Institutional museum collection abbreviations follow Evenhuis (2017). The material on which this study is based is located and/or was examined at the following institutions:

BMNH The Natural History Museum, London, U.K.

CASC California Academy of Sciences, San Francisco, U.S.A.

MCZC Museum of Comparative Zoology, Harvard University, Cambridge, U.S.A.

NMKE National Museums of Kenya, Nairobi, Kenya

ZFMK Zoological Research Museum Alexander Koenig, Bonn, Germany

The general terminology for ant morphology predominantly follows Keller (2011) and to a lesser extent Bolton (1990) and Borowiec (2009, 2016). For the description of mouthparts, we used the terminology of Gotwald (1969) and Keller (2011). The terminology for the description of surface sculpturing follows Harris (1979).

All raw images were taken with a Leica DFC450 camera attached to a Leica M205C microscope and Leica Application Suite (version 4.1). The raw photo stacks were then processed to single montage images with Helicon Focus (version 6). All montage images used in this publication are available online and can be seen on AntWeb. Vector illustrations were created with Adobe Illustrator (version CS 5) by tracing specimen photographs.

We measured 17 physical workers with a Leica M125 equipped with an orthogonal pair of micrometers under magnifications of 80 to 100 ×. Measurements and indices are presented as minimum and maximum values with arithmetic means in parentheses. In addition, measurements are expressed in mm to two decimal places. Since the workers of all three species treated herein are eyeless we omit any eye measurements and do not generate an ocular (or eye) index. We refrain from using total length since it is difficult to measure in already dry-mounted specimens that are not orientated in a straight line. The standard measurements HW and WL provide sufficient information about general body size dimensions. The following measurements and indices partly follow Bolton and Fisher (2012), Hita Garcia and Fischer (2014) and Hita Garcia et al. (2014) or are used here for the first time (Fig. 1):

HL Head Length: maximum distance from the midpoint of the anterior clypeal margin or from a line spanning the anterior-most points of the frontal lobes (depending on which projects farthest forward) to the midpoint of the posterior margin of head, measured in full-face view (Fig. 1C).

HW Head Width: the maximum width of the head capsule, measured in full-face view (Fig. 1C).

SL Scape Length: the maximum straight-line length of the scape, excluding the basal constriction or the neck (Fig. 1C).

PH Pronotal Height: the maximum height of the pronotum in profile (Fig. 1A).

PW Pronotal Width: the maximum width of the pronotum in dorsal view (Fig. 1B).

DML Dorsal Mesosoma Length: maximum length of mesosomal dorsum from anterodorsal margin of pronotum to dorsal margin of propodeal declivity (Fig. 1B).

WL Weber’s Length of Mesosoma: the maximum diagonal length of the mesosoma in profile, from the angle at which the pronotum meets the cervix to the posterior basal angle of the metapleuron (Fig. 1A).

MFL Metafemur Length: the maximum straight-line length of the metafemur, measured in dorsal view (Fig. 1D).

PTL Abdominal Segment II (petiole) Length: the maximum length of abdominal segment II (petiole), measured in dorsal view (Fig. 1B).

PTH Abdominal Segment II (petiole) Height: the maximum height of the petiolar tergum in profile view, including laterotergite, excluding petiolar sternum (Fig. 1A).

PTW Abdominal Segment II (petiole) Width: the maximum width of abdominal segment II (petiole), measured in dorsal view (Fig. 1B).

A3L Abdominal Segment III Length: the maximum length of abdominal segment III, measured in dorsal view (Fig. 1B).

A3W Abdominal Segment III Width: the maximum width of abdominal segment III, measured in dorsal view (Fig. 1B).

A4L Abdominal Segment IV Length: the maximum length of abdominal segment IV, measured in dorsal view (Fig. 1B).

A4W Abdominal Segment IV Width: the maximum width of abdominal segment IV, measured in dorsal view (Fig. 1B).

A5L Abdominal Segment V Length: the maximum length of abdominal segment V, measured in dorsal view (Fig. 1B).

A5W Abdominal Segment V Width: the maximum width of abdominal segment V, measured in dorsal view (Fig. 1B).

A6L Abdominal Segment VI Length: the maximum length of abdominal segment VI, measured in dorsal view (Fig. 1B).

A6W Abdominal Segment VI Width: the maximum width of abdominal segment VI, measured in dorsal view (Fig. 1B).

CI Cephalic Index: HW / HL × 100

SI Scape Index: SL / HL × 100

DMI Dorsal Mesosoma Index: PW / WL × 100

DMI2 Dorsal Mesosoma Index 2: DML / WL × 100

LMI Lateral Mesosoma Index: PH / WL × 100

MF Metafemur Index: MFL / HW × 100

LPI Lateral Petiole Index: PTL / PTH × 100

DPI Dorsal Petiole Index: PTW / PTL × 100

DA3I Dorsal Abdominal Segment III Index: A3W / A3L × 100

DA4I Dorsal Abdominal Segment IV Index: A4W / A4L × 100

DA5I Dorsal Abdominal Segment V Index: A5W / A5L × 100

DA6I Dorsal Abdominal Segment VI Index: A6W / A6L × 100

Figure 1. 

Schematic line drawings illustrating the measurements used in this study. A Body in profile with measuring lines for PH, PTH, and WLB Mesosoma and metasoma in dorsal view with measuring lines for A3L, A3W, A4L, A4W, A5L, A5W, A6L, A6W, DML, PW, PTL, and PTWC Head in full-face view with measuring lines for HL, HW, and SLD Metafemur in dorsal view with measuring line for MFL.

Micro-CT scans were performed using a ZEISS Xradia 510 Versa 3D X-ray microscope and the ZEISS Scout and Scan Control System software (version 10.7.2936). The scanned specimens were left attached to their paper point, which was clamped to a holding stage. Scan settings were selected according to yield optimum scan quality: optical magnification of 4 ×, exposure times of 1–3 s, binning of two by two pixels, source filter “air”, voltage of 35–85 keV, power of 3–7.5 W, current of 71–88 μA, and field mode “normal”. The combination of voltage, power and exposure time was set to yield intensity levels of between 10,000 and 15,000 across the whole specimen. Scanning duration varied from 1.2 to 2.2 h, depending on exposure time. Full 360 degree rotations were done with a number of 1601 projections. The resulting scans have resolutions of 1013 × 1013 pixels and voxel sizes are in range of 0.94–4.6 μm. The original file size was 3113.577 MB for all scans. We scanned a varying number of specimens per species, depending on specimen availability and character suitability. An overview of the specimens used and scanning settings is provided in Table 1.

3D reconstructions of the resulting scans were done with XMReconstructor (version 10.7.2936) and saved in DICOM file format (default settings; USHORT 16 bit output data type). Post-processing of DICOM raw data was performed with Amira software (version 6.1.1). Virtual examinations of 3D surface models were performed by using either the “volren” or “volume rendering” functions. The desired volume renderings were generated by adjusting colour space range to a minimum so that the exterior surface of specimens remained visible at the highest available quality. The 3D models were rotated and manipulated to allow a complete virtual examination of the scanned specimens. Images of shaded surface display volume renderings were made with the “snapshot” function at the highest achievable resolution (usually at around 1500 by 893 pixels). Volumetric surface rendering rotational videos of head, mesosoma, metasoma, and full body scans were created with the “camera path” object (5–10 keyframes, constant velocity for constant rotation speed) and “movie maker” function (parameters: MPEG format, AntiAlias2, total of 1200 frames at 60 frames per second, and resolution of 1920 × 1080 pixels).

In addition to the traditional morphological examination of the physical specimens under a light microscope with magnifications up to 100 ×, we virtually examined the full external morphology of the treated species in Amira. For this we compared more than 50 morphological characters potentially significant for dorylines (Bolton 1990; Keller 2011; Borowiec 2016), especially the genera previously grouped as Cerapachyinae, in the 3D models and made more than 350 snapshots to assess which characters have diagnostic value. A series of characters were hidden or obscured by other body parts, thus not observable by light microscopy, or only after destructive dissection. Due to the severe lack of material the latter was not an option. In order to examine such characters and explore a wider range of morphology, we used the segmentation function in Amira to deselect body parts obscuring segmented body parts. By doing so we were able to expose every desired structure. We examined the head capsule from all sides, which is usually ventrally and posteriorly obscured by antennae, legs, or anterior mesosoma, as well as ventral metasomal characters usually hidden between propodeum, legs, and different abdominal segments. Based on more than 110 character images per species, we chose to highlight 24 characters of high diagnostic significance for our newly developed species delimitation system (see Table 2 for complete list of examined characters). Furthermore, the volume reconstructions of the mouthparts and musculature were generated by using the segmentation function in Amira. Targeted structures were first visualised by adjusting density and contrast and then segmented by manually tracing their outline slide by slide.

In addition to the taxonomic standard measurements of external morphology given above, we also measured the thickness of the exoskeleton cuticle of the cephalic capsule, the pronotum, and abdominal segments II (petiole) and III. Measuring was performed with Amira by using the 2D measuring tool on slices representing sagittal sections along the median axis of the chosen body parts. For each body part, we measured five times over a defined area (Fig. 2) and calculated the average thickness. Based on Peeters et al. (2017) we put the measurements in context to body size by using the following measurements and indices:

Cephalic capsule cuticle thickness (CCC): thickness of cuticle of head measured in profile a short distance posterior of torulo-posttorular complex (Fig. 2A).

Dorsal pronotum cuticle thickness (PRC): thickness of cuticle of pronotum measured in profile a short distance posterior of anterodorsal margin (Fig. 2B).

Dorsal abdominal segment II (petiole) cuticle thickness (ASIIC): thickness of cuticle of abdominal segment II measured in profile a short distance posterior of anterodorsal margin (Fig. 2C).

Dorsal abdominal segment III cuticle thickness (ASIIIC): thickness of cuticle of abdominal segment III measured in profile a short distance posterior of anterodorsal margin (Fig. 2D).

Cephalic capsule cuticle thickness index (CCCI): CCC / HW × 1000

Dorsal pronotum cuticle thickness index (PRCI): PRC / HW × 1000

Dorsal abdominal segment II (petiole) cuticle thickness index (ASIICI): ASIIC / HW × 1000

Dorsal abdominal segment III cuticle thickness index (ASIIICI): ASIIIC / HW × 1000

Figure 2. 

Microtomographic slides showing cuticle thickness measurements (with measuring lines in white). A Head in profile B Mesosoma in profile C Petiole (abdominal segment II) in profile D Abdominal segment III in profile.

The first step to creating unicoloured 3D PDFs was to make 3D renderings of ant specimens in Amira using the Isosurface function (deselect compactify) for exporting surface meshes in the STL file format. These were imported into Meshlab (version 1.3.3) where the number of vertices per specimen was reduced in three steps to decrease total file size and before importing into Adobe Acrobat. First, the scan files were cleaned from isolated vertices (Filters > Cleaning and Repairing > Remove isolated pieces (wrt diameter) [set max diameter: 0.05–1%]) and the paper tips on which the ants are mounted were digitally removed as much as possible. The next step removed all internal vertices so that only the exoskeleton remained (1. Filters > Color Creation and Processing > Ambient Occlusion Per Vertex; 2. Filters > Selection > Select Faces By Vertex Quality (min = 0, max = 0.001); 3. Remove Selected Faces). In the last step, the number of total vertices was reduced to the final number of <750,000 (Filters > Remeshing, Simplification and Reconstruction > Quadratic Edge Collapse Decimation) in order to get a manageable resolution resulting in 3D PDF files of approximately 20 MB in size for supplementary files (the final step was omitted for files uploaded to Dryad). The processed STL files were annotated and exported as 3D PDFs in Adobe Acrobat Pro DC (version 2015.006.30119) using the Tetra4D Converter plug-in (version 5.1.2). When viewing the 3D PDFs with Adobe Acrobat Reader (version 8 or higher), trusting the document by clicking on the image will activate the interactive 3D-mode and allows rotating, moving and zooming into the 3D model.

To generate the coloured 3D PDF of the mouthparts, we first segmented each mouthpart (maxillae, labium, labrum) independently and labelled each with a different colour. A surface mesh of the combined segmentation data was then generated using Generate Surface function in Amira with Unconstrained Smoothing (Smoothing Extent set to 1.5). We exported the surface data into Open Inventor Format, where it was converted to U3D format using IvTuneViewer plugged in Amira. Finally, the 3D PDF was generated by importing the U3D file to Adobe Acrobat Pro DC (version 2015.006.30119) with Tetra4D plugged in (version 5.1.2).

The small number and the preservation conditions of the specimens available for this study provided some limitations for the examination of mouthparts. It was not possible to dissect in vivo or micro-CT scan the open mouthparts of Z. obamai, nor of Z. wilsoni. Fortunately, the mouthparts of one pinned specimen of Z. sarowiwai were open and mostly exposed, thus available for superficial examination under the light microscope and for micro-CT scanning. Consequently, we were unable to test mouthpart morphology in detail for species delimitation. However, based on the limited information observable in closed condition, there appears to be no significant difference between the three species (Fig. 13). In the following we briefly describe the open mouthparts of Z. sarowiwai based on a 3D reconstruction of segmented micro-CT data (Fig. 14 and Video 4):

Labrum: distal margin conspicuously cleft medially; median area from anterior cleft to proximal articulation very thin, dividing labrum into two lobes; each lobe bulging medially; lateroventrally with two conspicuous hook-like labral arms projecting parallel to remainder of labrum; row of ten to twelve setae (1 very long pair plus four/five shorter pairs) on basal third of exterior face; row of four to six setae (1 very long pair plus one/two shorter pairs) on exterior face close to distal margin; labral tubercles absent.

Maxillae: maxillary palp three-segmented with second segment being greatly enlarged, third segment with very long seta, second with two long setae; deep and conspicuous diagonal, transverse stipital groove present dividing stipes into proximal external face and distal external face; articulation of labrum with maxillae of labro-stipital type via lateral extension/shoulder; proximal faces projecting beyond inner margin of stipites, thus almost completely concealing prementum; galea with well-developed galeal crown and maxillary brush, galeal comb apparently absent; lacinial comb not observable.

Labium: labial palp three-segmented with first segment being greatly enlarged, first and second segment with one long seta, third segment with three long setae; premental shield with several moderately long setae; shape of glossa not observable (structure collapsed); subglossal brush present and conspicuous with numerous long and thick setae; paraglossae absent.

Figure 13. 

Shaded surface display volume renderings of 3D models of mouthparts (excluding mandibles) in closed configuration (green=maxillae; yellow=labrum; orange=labium). A Zasphinctus obamai sp. n. (CASENT0764127) B Zasphinctus sarowiwai sp. n. (CASENT0764654) C Zasphinctus wilsoni sp. n. (MCZ-ENT-00512764).

Figure 14. 

Volumetric 3D model of segmented surface reconstructions of the mouthparts of Zasphinctus sarowiwai sp. n. (CASENT0764652) in open configuration (green=maxillae; yellow=labrum; orange=labium). A Frontal view B Lateral view C Posterior view D Dorsal view.

Based on our virtually reconstructed and segmented data, we can show that the mesosoma and metasoma both contain high degrees of musculature (Fig. 15; Video 5). The propodeum is tightly packed with dorsal and ventral muscles moving the abdominal segment II (petiole) and stabilizing the weight of the abdominal segments III to VII. Due to the massive size of the latter, the volume of the propodeal muscles attaching to the anterior petiole is high and comparable to that of the neck muscles in the pronotum. The petiole and the following segments also have a high muscle density, which is prominently visible in lateral (Fig. 15A, 15B), dorsal (Fig. 15C, 15D), and posterodorsal views (Fig. 15E, 15F) in the segmented 3D models of the metasoma. While the muscles in abdominal segment II – and to a lesser extent in segment III – evenly fill almost the entire segment, those in segments IV to VII are mostly limited to positions along the lateral and ventral walls. Finally, attached to the sting apparatus are two separate muscle sets, the protractors and retractors of the sting shaft. The former set is responsible for extending the sting from the tip of the abdomen during attack or defence and the latter for retracting it back to its resting position within the abdomen (visible in dorsal view in Fig. 15F).

Figure 15. 

Still images of 3D model of full body of Zasphinctus sarowiwai sp. n. holotype worker (CASENT0764654). False-colour volume rendering of segmented mesosoma and metasoma musculature (red) and sting apparatus (green) superimposed on semi-transparent surface model (A, C, E) or stand-alone (B, D, F). A, B Body in profile view C, D Body in dorsal view E, F Body in posterodorsal view.

Almost all previous studies that used micro-CT for invertebrate taxonomy encountered problems with the achievable voxel resolution in relation to body size resulting in a poor recovery of certain, very fine or small structures, such as setae, ommatidia, and microsculpture (Faulwetter et al. 2013, 2014; Fernández et al. 2014; Carbayo et al. 2016; Fischer et al. 2016). In the case of ant taxonomy this was intensively discussed by Hita Garcia et al. (2017a) who achieved voxel sizes of around 5 μm for full body scans of the two treated species. Based on these results and in order to improve the voxel resolution and present better resolved 3D reconstructions, we scanned the head, the mesosoma, and the metasoma separately in addition to a full body scan for each species. As a consequence, we attained smaller voxel sizes for the 3D models of the single body parts (0.95–2.83 μm versus 3.00–4.61 μm) resulting in a much higher resolution, and significantly reduced or eliminated the problems encountered by Hita Garcia et al. (2017a) (Fig. 16), except for ommatidia that are absent in Zasphinctus workers. While setae were poorly recovered by Hita Garcia et al. (2017a), they are very well visible in our 3D models of single body parts presented in this study. However, due to the higher voxel size, our full body scans of the Zasphinctus species have a weaker resolution of setae. Furthermore, compared to the physical specimens, surface sculpture was recovered with high morphological accuracy in the 3D models of single body parts, whereas it was only poorly noticeable in the full body scans. Surprisingly, fine surface sculpture on some body parts was even more observable in the 3D models than in the physical material, due to a limited magnification of our light microscope to 100 ×.

Figure 16. 

Comparison of full body scan versus single body part scans based on Zasphinctus obamai sp. n. holotype (CASENT0764125) and paratype worker (CASENT0764127). A Full body scan (CASENT0764125) B Scan of head (CASENT0764125) C Scan of mesosoma (CASENT0764127) D Scan of metasoma showing abdominal segments III to VII in profile (CASENT0764127).

The visualised reconstruction of the mouthparts provides a comparatively adequate and detailed 3D model of the maxillae, labium, and labrum, but also presents some important limitations. The general morphology of maxillae, labium and labrum are well recovered, and they are very similar to the mouthparts of Z. steinheili that were described by Gotwald (1969). The vast majority of setae are well visible, as is the surface sculpture of most structures. More importantly though, for the first time it is possible to examine these structures in their natural position in 3D, as well as their configuration with respect to each other. This is a significant advantage compared to traditional histological dissections that always remove all parts from the head and then separate each structure for separate examination. Our 3D volume reconstruction allows detailed examinations from all possible angles and the segmented mouthparts can be observed independently or in combination with each other.

Nevertheless, there are some problems with our 3D reconstructed model. The most problematic structure is the glossa. As already pointed out by Gotwald (1969), due to its highly membranous nature it is deformed in most dead specimens independently of preservation agent. In our specimen, the glossa was already collapsed to a crater-like appearance prior to micro-CT scanning, thus not available for any shape examination.

Another important limitation is that not all structures could be satisfactorily outlined during the segmentation process. This was especially difficult for the delineation of some components, such as the cardo, the lacinial comb, the regions where labium and maxillae meet, and generally everywhere where membranous and chitinous tissues are in contact. These problems are caused by scanning a dry mounted specimen, in which most internal structures have undergone desiccation, shrinkage, and deformation. In such specimens, the dissimilarities in density and contrast between different tissues or components are minimal to zero, thus causing significant problems for the proper recognition and subsequent outlining of borders between structures. In general, our 3D reconstructed model provides fewer details compared to histological dissections. However, these problems might be solvable in future studies if specimens are preserved and prepared in a way more suitable for micro-CT scanning and virtual reconstruction. Based on unpublished data, the use of freshly killed material or specimens in alcohol combined with the use of potassium hydroxide (KOH) and iodine staining provides much better resolution of internal structures than the use of dry material. This allows a much more sophisticated recovery of mouthparts morphology.

The application of micro-CT scanning to obtain information about cuticle thickness is novel. Based on our data however, we refrain from using it for taxonomic diagnostics at the moment. There are some differences in the cuticle thickness among the three species, most notably the very thick head of Z. obamai (CCCI 44 vs. CCCI 31 in Z. sarowiwai and Z. wilsoni). It also appears that the head of Z. obamai is thicker than the pronotum and abdominal segments II and III, whereas the heads of the other two species are thinner than or as thin as most other body parts (see Table 5). However, these results could be based on measuring the cuticle at a wrong angle resulting in a distorted result. Even though we have tried to find the best possible sagittal section slides for virtual measuring, it cannot be ruled out that we did not measure the thickest part of the cuticle. For future use, we recommend measuring more slides and more body parts in order to achieve a more complete picture of cuticle thickness. Even though we do not use it for taxonomic diagnostics at the moment, we believe there is potential for such use (unpublished data). However, this should be investigated with more taxa and more specimens first. Nevertheless, our results permit to place the treated species in an evolutionary context and make interpretations about natural history. Currently, the use of micro-CT data for virtual cuticle measuring has rather limited applicability due to the sparse availability of scanning resources in the myrmecological community. Nevertheless, we believe such data provides important information that might be valuable for future systematic and evolutionary studies.

As pointed out in previous studies, a crucial advantage of using 3D models based on micro-CT data is its potential application as cybertypes (Faulwetter et al. 2013, 2014; Akkari et al. 2015; Hita Garcia et al. 2017a, 2017b). The concept of a cybertype is to present a detailed and as complete as possible virtual reconstruction of a physical type specimen that is freely accessible. Hita Garcia et al. (2017a) critically assessed the usefulness of cybertypes for ant taxonomy, and due to the limitations in voxel resolution in their scan data suggested to use a combination of micro-CT data (raw data, 3D PDF, and 3D rotation video) and montage photos (three standard view: head in full-face view, full body in profile and dorsal view) as a minimal cybertype for ants. Although we have achieved much higher quality 3D reconstructions compared to previous ant studies using micro-CT and strongly reduced the limitations discussed by Hita Garcia et al. (2017a), we still believe that a minimum ant cybertype should include optical montage photographs. The main reason for this is the lack of natural colour in the micro-CT, which can only be shown with visible light photography. However, in this study we improve the previous ant cybertypes by providing micro-CT scans of single body parts and scan data for the holotype and one paratype, if available, thus increasing the usefulness of the cybertype datasets.

One aim of this study was to evaluate new taxonomic characters for species level taxonomy on the basis of traditional morphological analysis and virtual examination of 3D reconstructions. Unfortunately, only dry mounted material was available for this study. As pointed out in Hita Garcia et al. (2017a), most internal anatomical structures of such specimens have undergone significant desiccation, shrinking and deformation. Consequently, micro-CT data from dry specimens provides much less useful information for comparative examination. Based on our initial investigation, however, some internal sclerotized structures, such as the tentorium, several apodemes, and the endosternum appear to have significant potential for comparative morphology among species. However, due to the poor recovery of these structures in our raw data, we could not examine these in detail and focused our character evaluation on external morphology, with the exception of cuticle thickness. For future taxonomic studies using micro-CT, we propose to examine internal characters in more detail by using material preserved in ethanol.

Initially, our intention was to omit a species identification key, which may appear counterintuitive and substandard. However, there are several problems with dichotomous identification keys that lead us to take a different approach in this study. Identification keys for well-studied regions, such as Japan and Central Europe, generally work well and are very stable since new species are rarely encountered and nomenclatorial benchmarking is rare. This is certainly not the case for most tropical and subtropical regions because our knowledge of the local and regional diversity is fragmentary to non-existent. One major limitation of keys for such regions is that they usually only work for the known species at the time of publication. Later discoveries of new species render keys less useful and often less reliable for identification purposes. To avoid this, it is necessary to update older keys in additional publications after new species are discovered, as done by Hita Garcia and Fischer (2014) to update Hita Garcia et al. (2010). However, this is comparatively work-intensive and most authors of revisionary taxonomy studies are hesitant to revisit previously “finished” groups. This situation is especially problematic in hyperdiverse genera, for which there is no obvious solution except for continuously revising the taxa until most or “all” species are known and described. However, for genera with small or moderate species richness there might be alternatives to traditional identification keys with a few characters per key couplet.

In the case of Zasphinctus it is very likely that future collecting in the Afrotropical region will reveal additional species, even though not too many. This assessment is based on the apparent rarity of these ants and the fact that the region is largely undersampled. Accordingly, any identification key that covers only the three species treated here is doomed to obsolescence with the discovery of additional species, especially if only a few characters are listed per key couplet. Instead of simply providing a short key, we decided to present an illustrated matrix with numerous characters, in which we only present the ones that have proven to be diagnostic. Future users of our identification system can check multiple character illustrations and compare them with their specimens at hand. Nevertheless, despite our concerns with a short key with few diagnostic characters, we understand that some users would still prefer to use a short dichotomous key and we still provide one in this study.

Furthermore, the characters chosen are suited for diverse audiences with different interests and resources. For users with limited microscopy resources or little taxonomic training, we have included many easy-to-examine characters that are visible at lower magnifications, such as the shapes of head, mesosoma, or abdominal segment II in profile (e.g. Figs 4A–F, J–L, 5G–R). Based on that, we have added numerous further characters that are not completely necessary for simple identification purposes of faunistic or ecological studies, but target a more taxonomically oriented audience with better microscopy resources and deeper knowledge of ant morphology. We present many characters that require detailed examinations, such as structures on the posterior or ventral head (Figs 4P–R, 5A–F) and the ventral metasoma (Fig. 6A–C, G–L, P–R). These are intended to provide important comparative data for future systematic studies, and will very likely improve delimitations of any additional species.

As outlined above, compared to the taxonomy of most insect groups, the character sets used in the field of ant worker taxonomy are very often rather limited and rely heavily on setation, sculpture, body size, and colour. These are often problematic since they can be highly variable within species and prone to geographic variability, such as shown for the Neotropical Tatuidris Brown & Kempf (Donoso 2012), Malagasy Tetramorium Mayr (Hita Garcia and Fisher 2011) and Malagasy Crematogaster Lund (Blaimer 2012). Against the background of a resilient taxonomic impediment with continuously declining taxonomic resources and funding (e.g. Wheeler et al. 2004; Ebach et al. 2011), it is imperative to deliver taxonomic works that provide high quality species delimitations at a more accelerated speed that offer a stable taxonomic foundation upon which future discoveries can be based. We believe that the application of 21^st century taxonomic tools, such as molecular phylogenetics/phylogenomics and 3D next-generation morphology techniques, can strongly improve ant taxonomy.

For the revision of Afrotropical Zasphinctus, we have evaluated every single character that could be of diagnostic importance based on the literature record (Bolton 1990; Keller 2011; Borowiec 2016) through a combination of examination of physical material under the light microscope and virtually reconstructed 3D models of micro-CT scans. The latter is of crucial importance since it provides numerous advantages. The most important turned out to be the use of micro-CT data for virtual character examinations and dissections. As noted above, the available material was too scarce to perform physical dissections or dangerous manipulations of specimens. Fortunately, we were able to examine the virtual specimens from all imaginable angles by rotating the 3D models. In addition, in order to examine characters that were hidden behind other body parts, we virtually removed any obstructing structure. By doing this we were able to observe and reveal characters that are challenging to see in most dry-mounted specimens, thus rarely used for ant taxonomy. In particular, we found that the ventral and posterior head possesses a series of useful diagnostic characters, such as the hypostoma (Fig. 5D–F), the vertex (Fig. 4M–O), the occiput (Fig. 4P–R), and several margins around these structures (Fig. 4M–O, 5A–C). The same applies for the ventral metasoma since we found that abdominal segments II and III in ventral view provided some valuable characters, such as the subpetiolar process (Fig. 5M–O, 6A–C) or the prora (Fig. 6J–L). Physical examination of most of these would require damaging the specimens by removing legs, moving the head, separating abdominal segments, or in order to examine the occiput, detaching the head from the pronotum. By using micro-CT based 3D models, we were able to accomplish this with valuable type material without any damage.

Furthermore, the use of virtually reconstructed 3D models permits a quick and effective use of time and resources. Dissections and manipulations of physical specimens are usually very time-consuming, especially if histology and SEM are involved. By contrast, the application of micro-CT scanning enables highly accelerated examinations of morphology compared to these methods (Faulwetter et al. 2013; Friedrich et al. 2014). The initial data generation with a powerful scanner and subsequent 3D reconstructions can be done quickly with minimal effort, if the necessary scanning and visualisation resources are available (see limitations below). The virtual examination of characters is easy and very straightforward. The 3D models can be manipulated in many ways to observe the targeted morphological structures and once a character is in focus one can generate high-quality images within seconds. It is also easy to make images of structures from different angles in order to find the best one for presentation purposes. In our study, we generated several hundred character snapshots within a few days, thus allowing us to choose the most suitable ones for the illustrations used in this study. Generating so many images by using montage photography, SEM, or histology would require much more time and significantly slow down the speed of the publication process.

As already discussed in previous studies employing micro-CT data (Faulwetter et al. 2014; Carbayo et al. 2016; Hita Garcia et al. 2017a), the most important weakness is the limited access to scanning resources since most universities or museums do not have their own micro-CT scanners. Currently, the acquisition of scanners and access to external scanning facilities require substantial economic resources, and this situation will remain unchanged for some time. Some natural history museums and universities already have or are in the process of establishing scanning resources, but presently only a small minority of the taxonomic community has access to the technology. In addition, generating, post-processing, and handling the usually rather large data requires time and technical skills. Nonetheless, as with all new technologies, it is likely that technological and computational developments will reduce the costs of scanning, increase the availability for the taxonomic community, and simplify data management (Faulwetter et al. 2014; Hita Garcia et al. 2017a).

As outlined above, there is no knowledge of the natural history of Afrotropical Zasphinctus, except that they might live in leaf litter since most specimens were collected in litter samples. Against the background that they are dorylines and that their Australian congeners are predators of other ants, it is likely that the species treated in this study pursue a similar lifestyle. The examination of the micro-CT data generated during this study allows some inference about the lifestyle of the studied species.

As mentioned above, all three Afrotropical Zasphinctus species possess a very thick cuticle. Peeters et al. (2017) found the species with the thickest cuticle in relation to body size to be predominantly large ponerine genera, such as Diacamma Mayr, Odontoponera Mayr, Leptogenys Roger, and Ectomomyrmex Mayr. However, the thickest cuticle was observed in the species Ooceraea biroi (Forel), which is a doryline in relative close phylogenetic proximity to Zasphinctus. Based on that result, it is not that surprising that African Zasphinctus display such thick cuticle. Notwithstanding that Peeters et al. (2017) found that the best predictors for thick cuticle were body size (larger ants have thicker cuticle) and phylogeny (poneroid ants have thicker cuticle), the thick cuticle of African Zasphinctus is likely related to a predatory lifestyle. Based on observations of other Zasphinctus species mentioned above, it is highly probable that African Zasphinctus are top predators that feed predominantly on other ants, which is also the case in the clonal raider ant Ooceraea biroi, which feeds primarily on ant brood. Furthermore, Peeters et al. (2017) concluded that cuticle thickness was also negatively correlated with larger colony size in more phylogenetically derived ant lineages (formicoid clade). Despite a severe lack of observation and natural history data, it appears that African Zasphinctus live in small colonies, which is well in accordance with the findings of Peeters et al. (2017).

In Zasphinctus, the relative amount of muscles responsible for moving the abdomen seems to be largely increased compared to other ants from the formicoid clade, e.g. Pheidole Westwood & Terataner Emery, studied in previous publications (Sarnat et al. 2016; Hita Garcia et al. 2017a). It may be tempting to think that the genus Zasphinctus – as compared to the majority of ant lineages which have (evolved) greatly reduced abdominal musculature – has retained a more ‘primitive’ and wasp-like internal morphology in its abdomen. Yet, its species have a morphology that makes them rather special among ants: their relatively long abdomen is serially constricted between individually rotating presclerital plates. This apparent adaptation may have gained Zasphinctus additional functionality during predation and defence by increasing overall flexibility of use for its well-developed sting apparatus (Fig. 15) in the apical abdominal segment. It seems that other doryline genera such as for example Eusphinctus Emery, Sphinctomyrmex Mayr, and to a lesser degree possibly some Cylindromyrmex Mayr, and Leptanilloides Mann have evolved very similar features independently from Zasphinctus. It may be interesting to investigate the evolution of these specialised morphologies with respect to the different army ant lifestyles. Since studies on internal ant morphology are generally rare, we have little opportunity to compare our present results with those of others. Hashimoto (1996) for example found that, in some ant subfamilies, the muscles in abdominal segment II and III (petiole and postpetiole) show ‘positional and functional modifications’. Although he also gives an anatomically based discussion on the functional morphology of these modifications, there is no information given on related behavioural or ecological functions nor on musculature modifications in the posterior abdominal segments. With our main focus on the taxonomy of the three newly described species we refrain from further speculations and more detailed analyses in this publication and defer to future internal functional ant morphology studies using micro-CT scanning technology.