Optical Imaging of the Small Intestine Immune Compartments Across Scales

The limitations of 2D microscopy constrain our ability to observe and understand tissue-wide networks that are, by nature, 3-dimensional. Optical projection tomography enables the acquisition of large volumes (ranging from micrometres to centimetres) in various tissues, with label-free capacities for the observation of auto-uorescent signals as well uorescent-labelled targets of interest in multiple channels. We present a multi-modal work�ow for the characterization of both structural and quantitative parameters of the mouse small intestine. As proof of principle, we evidence its applicability for imaging the mouse intestinal immune compartment and surrounding mucosal structures. We quantify the volumetric size and spatial distribution of Isolated Lymphoid Follicles (ILFs) and quantify density of villi throughout centimetre long segments of intestine. Furthermore, we exhibit the age-and microbiota-dependence for ILF development, and leverage a technique that we call reverse-OPT for identifying and homing in on regions of interest. Several quanti�cation capabilities are displayed, including villous density in the auto�uorescent channel and the size and spatial distribution of the signal of interest at millimetre-scale volumes. The concatenation of 3D image acquisition with the reverse-OPT sample preparation and a 2D high-resolution imaging modality adds value to interpretations made in 3D. This cross-modality referencing technique is found to provide accurate localisation of ROIs and to add value to interpretations made in 3D. Importantly, OPT may be used to identify sparsely-distributed regions of interest in large volumes whilst retaining compatibility with high-resolution microscopy modalities, including confocal microscopy. We believe this pipeline to be approachable for a wide-range of specialties, and to provide a new method for characterisation of the mouse intestinal immune compartment.


Introduction
The intestine forms an interface between the external environment and the rest of the body, ful lling many essential functions in the process.Among these are immune system education, and the regulation of the microbiome -which are incidentally interdependent 1,2 .Indeed, we now know that the microbiome is necessary for the correct education of the immune system 3,4 and that this has long-term repercussions on immunity in the gut.As such, the gut microbiome has been linked to distinctly immune-related disorders ranging from obesity 5 and diabetes 6 to auto-immune 7 and allergic diseases [8][9][10] .Importantly, signi cant advances in the aetiology of such disorders were made through studying gut structure [11][12][13] .
For example, de cits in gut morphology and barrier permeability have been implicated in both obesity 14,15 and diabetes 16 , with the latter showing signs of decline prior to type-1-diabetes onset in non-obese diabetic mice (NOD).The relevance of observing cell localization within tissue regions of the gut is further exempli ed by microscopic observations of intraepithelial lymphocyte localisation, which highlighted the impact of high-fat-diet-induced obesity on the control of pathogen transepithelial migration and the mediation of intestinal epithelial repair 15 .
Secondary and tertiary lymphoid structures (LT) are strategically positioned along the gastrointestinal tract to orchestrate immuno-surveillance against pathogens and invading microorganisms.These are known to develop during early life 17 , and to require stimuli from the microbiome 18,19 .Peyer's patches (seconday LT) and isolated lymphoid follicles (ILFs, tertiary LT) form key networks where microbial antigens are presented to T-and B-cells, effectively dictating both regulation and tolerance to commensals.While new methods are continuously being developed for immuno-phenotyping and isolation of immune cells from ILFs 20,21 , their function -deriving from their location at the interface with the microbiota -also necessitates further research in their natural context.Concurrently, the awareness that gut spatial structures have a powerful impact on establishing and sustaining the signaling and metabolic exchanges demanded techniques that will deliver in large-area, volumetric mapping.Existing solutions include confocal microscopy in combination with FISH 22 and lattice light-sheet microscopy 23 .
Optical projection tomography (OPT) is a 3D imaging modality that is ideal for mesoscale imaging, offering broad applicability while maintaining compatibility with other microscopy techniques.With elds of view spanning from a few millimetres to 60mm in length 24 and full 3D-volume acquisition times ranging from minutes to an hour 25 , OPT is a time-and cost-effective technique with which large-scale structural and functional parameters can be imaged.OPT has been leveraged for concatenated, functional multi-channel plant imaging 24 , functional cell proliferation imaging in zebra sh 26 , multiorientation digital sectioning of whole mouse embryos 27 and mouse organ imaging of the liver, pancreas 25 and brain 28 .
Recently, we developed a multi-spectral OPT modality to image the mouse gut 29 , with a 3D resolution of 28µm allowing the distinction of mucosal layers and villi in samples several centimetres in length.To demonstrate the wide range of applications for this method, we present a work ow that enables mesoscale observation of signal distribution throughout millimetre-long gut sections with auto uorescent contextualization, as well as the identi cation of regions of interest that can be characterized at higherresolution following reverse OPT (RevOPT) processing.We apply this work ow to visualise the gut immune compartment and present the rst observation of isolated lymphoid follicle (ILF) distribution in a single acquisition spanning several millimetres of gut tissue.Furthermore, we con rm the lack of large organised lymphoid structures in young SPF and old germ-free mice, likely re ecting the age-and microbiome-dependence of the gut immune system development.Finally, by implementing RevOPT and subsequent confocal microscopy, we show the feasibility of tracking regions of interest (ROIs) initially selected in 3D for subsequent higher resolution imaging using traditional histology methods.This method is a powerful approach to characterizing tissues at multiple scales while providing high resolution data with ready-to-use processing pipelines optimized for the mouse gut.

Sample preparation pipeline and the power of tissue auto uorescence
To achieve the multi-scale observation of spatially-distributed biological signals of interest, we developed a sample preparation pipeline that includes two imaging modalities: high-volume optical projection tomography and high-resolution confocal microscopy after reverse OPT (RevOPT, Fig. 1).The pipeline is divided into four phases, sample preparation (Fig. 1 a-k), imaging and image processing (Fig. 1 l-m), reverse OPT (Fig. 1 n-r) and secondary imaging (Fig. 1 s), spanning a duration of approximately three weeks.First, tissue preservation, auto uorescence quenching and tissue permeabilisation are performed to prepare the samples for staining.This is followed by uorescent antibody staining of select markers.A clearing step precedes the acquisition of optical projections over a 360° sample rotation.Next, a ltered back-projection algorithm is used to reconstruct the projections into a 3D image as described previously 29 .Using RevOPT we revert the sample to a state compatible with freezing in Optimal Cutting Temperature (OCT) compound, allowing for cryostat sectioning and counterstaining (Fig. 1, p-r).Finally, in amongst several methods requiring thin sectioning including electron microscopy or single-molecule FISH, we selected confocal microscopy to image regions of interest identi ed by OPT with improved resolution.
Tissue auto uorescence is an inherent signal produced by extracellular matrix components and certain pigmented cell types.In OPT, auto uorescence quenching is required to reduce noise and retain targeted uorescent signals 29 (Fig. 1, steps a and c).However, low levels of auto uorescence enable the discrimination of the outer and inner layers of the gut when samples are illuminated at 415nm spectrally ltered between 400-440nm, whilst emission is collected within the range of 455-520nm 29 .A longitudinal portrayal of the gut (Fig. 2a) provides an overview of the structures present in the tissue.In reconstructions made up of 1200 projections, well-resolved villi can be observed in a 3D visualisation software (Fig. 2b).When taking a cross-sectional view, the mucosal layers can be distinguished from the villi in the OPT scan (Fig. 2c; mu = muscularis, sm = submucosa, m = mucosa and L = lumen) whilst a greater resolution is achieved by confocal microscopy on the same sample having undergone RevOPT (Fig. 2d).During RevOPT, counterstaining is possible and demonstrated here by the staining of DNA with DAPI (Fig. 2d).The improved preservation of cross-sectional structure in OPT is evident when comparing the virtual section (Fig. 2c) and its histological counterpart (Fig. 2d), with loss of tissue and distortion being apparent in the confocal image.
We implemented a virtual unfolding technique 30 (for more detail, see supplementary SI Fig. 1) to observe the gut tissue from within the lumen, with sections spanning from this point to the serosa (Fig. 2e, section closest to lumen).Villous density could be calculated by segmenting the unfolded image and nding local maxima (Fig. 2f).This can be performed for the whole tissue region or applied to smaller regions of interest to probe different areas of the tissue.In this healthy tissue, overall villous density is mostly homogeneous (Fig. 2f).This data can be transformed into a quantitative visualisation of different sectors (Fig. 2g).Virtual unfolding also yields a straightened image (Fig. 2h) of the tissue cross section seen in Fig. 2c.
Virtual unfolding of 3D-reconstructed data can lead to detailed visualizations of structures that are di cult to visualise in a 3D image such as Fig. 2a or in a virtual cross-section as in Fig. 2c.We found a suspected lymphoid follicle in the auto uorescence channel of a different sample (top view Fig. 2i, side view Fig. 2j), whose structural context is made clear by virtual unfolding (Fig. 2k and straightened Fig. 2l).The follicle is made up of three lobes, with a concentration of uorescent vessels in the centre.In the areas surrounding the follicle, gaps in the villi suggests the potential presence of lymphatic vasculature.Typically, large vascular networks are di cult to observe by visualization of cross-sections.In Fig. 2m, an example of such a network is shown, highlighting the added value that processing the auto uorescence channel can bring to gut structure characterization.
3.2 Cell-type speci c signal distribution throughout the gut volume OPT can also be used for visualisation of cell types according to stainings of selective markers.To demonstrate this, we chose to stain the gut immune compartment, due to its structured organisation under healthy conditions and its common deregulation in intestinal diseases (e.g.IBD 31 ) and other systemic disorders (e.g.metabolic diseases 32,33 , autoimmunity 34 and neurodegeneration 35,36 ).For this we stained CD45-positive cells using uorescently labelled antibodies.In healthy adult mice, immune cells are found interspersed at regular intervals or compartmentalised in gut-associated lymphoid structures (GALTs) known as isolated lymphoid follicles (ILFs, Fig. 3a triangle).Overlaying the auto uorescence channel reveals other adjacent structures such as blood vessels and luminal dietary bers (Fig. 3a, cross and square respectively).In order to view the three-dimensional characteristics of such an OPT image, a movie is provided in the supplementary information (SI Movie 1).
It is known that age-and microbiota-dependent education of the immune system is responsible for the formation of lymphoid structures such as Peyer's patches and ILFs 37,38 .We con rm that with OPT, we are able to identify differences in the immune cell compartments in the contexts of young (14 days) SPF mice and old (30+ weeks) germ-free mice (Fig. 3b and 3c respectively), compared to the old SPF sample shown in Fig. 3a.At a young age under normal rearing conditions, no dense regions of immune cells are observed (Fig. 3b).Intestines of older, germ-free mice also exhibit reduced CD45-positive cell clusters in the mucosal layers.A two-channel cross-sectional view of these samples (Fig. 3d-f) frames the immune signal within the structured layers of the gut.An isolated lymphoid follicle is located within the submucosal layers and is surrounded by smaller, less dense CD45-positive clusters in gure 3d.Conversely, there is no speci c uorescence in both old germ-free and young SPF animals, thus indicating a lack of well-de ned GALT structures in these mouse models (Fig. 3b, 3c, 3e, and 3f).
OPT reconstructions can thus be used for broader, organ-scale characterisation of the mouse intestine.
Furthermore, current 3D image processing tools allow for accurate quanti cation of different parameters.
We segmented the ILFs in the 625nm channel alone (Fig. 3g) and found their size ranges from approximately 1 to 5 million μm 3 (Fig. 3h).Their spatial distribution along the small intestine is uniform, averaging at 500μm in between ILFs (Fig. 3i).

gutOPT pipeline for multi-modal imaging and high resolution characterisation of the gut
Because sample preparation for optical projection tomography is compatible with downstream processing for additional imaging modalities, we wondered whether we could incorporate a single pipeline for imaging at different scales.We performed reverse-OPT (Fig. 1 n and o) on the samples shown in Fig. 3 and imaged them using confocal microscopy (Fig. 4).
We wondered whether we could use OPT to pre-select regions of interest, and retrace them and image them using high-resolution techniques.To do this, we selected isolated lymphoid follicles in the OPT reconstruction and calculated their distance from the edge of the sample (Fig. 4a i and ii).Once samples had undergone RevOPT and were mounted in optimal cutting temperature (OCT) compound, the depth of each cryosection was used to track the localization of the ROIs.We nd that the uorescence signal was maintained from the OPT staining, and sections do not require further immuno-staining for confocal imaging.
Speci cally, we nd that preselected ILF regions observed by OPT are high-density cell clusters rich in CD45-positive cells (Fig. 4b and c).In both ROIs containing ILFs, the calculated distances were accurate, and we nd the immune cell-dense region situated within the submucosa as expected from the OPT reconstructions and their known localisation 39,40 .Areas lacking GALTs in 3D (Fig. 4a iii) only contain sparse positive cells in the lamina propria at higher resolution (Fig. 4d).By measuring the immune cell density in the whole-sectioned GALT regions we nd that the signal density threshold for visibility in OPT is approximately 400 uorescent cells per mm 2 of DAPI signal (Fig. 4e).We imaged the adult germ-free mice that display no GALTs by OPT (Fig. 4f).Here, we nd no CD45+ cells along the length of the villi nor in the submucosa, con rming that no ILFs are present, and suggesting that the lack of a microbiome indeed alters the immune compartment in the gut (Fig. 4f, triangle).The number of immune cells is also signi cantly reduced compared to that observed in the gut of SPF mice (Fig. 4g).Thus, OPT can be used to identify speci c structures and markers of interest using tissue-wide staining, and given a su ciently dense uorescent signal ROIs can be traced by confocal microscopy using RevOPT and cryosectioning.

Discussion
The choice of microscopy technique for gut characterization relies on certain features of the signal of interest, including the scale, the required resolution, the need for staining, the sample preparation and its application in-vivo or ex-vivo.In studying the gut, common imaging techniques include confocal microscopy, light sheet uorescence microscopy and two-photon microscopy whose applications range from studying villous vascularisation 41 , structural integrity 30,42 , local in ammatory status 43 and microbial community dynamics 23 .The heterogeneity of gut tissue structure and the dynamic recruitment and tra cking of cells involved in gut health makes 3D microscopy particularly adapted for this setting.
In addition, volumetric imaging modalities are continually being developed in parallel with advanced image processing techniques [44][45][46] , as the weight of data grows rapidly with 3D imaging.
We describe a multi-scale and multi-modal pipeline for visualising the gut architecture and associated isolated lymphoid follicles (ILFs) at the scale of organs.We provide quanti cations of volumetric sizes and spatial distribution of ILFs in adult mice throughout centimetre lengths of mouse intestine.We leverage the 3-dimensional nature of OPT data to facilitate the observation of vascular networks in the submucosa, as shown by virtual unfolding.As proof-of-concept, we have evidenced the requirement of the microbiome for the maturation of ILFs, and during development.Finally, we have incorporated a technique -which we have called RevOPT -for higher-resolution imaging of selected ROIs.This bridges imaging of tissue at the organ and histology level, and allows quantitation at different scales.While this method requires specialized manipulation lasting up to three weeks, the gain in information and wide eld of view is attractive for studying the distribution and localisation of distinct cellular structures.
Tissue auto uorescence serves multiple roles in the interpretation of microscopy images.It can provide crucial contextualization for uorescent labels within tissues and facilitate the interpretation of functionality based on uorescent signals.In addition, the intestinal architecture observed by auto uorescence imaging provides a label-free method for the characterization of diverse parameters 47 which may be used as comprehensive measures of gut integrity and leakiness 48 .Such a technique could be applied to histopathologic scoring where structural deformation is symptomatic of disease.For example, coeliac disease (CD) is characterized by a destruction of the intestinal epithelium driven by gluten-activated in ammation, resulting in observable villous atrophy and lymphocytic in ltration of the epithelium, as shown recently in a novel mouse model of CD 49 .Current methods for the diagnosis of coeliac disease rely heavily on histological observations of prepared endoscopic biopsies, with a necessity for multiple collections due to non-homogeneous tissue alterations 50 .With the ability to accurately reconstruct mesoscale volumes, the presented optical projection tomography pipeline offers an alternative approach that maintains structural integrity whilst multiplying the eld of view available for diagnostic observation.
The gastrointestinal tract constitutes an essential site for crosstalk between the external environment and the host, and which dictates immune development 51,52 .Gut immunity is implicated in intestinal diseases, such as in ammatory bowel disease (IBD) 53 , and to systemic disorders ranging from metabolic diseases such as diabetes 54 to neurodevelopmental, neuroin ammatory, and neurodegenerative diseases [55][56][57][58] .Thus, imaging and characterizing gut-associated lymphoid structures (GALT) is important for understanding how immune development impacts health.Yet, the spatial distribution of immune structures in the gut is not well documented at the mesoscale.To our knowledge, our method is the rst to image GALTs in 3D at a centimetre scale, with subsequent high-resolution 2D ROI referencing.
In order to show the applicability of OPT to characterize the mesoscale organization of cell-types within tissues, we explored the development of GALTs in models where age and the microbiota are manipulated.
We stained the CD45 antigen that is found on hematopoietic cells 59 from which almost all immune cell types are derived 60 .CD45-rich regions identi ed as isolated lymphoid follicles (ILF) were found in the gut of a 30-week-old SPF C57BL6 mice.ILFs are a sub-category of gut-associated lymphoid tissues 40 whose functions are to limit contact between luminal microbiota and the epithelium via IgA secretion, and to sense epithelial breaching by bacteria and signal the need for phagocytosis to surrounding macrophages 61 .These are thought to depend on broblastic reticular cells (FRCs) and follicular dendritic cell (FDC)-like broblasts, to be seeded by Lymphoid Tissue inducer (LTi) cells, and to require microbiotainduced IL-25 and IL-23 62 .Yet the exact mechanism remains to be fully elucidated and their development characterized along long stretches of the intestine in 3D.For the rst time, we are able to measure the variation in volume and the spatial distribution of ILFs in 3D throughout an uninterrupted section of tissue at the millimetre scale.This technique may prove useful in the tissue-wide imaging and characterisation LTi cell clusters and their development into mature GALTs.Furthermore, with concatenated overlapping data acquisition, this is easily extended to the centimetre scale.
The age-and microbiome-dependence of the maturation and regulation of gut immune responses has become evident in recent years 63 .Initial exposure to a microbial environment during the neonatal period shapes the immune system throughout development 64 .In accordance with this, we nd a sparse immune signal in young SPF mice, with no discernible GALTs.In addition, the study of germ-free and gnotobiotic models has proven that the gut microbiota is necessary for the development of a mature and complete immune system 65 , with post-gnotobiotic colonization with commensal bacteria resulting in the acute induction of lymphoid tissue genesis 19 .Thus, the lack of isolated lymphoid follicles in the small intestine of old germ-free mice observed in OPT is indicative of the expected stunted immune germ-free phenotype.
We also demonstrate the traceability of ROIs between imaging modalities by selecting ILFs in OPT reconstructions and performing RevOPT and confocal microscopy.The distances measured by image processing and by tracking the number of sectioning depths leads to an accurate correlation of signal localization.With signal density quanti cation, we are able to determine a limit of detection for regions of interest in OPT scans that may become a benchmark for the selection of targets of interest for OPT imaging.RevOPT adds value to our pipeline as it addresses the need for microscopic analysis of biological landscapes whilst offering the opportunity to interpret signals at the mesoscale.
Collectively, we provide an imaging pipeline for versatile multi-modal imaging of the mouse intestine and its associated immune compartment.Furthermore, we demonstrate its ability to quantitatively characterise sparsely distributed structures throughout centimetre long segments of the intestine, in volumetric terms.Cut-free sections reduce the presence of common artefacts that distort the sample and impede large-scale histopathological interpretations.The virtual sections also simplify the registration of multiple imaging depths for 3D or concatenated 2D segmentation of regions of interest, as we have demonstrated here.Finally, we demonstrate the advantage of combining of OPT with RevOPT for the empirical selection of regions of interest for high-resolution downstream imaging.We believe the ease of implementation and the resulting possibilities of analysis in large volumes and at high resolution make the gutOPT pipeline an attractive method for preclinical characterization of gut tissues in mice.Its potential for implementation in whole human tissue biopsy imaging raises exciting prospects for clinical diagnostics.Thus, gutOPT addresses the need for a detailed yet holistic approach to understanding the complex physiological interactions involved in gut health and disease.

Animal handling
Speci c pathogen free C57BL/6J mice were purchased and housed at the École Polytechnique Fédérale de Lausanne (Switzerland) under speci c pathogen free conditions with ad libitum access to food and water, according to guidelines and regulations of the state of Vaud, Switzerland (authorization VD3448).Germ-free C57BL/6J mice were obtained from the Clean Mouse Facility, University of Bern (Switzerland).Germ-free status was routinely monitored by culture-dependent and -independent methods and con rmed to be microbial-free.Experiments were performed in accordance with regulations approved by the ethical and veterinary committee of the Canton of Vaud, Switzerland.As was described previously 29 , mice were deeply anesthetised by intra-peritoneal injection of 50mg/kg sodium pentobarbital prior to a transcardiac perfusion of 10ml heparinised PBS (5 I.U./mlLiquemin).Tissues were xed by perfusing with 10ml of freshly-prepared 4% paraformaldehyde (CAS 30525-89-4, Carl Roth AG 0964.1) and an overnight postxation step at 4°C.

Sample preparation
All the following steps took place in the dark.Samples were washed for 30 minutes in PBS after the overnight xation.A 45-minute step-wise dehydration in methanol precedes overnight auto uorescence quenching in a 2:1:3 ratio solution of MetOH:DMSO:H 2 O 2 overnight at room temperature.The samples are washed twice in pure MetOH in preparation for three freeze-thaw cycles between -80°C and room temperature (1 hour and 30 minute cycles respectively) in order to permeabilize the tissue before antibody-mediated staining.A step-wise rehydration to TBS-Tween prepares the samples for antibodymediated staining of targets.This begins with blocking for 24h, is followed by a primary antibody incubation for 48h and a 24h washing step and ends with a 48h incubation in a secondary antibody and a nal 24h washing step.To stain immune cells, we used a rat anti-mouse CD45 monoclonal antibody conjugated to APC (BioLegend 147708) and a goat anti-rat IgG (H+L) Alexa Fluor 647 (Invitrogen A21247).

Microscopy
OPT Detailed descriptions of the optical setup are available at Schmidt, C. et al. ( 2021) 29 .Multi-channel sets of projections were acquired rst with a 625nm LED ( ltered within a 595 nm and 645 nm range, AT620/50x, Chroma) for the CD45 positive signal followed by a 415nm LED (400-440nm lter range, AT420/40x, Chroma) to acquire tissue auto uorescence.In each channel, either 400 or 1200 projections were acquired and subsequently reconstructed using our version of the ltered back-projection algorithm 29 .Reconstructed stacks were cropped in ImageJ 66 and visualised in 3D.

Reverse-OPT
After OPT imaging, the samples were dehydrated in pure methanol and rehydrated in PBS for 24h each.At this stage, it is possible to carefully track regions of interest identi ed in the 3D OPT images to be speci cally observed using other imaging modalities downstream.The samples were extracted from the agarose moulds and frozen in optimal cutting temperature (OCT) medium on dry ice.Using a Leica CM3050S cryostat, 25 micrometre-thick sections were collected and mounted on coated glass slides (Epredia™ J1800AMNZ).The sections were counterstained with DAPI (Thermo sher D1306) at a concentration of 5µg/ml for 10 minutes with a 5 minute pre-and post-wash with 0.3% Triton-X100 in PBS.

Confocal microscopy
Confocal microscopy was performed using a Leica SP8 inverted microscope, producing two-channel images encompassing the CD45-positive and nuclei signals.Sequential acquisition began with the AlexaFluor647 channel followed by the DAPI channel.Exposure times were determined according to live observation of pixel intensities in order to avoid over-exposure of the tissue.Tiled acquisitions of wholegut sections were performed using the automated tile function in the LAS-X software.

Virtual unfolding
The image processing pipeline for virtual unfolding was inspired by previous reports 30 applied using Fiji and is available in the form of a macro algorithm.All steps outlined below are applied to all of the sections within the ltered-back-projection stacks.First, the gut tissue is segmented from the lumen and surrounding background and a mask is created.For each section, the segmented tissue outline is added to the ROI manager.The centroid coordinates within the tissue outline are calculated and used as an origin for the identi cation of a sectioning origin at a 45° angle from the centre.From this point, we interpolate a polygon shape to draw the line along which the unfolding takes place.A stack of the straightened images is produced and re-sliced orthogonally to create the unfolded image whose eld of view include the entire surface of the sample, spanning the lumen in the innermost slice to the outermost layers of the gut.From this image, the apex of each villus is identi ed by applying a Laplacian lter and extracting local maxima ROIs, whose density can then be calculated within a de ned area.

ILF segmentation
Quanti able characteristics were extracted from FBP reconstructions showing the isolated lymphoid follicles using the surface tool in Imaris.The smoothing of surface areas was set to 2µm and thresholding based on absolute intensity, whose values were set visually by the user.Larger structures were segmented by implementing the "number of voxels" lter.Volume and distance statistics were exported in csv format for data plotting.

Declarations
Funding.H2020 Framework Program of the European Union (grant no 686271) and Innosuisse (grant no 31434.1IP-ICT).

Figure 3 Multi
Figure 3