., Dulac, C., Osten, P., & Tessier- Osten, P. (2019). Development of brain templates for whole brain Lavigne, M. (2016). Mapping of brain activity by automated volume atlases. Progress in Biomedical Optics and Imaging - Proceedings analysis of immediate early genes. Cell, 165(7), 1789–1802. https:// of SPIE, 10865, 1086511. https://doi.org/10.1117/12.2505295. doi.org/10.1016/j.cell.2016.05.007. Van Leemput, K., Maes, F., Vandermeulen, D., & Suetens, P. (1999). Renier, N., Wu, Z., Simon, D. J., Yang, J., Ariel, P., & Tessier-Lavigne, Automated model-based Bias Field correction of MR images of the M. (2014). iDISCO: A simple, rapid method to immunolabel large brain. IEEE Trans Med Imaging, 18(10), 885–896. https://doi.org/ tissue samples for volume imaging. Cell, 159(4), 896–910. https:// 10.1109/42.811268. doi.org/10.1016/j.cell.2014.10.010. Vigouroux, R. J., Belle, M., & Chédotal, A. (2017). Neuroscience in the Rocha, M. D., Düring, D. N., Bethge, P., Voigt, F. F., Hildebrand, S., third dimension: Shedding new light on the brain with tissue clear- Helmchen, F., Pfeifer, A., Hahnloser, R. H. R., & Gahr, M. (2019). ing. Molecular brain, 10(1), 33. https://doi.org/10.1186/s13041- Tissue clearing and light sheet microscopy: Imaging the 017-0314-y. Unsectioned adult Zebra finch brain at cellular resolution. Front Villares, R., Gutiérrez, J., Fütterer, A., Trachana, V., Del Burgo, F. G., & Neuroanat, 13,1–7. https://doi.org/10.3389/fnana.2019.00013. Martínez-A, C. (2015). Dido mutations trigger perinatal death and Roostalu, U., Salinas, C. B. G., Thorbek, D. D., Skytte, J. L., Fabricius, generate brain abnormalities and behavioral alterations in surviving K., Barkholt, P., John L. M., Jurtz V. I., Knudsen L. B., Jelsing J., adult mice. Proc Natl Acad Sci U S A, 112,4803–4808. https://doi. Vrang N., Hansen H. H., Hecksher-Sørensen J. (2019). Quantitative org/10.1073/pnas.1419300112. whole-brain 3D imaging of tyrosine hydroxylase-labeled neuron Wan, P., Zhu, J., Xu, J., Li, Y., Yu, T., & Zhu, D. (2018). Evaluation of architecture in the mouse MPTP model of Parkinson’s disease. seven optical clearing methods in mouse brain. Neurophotonics., 5. DMM Disease Models and Mechanisms https://doi.org/10.1242/ https://doi.org/10.1117/1.nph.5.3.035007. dmm.042200, 12, dmm042200. Wang, Q., Ding, S. L., Li, Y., Royall J., Feng D., Lesnar P., Graddis N., Salinas, C. B. G., Lu, T. T. H., Gabery, S., Marstal, K., Alanentalo, T., Naeemi M., Facer B., Ho A., Dolbeare T., Blanchard B., Dee N., Mercer, A. J., Cornea, A., Conradsen, K., Hecksher-Sørensen, J., Wakeman W., Hirokawa K. E., Szafer A., Sunkin S. M., Oh S. W., Dahl, A. B., Knudsen, L. B., & Secher, A. (2018). Integrated brain Bernard A., Phillips J. W., Hawrylycz M., Koch C., Zeng H., Harris atlas for unbiased mapping of nervous system effects following J. A., & Ng, L. (2020). The Allen mouse brain common coordinate Liraglutide treatment. Sci Rep, 8(1), 10310. https://doi.org/10. framework: A 3D reference atlas. Cell 181,(4), 936–953.e20. https:// 1038/s41598-018-28496-6. doi.org/10.1016/j.cell.2020.04.007. Schneeberger, M., Parolari, L., Das Banerjee, T., Renier, N., Friedman, J. M., & Nectow, A. R. (2019). Regulation of energy expenditure by brainstem GABA neurons in brief. Cell, 178(3), 672–685.e12. Publisher’sNote Springer Nature remains neutral with regard to juris- https://doi.org/10.1016/j.cell.2019.05.048. dictional claims in published maps and institutional affiliations. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Neuroinformatics Springer Journals

An Optimized Mouse Brain Atlas for Automated Mapping and Quantification of Neuronal Activity Using iDISCO+ and Light Sheet Fluorescence Microscopy

Loading next page...
 
/lp/springer-journals/an-optimized-mouse-brain-atlas-for-automated-mapping-and-jnSPv1UIKx
Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2020
ISSN
1539-2791
eISSN
1559-0089
DOI
10.1007/s12021-020-09490-8
Publisher site
See Article on Publisher Site

Abstract

In recent years, the combination of whole-brain immunolabelling, light sheet fluorescence microscopy (LSFM) and subsequent registration of data with a common reference atlas, has enabled 3D visualization and quantification of fluorescent markers or tracers in the adult mouse brain. Today, the common coordinate framework version 3 developed by the Allen’s Institute of Brain Science (AIBS CCFv3), is widely used as the standard brain atlas for registration of LSFM data. However, the AIBS CCFv3 is based on histological processing and imaging modalities different from those used for LSFM imaging and consequently, the data differ in both tissue contrast and morphology. To improve the accuracy and speed by which LSFM-imaged whole-brain data can be registered and quantified, we have created an optimized digital mouse brain atlas based on immunolabelled and solvent-cleared brains. Compared to the AIBS CCFv3 atlas, our atlas resulted in faster and more accurate mapping of neuronal activity as measured by c-Fos expression, especially in the hindbrain. We further demonstrated utility of the LSFM atlas by comparing whole-brain quantitative changes in c-Fos expression following acute administration of semaglutide in lean and diet-induced obese mice. In combination with an improved algorithm for c-Fos detection, the LSFM atlas enables unbiased and computationally efficient characterization of drug effects on whole- brain neuronal activity patterns. In conclusion, we established an optimized reference atlas for more precise mapping of fluorescent markers, including c-Fos, in mouse brains processed for LSFM. . . . . . Keywords Light sheet fluorescence microscopy iDISCO Tissue clearing Brain atlas C-Fos Whole brain imaging Introduction expression patterns of c-Fos, a proxy for neuronal acti- vation (Dragunow and Faull 1989). Rodent models are important tools in preclinical drug devel- Recent advances in immunohistochemical methods and opment for central nervous system (CNS) disorders (Bobela optical clearing techniques have, together with ex vivo imag- et al. 2014; Esquerda-Canals et al. 2017; Leung and Jia ing technologies such as light sheet fluorescence microscopy 2016). A common method for characterizing central ef- (LSFM), enabled whole-organ imaging (Chung et al. 2013; fects of potential novel therapies is to quantify Ertürk et al. 2012;Jensen et al. 2015; Kjaergaard et al. 2019; Renier et al. 2014; Rocha et al. 2019; Secher et al. 2014). As a Electronic supplementary material The online version of this article result, it is now possible to visualize c-Fos expression at the (https://doi.org/10.1007/s12021-020-09490-8) contains supplementary single cell level in the intact adult mouse brain (Kjaergaard material, which is available to authorized users. et al. 2019;Nectowetal. 2017;Renier et al. 2016). In recent years, automated image analysis algorithms have * Jacob Hecksher-Sørensen been developed enabling 3D quantification of activated neu- jhs@gubra.dk rons and their signal intensities in the adult mouse brain (Detrez et al. 2019;Jensenetal. 2015;Liebmannetal. Gubra ApS, 2970 Hørsholm, Denmark 2016;Nectow et al. 2017; Salinas et al. 2018;Schneeberger Department of Applied Mathematics and Computer Science, et al. 2019). The first step of the analysis process is to register Technical University Denmark, 2800 Kongens Lyngby, Denmark LSFM imaging data onto a common reference brain which Danish Research Centre for Magnetic Resonance, Centre for contains annotated brain regions. Today, the most widely used Functional and Diagnostic Imaging and Research, Copenhagen University Hospital Hvidovre, 2650 Hvidovre, Denmark mouse brain atlas is the common coordinate framework 434 Neuroinform (2021) 19:433–446 version 3 (CCFv3), developed by the Allen Institute for Brain Materials and Methods Science (AIBS) (Allen Institute for Brain Science 2011, 2015, 2017;Kuan et al. 2015;Wang et al. 2020). For quantification Animals of fluorescent signals, registration is followed by cell detec- tion, e.g. ClearMap, to segment and count c-Fos positive cells Male C57Bl/6 J mice were obtained from Janvier Labs (Le (Nectow et al. 2017; Renier et al. 2016) or extract voxel in- Genest-Saint-Isle, France), and were maintained in standard tensities (Salinas et al. 2018). Finally, the results can be housing conditions (12 h light/dark cycle and controlled tem- assigned to specific regions using the anatomical reference perature of 21–23 °C). Mice had ad libitum access to tap water atlases such as those provided by AIBS. and regular chow (Altromin 1324, Brogaarden, Hørsholm, LSFM image processing pipelines have improved quanti- Denmark) or high fat diet (60% fat, 21% carbohydrates, tative whole-brain 3D imaging. However, the quality of the 19% protein; Ssniff Spezialdiäten GmbH, Soest, Germany). LSFM results is highly dependent on sample processing and The LSFM atlas was established based on analysis of 139 the imaging methods applied. Whole-organ immunolabelling brains from 8 to 10 weeks old male chow-fed mice. The requires lipid extraction to make the tissue permeable to anti- pharmacology-induced neuronal activity study involved two bodies (Kim et al. 2018) and enable deep tissue imaging groups of lean mice and two groups of DIO mice. All groups (Vigouroux et al. 2017). In particular, myelin fibers which were aged matched (38 weeks) and consisted of n = 6. Lean are lipid-rich (Villares et al. 2015), are more likely to be af- and DIO control group animals received phosphate buffered fected by lipid extraction, leading to non-uniform morpholog- saline with BSA, lean and DIO treatment group animals re- ical changes within the brain. Also, various clearing medias ceived semaglutide (Ozempic®, Novo Nordisk A/S, have different chemical properties which will result in either Bagsværd, Denmark) dose of 0.04 mg/kg. Both groups were shrinkage or expansion of brain structures (Wan et al. 2018). administered subcutaneously 5 ml/kg and the animals were In contrast, the AIBS CCFv3 is based on vibratome-sectioned sacrificed 4 h post-dose. All animal procedures were conduct- and two-photon microscopy imaged brains. Consequently, ed in compliance with internationally accepted principles for brains imaged with LSFM differ from the AIBS CCFv3 atlas the care and use of laboratory animals and were approved by with respect to morphology and signal intensity. This affects the Danish Animal Experiments Inspectorate (license #2013- the registration accuracy and because the morphological 15-2934-00784). changes introduced by the sample processing are tissue-de- pendent, some brain regions are more prone to erroneous Sample Preparation for Immunohistochemistry alignment than others. As result, subsequent data analysis re- quires time-consuming validation and manual correction to Animals were transcardially perfused with heparinized PBS ensure accurate quantification. This is particularly relevant in and 40 ml of 10% neutral buffered formalin (CellPath, pre-clinical research where group sizes are often relatively Newtown, UK) under Hypnorm-Dormicum (fentanyl large in order to provide better statistical power. 788 μg/kg, fluanisone 25 mg/kg and midazolam 12.5 mg/kg, In our experience, the hindbrain is particularly sensitive to subcutaneous injection) anesthesia. Brains were carefully dis- erroneous registration when cleared samples are mapped di- sected and immersion-fixed in 10% neutral buffered formalin rectly onto the AIBS CCFv3. High quality registration can be overnight at room temperature on a horizontal shaker. Whole- achieved using a multi-regional approach where each larger brain samples were washed 3 × 30 min in PBS with shaking part of the brain, e.g. the hindbrain, is registered separately. and dehydrated at room temperature in methanol/H Ogradi- However, this procedure reduces analysis speed as initial seg- ent to 100% methanol (20%, 40%, 60%, 80%, 100% metha- mentation of the larger brain structures is required. We aimed nol, each step 1 h). The brains were stored in 100% methanol to preserve both data flow and quality by generating a refer- (VWR International A/S, Søborg, Denmark) at 4 °C until ence template based on iDISCO+ processed and LSFM- further processing. imaged mouse brains and aligning the AIBS CCFv3 with the template through multi-regional registration. Whole-Brain Immunohistochemistry for Labeling of c- The LSFM atlas enables fast brain-wide inter-modality regis- Fos Positive Cells and Clearing tration of other LSFM samples. To confirm accuracy and demon- strate the utility of the LSFM-based reference brain atlas, we de- The iDISCO+ (immunolabeling-enabled three-dimensional termined the c-Fos expression signature of semaglutide, a long- imaging of solvent-cleared organs) protocol was used for acting glucagon-like peptide-1 (GLP-1) receptor agonist. The whole brain immunolabelling (Renier et al. 2014, 2016). LSFM atlas enabled precise mapping of semaglutide-induced c- Samples were washed with 100% methanol for 1 h and incu- Fos expression in the mouse whole-brain. In addition to c- bated overnight in 66% dichloromethane/33% methanol Fos imaging, application of the atlas includes also map- (VWR International A/S, Søborg, Denmark) at room temper- ping other fluorescent markers imaged by LSFM. ature. Then, samples were washed twice in 100% methanol Neuroinform (2021) 19:433–446 435 for 30 min and bleached in chilled fresh 5% H O (Acros Images were acquired at 0.63 x magnification (1.2 × total 2 2 Organics, Fisher Scientific Biotech Line A/S, Slangerup, magnification) with an exposure time of 254.47 ms in a z- Denmark) in methanol overnight at 4 °C. Subsequently, the stack at 10 μm intervals. Acquired volumes (16-bit tiff) had samples were rehydrated in methanol/PBS series (80%, 60%, an in-plane resolution of 4.8 μm and z-resolution of 3.78 μm 40%, 20% methanol with 0.2% Triton X-100 (Merck, (NA = 0.156). Horizontal focusing was captured in 9 planes Darmstadt, Germany), each step 1 h) at room temperature, with blending mode set to the centre of the image to merge the washed in PBS with 0.2% Triton X-100 twice for 1 h at room individual raw images. Data was acquired in two channels, temperature and in permeabilization solution (PBS with 0.2% autofluorescence and antibody-specific channel, because the Triton X-100, supplemented with 20% volume of DMSO former provides information on tissue structure and the latter (Merck, Darmstadt, Germany) and 2.3% weight/volume gly- on neuronal activity. Autofluorescence volumes were ac- cine (Merck, Darmstact, Germany)) for 3 days at 37 °C. For c- quired at excitation wavelength of 560 ± 20 nm and emission Fos labeling, unspecific antibody binding was blocked in wavelength of 650 ± 25 nm, laser power was set to 80%. blocking solution (PBS, 0.2% TritonX-100, 10% DMSO/6% Fluorescently labelled c-Fos positive cells were captured donkey serum (Jackson ImmunoResearch, Cambridgeshire, in a specific channel at excitation wavelength of 630 ± UK)) for 2 days at 37 °C before incubated in the primary 15 nm and emission wavelength of 680 ± 15 nm, laser antibody buffer (PTwH, 5% DMSO, 3% donkey serum, power was set to 100%. 0.2% of 10% NaN (Merck, Darmstadt, Germany)) for 7 days at 37 °C. For visualization of c-Fos expression, rabbit anti-c- Fos antibody (1:5000, Cell Signaling Technology, Image Processing for Creating the Mouse Brain Atlas Massachusetts, US, cat number #2250) was used. Following incubation with primary antibody, the brains were washed in An average LSFM mouse brain volume was created from PTwH (PBS, 0.2% Tween 20 (Merck, Darmstadt, Germany), 139 individual mouse brain autofluorescence datasets by an 0.1% of 10 mg/ml heparin solution) for 1 × 10 min, 1 × iterative multi-resolution image registration algorithm 20 min, 1 × 30 min, 1 × 1 h, 1× 2 h and 1× 2 days. (Kovačević et al. 2005;Kuanetal. 2015;Umadevi Subsequently, the brains were incubated in secondary anti- Venkataraju et al. 2019). Pre-processing was initiated by body solution (PTwH, 3% donkey serum, 0.2% of 10% down-sampling to 20 μm isotropic resolution. N3 method NaN ) for 7 days at 37 °C with donkey anti rabbit Cy-5 anti- (Larsen et al. 2014; Sled et al. 1998; Van Leemput et al. body (1:1000, Jackson ImmunoResearch, Cambridgeshire, 1999) was applied to correct intensity inhomogeneity. UK, cat no #711–175-152) and washed in PTwH for 1 × Subsequently, the intensity histograms of the individual 10 min, 1 × 20 min, 1 × 30 min, 1 × 1 h, 1× 2 h and 1× 3 days. volumes were normalized and, contrast adaptive histogram For clearing, the brains were dehydrated in a methanol/H2O equalization was performed (Fig. 1a, left). For generating an series (20%, 40%, 60%, 80% and 100% methanol, each step average mouse brain template, a reference volume was ran- 1 h) at room temperature, incubated in 66% dichloromethane/ domly selected as a starting point. Six iterative multi- 33% methanol for 3 h at room temperature with shaking and in resolution registration steps – one affine and five B-spline 100% dichloromethane twice for 15 min with shaking to re- transformations were performed for the remaining samples move traces of methanol. Finally, the samples were trans- (Fig. 1a, middle). In the first step the brains were registered ferred to dibenzyl ether (Merck, Darmstadt, Germany) and to the chosen reference brain and in subsequent steps stored in closed glass vials until imaged with light sheet fluo- aligned to the average of all brains from the previous step. rescence microscope. Due to the limit in scanning depth in the Z-dimension, which is about 6 mm for our LSFM setup, about half a Light Sheet Fluorescence Microscopy of Labeled and millimetre of the dorsal cortex was not imaged. To produce Cleared Mouse Brains a template with full cortex, 15 additional image stacks of cor- tices were acquired, pre-processed and aligned to the average All whole-brain samples were imaged in an axial orientation mouse brain volume. Subsequently, both volumes were on a LaVision ultramicroscope II setup (Miltenyi Biotec, merged. Satisfactory axial symmetry was achieved by divid- Bergisch Gladbach, Germany) equipped with a Zyla 4.2P- ing the template brain volume into three coronal slabs with CL10 sCMOS camera (Andor Technology, Belfast, UK), equal thickness and manually rotating them into correct posi- SuperK EXTREME supercontinuum white-light laser EXR- tion. The final template was created by mirroring one hemi- 15 (NKT photonics, Birkerød, Denmark) and MV PLAPO sphere to the opposite side and merging the hemispheres with 2XC (Olympus, Tokyo, Japan) objective lens. The samples a sigmoidal blending function for receiving a symmetric tem- were attached to the sample holder with neutral silicone and plate brain (Fig. 1a, right) Additionally, a tissue mask and a imaged in a chamber filled with dibenzyl ether. Version 7 of ventricular mask were added to the LSFM template from the the Imspector microscope controller software was used. AIBS CCFv3 and manually adapted to fit the template. 436 Neuroinform (2021) 19:433–446 3D autofluorescence images Iterative normalization and averaging Final template I A A 1 1 2 6 + post-processed 1, pre-processed I A 6 + post-processed AIBS CCFv3 Final LSFM atlas AIBS CCFv3 A + segmentations 6 + post-processed Multi-regional registration 6 + post-processed Fig. 1 LSFM-based mouse brain atlas. a) Generation of a brain where y stands for the iteration step. B) Transfer of brain region template based on the LSFM autofluorescence volumes of 139 mice segmentations from the AIBS CCFv3 to the LSFM mouse brain brains using an iterative registration and averaging algorithm. Raw light template. Brain regions of the AIBS CCFv3 were mapped to the LSFM sheet scans are annotated with I where x stands for the animal number, template in six parts, e.g. cortex to cortex, hindbrain to hindbrain etc. and the intermediate average mouse brain volumes are annotated with A Brain regional annotations were transferred to the LSFM Segmentation refinements were performed with micros- template from the AIBS CCFv3 (Fig. 1b) (Allen Institute for copy image analysis software Imaris™ version 2 Brain Science 2011, 2015, 2017;Kuan etal. 2015;Wang etal. (Oxford instruments, Abington, UK). Image processing 2020). First, the mouse brain template of AIBS was registered was performed in Python and Elastix toolbox (Klein onto the LSFM template using multi-resolution affine and B- et al. 2010; Shamonin et al. 2014) was used to imple- spline registration. Subsequently, the registered AIBS CCFv3 ment the registrations. Detailed description of the atlas template and its segmentations were divided into six parental creation procedure and full sets of parameters can be brain regions – cerebral cortex, cerebral nuclei, hindbrain, found in the Online Resource 1. cerebellum, septal regions and interbrain together with midbrain. The parental regions were then separately reg- Quantification of c-Fos Positive Cells istered to the corresponding areas of the LSFM tem- plate. Manual corrections were performed for regions Neuronal activity was quantified by detecting and counting c- near to ventricular system, such as AP and SFO. Fos positive cells using an adapted ClearMap routine (Renier Neuroinform (2021) 19:433–446 437 et al. 2016). In brief, the volume pairs collected from the generalized linear model provided a suitable fit to our c-Fos cell autofluorescence and c-Fos specific channel were aligned count data. For each generalized linear model, a Dunnett’stest slice-by-slice using affine registration in 2D with mattes mu- was performed. Statistical analysis for determining differences in tual information as a similarity measure and background c-Fos expression between semaglutide and vehicle treated mice subtracted through morphological opening using a disk ele- involved p value adjustment using a multiple comparison meth- ment. For removing false positive c-Fos signal originating od called false discovery rate. Statistical analysis of the data was from increased tissue autofluorescence, a signal appearing performed using R statistics library. both in the autofluorescence and the c-Fos specific channel Further, all significantly regulated brain regions underwent was removed from the specific channel. For identifying c-Fos a two-step manual validation procedure for checking if the positive cells, local intensity peaks were monitored by moving used statistical model fits the data points, the significance of a filter cube over the specific channel volume followed by the brain regions is not achieved due to outliers and the raw seeded watershed for segmenting the c-Fos positive cells. signal is truly originating from the region. First, the fit of cell The initial parameters were taken from the original counts to the generalized linear model was evaluated. This ClearMap implementation (Renier et al. 2016) but optimized was done by investigating deviance residuals and checking to fit our data, being acquired under different conditions, in- if the residuals aligned with the assumptions of normality cluding image resolution. The size of the filter cube was set to and homoscedasticity. Furthermore, Cook’s distance was cal- 5x5x3 pixels for effectively detecting all possible c-Fos posi- culated for each cell count data point in the model as a mea- tive cell candidates. The third dimension of the filter cube was sure of model influence. Regions where the generalized linear chosen to be smaller than the first and second dimension of the model showed severe violations of the assumptions, or the cube since z-resolution of the LSFM volumes was lower than model contained overly influential data points, were the in-plane resolution. The coordinates of the detected local discarded. Secondly, the remaining brain regions were visual- intensity peaks were used as seeds in watershed segmentation ly studied for possible spillover signal from neighboring re- with a background intensity cut-off of 800 and the resulting gions. If the c-Fos response in a region seemed to originate segmentations were filtered by removing cell segmentation from the neighboring region, e.g. very few c-Fos positive cells regions smaller than 8 voxels and bigger than 194 voxels. were observed only in the border areas of the region while the Following c-Fos positive cell detection in the specific channel, neighboring areas were exhibiting very high signal, it was the corresponding autofluorescence volumes underwent bias declared as not significant. field correction and contrast limited adaptive histogram equal- ization (similar procedure as for the LSFM mouse brain tem- plate creation). For quantifying the number of c-Fos positive Results cells in individual brain regions, the LSFM atlas was aligned to c-Fos specific channel volumes of individual mice over pre- LSFM Reference Atlas of the Adult Mouse Brains processed autofluorescence volumes and the number of c-Fos positive was counted in every brain region. Heatmaps visual- The standard way of aligning a LSFM scanned mouse brain izing the density of the c-Fos positive cells were created by with the AIBS CCFv3 is to perform a single cross-modality mapping the specific channel volumes to the LSFM atlas registration of the full brains by computing a global affine and using the inverse transform, generating and summing the local B-spline transformation in a one-to-one manner spheres of uniform value and 20 μm radius around the centers (Fig. 2a). However, an alternative strategy is to perform mul- of the c-Fos positive cells (Renier et al. 2016). Image tiple registrations, where each of the major brain structures is processing and analysis was performed in Python. 3D aligned individually (Fig. 2b). By comparing the two ap- visualizations of heatmaps were created with microscopy proaches we observed that multiple registrations yield higher image analysis software Imaris™ version 2 (Oxford in- quality registrations in some parts of the brain, e.g. the area struments, Abington, UK). postrema (Fig. 2c). However, aligning LSFM-imaged brains using multiple registrations is time-consuming and require Statistics both initial segmentation of the larger brain structures and manual validation for each brain which is not compatible with For simplicity, 666 individual brain region segmentations of the high-throughput analysis. Our solution to this dilemma was to LSFM atlas were collapsed to their parental regions using the build an LSFM-based reference atlas by aligning the AIBS hierarchy tree of the atlas ontology (Online Resource 2) resulting CCFv3 to the LSFM-based mouse brain template through in 284 regions in which the statistical analysis was performed. multi-regional registrations. The present LSFM-based mouse For determining the difference in the c-Fos positive cell counts, a brain reference atlas can be used to analyze individual LSFM- generalized linear model was fitted to the cell counts observed in imaged samples directly by fast one-to-one registrations or for each brain region in every animal group. A negative binomial improved alignment to the AIBS CCFv3 space if needed (Fig. 438 Neuroinform (2021) 19:433–446 Whole-brain one-to-one Multi-regional Example of one-to-one and ab c registration registration multi-regional registration Cleared LSFM-imaged brain (raw data) Cleared LSFM- Cleared LSFM- AIBS CCFv3 imaged brain imaged brain template AP Cerebral cortex Septal AIBS CCFv3 registered to the regions cleared LSFM-imaged brain Hindbrain Cerebral nuclei AP AP Mid- and interbrain Whole-brain Multi- AIBS CCFv3 one-to-one regional Cerebellum template registration registration Fast Slow Fast inaccurate accurate accurate Whole-brain one- Multi-regional Whole-brain one- to-one registration registration to-one registration (performed once) Cleared LSFM- Cleared LSFM- LSFM AIBS CCFv3 imaged brain imaged brain template Fig. 2 Techniques for registering LSFM-imaged samples with the aligning cleared LSFM-imaged samples with the AIBS CCFv3 template AIBS CCFv3. a) Illustration of one-to-one registration between a cleared provides better accuracy but is relative slow compared to the one-to-one LSFM-imaged sample and the AIBS CCFv3 template. b) Illustration of registration. By generating a template from cleared LSFM-imaged brains multi-regional registration between a cleared LSFM-imaged sample and and registering the AIBS CCFv3 with it once using multi-regional regis- the AIBS CCFv3 template, where the brain volumes have been divided tration approach we ensure good alignment accuracy between the two into six larger brain areas that are mapped individually. c) Example of the templates. Subsequent registrations of cleared LSFM-imaged brains with registration quality in area postrema (AP) using either one-to-one or the LSFM template can then be done directly using fast one-to-one reg- multi-regional registration. d) Illustration of the registration flow de- istrations. This way it is possible to achieve both fast as well as accurate scribed in this manuscript. Using one-to-one registration for aligning registration of cleared LSFM-imaged brains. Regardless of computer per- cleared LSFM-imaged samples with the AIBS CCFv3 is fast but inaccu- formance the speed of analysis improved by a factor of six compared to rate in some brain regions like the AP. Multi-regional registration for the multiregional registration 2d). Regardless of computer performance we found that direct viewed from the coronal and horizontal orientation. The axial alignment to the LSFM atlas improved the registration speed resolution of the mouse brain template is 20 μm. Brain region for each brain sample volume by a factor of six. annotations for the LSFM template were imported from the An LSFM-based mouse brain reference atlas containing an AIBS CCFv3 by image registration (Fig. 1b). The annotations average anatomy template with corresponding brain region were imported as six separate pieces with manual corrections annotations was created. The mouse brain template was gen- to mitigate the challenge of cross-modality registration. The erated from 139 3D autofluorescence-scanned brain volumes final atlas contains 666 brain region segmentations with by an iterative multi-resolution image registration algorithm anatomical nomenclature corresponding to the AIBS (Fig. 1a). Post-processing of the template involved refinement CCFv3 (hierarchy tree of the atlas ontology in of the axial symmetry to obtain a midline symmetric atlas Online Resource 2) (Dong 2008). Neuroinform (2021) 19:433–446 439 Improved Registration of LSFM-Imaged Mouse Brains the AIBS CCFv3 templates, as well as in the same ten indi- vidual brain volumes which were previously used for registra- To validate that the LSFM reference atlas improved alignment tion evaluation (Fig. 3c; an atlas template containing the 27 of LSFM-acquired brain volumes, we tested alignment of ten landmarks together with the intensity variance map is raw LSFM-imaged mouse brain volumes and compared the available at GitHub and the atlas coordinates for each results to alignment with the AIBS CCFv3 using identical landmark can be found in the Online Resource 5). For the registration procedures. By computing the amount of defor- placement of each landmark several factors were considered. mation needed to register each brain into the two atlases, we The landmarks should be: 1) easily recognizable in both the evaluated the voxel-wise magnitude of displacement neces- AIBS CCFv3 and LSFM templates; 2) distributed brain-wide sary to convert the individual brain volumes to either of the such that several landmarks were located in cerebral cortex, atlas template (Fig. 3a). As expected, the LSFM-imaged brain cerebral nuclei, interbrain, midbrain, hindbrain and cerebel- volumes are less deformed when aligned with the generated lum; 3) distributed along the midline as well as in more lateral LSFM atlas compared to alignment with the AIBS CCFv3. parts of the brain; 4) placed in regions with increased local We found deformations ranging up to 13 voxels with the intensity variance, if possible (Fig. 3b). Following the regis- AIBS CCFv3 compared to deformations ranging up to 8 tration of the individual brains to the LSFM atlas and voxels with the LSFM atlas. Furthermore, the volume of the the AIBS CCFv3, the Euclidean distance between the regis- area affected by the deformation is smaller for the brains tered and atlas landmarks was calculated. Although this ap- aligned to the LSFM atlas compared to the brains aligned to proach also reflects the inherent variation that occurs when the AIBS CCFv3. The results show that deformations are placing landmarks, it consistently showed more accurate reg- most pronounced in the midbrain and hindbrain (Fig. 3a) istration when the LSFM atlas was used as a template. and most likely the reflect they morphological changes inflicted by tissue processing and clearing. Accurate c-Fos Quantification in LSFM-Imaged Mouse As the magnitude of the deformation is only an indicative Brains measure by which the registration quality cannot be fully assessed, we further investigated the alignment quality using For evaluating the performance of the LSFM atlas to assign c- a standardized metric called intensity variance developed by Fos positive cells to anatomical brain regions, we conducted a the Non-Rigid Image Registration Evaluation Project separate experiment where we mapped the brains from (NIREP) (Christensen et al. 2006). Intensity variance quan- semaglutide-dosed lean mice onto the LSFM and AIBS tifies how much the signal intensity differs per voxel between CCFv3 atlas, respectively, and compared the distribution the set of registered brain volumes and hence, estimates the and number of c-Fos positive cells counted using each atlas amount of noise in the data set. We therefore computed the (Fig. 4a). The two atlases showed highly overlapping results intensity variance for all brain regions using ten LSFM- in the majority of brain regions. However, 11 regions showed imaged mouse brains registered to both the LSFM atlas and significant differences in the number of c-Fos positive cells the AIBS CCFv3 (Fig. 3b). The mean intensity variance de- when comparing data analyzed with the two atlases (Fig. 4b- termined for the six major brain volumes registered to the c). Hence, to determine how registration accuracy impacts the AIBS CCFv3 was 17.42 for cerebral cortex, 19.09 for cerebral localization of c-Fos positive cells, we compared c-Fos signa- nuclei, 21.77 for interbrain, 32.38 midbrain, 49.34 for cere- tures in the hindbrain regions, i.e. the nucleus of the solitary bellum and 53.90 for hindbrain. In contrast, the mean intensity tract (NTS) and the dorsal motor nucleus of the vagus nerve variance computed for volumes registered to the LSFM atlas (DMX). According to the LSFM atlas, most c-Fos positive was 19.00 for cerebral cortex, 16.01 for cerebral nuclei, 18.64 cells were localized to the NTS (234 ± 38 cells) compared to for interbrain, 24.13 for midbrain, 44.19 for cerebellum, 30.31 the DMX (144 ± 14 cells) (Fig. 4d). In contrast, the AIBS for hindbrain. To analyse these findings in more detail, the CCFv3 revealed an opposite pattern (NTS, 95 ± 16 cells; intensity variance for all sub-regions within the six major DMX, 205 ± 25 cells) (Fig. 4e). To clarify which atlas is more brain regions, were plotted in scatter plots with AIBS accurate in the signal localization, we compared the raw mi- CCFv3 values on the y-axis and LSFM atlas values on the croscope images to heatmaps representing c-Fos signal densi- x-axis. As for the deformation (Fig. 3a), the most substantial ty using either atlas (Fig. 4f). The autofluorescence intensity differences in intensity variance were observed in the mid- of NTS is brighter than the intensity of surrounding tissue brain and hindbrain. The improvement of the registration ac- making it easy to delineate and shows that the raw c-Fos signal curacy using the LSFM atlas was particularly notable for hind- is indeed localized in the NTS, thus validating the LSFM atlas brain due to significantly lower intensity variance for majority mapping. Signal localization accuracy of the LSFM atlas was of the sub-regions when LSFM atlas was used for registration. also assessed for the other nine brain regions with conflicting To further compare the registration quality between the two c-Fos data (data not shown). While improved c-Fos signal atlases, 27 landmarks were identified in both the LSFM and localization by the LSFM atlas was confirmed for additional 440 Neuroinform (2021) 19:433–446 Neuroinform (2021) 19:433–446 441 Fig. 3 Improved registration of LSFM-acquired brain volumes using autofluorescence channel was applied for correction in the LSFM atlas. a) Heatmaps illustrate the average magnitude of the whole-brain mounts in both lean and DIO mice deformation resulting from the registration of ten random raw LSFM (Supplementary Fig. 1), resulting in significantly improved brain volumes to the AIBS CCFv3 and to the LSFM atlas. b) Registration using the LSFM mouse brain atlas enables improved signal-to-noise ratio specifically in DIO mice alignment between individual brains. Intensity variance, a measure for (Supplementary Fig. 2). To identify the differences between registration performance, was calculated per brain region for the ten the semaglutide and the vehicle dosed mice, average signal random brain volumes aligned to the LSFM atlas and for the same ten heatmaps in semaglutide-treated lean and obese mice were brain volumes aligned to the AIBS CCFv3. Highest intensity variance was detected in both cases in ventricular and hindbrain regions (example subtracted voxel-wise from the corresponding vehicle control sections, left). Statistical analysis of the intensity variance was performed group (Fig. 5a, c) with statistical analyses on the raw c-Fos using two-tailed Welch’s t-test and the resulting significant regions are positive cell counts (Fig. 5b, d). Compared to vehicle controls, visualized in the scatter plot (right) along with the mean intensity variance 9 brain regions were significantly regulated by semaglutide per major brain region for both atlases (denoted as mean IV). The results indicate that the difference in intensity variance values was small for treatment in both lean and DIO mice. Semaglutide treated lean cortical areas. However, majority of brain regions in cerebral nuclei, and DIO mice showed similar increased c-Fos expression in interbrain, midbrain, cerebellum and hindbrain exhibited higher intensity the bed nuclei of the stria terminalis (BST), paraventricular variance when the AIBS CCFv3 was used for registration compared to nucleus of the thalamus (PVT), xiphoid thalamic nucleus when the LSFM atlas was used for registration. c) Registration of the ten brain volumes was further evaluated using 27 landmarks distributed over (Xi), central amygdalar nucleus (CEA), parabrachial nucleus the whole brain (overview of the landmark positions, left). The landmarks (PB), nucleus of the solitary tract (NTS) and dorsal motor were divided between the six major brain areas in both atlases as well as nucleus of the vagus nerve (DMX) compared to the vehicle in the ten brain volumes. Distances between the corresponding landmarks treated controls. Additionally, semaglutide treated DIO mice in the individual brains and the atlas templates were calculated after registering the ten brain volumes to the LSFM atlas and the AIBS exhibited increased c-Fos expression in the parataenial nucle- CCFv3 (bar plot, right). For most landmarks, the calculated distances us (PT) and parasubthalamic nucleus (PSTN), whereas are lower when the LSFM atlas is used as template. Significant differ- semaglutide treated lean mice showed increased c-Fos ex- ences in distances between the two atlases was consistently observed in pression in the pedunculopontine nucleus (PPN) and cerebral cortex and hindbrain. Two-tailed Welch’st-test was appliedfor determining statistical significance in landmark distances between the mediodorsal nucleus of thalamus (MD) compared to the re- atlases: ∗ for 0.01 ≤ p <0.05, ∗∗ for 0.001 ≤ p <0.01 and ∗∗∗ for spective vehicle treated controls. p <0.001 Discussion five regions (hypoglossal nucleus (XII), presubiculum (PRE), nodulus (NOD), nucleus of the optic tract (NOT) and We present here the generation of an LSFM-based mouse postsubiculum (POST)). The AIBS CCFv3 performed better brain atlas. Compared to the AIBS CCFv3 (Allen Institute in one region, flocculus (FL), while three regions (lateral part for Brain Science 2011, 2015, 2017;Kuan et al. 2015), the of the central amygdalar nucleus (CEAl), parabrachial nucleus LSFM reference mouse brain atlas provides more accurate (PB) and pedunculopontine nucleus (PPN)), could not be anatomical segmentation and quantitative detection of properly evaluated because the ground truth could not be iden- immunolabelled markers in iDISCO+ processed and LSFM- tified due to dispersed c-Fos signal. imaged mouse brains, exemplified by characterization of whole-brain c-Fos responses to semaglutide treatment in both C-Fos Detection in a Pharmacological Study lean and DIO mice. To create an atlas template that is fully representative for To exemplify the use of the LSFM atlas we performed a study average brain anatomy we developed the LSFM atlas based on with the aim of quantifying c-Fos expression in mice dosed the variational atlas algorithm previously described with the GLP-1 receptor agonist semaglutide. Semaglutide (Kovačević et al. 2005;Kuanetal. 2015; Umadevi and vehicle was administered peripherally to lean and DIO Venkataraju et al. 2019). This algorithm avoids bias of the mice, and the c-Fos expression was evaluated 4 h post- template towards the shape of a single chosen reference brain dosing (Fig. 5). When examining the raw LSFM volumes of and accounts for morphological differences between the indi- DIO mice we observed increased autofluorescence in both the vidual brains. specific and the autofluorescence channel, which could poten- In comparison with the AIBS CCFv3, the created LSFM tially lead to false positive c-Fos signals (Supplementary brain template resulted in registrations with lower amount of Figs. 1 and 2). Increased autofluorescence in DIO mice was deformation, and the intensity variance as well as landmark present throughout the brain, but strongest in the cerebellum distances showed improved alignment for LSFM-imaged (Supplementary Fig. 2). Since the increased tissue fluores- samples. This is particularly relevant for tissue samples im- cence was apparent in both channels, but true positive c-Fos aged with LSFM since the samples have been cleared and/or signal was only present in the specific channel, the immunolabelled prior to scanning which affect brain 442 Neuroinform (2021) 19:433–446 b c Average regional cell count Top 20 regions Regions with significant differences Higher in AIBS CCFv3 AIBS CCFv3 LSFM atlas Higher in LSFM atlas Significant regions d e LSFM atlas AIBS CCFv3 NTS NTS DMX DMX 200 200 NTS DMX NTS DMX Raw data LSFM atlas AIBS CCFv3 NTS NTS NTS c-Fos c-Fos c-Fos signal cells signal No. c-Fos positive cells No. c-Fos positive cells Neuroinform (2021) 19:433–446 443 Fig. 4 Choice of brain atlas influences the number of c-Fos positive Our results demonstrate that c-Fos signal distribution in cells per brain region. Comparison of number of c-Fos positive cells in hindbrain regions is less accurately mapped using the AIBS response to semaglutide treatment using the LSFM atlas and the AIBS CCFv3 compared to delineation of signals using our LSFM CCFv3. a) Average number of detected c-Fos expressing cells in every brain region after registration to either the AIBS CCFv3 or the LSFM reference brain atlas. The large difference may be explained atlas. Regions in which the c-Fos positive cells are differentially quanti- by the highamountoflipid-richmyelinfibersinthispartof fied are highlighted by a circle surrounding the data points. An average the brain (Smith 1973). As solvent-based tissue clearing cell count per group below ten is considered too low to judge. b) The bar removes lipids, this could explain the difficulty of mapping chart lists the brain regions and the corresponding mean log fold changes of quantified c-Fos positive cells in these regions according to the p value. certain brain volumes to the AIBS CCFv3 which is based on Blue = higher with LSFM atlas, red = higher with CCFv3. NS stands for non-cleared tissue samples. In addition to the NTS and DMX, not significant, ∗ for 0.01 ≤ p < 0.05, ∗∗ for 0.001 ≤ p < 0.01 and ∗∗∗ for we found improved signal localization using the LSFM atlas p < 0.001. c) Horizontally and sagittally depicted brain volumes highlight in five other brain areas. In four of these areas the improve- the regions in which the c-Fos cells were differentially quantified while using the LSFM atlas and the AIBS CCFv3 for the analysis (same colour ment could be assigned to the detailed ventricular mask creat- code as in b and c). See Online Resource 2 for full names of the brain ed for the LSFM atlas. Because the AIBS CCFv3 template regions. d-e) Comparison of total number of c-Fos positive cells quanti- depicts a narrower ventricular system compared to the LSFM fied in 3D-volumes of the nucleus of the solitary tract (NTS) and the atlas template, this may have resulted in incorrectly assigned dorsal motor nucleus of the vagus nerve (DMX) using the LSFM atlas and the AIBS CCFv3. DMX (blue) and NTS (grey) volumes of both c-Fos signal from the choroid plexus to nearby brain regions. atlases in which the signal (glow colormap) was quantified is visualized In the FL, the AIBS CCFv3 performed better than the LSFM in 3D renderings. d) Quantification of c-Fos positive cells following reg- atlas. However, as the FL is often damaged or dislocated dur- istration of the LSFM atlas to the LSFM-acquired brain volumes showed ing dissection of the brain this may impact the subsequent that in average 234 ± 38 c-Fos positive cells were found in the NTS and 144 ± 14 in the DMX. e) Quantification of c-Fos positive cells following mapping. In three brain regions we detected a significant dif- registration of the AIBS CCFv3 to the LSFM-acquired brain volumes. ference in the mapping, but we were unable to determine Here the majority of the signal is found in the DMX. Quantification which of the two atlases performed best because the c-Fos revealed that on average 95 ± 16 c-Fos positive cells are counted in the signal was too scattered. NTS and 205 ± 25 in the DMX. f) Comparing the raw data to the data in alignment with atlases. DMX has a dense dark appearance compared to In terms of c-Fos detection, DIO mice exhibited relatively NTS high unspecific background signals as compared to lean con- trols, most likely attributed to lipid-associated autofluores- cence. Lipid-containing residues of lysosomal digestion, lipofuscins, have also been reported to increase during aging morphology by shrinkage/expansion and de-lipidation (Kim and oxidative stress (Boellaard et al. 2004) leading to in- et al. 2018; Wan et al. 2018). Furthermore, the contrast within creased autofluorescence (Cho and Hwang 2011;Di Guardo anatomical structures in the brain that are important for sub- 2015; Schnell et al. 1999). When comparing the c-Fos activity sequent image registration differs between the AIBS CCFv3 maps between lean and DIO mice we found that the response template and brain processed for LSFM. These issues have to semaglutide looked overall similar in both phenotypes with also been recognized by other researchers and a need for a significant c-Fos activation in BST, PVT, Xi, CEA, PB, NTS dedicated atlas for cleared LSFM-imaged brains has previous- and DMX. Semaglutide is a glucagon-like peptide-1 (GLP-1) ly been highlighted (Umadevi Venkataraju et al. 2019). analogue which has been shown to activate GLP-1 receptors In this study annotations from the AIBS CCFv3 were in the hypothalamus and brainstem (Secher et al. 2014)and mapped to the LSFM atlas template (Wang et al. 2020). markedly stimulates c-Fos expression in mice (Kjaergaard However, as annotation volumes are continuously refined et al. 2019; Salinas et al. 2018). The observed c-Fos expres- (Chon et al. 2019), these can also be aligned to the LSFM sion pattern observed in this study fits well with these previous template. The process of mapping annotations from an reports. Only slight differences were seen between lean and existing atlas to the LSFM-template depends on cross- DIO as exemplified by only DIO mice showed significantly subject cross-modality registration (i.e. different brain, differ- upregulated c-Fos expression in the PT and PSTN. It should ent microscope) which is difficult and often requires manual be noted that lean mice demonstrated a similar c-Fos expres- corrections. With the respect to mapping annotations from the sion pattern in these regions which, however, did not attain AIBS CCFv3 to the LSFM-template, the main difficulty was statistical significance. related to morphology differences in the hindbrain and ven- In this study a c-Fos was detected using a Cy5 labelled tricular system. This was solved by stepwise mapping of the secondary antibody. Consequently we used the 560 nm to annotations for larger parts of the brain such as the hindbrain, record the autofluorescence which is different from the map- together with manual corrections around the ventricular sys- ping reported in the original ClearMap protocol (Renier et al. tem. Now complete, the LSFM atlas provides the benefit of 2016). However, since the choice of fluorophores might vary improved registration of other LSFM-samples together with from study to study, we tested how the choice of autofluores- cence impacts the subsequent mapping (atlas-registered detailed brain region annotations. 444 Neuroinform (2021) 19:433–446 a b Vehicle vs semaglutide, lean mice BST CEA PVT PB NTS DMX c d Vehicle vs semaglutide, DIO mice BST CEA PVT PSTN PB NTS DMX Fig. 5 Differentially regulated c-Fos expression in response to vehicle treatment and corresponding mean log fold changes of c-Fos semaglutide administration. Up (red) and down (blue) regulation of c- positive cells in these regions in b) lean and d) DIO mice. ∗ stands for Fos expression in a) semaglutide treated lean mice in comparison to 0.01 ≤ p < 0.05, ∗∗ for 0.001 ≤ p<0.01 and ∗∗∗ for p < 0.001. P-values vehicle treated lean mice and c) in semaglutide treated DIO mice in were adjusted for multiple comparisons using the false discovery rate. See comparison to vehicle treated DIO mice. Differentially regulated brain Online Resource 2 for full names of the brain regions regions in response to semaglutide administration in comparison to autofluorescence volumes can be found in Github). Although, accumulation. In this case it did not impact on the registration, we obtained the best registration using the 560 nm channel to but it will always be important to consider the possibil- record the autofluorescence, channels below 700 nm ity that the choice of model may influence registration worked as well. When reaching the NIR spectrum the and quantification. endogenous fluorescence become so weak it can no lon- In conclusion, we developed a dedicated reference atlas ger be used for registration. allowing faster and more accurate mapping of iDISCO+ proc- The average brain generated in this study was created from essed and LSFM-imaged whole mouse brains. In combination 8 to 10 week old C57Bl/6 J male mice. Since factors such as with an improved c-Fos detection algorithm, our pipeline en- age, sex and strain are known to affect brain size and anatomy, ables for unbiased, automated and computationally efficient it is possible deviations from the average parameters may have quantitative analysis of drug-induced c-Fos expression in the a slight impact on the overall quality of registration and quan- intact mouse brain. The LSFM atlas is highly applicable for tification. Indeed, we observed that obesity led to an unexpect- fast and precise mapping of fluorescent markers in both the ed increase in autofluorescence, presumably due to lipofuscin normal mouse brain and mouse models of CNS diseases as Neuroinform (2021) 19:433–446 445 Boellaard, J. W., Schlote, W., & Hofer, W. (2004). Ultrastructural pathol- well for improved delineation of compound distribution in the ogy species-specific ultrastructure of neuronal Lipofuscin in CNS imaged by LSFM (Liebmann et al. 2016; Roostalu et al. Hippocampus and Neocortex of subhuman mammals and humans. 2019; Salinas et al. 2018;Secher etal. 2014). Ultrastruct Pathol, 28(5–6), 341–351. https://doi.org/10.1080/ Cho, S., & Hwang, E. S. (2011). Fluorescence-based detection and quan- tification of features of cellular senescence. Methods Cell Biol, 103, Information Sharing Statement 149–188. https://doi.org/10.1016/B978-0-12-385493-3.00007-3. Chon, U., Vanselow, D. J., Cheng, K. C., & Kim, Y. (2019). Enhanced and unified anatomical labeling for a common LSFM reference atlas files are freely accessible at https:// mouse brain atlas. Nat Commun, 10, 5067. https://doi.org/ github.com/Gubra-ApS. Quantitative c-Fos data for all brain 10.1038/s41467-019-13057-w. regions is available as Online Resources 3 and 4. Source code Christensen, G. E., Geng, X., Kuhl, J. G., Bruss, J., Grabowski, T. J., Pirwani, I. A., et al. (2006). Introduction to the non-rigid image used for generating the LSFM reference atlas along with the registration evaluation project (NIREP). Lecture Notes in code for detecting and quantifying the number of c-Fos posi- Computer Science (including subseries Lecture Notes in Artificial tive cells in LSFM mouse brain volumes is accessible at Intelligence and Lecture Notes in Bioinformatics), 4057,128–135. https://github.com/Gubra-ApS. https://doi.org/10.1007/11784012_16. Chung, K., Wallace, J., Kim, S.-Y., Kalyanasundaram, S., Andalman, A. Acknowledgements The authors would like to acknowledge Lotte S., Davidson, T. J., Mirzabekov, J. J., Zalocusky, K. A., Mattis, J., Denisin, A. K., Pak, S., Bernstein, H., Ramakrishnan, C., Grosenick, Ankjær Frederiksen and Hanne Duus Laustsen for skillful technical L., Gradinaru, V., & Deisseroth, K. (2013). Structural and molecular assistance. interrogation of intact biological systems. Nature, 497(7449), 332– 337. https://doi.org/10.1038/nature12107. Funding The work was funded by Gubra ApS and Innovation Fund Detrez, J. R., Maurin, H., Van Kolen, K., Willems, R., Colombelli, J., Denmark grant number (8053-00121B). Lechat, B., et al. (2019). Regional vulnerability and spreading of hyperphosphorylated tau in seeded mouse brain. Neurobiol Dis, Compliance with Ethical Standards 127,398–409. https://doi.org/10.1016/j.nbd.2019.03.010. Di Guardo, G. (2015). Lipofuscin, lipofuscin-like pigments and autoflu- orescence. Eur J Histochem, 59(1), 1–2. https://doi.org/10.4081/ejh. Conflict of Interest Johanna Perens, Casper Gravesen Salinas, Jacob 2015.2485. Lercke Skytte, Urmas Roostalu, Franziska Wichern, Pernille Barkholt, Dong, H. . (2008). The Allen reference atlas: A digital color brain atlas of and Jacob Hecksher-Sørensen are currently employed by Gubra ApS. Niels Vrang and Jacob Jelsing are the owners of Gubra ApS. the C57BL/6J male mouse. Wiley. Dragunow, M., & Faull, R. (1989). The use of c-fos as a metabolic marker in neuronal pathway tracing. J Neurosci Methods, 29(3), 261–265. Ethics Approval All animal procedures were conducted in compliance https://doi.org/10.1016/0165-0270(89)90150-7. with internationally accepted principles for the care and use of laboratory Ertürk, A., Becker, K., Jährling, N., Mauch, C. P., Hojer, C. D., Egen, J. animals and were approved by the Danish Animal Experiments G., Hellal, F., Bradke, F., Sheng, M., & Dodt, H. U. (2012). Three- Inspectorate (license #2013-15-2934-00784). dimensional imaging of solvent-cleared organs using 3DISCO. Nat Protoc, 7(11), 1983–1995. https://doi.org/10.1038/nprot.2012.119. Open Access This article is licensed under a Creative Commons Esquerda-Canals, G., Montoliu-Gaya, L., Güell-Bosch, J., & Villegas, S. Attribution 4.0 International License, which permits use, sharing, (2017). Mouse models of Alzheimer’s disease. J Alzheimers Dis, 57, adaptation, distribution and reproduction in any medium or format, as 1171–1183. https://doi.org/10.3233/JAD-170045. long as you give appropriate credit to the original author(s) and the Jensen, C. B., Secher, A., Hecksher-Sørensen, J., Conradsen, K., & source, provide a link to the Creative Commons licence, and indicate if Larsen, R. (2015). Quantification of brain access of exendin-4 in changes were made. The images or other third party material in this article the C57BL mouse model by SPIM fluorescence imaging and the are included in the article's Creative Commons licence, unless indicated Allen mouse brain reference model. In Lecture Notes in Computer otherwise in a credit line to the material. If material is not included in the Science (including subseries Lecture Notes in Artificial Intelligence article's Creative Commons licence and your intended use is not and Lecture Notes in Bioinformatics) https://doi.org/10.1007/978-3- permitted by statutory regulation or exceeds the permitted use, you will 319-19665-7_38. need to obtain permission directly from the copyright holder. To view a Kim, J. H., Jang, M. J., Choi, J., Lee, E., Song, K. D., Cho, J., et al. copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. (2018). Optimizing tissue-clearing conditions based on analysis of the critical factors affecting tissue-clearing procedures. Sci Rep, 8, 12815. https://doi.org/10.1038/s41598-018-31153-7. Kjaergaard, M., Salinas, C. B. G., Rehfeld, J. F., Secher, A., Raun, K., & References Wulff, B. S. (2019). PYY(3-36) and exendin-4 reduce food intake and activate neuronal circuits in a synergistic manner in mice. Allen Institute for Brain Science. (2011). Allen Mouse Brain Atlas. Neuropeptides, 73,89–95. https://doi.org/10.1016/j.npep.2018.11. Available from: http://mouse.brain-map.org. Allen Institute for Brain Science. (2015). Allen mouse common coordi- Klein, S., Staring, M., Murphy, K., Viergever, M. A., & Pluim, J. P. W. nate framework. Technical White Paper. (2010). Elastix: A toolbox for intensity-based medical image regis- Allen Institute for Brain Science. (2017). Allen mouse common coordi- tration. IEEE Trans Med Imaging, 29(1), 196–205. https://doi.org/ nate framework and reference atlas. Technical White Paper. 10.1109/TMI.2009.2035616. Bobela, W., Zheng, L., & Schneider, B. L. (2014). Overview of mouse Kovačević, N., Henderson, J. T., Chan, E., Lifshitz, N., Bishop, J., Evans, models of Parkinson’sdisease. Current protocols in mouse biology, A. C., Henkelman, R. M., & Chen, X. J. (2005). A three- 4(3), 121–139. https://doi.org/10.1002/9780470942390.mo140092. dimensional MRI atlas of the mouse brain with estimates of the 446 Neuroinform (2021) 19:433–446 average and variability. Cereb Cortex, 15(5), 639–645. https://doi. Schnell, S. A., Staines, W. A., & Wessendorf, M. W. (1999). Reduction org/10.1093/cercor/bhh165. of Lipofuscin-like autofluorescence in fluorescently labeled tissue. J Kuan, L., Li, Y., Lau, C., Feng, D., Bernard, A., Sunkin, S. M., Zeng, H., Histochem Cytochem, 47(6), 719–730. https://doi.org/10.1177/ Dang, C., Hawrylycz, M., & Ng, L. (2015). Neuroinformatics of the 002215549904700601. Allen mouse brain connectivity atlas. Methods, 73,4–17. https://doi. Secher, A., Jelsing, J., Baquero, A. F., Hecksher-Sørensen, J., Cowley, org/10.1016/j.ymeth.2014.12.013. M. A., Dalbøge, L. S., Hansen, G., Grove, K. L., Pyke, C., Raun, K., Larsen, C. T., Iglesias, J. E., & Van Leemput, K. (2014). N3 Bias Field Schäffer, L., Tang-Christensen, M., Verma, S., Witgen, B. M., correction explained as a Bayesian modeling method. Bayesian and Vrang, N., & Bjerre Knudsen, L. (2014). The arcuate nucleus me- grAphical Models for Biomedical Imaging. Lect Notes Comput Sci, diates GLP-1 receptor agonist liraglutide-dependent weight loss. J 8677,1–12. https://doi.org/10.1007/978-3-319-12289-2. Clin Investig, 124,4473–4488. https://doi.org/10.1172/JCI75276. Leung, C., & Jia, Z. (2016). Mouse genetic models of human brain dis- Shamonin, D. P., Bron, E. E., Lelieveldt, B. P. F., Smits, M., Klein, S., & orders. Front Genet, 7, 40. https://doi.org/10.3389/fgene.2016. Staring, M. (2014). Fast parallel image registration on CPU and GPU for diagnostic classification of Alzheimer’sdisease. Liebmann, T., Renier, N., Bettayeb, K., Greengard, P., Tessier-Lavigne, Frontiers in Neuroinformatics, 7,50. https://doi.org/10.3389/fninf. M., & Flajolet, M. (2016). Three-dimensional study of Alzheimer’s 2013.00050. disease hallmarks using the iDISCO clearing method. Cell Rep, Sled, J. G., Zijdenbos, A. P., & Evans, A. C. (1998). A nonparametric 16(4), 1138–1152. https://doi.org/10.1016/j.celrep.2016.06.060. method for automatic correction of intensity nonuniformity in mri Nectow, A. R., Schneeberger, M., Zhang, H., Field, B. C., Renier, N., data. IEEE Trans Med Imaging, 17(1), 87–97. https://doi.org/10. Azevedo, E., Patel, B., Liang, Y., Mitra, S., Tessier-Lavigne, M., 1109/42.668698. Han, M. H., & Friedman, J. M. (2017). Identification of a brainstem Smith, M. E. (1973). A regional survey of myelin development: Some circuit controlling feeding. Cell., 170,429–442.e11. https://doi.org/ compositional and metabolic aspects. J Lipid Res, 14(5), 541–551. 10.1016/j.cell.2017.06.045. Umadevi Venkataraju, K. U., Gornet, J., Murugaiyan, G., Wu, Z., & Renier, N., Adams, E. L., Kirst, C., Dulac, C., Osten, P., & Tessier- Osten, P. (2019). Development of brain templates for whole brain Lavigne, M. (2016). Mapping of brain activity by automated volume atlases. Progress in Biomedical Optics and Imaging - Proceedings analysis of immediate early genes. Cell, 165(7), 1789–1802. https:// of SPIE, 10865, 1086511. https://doi.org/10.1117/12.2505295. doi.org/10.1016/j.cell.2016.05.007. Van Leemput, K., Maes, F., Vandermeulen, D., & Suetens, P. (1999). Renier, N., Wu, Z., Simon, D. J., Yang, J., Ariel, P., & Tessier-Lavigne, Automated model-based Bias Field correction of MR images of the M. (2014). iDISCO: A simple, rapid method to immunolabel large brain. IEEE Trans Med Imaging, 18(10), 885–896. https://doi.org/ tissue samples for volume imaging. Cell, 159(4), 896–910. https:// 10.1109/42.811268. doi.org/10.1016/j.cell.2014.10.010. Vigouroux, R. J., Belle, M., & Chédotal, A. (2017). Neuroscience in the Rocha, M. D., Düring, D. N., Bethge, P., Voigt, F. F., Hildebrand, S., third dimension: Shedding new light on the brain with tissue clear- Helmchen, F., Pfeifer, A., Hahnloser, R. H. R., & Gahr, M. (2019). ing. Molecular brain, 10(1), 33. https://doi.org/10.1186/s13041- Tissue clearing and light sheet microscopy: Imaging the 017-0314-y. Unsectioned adult Zebra finch brain at cellular resolution. Front Villares, R., Gutiérrez, J., Fütterer, A., Trachana, V., Del Burgo, F. G., & Neuroanat, 13,1–7. https://doi.org/10.3389/fnana.2019.00013. Martínez-A, C. (2015). Dido mutations trigger perinatal death and Roostalu, U., Salinas, C. B. G., Thorbek, D. D., Skytte, J. L., Fabricius, generate brain abnormalities and behavioral alterations in surviving K., Barkholt, P., John L. M., Jurtz V. I., Knudsen L. B., Jelsing J., adult mice. Proc Natl Acad Sci U S A, 112,4803–4808. https://doi. Vrang N., Hansen H. H., Hecksher-Sørensen J. (2019). Quantitative org/10.1073/pnas.1419300112. whole-brain 3D imaging of tyrosine hydroxylase-labeled neuron Wan, P., Zhu, J., Xu, J., Li, Y., Yu, T., & Zhu, D. (2018). Evaluation of architecture in the mouse MPTP model of Parkinson’s disease. seven optical clearing methods in mouse brain. Neurophotonics., 5. DMM Disease Models and Mechanisms https://doi.org/10.1242/ https://doi.org/10.1117/1.nph.5.3.035007. dmm.042200, 12, dmm042200. Wang, Q., Ding, S. L., Li, Y., Royall J., Feng D., Lesnar P., Graddis N., Salinas, C. B. G., Lu, T. T. H., Gabery, S., Marstal, K., Alanentalo, T., Naeemi M., Facer B., Ho A., Dolbeare T., Blanchard B., Dee N., Mercer, A. J., Cornea, A., Conradsen, K., Hecksher-Sørensen, J., Wakeman W., Hirokawa K. E., Szafer A., Sunkin S. M., Oh S. W., Dahl, A. B., Knudsen, L. B., & Secher, A. (2018). Integrated brain Bernard A., Phillips J. W., Hawrylycz M., Koch C., Zeng H., Harris atlas for unbiased mapping of nervous system effects following J. A., & Ng, L. (2020). The Allen mouse brain common coordinate Liraglutide treatment. Sci Rep, 8(1), 10310. https://doi.org/10. framework: A 3D reference atlas. Cell 181,(4), 936–953.e20. https:// 1038/s41598-018-28496-6. doi.org/10.1016/j.cell.2020.04.007. Schneeberger, M., Parolari, L., Das Banerjee, T., Renier, N., Friedman, J. M., & Nectow, A. R. (2019). Regulation of energy expenditure by brainstem GABA neurons in brief. Cell, 178(3), 672–685.e12. Publisher’sNote Springer Nature remains neutral with regard to juris- https://doi.org/10.1016/j.cell.2019.05.048. dictional claims in published maps and institutional affiliations.

Journal

NeuroinformaticsSpringer Journals

Published: Oct 16, 2020

Keywords: Light sheet fluorescence microscopy; iDISCO; Tissue clearing; Brain atlas; C-Fos; Whole brain imaging

References