Phylogenomics of the bumblebee catfishes (Siluriformes: Pseudopimelodidae) using ultraconserved elements

2.1 Taxon sampling

Ingroup sampling included 33 specimens spanning all seven genera and 21 species, (19 valid plus two undescribed), representing 36% of the valid species diversity of Pseudopimelodidae (Table 1). Related taxa included 18 species of the siluriform families Pimelodidae (six species), Heptapteridae (six species), Ariidae (one species), Aspredinidae (one species), Callichthyidae (one species), Doradidae (two species), and Ictaluridae (one species; Table 1). Related taxa were chosen based on a previous phylogenetic hypothesis within Siluriformes (Sullivan et al., 2006). Trees were rooted in the most distant taxon Charax metae (Characiformes: Characidae).

2.2 DNA extraction and sequencing

Tissue samples were taken from fresh voucher specimens and preserved in 95% ethanol. Vouchers were then fixed in 10% formaldehyde and transferred to 70% ethanol for permanent storage (see Table 1 for catalog and locality data). Institutional codes follow Sabaj (2020). Whole genomic DNA was extracted from tissue samples with the DNeasy Tissue Kit (Qiagen) and quantified using the Qubit® dsDNA broad range (BR) Assay Kit (Invitrogen, Life Technologies) following manufacturer's instructions. We used a newly developed probeset for Ostariophysi to capture sequence data for 2708 UCE loci (Faircloth et al., 2020). Library preparation, sequencing, and data pipelining were performed at Arbor Biosciences (AB) using the following protocol: DNA libraries were prepared for the 52 specimens (33 ingroup, 18 related taxa, and one outgroup taxon) by modifying the Nextera (Epicentre Biotechnologies) library preparation protocol for solution-based target enrichment following Faircloth et al. (2012) and increasing the number of PCR cycles following the tagmentation reaction to 20 as recommended by Faircloth et al. (2013). AB staff used the Nextera library preparation protocol of in vitro transposition followed by PCR to prune the DNA and attach sequencing adapters, then used the Epicentre Nextera kit to prepare transposase-mediated libraries with insert sizes averaging 100 bp (95% CI: 45 bp) following Adey et al. (2010).

Whole genomic DNA (40 ng/µl) was first sheared with a QSonica Q800R instrument and selected to modal lengths of approximately 500 nt using a dual-step SPRI bead cleanup to prepare the libraries. AB staff then converted the DNA to Illumina sequencing libraries with a slightly modified version of the NEBNext(R) Ultra(TM) DNA Library Prep Kit for Illumina(R). After ligation of sequencing primers, libraries were amplified using KAPA HiFi HotStart ReadyMix (Kapa Biosystems) for six cycles using the manufacturer's recommended thermal profile and dual P5 and P7 indexed primers (Kircher et al., 2012). After purification with SPRI beads, libraries were quantified with the Quant-iT (TM) Picogreen (R) dsDNA Assay kit (ThermoFisher). AB staff then enriched pools comprising 100 ng each of eight libraries (800 ng total) using the MYbaits(R) Target Enrichment system (MYcroarray) following manual version 3.0. After capture cleanup, the bead-bound library was re-suspended in the recommended solution and amplified for 10 cycles using a universal P5/P7 primer pair and KAPA HiFi reagents. After purification, each captured library pool was quantified with PicoGreen and combined with all other pools in projected equimolar ratios prior to sequencing. Sequencing was performed across two Illumina HiSeq paired-end 100 bp lanes using v4 chemistry.

2.3 Raw data analysis

After sequencing, adapter contamination, low-quality bases, and sequences containing ambiguous base calls were trimmed using the Illumiprocessor software pipeline developed by Faircloth et al. (2013); https://github.com/faircloth-lab/illumiprocessor). After trimming, we assembled Illumina reads into contigs on a species-by-species basis using Abyss pipeline (Simpson et al., 2009; https://github.com/bcgsc/abyss). We then used a custom Python program (match_contigs_to_probes.py) implemented in PHYLUCE (Faircloth, 2016) integrating LASTZ (Harris, 2007) to align species-specific contigs to the probe-UCE set. This last program creates a relational database of matches to UCE loci by taxon. We then used the get_match_counts.py program (also included in PHYLUCE) to query the database and generate fasta files for UCE loci that were identified across all taxa. A custom Python program (seqcap_align_2.py) was then used to align contigs using the MUSCLE algorithm (Edgar, 2004) and to perform edge trimmings (i.e., cutting edges of each alignment, eliminating highly variable and saturated regions). We also performed phylogenetic analyses on three matrices with varying amounts of data (70%, 80%, and 90% of UCEs present in the complete alignment matrices) to explore the potential effects of missing data on tree reconstructions (Hosner et al., 2016; Streicher et al. 2016). All matrices are available in Figshare [https://doi.org/10.6084/m9.figshare.14182865]. Information about data for each matrix is summarized in Table 1; species read information is presented in Table S1. All sequences are available at NCBI Sequence Read Archive (SRA) submissions: SAMN18849693-SAMN18849744. Details on UCE sequence analyses are available online in the PHYLUCE documentation (Faircloth, 2016).

2.4 Phylogenetic analyses

We analyzed the Pseudopimelodidae dataset, with 52 specimens, using maximum likelihood (ML; RAxML v8; Stamatakis, 2014), Bayesian (BI; ExaBayes v1.4; Aberer et al., 2014), and coalescent tree (ASTRAL-II; Mirarab & Warnow, 2015) approaches. For all analyses, we partitioned the UCE data using the Partition-UCE (Tagliacollo & Lanfear, 2018) and performed model selection in PartitionFinder (Lanfear et al., 2012). The RAxML analysis was performed on 70%, 80%, and 90% complete matrices using partitions with edge-trimming alignment (Table 2). Five alternative runs using GTRGAMMA model and distinct parsimony-starting trees were performed to find the best ML tree. Pseudo-replicates applied the autoMRE function for the extended majority-rule consensus tree criterion available in RAxML v8 (Stamatakis, 2014) to assess bootstrap support for branches. This option allows tests for bootstrap convergence, determining if pseudo-replicates are getting stable support values (Pattengale et al., 2010).

BI of the concatenated alignment was performed using ExaBayes (Aberer et al., 2014) on two independent runs, each with two chains, (one cold and one hot), of 1,000,000 generations using the partition schemes for all three complete matrices (Table 2). Tree space was sampled every 100 generations to yield 10,001 trees. Parameter estimates and ESS values were visualized in Tracer v1.6 (Rambaut et al., 2018), and the last 7500 trees were sampled after checking results for convergence (25% burn-in). This procedure allowed us to visualize the posterior probability log within and between independent runs and ensure that the average standard deviation of split frequencies was <1%. The effective sample sizes (ESS) were >200, and the potential scale reduction factor for estimated parameters was approximately 1.0. We generated the 50% most credible set of trees from the posterior distribution of possible topologies using the consensus algorithm of ExaBayes (burn-in 25%; thinning 500).

We inferred a coalescent tree analysis from individual gene trees using a two-step process accounting for coalescent stochasticity among individual UCE loci and addressing the related problem where concatenated analyses can return highly supported but incorrect trees process. First, we used PHYLUCE to resample the 70%, 80%, and 90% complete matrices by loci and generated a best tree using RAxML bootstraped for each locus in each of those matrices. Then, we used ASTRAL-II (Mirarab & Warnow, 2015) to infer species trees from each of the best tree subsets of loci and generated a majority-rule consensus tree of the results (minimum clade frequency 0.7). Although ASTRAL is not strictly a coalescent method, it is statistically consistent with the multispecies coalescent model (Nute et al., 2018) and scales well for a high number of loci.

3.1 Relationships within Pseudopimelodinae Lundberg et al. 1991

Included genera: Cruciglanis Ortega-Lara, 2006; Pseudopimelodus Bleeker, 1858 (type genus); Rhyacoglanis Shibatta & Vari, 2017.

Diagnosis. Pseudopimelodinae is distinguished within Pseudopimelodidae by five morphological synapomorphies: (a) pectoral-fin spine covered with thick (vs. thin) skin (Shibatta & Vari, 2017: character 3, state 0); (b) tip of pectoral-fin spine bifurcate (vs. pointed) (Shibatta & Vari, 2017: character 29, state 1); (c) gas bladder small, dumbbell-shaped (vs. large, heart-shaped) (Birindelli & Shibatta, 2011; Shibatta & Vari, 2017, character 35, state 1); (d) pseudotympanum opening small (vs. large) (Birindelli & Shibatta, 2011; Shibatta & Vari, 2017, character 38, state 1); (e) anterior portion of the cerebellum extremely prolonged, extending beyond mesencephalon, by less than half the length of telencephalon (vs. moderately prolonged, extending only as far as boundary between the mesencephalon and telencephalon and not contacting the latter) (Abrahão et al., 2018).

Pseudopimelodinae is herein restricted to the genera Cruciglanis, Pseudopimelodus, and Rhyacoglanis (Figures 2 and 3). In our results, the monotypic Cruciglanis is sister to Pseudopimelodus + Rhyacoglanis. Morphological (Shibatta & Vari, 2017) and genetic (Rangel-Medrano et al., 2020) analyses support a monophyletic group composed of the same three genera, but with Rhyacoglanis sister to Cruciglanis + Pseudopimelodus. According to Shibatta and Vari (2017), the sister group relationship between Cruciglanis and Pseudopimelodus is supported by the shared presence of a dark band on the predorsal region (vs. predorsal region light in Rhyacoglanis) and pectoral fin with seven branched rays (vs. usually six in Rhyacoglanis). However, Shibatta and Vari (2017) also noted some variation in the number of branched pectoral-fin rays in Rhyacoglanis with some specimens from the Rio Madeira having seven branched pectoral-fin rays. Give the UCEs phylogeny, the absence of a dark band on the predorsal region in Rhyacoglanis might be interpreted as a secondary loss within Pseudopimelodidae.

In the phylogeny of Shibatta and Vari (2017), Rhyacoglanis paranensis (upper Paraná basin) was sister to a clade of species from the Amazon and Orinoco basins, with Rhyacoglanis epiblepsis sister to Rhyacoglanis annulatus, and Rhyacoglanis seminiger sister to Rhyacoglanis pulcher. In our molecular genetic analysis, R. pulcher was the first species to diverge in the genus, and R. paranensis was sister to R. seminiger plus an undescribed species from Xingu basin (R. epiblepsis and R. annulatus not analyzed).

3.2 Relationships within Batrochoglaninae Shibatta & Silva, new subfamily

LSID: urn:lsid:zoobank.org:pub:EF11973E-91D7-4267-9D29-86457B34706A

Included genera: Lophiosilurus Steindachner, 1877; Cephalosilurus Haseman, 1911; Batrochoglanis Gill, 1858 (type genus); Microglanis Eigenmann, 1912.

Diagnosis. Batrochoglaninae is distinguished within Pseudopimelodidae by three morphological synapomorphies: (a) anterior nostril near (vs. distant from) margin of mouth (Shibatta & Vari, 2017: character 15, state 1); (b) vomer absent (vs. present) (Shibatta & Vari, 2017, character 16, state 1); and (c) anterior and posterior margins of pectoral-fin spine with serrae of approximately the same length (vs. anterior serrae shorter than posterior ones) (Shibatta & Vari, 2017: character 27, state 1).

One highly supported clade within Batrochoglaninae includes the monotypic Lophiosilurus (endemic to the São Francisco basin) and Cephalosilurus with four species distributed in the São Francisco and Orinoco basins and coastal drainages of the Guianas. Based on our results, Cephalosilurus is paraphyletic with Cephalosilurus fowleri (type species) more closely related to Lophiosilurus alexandri than to Cephalosilurus apurensis (Figure 2). Lophiosilurus alexandri is quite distinct morphologically, and all species in Cephalosilurus are more similar to each other than any are to Lophiosilurus. The L. alexandri phenotype emerges from an ancestral Cephalosilurus phenotype, and the close resemblance of C. fowleri to other Cephalosilurus is not convergent. According to Shibatta and Vari (2017), Lophiosilurus and Cephalosilurus share four synapomorphies: (a) well-developed unculiferous epidermal structures (vs. little developed; character 2, state 1); (b) seven branched pectoral-fin rays (vs. five, six, eight or more; character 26, state 1); (c) presence of a constrictor muscle of gas bladder (vs. absent; character 36, state 1); and (d) lateral trabeculae on internal T-shaped gas bladder septum (vs. absence of lateral trabeculae; character 37, state 1). Abrahão et al. (2018) also pointed out the reduced lobus facialis, comprising less than half length of the lateral line lobe (vs. elongated; state 1). Our results strongly support and suggest Cephalosilurus Haseman, 1911 as a junior synonym of Lophiosilurus Steindachner, 1877. A study assessing the synonymy of these genera has been conducted by Shibatta et al. (in review), including all species of Cephalosilurus.

Our analyses included three out of the six valid species of Batrochoglanis: Batrochoglanis raninus, Batrochoglanis melanurus and specimens of Batrochoglanis villosus (from Guamá, Madeira, Xingu, and Negro river basins). Batrochoglanis raninus is the sister of a major clade with B. melanurus nesting deeply within a paraphyletic B. villosus (Figure 3). Batrochoglanis villosus is a widely distributed species known from the Orinoco and Amazonas basins as well as coastal drainages of the Guianas, whereas B. melanurus is restricted to the upper Paraguay in the La Plata basin (Shibatta & Pavanelli, 2005). In their description of B. melanurus, Shibatta and Pavanelli (2005) noted its morphological similarity to B. villosus. They also identified B. villosus from the upper Tapajós (Amazon basin) which shares a low divide with the upper Paraguay basin. Based on our phylogeny, B. melanurus might represents a junior synonym of B. villosus; however, the lack of DNA sequence data of B. villosus from the type-locality (Potaro river, Guyana) prevents a formal proposal of synonymy.

Microglanis is the most diverse genus of Pseudopimelodidae, with 29 valid species (Tobes et al., 2020) broadly distributed in the Orinoco, Amazon, and La Plata basins as well as Atlantic coastal drainages from Guyana to Uruguay. A couple species occur west of the Andes in Pacific coastal drainages of Ecuador (Tobes et al., 2020). Despite its substantial diversity, no phylogenetic hypothesis has been proposed for the genus. In our topology, Microglanis can be divided into four large subclades. The first is represented by an undescribed species from the Orinoco basin. The second group consists of two subclades: species from the Atlantic coastal drainages (Microglanis cottoides, Microglanis cf. parahybae) and the other with two species from La Plata basin Microglanis garavelloi (upper Paraná River) and Microglanis malabarbai (Uruguay River). Souza-Shibatta et al. (2018) using genetic characters supported a clade composed by the same species, with M. cottoides + M. parahybae sister to M. garavelloi + M. malabarbai. Alternatively, our results nest M. parahybae within a paraphyletic M. cottoides (Figure 3).

Microglanis poecilus is the most widely distributed species in the genus (Ruiz & Shibatta, 2011) confirmed to be monophyletic based on our sampling from four distinct Amazonian sub-basins (Arinos, middle Tapajós, Juruá, and Branco). Ruiz and Shibatta (2011) noted that M. poecilus shares 11 features with Microglanis oliveirai, a species they described from the upper-middle Araguaia basin with drains into the Tocantins River. Our results, however, did not support a close relationship between M. poecilus and M. oliveirai and placed the latter species as sister Microglanis sparsus, a species described from the upper Xingu basin (Ruiz, 2016). Furthermore, our analyses revealed deep splits among lineages of M. poecilus suggestive of undescribed taxa.

Finally, our analyses supported a close relationship between Microglanis leptostriatus, M. oliveirai, and M. sparsus in a clade sister to M. poecilus. Mori and Shibatta (2006) considered M. leptostriatus to be a member of the M. parahybae species complex along with M. garavelloi and M. parahybae. Alcaraz et al. (2008) expanded this group to include Microglanis pataxo and Microglanis carlae. Our results placed M. leptostriatus closer to the Amazonian species M. sparsus and M. oliveirai than M. garavelloi and M. parahybae calling into question the cohesiveness of the M. parahybae species complex.