Investigations into the ancestry of the Grape-eye Seabass (Hemilutjanus macrophthalmos) reveal novel limits and relationships for the Acropomatiformes (Teleostei: Percomorpha)

For 175 years, an unremarkable bass, the Grape-eye Seabass ( Hemilutjanus macrophthalmos ), has been known from coastal waters in the Eastern Pacifi c. To date, its phylogenetic placement and classifi cation have been ignored. A preliminary osteological examination of Hemilutjanus hinted that it may have affi nities with the Acropomatiformes. To test this hypothesis, we conducted a phylogenetic analysis using UCE and Sanger sequence data to study the placement of Hemilutjanus and the limits and relationships of the Acropomatiformes. We show that Hemilutjanus is a malakichthyid, and our results corroborate earlier studies that have resolved a polyphyletic Polyprionidae; accordingly, we describe Stereolepididae, new family, for Stereolepis . With these revisions, the Acropomatiformes is now composed of the: Acropomatidae; Banjosidae; Champsodontidae; Glaucosomatidae; Malakichthyidae; Ostracoberycidae; Pempheridae; Pentacerotidae; Polyprionidae; Scombropidae; Stereolepididae, new family; Symphysanodontidae; Synagropidae; and Schuettea . Finally, using our new hypothesis, we demonstrate that acropomatiforms repeatedly evolved bioluminescence and transitioned between shallow waters and the deep bootstrap


INTRODUCTION
In 1846, Johann Jakob von Tschudi described a grouper-like fi sh, Plectropoma macrophthalmos (von Tschudi, 1846), from multiple coastal locations in the tropical Eastern Pacifi c near Lima, Peru (Fig. 1). Since its original description, this species has been collected from Anto fasta, Chile, in the south, to the Galápagos Islands in the north, typically among rock outcroppings at depths ranging from 10 to 55 m (Grove, Lavenberg, 1997;Froese, Pauly, 2021). Following von Tschudi's work, Bleeker (1876) reclassifi ed this species in his Lutjanini, an assemblage that included species currently classifi ed in groups as varied as the Arripidae, Banjosidae, Haemulidae, and Lutjanidae. Because of similarities between members of Bleeker's Lutjanini and Plectropoma macrophthalmos and dissimilarities between this species and sea basses and groupers, Bleeker described a new genus, Hemilutjanus, for P. macrophthalmos. Most subsequent authors in the late 19th and early 20th century followed Bleeker's generic assignment but continued to follow von Tschudi's (1846) placement of Hemilutjanus macrophthalmos by aligning this species with the Epinephelidae or Serranidae (Jordan, Eigenmann, 1890;Boulenger, 1895;Jordan, 1923;Hildebrand, 1946). Jordan, Eigenmann (1890:344) went so far as to state, "the name selected by Dr. Bleeker for this genus is peculiarly unfortunate, for besides the lack of euphony in the name, the genus has neither resemblance to nor affi nity with the genus Lutjanus". Most systematic ichthyologists have not discussed 4/41 ni.bio.br | scielo.br/ni Phylogeny of Hemilutjanus and the Acropomatiformes 1997; McCosker, Rosenblatt, 2010). In 2016, Nelson et al. returned Hemilutjanus to the Epinephelidae (their Epinephelinae) without discussion. Most recently, Parenti, Randall (2020) separated Hemilutjanus from their "Serranidae" and described a new "closely related" family, Hemilutjanidae, for this species in their annotated checklist of fishes of the family "Serranidae". This monotypic Hemilutjanidae was classified by van der Laan et al. (2021) as a member of their "Perciformes *sedis mutabilis*", a group that includes the traditional "Serranidae" as well as families as phylogenetically divergent as the Apogonidae, Centrogenyidae, Lutjanidae, and Moronidae (see Smith, Craig, 2007;and Betancur-R et al., 2017 for family-level placement in molecular phylogenies). Despite Hemilutjanus macrophthalmos being known to science for 175 years, the closest relative of the Grape-eye Seabass remains obscure, with the publication by Parenti, Randall (2020) recognizing the species as a monotypic family and effectively declaring that it is not a "serranid" but that its placement among percomorphs remains unknown.
Given that recent results have placed many former "serranids" either among the mail-cheeked fishes or the acropomatiforms, this study was designed to look at the placement of the enigmatic Hemilutjanus with a particular focus on acropomatiforms. This placement seemed most likely given that Hemilutjanus lacks the characteristic third opercular spine, suborbital stay, extensive head spination, and the expected distal insertion condition of the epaxial musculature on the dorsal-fin pterygiophores that are common to the "serranids" allied with the mail-cheeked fishes (Johnson, 1983;Mooi, Gill, 1995;Smith, 2005; current study). Therefore, we conducted a genome-scale molecular analysis with several goals associated with the phylogenetic placement of Hemilutjanus macrophthalmos. First, we tested whether Hemilutjanus was most closely related to the traditional "Serranidae" as suggested by von Tschudi (1846) and Parenti, Randall (2020), the members of Bleeker's (1876) Lutjanini (which includes several acropomatiforms), the modern Acropomatiformes, or a separate percomorph group altogether. Secondarily, we assessed the limits and relationships of the Acropomatiformes (including testing the monophyly of the "Acropomatidae" [viz. Acropomatidae, Malakichthyidae, and Synagropidae]) using genome-scale DNA-sequence data with the goal of resolving the limits of the order and clarifying the conflicting familial interrelationships by including dramatically more sequence data and representatives of all putative families. Finally, we will use our resulting hypothesis to trace the evolution of bioluminescence and the invasions of the deep sea among the acropomatiforms.  For Sanger sequence data, polymerase chain reaction (PCR) was used to amplify seven gene fragments (16S, COI, ENC1, GlyT, HH3, PLAGL2, and RAG1). Sanger molecular protocols for amplifying and cleaning these markers can be found in Grande et al. (2013) and Smith, Busby (2014). Both strands of the purified PCR fragments were used as templates and amplified for sequencing using the amplification primers and a Prism Dye Terminator Reaction Kit v1.1 (Applied Biosystems). The sequencing reactions were cleaned and desalted using cleanSEQ (Beckman Coulter). The nucleotides were sequenced and called on a 3730 or 3730xl automated DNA sequencer (Applied Biosystems) or by Beckman Coulter Genomics (Danvers, MA).
For high-throughput sequencing, Promega extractions were eluted into a 102 µL volume or the first and second Qiagen elutions were combined and dried down with a DNA SpeedVac Concentrator (Thermo Fisher) to a 102 µL volume. Two microliters of the raw or concentrated extracts were quantified using a Qubit fluorometer (Life Technologies) using the dsDNA BR Assay Kit. Quantified samples (100 µL volume) were sent to Arbor Biosciences (Ann Arbor, MI) for library preparation (e.g., DNA shearing, size selection, cleanup), target capture (using the 500 UCE actinopterygian loci probe set; Faircloth et al. (2013)), enrichment, sequencing using an Illumina HiSeq 2500 or NovaSeq 6000, and demultiplexing of samples.
Arbor Biosciences generated DNA-sequence data using genomic extractions and the 500 UCE actinopterygian loci probe set (Faircloth et al., 2013). We processed the raw FASTQ files from Arbor Biosciences using the PHYLUCE 1.71 (Faircloth, 2016) workflow to retrieve UCE and flanking regions from newly sequenced specimens. Using a parallel wrapper (https://github.com/faircloth-lab/illumiprocessor), we trimmed reads to remove adapter contamination and low-quality bases using Trimmomatic (Bolger et al., 2014). The cleaned sequencing reads were submitted to GenBank and have been assigned BioProject PRJNA831283. We assembled cleaned reads from new and previously published samples (data from Alfaro et al., 2018;Friedman et al., 2019;Girard et al., 2020; Tab. 1; Tab. S1) using a python script (assemblo_abyss.py) with PHYLUCE and SPAdes v3.14.1 (Prjibelski et al., 2020) under the default settings. To identify assembled, orthologous contigs for the UCE loci, we aligned species-specific contig assemblies to a FASTA file of all enrichment baits using match_contigs_to_ probes. This PHYLUCE program implements a matching process using LASTZ (Harris, 2007) and ensures that UCE matches are at least 80% identical over 80% of their length to avoid contamination and paralogy. Further, this program assesses and removes apparent duplicate contigs and contigs hit by baits targeting more than one locus. As noted by Faircloth (2016), the program then creates a relational database containing several tables that map the contig names generated by the assembler to the names of each corresponding locus across all selected taxa. Next, we extracted the contigs corresponding to non-duplicate conserved loci into a monolithic FASTA-formatted file (all UCEs for all species) using get_fastas_from_match_counts. We then aligned the sequence data for UCEs containing more than four taxa using seqcap_align that parallelizes MAFFT 7.130b (Katoh, Standley, 2013). The alignment was refined using GBlocks (Talavera, Castresana, 2007) using the default PHYLUCE settings. For a final PHYLIP-formatted data matrix, we concatenated the resulting alignments for all UCEs present for ≥ 75% of UCE taxa (i.e., loci with data for 40 or more of the 54 species with UCEs) using align_get_only_loci_with_min_taxa followed by align_concatenate_ alignments. The resulting 75% complete "UCE matrix" was based on 457 UCEs or 273.579 bps that were present for the 54 species that had UCE data; this UCE matrix was 95% complete at the locus level (Tab. S4). Across all UCE loci, median sequence fragment length was 599 bps, with a range of 163-1.055 bps (Tab. 2; Tab. S1). The UCE and flanking region sequences were partitioned using the sliding-window site characteristics-entropy method (hereafter, SWSC-EN; Tagliacollo, Lanfear, 2018) to split each UCE locus into left and right flanking regions and the ultraconserved core (i.e., center segment) by rate of evolution.
The final concatenated molecular matrix or "expanded matrix" included 457 UCE loci and 9 Sanger-based loci that encompassed 279.979 aligned base pairs and 70.230 parsimony-informative characters. The resulting left, central, and right UCE segments from SWSC-EN were then used as input along with the independent 16S locus and the three independent codon positions for each of the eight protein-coding Sanger genes to PartitionFinder v2.1.1 (Lanfear et al., 2014(Lanfear et al., , 2017Stamatakis, 2014) for this software to find the best-fitting nucleotide substitution model for each data partition. PartitionFinder selected among models using AIC c and the rclusterf search method with the setting -raxml (Lanfear et al., 2014). PartitionFinder designated 1.310 subsets with associated models for these regions. A list of the subsets of UCEs, partitions, and 12/41 ni.bio.br | scielo.br/ni Phylogeny of Hemilutjanus and the Acropomatiformes associated models can be found in Tabs. S1, S2 and S3. The Sanger alignment can be found in Tabs. S1 and S4 and the expanded alignment can be found in Tabs. S1 and S5.

Phylogenetic analyses.
Concatenated maximum-likelihood and species-tree analyses were conducted on our "core" 54-taxon dataset that includes 54 percomorph terminals and both UCE and Sanger data. Additionally, we analyzed an "expanded" 57-taxon concatenated dataset that included three species that were only represented by 2-9 Sanger fragments. Both concatenated analyses used IQ-Tree v2.1.1 (Nguyen et al., 2015;Chernomor et al., 2016). These analyses began with ten independent runs of the software with the perturbation strength (-pers) set to 0.2 and using the 1.310-partition scheme from PartitionFinder. For each dataset, the optimal trees from each of the ten independent runs were submitted as starting trees with a final round of refinement with -pers set to 0.2 and a larger number of stop iterations (-nstop) set to 2.000, with more thorough nearest neighbor interchange branch swapping (-allnni), and the number of candidate trees (-nbest) to be maintained during the maximum-likelihood search set to 25. Support for the concatenated phylogenies for each analysis was assessed using IQ-Tree (-bo), and the results from 200 bootstrap replicates are summarized using majorityrule consensus trees. We recognize three levels of nodal support: ≥50% bootstrap support represents a supported node or clade, ≥70% bootstrap support represents a moderately well-supported node or clade, and ≥95% bootstrap support represents a well-supported or strongly supported node or clade.
In addition to the concatenated maximum-likelihood analyses above, each of the 457 UCE, 16S, and protein-coding "Sanger" loci described above were analyzed individually using RAxML v 8.2 (Stamatakis, 2014) using a GTRGAMMA substitution model for the core taxa (54 taxa that have UCE and Sanger data). The best likelihood result from five independent analyses for each UCE, 16S, and protein-coding locus were retained and combined with the results from each locus for subsequent analysis in ASTRAL-III (Zhang et al., 2018) to infer a species tree from the individual gene trees. Branch supports for this species-tree analysis represent support for the quadripartition across all loci or local posterior probability (LPP) for the nodes in the "species-tree analysis". As above, we recognize three levels of nodal support: ≥50% LPP support represents a supported clade, ≥70% LPP support represents a moderately well-supported clade, and ≥95% LPP support represents a well-supported or strongly supported clade. Depth and bioluminescence data. We collected depth data for each clade from the institutions sharing their data on FishNet2 (2021). Following clade-level data downloading, we used the following procedure for data aggregation: 1) we removed all samples without depth data; 2) we removed all samples with depth listed as meterwire-out; 3) we removed all unclear depth data; 4) we converted the data for all samples with a depth range of 0 to a number to only the non-zero number (e.g., if the depth was listed as 0-2.500 m we converted the number to 2.500); 5) we converted all depth data to meters; 6) we calculated the mean depth for samples with a collection depth range and used this mean depth for the depth of these samples; 7) we removed samples that listed the depth of collection as 0; 8) finally, we removed outlier samples that were notably more extreme than other lots (typically 2-6 lots per clade). Once these cleaning procedures were completed, we calculated a mean depth value and standard deviation of depth for each family-level clade (along with median, minimum, and maximum values); we considered a fish clade as part of the deep sea if the mean + standard deviation extended beyond a depth of 200 m. We took the presence or absence of bioluminescence data from Ghedotti et al. (2018). We examined and analyzed the dataset (ancestralstate reconstructions) in Mesquite v3.5 (Maddison, Maddison, 2018). We examined ancestral-state reconstructions using parsimony and maximum likelihood.

Morphological investigation.
We examined formalin-fixed and ethanol-preserved fishes, cleared-and-stained specimens, and dried skeletal material using multiple stereomicroscopes with varying magnification and lighting regimes. We explored new morphological data along with previously described variation as an initial investigation into acropomatiform relationships with the goal of finding characteristic features for new clades (Tab. 2). The specimens examined for this study are listed in the material examined, and a more complete morphological examination is underway.

RESULTS
The core 54-taxon concatenated maximum-likelihood analysis (ML) resulted in a single optimal tree (Fig. 3) with a likelihood score of -2087266.778. Most of the 51 nodes (41 or 80%) were well supported by a bootstrap value ≥95, 45 nodes (88%) were moderately well supported by a bootstrap value ≥70, and 49 nodes (98%) were supported in the 50% majority-rule tree (Fig. 3). The core 54-taxon species-tree analysis (ST) resulted in a single tree with most of the 51 nodes (34 nodes or 67%) well supported by an LPP value ≥95, 42 nodes (82%) moderately well supported by an LPP value ≥70, and 49 nodes (96%) supported in the 50% majority-rule tree (Fig. 4). The expanded 57-taxon concatenated analysis resulted in a single optimal tree (Fig. 5) with a likelihood score of -2089599.873. Half of the 54 nodes (27 or 50%) were well supported by a bootstrap value ≥95, 44 nodes (81%) were moderately well supported by a bootstrap value ≥70, and 48 nodes (89%) were supported in the 50% majority-rule tree (Fig. 5). Clearly, the addition of three Sanger-only taxa reduced bootstrap support, but it did not alter the recovered relationships among the taxa with both UCE and Sanger data. However, the placement of Sanger-only species (Acanthaphritis, Crystallodytes, and Schuettea) was not generally well supported with <50 to 88% bootstrap support for these Sanger-only taxa and their sister groups (Fig. 5). The results of all three analyses (54-taxon ML, 54-taxon ST, and 57-taxon ML) agreed on the most important findings of this study. First, all analyses separated Hemilutjanus from its frequent historic taxonomic allies in the Acanthistiidae, Anthiadidae, Epinephelidae, and Serranidae. As has been shown in most recent analyses, the traditional "Serranidae" did not form a monophyletic group (e.g., Chen et al., 2003;Smith, Wheeler, 2004;Smith, Craig, 2007;Lautredou et al., 2013;Figs. 3-5), and we show that these taxa are not particularly closely related to Hemilutjanus. Instead, all analyses consistently place Hemilutjanus sister to Malakichthys with moderate to strong support, deeply nested within the Acropomatiformes. Given the moderate to strong support for this close relationship and overall similarity of the included groups, we formally classify Hemilutjanus macrophthalmos as a member of the Malakichthyidae rather than a separate monotypic family. As has been shown in previous analyses (Smith, Craig, 2007;Betancur-R et al., 2013a;Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;Mirande, 2016;Sanciangco et al., 2016;Ghedotti et al., 2018;Rabosky et al., 2018;Fig. 2), the traditional "Acropomatidae" (e.g., Nelson, 2006) is polyphyletic. Instead, we recover three independent "acropomatid" families that are consistent with the classification of Ghedotti et al. (2018)  In the core 54-taxon ML analysis, our family-level phylogeny for the Acropomatiformes resulted in several relationships that had not been previously proposed ( Fig. 2; Tab. 1). Malakichthyidae was recovered sister to a clade composed of Stereolepis, Bathyclupeidae, Champsodontidae, Creediidae, and Synagropidae. Together, these taxa were resolved sister to a clade that included the Banjosidae and Pentacerotidae. All these taxa were recovered sister to a clade composed of the Dinolestidae. The sister group to the clade composed of Stereolepis, Banjosidae, Bathyclupeidae, Champsodontidae, Creediidae, Dinolestidae, Malakichthyidae, Pentacerotidae, and Synagropidae is a clade composed of Polyprion, Glaucosomatidae, Lateolabracidae, and Pempheridae. Finally, the acropomatiform clade sister to all other previously discussed acropomatiform groups is composed of the Acropomatidae, Epigonidae, Howellidae, Ostracoberycidae, and Scombropidae.
As with the core concatenated 54-taxon ML analysis, our ST analysis separated Hemilutjanus from the included "serranid" taxa, and the "serranids" that were included were recovered as paraphyletic. As with the concatenated 54-taxon ML analysis, the ST analysis recovered Hemilutjanus sister to Malakichthys (i.e., we recovered a modified Malakichthyidae) inside a monophyletic Acropomatiformes with identical composition (for the included taxa). The ST analysis also separated the Acropomatidae, Malakichthyidae, and Synagropidae into three distinct and independent clades. Therefore, the core findings of our study were consistent across methods and datasets analyzed. Despite identical findings for the core goals of this study, there were differences between the 54-taxon ML and ST results. Most of the disagreements between these trees were due to the differential placement of Champsodon+Crystallodytes as the sister group to a clade composed of Stereolepis, Bathyclupeidae, and Synagropidae (in ML) versus a sister group to all other acropomatiforms (in ST). The ML and ST results also differed in the placement of Stereolepis sister to Bathyclupeidae+Synagropidae in ML  and sister to Banjosidae+Pentacerotidae in ST, the placement of Dinolestidae sister to Banjosidae+Pentacerotidae in ML and sister to Stereolepis+Banjosidae+Pentacerotidae in the ST analysis, the monophyletic (in ML) versus paraphyletic (in ST) resolution of Parascombrops relative to Synagrops, and the placement of gobiiforms at the base of our resulting phylogeny in ML versus the sister group to all analyzed taxa except the included Ophidiiformes, Scombriformes, and Syngnathiformes in the ST analysis. In total, 42 of 51 nodes were identical between the core ML and ST analyses with six of the nine (67%) conflicting nodes resulting from the movement of Champsodon and Crystallodytes (Figs. 3-4).
In all three analyses, the Acropomatiformes was recovered sister to the Uranoscopiformes, and the overall phylogeny was similar to other large-scale studies (Figs. 3-5;Davis et al., 2016;Sanciangco et al., 2016). Together, Acropomatiformes and Uranoscopiformes were recovered sister to the Acanthuriformes. Among the taxa sampled in this study, this clade of three orders had a series of progressively less closely related sister groups: Centrarchiformes, Scorpaeniformes, and Carangiformes+Ovalentaria. The subsequent sister group was Scombriformes+Syngnathiformes in both ML analyses, followed by the Gobiiformes and Ophidiiformes. In contrast, the earliest splits in the ST analysis were composed of Ophidiiformes and Gobiiformes+Scombriformes+Syngnathiformes .
In our expanded 57-taxon concatenated maximum-likelihood analysis that included Acanthaphritis, Schuettea, and Symphysanodon and looked more broadly at the interfamilial relationships among the Acropomatiformes, we recovered identical relationships with the core 54-taxon ML analysis except for the inclusion of the three Sanger-only taxa (Fig. 5). The inclusion of these three species with only 2-9 Sanger sequences resulted in a notable decrease in support for relationships among the Acropomatiformes but the same general relationships (Figs. 3 and 5). The placement of Acanthaphritis as the sister group to Creediidae had moderate support, and the placement of Symphysanodon as the sister group of Epigonidae+Howellidae was supported with a bootstrap value of 52%. In contrast, the sister-group relationship between Schuettea and Champsodon was not supported. During the individual IQ-Tree replicates (results not shown), Schuettea was always recovered in the clade composed of Stereolepis, Banjosidae, Bathyclupeidae, Champsodontidae, Creediidae, Dinolestidae, Hemerocoetidae, Malakichthyidae, Pentacerotidae, and Synagropidae, and it was never allied with its traditional ally, Monodactylus. This surprising and well-supported acropomatiform placement for Schuettea rather than sister to Monodactylus lends support to Tominaga's (1968) suggestion that Schuettea may be more closely related to pempherids than monodactylids. While a somewhat close relationship between Schuettea and Pempheridae (as members of the Acropomatiformes) was recovered, the analyzed DNA-sequence data do not support the proposed sister-group relationship between Schuettea and Pempheridae noted by Tominaga (1968), but clearly much additional work is needed on the placement of Schuettea among acropomatiforms given the paucity of data for Schuettea in this study.
Using the results of the 57-taxon ML analysis, we examined habitat transitions between shallow and deep water at the family level. Using both parsimony and likelihood, we found remarkably large numbers of transitions between shallow and deep-water habitats for a clade of approximately 300 species. Using maximum likelihood, our optimization suggests two independent invasions into the deep sea: the ancestor of the Acropomatiformes and the Champsodontidae. Further, this optimization estimated seven independent returns to shallow water in the Glaucosomatidae+Lateolabraci-dae+Pempheridae, Champsodontidae+Schuettea, Creediidae, Dinolestidae, Hemilutjanus, Scombropidae, and Stereolepis (Fig. 6). Using parsimony, the Acropomatidae+Epigoni-dae+Howellidae+Ostracoberycidae+Symphysanodontidae, Polyprion, and one or more clades in the Banjosidae+Bathyclupeidae+Champsodontidae+Creediidae+Dinolesti-dae+Hemerocoetidae+Malakichthyidae+Stereolepis+Synagropidae+Pentacerotidae invaded the deep sea. Using maximum likelihood and parsimony, we recover three independent evolutions of bioluminescence at the family level (treating Schuettea as a family): Epigonidae+Howellidae, Acropomatidae, and Pempheridae (Fig. 6).

DISCUSSION
This study was first and foremost designed to resolve the placement of Hemilutjanus. Given that all our analyses recovered Hemilutjanus sister to Malakichthys within a monophyletic Acropomatiformes, this discussion begins with this sister-group relationship and its taxonomic implications. Next, this study describes a new family of fishes for Stereolepis, discusses the somewhat surprising placement of Schuettea in the acropomatiforms, explores the limits and relationships of the Acropomatiformes (including a discussion on the polyphyly of the traditional "Acropomatidae"), and ends with a look at the implications of this revised phylogeny of the Acropomatiformes on the invasion of the deep sea and the evolution of bioluminescence.   The phylogenetic and taxonomic placement of Hemilutjanus. Hemilutjanus macrophthalmos has not been included in any previous explicit phylogenetic analyses, but previous studies based on external anatomy or radiographs have suggested possible placements for this species as a "serranid", a close relative of "serranids", among the Arripidae, Banjosidae, Haemulidae, and Lutjanidae, or as incertae sedis in the traditional Percoidei (von Tschudi, 1846;Bleeker, 1876;Jordan, Eigenmann, 1890;Boulenger, 1895;Johnson, 1984;Mooi, Gill, 1995;Nelson et al., 2016;Parenti, Randall, 2020). No molecular phylogenetic studies have included Hemilutjanus macrophthalmos, and the detailed anatomy of this species has not been documented using either cleared-andstained specimens, dried skeletal material, or micro-computed-tomography scans, so it is not surprising that its placement has remained obscure.
The results of this molecular study clearly place Hemilutjanus in the Acropomatiformes and sister to Malakichthys. There are no published molecular phylogenetic studies looking broadly at relationships among malakichthyids. There have been specieslevel external morphological studies across the malakichthyids (Yamanoue, Matsuura, 2004; Yamanoue, 2016) that have documented substantive variation among the species in the family, but there have been limited comparative osteological studies across the "Acropomatidae" or Malakichthyidae (best detailed in Katayama, 1959, andSchwarzhans, Prokofiev, 2017). Katayama (1959) described variation across a diversity of systems for all the known "acropomatids" and allies in Japan, and his primary characters used to separate Malakichthys (his Malakichthyinae) from other examined fishes were the absence of a metapterygoid lamina and the lack of canine teeth. Schwarzhans, Prokofiev (2017) provided some osteological differences between malakichthyids and other "acropomatids", but their focus was on the Synagropidae, and they did not discuss the limits and relationships of malakichthyids. Subsequent work by Yamanoue, Matsuura (2004) and Yamanoue (2016) showed that canine teeth were found in some malakichthyids, primarily in Verilus, but neither they nor other authors have noted whether species of Verilus or species of Malakichthys outside of Japan have a metapterygoid lamina. Based on our examination, Hemilutjanus has a well-developed metapterygoid lamina and canine teeth, which were absent in our examined material of Malakichthys. Thus, previously identified morphological characters are of little use for placing Hemilutjanus with Malakichthys.
The conclusive placement of Hemilutjanus sister to Malakichthys with molecular data, however, raises the question whether Hemilutjanidae should be retained as a monotypic family or whether Hemilutjanus should be classified as a member of the Malakichthyidae as either choice would be a classification that recognizes only monophyletic groups and both family-level names have already been described and are valid and available. We have chosen to place Hemilutjanidae in the synonymy of Malakichthyidae for several reasons. First, there is virtually no tradition of recognizing Hemilutjanidae. The family was only recently described in 2020 by Parenti and Randall, and that description was in the context of a checklist of the "Serranidae". The authors emphasized the "serranids", and they appear to have described Caesioscorpididae and Hemilutjanidae, so that they could properly exclude them from their "Serranidae". There was no material of Hemilutjanus listed as examined in that study, and the family-level diagnosis obviously drew heavily from Hildebrand (1946). The lack of comparative material appears to have resulted in some inaccuracies or misinterpretations of the original study. For 23/41 ni.bio.br | scielo.br/ni W. Leo Smith, Michael J. Ghedotti, Omar Domínguez-Domínguez, Caleb D. McMahan, Eduardo Espinoza, Rene P. Martin et al. example, Parenti, Randall (2020) listed the number of lateral-line scales in Hemilutjanus as 108-115, but the lateral-line scale count in Hemilutjanus is 64-66, which overlaps with the 37-69 lateral-line scales found in other malakichthyids (Jordan, Eigenmann, 1890; Tab. 2; current study). Given their focus elsewhere, the limited time since the publication of their work, and the limited discussion of Hemilutjanus over the last 175 years, we believe that it is preferable to refer Hemilutjanus to Malakichthyidae to emphasize the close relationship between Hemilutjanus, Malakichthys, and Verilus and to avoid the recognition of a monotypic family that is substantially like its sister group in physiognomy and meristics. Additionally, the placement of Hemilutjanus in the Malakichthyidae represents the addition of an Eastern Pacific representative in the family such that the family is now distributed (generally in deep water) in the eastern and western tropical and temperate regions of the Atlantic, Indian, and Pacific Oceans. A preliminary morphological examination across the Acropomatiformes suggests that Hemilutjanus, Malakichthys, and Verilus can be united as a clade and separated from all other acropomatiforms by a suite of four characters: 25 vertebrae (10 precaudal and 15 caudal), 10 dorsal-fin spines, the presence of a supramaxilla, and 37-69 lateral-line scales. Comparative meristic and character data that are useful for recognizing an expanded Malakichthyidae among all acropomatiform families can be found in Tab. 2.
Given the revised classification of Hemilutjanus among the malakichthyids, the implications for these findings relative to the historic allies in Bleeker's Lutjanini and the "Serranidae" need clarification. The lutjanins, although only a handful of families, have been consistently recovered across multiple orders Davis et al., 2016;Sanciangco et al., 2016;current study) including the Acanthuriformes (Haemulidae and Lutjanidae), Acropomatiformes (Banjosidae), and Scombriformes (Arripidae). The "Serranidae" used by Katayama (1959) is demonstrably polyphyletic with representatives spread across the Acanthuriformes, Acropomatiformes, Centrarchiformes, and Scorpaeniformes (e.g., Davis et al., 2016;Sanciangco et al., 2016). Unlike Bleeker's Lutjanini, the "Serranidae" has received a lot of attention. Initially, Gosline (1966) and later Johnson (1983) progressively restricted Katayama's (1959) "Serranidae" to a subset of its former taxa with four characters, most notably the presence of an opercle with three spines. Beginning with Imamura, Yabe (2002), ichthyologists began recognizing that the "serranids" may be more closely related to mail-cheeked fishes than most other so-called percoid fishes. Shortly thereafter, numerous molecular studies using 3-10 loci corroborated this close relationship with mail-cheeked fishes, notothenioids, and percids; these studies demonstrated that the "Serranidae" was not monophyletic, and that many of the historic "serranids" excluded by Gosline (1966) and Johnson (1983) belong among the Acropomatiformes (e.g., Chen et al., 2003;Dettaï, Lecointre, 2004, 2005Smith, Wheeler, 2004, 2006Craig, Hastings, 2007;Smith, Craig, 2007;Li et al., 2009;Smith et al., 2009;Lautredou et al., 2013;. Our analysis included seven "serranids" and one percid in our genome-scale analysis, and we recovered a paraphyletic "Serranidae" relative to the one included percid. Our results provide further evidence for the non-monophyly of the "Serranidae" and the recognition that several former "percoid" and mail-cheeked fish groups are nested among a non-monophyletic "Serranidae" as was first studied in detail by Smith, Craig (2007).

Stereolepis.
Our results separated Stereolepis from its frequent ally, Polyprion. While 24/41 ni.bio.br | scielo.br/ni Phylogeny of Hemilutjanus and the Acropomatiformes these two genera have been frequently classified in the Polyprionidae (e.g., Nelson, 2006;Nelson et al., 2016), there are no synapomorphies that unite these two genera, and several studies have also treated them as independent clades (e.g., Gosline, 1966;Johnson, 1983Johnson, , 1984. There are no morphological phylogenetic analyses that have included both genera, and half of the eight prior molecular studies that included both Polyprion and Stereolepis have recovered the two genera independently in their resulting phylogenies (Smith, Wheeler, 2006;Smith, Craig, 2007;Betancur-R et al., 2013a;Rabosky et al., 2018;Fig. 2).
Given the separation of Polyprion and Stereolepis in most molecular phylogenetic analyses (Figs. 2-5), the lack of morphological evidence placing them together in the Polyprionidae, and the strong support for their separation in this and previous studies, we herein describe a new family for the two charismatic species in Stereolepis. Correspondingly, Polyprionidae is hereby restricted to species in the genus Polyprion. The recognition of new and revised families for separate clades is critical as we move toward a well-supported monophyletic classification of the Acropomatiformes and Percomorpha. Diagnosis. Species in Stereolepididae can be distinguished from all other acropomatiforms by a unique combination of 26 vertebrae (12 precaudal and 14 caudal), 9-12 dorsal-fin spines, and the presence of a supramaxilla (Tab. 2). Species in Acropomatidae, Banjosidae, Epigonidae, Glaucosomatidae, Malakichthyidae, Pempheridae, Synagropidae, and Schuettea have 24-25 vertebrae, and species in Champsodontidae, Creediidae, Dinolestidae, and Lateolabracidae have 27 or more vertebrae. The remaining acropomatiform families have overlapping total vertebral counts, but there is also diagnostic variation in the number of precaudal vertebrae. Among acropomatiforms with 26 vertebrae, Howellidae, Ostracoberycidae, Scombropidae, and Symphysanodontidae have 10-11 precaudal vertebrae and species in Polyprion have 13 precaudal vertebrae. These counts differ from the 12 precaudal vertebrae found in Stereolepis. Among acropomatiforms with 26 vertebrae (and 12 precaudal vertebrae), species in the Bathyclupeidae and Hemerocoetidae have eight or fewer dorsal-fin spines compared with the 9-12 dorsal-fin spines found in Stereolepis. Finally, the Pentacerotidae can be separated by its lack of a supramaxilla; whereas, this element is present in the jaws of stereolepidids. These features and a variety of other meristic counts and character states for acropomatiforms are summarized in Tab. 2.  Martin et al. al., 2016). In contrast, Tominaga (1968) suggested that Schuettea belonged in its own family that is closely allied with the Pempheridae (a family Regan [1913] allied with Monodactylidae). We included Monodactylus and the available sequences for Schuettea in our phylogenetic analysis of acropomatiforms following Tominaga's (1968) suggestion.

Schuettea. The two species in
In our analysis, Schuettea was recovered among the Acropomatiformes with strong support, and it was separated from its traditional ally, Monodactylus. The specific placement of Schuettea was not well supported in our analysis (Fig. 5); it was resolved sister to Champsodon in our most likely hypothesis (Fig. 5). Our analysis did not place Schuettea sister to Pempheridae, and there are multiple moderately or strongly supported nodes separating these two clades (Fig. 5). In support of his pempherid-Schuettea hypothesis, Tominaga highlighted that these clades share 10+15 vertebrae, a lateral line that reaches to the posterior margin of the caudal fin, and anteriorly extended epaxial muscles that reach the frontals. Additionally, Tominaga highlighted three features that were unique to Schuettea in his study: rib-like ossicles on the first and second hemal spines, a posterior extension of the gas bladder, and three anal-fin pterygiophores inserted anteriorly to the first hemal spine. Our preliminary osteological investigation focusing on acropomatiforms found that Schuettea (based on a dried skeleton) has cycloid scales (also shared with some or all acropomatids, bathyclupeids, creediids, dinolestids, hemerocoetids, pempherids, scombropids, synagropids, and Verilus), has a reduced number of dorsal-fin spines (five) and a larger number of dorsal-and analfin rays (28-32), has a spineless first dorsal-fin pterygiophore, and lacks a procurrent spur and supramaxilla (Johnson, 1984;Rosa, 1995;Nelson, 2006; Yamanoue, 2016; current study; Tab. 2). While externally dissimilar, many of these features are shared with the Champsodontidae, Creediidae, and Hemerocoetidae that have been typically recovered as a clade (Near et al., , 2015Thacker et al., 2015;Sanciangco et al., 2016;Ghedotti et al., 2018;Satoh, 2018;Fig. 2) and where we recovered Schuettea in our analysis (Fig. 5). Clearly, a more detailed examination using morphological and molecular data is needed to conclusively place Schuettea. Because of the nature of the preliminary morphological examination and the poor molecular support, we are not prepared to describe a new family for Schuettea. There remains a good chance that it is sister to or within a currently recognized acropomatiform family, and subsequent research will be needed to determine whether it should be classified within that family or in its own family. For now, we classify both species of Schuettea as incertae sedis in the Acropomatiformes, and this clade can be diagnosed from all other acropomatiforms by having 24 vertebrae (10 precaudal and 14 caudal), five dorsal-fin spines, and 28-32 anal-fin and 28-31 dorsal-fin rays (Tab. 2).
Phylogeny of the Acropomatiformes. The Acropomatiformes is one of the major clades in the recently recognized Eupercaria, and several early large-scale molecular phylogenies circumscribed this clade without recognizing it or focusing on its intrarelationships (Smith, Wheeler, 2006;Smith, Craig, 2007;Betancur-R et al., 2013a;Near et al., , 2015Thacker et al., 2015;Fig Ghedotti et al. (2018) and Satoh (2018) were the first studies to explicitly focus on the intrarelationships of the Acropomatiformes (Fig. 2). Across previous acropomatiform studies (Fig. 2), the 26/41 ni.bio.br | scielo.br/ni Phylogeny of Hemilutjanus and the Acropomatiformes lack of consistent clades, the few repeated results, and the limited number of strongly supported nodes is somewhat surprising. In an attempt to resolve relationships within the Acropomatiformes with strong support, we purposefully chose taxa to cover the diversity of acropomatiforms, and we greatly expanded the number of base pairs (Tab. 1). Increasing the number of base pairs in an analysis often results in phylogenies with more support and well-supported hypotheses (e.g., Harrington et al., 2016;Longo et al., 2017;Martin et al., 2018). This increase in data combined with being the first study to include all families of acropomatiforms resulted in phylogenies with considerably more support than previous studies (Figs. 3-4). Despite many new hypothesized relationships in our study, there are several clades of acropomatiforms that we recovered that recent molecular studies have also recovered. There are four clades shared among our analyses and many previous studies that we will focus on beyond the placement of Hemilutjanus: 1) Acropomatidae, Epigonidae, Howellidae, Ostracoberycidae, Scombropidae, and Symphysanodontidae; 2) Banjosidae and Pentacerotidae; 3) Champsodontidae, Creediidae, Hemerocoetidae, and potentially Schuettea; and 4) Glaucosomatidae and Pempheridae .
The largest repeated clade found within the acropomatiforms is the Acropomatidae, Epigonidae, Howellidae, Ostracoberycidae, Scombropidae, and Symphysanodontidae or what we refer to as the Acropomatoidei. The acropomatoids were first grouped together by  who included the Acropomatidae, Howellidae, and Ostracoberycidae in their study. Subsequent studies by Near et al. (2015), Thacker et al. (2015), Davis et al. (2016), Sanciangco et al. (2016, Ghedotti et al. (2018), andSatoh (2018) continued to recover this clade as new families and more data were added to the analyses. Interestingly, the acropomatoids were not recovered in analyses based on fewer than 5.000 base pairs (i.e., Smith, Wheeler, 2006;Smith, Craig, 2007;Tab. 1) or in analyses with substantial (>50%) missing data (e.g., Betancur-R et al., 2013a;Mirande, 2016;Rabosky et al., 2018;Tab. 1). Given the recurrent discovery of this clade in studies that had sufficient DNA-sequence data (i.e., more than 5.000 bps) and studies that emphasized overlapping data relative to increased taxonomic sampling (i.e., studies with less than 30% missing data), it is possible that many of the conflicting results depicted in Fig. 2 are due to data insufficiency rather than data conflict. The acropomatoids have substantial morphological variation, and our preliminary morphological investigation did not identify any synapomorphies for this group. Species in this clade are predominantly found in the deep sea (although many deep-sea families have species that reside in shallower waters; Fig. 6), and the acropomatoids include many of the acropomatiforms that were formerly included in the "serranids" (sensu Katayama, 1959).
In addition to the acropomatoids, one of the most frequently recovered clades in studies that have included many acropomatiforms is the sister-group pairing of Banjosidae and Pentacerotidae. This sister-group pairing has been recovered in every molecular analysis that included both families (Fig. 2) except Satoh (2018). In a detailed morphological study of the Pentacerotidae, Kim (2012) included Banjosidae as one of his outgroup taxa. Although there was no formal outgroup analysis, Kim suggested that Chaetodontidae and Ostracoberycidae were the most likely sister groups to the Pentacerotidae. Subsequent molecular analyses consistently place the Chaetodontidae within the Acanthuriformes and Ostracoberycidae in the Acropomatiformes Davis et al., 2016;Sanciangco et al., 2016;Smith et al., 2016). Most molecular 27/41 ni.bio.br | scielo.br/ni W. Leo Smith, Michael J. Ghedotti, Omar Domínguez-Domínguez, Caleb D. McMahan, Eduardo Espinoza, Rene P. Martin et al. studies have recovered ostracoberycids among the acropomatoids (Fig. 2). Interestingly, Kim (2012) noted that he allied the chaetodontids with the pentacerotids because chaetodontids shared four of the pentacerotid synapomorphies (his SA3, SA4, SA5, and SA7) and because they have strongly compressed bodies. Banjosids share these four formal synapomorphies and this one informal synapomorphy (compressed bodies) with the pentacerotids (Kim, 2012). There is no explanation in Kim (2012) for why chaetodontids were preferred over banjosids, but given the molecular results presented by Near et al. ( , 2015, Thacker et al. (2015), Davis et al. (2016), Mirande (2016), Sanciangco et al. (2016), Ghedotti et al. (2018), Rabosky et al. (2018, and the current study , it is clear that this banjosid sister-group relationship is supported and should be explored more explicitly with morphological data. Another clade of acropomatiforms that is consistently recovered is the group that includes Champsodontidae, Creediidae, Hemerocoetidae, and, potentially, Schuettea. We refer to this clade informally as the "champsodontoids". While Creediidae and Hemerocoetidae have been consistently recovered together in previous studies ( Fig. 2; Tab. 1), the placement of Champsodontidae among, or even in, the acropomatiforms has been the most problematic family-level placement in the order (Tab. 1). Despite some ambiguity, this clade has been consistently recovered in previous studies (Near et al., , 2015Thacker et al., 2015;Sanciangco et al., 2016;Ghedotti et al., 2018;Satoh, 2018;Tab. 2) with a few exceptions that potentially lacked sufficient data (Smith, Craig, 2007) or had extensive missing data (Mirande, 2016;Rabosky et al., 2018). Relative to other acropomatiforms, the champsodontoids are characterized generally by larger vertebral counts, the loss of both the procurrent spur and supramaxilla, more median fin-ray elements, and few to no dorsal-and anal-fin spines (Tab. 2). The monophyly and support of this clade will depend on the inclusion, or not, of Schuettea and the phylogenetic placement of the clade given its movement in our analyses .
The final acropomatiform clade that is consistently recovered is the grouping of Glaucosomatidae and Pempheridae (Betancur-R et al., 2013a;Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;Mirande, 2016;Sanciangco et al., 2016, Ghedotti et al., 2018Rabosky et al., 2018 [albeit outside of the Acropomatiformes]; Satoh, 2018;current study;Figs. 2-5). This sister-group pairing is the most consistent result across all acropomatiform studies, and it has been found in all analyses with more than six loci ( Fig. 2; Tab. 1). While there are no known morphological synapomorphies to unite this group, the ubiquity of their relationship across molecular studies (Fig. 2) provides striking support for the placement of these two families together. This clade, while frequently recovered with DNA-sequence data, has not been recovered in morphological analyses and would benefit from morphological investigations, particularly considering Tominaga's (1968) hypothesis that pempherids and Schuettea are closely related.
While we are beginning to recognize well-supported clades across a diversity of studies examining the Acropomatiformes, the monophyly of the traditional "Acropomatidae" is continually being rejected, and it should not be recognized further. The family Acropomatidae was described (as Acropomidae) by Gill (1893) as one of the families of his Percoidea. Most studies in the late 19th century and early 20th century allied members of the modern Acropomatidae, Malakichthyidae, and Synagropidae with the Apogonidae or "Serranidae" (sensu Katayama, 1959) and did not treat them as a single natural group (e.g., Jordan, Richardson, 1910;Regan, 1913;Schultz, 1940 Katayama (1959), in the most extensive comparative morphological investigation of the "Acropomatidae", split these taxa across three subfamilies (his Acropominae, Döderleininae, and Malakichthyinae), none of which has the same composition as any modern "acropomatid" family. Despite Katayama (1959: fig. 39) classifying these species in three subfamilies, he did illustrate the included genera as what we would refer to as a monophyletic group in his pre-cladistic phylogeny. Further, he placed this assemblage in a trichotomy with (Stereolepis + (Coreoperca+Siniperca)) and Lateolabrax. This clade was sister to a clade composed of Niphon and Ostracoberyx, and these taxa were referred to as the "Acropoma-stem group". Gosline (1966) treated the "acropomatids" as part of his "oceanic percichthyids". Johnson (1984: 464) recognized the modern "Acropomatidae", but he noted that he knew "of no synapomorphy that unites the acropomatids, and further work will be necessary to test their monophyly". Following Johnson's (1984) study, many studies treated the "Acropomatidae" as a family that included the modern Acropomatidae, Malakichthyidae, and Synagropidae (Nelson et al., 2016), but other authors variously included members of the Howellidae, Scombropidae, Symphysanodontidae, and Polyprionidae within the Acropomatidae (e.g., Heemstra, 1986;Nelson, 1994Nelson, , 2006Heemstra, Yamanoue, 2003). More recent authors conducting morphological analyses (e.g., Prokofiev, 2007;Schwarzhans, Prokofiev, 2017) have provided evidence to refute the monophyly of the "Acropomatidae". Across all these morphological studies, it is clear that a close relationship among a number of acropomatiform clades were being recognized, but the limits and relationships of the families and the order as a whole remained unclear.
Beginning with Smith, Craig (2007), molecular studies began to include multiple genera of "acropomatids" and were consistently finding the family polyphyletic (Betancur-R et al., 2013a;Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;Mirande, 2016;Sanciangco et al., 2016;Ghedotti et al., 2018;Rabosky et al., 2018;Fig. 2;Tab. 1). Across the previous and current molecular studies, the Acropomatidae, Malakichthyidae, and Synagropidae were always found independent of each other . The sister groups of each family varied among studies. Seven of the previous studies and the current study recovered a clade composed of the acropomatoids (Near et al., , 2015Thacker et al., 2015;Sanciangco et al., 2016;Ghedotti et al., 2018;Satoh, 2018;Figs. 2-5). Among these studies and the current study, the most common sister group for the Acropomatidae was the Ostracoberycidae, which was not one of the families that traditional studies had classified within the Acropomatidae. The sister group to Malakichthyidae was less consistent across studies or across methods in our study . Our core ML analysis recovered Malakichthyidae sister to a clade composed of Bathyclupeidae, Champsodontidae, Creediidae, Stereolepididae, and Synagropidae (Fig. 3). In contrast, our results from the ST analysis place a clade composed of Banjosidae, Dinolestidae, Pentacerotidae, and Stereolepididae sister to Malakichthyidae (Fig. 4). The only shared member from the malakichthyid sistergroup clades between our two analyses was Stereolepididae, which is, overall, the most consistent sister group in previous studies (Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;Rabosky et al., 2018). The specific sister group to Malakichthyidae remains contentious (Figs. 2-5). The third "acropomatid" family is Synagropidae, which we consistently recovered sister to Bathyclupeidae across methods and datasets with strong support (Figs. 3-5). In contrast, no previous studies recovered a bathyclupeid sister group for Synagropidae, and the only repeated sister group in previous analyses was Howellidae (Mirande, 2016;Rabosky et al., 2018;Fig. 2). While these two studies recovered a howellid sister group, most studies (noted above) place Howellidae in a distantly related and well-supported clade with Acropomatidae, Epigonidae, Ostracoberycidae, Scombropidae, and Symphysanodontidae. Therefore, every previous molecular study with multiple "acropomatid" families and the current study show that the "Acropomatidae" is polyphyletic (Figs. 2-5). Most studies have found the Acropomatidae sister to Ostracoberycidae (Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;current study;Figs. 2-5). The current study shows that Synagropidae is sister to  and the placement of Malakichthyidae among the acropomatiforms has conflicting results (Figs. 2-5).
Our study included considerably more data than all previous studies that included substantive acropomatiform taxa (Tab. 1), and it was the first study to include every acropomatiform family and every genus classified as incertae sedis. Our results were recovered with considerable support (>80% of nodes were well or moderately well supported across all three analyses; Figs. 3-5), and outside of the placement of the champsodontoids, most relationships were shared among the acropomatiforms across methods (Figs. 3-4). Relative to studies with limited sequence data (e.g., Smith, Craig, 2007) and studies with extensive missing data or goals well outside of the Acropomatiformes (e.g., Betancur-R et al., 2013a;Sanciangco et al., 2016;Rabosky et al., 2018), there is more consistency of relationships than not (e.g., Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;Ghedotti et al., 2018;Satoh, 2018;current study). Perhaps much of the conflict that researchers are finding among studies of percomorph groups is due more to insufficient data rather than conflicting data. Certainly, including as many species as possible has benefits (Wiens, Tiu, 2012;Borden et al., 2013;Tang et al., 2021), but the inclusion of taxa with insufficient comparable data has resulted in contradictory phylogenies for the Acropomatiformes (Figs. 2-5; Tab. 1).
Evolution of the Acropomatiformes. Through our work to resolve the placement of Hemilutjanus, we have taken this opportunity to examine the limits and relationships of the Acropomatiformes. One of the striking changes in our understanding of percomorph relationships that came following our improved understanding of fish relationships or Smith's (2010:523) impending "renaissance" brought on by molecular systematics is that we have approximately 20 repeatedly recovered clades (orders in Davis et al., 2016) of percomorphs of which three-quarters of these clades are completely new to science. These are newly recognized clades, so we have studied little more than their phylogeny since their identification over the last decade (Betancur-R et al., 2013a;Davis et al., 2016;Sanciangco et al., 2016;Smith et al., 2016;Rabosky et al., 2018). This is a dynamic time in fish phylogenetics because the classification of fishes is fluid and changing, and we are now able to explore the morphology, biology, and evolution of these percomorph orders (e.g., Thacker, 2014;Davis et al., 2016;Harrington et al., 2016;Ghedotti et al., 2018;Rabosky et al., 2018;Girard et al., 2020). The dominant biological phenomena that have been recognized among acropomatiforms are that this relatively small order of ~300 species includes a surprisingly large number of bioluminescent and deep-water species (for percomorphs) that previous studies have suggested evolved independently multiple times (Davis et al., 2016;Ghedotti et al., 2018) and that they are 30/41 ni.bio.br | scielo.br/ni Phylogeny of Hemilutjanus and the Acropomatiformes poorly represented in the Eastern Pacific (Schwarzhans, Prokofiev, 2017).
The evolution of bioluminescence among acropomatiforms has been explicitly studied by Davis et al. (2016) and Ghedotti et al. (2018). Both studies highlighted that approximately 10% of acropomatiforms are bioluminescent, and both studies included representatives of all four acropomatiform families that have bioluminescent species (Acropomatidae, Epigonidae, Howellidae, and Pempheridae; Tab. 1). Davis et al. (2016) suggested that bioluminescence evolved in the Acropomatidae, Epigonidae+Howellidae, and Pempheridae. Ghedotti et al. (2018) hypothesized that representatives of each family with bioluminescent species evolved bioluminescence independently in their analysis. Their study included more acropomatiforms, but it included fewer acropomatiform families and less DNA-sequence data. Further, they noted that bioluminescence may have evolved multiple times independently among epigonids. Our results present a different phylogeny of acropomatiform fishes relative to these two prior studies, and our phylogeny suggests three independent evolutions of bioluminescence: Acropomatidae, Pempheridae, and the ancestor of Epigonidae and Howellidae (Fig. 6). However, and as noted by Ghedotti et al. (2018), none of these families is wholly bioluminescent with between 6% and 92% of the species in each family being bioluminescent (Acropomatidae: 12 of 13 species are bioluminescent; Epigonidae: 3 of 47 species; Howellidae: 1 of 9 species; Pempheridae: 5 of 85 species). Further, Ghedotti et al. (2018) noted that the epigonids in Epigonus and Rosenblattia have different anatomies for their bioluminescent organs (Ghedotti et al., 2018). Taken together, this not only suggests that each bioluminescent family of acropomatiforms evolved bioluminescence independently, but it also suggests that epigonids may have evolved bioluminescence twice. If our higher-level phylogeny is correct and the species-level phylogeny and anatomical descriptions for the Epigonidae in Ghedotti et al. (2018) are correct, bioluminescence would have likely evolved at least five times in the Acropomatiformes, once in shallow water (Pempheridae) and four times in the deep sea (Acropomatidae, Howellidae, and twice in the Epigonidae). To further resolve questions around the evolution of bioluminescence in the Acropomatiformes, we need denser taxon sampling in the Acropomatidae, Epigonidae, Howellidae, and Pempheridae, with a particular emphasis on the Epigonidae and Pempheridae. Given our phylogenetic hypothesis, it is noteworthy that the shallow water bioluminescent pempherids obtain all necessary bioluminescent molecules through their diets, whereas the deep-water bioluminescent acropomatiforms rely on symbiotic bacteria (Ghedotti et al., 2018;Bessho-Uehara et al., 2020) and are restricted to the acropomatoids that invaded the deep sea prior to the evolution of bioluminescence.
Although not explicitly discussed in previous acropomatiform studies, the acropomatiforms are unusual for percomorphs in that more than half of the species can be found in deep water ≥ 200 m (Froese, Pauly, 2021). Relative to other percomorph ordinal-level clades identified by Davis et al. (2016), no other order is dominated by deep-sea fishes (Davis et al., 2016;Nelson et al., 2016;Froese, Pauly, 2021). Optimizing the invasions of the deep sea among acropomatiforms using parsimony and maximum likelihood demonstrate that acropomatiforms invaded the deep sea and shallow water multiple times (Fig. 6). Using parsimony, there is much ambiguity in the specific acropomatiform clades that have invaded the deep sea. Using maximum likelihood, we found that there was one invasion in the ancestor of the Acropomatiformes and 31/41 ni.bio.br | scielo.br/ni W. Leo Smith, Michael J. Ghedotti, Omar Domínguez-Domínguez, Caleb D. McMahan, Eduardo Espinoza, Rene P. Martin et al. one invasion in the ancestor of the Champsodontidae. The number of returns to shallow water suggested by our parsimony optimization are hampered by many ambiguous or equally parsimonious reconstructions, but in maximum likelihood, we see independent invasions in the Glaucosomatidae+Lateolabracidae+Pempheridae, Champsodontidae+Schuettea, Creediidae, Dinolestidae, Hemilutjanus, Scombropidae, and Stereolepididae (Fig. 6). Transitions between shallow water and the deep sea and the evolution of bioluminescence are not that common in the Eupercaria (Froese, Pauly, 2021). Repeated transitions between shallow and deep environments within the 300 acropomatiform species is noteworthy. Similarly, the multiple evolutions of bioluminescence among just 300 species of acropomatiforms is also startling. These habitat invasions and luminescent adaptations are uncommon (but not rare) among percomorphs, but the frequency of these specializations and transitions demands further research on this largely unexplored and newly discovered order of fishes.
Finally, this study highlights our natural biases to compare fishes from similar habitats. One of the likely reasons that no one had found the closest relatives to Hemilutjanus is that, while a shallow water fish, its closest relatives are in the deep sea. Further, its deepsea relatives are poorly represented in the Eastern Pacific Ocean. By searching broadly for its potential relatives and benefiting from the molecular phylogenies that have grouped many of the species excluded from the "serranids" (sensu Johnson, 1993) into the Acropomatiformes (e.g., Smith, Wheeler, 2006;Smith, Craig, 2007;Betancur-R et al., 2013a;Near et al., , 2015Thacker et al., 2015;Davis et al., 2016;Mirande, 2016;Sanciangco et al., 2016;Rabosky et al., 2018; Tab. 1), we have been able to place Hemilutjanus in the Malakichthyidae. This phylogenetic placement also adds to the diversity of acropomatiforms in the Eastern Pacific Ocean, particularly in shallow waters. The addition of another shallow water fish in the Acropomatiformes, particularly one sister to a deep-sea clade, serves as a good reminder that acropomatiforms have many transitions between deep-sea and shallow-water habitats (Fig. 6), particularly for a recent group in the Eupercaria. Most orders dominated by deep-sea fishes have few to no transitions between shallow and deep-water habitats (e.g., Alepocephaliformes, Myctophiformes, Stomiiformes; Davis et al., 2016;Nelson et al., 2016). These transitions and the biases scientists have for making comparisons of fishes from similar locations and habitats may best explain why the Acropomatiformes was not recognized earlier and why its relationships have been so poorly understood.
Comparative osteology and myology of the caudal fin in the Paracanthopterygii (Teleostei: Acanthomorpha