New phylogenetic insights into the African catfish families Mochokidae and Austroglanididae.

Several hundred catfish species (order: Siluriformes) belonging to 11 families inhabit Africa, of which at least six families are endemic to the continent. Although four of those families are well-known to belong to the 'Big-Africa clade', no previous study has addressed the phylogenetic placement of the endemic African catfish family Austroglanididae in a comprehensive framework with molecular data. Furthermore, interrelationships within the 'Big-Africa clade', including the most diverse family Mochokidae, remain unclear. This study was therefore designed to help reconstruct inter- and intrarelationships of all currently valid mochokid genera, to infer their position within the 'Big Africa clade' and to establish a first molecular phylogenetic hypothesis of the relationships of the enigmatic Austroglanididae within the Siluriformes. We assembled a comprehensive mitogenomic dataset comprising all protein coding genes and representing almost all recognized catfish families (N = 33 of 39) with carefully selected species (N = 239). We recovered the monophyly of the previously identified multifamily clades 'Big Asia' and 'Big Africa' and determined Austroglanididae to be closely related to Pangasiidae, Ictaluroidea and Ariidae. Mochokidae was recovered as the sister group to a clade encompassing Auchenoglanididae, Claroteidae, Malapteruridae and the African Schilbeidae, albeit with low statistical support. The two mochokid subfamilies Mochokinae and Chiloglanidinae as well as the chiloglanid tribe Atopochilini were recovered as reciprocally monophyletic. The genus Acanthocleithron forms the sister group of all remaining Mochokinae, although with low support. The genus Atopodontus is the sister group of all remaining Atopochilini. In contrast to morphological reconstructions, the monophyly of the genus Chiloglanis was strongly supported in our analysis, with Chiloglanis macropterus nested within a Chiloglanis sublineage encompassing only other taxa from the Congo drainage. This is an important result because the phylogenetic relationships of C. macropterus have been controversial in the past, and because we and other researchers assumed that this species would be resolved as sister to most or all other members of Chiloglanis. The apparent paraphyly of Synodontis with respect to Microsynodontis provided an additional surprise, with Synodontis punu turning out to be the sister group of the latter genus.

INTRODUCTION

The order Siluriformes (Superorder: Ostariophysi; Series: Otophysi), commonly known as catfishes, represents one of the most diverse vertebrate groups on our planet, with currently 4019 valid described species and 500 genera assigned to 39 families (Fricke et al., 2021). However, the exact number of catfish families remains debated because of conflicting views about familial classifications (Betancur‐R. et al., 2017; Diogo & Peng, 2010; Ferraris Jr., 2007; Nelson et al., 2016; Sullivan et al., 2006). Nevertheless, the monophyly of the Siluriformes is well supported by morphological and molecular data (Betancur‐R. et al., 2017; Fink & Fink, 1981, 1996; Hughes et al., 2018). Systematists typically recognize three major clades within the siluriform catfishes, two of which (Loricarioidei and Diplomystoidei) are restricted to the freshwater systems of South America, whereas the third (Siluroidei) is globally distributed and contains some families that have secondarily adapted to brackish and even marine environments (Betancur‐R, 2009; Covain & Fisch‐Muller, 2007; Diogo, 2004; Stange et al., 2016). Although the order has a global distribution, many catfish families occur on only one continent, and are sometimes restricted to a comparatively small geographic area. For example, the family Diplomystidae is endemic to southern South America, the family Austroglanididae is confined to Southern Africa, and the recently described family Lacantuniidae inhabits only Central America (Armbruster, 2011; Rodiles‐Hernandez et al., 2005; Skelton, 2001). Because of their Pangaean origin probably dating back to the Early Cretaceous, siluriform catfishes became important model systems for testing key biogeographic questions (Kappas et al., 2016; Rivera‐Rivera & Montoya‐Burgos, 2017; Sullivan et al., 2006). Considerable progress has been made over the past two decades to reconstruct the evolutionary history of the Siluriformes through use of different molecular markers (Arcila et al., 2017; Betancur‐R. et al., 2017; Hardman, 2005; Kappas et al., 2016; Nakatani et al., 2011; Sullivan et al., 2006). These studies have helped to shed some light on relationships within some superfamily and multifamily clades, for example by providing consistent support for the monophyly of the ‘Big Asia clade’ and ‘Big Africa clade’ sensu Sullivan et al. (2006). However, many interfamilial relationships remain poorly resolved and hence controversial.

Uncertainties also still persist for intrafamilial relationships, even for some of the comparatively well‐studied catfish families such as Mochokidae. This family is endemic to Africa and belongs to the ‘Big Africa clade’ (sensu Sullivan et al., 2006). It is the most species‐rich catfish family on the continent and it is currently represented by 224 valid species (Fricke et al., 2021). Their largest density of species occurs in the Congo basin, but mochokids are ubiquitous in all tropical African river drainages and basins. Synodontis Cuvier, 1816 has radiated into several species flocks within the East African Great Lakes (Day et al., 2009, 2013; Koblmüller et al., 2006). The oldest mochokid fossil, which is tentatively assigned to Synodontis, dates to the Early Oligocene and was found in Oman, suggesting that the family's distribution was wider in the past (Otero & Gayet, 2001).

Following the most recent taxonomic revision by Vigliotta (2008) and the description of the genus Atopodontus Friel & Vigliotta, 2008, nine mochokid genera are currently recognized. Three of these, Mochokiella Howes, 1980, Acanthocleithron Nichols & Griscom, 1917 and Atopodontus are monotypic, whereas two, Synodontis and Chiloglanis Peters, 1868, are extremely species‐rich. Two subfamilies are recognized. These are Chiloglanidinae Riehl & Baensch, 1990, which is characterized by species with lips and barbels modified into an oral disc (suckermouth), and Mochokinae Jordan, 1923, which comprises species without suckermouths (Seegers, 2008). A comprehensive morphological investigation of all mochokid genera revealed strong support for the monophyly of Chiloglanidinae, but recovered the Mochokinae as paraphyletic (Vigliotta, 2008). That study also recovered a monophyletic group encompassing the chiloglanidin genera Microsynodontis Boulenger, 1903 and Synodontis, and suggested the paraphyly of Chiloglanis and Euchilichthys Boulenger, 1900. Interestingly, the single representative of Atopochilus Sauvage, 1879 in this study nested phylogenetically within Euchilichthys, while Chiloglanis macropterus Poll & Stewart, 1975 was recovered as a sister group to all remaining chiloglanidin taxa.

Mochokids have also attracted considerable attention in biogeographical research because of their almost pan‐African distribution. The species‐rich genus Synodontis provides a particularly excellent model for testing the role of geological processes in promoting lineage diversification and shaping present‐day biogeographic patterns on a continental scale (Day et al., 2009, 2013; Pinton et al., 2013). Furthermore, a growing number of studies focus on mochokid biogeography at a regional scale, most notably within the species‐rich and widely distributed genus Chiloglanis (Chakona et al., 2018; Morris et al., 2016; Schmidt et al., 2014, 2016; Watson, 2020).

The discovery of many endemic and often cryptic candidate species of Chiloglanis underscores the fact that our knowledge of mochokid diversity and phylogenetic relationships is still incomplete. This gap persists mainly because previous phylogenetic studies focused only on Synodontis or Chiloglanis, or considered only higher‐level relationships of Siluriformes and included relatively few mochokid taxa. Day et al. (2013) achieved the most complete mochokid taxon sampling so far. This study of continental diversification within Synodontis used a molecular dataset with four genes and included representatives of most known mochokid genera as outgroups. Only Atopodontus and Acanthocleithron were missing. Interestingly, the generic interrelationships recovered from this study contradicted those inferred from the morphological data of Vigliotta (2008) in some respects. Most importantly, Day et al. (2013) recovered the monophyly of the Mochokinae, with Mochokiella being the sister group of Synodontis, whereas in Vigliotta's (2008) study Mochokus followed by Mochokiella were recovered as lineages originating at the oldest splits within Mochokidae.

To date, the phylogenetic positions of Atopodontus and Acanthocleithron remain unknown. To our knowledge no DNA sequences of Atopodontus have been published but there is a single COX1 barcode sequence for Acanthocleithron chapini Nichols & Griscom, 1917 available. That sequence was generated in a barcoding project focusing on the ichthyological diversity of the north‐eastern Congo basin (Decru et al., 2016) and has never been included in an evolutionary analysis.

Similarly, the South African family Austroglanididae remains one of the most enigmatic clades in terms of its phylogenetic placement because no comprehensive molecular phylogeny of the order has included it (Kappas et al., 2016; Rivera‐Rivera & Montoya‐Burgos, 2018; Sullivan et al., 2006). The only available genetic data for this family originated from a pair of studies that focused on population structure and the relationships among the three species of Austroglanis. Skelton et al. (1984) used mitochondrial markers (cytochrome b and 16 S) for this purpose while Cunningham, Bills & Swartz (unpublished data, as cited in Bills and Impson (2013)) employed allozymes. In contrast, several cladistic studies aimed to elucidate the placement of this catfish family using morphological data, but the resulting phylogenetic hypotheses were diverse and often contradictory. For example, Skelton et al. (1984) inferred that Austroglanis was closely related to the family Bagridae (which at that time included the family Claroteidae), whereas Mo (1991) inferred a close relationship with the family Cranoglanididae (cladogram II) and raised the genus to family level. Subsequently, de Pinna (1993) recovered a sister group relationship of Austroglanididae with a large clade encompassing several families, including Ariidae, Anchariidae, Bagridae, Claroteidae, Schilbeidae, Pangasiidae, Pimelodidae and the genus Horabagrus Jayaram, 1955. Finally, Diogo (2005) and Diogo and Bills (2006) suggested a close relationship between Austroglanididae and the families Ariidae, Claroteidae, Ictaluridae and Cranoglanididae based on four derived morphological characters.

This present study aimed to resolve the unsettled phylogenetic question regarding the placement of Austroglanididae within Siluriformes and to improve the understanding of the inter‐ and intrageneric relationships of Mochokidae. To achieve these tasks, we took advantage of the constantly growing body of mitochondrial genome data available for Siluriformes (i.e., Kappas et al., 2016; Ma et al., 2015; Moreira et al., 2017; Nakatani et al., 2011) which we combined with newly sequenced partial mitochondrial genomes of representative taxa of all mochokid genera as well as of all described Austroglanis species. Furthermore, we compiled a small nuclear dataset (rag 1 and rag 2) based on the dataset provided by Sullivan et al. (2006) and newly obtained sequence data of the family Austroglanididae. We used these data to reconstruct the most comprehensive phylogeny of Mochokidae to date, showing the intrageneric relationships of all currently valid mochokid genera, including the previously omitted Atopodontus and Acanthocleithron. The reconstruction also confirms Mochokidae's position within the ‘Big Africa clade’ and infers the placement of Austroglanididae within Siluriformes for the first time.

MATERIALS AND METHODS

Taxon sampling

We included mitogenome data of 256 specimens (approx. 239 species) composed of concatenated alignments of all mitochondrial protein coding genes (see Supporting Information Table S1; hereafter referred to as the ‘all‐catfish dataset’). The taxon sampling for newly generated sequences targeted mainly the catfish clade ‘Big Africa’ (sensu Sullivan et al., 2006), with a specific focus on Mochokidae (40 specimens, 32 species) and its two subfamilies Mochokinae (14 specimens/species) and Chiloglanidinae (26 specimens, 18 species). This is the first dataset to ever include representatives of all nine mochokid genera. Furthermore, the taxon sampling includes representative taxa of all nominal African catfish families associated with the ‘Big Africa clade’: Auchenoglanididae (N = 2), Amphiliidae (N = 2), Claroteidae (N = 4), Schilbeidae (N = 4) and Malapteruridae (N = 1). Unfortunately, we were unable include Lacantunia enigmatica Rodiles‐Hernández, Hendrickson & Lundberg 2005 in our mitogenome taxon sampling, the only member of the Mesoamerican family Lacantuniidae and a member of the ‘Big Africa clade’ (Lundberg et al., 2007). In addition, we sequenced partial mitochondrial genomes of all three described species of the phylogenetically unplaced family Austroglanididae. A comprehensive overview of the newly sequenced catfish specimens for this study (52 specimens and 43 species, see Supporting Information Table S1), including photos of either of living or of preserved specimens and X‐rays, is available in Supporting Information S1.

The all‐catfish dataset was compiled by downloading all available mitochondrial genomes for the catfish sister group order Gymnotiformes (N = 8) and Siluriformes, including representative members of all three major catfish lineages: Loricarioidei (N = 40), Siluroidei (N = 157) and Diplomystoidei (N = 1). All taxon names were checked for current validity using Eschmeyer's Catalogue of Fishes (Fricke et al., 2021). In cases where more than one mitochondrial genome was available for a species, we randomly selected one representative sequence to be included in our data set (see Supporting Information Table S1 for GenBank accession numbers). Some sequences/taxa obtained from GenBank were identified as of doubtful quality or identity and excluded after a preliminary phylogenetic analysis due to their highly unlikely phylogenetic position (for further details see Supporting Information S2).

Furthermore, we downloaded all available COX1 sequence data for the family Mochokidae from GenBank (646 sequences, see Supporting Information Table S2) and extracted the COX1 sequence data of the partial mitochondrial genomes of the newly sequenced mochokids. This second data set (hereafter referred to as the ‘mochokid COX1 dataset’) included 688 specimens (representing approximately 105 mochokid species, including several undetermined and/or undescribed species) and covered much of the natural distribution of the family. This COX1 dataset enabled us to investigate the taxonomic assignment of selected mochokid taxa, such as Chiloglanis niloticus Boulenger, 1900 and Synodontis punu Vreven & Milondo, 2009.

In addition to the mitochondrial data, we obtained partial sequences for the nuclear recombination activating genes rag1 and rag2 from GenBank. This nuclear dataset is mainly based on the data provided by Sullivan et al. (2006) and included 114 Siluriformes, to which we added new sequence data of all three Austroglanis species (see Supporting Information Table S2). The nuclear dataset includes two species of Gymnotiformes and one characiform as outgroups. This third DNA dataset provided an opportunity to investigate phylogenetic relationships of the family Austroglanididae with more slowly evolving nuclear markers. We refrained from constructing a combined dataset of nuclear and mitochondrial data because matching mitochondrial and nuclear data were available for only for 29 species, and thus the combined matrix would have contained a very high proportion of missing data.

Sampling procedures

The tissues used in this study were collected on various field expeditions and sourced from available museum specimens. Fish were caught using various techniques (i.e., fish traps, frame and dip nets, gillnets and beach seines) depending on the sampled habitats and local conditions or were obtained from local fishermen. Freshly caught fish were sacrificed by an overdose of approved fish anaesthetic (i.e., MS222) and photographed in a cuvette (most specimens). Fin clips or muscle plugs were taken and immediatley preserved in 96% ethanol, and corresponding specimens were fixed in formalin (4% formaldehyde in aqueous solution) and later transferred into 75 % ethanol for long‐term storage following Neumann (2010). All relevant ethical standards and applicable national laws for collecting, sampling and export of specimens were considered. Tissue samples and corresponding export documentation for most investigated specimens are stored at the SNSB‐bavarian state collection of zoology (SNSB‐ZSM). Those for Atopodontis adriaensi, Atopochilus savorgnani and Synodontis punu are available from oregon state university ichthyology collection (OS) and were collected with all appropriate permits following Animal Care and Use Protocol #4909 from Oregon State University. Samples and specimens of Austroglanis are deposited in the National Fish Collection Facility at the NRF‐SAIAB. The sampling approaches that were used for collecting the corresponding Austroglanis samples were approved by the National research foundation – South African institute for aquatic biodiversity (NRF‐SAIAB) animal ethics committee. Research permits were issued by Cape Nature and Northern Cape authorities and are deposited at the National Fish Collection Facility at the NRF‐SAIAB. See Acknowledgements for more detailed permit information.

Molecular methods and phylogenetic analysis

Partial mitochondrial genomes were sequenced by the Sequencing Service of the Ludwig Maximilian University of Munich on an Illumina MiSeq platform (MiSeq Reagent Kit v2; 2X250). Individual library preparations were based on two different approaches. First, total genomic DNA was extracted for all samples following a custom cetyltrimethylammonium bromide (CTAB) DNA extraction protocol and DNA concentrations were quantified using a spectrophotometer (NanoDrop ND‐1000, Thermo Scientific, Wilmington, USA) and adjusted to 25 ng μl−1 per sample. Subsequently, the libraries were prepared following the first approach, which was based on the amplification of a large fragment of the mitochondrial genome (~13,000 bp, including all mitochondrial protein coding genes) using the TaKaRa LA Taq DNA (Takara Bio Inc., Shiga, Japan) polymerase kit and with the following primer pair (taken from Abwe et al., in prep): Amp2571F (TTC AAC GAT TAA AGT CCT ACG TGA TCT GAG) and Amp15543R (TTT AAC CTT CGA TCT CCG GAT TAC AAG AC). Furthermore, we used a modified thermal profile from Schedel et al. (2019) for the amplification reactions: initial denaturation at 98°C (60 s), followed by 35 cycles of denaturation 98°C (10 s), annealing at 58°C for 60s, elongation at 68°C (15 min) and a last extension step at 72°C (10 min). Successfully amplified PCR products were excised from agarose gels and purified using the GeneJET Gel Extraction Kit (Thermo Fisher Scientific) following the manufacturer's protocol. Subsequently, DNA concentrations of cleaned amplification products were adjusted to 0.21 ng μl−1 per sample and individual libraries were prepared using the Nextera XT DNA Sample Preparation Kit (Illumina Inc, San Diego, CA, USA) following the manufacturer's protocol until to the library normalization step. The latter was conducted by pooling all libraries equimolarly based on the DNA concentration and the size distribution of the individual libraries, resulting in a final pool containing the ‘amplified partial mitochondrial genomes’.

The second library preparation approach was only applied for selected samples for which the long‐range PCR failed. In those cases, we used directly the extracted DNA (adjusted to 0.21 ng μl−1 per sample) as input template for the Nextera XT DNA Sample Preparation Kit (Illumina), which was conducted as described above. The resulting shotgun libraries were pooled equimolarly as described above. Once sequenced we used Geneious v.11.0.4 (Kearse et al., 2012) and the plugin BBDuk Trimmer for quality control, and for adaptor trimming of the demultiplex reads. We de novo assembled the amplified partial mitochondrial genomes using the ‘De Novo Assembly’ function implemented in Geneious. Sequence reads derived from the shotgun libraries were first mapped against the complete mitochondrial genome of Synodontis schoutedeni (GenBank accession number: AP012023.1; Nakatani et al., 2011) and the resulting contigs were subsequently used for further analysis. The assembled contigs were finally annotated using the S. schoutedeni mitochondrial genome as a reference and uploaded to GenBank (GenBank accession numbers: MZ930069‐MZ930120; see Supporting Information Table S1).

Mitochondrial analysis

To prepare a protein‐coding gene alignment we extracted from newly generated and GenBank mitogenome data the sequence information of all mitochondrial protein‐coding genes (ND1, ND2, COX1, COX2, ATP8, ATP6, COX3, ND3, ND4L, ND4, ND5, ND6, CytB) of all specimens included in our all‐catfish dataset (N = 256). Subsequently, we aligned sequences of individual genes using the Geneious alignment tool (default settings). In a few cases where the sequence information for an individual gene was incomplete or missing (i.e., due to poor sequence quality) a multi‐N string was inserted in the respective position of the corresponding alignment. Finally, single gene alignments were concatenated resulting in an alignment of 11,400 bp with relative base frequencies of A = 28.8%, T = 28.6%, C = 15.1% and G = 27.4% and about 0.5% of missing data (excluding gaps and ambiguous sites). We assessed the extent of substitution saturation of our alignment using Xia's test (Xia et al., 2003; Xia & Lemey, 2009) and plotted the observed number of transitions and transversions against the corrected genetic distances using the program DAMBE v.7.2.152 (Xia, 2018). The program jModelTest (Posada, 2008), implemented on the CIPRES Science Gateway (Miller et al., 2010), was used to calculate the best substitution model based on the Akaike information criterion: GTR + I + G. RAxML v.8.2.12 (Stamatakis, 2014), as implemented on the CIPRES Science Gateway was used for maximum likelihood (ML) inference of phylogenetic relationships by further partitioning the dataset according to first, second and third codon positions, and assigning all included gymnotiform taxa as outgroups based on evidence from previous phylogenetic studies (Arcila et al., 2017; Betancur‐R. et al., 2017; Betancur‐R et al., 2019; Hughes et al., 2018). Since the third codon position of the all‐catfish dataset showed signatures of saturation we also ran an ML analysis on the all‐catfish dataset using the same parameters but excluding the third codon position to explore whether the saturation affected the inference.

Furthermore, a Bayesian Inference (BI) analysis was conducted for the all‐catfish dataset using MrBayes 3.2.7a (Huelsenbeck & Ronquist, 2001; Ronquist et al., 2012). Two Markov Chain Monte Carlo (MCMC) runs were calculated simultaneously for 15 million generations, with tree space sampled every 1000th generation. We discarded the first 3.75 million generations as burn‐in before checking the convergence of all parameters, that is, that all estimated sample size (ESS) values were above 200. The remaining trees were then used to calculate the consensus tree.

We also aligned all available COX1 sequences of the family Mochokidae and representative taxa of the family Amphiliidae (mochokid COX1 dataset) resulting in alignment of 1375 bp and 688 specimens. An ML analysis was conducted on this alignment in RAxML using the same substitution model and partition scheme as for the all‐catfish dataset. Amphiliidae sequences were used as the outgroup taxa.

Nuclear analysis

In a final step, we mapped the demultiplexed, trimmed reads of the four shotgun‐sequenced Austroglanis libraries (~4.326.187 reads per specimen) against two nuclear markers which were previously used to infer phylogenetic relationships among Siluriformes (see Sullivan et al., 2006): rag1 (exon 1, 2 and 3) and rag2. For doing so we used the ‘Map to Reference’ option in Geneious v.11.0.4 and, as references, sequences of the corresponding markers [GenBank accession numbers: rag1 (exons 1 + 2): DQ492636, rag1 (exons 3): JN020097, rag2: JN020130] derived from different Pangasius Valenciennes 1840 (in Cuvier & Valenciennes, 1840) species. This choice was based on the phylogenetic analysis conducted on our all‐catfish dataset, which recovered Pangasiidae as sister to Austroglanididae (see below). The resulting contigs were extracted and aligned to the corresponding sequence data obtained from GenBank (see Supporting Information Table S2). For both gene alignments we first obtained neighbour‐joining trees calculated using the Geneious tree builder. Then we concatenated both alignments and inferred phylogenetic relationships using RAxML. As in the other analyses we partitioned the dataset into first, second and third codon positions, and defined the included gymnotiform taxa as outgroup. The newly obtained rag1 and rag2 sequences are available on GenBank under the accession numbers MZ895066–MZ895072 (see Supporting Information Table S2).

RESULTS

Partial mitogenomes and nuclear makers

Amplification of partial mitochondrial genomes and subsequent sequencing on an Illumina MiSeq platform was successful for 44 of the 52 specimens considered herein, yielding for most samples a single contig sequence; in a few cases two sequences and in a single case three contigs were retrieved (mean coverage 400, mean sequence length 10,704 bp). The remaining eight samples, including four mochokids and all four studied Austroglanis samples, were shotgun‐sequenced from total genomic DNA. These efforts yielded an average of 2.75 million raw reads per sample of which approximately 0.05% mapped against the mitochondrial reference genome, returning between two and four contig sequences per sample. Almost complete mitochondrial genomes were recovered for Chiloglanis sp. Nigeria and Austroglanis gilli (Barnard, 1943) (mean coverage 52, mean sequence length 5832 bp).

Mapping of shotgun‐sequenced Austroglanis reads against the nuclear markers was only partially successful. All four Austroglanis samples reads could be mapped against the reference sequence of rag1 (exon 1, 2) thereby yielding four contig sequences (mean coverage 3, mean sequence length 1374 bp) but no reads mapped against the reference sequence for rag1 (exon 3). Therefore, we decided to exclude the rag1 (exon 3) from downstream analysis. Mapping against the reference sequence of rag2 was successful for three of the four Austroglanis samples, yielding three contig sequences (mean coverage 3.4, mean sequence length 933 bp).

Siluriform phylogenetic relationships based on partial mitochondrial genomes

The ML analysis (Figures 1 and S1) based on all mitochondrial protein coding genes (all‐catfish dataset) strongly supported the monophyly of the Siluroidei (BS: 100) but the monophyly of the Loricarioidei was only moderately supported (BS: 80). Diplomystoidei was recovered as the sister group to the Siluroidei, albeit with low support (BS: 61). A weakly supported clade (BS: 31) composed of the families Plotosidae, Chacidae, Cetopsidae and Ritidae was recovered as originating at the earliest split within Siluroidei.

FIGURE 1

ML phylogeny (RAxML) of the order Siluriformes, including 30 catfish families (indicated either in blue or in red = target families), based on all mitochondrial protein coding genes (11,400 bp). Bootstrap proportion based on 1000 BS replicates are indicated at nodes either by numbers or by black dots (BS = 100). Genera including more than one taxon are collapsed (see Supporting Information Figure S1 for the noncollapsed tree). Representative species of most included catfish families/subfamilies are depicted from top to bottom (photographers in brackets, specimen/species not necessarily included in the dataset): Mochokidae; Chiloglanidinae: Chiloglanis sp. Ntumbachushi (F.D.B. Schedel), Mochokidae; Chiloglanidinae: Synodontis petricola (F.D.B. Schedel), Claroteidae: Chrysichthys (F.D.B. Schedel), Auchenoglanididae: Parauchenoglanis cf. monkei (F.D.B. Schedel), Schilbeidae: Schilbe grenfelli (F.D.B. Schedel), Malapteruridae: Malapterurus sp. (F.D.B. Schedel), Amphiliidae: Amphilius sp. (F.D.B. Schedel). Ictaluridae: Ameiurus nebulosus (F.D.B. Schedel), Pangasiidae: Pangasius pangasius (E. Schraml), Austroglanididae: Austroglanis barnardi (Roger Bills), Ariidae: Ariopsis cf. guatemalensis (preserved specimen; F.D.B. Schedel), Siluridae: Silurus glanis (E. Schraml), Pimelodidae: Sorubim lima (E. Schraml), Pseudopimelodidae: Lophiosilurus alexanderi (E. Schraml), Sisoridae: Glyptothorax cf. sinensis (E. Schraml), Sisoridae: Hara jerdoni (E. Schraml), Amblycipitidae: Liobagrus reinii (E. Schraml), Bagridae: Tachysurus fulvidraco (F. Schäfer), African Bagridae: Bagrus sp. (F.D.B. Schedel), Horabagridae: Horabagrus brachysoma (J. Geck), Clariidae: Clarias gariepinus (F.D.B. Schedel), Heteropneustidae: Heteropneustes fossilis (preserved specimen, F.D.B. Schedel), Auchenipteridae: Tetranematichthys wallacei (E. Schraml), Doradidae: Platydoras armatulus (E. Schraml), Aspredinidae: Bunocephalus coracoideus (E. Schraml), Plotosidae: Plotusus lineatus (E. Schraml), Chacidae: Chaca bankanensis (preserved specimen, F.D.B. Schedel), Cetopsidae: Cetopsis coecutiens (preserved specimen, F.D.B. Schedel), Diplomystidae: Diplomystes sp. (preserved specimen, F.D.B. Schedel) Loricariidae: Hypostominae: Hypostomus cf. plecostomus (E. Schraml) Loricariidae: Hypoptopomatinae: Otocinclus sp. (J. Geck), Callichthyidae: Corydoras rabauti (E. Schraml)

As in previous phylogenetic studies (i.e., Arcila et al., 2017; Betancur‐R. et al., 2017; Kappas et al., 2016; Silva et al., 2021; Sullivan et al., 2006), several multifamily clades were recovered, including (a) a clade (BS: 73) composed of Aspredinidae + Doradoidea (Doradidae + Auchenipteridae), (b) Clarioidea (Clariidae + Heteropneustidae; BS: 100), (c) Pimelodoidea (Pimelodidae + Pseudopimelodidae; BS: 100, note that the family Heptapteridae and the genus Conorhynchos Bleeker, 1858 are missing in our analysis), (d) Ictaluroidea (Ictaluridae + Cranoglanididae (BS: 100), (e) the informal clade ‘Big Asia’ (BS: 100) and (f) the informal clade ‘Big Africa’ (BS: 78). Furthermore, we recovered a clade (BS: 70) encompassing Ictaluroidea, Ariidae, Pangasiidae and Austroglanididae as the sister group of the ‘Big Africa clade’ again with moderate support (BS: 71). Within this clade the clade comprising all Austroglanididae was recovered as sister to Pangasiidae with moderate support (BS: 72). Interrelationships of these siluroid multifamily clades were only poorly supported, as in previous studies focusing on the reconstruction of siluriform interfamily relationships (i.e., Kappas et al., 2016; Sullivan et al., 2006).

Apart from the megadiverse family Loricariidae (BS: 38), the monophyly of most catfish families was highly supported (BS: >95). One unexpected discovery was that Eutropiichthys vacha (Hamilton, 1822) nests within the pangasid genus Pangasianodon Chevey, 1931, which contradicts the current taxonomic assignment of Eutropiichthys to the family Schilbeidae (Wang et al., 2016). Furthermore, our analyses recovered paraphyletic or polyphyletic genera within the Loricariidae, Bagridae, Sisoridae, Siluridae, Ariidae, Pangasiidae, Schilbeidae and Mochokidae (see Supporting Information Figure S1).

Results excluding saturated third codon positions

The topology of the ML analysis (see Supporting Information Figure S2) based on the all‐catfish dataset but with the third codon position removed from the alignment was widely congruent with that of the full dataset. Nevertheless, several topological differences concerning interfamily and multifamily clade relationships are detectable between the two analyses. For example, the multifamily clade (including the Plotosidae, Chacidae, Cetopsidae and Ritidae) recovered as originating at the earliest divergence within Siluroidei based on the full dataset was not recovered in this analysis. Instead, Ritidae and Ariidae were reconstructed as successive sister groups to all remaining members of Siluroidei. Several traditionally recognized multifamily clades were recovered with lower support compared to the full data set, that is, the ‘Big Africa clade’ (BS: 42 vs. 71); others were not recovered at all, that is, the ‘Big Asia clade’. Loricariidae, Doradidae and Auchenipteridae were recovered as paraphyletic or polyphyletic. Austroglanididae was recovered as sister group to Ictaluroidea, but with low support (BS: 32). The recovered intrageneric and intergeneric relationships within Mochokidae were generally identical to those of the full dataset but often with lower BS support (i.e., that of Chiloglanis niloticus). However, see below for some important exceptions. These comparisons suggest that although third codon positions in the all‐catfish dataset show signatures of saturation, those loci still carry critical phylogenetic information, especially at the intrafamilial level.

Congruence of ML and BI analyses

Likewise, the topology of the BI analysis (see Supporting Information Figure S3) based on the all‐catfish dataset was widely congruent with that of the ML analysis with respect to intrafamilial relationships. The monophyly of the three superfamilies Siluroidei, Diplomystoidei and Loricarioidei and their relationships to each other were identical to the results of the ML analysis. Both analyses recovered the major multifamily clades, but the interrelationships among these clades showed some differences. For example, the BI analysis obtained the ‘Big Asia clade’ as sister to a clade encompassing the ‘Big Africa clade’ and the multifamily clade encompassing Ictaluroidea, Ariidae, Pangasiidae and Austroglanididae.

Phylogenetic relationships within Mochokidae

The monophyly of Mochokidae (see Figure 2) was highly supported (BS: 96) by our ML analysis based on the all‐catfish dataset. While our analysis strongly supported the monophyly of the subfamily Chiloglanidinae (BS: 100), it only weakly supported that of the Mochokinae (BS: 64). Within Chiloglanidinae, the genus Chiloglanis was recovered as sister group to all remaining chiloglanidin genera (Atopodontus, Euchilichthys, Atopochilus). Acanthocleithron was recovered as sister group to a strongly supported clade encompassing all other mochokin taxa (BS: 100). Within this clade the first split separated Mochokus from all other lineages. Unexpectedly, Synodontis punu was recovered as sister to Microsynodontis, rendering the genus Synodontis polyphyletic. This finding was congruent with the ML analysis of the mochokid COX1 dataset which included additional specimens of S. punu as well as of Microsynodontis (see Supporting Information Figure S4).

FIGURE 2

Detailed relationships among the ‘Big African clade’ (pruned from the ML phylogeny depicted in Figure 1). Bootstrap proportion based on 1000 BS replicates are indicated at nodes either by numbers or by black dots (BS = 100). Represented mochokid species depicted from top to bottom (photographers and corresponding sample ID in brackets): Chiloglanis macropterus (F.D.B. Schedel, DRC‐2012/3637), Chiloglanis sp. ‘Lufupa’ (F.D.B. Schedel, DRC‐2016/4684), Chiloglanis sp. ‘Ntumbachushi’ (F.D.B. Schedel, DRC‐2012/3568), Euchilichthys sp. Mukuleshi (F.D.B. Schedel, DRC‐2012‐4612), Atopochilus savorgnani (B. Sidlauskas, GAB17‐365), Atopodontus adriaensi (B. Sidlauskas, GAB17‐364), Synodontis petricola (F.D.B. Schedel, P‐AA‐1422), Mochokiella paynei (F.D.B. Schedel, P‐AA‐1344), Microsynodontis cf. batasii (F.D.B. Schedel, P‐AA‐1504), Mochokus brevis (F.D.B. Schedel, SUD‐2016‐847), Acanthocleithron chapini (E.J.W.M.N. Vreven, P‐AA‐1382)

Phylogenetic relationships of Austroglanididae

The ML analysis based on the mitochondrial all‐catfish dataset (Figure 3), as well as the one based on the nuclear dataset (Figure 4), recovered Austroglanis sclateri (Boulenger, 1901) as sister to the remaining species of Austroglanididae. Austroglanis gilli and Austroglanis barnardi (Skelton, 1981) were consistently recovered as sister taxa in all analyses with high support (BS: >98, see also Supporting Information Figures S1 and S2). The mitochondrial data supported a close relationship of Austroglanididae with Pangasiidae (Asia) and Ictaluroidea (North America & Asia). Together the three families form a weakly supported clade (BS: 61) which was recovered to be the sister group to Ariidae (BS: 70). The BI analysis (Supporting Information Figure S3) recovered Austroglanididae instead as sister group to Ariidae although with weak support (BPP: 0.62), and the Pangasiidae as sister group of this clade, again with low support (BPP: 0.7). Together with the Ictaluroidea these three families formed a well‐supported clade (BPP: 1). These results contrast with our ML analysis of the nuclear dataset which recovered a close relationship between Austroglanididae and the southeast Asian family Ritidae, albeit with very weak support (BS: 32).

FIGURE 3

Detailed relationships of the family Austroglanididae and related families (pruned from the ML phylogeny depicted in Figure 1). Bootstrap proportion based on 1000 BS replicates are indicated at nodes either by numbers or by black dots (BS = 100)

FIGURE 4

ML phylogeny (RAxML) of the order Siluriformes (focal families are indicated by pictures of representative species) based on the ‘nuclear dataset’, including the sequence data of rag1 and rag2 provided by the study of Sullivan et al. (2006) and the newly created sequences of four Austroglanis specimens (117 individuals, 1983 bp). Bootstrap proportions based on 1000 BS replicates are indicated at nodes either by numbers or by black dots (BS = 100)

DISCUSSION

Mochokidae: Phylogenetic placement, intrageneric relationships and taxonomic implications

This study tentatively identifies the family Mochokidae as the sister to the clade containing the Claroteidae, Auchenoglanididae, (African) Schilbeidae and Malapteruridae. The Amphiliidae were recovered as originating at the earliest split within the ‘Big Africa clade’ (Figure 1). These placements partially conflict with the results of Kappas et al. (2016), which is the only other large‐scale phylogenetic study of Siluriformes based on mitochondrial genomes. Kappas et al. (2016) recovered a sister relationship of Mochokidae to a clade encompassing the Schilbeidae and Auchenoglanididae (the later referred in the corresponding study as Claroteidae; BS: 60), while the Malapteruridae formed a sister group to the Amphiliidae (BS: 56). As in the present study, interfamilial relationships within the ‘Big Africa clade’ were rather weakly supported, indicating that the phylogenetic information carried by the mitochondrial genome might not suffice to resolve those relationships. However, the topological differences could also originate from the low overall taxon sampling for the ‘Big Africa clade’ in the study by Kappas et al. (2016), which lacked members of Claroteidae entirely. Likewise, the phylogeny obtained by Sullivan et al. (2006), based on two nuclear genes and including representatives of all catfish families currently associated with the ‘Big Africa clade’, differs in some aspects from our mitogenomic phylogeny. For example, Sullivan et al. (2006) recovered Mochokidae as sister group to a clade encompassing Malapteruridae and Amphiliidae (BS: 99), with the remaining members of the ‘Big Africa clade’ forming the sister to that clade (BS: 99). However, interfamily relationships were rather poorly supported in this study too. Though other published multilocus studies have reconstructed relationships using many more genes (i.e., Arcila et al., 2017; Betancur‐R. et al., 2017; Rivera‐Rivera & Montoya‐Burgos, 2018) and could potentially allow for further insights into interfamilial relationships, all these efforts included very few representatives of the ‘Big Africa clade’. Thus, DNA‐based results to date do not unambiguously favour any of the alternative hypotheses about the exact placement of Mochokidae within the order.

Neither has morphological evidence proved sufficient in resolving this conundrum. Several authors (i.e., de Pinna, 1993, 1998; Diogo, 2005; Vigliotta, 2008) have suggested a sister relationship between Mochokidae and Doradoidea based on morphological similarities of the compound centrum, elements at the junction between the dorsal fin and the neurocranium and the nuchal shield. However, Vigliotta (2008) argued that these hypothesized synapomorphies grouping Doradoidea and Mochokidae might represent striking examples of homoplasies. Alternatively, Vigliotta (2008) tentatively suggested some synapomorphies for a clade encompassing Malapteruridae, Amphiliidae and Mochokidae, namely the absence of the ascending Meckel's cartilage (although present in the amphiliid subfamily Leptoglanidinae) and a characteristically shaped autopalatine. No study has yet rigorously tested these alternative hypotheses using morphological data alone or in a total evidence approach.

Given the conflicting reconstructions yielded by molecular and morphological data alike, it is premature at this point to draw any further conclusions about the interfamilial relationships within the ‘Big Africa clade’. The best path forward will involve analysis of a more comprehensive dataset with deeper sampling of the nuclear genome and even broader taxon sampling. For example, it will be critical to include the Mesoamerican Lacantunia, which currently is assumed to be the only non‐African representative in the ‘Big Africa clade’ (Lundberg et al., 2007). Likewise, revisiting the morphological data in light of and in combination with the molecular data has the potential to enhance tests of alternative hypotheses, reveal additional synapomorphies and test hypotheses about character evolution in this diverse clade of fishes.

Despite the need to investigate deep phylogenetic relationships with more encompassing nuclear and morphological data, analysis of the mitochondrial dataset still yielded substantial insight about intrafamilial relationships. Our phylogenetic reconstruction based on a large fraction of the mitochondrial genome included for the first time all mochokid genera, and tentatively supports the postulated monophyly of Mochokinae and Chiloglanidinae sensu Seegers (2008). Within subfamilies, intrageneric relationships were all strongly supported, except for the phylogenetic placement of Acanthocleithron at the earliest split within Mochokinae. Overall, our intergeneric topology is largely congruent with the results of Day et al. (2013), which specifically focused on the genus Synodontis and did not include Acanthocleithron and Atopodontus. The limited taxon sampling outside of Synodontis, might explain why Day et al. (2013) recovered Microsynodontis rather than Mochokus as the sister group of a clade encompassing Mochokiella and Synodontis.

The genus Acanthocleithron

The phylogenetic placement of Acanthocleithron within Mochokidae remains elusive due to the weak support for its origination at the earliest split within Mochokinae (BS:64). The difficulty of this reconstruction might result from ancient incomplete lineage sorting during rapid early cladogenesis at the base of the Mochokidae. As such, it is possible that the Acanthocleithron lineage is even older, potentially originating at the earliest split within the entire family. Such a scenario would render the Mochokinae as currently defined (sensu Seegers, 2008) paraphyletic and further suggest a Congo basin origin for the family Mochokidae. However, this hypothesis has not yet been critically tested.

Specimens of Acanthocleithron are rare in ichthyological collections and apart from the few specimens known from the Ituri/Aruwimi drainage, a tributary of the Congo basin from which A. chapini was originally described (Seegers, 2008), specimens of A. chapini have been collected at Bamu Island (Pool Malebo) and at Yangambi (Poll, 1959; Poll & Gosse, 1963). Although Acanthocleithron is currently monotypic, Seegers (2008) speculated that specimens collected outside of the Ituri/Arwuimi drainage might not be conspecific with A. chapini but rather represent undescribed species of Acanthocleithron. He based that hypothesis on differences in coloration and morphological features, particularly differences in the serration of the pelvic fin spine. For a better understanding of the evolutionary history of this little‐studied genus and to help stabilize the reconstruction of relationships within Mochokidae, it might be essential to include these enigmatic Acanthocleithron populations in upcoming molecular analysis.

The genus Chiloglanis

The present study recovered a molecular hypothesis of intergeneric mochokid relationships that partially conflicts with those based on morphological data (Vigliotta, 2008). While both datasets support the monophyly of Chiloglanidinae and the chiloglanid tribe Atopochilini Vigliotta, 2008 (including Atopodontus, Atopochilus and Euchilichthys), the morphological analysis of Vigliotta (2008) does not support the monophyly of the subfamily Mochokinae (sensu Seegers, 2008), and the studies differ in other inferred relationships among mochokin genera.

Vigliotta (2008) also recovered the genus Chiloglanis as paraphyletic, with C. macropterus and Chiloglanis sp. ‘Burundi’ (subsequently described as Chiloglanis kazumbei Friel & Vigliotta, 2011) not recovered within his Chiloglanis sensu stricto clade. Rather, he inferred the lineages leading to those species as originating from splits closer to the base of Chiloglanidinae. That finding was perhaps not surprising, given that Balon and Stewart (1983) had noted that C. macropterus in many ways is an unusual Chiloglanis and rather might be a ‘primitive sister species to all Chiloglanis’. A robust test of the monophyly of the genus Chiloglanis would therefore depend on inclusion of C. kazumbei and C. macropterus. We included four specimens of C. macropterus for the first time in a molecular analysis and recovered a deeply nested placement of this taxon within the genus (see Figure 2). Unfortunately, we were unable to obtain samples from C. kazumbei.

The four C. macropterus specimens were collected at two different locations, one of which expands the known distribution of this species. The first location was situated on the Luongo River (a right‐hand tributary of the Luapula River), most likely at the type locality of C. macropterus which Poll and Stewart (1975) indicated to be situated on the Luongo River at the ferry crossing 53 km south of Kawambwa (Poll & Stewart, 1975) and which corresponds to Locality 2 in the study of Balon and Stewart (1983). The two other specimens were surprisingly caught above Kundabikwa Falls on the Kalungwishi River (a tributary of Lake Mweru), which appears to be a natural barrier for upstream migration of fish (Schedel, 2020). Previously, C. macropterus was only recorded from the Luongo River, although Balon and Stewart (1983) speculated that a record by Malaisse (1968) of an undescribed Chiloglanis from the Luanza River (a left‐hand tributary of the Luapula River) might be conspecific with C. macropterus. Our records expand the known distribution area of this species to the Kalungwishi River. This adds further support to the hypothesis of Balon and Stewart (1983) that the Luongo River and Kalungwishi had been once connected. Interestingly, it seems that genetic divergence between the populations from the Luongo and Kalungwishi River is quite low if not absent (see Figure 2). These findings suggest that the breakup of the Luongo–Kalungwishi connection took place rather recently, or alternatively that movement between these systems still occurs. For example, the extensive dambo wetland system situated between the two river systems (~25 km east of Kawamba) might permit occasional exchanges and gene flow.

Our phylogenetic data not only strongly supported the monophyly of Chiloglanis (BS: 100) but also revealed for the first time the presence of at least two major lineages (BS:100) within the genus (see Figure 2). One clade includes taxa from the Zambezi drainage (including Lake Malawi), Nile drainage, Niger drainage, and the Luapula and Chambeshi rivers (both part of the upper Congo drainage). The second clade includes only taxa from the Congo basin and its subdrainage systems (including Lake Tanganyika and the Luapula drainage systems) and it is hence referred to as the ‘Congo‐Chiloglanis’ clade. Strikingly, C. macropterus was nested within the ‘Congo‐Chiloglanis’ clade with Chiloglanis productus Ng & Bailey, 2006 as its sister group. This contrasts with Balon and Stewart (1983) suggestion that C. macropterus could be a primitive member of the genus Chiloglanis. Our findings also differ from those of Vigliotta (2008), who recovered the lineage leading to C. macropterus as originating at the earliest split within Chiloglanidinae. Morphological characters distinguishing C. macropterus from other Chiloglanis include comparatively large fins with elongated pectoral and dorsal fin spines, a roundish caudal fin and elongated mandibular barbels (see Poll & Stewart, 1975). These morphological features could be easily considered as secondarily derived from a more bottom‐dwelling Chiloglanis mode of life because our own underwater observations in the Luongo River (F.D.B. Schedel) revealed that C. macropterus is epibenthic, that is, it tends to hover several centimetres above the substrate. This contrasts with other Chiloglanis, which are typically benthic.

The phylogeographic signal recovered from the major Chiloglanis subclades should be interpreted with great caution for the moment because our all‐catfish dataset included only a small fraction of the impressive Chiloglanis diversity and highly likely missed potential additional lineages from ichthyological provinces not sampled in this study. Indeed, the first phylogenetic analysis based on the mochokid COX1 dataset (Supporting Information Figure S4) suggests the presence of additional sublineages and implies that the ‘Congo‐Chiloglanis’ clade does not exclusively includes taxa from the Congo basin but likely also species from Western Africa and the Nile drainage. Other recent studies have suggested a past linkage and exchange between the Congo and Nile drainages as well, such as the presence of syntopically collected Garra specimens in the Main Nile suggest (Moritz et al., 2019), and the sister relationship recovered between the cichlid species Pseudocrenilabrus nicholsi (Upper Congo drainage) and Pseudocrenilabrus multicolor (Lower Nile drainage), with an estimated divergence time of approximately 2.2 million years (Schedel et al., 2019).

Surprisingly, two specimens identified as Chiloglanis niloticus (GenBank accession number: HF565846; Day et al., 2013) from the Main Nile did not cluster together in the analysis of the COX1 dataset. Instead, the C. niloticus specimen clustered within the ‘Congo‐Chiloglanis’ clade while our newly sequenced specimen of C. cf. niloticus clustered with a Chiloglanis from Niger drainage. The sequence similarity of these two individuals on COX1 therefore was only 86.7%. This result strongly suggests the presence of at least two Chiloglanis species in the Main Nile (see Neumann et al. (2016) for an overview of what is known about Chiloglanis populations in that system). Given that both specimens were collected hundreds of river kilometres upstream from the type locality, it is very possible that neither specimen is not conspecific with C. niloticus. This conjecture would have to be investigated further using both morphological and molecular data. Overall, these results underline our limited knowledge about the alpha‐taxonomy and distribution of the genus Chiloglanis, starting with its alpha‐taxonomy as well as its biogeographical history.

The genus Microsynodontis

Contrasting with the morphological study of Vigliotta (2008) and the molecular study of Day et al. (2013), our analysis did not recover the monophyly of Synodontis, but rather obtained Synodontis punu as sister to Microsynodontis. Vreven and Milondo (2009) originally assigned this species to Synodontis based on a combination of characters diagnostic for the genus, including a forked caudal fin (see Vreven & Milondo, 2009). Microsynodontis is the only mochokid genus that includes species with a rounded caudal fin (Howes, 1980), which Synodontid punu clearly lacks. However, the latest revision of Microsynodontis and the description of several new species by Ng (2004) clarified that caudal fin shape ranges from emarginate to truncate among Microsynodontis species. In addition, Seegers (2008) reported that mochokid catfish imported by the ornamental fish trade to Germany had an overall ‘Microsynodontis‐appearance’, but with deeply forked caudal fins (potentially associated with sexual dimorphism). This apparent variability in caudal fin shape among Microsynodontis species in combination with phylogenetic results from our study suggests that S. punu should be transferred to Microsynodontis. Another small‐sized species, Synodontis acanthoperca Friel & Vigliotta, 2006, was described from the Ogowe drainage in Gabon and found to be morphologically quite similar to S. punu (Vreven & Milondo, 2009). It might likewise represent a member of Microsynodontis, but so far no molecular data are available for that species.

Phylogenetic placement of Austroglanididae

This study provides the first molecular reconstruction of the phylogenetic affinities of the African catfish family Austroglanididae within Siluriformes. While our analysis resolved the intrageneric relationships of the three currently known Austroglanis species quite well, the phylogenetic placement of the family received only weak support. Our data suggest a closer relationship of Austroglanididae to Pangasiidae, Ictaluroidea and/or Ariidae. These four families were consistently recovered in a monophyletic group in all phylogenetic analyses based on mitochondrial data.

Substantial efforts over the past five decades have attempted to infer the relationship of this family using morphological characters (de Pinna, 1993; Diogo, 2005; Diogo & Bills, 2006; Mo, 1991; Skelton et al., 1984). Some of those studies have suggested Austroglanididae to be related closely to one or more of three clades with which it clusters in the molecular analysis, but close ties to Anchariidae, Bagridae, Claroteidae, Schilbeidae and Pimelodidae have also been obtained. Diogo and Bills (2006) provided a detailed discussion on potential autapomorphies of Austroglanididae as well as of morphological similarities between Austroglanididae and Ictaluroidea. However, to our knowledge no study has identified clear synapomorphies supporting a clade containing Austroglanididae and one or more of the families Ictaluroidea, Pangasiidae and Ariidae. Of the various catfish families proposed to be related to Austraoglanididae, our dataset was missing only Anchariidae, a family endemic to Madagascar and which has been generally considered to be closely related to Ariidae, with which it forms the superfamily Arioidea (Sullivan et al., 2006).

Despite all this uncertainty, our findings are congruent with the general morphological consensus that the closest relatives of Austroglanididae are not part of the ‘Big Africa clade’ of Sullivan et al. (2006). That consensus raises intriguing questions about the evolutionary and biogeographic origin of Austroglanididae, which is endemic to Southern Africa. A close relationship with Pangasiidae and Ictaluroidea would suggest an Asian origin for the family and imply that their ancestors colonized Africa without leaving any modern descendants in Asia or in remaining parts of Africa. Alternatively, the family might have derived from marine/brackish ancestors that ascended and colonized exclusively the Southern African river systems. This hypothesis is appealing given the fact that members of Ariidae (one of the clades to which Austroglanidiade appears to be closely related) are primarily marine but known to have colonized freshwater systems (Nelson et al., 2016). Undoubtedly, an in‐depth phylogenomic approach will be needed to test the validity of a clade containing Austroglanididae, Ictaluroidea, Pangasiidae and potentially Arioidea, and to clarify the biogeographical origin of Austroglanididae.

CONCLUSIONS

Our study provides 52 new partial mitochondrial genomes (i.e., covering all protein‐coding genes) of representative taxa of all currently valid mochokid genera, all described Austroglanis species, and selected species of other catfish families belonging to the ‘Big Africa clade’ sensu Sullivan et al. (2006). Hence, this study substantially increases the mitochondrial genomic resources for African catfish available on public databases such as GenBank, which were limited previously to seven species. Our increased taxon sampling allowed reconstruction of the first robust phylogeny that resolves all intergeneric relationships within Mochokidae. It is also the first to infer the relationships of the family Austroglanididae, which is apparently not closely related to the ‘Big Africa clade’ but rather linked to the Asian Pangasiidae, the Asian/North American Ictaluroidea and perhaps even the marine Ariidae. As in previous studies based primarily on mitochondrial data, our analyses only weakly supported a reconstruction of the earliest divergences within Siluriformes. We encourage large‐scale genomic studies and a renewed examination of the phylogenetic signal in their morphological characters in the future. Such studies would increase the robustness of the phylogenetic framework needed to resolve the spatiotemporal diversification of catfishes throughout Africa and beyond, further test the placement of the enigmatic family Austroglanididae, and help reconstruct the history of morphological and ecological evolution in this rich and diverse clade of fishes.

DISCLAIMER

Data on genetic material contained in this paper are exclusively published for noncommercial use only. Access to specimens and/or tissue samples complies with relevant national laws implementing the Convention on Biological Diversity and Nagoya Protocol agreement where relevant, that is, where national access laws were established and applicable at the time of collecting. Nonmonetary benefits have been shared through taxonomic training and capacity building, e.g., during joint field work or joint publication. Use by third parties for purposes other than noncommercial scientific research may infringe the conditions under which the genetic resources were originally accessed and should not be undertaken without obtaining consent from the corresponding author of the paper and/or obtaining permission from the original provider of the genetic material.

AUTHOR CONTRIBUTIONS

F.D.B.S. conceived the study and wrote the first draft of the manuscript. F.D.B.S., A.C., M.O.P., B.L.S., D.N., E.J.W.M.N.V. and U.K.S. were involved with the collection of the specimens and tissue samples studied herein. F.D.B.S., A.C. and N.U.W. performed the molecular work associated with the sequencing of the partial mitochondrial genomes. F.D.B.S. conducted all molecular analyses and prepared all figures. B.L.S, D.N. and U.K.S. contributed to the improvement of all versions of the manuscript. All authors read and approved the final manuscript.

Supporting Information Table S1 Overview of the taxon sampling (all‐catfish dataset) with corresponding GenBank accession numbers and where applicable additional metadata is given (i.e., repository numbers of specimens and origin)

Click here for additional data file.

Supporting Information Table S2 Overview of sequence data obtained from GenBank: COX1 (mochokid COX1 dataset), rag1 and rag2 ‘nuclear dataset’

Click here for additional data file.

Supplementary Material S1 Comprehensive overview of the specimens and taxa sequenced for this study.

Click here for additional data file.

Supplementary Material S2 Overview of sequences obtained from GenBank but excluded from our analysis

Click here for additional data file.

Supporting Information Figure S1 ML phylogeny (RAxML) of the order Siluriformes, including 30 catfish families (focal families are indicated by pictures of representative species), based on all mitochondrial protein coding genes (11,400 bp). Bootstrap proportions based on 1000 BS replicates are indicated at nodes by corresponding BS values. For a collapsed version of the same analysis see Figure 1 in the main manuscript

Click here for additional data file.

Supporting Information Figure S2 ML phylogeny (RAxML) of the order Siluriformes, including 30 catfish families (focal families are indicated by pictures of representative species), based on all mitochondrial protein coding genes but with third codon position stripped from the alignment ‘all‐catfish dataset’ (7600 bp). Bootstrap proportions based on 1000 BS replicates are indicated at nodes by corresponding BS values

Click here for additional data file.

Supporting Information Figure S3 BI phylogeny (MrBayes) of the order Siluriformes, including 30 catfish families (focal families are indicated by pictures of representative species), based on all mitochondrial protein coding genes (11,400 bp). Numbers next to branch nodes represent Bayesian posterior probabilities (BPP)

Click here for additional data file.

Supporting Information Figure S4 ML phylogeny (RAxML) for the family Mochokidae based on all available COX1 sequences available GenBank as well as of the newly sequenced mochokid specimens (688 individuals; 1545 bp). Specimens included in the in the ‘all‐catfish dataset’ (see Figure 2 in the main manuscript) indicated by blue font. Bootstrap proportions based on 1000 BS replicates are indicated at nodes by corresponding BS values

Click here for additional data file.