Background & Summary

Cichlids are one of the most speciose families of vertebrates, with over 1000 species in the African Rift Valley alone1,2. Multiple, independent, adaptive radiations of these fishes have evolved in the Great Lakes of East Africa, their associated satellite water bodies, as well as their connecting riverine systems. The radiations of these fishes (Subfamily: Pseudocrenilabrinae3), particularly those associated with Lakes Malawi, Victoria and Tanganyika, have become powerful models for the study of macroevolutionary processes4,5,6,7,8,9,10, behaviour and physiology11,12,13,14,15, and have emerged more recently as models in evolutionary developmental biology16,17,18,19.

Lake Malawi haplochromine cichlids represent a particularly speciose and phenotypically diverse adaptive radiation of lacustrine fishes. This diversity, comprising approximately 850 species of maternal mouthbrooders, is the most extensive adaptive radiation of vertebrates so far identified1,9. Molecular clock analyses estimate the radiation to be approximately 800 thousand years old4, a relatively young radiation when compared to the older system of Lake Tanganyika (~10myr) which contains just 250 species7,20. Despite their high phenotypic diversity, genetic variation between Lake Malawi cichlids is extremely low. Whole genomic comparisons of representatives from all seven distinct ecomorphological groups within Lake Malawi, estimated an average DNA sequence divergence of just 0.19–0.27%4 - a range comparable to that within human populations6. In addition, a relatively low DNA mutation rate; that alone cannot account for the estimated divergence time of Lake Malawi cichlids4,6 and overlapping distributions of inter- and intraspecific (heterozygosity) genetic variation4 only further complicates this enigmatic adaptive radiation.

East African cichlids, including those belonging to the Lake Malawi radiation, have recently emerged as powerful models in evolutionary developmental biology16,17,18,19. Evolutionary modification of embryological mechanisms drives the evolution of novel adaptations and requires genetic variation21. Thus, comparing the embryological development of cichlids, which have limited genetic variation, can enable us to identify specific cases where evolution has modified developmental mechanisms17. The diversity of feeding habits of Lake Malawi cichlids, and the ability to causally link morphological differences in craniofacial morphology to these ecological niches, has enabled integrative genetic and morphological studies examining the evolution of these traits22,23,24,25. More recent studies have expanded the scope beyond craniofacial phenotypes, including pigmentation patterning26,27,28; body and fin shape19,29 and axial elongation18. In parallel, aided by developments in whole-genome sequencing technologies6, it has been possible to considerably improve our understanding of the phylogenetic relationships among Lake Malawi cichlids4,9,30,31. Previously intractable macroevolutionary studies, such as the convergent evolution of hypertrophied lips32 can now take advantage of relatively robust phylogenies based on whole-genome sequences. Moreover, there are now opportunities to use this new phylogenetic information to focus on the evolution of other traits, such as the axial and appendicular skeleton, that is of key importance in teleost diversification33,34. However, a whole-body μCT-scan dataset of Lake Malawi cichlid fishes that captures the skeletal diversity present in the adaptive radiation has not yet been described.

Here we present a new database of high-resolution X-ray micro-computed tomography (μCT) scans of Lake Malawi cichlids, providing 3D data on skeletal morphology for the whole body of 56 species across 26 genera. In total these data comprise 116 individuals (56 species, 26 genera) from seven recognized ecomorphological groupings4 (Fig. 1, Table 1), contrasting in multiple aspects of morphology, size, behaviour, and habitat preference. We demonstrate the resolution and utility of our dataset by illustrating 3D whole-body renderings of several species, and of several skeletal regions of interest. Our dataset now joins two other East African adaptive radiation datasets, including the recent haplochromine Lake Victoria library35 and extensive μCT-scan dataset of Lake Tanganyika cichlid fishes7, permitting the examination of macroevolutionary patterns common to adaptive radiations36,37 and the macroevolutionary dynamics of convergent evolution. We also describe a methodology to efficiently μCT-scan multiple specimens simultaneously; reducing scanning time and financial cost, whilst maintaining scan quality and demonstrate the utility of this method by reconstructing 3D-models of multiple bones from multiple specimens within our dataset. We hope the availability of these data will inspire others to address some of the many questions still left to understand this remarkable adaptive radiation, permit wider-scale comparisons with other cichlid adaptive radiations, and set a precedent to make whole-body μCT scans the automatic standard for any sampling efforts involving cichlids.

Fig. 1
figure 1

A summary of the μCT-scan dataset. We were able to sample species from all seven ecomorphological groups in the Lake Malawi haplochromine radiation. The phylogenetic relationships between the majority of the species scanned is indicated and coloured according to the respective ecomorphology. The tree is a pruned version of the full (no intermediates) neighbour-joining tree published by Malinsky et al.4, which is rooted to Neolamprologous brichardi, a non-haplochromine cichlid endemic to Lake Tanganyika20. Longer terminal branches reflect a higher ratio of within-species to between-species variation. A cladogram depicting the relationship between the Lake Victoria, Lake Malawi and the Astatotilapia species native to the Great Ruaha River is indicated in the black box. We also scanned 18 species of cichlid whose phylogenetic relationships are not resolved in the phylogeny shown. The names of these species, most of which are undescribed, are indicated in their respective ecomorphological group in bold. Pictures (not to scale) of example species belonging to each ecomorphological group are also shown. Black bar: 2 × 10−4 substitutions per base pair. Fish images used with permission from Ad Konings (Alticorpus macrocleithrum, Diplotaxodon greenwoodi, Genyochromis mento, Hemitilapia oxyrhynchus, Iodotrophesus sprengerae, Nimbochromis polystigma, Placidochromis milomo and Trematocranus placodon), George F. Turner (Astatotilapia sp. ‘Ruaha blue’183, Diplotaxodon macrops, Mylochromis anaphyrmus, Otopharynx speciosus and Rhamphochromis woodi), Martin J. Genner (Diplotaxodon sp. ‘similis white-back north’, Diplotaxodon sp. ‘macrops ngulube’ and Rhamphochromis sp. ‘Chilingali’), Hannes Svardal (Copadichromis virginalis) and Callum V. Bucklow (Maylandia zebra). Fish images are not to scale.

Full size image
Table 1 Genera and species represented within the dataset.
Full size table

Methods

Sample Selection

There are an estimated 850 species of Lake Malawi cichlid fishes4. Many of which have not been described, preserved in museum collections or are available on phylogenies based on whole genome evidence. Therefore, to maximise the utility and morphological variation captured by our dataset, we focused on species present in published phylogenies4,9,24, and sought to include as many genera as possible. Moreover, we prioritised scanning the type species for genera, and avoided inclusion of species which already had whole body scans available on Morphosource. We were able to sample 26 of the 53 (49.05%) endemic genera listed on the curated species list on malawi.si38, or 48.15% of the 54 genera previously cited4 (see Table 1). Our specimens were sourced from the collections at the Natural History Museum in London (NHMUK), from the School of Biological Sciences of the University of Bristol (Martin J. Genner) and from the School of Natural Sciences of Bangor University (George F. Turner). In total we scanned 116 specimens from 56 species (Supplementary Table S1). Of these, 99 were wild-caught and 17 were laboratory-reared. Laboratory-reared specimens included Astatotilapia calliptera (Mbaka River, n=10), Maylandia zebra (Boadzulu island, n=5) and Rhamphochromis sp. ‘Chilingali’ (n=2), all of which died naturally or were compassionately euthanised by anaesthetic overdose [Schedule 1; Animals (Scientific Procedures) Act 1986]. All animal work was conducted following approval by the Departmental Animal Welfare Ethical Review Body (AWERB) for the Department of Biology at the University of Oxford.

μCT-Scanning

A flowchart describing all the necessary decisions and required processing steps is provided in Fig. 2. Since there was already a large collection of specimens present at the Natural History Museum in London (NHMUK) and in the extensive research collections of Martin J Genner (School of Biological Sciences, University of Bristol) and George F Turner (School of Natural Sciences, Bangor University) we decided to take advantage of the scanners present in the CT facility of NHMUK and at the XTM Facility based in the Paleobiology Research Group at the University of Bristol, respectively. Of the total 116 individuals scanned (56 species), 56 specimens (28 species) were scanned at the NHMUK Imaging and Analysis Centre and 60 specimens (28 species) were scanned at the XTM Facility at the University of Bristol.

Fig. 2
figure 2

Flowchart of μCT-scanning, image processing and segmentation methodology. The flowchart outlines the necessary decisions that were made during collation of the described μCT scan dataset. Rectangles represent processes; parallelograms represent inputs or outputs; diamonds represent decisions. It is sufficiently generalised that it can be reused for future data collection. We were focused on generating data for a specific macroevolutionary study, so we restricted the dataset to species with known phylogenetic placements but this is not strictly necessary. Software associated with data processing steps are indicated in purple. Further information about processing and segmentation is provided in the Usage Notes.

Full size image

Scanning Arrangement

To maximise the utility of our time and the number of species scanned, multiple specimens were scanned in each individual scan (Fig. 3). Each batch of specimens was fit to the width of the scan field of view to maximise resolution, and multiple scans were conducted along a the vertical axis in order to scan the full body length of each specimen. Therefore, batches of similarly sized fish were packaged together prior to scanning to ensure multiple individuals could be simultaneously captured within the field of view during scanning. Batch sizes varied between two and five specimens, with the number of each batch ultimately dependent upon the overall size of the specimens within the batch. Of the 32 batches scanned: 20 were comprised of four specimens; nine of three specimens, and two and one batch(es) of one and five specimens, respectively. Since multiple individuals of different species were often scanned together (Fig. 3A,B) it was critical that individuals of the same species could be readily identified. Therefore, unique, low density objects such as plastic bricks, pipette tips and rubber bands were placed in physical proximity to each specimen, to act as recognisable markers (Fig. 3C) which would readily resolve in the reconstructed image stacks (see Post-Scanning Processing). These objects were attached to each individual specimen-containing bag, and the specimens were bundled together, ideally ensuring that the objects faced outwards (Fig. 3D–F). Specimens were tightly packed into plastic containers, sealed with tape, and allowed to rest upright (head-up) for at least ten minutes so the contents could settle to prevent movement during scanning (Fig. 3G–K). We were able to scan, on average, 23 specimens per hour at maximal efficiency, an efficiency that was primarily the result of having two people per scanning visit (one scanning and one packing). The scanning rate could be further increased by packaging specimens in advance of the scanning so that subsequent batches can be scanned with no delay.

Fig. 3
figure 3

Specimen preparation for μCT-scanning. Multiple fish were scanned at the same time (A). Individual fish were labelled and placed in separate plastic bags so they could be correctly identified and stored after imaging (B). Unique objects (C) that would be readily identifiable following μCT-scanning were attached to the outside of these bags, ideally close to the heads, positioned outwards (F, arrows), and bundled together with tape (D–F) all with the same orientation (head-up). Bundles were then wrapped in bubble wrap and other packaging material (G,H) and tightly sealed inside a plastic container, again head-up (I). Containers were left for at least ten minutes to settle to prevent movement during scanning (J) and an additional label was placed on the container to permit future identification if multiple batches were prepared together (K).

Full size image

Scanning Procedure

Prior to scanning, a visual inspection of the specimens was made following X-ray exposure to check for internal damage (such as damage to the vertebral column). If present, and possible, specimens were switched for another specimen of the same collection. To maximise scan resolution, each batch of fish was scanned using the full width of the scanner field of view. This necessitates that multiple, overlapping scans were conducted along the vertical axis of the scanner (Z-axis) in order to capture the full body length of each fish. The overlapping scans were subsequently stitched together in the processing stage using the software, VGStudio Max 3.2.5 64-bit (see below). The fish scanned ranged in standard length of 3.8cm to 33cm, which meant that scanning parameters varied between batches. The number of projections generated range from 901 to 2001. Exposure times varied between 250 and 500 seconds, with a median of 354 seconds. Similarly, power varied between scans, ranging between 12.480 and 37.995 watts (W) (mode 37.440W). All scanning parameters, for each batch conducted, as well as for each individual scan can be found in the metadata supplied in Supplementary Table S1.

Post-Scanning Processing

All processing and segmentation was conducted on a machine specifically built for image analysis, with the following specifications: 2 x Intel® Xeon® CPU-ES-2640 @2.60GHz, 2601MHz, 8 Core(s) processors, 128GB of dedicated DDR3 RAM running on Microsoft Windows 10 Pro (Build Number: 10.0.19045). Raw, isometric volumes generated from the CT-scanning were imported into VGStudio Max 3.2.5 64-bit and anterior and posterior halves were subsequently stitched together by defining overlapping regions of interest in both the anterior, middle (if applicable) and posterior volumes. 16-bit tiff stacks were exported from VGStudio Max 3.2.5 64-bit and imported into FIJI39, a GUI for ImageJ40. In FIJI, individual fish were cropped out of the 16-bit stacks, which were identifiable due to the unique objects associated with each individual (see above). The brightness was adjusted by extending the distribution of pixel values to remove 0 values, and the tiff stacks were converted into 8-bit to decrease file size. Total stack file size was further decreased by removing images at the beginning and end of the exported tiff stacks that did not contain readily identifiable bone or tissue.

Segmentation and Visualisation

Reconstructed image stacks were imported into Avizo Lite (v9.3.0), a proprietary software developed by Thermo Fisher Scientific and generated either a full body volume render using the volume generation tool or manually rendered 3D models of the whole body following a rough manual segmentation of the whole body. Optimal threshold values were manually chosen based on the volume or 3D rendering and all whole-body 3D models were smoothed with a smoothing factor of 2.5. Surfaces from 3D whole-body renderings were exported from AvizoLite as Polygon File Format (.ply) files and imported into MeshLab for visualisation and manipulation. See the Usage Notes for further notes on segmenting the μCT-scans in the dataset.

Data Records

All data is freely avaliable and has been deposited into a dedicated project on Morphosource41, which includes cropped 8-bit tiff stacks of 116 specimens42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,156,157, representing 56 species. Coverage of the Lake Malawi radiation by our dataset41 is outlined in Table 1. The names of all 56 species present in the dataset, including their respective ecomorphological group, is listed in Fig. 1. Full specimen details, including scanning parameters, can be found in Supplementary Table S1. Detailed discussion of the species in the dataset, including suggestions for possible macroevolutionary and systematic studies can be found in the Usage Notes (see below).

Technical Validation

Example whole-body, scaled, 3D renderings of representatives from each ecomorphological group can be found in Fig. 4 and for every specimen (mix of volume and model renders) in the dataset in the Supplementary Material. To demonstrate the quality of 3D-models that can be segmented from specimens in our dataset we manually segmented multiple bones, including the dentary, premaxilla, lower pharyngeal jaw and multiple vertebral types in several species discussed below (see Usage Notes). This included: the ‘generalist’ Astatotilapia calliptera; the ‘mbuna’ and lepidophage (scale-eater), Genyochromis mento; the ‘shallow benthic’ and snail crusher, Trematocranus placodon (Fig. 5); the pelagic piscivores, Rhamphochromis esox and Pallidochromis tokolosh and the zooplanktivorous (utaka), Copadichromis trimaculatus (Fig. 6). Lower pharyngeal jaws that are highly variable among species of Lake Malawi cichlids25 were particularly well resolved. For example, newly erupting teeth were visible on the relatively large, and dense, lower pharyngeal jaw of Trematocranus placodon (Fig. 5B). Similarly, renderings of multiple vertebral types (Figs. 5–6) were also of good quality. The zygapophyses and fine structure of the vertebral centra, sometimes including the neural foramen, were also well resolved. All 3D-renderings of these bones can be found in the Supplementary Materials as downloadable .ply files.

Fig. 4
figure 4

Whole-body 3D models of select specimens from the dataset. Specimens are arranged according to the ecomorphological group they belong to. Species names are indicated. The ring structure in Diplotaxodon sp. ‘holochromis’ and Lethrinops gossei is a rubber band used for identification purposes. Scale for all images is shown as 1cm. See Supplementary Table S1 for details of the specimens used.

Full size image
Fig. 5
figure 5

Segmented Bones from Astatotilapia calliptera, Genyochromis mento (mbuna) and Trematocranus placodon (shallow benthic). (A, left) A close up, lateral view of the head of each species (species name indicated on right), showing the dentary (green), premaxilla (pink) and lower pharyngeal jaw (purple) positioned within a volume render of the head. (A, right) A whole body lateral view showing the aforementioned jaw bones, as well as the first non-rib-bearing vertebra (orange), the first rib-bearing (precaudal, PC) vertebrae (light blue), PC8 (green), non-rib bearing (caudal, CV), CV3 (orange), CV10 (gold) and the pre-urostyle vertebrae (red). (B) Anterior (top) and anterolateral (bottom) view of the lower pharyngeal jaws for each species in (A). Scale for all images is 1cm. See Supplementary Table S1 for details of the specimens used. 3D models for all segmented bones can be found in the Supplementary Material.

Full size image
Fig. 6
figure 6

Segmented Bones from Rhamphochromis esox (Rhamphochromis), Pallidochromis tokolosh (Diplotaxodon) and Copadichromis trimaculatus (Utaka). (A) Left, lateral view of the head of each species, showing the dentary (green), premaxilla (pink) and lower pharyngeal jaw (purple). Right, whole body lateral views showing aforementioned jaw bones, as well as the first non-rib-bearing vertebrae (orange), the first rib-bearing (precaudal, PC) vertebrae (light blue), PC8 (green), non-rib bearing (caudal, CV), CV3 (orange), CV10 (gold) and the pre-urostyle vertebrae (red). (B) Images of select vertebrae (indicated) from each species shown in (A). Axes are indicated (A, anterior; P, posterior; D, dorsal; V, ventral). Vertebral models are labelled (ac, anterior cone span; cn, centrum; ec (dr), epicentrals (dorsal ribs); hc, haemal canal; hs, haemal spine; nc, neural canal; nf, neural foramen; ns, neural spine; pc, posterior cone span; pr, pleural ribs; zg, zygapophyses). Scale for all images is 1cm, besides CV10 for Utaka which is 0.1cm. See Supplementary Table S1 for details of the specimens used. 3D models for all segmented bones can be found in the supplementary material.

Full size image

In addition, to demonstrate that our data will also be useful for the collection of meristic data, we also counted the number of precaudal, caudal and total number of vertebrae (including the urostyle), as well as an estimation of the body aspect ratio for 113 of the 116 specimens in Supplementary Table S2. The three remaining specimens were deformed or poorly rendered and vertebral counts or body aspect ratios could not be taken. Precaudal, caudal and total vertebral counts as all as an estimation of the body aspect ratio were estimated from 2D lateral images of either a volume rendering or 3D-model of the whole body of the specimen (Fig. 7). Lateral images of a volume or 3D whole body rendering can be found in the Supplementary Material for every specimen in the dataset. Length landmarks were placed on the anterior tip of the premaxilla and in the centre of the urostyle and width landmarks were placed at the base of the dorsal fin spine and pelvic fin spine, respectively (see Fig. 7). We note that these data could also be collected with the use of 2D radiographs. However, μCT-scanning adds the additional possibility of also generating 3D models which can be the basis of studies that consider and compare overall, complex bone shape that would otherwise be missed (3D geometric morphometrics), or undetectable with the use of 2D radiographs. Therefore, the μCT-scans in our dataset provide can both provide well resolved models for geometric morphometric studies comparing complex bone shape between Lake Malawi cichlids as well as a useful source of meristic data for macroevolutionary studies. We have included suggestions of possible macroevolutionary studies that could be conducted with these data in the Usage Notes.

Fig. 7
figure 7

Example specimen volume rendering with body aspect ratio landmarks. The number of precaudal (pink) and caudal vertebrae (blue), including the urostyle, the total number of vertebrae (sum of the precaudal and caudal vertebrae, including the urostyle) and the body aspect ratio was estimated for 113 of the 116 specimens in the dataset. The landmarks used to estimate the body aspect ratio are indicated in the figure on an example specimen (Maylandia zebra, L-BV:M6, see Supplementary Table S1). Landmarks 1 and 2 were used to calculate the length, and landmarks 3 and 4 the width by calculating the length of a straight line between the x,y coordinates. Landmark 1, anterior tip of the premaxilla; Landmark 2, centre of the urostyle; Landmark 3, Base of the first dorsal fin spine; Landmark 4, base of the pelvic fin spine.

Full size image

Usage Notes

Diplotaxodon and Rhamphochromis

Nine species of the Diplotaxodon group (47%; Table 1), including the type species Diplotaxodon argenteus (n=1) and Pallidochromis tokolosh (n=2) are present within our dataset41. In addition, we sampled 10 species of the Rhamphochromis genus (71%; Table 1), including the type species Rhamphochromis longiceps (n=2) and the remarkably large Rhamphochromis woodi (n=2, see below), that are endemic to Lake Malawi. We were also able to sample two sympatric species from the crater lake, Lake Kingiri, Rhamphochromis sp. ‘Kingiri dwarf’ (n=2) and Rhamphochromis sp. ‘Kingiri large’ (n=2), as well as Rhamphochromis sp. ‘Chilingali’ (n=4) from the satellite, Lake Chilingali (Fig. 4), which is now presumed extinct in the wild158.

The Diplotaxodon and Rhamphochromis groups are two reciprocally monophyletic diverging lineages of Lake Malawi cichlids4 that have adapted to the pelagic-limnetic zone of Lake Malawi159. The majority of species in the groups are piscivorous, although several species, including Diplotaxodon limnothrissa (n=2) are predominantly zooplanktivorous160. Large-bodied Rhamphochromis primarily feed on Lake Malawi sardines (usipa; Engraulicypris sardella) and endemic cichlids (e.g. utaka). Members of Diplotaxodon and Rhamphochromis are among the deepest-living of all Lake Malawi cichlids, with representatives of both being caught at depths exceeding 200 metres - the ‘twilight zone’ where light is almost completely absent159. Species within the Diplotaxodon macrops complex, which is represented in the scanned samples by Diplotaxodon macrops (n=1) (Fig. 4), Diplotaxodon sp. ‘macrops north’ (n=2), Diplotaxodon sp. ‘macrops black dorsal’ (n=2) and Diplotaxodon sp. ‘macrops ngulube’ (n=2), have been found between 100 and 220m, a depth similarly reported to be occupied by Rhamphochromis during the day161.

Morphological comparisons of Diplotaxodon and Rhamphochromis with Lake Malawi cichlids from other habitats could provide valuable insights into convergent adaptation of traits enabling occupation of pelagic niches. Divergence along depth gradients is associated with the evolution of reproductive isolation in many marine and freshwater species groups, likely a consequence of the strong selective pressures associated with deeper water, such as the absence of sunlight, greater hydrostatic pressure, and reduced levels of dissolved oxygen162,163. Morphological comparisons of Diplotaxodon and Rhamphochromis, against closely related littoral species, could be a powerful model for the study of evolution of convergent phenotypes necessary for adapting to pelagic environments. Body elongation, supported by increased vertebral counts, is a adaptive trait common in teleosts adapted to pelagic (and piscivorous) niches164, including in Rhamphochromis165. Moreover, the evolutionary modification of vertebral morphology has been linked to changing swimming kinematics, body shape and habitat preference166, including adapting to pelagic environments34. Given the remarkable depth preference and rapid divergence of the Diplotaxodon and Rhamphochromis lineages, a study of body shape, as well as vertebral count and shape could better determine the rate at which these phenotypes can become fixed and provide further insights into the role these morphological adaptations play along the benthic-pelagic speciation axis162,163.

Remarkable size variation is present within the Rhamphochromis genus. Rhamphochromis woodi is considered to be one of the largest Lake Malawi cichlids, measuring a standard length (SL) of up to 40 cm167. In contrast, the smallest known member of Rhamphochromis, Rhamphochromis sp. ‘Kingiri dwarf’, endemic to the crater lake Kingiri, do not exceed 7.5 cm SL in the wild158 – a 5.33x length difference just within the same genus. Similarly, wild caught Rhamphochromis sp. ‘Chilingali’ are also small bodied, with maximum observed standard length of 10.6 cm8, which makes them relatively amenable to laboratory study. Its elongate body, supported by relatively high vertebral counts, has made it a useful model in evolutionary developmental biology17, particularly for the study of somitogenesis18, the developmental process that gives rise to the vertebral precursors. Nonetheless, given the exceptional size difference in the genus Rhamphochromis, our dataset represents a potentially valuable resource for the study of the evolution of allometric scaling, which has not been well studied in cichlids168.

Shallow Benthic

The shallow benthic species group is extremely speciose, comprising hundreds of species that exhibit remarkable morphological4,169 diversity. The majority of shallow benthic species inhabit relatively shallow inshore habitats of Lake Malawi, such as the sand or mud lake floor, or sand-rock transitional zones. Our dataset41 includes 20 shallow benthic species in 12 genera (Table 1), including several large, ambush predators, as well as a collection of trophic specialists. For a complete list of shallow benthics in our dataset41 see Supplementary Table S1.

Large ambush predators represented in the dataset41 include Dimidochromis strigatus (n=1), Dimidochromis compressiceps (n=1), Tyrannochromis macrostoma (n=1), Nimbochromis livingstonii (n=1) and Nimbochromis polystigma (n=2). Dimidochromis compressiceps has a generalist piscivore lifestyle, occupying the reed-beds of the Lake. Nimbochromis livingstonii and N. polystigma are both considered to be ‘sleeper[s]’ (Chichewa: “kaligono”), which bury themselves within the sandy substrate and snatch unsuspecting prey attracted by the disturbed sediment169. Another member of Nimbochromis, Nimbochromis linni (n=1) has a characteristic downward-projecting snout (Fig. 4), enabling it to extract prey from rock crevices169,170.

We sampled several shallow-benthic predators, including Otopharynx speciosus (n=2), one of the few piscivores within Otopharynx. Males of this species have been encountered at depths exceeding 25m169, suggesting tolerance of relatively deep water, and suggesting the species may have morphological adaptations enabling occupation of deep-water niches similar to Rhamphochromis and Diplotaxodon. Of the approximately 20 species of Otopharynx171 we were able to sample an additional three species: Otopharynx lithobates (n=3, including the holotype NHMUK 1974.7.5.1); Otopharynx tetrastigma (n=2); and the undescribed Otopharynx sp. “brooksi nkhata” (n=1). We also sampled several specialised trophic specialists including the molluscivores Mylochromis anaphyrmus (n=1) and Trematocranus placodon (n=1) and the invertebrate picker Placidochromis johnstoni (n=1, Fig. 4). The diet of T. placodon predominately comprises the gastropods Bulinus nyassanus and Melanoides tuberculata172. Enlarged sensory pores and lateral lines form a sonar-like detection system that allows T. placodon to sense the movement of these prey within the sediment. Curiously, this strategy and associated morphological characteristics are also associated with Aulonocara and Lethrinops, both ‘deep benthics’, suggesting convergent evolution of lateral line phenotypes9. The specimens in our dataset41 may enable morphological comparisons to further investigate differences in sensory pore characteristics among species.

‘Rock-dwelling’ Mbuna

The mbuna group dominate the rocky shores of Lake Malawi, and are used as a model system for the study of rapid speciation and adaptive radiation25,173,174. Similar to the shallow-benthics, there are hundreds of species, many of which are undescribed169,174. We aimed to maximise our coverage of the phenotypic diversity in the group by sampling multiple genera, which are largely differentiated on the basis of head, jaw and tooth morphology174. Our dataset41 includes 7 species (15 individuals) of mbuna, covering 7 of the 14 described mbuna genera (Table 1).

Cynotilapia can be distinguished from other genera by the presence of unicuspid (conical) teeth169,175,176 and is represented in our dataset by Cynotilapia axelrodi (n=1, Fig. 4). This is a genus of typically planktivorous species169 and their relatively simple dentition may reflect this lifestyle177. By contrast, Maylandia (Metriaclima178), represented by Maylandia zebra (n=5), has closely arranged bicuspid teeth, that is uses for pulling and scraping loose Aufwuchs (periphyton) attached to the rocks found in their preferred preferred rocky habitats175,179. Tropheops and Iodotropheus, represented by Tropheops tropheops (n=2) and Iodotropheus sprengerae (n=2), also have closely packed bicuspid teeth, that they use to feed on epilithic algae which they pluck with sideways, upwards head jerks, a behaviour likely supported by Tropheops’ characteristic steeply sloped vomer (71-96°)169. Members of Petrotilapia, represented by Petrotilapia genalutea (n=1) have a mixed combination of tricupsid and unicupsid teeth that they use to comb loose peripyton from rock surfaces180. A further represented mbuna genus is the monotypic Genyochromis, represented by Genyochromis mento (n=2). Like the majority of mbuna, G. mento has prominent outer bicupsid teeth that are supported by smaller inner tricupsid teeth169,175. In contrast to most other mbuna, however, G. mento is a highly specialised feeder, a lepidophage (scale-eater), that targets the the caudal and anal fins of other cichlids in rocky habitats169,175,181. The preferred striking side of G. mento significantly correlates with left-right asymmetry of the dentary, with right and left-leaning individuals preferring to strike the corresponding side, respectively, of their prey. Interestingly, however, a comparison of their jaw laterality with Perissodus microlepis, a lepidophage endemic to Lake Tanganyika20, showed that laterality in G. mento is weaker than in P. microlepis – likely a result of phylogenetic constraint from their shorter evolutionary history and their herbivorous ancestors181.

The craniofacial bones commonly studied in mbuna, such as the dentary, premaxilla, pharyngeal jaws, as well as their associated teeth, can be segmented from specimens in the dataset (see G. mento, Fig. 5) and will be helpful for geometric morphometric analyses focused on examining craniofacial shape differences between mbuna species. Future sampling should focus on the seven remaining genera not sampled in our dataset41: Abactochromis, Chindongo, Cyathochromis, Gepyrochromis, Labeotropheus, Melanochromis and Psuedotropheus.

Astatotilapia calliptera and Ruaha Catchment

Astatotilapia is polyphyletic and current members of the genus are widespread across East and North Africa6,182,183. Only one species of Astatotilapia is native to Lake Malawi, Astatotilapia calliptera, which is also found in East African rivers flowing eastward to the Indian Ocean, from the Rovuma River in the north, to the Save River in the south. Given the wide distribution of the species, it is perhaps unsurprising that intraspecific genetic variation within the species is comparable to that of the whole Lake Malawi radiation4,6. Despite their wide distribution and relatively large intraspecific genetic variation, they phylogenetically cluster within the Lake Malawi radiation (Fig. 1), forming a sister clade to the mbuna, with which they share an excess of alleles4. This pattern, alongside a perceived riverine ‘generalist’ lifestyle, has led to the hypothesis that either Lake Malawi cichlids radiated from an A. calliptera-like ancestor or that A. calliptera is the sympatric ancestor of all Lake Malawi cichlids4,6,169,183.

Given the importance of A. calliptera in the Lake Malawi radiation, we sampled multiple individuals from multiple populations. We scanned nine laboratory-reared individuals from the Mbaka river population, which flows into the northern end of Lake Malawi184. We also scanned individuals from Lake Chilwa (an endorheic lake south-east of Lake Malawi185; n=2), Lake ‘Misoko’, presumably Lake Masoko (a crater lake north of Lake Malawi184, n=2), and wild-caught individuals from the main body of Lake Malawi (n=2). Populations of A. calliptera differ in life history strategies186 and are also undergoing sympatric speciation along a depth gradient in at least one location (Lake Masoko)8, where littoral and benthic A. calliptera ecomorphs have diverged in multiple characteristics, including body shape and trophic specialism, in approximately 1000 years8. Therefore, it is possible that morphological evaluations of more populations of A. calliptera will reveal further diversity, potentially providing greater insight into the role it has taken in generating the wider Lake Malawi haplochromine radiation and we would suggest increasing the sampling of A. calliptera populations to further investigate this.

A key part of macroevolutionary studies is the estimation of ancestral state of traits based on the morphology of their descendants187. This necessitates a comprehensive understanding of trait diversity across taxa, where such data is critical for the construction of models of morphological evolution, including estimating rates of phenotypic evolution. Since the genetic diversity of the Lake Malawi radiation was possibly seeded by multiple riverine species5, we sought to add specimens to the dataset that could enable the morphological reconstruction of the common ancestor of the Lake Malawi radiation. Therefore, we sampled two additional species of Astatotilapia. These included Astatotilapia gigliolli (n=2) and Astatotilapia sp. ‘Ruaha blue’ (n=2), native to the Great Ruaha River182,183. Construction of a mtDNA-based phylogeny initially placed Astatotilapia sp. ‘Ruaha blue’ as a sister taxa to the Lake Malawi radiation182. However, a phylogeny based on variation within whole-genome sequences has shown A. gigliolli and A. sp. ‘Ruaha blue’, sister taxa, form a sister clade with both the Lake Malawi and Lake Victoria radiations (see Fig. 1). This topology is likely the result of an ancestral hybridisation event with the ancestors of both lineages prior to their respective adaptive radiations5. Therefore, the addition of species from the Ruaha catchment, may therefore enable a more robust estimation of the ancestral phenotype of Lake Malawi cichlids.

Deep Benthic and ‘Utaka’

We sampled deep-water benthic species from two genera; Alticorpus and Lethrinops9,188 (Table 1). Alticorpus is characterised by the presence of greatly enlarged cranial sensory openings and lateral line canals used to detect prey in the sediment. Deep-water benthic species are found below 50m, a ‘twilight’ zone with very little visible light. Alticorpus macrocleithrum (n=3) is found between 75m and 125m, with abundance peaking above 100m189, a depth similarly occupied by deep-water Lethrinops190, including Lethrinops gossei (n=1). Several species of Lethrinops, however, inhabit shallower water4,169. We sampled two species of shallow water Lethrinops, including Lethrinops auritus (n=2), and Lethrinops albus (n=2), both of which phylogenetically cluster within the ‘shallow benthic’ lineage (Fig. 1).

Our dataset41 also contains four species of zooplankton-feeding, shoaling cichlids which are commonly referred to as ‘utaka’. Utaka is primarily made up of species belonging to Copadichromis191, with a small number of species also belonging to Mchenga and Nyassachromis192. However, their placement within the utaka is disputed and they have not been considered in our species/genera counts (see Table 1). We sampled four species of Copadichromis: Copadichromis likomae (n=2), Copadichromis quadrimaculatus (n=2), Copadichromis trimaculatus (n=2, see Fig. 6) and Copadichromis virginalis (n=2). Utaka feed in the water column, and can be commonly found close to the shore169. Copadichromis are generally characterised by their relatively small, highly protrusible mouths, that they use to suck zooplankton into their mouths, as well as numerous long gill rakers which strain plankton from the water that enters their mouths as a result of their sucking feeding mechanism169,193.

Both the deep benthics and utaka are currently underrepresented within our dataset and future sampling should aim to add additional species of Copadichromis. In addition, sampling missing ‘deep benthic’ genera, such as Aulonocara and Tramitichromis should be prioritised for future sampling efforts. Aulonocara stuartgranti and Aulonocara steveni would be particularly interesting future additions and could offer interesting morphological comparisons with deeper living species. Moreover, given that Lethrinops is polyphyletic190,194, additional sampling of Lethrinops species could provide morphological data to support future systematic studies.

Whilst we were able to generate relatively good models for the specimens within this group (see Fig. 4 and Fig. 6), it is clear that some of the jaw did not resolve as well as in other specimens. Given the preferred ‘sucking’ zooplanktivore feeding mechanism of the species within Copadichromis it is possible that the jaw bones of these fish species are not particularly dense. This perhaps made it difficult to image these specimens using the same scanning procedure used for all the other remaining specimens. Therefore, in future, we would recommend care when segmenting craniofacial bones from the Copadichromis in this dataset and future sampling efforts should increase scanning exposure time and power to optimise specimen imaging.

Processing and Segmentation Notes

The computer specifications we used for all processing steps (see Methodology) are hard to find on personal, or older machines and some users may find it difficult to work with some of our larger image stacks. To minimise memory usage during segmentation and speed up processing, cropped reconstructed stacks can be loaded in multiple increments (note that the Z-voxel size must be multiplied by said increment). We tested this and found that roughly comparable models could be generated, although it was clear that finer morphological detail was absent (data not shown). Therefore, where possible, the whole stack should be used when segmenting regions of interest. In addition, since these regions were manually segmented, many of the segmentation steps rely on the judgment of the individual segmenting and rendering the regions of interest. We found that segmenting from median-filtered reconstructed image stacks drastically lowered the quality of the rendered models (data not shown) and we suggest refraining from segmenting from a median-filtered image stack. In addition, we found that relatively low smoothing factors were best for rendering surfaces from segmented regions of interest. In Avizo Lite (v9.3.0), a smoothing factor between 0-10 (including rational intermediates) can be applied when rendering surfaces of segmented regions of interest. We rarely found it necessary to use a value above 3; indeed, all whole-body 3D models were smoothed with a factor of 2.5. Therefore, we suggest that regardless of the tool used to smooth and render segmented surfaces that smoothing be used conservatively. We also note that there are free, open access alternatives to Avizo Lite (v9.3.0) for the segmentation of 3D-image data, such as 3D-Slicer, which has a large and active community of users195 and Dragonfly which supports the use of deep learning to automatically segment 3D image data and offers non-commercial licenses, for academic use, free-of-charge. Both of these tools could be used in place of Avizo Lite (v9.3.0) for the segmentation steps outlined in Fig. 2.