Taxa sampling

The specimens were collected during two oceanographic cruises onboard the Brazilian R/V Alpha Crucis using a semi-balloon bottom open trawl with 27 meters in the lower rope, mesh sizes of 100 mm in the wings and body, and 25 mm in the codend, and fish traps. The cruises were launched on the continental slope of southern Brazil off Ilhabela, São Paulo (24°22’S– 25°01’S, 43°45’W–44°32’W), between 390 and 1,500 m deep, and off Florianopolis, Santa Catarina (28°15’S–28°42’S, 46°44’W–47°12’W), between 274 and 1,232 m deep. Both cruises sampled at target depths in a range of 100 m between the minimum and maximum depth; however, due to the use of an open trawl, specimens may be collected during the release and recovery net operations. Thus, all depth mentions refer to the target sampling depths. The sample points are represented by red circles on a map produced in QGIS v. 3.24.2 [28] (Fig 1). Soon after the net recovery, the specimens were sorted and selected, and a musculature sample was removed from the right side. Each specimen received an identification number that matched the tissue sample. About 2,000 tissue samples were obtained and preserved in 96% ethanol, including 1,400 from actinopterygians.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Map of the western South Atlantic, Brazilian margin, indicating the sampling areas (red circles).

Sampling areas located off Ilhabela, São Paulo, Brazil (24°22’S– 25°01’S, 43°45’W–44°32’W) between 390 and 1,500 m; and off Florianopolis, Santa Catarina, Brazil (28°15’S–28°42’S, 46°44’W–47°12’W) between 274 and 1,232 m.

https://doi.org/10.1371/journal.pone.0347925.g001

The specimens were preserved in formalin and, in the laboratory, transferred to ethanol for collection preservation. The tissue samples are deposited at the Coleção Biológica Prof. Edmundo Nonato – COLBio.

Collection permits were issued by the Instituto Chico Mendes de Conservação da Biodiversidade (SISBIO permits #28054−4, 82624−1), Secretaria da Comissão Interministerial para Recursos do Mar da Marinha do Brasil (Portaria No. 223), and the Comitê de Ética em Uso de Animais em Pesquisa e Ensino do Instituto Oceanográfico da Universidade de São Paulo (CEUA permit #16 to MRSM).

Laboratory procedures

The specimens were previously identified at the species level using taxonomic revisions and species keys [29,30]. The DNA was extracted using the Wizard Genomic DNA Purification (Promega Corp., WI, USA) and DNeasy Blood & Tissue Kit (Qiagen, USA) extraction kits, following the manufacturer’S protocols. Subsequently, the extracted DNA was quantified using a Nanodrop ND-1000 spectrophotometer. The mitochondrial cytochrome c oxidase subunit I (COI) gene was amplified and sequenced using universal primers (Table 1) [31,32].

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Universal primers used in this study.

https://doi.org/10.1371/journal.pone.0347925.t001

The Polymerase chain reaction (PCR) was performed using 12.5 µL of 2X GoTaq® Green Master Mix, 0.5 µL of each primer, 2 µL of DNA, and Nuclease–Free Water to a final volume of 25 µL. Fragments of the COI gene were amplified using an initial denaturation at 95°C for 3 minutes, followed by 30–35 cycles of amplification: denaturation at 94°C for 30 seconds, annealing at 50°C to 54°C for 45 seconds, and extension at 72°C for 1 minute. The final extension was at 72°C for 7 minutes. PCR products were visualized on a 1% agarose gel and purified with the SAP–Exo purification kit (Jena Bioscience, Jena, Germany). Bidirectional sequencing was performed using the BigDye Terminator v3.1 Cycle Sequencing kit on an ABI 3500 Series Genetic Analyzer sequencer (Applied Biosystems®) at the Instituto de Biotecnologia (IBTEC), Botucatu, São Paulo, Brazil

Sequence identification and molecular data analysis

The chromatograms were edited and quality-checked, and the consensus sequences were generated in Geneious v. 4.8.5 [33]. Each sequence was verified by the Basic Local Alignment Search – BLASTn [34] and BOLD Barcoding ID [35] from a threshold of similarity of 98% with the sequences available on those databases.

All sequences were submitted to the BOLD (https://www.boldsystems.org/) database under the project “DEEP OCEAN: Diversity Ecology and Evolution of Deep-Sea Fishes” (code: DOP). The multiple alignment was performed in Geneious v. 4.8.5, and the nucleotide substitution model Kimura2-Parameters (K2P) was used to determine distances within taxonomic levels using MEGA software v11.0 [36].

A Maximum Likelihood (ML) phylogenetic tree was performed by the software RAxML v8.2.10 [37]. The best ML topology was found using the evolutionary model GTR + G, and node support was estimated from 1,000 bootstrap pseudoreplicates. The phylogenetic tree was visualized and edited using FigTree v. 1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/). As outgroup, sequences of three chondrichthyans also collected during the cruises were used: Scyliorhinus haeckelii (Miranda Ribeiro, 1907) (DOP145−24); Galeus mincaronei Soto, 2001 (DOP146−24); and Zameus squamulosus (Günther, 1877) (DOP147−24). The taxonomic classification adopted herein follows Betancur et al. [38], except for Lopholatilus villari Miranda Ribeiro, 1915, which follows Fricke et al. [39].

For the species with unclear similarity results, i.e., those for which the threshold considered matched more than one species, the results were further investigated using all database sequences for each genus using both the maximum likelihood tree and genetic divergence. In these cases, the multiple alignment was performed using the WSA sequences, database sequences from BOLD Systems and GenBank [35,40] of congeners, and sequences from the closest genus and family as outgroups. The alignment was submitted to the online tool ElimDupes (https://www.hiv.lanl.gov/content/sequence/elimdupesv2/elimdupes.html) to compare sequences and remove duplicates, thereby reducing computational time for analysis. The intraspecific and interspecific K2P distances were calculated for each genus. Sequence groups were considered the same taxon if the distance within the group was < 2% [41]. The Maximum Likelihood tree by genus was performed using the evolutionary model GTR + G and 1,000 bootstrap pseudoreplicates. Therefore, species identification was performed using results from similarity searches, K2P distances, and ML phylogenetic trees by genus.

Deep-sea fish diversity from the western South Atlantic

While DNA barcoding has been extensively used to assess the diversity and taxonomy of deep-sea fish worldwide, regions such as the Brazilian EEZ remain largely underexplored [13,14,16,17]. The present work successfully identified a high diversity of deep-sea fishes from the Brazilian margin.

A total of 102 species of deep-sea actinopterygians were identified from 170 COI sequences, of which 84 species (82%) were previously reported in the Brazilian EEZ. Ten of those species were originally described from southern Brazil: Argentina brasiliensis Kobyliansky, 2004; Bembrops heterurus (Miranda Ribeiro, 1903); Genypeterus brasiliensis Regan, 1903; Kali colubrina Melo, 2008; Lophius gastrophysus Miranda Ribeiro, 1915; Lopholatilus villarii Miranda Ribeiro, 1915; Paralichthys triocellatus Miranda Ribeiro, 1903; Sciadonus alphacrucis Melo, Gomes, Møller & Nielsen, 2021; Synaphobranchus calvus Melo, 2007; and Polymixia carmenae Caixeta, Oliveira & Melo, 2024 [8,24,39,43–48]. In addition, nine species sequenced are endemic to southern Brazil or also occur in nearby areas in Uruguay and Argentina: Anthias menezesi Anderson & Heemstra, 1980; Argentina brasiliensis Kobyliansky, 2004; Coelorinchus marinii Hubbs, 1934; Genypterus brasiliensis Regan, 1903; Lopholatilus villarii Miranda Ribeiro, 1915; Merluccius hubbsi Marini, 1933; Polymixia carmenae Caixeta, Oliveira & Melo, 2024; Sciadonus alphacrucis Melo, Gomes, Møller & Nielsen, 2021; and Synaphobranchus calvus Melo, 2007.

The findings include two species recently described, the bythitid Sciadonus alphacricus Melo, Gomes, Møller & Nielsen, 2021, described by Melo et al. [24] and subsequently barcoded by dos Reis Júnior et al. [6], and the polymixiid Polymixia carmenae Caixeta, Oliveira & Melo, 2024 [8], which was described based on morphology (morphometrics, meristics, osteology) and molecular characters (COI, 12S, 16S), in addition to one putative new stomiid species is under investigation, assigned to the genus Photonectes. (Fig 7B). The genus Alepocephalus urges for  a thorough taxonomic review and Alepocephalus sp. can be a new species that deserves further investigation. There are three new records in the Brazilian continental slope, the macrourids Coryphaenoides striaturus Barnard, 1925, and C. subserrulatus Makushok, 1976 [49] (Fig 7C, 7D), and the beryciform Sio nordenskjoldii – interestingly, the three species are widely spread in the southern hemisphere, but C. subserrulatus was previously known from Uruguay and Argentina, S. nordenskjoldii from Uruguay, and C. striaturus from the Rio Grande Plateau [42,50–58].

Even though deep-sea zones are remote and difficult to access, many deep-sea species are subject to anthropogenic impacts from fishing and marine pollution. [59–61]. Indeed, five barcoded species are targets for the Brazilian commercial fisheries fleet, including the tilefish Lopholatilus villarii, the Argentinean hake Merluccius hubbsi, the blackfin goosefish Lophius gastrophysus, the pink cusk-eel Genypterus brasiliensis, and the silvery John dory Zenopsis conchifer (Lowe, 1852) [62,63]. According to the Brazilian official endangered species list [64], 16 deep-sea fish species are at risk of extinction, including Lopholatilus villarii, which is listed as Vulnerable.

Considering species already represented in online databases, we include sequences for 16 species that have sequences only from the Southern, Indian, or Pacific Oceans. Additionally, 26 species had sequences from the North Atlantic but not from the South Atlantic, and another 28 species had sequences from the eastern South Atlantic. In summary, a total of 70 species had the sequence coverage expanded to the WSA.

The COI sequences of nine species were not previously represented in any online database: the gadiform Nezumia atlantica (Parr, 1946); the ophidiiforms Selachophidium americanum (Nielsen, 1971) and Sciadonus alphacrucis; the pempheriform Parascombrops aff. spinosus (Schultz, 1940); the perciforms Bembrops heterurus and Peristedion imberbe Poey, 1861; the pleuronectiform Paralichthys triocellatus Miranda Ribeiro, 1903; the scombriform Thyrsites lepidopodea Lesson, 1831; and the stomiiform Aristostomias cf. grimaldii Zugmayer, 1913.

We could not identify sequences at the species level for four taxa: the aulopiform Lestidiops sp., the gadiform Nezumia sp., and the stomiiforms Photonectes sp. and Diplophos sp. These sequences do not match any sequences in the online databases with a similarity above 98%. For Lestidiops sp. and Nezumia sp., we were also unable to identify them to the species level using taxonomic keys. These results highlighted the effect of the absence of comparative sequences in online databases on species identification by DNA barcoding. In addition, the challenge of identifying it using both taxonomic keys and COI sequences can provide insights, such as the need for taxonomic revisions and the identification of putative new species, as in the case of Photonectes sp. The morphological characteristics of the specimen differ from those of other species in the genus, and combined with the genetic divergence, we consider it a putative new species.

The sequence of the gonostomatid Diplophos sp. has a similarity of 94% with sequences of Diplophos australis Ozawa, Oda & Ida, 1990 (S1 Table). Examining the specimen, we supposed that it could be a representative of the genus Manducus, however, the specimen is not in good condition because it was cut in half during collection, and only the anterior half was recovered, damaging the key morphological characteristics for identification, such as the skin, body photophores, and the relative position of dorsal and anal fins. Compared with the sequences of Manducus maderensis (Johnson 1890) available online, our sequence differs by 21.5%.

Among the bothids, the single species sequenced was identified as Monolene sessilicauda Goode, 1880, with 98% similarity to M. sessilicauda from Canada and Florida, the western North Atlantic, and the Gulf of Mexico, respectively. Another congener reported from Brazilian waters, M. antillarum Norman, 1933, has similarities in pigmentation, meristics, and morphometric data, suggesting the need of a comprehensive taxonomic revision of Monolene occurring in the western Atlantic [65].

According to Ward [41], individuals showing less than 2% genetic divergence in COI are more likely to belong to the same species, whereas those with greater divergence are more likely to be distinct species. Among the generated sequences, the mean distances within a species (0.4%) are congruent with those of intraspecific values. The mean genetic divergence within the genera, families, and orders is 2.4%, 7.9%, and 14.7%, respectively. Ward [41] also explored the divergences at those supraspecific taxonomic levels, finding values considerably higher than ours; however, the correspondence between genetic divergence and the taxonomic limits of genus, family, and order is not straightforward.

The barcoding tree generated by ML inference recovered 90% of the families and 47% of the orders as monophyletic. In addition, bootstrap values are below 60% in most clades that group orders and families, being higher in terminal branches. This reflects the effectiveness of the COI gene in identifying species, as discussed previously, and in recovering hypotheses of genus and family relationships. However, the use of the COI gene to resolve deeper phylogenies (i.e., order, class) is limited [12,32], resulting in a high number of polytomies and less efficient recovery of monophyletic clades.

Are online databases useful to identify deep-sea fishes?

Worldwide, there are many challenges to scientifically exploring the deep sea, including the need for an oceanographic vessel, the availability of technology (e.g., ROVs, HOVs, and deep-water nets), and the expensive operational costs [66,67]. In the context of better understanding Brazilian deep-sea biodiversity, even with these challenges, there was a significant increase in data from the 2000s onwards especially after the cruises of the French RV Thalassa (1999, 2000), the Brazilian RV Atlântico Sul (1996–1999) and the Brazilian FVs Diadorim and Solancy Moura (1996–2002), Brazilian OSV Astro Garoupa (2001, 2007), the USA RVs Luke Thomas (2009) and Seward Johnson (2011), the French R/V Antea (2015–present), and the Braziilan RV Alpha Crucis (2019, 2020) [20].

However, beyond the economic scope, the recent challenges in studying Brazilian deep-sea fishes are also due to the limited or no availability of tissues for molecular analyses. It was only in the past few years that tissue samples became available due to the recent expeditions, such as those conducted onboard the RV Alpha Crucis. As a result, the number of COI sequences available in online databases is limited, especially for endemic species. Such a paucity of samples and sequences poses many challenges for the use of DNA barcoding to properly identify WSA deep-sea fishes.

When analyzing the WSA sequences generated during this study for molecular species identification, we encounter several challenges: unrepresentative taxon coverage in online databases, unclear similarity results, and sequences that do not correspond to the actual taxa, likely due to misidentifications. S1 Table illustrates instances in which the closest sequences detected by BLASTn and the BOLD Barcoding ID resulted in unclear species identification, which could only be verified by reviewing the specimen morphologically, by further genetic distances, and by comparing all database sequences from each genus. Results from similarity analysis are efficient when the queried species is represented in the database; however, if it is not, these methods can yield a false identification [68].

In 34 cases, the similarity results showed that more than one species matched the generated sequences (>98%). For some cases, the identifications were then confirmed morphologically by examining museum specimens or checking original descriptions, such as Phenacoscorpius nebris, Anthias menezesi, and Argyropelecus aculeatus, while others need to be further investigated with all database sequences, i.e., not only the sequences presented in the similarity result, but by maximum likelihood tree and distance methods for each genus, including Rouleina attrita, Sphagemacrurus grenadae, and Merluccius hubbsi (S1–S3 Figs; S6–S8 Tables). A review of the literature indicates that most studies over the past decade have used tree topology for other analyses, with likelihood and Bayesian inference gaining strength, even though neighbor-joining remains the most widely used [12]. Additionally, in some cases, distance-based methods for species identification may not adequately represent species relationships and underestimate the potential of DNA barcoding [12].

Incongruences were found in two commercially important species, Merluccius hubbsi and Lopholatilus villarii, but for different reasons. The sequences of the Argentinean hake M. hubbsi, matched those also identified as M. hubbsi, but also with the North Pacific hake M. productus (Ayres, 1855), and the Southern hake M. australis (Hutton, 1872). The sequence of M. productus (Genbank accession number GU324172.1) was produced from a sample commercially purchased from the Pacific Ocean, and the sequences of M. australis (BOLD accession numbers EHPQ015–19 and EHPQ016–19) are from Argentina. However, since our samples were obtained from the southern Atlantic Ocean and the morphological identifications indicate M. hubbsi, in addition to most sequences with a similarity of 99–100% being of M. hubbsi, and finally the maximum likelihood recovered our sequences in the M. hubbsi clade, we understand that as two distinct cases of identification error in the online databases.

Another case returns more than one species of the same genus, the tilefish Lopholatilus villarii and the great northern tilefish Lopholatilus chamaeleonticeps Goode & Bean, 1879. Our sequence matched those of both L. villarri from the South Atlantic and L. chamaeleonticeps from the United States and Mexico. Sequences of the two valid species of Lopholatilus are available online, with an interspecific distance of 1.3%, below the 2% threshold commonly used to delimit species. Overall, the COI gene successfully delineates species; however, rates of evolution vary across taxa and, in some cases, such as the youngest species and synonyms, make it difficult to recognize species limits [10,41]. Furthermore, integrating distance-based and character-based analyses, as well as morphology, might reveal a synonym for the two species.

Similarly, Allocyttus verrucosus showed high similarity with A. verrucosus from the South Atlantic, Pacific, and Indian Oceans, and with A. folletti Myers, 1960, but also with Neocyttus helgae (Holt & Byrne, 1908), N. psilorhynchus Yearsley & Last, 1998, and N. rhomboidalis Gilchrist, 1906. Those genera and species are morphologically very distinctive and therefore difficult to confuse [69], but the oreosmomatids exhibit low genetic divergence and are difficult to separate using DNA barcoding [41].

A remarkable case is Helicolenus lahillei Norman, 1937 (Fig 5C), which has a similarity index >98% (98.03–99.84%) with the sequences available of seven out of nine valid species of the genus: H. avius Abe and Eschmeyer, 1972; H. barathri (Hector, 1875); H. dactylopterus (Delaroche, 1809); H. hilgendorfii (Döderlein, 1884); H. lahillei Norman, 1937; H. lengerichi Norman, 1937; and H. percoides (Richardson & Solander, 1842). Eschmeyer (1969) questioned the validity of H. hilgendorfii and H. lahillei based on the morphological similarity, referring to the latter as a subspecies of H. dactylopterus. This seems to be a similar case to the scorpionfish Setarches guentheri Johnson, 1862, which was hypothesized to comprise five nominal species, but all of which were synonymized based on morphological and genetic evidence [70]. In this case, we suppose it might be the same, widespread species that needs taxonomic revision.

Another challenging case was identifying synaphobranchid eels. Seven species were sequenced: Diastobranchus capensis Barnard, 1923; Meadia cf. abyssalis (Kamohara, 1938); Synaphobranchus affinis Günther, 1877; S. brevidorsalis Günther, 1887; S. calvus; S. oregoni Castle, 1960; and Symenchelis parasitica Gill, 1879. The sequences avaibable on-line were assigned to four of the seven valid species of Synaphobranchus, however, the genus still proved difficult to identify using DNA barcoding methods. Both BLASTn and BOLD Barcoding ID results show more than one species with similarity above 98%. In our results, the sequences of Synaphobranchus affinis showed similarity with some sequences identified online as Synaphobranchus kaupii Johnson, 1862 and Synaphobranchus brevidorsalis; S. calvus, with S. brevidorsalis and Diastobranchus capensis; and S. oregoni, also with S. kaupii. Notably, the species of Synaphobranchus are extremely difficult to identify [30,71–73], leading to possible misidentifications or the existence of multiple cryptic species.

The similarity results can be misinterpreted because the closest sequence match depends on the data’s reliability [74]. Here, we can identify a possible misidentification in the sequence of Bembrops heterurus (Miranda Ribeiro, 1903) in databases based on morphological identification and the type species locality of our sequenced specimen. Another study focused on DNA barcoding of deep-sea fish has also identified misidentifications in databases [75].

Erroneous sequence identification can explain the observed increase in genetic divergence among fish [41]. For instance, Bembrops heterurus does not match any conspecific sequences in online databases, indicating a distance greater than 5%. However, our specimen was collected near its type locality [46], whereas the sequence available does not include the locality for the conference. Therefore, based on morphological identification and type locality, we believe that the online database sequence represents a misidentification.

On the other hand, the DNA barcoding was extremely useful for the identification of four species that could not be promptly achieved by morphology: the macrourid Coryphaenoides striaturus; two species of bristlemouths that are extremely similar, Cyclothone microdon Günther, 1878, and C. pallida Brauer, 1902; and the trichiurid Benthodesmus simonyi (Steindachner, 1891). For those cases, the molecular identification was later confirmed morphologically using the original descriptions and/or taxonomic revisions.