Exploring and illustrating the mouse embryo: virtual objects to think and create with

The teaching, learning, communication, and practice of Developmental Biology require interested parties to be at ease with the considerable spatial complexity of the embryo, and with its evolution over time as it undergoes morphogenesis. In practice, the four dimensionality of embryonic development (space and time) calls upon strong visual-spatial literacy and mental manipulation skills, generally expected to be innate or to come through experience. Yet it has been argued that Developmental Biology suffers the most from available traditional media of communication and representation. To date, few resources exist to engage with the embryo in its 3D and 4D aspects, to communicate such aspects in one’s work, and to facilitate their exploration in the absence of live observations. I here provide a collection of readily-usable volumetric models for all tissues and stages of mouse peri-implantation development as extracted from the eMouse Atlas Project (E5.0 to E9.0), as well as custom-made models of all pre-implantation stages (E0 to E4.0). These models have been converted to a commonly used 3D format (.stl), and are provided in ready-made files for digital exploration and illustration. Further provided is a step-by-step walkthrough on how to practically use these models for exploration and illustration using the free and open source 3D creation suite Blender. I finally outline possible further uses of these very models in outreach initiatives of varying levels, virtual and augmented reality applications, and 3D printing.

. The vicious cycle of scientific illustration. Blue does not have access to real mouse embryos, and thus their entire understanding depends on the mental reconstruction of pre-existing representations from the literature. Blue's output representations, regardless of the fidelity with which they match Blue's mental reconstructions, will in turn serve as the basis of understanding for other scientist (e.g. Pink, and Blue themselves). Blue is thus entirely dependent on the output of direct observers such as Green, and has to wait for Green to draw the specific scenario needed.
How then to facilitate and democratise both the visualisation and the illustrations of mouse embryos? An important resource 64 has been, and still remains, the Edinburgh Mouse Atlas Project (EMAP), the freely-accessible, annotated, online collection 65 of 3D volumetric data hosted at https://www.emouseatlas.org/emap/ema/home.php, covering mouse development from 66 pre-implantation, to gastrulation, to organogenesis and late post-implantation (Armit et al., 2017;Richardson et al., 2013).

67
3D models are easily navigable and allow users to define cutting planes as to see virtual histological sections at any desired 68 angle. The value of such resource is maybe understated, but here are on-demand, user-explorable, customisable, 3D (+ 69 timepoints) visualisations of mouse embryos at almost all stages of development. Clearly, this is a complete subversion of the 70 normal illustration-spectator dependant relationship built by traditional developmental visualisation, and a powerful enabler of 71 understanding (cfr. Figure 2) . Crucial in this, is the role played by the actual data-format (i.e. a 3D object in space) and how it 72 allows to completely evade the two-dimensional constraints of the printed page. Put very simply, an interactive, explorable 3D 73 model condenses the informational potential of as many 2D pictures as there are angles to visualise that model. With the evolution of our formats of scholarly communication, published data is also slowly starting to take life out of the 76 page. Concomitant with the increasing use of experimental strategies outputting volumetric microscopy data, and with their 77 application to the mouse embryo, such data is increasingly being presented as videos accompanying the online version of 78 manuscripts. In these videos, the models are usually animated to rotate around their main axes, they might toggle specific 79 structures on and off to reveal internal structure, or show internal cross-sections. Here is the dimensionality, here is the change 80 over time, features so intimately associated with developmental biology (Hardin, 2008). The models live and are communicated 81 in their 3D environment, if they are not user-explorable they are -at least -explored (cfr. Figure 2) . Even if just provided 82 in their printed version, 3D models still remain potent embryonic representations. One could say that the understanding of 83 e.g. tailbud structure, embryonic heart development, and uterine architecture is almost immediate when shown, respectively, 84 as in (Arora et al., 2016;Dias et al., 2020;Ivanovitch et al., 2017), even without the animated videos that may accompany 85 such publications. Still, access to these models and thus the ability to explore them, is entirely dependent on the sharing Fig. 2. The power of 3D models. Direct observers like Green do not just illustrate the object from a specific point of view, but also decide to provide a 3D model of the original object (here, a boat). Blue can now observe the object from the specific angle they needed (becomes Green by proxy): the cycle of dependence is bypassed. Notice however that Blue cannot necessarily use the model provided by Green (e.g. only available as a video, or as 2.5D) when communicating with Pink. The cycle of dependence is broken at the understanding/reconstruction step, but not at the communication step.
illustration softwares. These conversion steps, and the digital reuse of 3D objects, requires skills that -while not difficult -are 90 yet not mainstream compared e.g. to the relative familiarity one is nowadays expected to have for 2D illustration software.

92
In the light of the "illustration -> understanding -> new illustration" framework outlined above , we finally also have to reflect 93 on the usability of our visualisations. That is, to what extent do they allow the reader to transition from passive "visualiser" and 94 "learner", to active user, "illustrator", "teacher", "communicator"? To what extent do they empower them? To what extent to 95 they break this cycle of dependency? Is it possible to imagine a framework where illustrations not only catalyse understanding, 96 but they simultaneously facilitate creative experimentation? Where the lifecycle of an illustration/model does not end upon 97 publication but gets new life in the hands of the reader? There is an untapped value in 3D models, probably uniquely so for the 98 mouse developmental biology community given the existence of repositories such as EMAP. Tapping into it only requires easy 99 access to the models themselves and the technical know-how on how to use and explore such models digitally. And here are 100 scientist that will be able to explore these models on their own. Scientists and students that will be able to incorporate them 101 into their own illustrations. Outreach officers that will be able to deploy them in e.g. virtual or augmented reality applications. Green not only provides models, but makes them accessible too (e.g. copies of the model can be downloaded and manipulated). Blue can actively explore the object and is even more likely to find the point of view they needed. Pink developed the technical skills to use the model itself, and can now transform/reinterpret it, 3D print it, or use it in new visualisations. All of these elements feed back into the visual data formats available to the community, and thus further catalyse understanding. I here provide a collection of readily-usable volumetric models for all tissues and stages of mouse peri-implantation develop-105 ment as extracted from the eMouse Atlas Project (E5.0 to E9.0), as well as custom-made models of all pre-implantation stages 106 (E0 to E4.0). These models have been converted to a commonly used 3D format (.stl), and are provided in ready-made files for 107 digital exploration and illustration. I further provide a step-by-step walkthrough on how to practically use these models using 108 the free and open source 3D creation suite Blender. I finally outline possible further uses of these very models in outreach 109 initiatives of varying levels, virtual and augmented reality applications, and 3D printing. I hope these resources and instruc-110 tions may add to those already available to the mouse developmental biology community, serve as a soft introduction to more 111 members of our community to the world of 3D illustration, and encourage creative exploration and experimentation in such a 112 visually rich and stimulating field that is that of embryology.

113
A collection of volumetric models of embryonic stages.

115
An overview of all 16 models available for download is provided in Figure 4. These models cover Theiler Stage (TS) 01 116 to TS14, and include 8 custom-made models (covering TS01 to TS06, or E0.5 to E4.5), as well as 8 models assembled 117 from EMAP data (covering TS07 to TS14, or E5.0 to E9.0). Also provided are two additional models corresponding to  stage. These are provided as .stl format, as to be readily used not only for illustration, but also in 3D printing, gaming, and 127 virtual/augmented reality environments. Finally, an additional .blend file with the full embryonic collection is also provided (as Vianello | Exploring and illustrating the mouse embryo bioRχiv | 5 Fig. 5. List of pre-made materials available for download. By default, 3D models will be displayed as grayish-white, but the visual appearance of a model can be controlled by applying materials/shaders to it. To facilitate the process, the above materials can be downloaded and used. These correspond to all the main colours, in a neutral and in a glossy ("wet") version. "Special" materials include one made to mimic the surface of an epithelium, one showing mesenchymal cells migrating above it, and one making the structure transparent. Technical details on how these materials were created are provided in the Methods section.
Use of the models for exploration/understanding.
As mentioned in the introduction, one of the biggest advantages of having access to 3D models is that they allow the user to 132 take control over the exploration of these very models (as per the paradigm illustrated in Figure 3). A big part of the models 133 provided here have been long accessible and explorable on the EMAP portal, but they come in a dedicated file format to be 134 open and read with the Java-based viewer JAtlasViewer. Through this viewer, the user can relate 3D structure and histological 135 cross-sections at any cutting plane defined in the application. At the cost of losing the possibility to view cross-section 136 histology (which would however still be available on the EMAP website), Blender provides an alternative way of manipulating 137 and exploring 3D objects, while also allowing easy customisation of the models, rendering to 2.5D illustrations, the creation of 138 scenes with multiple models, sculpting, and more advanced functionalities. Furthermore, and for 3D models that do not benefit 139 from curated interactive visualisation platforms such as those of EMAP, Blender simply provides an easy-to use interaction 140 and exploration platform.

159
An alternative way to explore the embryo models provided is to move and rotate the model itself, and to move structures in and G (for "grab", followed by X, Y, or Z to lock the movement along these axes), and R (for "rotate", followed by X, Y, or Z to 165 lock the rotation around these axes). The same can be done on individual components, rather than on the whole embryo, by  While these models have been here made available as pre-prepared .blend file, one can exploit these same basic functions to 182 explore any type of volume data (.obj, .wrl, .stl; imported in a Blender scene via File>Import). This has been e.g. recently 183 exploited for 3D exploration of cell migration tracking data in Samal et al. (2020), and can also be applied to explore .obj/.stl 184 meshes generated from e.g. light-sheet imaging.
Use of the models for illustration. To explain how the models provided here can be used for illustration, we will take a simple case study: that of a scientist one can repeat the Append process to add each of the following embryonic stages (Figure 8.3). As explained before, specific 211 structures and layers can be then toggled off (eye icon) to expose the structures one is interested in (here, the endoderm) ( Figure   212 8.4).

Fig. 8. Importing models from other files by using File>Append>Collection
Once the scene is set with all the models in the right place, it is time to add shaders/materials. Materials determine the final 214 appearance (in its simplest form, the colour) of the models in the output illustration, and as such have a heavy influence on its 215 success. To get an idea of what the final image will look like one can toggle the Rendered Viewport Shading icon at the top 216 right of the viewport (Figure 9.2; allow some processing time). Since all models provided here come with a generic grayish 217 shader, this is indeed how the models will look like (Figure 9.2). An important note is that, depending on where you placed 218 your models into the scene, there might nor be enough light to actually see the real final colour of the model. This case is 219 shown in Figure 9.3B, and corrected in Figure 9.3C by providing more light to the rightmost model. In Blender, light is if clicking directly on the objects, Shift+RightClick), press Shift+D to duplicate, and then move them to your next model (in 224 Figure 9.2 you can indeed see how additional lights have been placed all along the lineup of models).
As a final illustration technique, a developmental biologist might often need to show cross-sections or cut-outs of a specific 267 tissue: that is, not toggle the tissue off view completely, but just remove a section of it as to show underlying structures. This 268 is achieved in Blender through the use of so-called Boolean modifiers, which essentially allow to substract/add shapes from/to 269 one another. A cube or a parallelepiped can thus be intersected with our tissue of interest and the intersection deleted out.

270
Upon removal of the cube, the tissue is left open: a process illustrated in Figure 12.

272
As a first step, one needs to create the shape to be subtracted. Making sure that one is in the topmost hierarchy (Scene 273 Collection), add a cube to the scene by selecting Add>Mesh>Cube (keyboard shortcut Shift+A) (Figure 12.1). To scale the 274 cube right-click on it (Figure 12.2) and select the Scale icon on the left, or use keyboard shortcut "S", followed by X, Y, or 275 Z to lock the scaling to either of the three axes. Since we here want to cut out the visceral endoderm both in the embryonic 276 and in the extraembryonic region, we make a parallelepiped that is as tall as the embryo (Figure 12.3). Once satisfied with the 277 shape, move it as to intersect your model over the regions you want to carve out (Figure 12.4), select the tissue that needs to 278 be cut (here starting with the embryonic visceral endoderm), and select the wrench icon of the bottom right menu (modifiers 279 tab, Figure 12.5). Under "Add Modifier", select "Boolean", and then make sure that the Operation is set to "Difference", and 280 the Object field displays the shape you want to subtract (our parallelepiped is called "Cube" in this example; Figure 12.5).

281
Click on "Apply" and then repeat the whole process for the extraembryonic visceral endoderm. Once both tissues have been 282 processed, delete the parallelepiped by pressing the keyboard shortcut "X" (Figure 12.6) to reveal what is left of the model (Fig-283 ure 12.7). We can now clearly see epiblast and primitive streak under this newly created window through the visceral endoderm.

285
Because the models provided here are not "solid" (they are actually hollow, thin-layered envelopes), the kind of editing de-286 scribed above might reveal exposed holes in the cross-sections (Figure 12.8A, the cut visceral endoderm is hollow). These are this gives the illusion of a solid model (Figure 12.8F), which can now be processed for illustration as described above. Vianello | Exploring and illustrating the mouse embryo bioRχiv | 15 Texture" node was used instead of a "Voronoi Texture" (fBM, Scale: 18.2, Detail: 4.6, Dimension: 1.614, Lacunarity: 10.5, Offse: 0.5, Gain: 0). Altering the Scale value of this node will allow to change the size of the cells as a function of the size of 378 the tissue of interest. The "Bump" node had values= Invert: unchecked, Strength: 1, Distance: 0.2. The "transparent" material 379 was created by mixing ("Mix Shader") a "Principled BSDF" node set as for Base and Neutral Materials (Base Color: FFFFFF) 380 and a "Transparent BSDF" node (Color: FFFFFF). The "Fac" value of the Mix Shader was set at 0.225 (for the transparent 381 input).

382
Preparation of pre-implantation models. Models for the 1-, 2-, and 4-cell stages (TS01.blend, TS02-2cell.blend, TS02-383 4cell.blend) were created from simple sphere shapes (Add>Mesh>UV Sphere). Models for the 8-cell stage, morula, early 384 blastocyst, late blastocyst, and implanting blastocyst stages (TS03-8cell.blend, TS03-16cell.blend, TS04.blend, TS05.blend, 385 TS06.blend) were created by using the Particle Emitter function in Blender. Briefly, an object of the desired shape (a sphere 386 for most models; a sculpted "implanting" mesh for TS06.blend) is added (Add>Mesh>UV Sphere) and selected as an "emitter" 387 object ("Particles" properties tab). Another object (the 1-cell model) is imported (File>Append) and selected as the "emitted" 388 object. If the particle settings of the emitter object is set as "Hair", running the simulation (keyboard shortcut: spacebar) will 389 lead to individual cells emerging out of (and thus covering) the faces of the object. The entire system was then saved as a single 390 model by applying the particle modifier ("Convert") of the emitter object. The entire setup used to create each of the models is 391 included in the .blend files provided, for reference and hidden from view in a dedicated collection called "Emitter_system".

392
The open versions of the early and mid-blastocyst models (TS04_HALF.blend, TS05_HALF.blend) were generated as above, 393 but the emitter object used was a hollowed out hemisphere for the surface that will be covered by trophectoderm cells, and a 394 separate shape for the surface that will be covered by inner cell mass cells (TS04_HALF.blend). For the latter particle system, 395 the "emitted" object was a collection of two different cell objects: a sphere to represent epiblast cells, and another sphere to 396 represent primitive endoderm cells. To instead avoiding having intermingled cells in TS05.blend, two separate inner emitters 397 were used (each producing one type of cell).