Micro-tomography
The small size and delicate bones make it impossible to remove the hard rock matrix without damaging the specimens and losing important information. In order to overcome this issue, PULR-V121 was analyzed through X-ray micro-tomography in YPF Tecnología S.A. (Y-TEC, Ensenada, Buenos Aires, Argentina) using the Bruker SkyScan 1173 instrument. The equipment was set up to 100kV and 80µA. A total of 900 images of the specimen were captured through a 360° tomography (rotation step 0.4°) with an exposure time of 250ms with two frames averaged. The experimental design resulted in a 40.01µm pixel size. The tomographic reconstruction was then produced with the software NRecon v. 1.6.9.8. Although obtaining acceptable results, the resolution was not ideal. Regrettably, the reconstructed images proved to be difficult to interpret, as bone and matrix were in some regions indistinguishable from each other. We acknowledged that this issue was probably technique-related due to the presence of ferruginous material in the sediment. Hence, we decided to perform a preliminary neutron-tomography at the RA-6 facility (Comisión Nacional de Energía Atómica, San Carlos de Bariloche, Argentina). Considering the promissory results obtained and the hypothesis that this specimen may represent a previously unregistered taxon of the very poorly represented mammalian ancestors in the late Late Triassic of Argentina, we performed a neutron tomography with the highest possible spatial resolution at the ANTARES instrument [56, 57] in the Forschungs-Neutronenquelle Heinz Maier-Leibnitz Zentrum (FRM II, Garching, Germany).
The specimen did not show previous radioactivity before introducing it directly to the reactor hall at ANTARES instrument. For the neutron tomography, PULR-V121 was wrapped in aluminum foil together with two additional specimens from the same stratigraphic levels (PULR-V222 and PULR-V223) to maximize the available beam-time. The package was placed in a 5cm long slot of an aluminum cylinder. A small aluminum plate was fixed to the cylinder using aluminum tape to act as a floor. This stabilized the specimens during the tomography and allowed to place them as close as possible to the detector (Supplementary Fig. S1). At ANTARES, a collimation ratio of L/D=500 was used. The Andor Neo sCMOS detector was equipped with a 100mm Zeiss Milvus f2.0 lens which allowed us to obtain high resolution images, with a 19.74µm pixel size. We performed a standard (white-beam), 360° tomography employing a Gd2O2S based neutron scintillator of 6 cm x 6 cm of 20µm thickness. The exposure time was 17s and each angular position (every ~0.192°) was acquired three times for improving statistics. The neutron tomography took circa 17 hours and 42 minutes. In order to normalize the images obtained, 19 open beam (open shutter, no sample in the beam) and 5 dark field (closed shutter) images were taken before and after the tomographic acquisition of the fossil remains, respectively.
The images were normalized and filtered using Image-J v. 1.52p [58] software and then reconstructed with Octopus Reconstruction v. 8.9.3.4 software at the Heinz Maier-Leibnitz Zentrum facility. Posteriorly, the reconstructed images were subjected to a new filtering process with the Inverse Scale Space Filter (ISS) module implemented in KipTool [59–61]. The ISS, an edge preserving de-noising filter based on the equation formulated by Burger et al. [62], notably increased the sharpness of the images without sacrificing morphological information. The segmentation of the specimen was performed by Dr. Gaetano who also generated the 3D model of PULR-V121. Approximately two weeks after the neutron tomography, PULR-V121 showed a very low decay ratio and was possible to remove it from the reactor hall.
Phylogenetic analysis
We put together a data matrix combining those of Liu and Olsen [42] and Ruta et al. [28] considering the subsequent modifications to both of them as well as modifying or deleting some of the characters and character states. We also added new characters and included the new specimen PULR-V121 as well as other relevant taxa. As a result, a comprehensive data matrix including 73 taxa and 151 characters was produced. Previous scorings were revised, and corrections implemented (Supplementary Appendix S1).
A first analysis was produced after the complete data matrix. After the recognition of three wild-card taxa (i.e., Diegocanis, Charruodon, and Microconodon), we pruned them from the matrix and performed a second analysis considering only the 70 remaining taxa. TNT 1.5 software program [63, 64] was used for searching of most parsimonious trees. Routine used was the command xmult=level 10, that produce 14 autoconstrained replications; each replication with random sectorial searches, drifting (36 iters) and fusing (10 round), finding best score 1 time; followed by bb (bbreak) command that perform branch-swapping (tree bisection reconnection) using pre-existing trees. Characters were unordered and equally weighted. The analysis of the complete matrix resulted in 29,304 most parsimonious trees (mpt) of 786 steps; whereas the second analysis resulted in 200 mpt of 782 steps.
Most scholars agree on the putative probainognathian nature of dromatheriids, a group of small cynodonts represented by isolated teeth or fragments of mandible, mostly discovered in the Laurasia subcontinent and India [47], yet its placement was not cladistically tested. Unfortunately, removal of Microconodon (one of the wildcard taxa in the complete analysis) from our second analysis, does not allow us to inspect the relationship of this lineage.
Paleobiogeographic analysis
A biogeographic analysis of the probainognathian clade as represented in the majority rule consensus tree of the pruned (70-taxa) phylogenetic analysis was performed in RASP 4.2 [65]. The cladogram branch-length was temporally calibrated with RStudio [66] using PaleoTree package [67] with EQUAL methodology. The First Appearance Datum (FAD) and Last Appearance Datum (LAD) as well as the geographical distribution of the taxa were obtained from the relevant published sources (see [28] and [4] for a review). We considered nine ancestral areas according to the regions that fossils were found (Supplementary Table S1). Most areas are so distant from each other that it is not necessary to determine discontinuity between them. Additionally, almost all of them have only endemic taxa. We performed a Bayesian Binary Markov Chain Monte Carlo (BBM; [68]) analysis with the calibrated cladogram, considering 100,000 cycles, 100 chains, and the maximum number of areas (nine) per node. We conducted an optimization analysis to test the possible ancestral area of relevant nodes using the Weighted Ancestral Area Analysis (WAAA; [44]). The number of weighted gain steps (GSW), weighted loss steps (LSW) and the probability index (PI = GSW/LSW) were calculated manually. The PI of each area indicates the probability of that area as a part of the ancestral area.