B. fissidens.

• Concreted anterior part of the skull with part of the maxilla and nasal bones and the left A3, A4, P4, M1 and M2 in their alveolus. The left A2 and right A3 are also present, but isolated and stuck to the palate. The part where M3 is inserted is missing in the maxilla. It is from level TE9c of Sima del Elefante. MPZ 2018/467.
• Fragment of maxilla with the right I, A1 and A2, from level TE9c. MPZ 2018/468.
• Fragment of maxilla with the left P4, M1, M2 and M3. The maxilla is broken in the part where M3 is inserted, but all the fragments are together. It is from level TE9a. MPZ 2005–491.
• Right mandible. Complete with all the teeth and partially concreted. It is from level TE9c. MPZ 2018/469.

D. glyphodon.

• Complete right mandible with all the teeth, from level TD6. MPZ 2010–579.
• Fragment of maxilla with right P4, M1, M2 and M3 and the posterior part of the palate preserved, from level TD5E. MPZ 2010–27.
• Concreted maxilla with the left I, A1, A2, A3, P4 and M1 and the right A1, A2, A3, P4, M1 and M2. The insertions of M3 are preserved. Nasals are not preserved. It is from level TD5BE. MPZ 2010–512.
• Maxilla and nasals concreted. With the left I, A1, A2, A3, P4 and M1 and right A2, A3, P4, m1 and M2. It is from level TD5E. MPZ 2010–201.

Extant species.

In addition, remains of extant shrews were used to complete the reconstruction of the skulls and to compare the final models of different species. The reason to choose them is that they are complete. In this work they come from owl pellets because of ease of access, as this kind of remains is available in museum collections:

• One skull and the two hemimandibles corresponding to the same individual of B. brevicauda, from North America. Material purchased already prepared. Universidad de Zaragoza.
• One skull and one hemimandible of Neomys fodiens, obtained from owl pellets in the Pyrenees. Material from the Instituto Pirenaico de Ecología (IPE)
• One skull and one hemimandible of Sorex coronatus, obtained from owl pellets in the Pyrenees. Material from the Instituto Pirenaico de Ecología (IPE)
• One skull and one hemimandible of Crocidura russula, obtained from owl pellets in the Pyrenees. Material from the Instituto Pirenaico de Ecología (IPE).

Recovery of the samples.

The fossils in this work are from the archaeological sites of Sierra de Atapuerca (Burgos, Spain). Specifically, they are from the cave sites of Gran Dolina and Sima del Elefante (Burgos, Spain) and were collected in different excavation campaigns from 1996 to 2015. The microfossils were recovered using the method of washing and sieving sediments [42]. Each sample corresponds to a volume of sediment of 1m x 1m x 10cm and is labelled with the exact level and spatial position in the site. In the course of each campaign the sediment was washed and sieved using sieve towers with decreasing meshes (5, 2 and 0.5 mm); water was extracted by water-pump from the River Arlanzón in order to remove the clay that covers the fossils. The resulting concentrate is formed of small clasts and fossils of small vertebrates or fragments of large vertebrates. Finally, the identifiable fossils were picked out and separated with soft forceps. We selected the fragments of shrews detailed in the “Material” section from this mixture of fossils collected in the field and from material stored in the Museo de Ciencias Naturales de la Universidad de Zaragoza (MCNUZ) [38] because of previous studies that had also selected the material from the concentrates.

Having described the selection and collection of the material, we now explain the protocol designed and followed in the reconstruction of both skulls:

Digitalization.

Prior to digitalization, images of all the elements were taken in the laboratory with an Olympus SZ61 trinocular microscope and an Olympus Soft Imaging Solutions LC20 camera, and measured with the software TpsDig v.2.17 [43].

The fossils and present-day items were scanned in the Centro Nacional para la Investigación de la Evolución Humana (CENIEH, Burgos) using a Phoenix v/tome/x s by GE Measurement & Control. The technique used was the microCT scan.

The voxel size was isometric, with the same value in X, Y and Z. It was 9.49978 μm in all samples except in B. fissidens, where it was 7.99958 μm, and MPZ 2010–201, where it was 9.50016 μm. The pixel size in X and Y was always 0.2 μm. The voltage used was 100 kV, and the current 140 ρA.

Beforehand, the pieces were set up for scanning. Some of them were put inside small plastic containers as centrifuge tubes and wrapped in plastic (HDPE) to avoid interference between the container and the piece, and make sure that the piece would not move inside the container during scanning, since the scanning involves the rotation of the piece. The biggest samples, such as the B. brevicauda skull, were fastened inside an expanded polystyrene container that had previously been adapted to the skull morphology. Some of the samples had been stored in a museum for years, glued with a reusable putty-like adhesive that had hardened during this time.

The computer that was used to make the reconstructions was a commercial laptop, so the process of creation is adapted to its characteristics: Windows 10 system with processor Intel Core i7-7700HQ CPU 2.80GHz, 16 GB RAM memory, 64-bit system, and the graphic card is NVIDIA GeForce GTX 1050 Ti, with 12 GB of memory.

The data were exported as TIFF images. They were treated with the software Fiji (ImageJ) [44–46] to optimize the size of the files. The quality was reduced from 16 bits to 8 bits, the work area was cut and adapted to the elements, and then the images without bony elements were deleted. The images were also rescaled.

With the software XnView v.2.38 [47], the TIFF format was converted to bmp because TIFF is not accepted in the software here used for the 3D reconstruction.

Creation of the 3D models.

The 3D models of each piece were created using the free software SPIERS [48, 49], with the application SPIERSEdit v.2.20.

No downsampling was applied to the models of fossils, but a downsampling of four was applied to the skulls in the recent material in order to reduce the size of the files and facilitate work with a commercial laptop. The reason we decided not to apply downsampling to the fossils, which are the objects of greatest interest, was that it reduces the resolution of the model, but we did apply it to the extant species, which are used for comparison [50].

To select the area of the images that had to be reconstructed, a different threshold was applied to each piece due to the varying range of greys obtained in the scans. The threshold was applied trying to exclude most of the external elements such as the container and protection of the pieces, the concretions, the hardened adhesive stuck to some of the pieces, and the glue that was reinforcing the pieces of B. brevicauda. As it was not possible to remove some of these elements completely (e.g. the concretions and putty-like adhesive) because their grey contrast coincided partially with that of the bone or enamel, we also tried to select a threshold that did not exclude the material under study.

After this, the Mask mode of SPIERSEdit was used, and different masks were used to separate elements of each piece. Both the teeth of the complete maxillae and those that belonged to incomplete maxillae that were needed for the reconstruction were separated.

If the external bones were broken, they were also separated in order to put them in their correct position later. Using masks, the external concretions and putty adhesive were avoided or marked. For the non-fossil material, the elements were not separated. Each masked element was converted into an object and exported to SPIERSview v.2.20. From SPIERSview the objects were exported into Virtual Anatomy eXtensible Markup Language/Standard Triangle Language (VAXML/STL). The parts unnecessary for the final model were not exported.

Each object was first importedinto MeshLab v2016.12 [51]. There we applied the filter “Invert Faces”. Then the objects were exported into PLY (Polygon) file format.

To continue with the treatment of the mesh, the PLY files were then imported into Meshmixer [52]. There we used the “Inspector” analysis tool and the “Auto Repair” option to remove the islands corresponding to the dirt on the model.

It was necessary to smooth the models using manual smoothing brushes. The smoothing affects the model to different degrees depending on the structure, so it is important to be careful in performing this step [50]. The “ShrinkSmooth” brush was used over the whole surface of the objects in order to delete the “step” effect produced by the reconstruction of the slices, which is a consequence of the method of digitalization and reconstruction and thus does not represent real structures. “RobustSmooth” was used if there was any irregularity associated with extra material such as concretion stuck to the pieces. If the imperfections were related to cracks or to missing material, they were treated using “BubbleSmooth”. The size and strength values of the brushes were low, adapted to the shape and size of the triangles so as not to distort the general shape of the pieces. The effects of the application of smoothing on different types of structures are explained in [50]. We show the effects produced on some elements in the figures in S1 Fig.

Finally, the number of triangles was reduced using the “Reduce” brush, which is a decimation filter. The strength and size of the brush was also adapted to the piece and the surface, being low in the teeth, especially in details such as crests and cusps, and higher in the bones, in the flatter surfaces. This was done with the “Wireframe” view to ensure that only the unnecessary triangles were deleted. It is important to be careful here in order not to lose structures [50].

Formation of the general models.

As each of the separated elements of each fossil was saved in the same position, they were opened in the same file. Firstly, the broken parts of bones such as the insertions for m3 were reconstructed using the “Transform” tool to collocate the fragments in their correct position with respect to the main fragment. The teeth were also relocated in the cases where they had popped out from their alveolus, totally or partially.

Later, with all the elements selected at the same time, the “Transform” tool was used again to rotate the elements in order to orientate them in the same way, that is:

• The antero-posterior anatomical axis corresponds with the front-back direction of Meshmixer, or the X axis.
• The lateral anatomical axis corresponds with the right-left direction of Meshmixer, or the Y axis.
• The dorso-ventral anatomical axis corresponds with the top-bottom direction of Meshmixer, or the Z axis.

After this, the elements were scaled to form the final model. In B. fissidens the most complete maxilla was used as a reference, whereas in Dolinasorex it was one of the snouts without the complete nasals that was used. The other models were added to the file and scaled using the length of the teeth “repeated” in the models, i.e. the length of A2 and the width of M2 for the upper elements. For this scaling, the tools used were the “Measure” selection in “Analysis-Units and Dimensions” and then, after calculating the scale value, “Edit-Transform-Scale”.

For the lower dentition, as the teeth were not separated in the initial model, a different process was applied to correct the position of the m3, which was also rather detached from its alveolus. The m3 was selected in the model by combining the “Lasso” tool, where it was possible to select all the structures of the tooth, and the “Brush” selection tool. Then it was separated and transformed so as to be put in its position. The scale of the mandible with respect to the upper dentition was calculated by measuring the distances between the cusps of the lower teeth and making them equal to the distances between the points where they occlude in the upper dentition.

Due to the symmetry of vertebrates, only the elements of one side were selected, although the most important factor was the preservation of the elements rather than whether they were all from the same side. For this reason it was necessary to mirror the pieces [7,15], which means duplicating them with the opposite orientation with respect to a plane of symmetry. Here the plane used was the sagittal plane.

The first step was to select the better-preserved half of the maxilla and nasals. In Beremendia it was the left half that was selected, but in D. glyphodon the anterior part of the left side was broken whereas the posterior part of the right side was missing, so the D. glyphodon maxilla was mirrored by sectors. To undertake this process, the models were cut using the “Plane cut” editing tool. The cutting plane was located in the sagittal plane of B. fissidens, and in the case of D. glyphodon parallel to this plane at two different points to avoid the higher quantity of imperfections, and also perpendicular to the coronal plane in the position of the premolars. The type of cut used was the “Slice (Keep Both)” option. To separate the slices as different objects, it was necessary to use the “Separate Shells” edit option. The interesting parts were then mirrored with the “Mirror” edit option in the correct position, taking the real object as a guide. Afterwards, the real objects that we wanted to replace were deleted. With the elements that had not been cut exactly in the sagittal plane but parallel to it, after the mirroring and separation it was necessary to transform them to set the correct distance to the middle.

With other pieces such as the teeth or mandible, the mirror was applied with respect to the sagittal plane, bearing in mind the correct insertion of teeth in the alveolus and the occlusion of upper and lower teeth.

Once this was done, the whole model was scaled using the height of the mandible. This was 6.15 mm for B. fissidens, a small shrew whose size lies within the range of measurements for Sima del Elefante and is rather smaller than the mean for Europe, 6.28mm [34]. The measurement for D. glyphodon was 7.16 mm, which coincides exactly with the mean from level TD6 of Gran Dolina [32].

Reconstruction using a complete skull of a related species.

At this point, the models are the most complete skulls that can be obtained from real fossils of these species. However, these models lack the posterior part of the skull, the whole braincase, including the temporomandibular joint. In order to gain an idea of what this part might be like, the next step was to attach the posterior part of B. brevicauda to the models of the fossils. In the “Materials” section, we have explained why we selected a skull of this species. To this end, the model of the B. brevicauda skull was imported into the Meshmixer file of the fossil reconstruction.

The model of the skull of B. brevicauda was scaled using the length of the maxilla and superimposed upon the models of B. fissidens and D. glyphodon.

In B. fissidens, when the length was scaled, the width of the maxilla and height of the snout did not coincide, so it was necessary to add a non-uniform scale of 110% to the Y and Z axes. In the case of D. glyphodon, by contrast, this latter transformation was not necessary.

After making the maxilla coincide in size and shape, the only part that could be inferred, at least in position, was the glenoid fossa. This is because one of the parts of the temporomandibular articulation, namely the condyles of the mandible, is oriented in the position of occlusion. In both cases, the glenoid fossa had to be extended outwards in a lateral direction, because in the fossil species the mandibles are wider in form and position than in B. brevicauda. As the transformation is not uniform, the transformation was here carried out using the “Drag” brush, dragging the whole of the articulation and the surrounding part to make it coincide with the mandible. In D. glyphodon it was also dragged in an anterior direction because the mandible is proportionally shorter in the posterior part.

The process is shown in Figs 2 and 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Reconstruction of Beremendia fissidens.

A. Separation of the elements of the main fossil, lower view. B. Upper view of the maxilla used. C. M3 insertion part reconstructed. D. Frontal view of the original maxilla with teeth mirrored. E. Lower view of the reconstructed skull. F. Oblique view of the skull with the mandibles mirrored and collocated. G. Superimposition of the skull of B. brevicauda. H. Final reconstruction with the adapted skull of B. brevicauda. Different colours indicate different 3D objects.

https://doi.org/10.1371/journal.pone.0213174.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Steps in the division and reconstruction of Dolinasorex glyphodon.

A. Reconstructed part of one skull, divided into the sectors that were subsequently mirrored. B-C. Upper and lower views of the reconstructed snout. D. Reconstruction with the snout and mandibles. E. Superimposition of the snout with complete nasals. F. Superimposition of the skull of B. brevicauda. The part with the worst adjustment is marked. G. Final reconstruction with the adapted skull of B. brevicauda. Different colours indicate different 3D objects.

https://doi.org/10.1371/journal.pone.0213174.g003

The reconstructions of the extant shrews were performed in a similar way, but mirroring only the mandible needed and without using an adapted B. brevicauda skull.

By this point we have obtained the two reconstructions that were the objective of the work. We also have four models of extant shrews to compare them with. The following section seeks to show some of the possibilities for working with the 3D models obtained from fragments, specifically in the study of these soricids.

The entire protocol to make the reconstructions is summarized in Fig 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Steps.

Flow chart that summarizes the main steps of the protocol to make a reconstruction.

https://doi.org/10.1371/journal.pone.0213174.g004

Example of an application of the reconstructions: Measurements.

Using the reconstructions, we took measurements of the anterior part of the skulls in order to compare the different species and characterise the anterior part of the skull of shrews. These measurements were taken with the “Analysis-Dimensions” tool in Meshmixer, except for the angle between the mandibles, which was measured in the frontal projection using TpsDig v.2.17 [43]. The measurements are modifications of those used in [53,54], which are shown in Fig 5. They are the rostral length (RL), distance from the palatine foramen to the anterior end (PF), length of the upper dental series (UTL), length of the unicuspid series (UL), length from P4 to M3 (P4-M3), infraorbital width (OW), interorbital width (IOW), anterior width in A2 (AW), width of M2 (WM2), zygomatic width (ZW), height of the snout at M1 (HM1), height of the snout in the posterior part of the palate (Hpost), height of the snout at A2 (HA2), coronoid separation (CW), postglenoid width, direct or inferred from the mandibular condyles (PGW), height of the coronoid process of the mandible (Hmand), and the angle formed by the two mandibles (αmand). Note that the measurements were selected to measure only the parts obtained with fossils, not the parts reconstructed with the B. brevicauda skull.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Measurements taken in this study for each species.

Measurements taken in this study. Some of them are modified from [53,54].

https://doi.org/10.1371/journal.pone.0213174.g005

In addition, in order to make the measurements comparable among the different species, each one was divided by the rostral length of each reconstruction.

Measurements.

Taking RL as the reference to infer the size of the snout, the biggest snouts belong to D. glyphodon, followed by B. fissidens, as expected given that they are the “giant shrews”. B. brevicauda and N. fodiens have very similar RL, N. fodiens being slightly bigger, and the smallest are C. russula and S. coronatus. These data are in accordance with Hmand, which also indicates that the biggest snout belongs to D. glyphodon. According to these measurements, N. fodiens is smaller than B. brevicauda, which is consistent with their general size if inferred from weight [60]. RL is lower in B. brevicauda as a consequence of it being a pug-nosed shrew. It is important to bear in mind that the measurements give information only from the temporomandibular articulation to the proximal end, which corresponds to the incisors. The angle formed by the mandibles changes from the base to the top in B. fissidens and D. glyphodon. It is most heterogeneous and has the highest angle in B. fissidens and the lowest in N. fodiens.

As the species differ in size, the measurements have been divided by RL for comparison with one another, since this is the longest measurement of the bone. As the reference is to length, the proportions refer to the proportional width and height of the snout.

The main difference between B. fissidens and the other species is that the snout is more elevated than in the others, with HA2, HM1 and Hpost showing the highest values, giving a more vertical face to this species. These measurements are also variable, and there is a significant difference from the anterior to the posterior part, so the snout becomes very high in a short distance.

B. brevicauda stands out due to the relative width of the snout, all the relevant measurements being the relatively highest among the species studied (OW, IOW, AW, ZW, CW and PGW). CW and PGW, however, are very close to Beremendia, which is also very wide in the posterior part.

The main characteristic of C. russula is the high relative length of the dental series, which occurs because the unicuspids are not parallel to the rostral length and are bigger than in the other species. Another characteristic is that the snout is wider in the centre than in the posterior part.

N. fodiens is important because it has the lowest relative values for most of the parameters. It has the lowest values in the relative length of the tooth series (UL/RL, UTL/RL, P4-M3/RL), probably due to the similar shape of the tooth profile and the total length of the snout. At the same time, it is also the shrew with the narrowest and flattest snout and with the most parallel mandibles. In addition, it has the most regular nose, with smaller differences in height values.

Some of the values are similar to S. coronatus, which is also characterised by a narrow, but not such a flat snout.

D. glyphodon does not have extreme relative parameter values compared with the other species studied here. It is the biggest species, but the rest of its characteristics are intermediate, falling between B. brevicauda and B. fissidens.

Internal characteristics: Subnumerary teeth.

One of the advantages of reconstructions based on microCT is that we can also use them to make observations of the interior of the model. In the specific case of the two studied soricids, we have identified for the first time that one of these species presents subnumerary teeth.

In the description of D. glyphodon in [32], it is said to have four upper unicuspids, but they are not present in the paratype, only the alveoli. Among the present D. glyphodon material, the A4 was not found either, so maxillae without this antemolar were selected. Although a rounded gap after A3 can be seen, in the microCT the alveolus was not found except on one side of one of the two skulls scanned. This indicates that the number of antemolars in D. glyphodon is variable.

[61] describe common anomalies related with teeth in shrews, and one of the most common (up to 5%) is subnumeracy of teeth, usually of the unicuspids. Although [62] do not encounter the anomaly in Sorex araneus, [63,64] say that in some individuals the last unicuspid is missing on one side or both. This phenomenon could be what is observed in D. glyphodon, since in one of the skulls there is only one A4. This indicates that the absence is not symmetrical, so this change does not seem to be the result of an adaptation to the environment or related to sexual dimorphism.