Identification of pre-flexion fish larvae from the western South Atlantic using DNA barcoding and morphological characters

Sampling sites and specimen collection

Ichthyoplankton were sampled using twin bongo nets (mesh sizes 300 and 500 µm) towed at 0.5 m depth from surface for 10 min (Kelso & Rutherford, 1996; Castro, Bonecker & Valentin, 2005; Food Agriculture Organization (FAO), 2017) in coastal waters near islands from the northeastern part of Rio de Janeiro State, in the Campos Basin, Brazil. Two collecting expeditions were conducted, one in March and a second in April 2022 (Fig. 1). The first expedition was aimed at collecting around the Santana Archipelago, off the city of Macaé (22°24′37.7″S/41°42′20.7″W), in six different sampling localities. In the second expedition, the same localities around the Santana Archipelago were again sampled using the same protocol, with additional collecting at Feia Island, another coastal island off the city of Armação dos Búzios (22°43′26.7″S/41°55′18.7″W). Towed material was directed to the bottom of bongo nets, where a hollow plastic cup accumulated the collected material. The cup was then detached and all the collected material was transferred to pots where samples were preserved in 95% ethanol and transported to be stored in freezers for later sorting of fish larvae and eggs using stereomicroscopes in the laboratories of the Instituto de Biodiversidade e Sustentabilidade, Universidade Federal do Rio de Janeiro (NUPEM/UFRJ).

Molecular techniques

In the laboratory, fish larvae and eggs were separated from the remaining plankton. Eighty-nine randomly selected individual fish larvae were then photographed with a high-resolution stereoscope (LEICA M205 FA) at the Unidade Integrada de Imagem NUPEM/UFRJ, to get high quality images for posterior anatomical investigation. After specimens were photographed, the WIZARD PROMEGA kit was used to perform DNA extraction, with individuals digested entirely, due to prevalence of small fish larvae (total length <5 mm). The molecular marker COI-5P was amplified through Polymerase Chain Reaction (PCR) using the primers of Ward et al. (2005). Thermal cycling conditions consisted of an initial denaturation step of 2 min at 95 °C, followed by 35 cycles of denaturation (94 °C, 45 s), annealing (53 °C, 30 s), and extension (72 °C, 45 s), with a final extension at 72 °C for 8 min, before being held at 8 °C at the end of reaction. Electrophoresis of the PCR products was conducted in a 1.5% agarose gel to confirm successful amplification, and sequencing of the PCR products was performed by the commercial services of Macrogen Ltd. (South Korea) with the same primers used in PCR reaction.

Forward and reverse sequences were aligned using the DAMBE software (Xia & Xie, 2001), resulting in the full-length COI-5P fragment (~654 bp) for each specimen. The closest molecular taxonomic identification was then determined by comparing these sequences to the reference databases of NCBI (GenBank), using the nBLAST algorithm, and BOLD Systems, using the Identification Engine (hereafter called BOLD IE). For NCBI nBLAST, high similarity (>98%) between sequences with good coverage (overlap between sequences—>95%) are indicative that sequences belong to the same species (Hebert et al., 2003; Hebert, Ratnasingham & Dewaard, 2003). For the BOLD IE, identification can be assessed both based on sequence identity and through the placement of the submitted sequence within a local tree.

Additional checking of taxonomic identification of sequences was performed through phylogenetic analyses and K2P intra/inter specific genetic distances for larvae sequences with different results for species identification on the two databases tested (NCBI and BOLD) with close percentages of similarity between NCBI and BOLD or if low percentages of similarity were observed (<85%). Phylogenetic analyses and K2P intra/inter specific genetic distances were performed in MEGA X (Kumar et al., 2018).

Since most uncertainty concerning larvae identification was found in the Labrisomidae (see Results), we used the phylogeny of the Blenniiformes proposed by Lin & Hastings (2013) to further assess the identification of sequences of the family. In order to do that, maximum likelihood (ML) and maximum parsimony (MP) were performed, and bootstrap analyses (Felsenstein, 1985) were implemented to produce topologies and give statistical support to nodes, using 1.000 pseudoreplicates. A total of 65 reference sequences, representing 23 species of the ingroup and one outgroup (Opistognathus aurifrons, Opistognathidae), in addition to 39 sequences from our larvae dataset that could not be confidently identified by nBLAST and BOLD IE (total of 104 sequences), were included in both ML and MP analyses.

Taxonomic classification above the genus level follow Ghezelayagh et al. (2022) and Near & Thacker (2024).

Anatomical counts and observations

After monophyletic lineages were determined and likely species identified, high-resolution images of larval specimens were grouped based on the molecular identifications. Anatomical characters were then explored for genera containing more than five specimens identified, in order to increase the possibility of detecting anatomical variation which would not be detected if based on a fewer number of specimens. To assess possible diagnostic anatomical features, pigmentation patterns, including the number and placement of individual melanophores, as well as the shape and size of pigmented patches, frequently recognized among the most taxonomically useful characters for discriminating closely related species, were investigated (Moser et al., 1984; Strauss & Bond, 1990; Moser, 1996; Richards, 2005).

Molecular identification

Among the COI-5P sequences produced for the 89 specimens submitted for NCBI nBLAST and the BOLD IE analyses, more than 84% (75 specimens) showed identifications with similarities above 96%, providing identity of specimens with a high degree of confidence at least to the genus level. Among these, 68 specimens (76.4%) presented similarity above 98%, as expected for individuals belonging to the same species. A total of 20 species were identified representing six orders (Blenniiformes, Acanthuriformes, Syngnathiformes, Carangiformes, Clupeiformes and Gobiiformes). More than 80% of the larvae identified were members of the Blenniiformes (75 specimens), whereas the Gobiiformes, and Clupeiformes were each represented by a single specimen. The Blenniidae and Labrisomidae were the most frequently identified families, with most specimens identified in the genera Parablennius and Labrisomus. Detailed information on the percentage of similarity for each specimen in relation to available reference sequences and Genbank accession numbers for the new sequences are presented in Table 2.

Thirty-nine sequences had different species identifications from the two databases or had low values of similarity for species identification in both databases. These sequences were phylogenetically analyzed using two different methodologies (ML/MP). Both methods resulted in the same monophyletic groups with strong bootstrap support (>98%, except for P. fasciatus, 73% in MP). Each main clade only contained individuals from a single species, reinforcing that the clades where larval sequences are clustered represent reliable taxonomic identifications (Fig. 2).

The genera with most species and specimens identified were Labrisomus (L. nuchipinnis, L. cricota and L. conditus, with 21 specimens) and Parablennius (P. marmoreus and P. pilicornis, also with 21 specimens), which represent all species of those genera currently reported for the western South Atlantic (Menezes et al., 2003; Levy et al., 2013; Lin & Hastings, 2013; Fricke, Eschmeyer & Van Der Laan, 2023). Eighteen larval sequences have a high degree of similarity (96.7–98.3%) with P. marmoreus from the western North Atlantic (Florida and Caribbean) deposited in the databases used as references in our study. Three further sequences are closely related to P. pilicornis, again with a high degree of similarity (>98%).

In the case of Labrisomus, nBLAST and BOLD IE species identifications are conflicting (Table 2), but the phylogenetic analyses agree that our sequences represent L. conditus (11 sequences), L. cricota (eight sequences,), and L. nuchipinnis (two sequences), with a high support for all clades (bootstrap = 100%, except for L. cricota, with 99% in MP; Fig. 2). Intra and interspecific genetic distances (K2P) for Labrisomus species, support taxonomic identification, presenting a major gap between the averages observed (0.4% and 17.7%, respectively) (Tables S1 and S2). Other species of the Labrisomidae were identified, including Malacoctenus delalandii (three sequences), both according to nBLAST and BOLD IE (>99% similarity for both) and the phylogenetic analyses (100% bootstrap). An additional sequence was identified as Malacoctenus brunoi, again based on both nBLAST and BOLD IE (>99% similarity for both) and the phylogenetic analyses (100% bootstrap). Three further sequences were assigned to species of the Pomacentridae, with two species identified and with high values of similarity for both (>99%)

Identification of those sequences at the species level is therefore highly likely, but in other cases taxonomic precision was not achieved based on the methods employed and comparative sequences included in our study. Fourteen sequences identified as related to Paraclinus have a lower degree of similarity according to nBLAST and BOLD IE (84.43–84.82%). Our phylogenetic analyses also failed to recover Paraclinus as monophyletic, but the 14 new sequences form a distinct clade (100% bootstrap) that is included in a larger clade containing other species of Paraclinus (P. sini and P. asper) with relatively low support (bootstrap < 70%; Fig. 2). A great gap is also observed between averages of intra and interspecific genetic distances for Paraclinus species (1.9% and 21%, respectively) (Tables S1 and S2). However, intraspecific divergence of P. nigripinnis and P. fasciatus are higher (3.9% and 9.7%, respectively) than the ones observed for the remaining species of Paraclinus (0.8% average) (Table S1).

The Acanthuriformes was the order with the second highest number of specimens identified in our analyses. Seven specimens were identified as Eucinostomus argenteus (Gerreidae), with high values of similarity (98.6–100%) with comparative sequences. A single specimen belonging to the Sciaenidae was also identified, again with high values of similarity with sequences from both databases (>99%). Two specimens were identified as a single species of the Syngnathiformes, and two other specimens were identified as distinct species of the Carangiformes, with high values of similarity in at least one of the databases used for analysis (>98%). Only one specimen was assigned to each of the orders Clupeiformes and Gobiiformes, but with differing degrees of similarity (Table 2). In the Clupeiformes, our specimen can be reliably identified as Opisthonema oglinum (>98%), but in the case of the specimen identified as a member of the Gobiiformes a relatively lower level of similarity was recovered (>96%). Therefore, in this case identification was confident only at the genus level (Evorthodus), with only E. lyricus currently reported for the region (Table 2).

Morphology

No visible pigmentation pattern with taxonomic diagnostic applicability was found among the larvae of the two Parablennius species identified here (Fig. 3). The number and form of ventral spots do not reliably distinguish the two species (P. marmoreus = 20–29/P. pilicornis = 22–23). Parablennius pilicornis seems to have a lower number of ventral spots, but this variation might be due to the low number of specimens examined (3–Table 3).

Different species of Labrisomus, in turn, could be reliably identified based on different numbers and relative size of the ventral spots. Specimens of Labrisomus conditus have a lower total number of ventral spots (02–06), L. nuchipinnis has an intermediate number (06-07), and L. cricota has the highest number of spots (07-11) (Table 3). Pre-anal and post-anal spot counts in particular seem to be more effective in terms of specimen identification. In L. conditus, one to three post-anal spots were identified, whereas both L. nuchipinnis and L. cricota have five or more post-anal spots (Table 3). Labrisomus conditus also has a relatively large spot on the ventral mid-section of the body, contrasting with the mostly similarly sized spots of L. cricota and L. nuchipinnis at this region of the body (Fig. 4). Labrisomus cricota, in turn, has two or more pre-anal spots, whereas L. nuchipinnis has one pre-anal spot. Overall, considering spot size and shape, pigmentation pattern of L. nuchipinnis seems to be more similar to that present in L. cricota than in relation to L. conditus, where spots along the ventral part of the body are more uniform (Fig. 4).

A pattern of ventral spots was identified for the E. argenteus larvae. These large larvae (>5 mm) all present well delimited spots compared to the starred or blurred spots observed in the Blenniiformes described above (Fig. 5). The total number of ventral spots ranges from 08 to 11 and there is a constant number of pre-anal spots (03–Table 3).

Diversity found

Overall, molecular identification of fish larvae was successful, despite limited sampling year periods that limits our power of inference and discussion. The dominance of cryptobenthic reef species (mostly species of the Blenniiformes and Gobiidae) was expected because the collected sites are near islands surrounded by reef formations. Brandl et al. (2019) observed a similar pattern, with a median of two-third (65.7%) of fish larvae belonging to these taxa near reefs (<10 km) worldwide. Specifically in the western Atlantic, Brandl et al. (2019) recorded a smaller proportion of cryptobenthic species among fish larvae (<60%) in comparison with other parts of the world. However, their study was mostly focused on the Caribbean Sea and the Gulf of Mexico in the Florida Peninsula, which represent relatively enclosed portions of the ocean in terms of circulation of water masses when compared to the region of the western South Atlantic where our samples were collected. It is possible that this condition results in an aggregation of larvae of usually larger and pelagic fishes near islands of the Caribbean and the Gulf of Mexico, likely explaining the differences observed between the results of Brandl et al. (2019) and our study, where more than 80% of identified species belong to cryptobenthic reef fishes.

Recognizing the different species composition of the cryptobenthic fish community is of great concern since this group has been highlighted as having a central role in “Darwin’s paradox”. This paradox questions how coral reefs can maintain such high diversity and productivity in oligotrophic tropical oceans. Brandl et al. (2019) proposed that the cryptobenthic species promote internal reef fish biomass production, accounting for around two-thirds of reef fish larvae and producing around 60% of consumed reef fish biomass. Thus, precise knowledge about the composition of cryptobenthic fish species is fundamental for understanding the maintenance and conservation of reef ecosystems.

Distinguishing between closely related fish species can be challenging, sometimes even when dealing with adult specimens, and the task is even more complicated for ichthyoplankton, resulting in an early use of molecular techniques for identification of these specimens. Pegg et al. (2006) explored COI-5P as a marker for fish larval identification in the Great Barrier Reef in Australia and since then there has been an exponential growth of studies using molecular techniques to identify ichthyoplankton, from six publications between 2000 and 2010, to 75 in the next decade (Lira et al., 2022). Fish larvae from various other regions of the world have been identified by sequencing their COI-5P fragment. Hubert et al. (2010), for instance, focused on 22 species of the Acanthuridae and 16 of the Holocentridae from French Polynesia (Pacific Ocean), highlighting the differences between intra and interspecific genetic divergences, with 20- to 87-fold higher interspecific divergence between congeners compared to intraspecific divergence, also supporting the reliability of molecular taxonomic identifications for fish larvae.

Similarly, Azmir et al. (2020) analyzed fish larvae from Malaysia and found 41 species among the 153 larvae sequenced, also with a great difference between intraspecific average K2P distance (1%) and interspecific divergence (20%). Despite the “gap” observed in their study, five specimens were not identifiable with precision, with low values of identity and similarity for the two databases (GenBank and BOLD—<90%). In Hawaiian waters 92 fish larvae were identified using the COI-5P marker with a mean intraspecific divergence of 0.72% vs. a mean interspecific but congeneric divergence of 25.9%, but even with these genetic divergences, 52 specimens could only be assigned to the genus level because of the low values of identity observed when comparing query sequences with the reference database (Xing et al., 2022). The difference between intra and interspecific genetic distances observed here for Labrisomus species, support the phylogenetic position and taxonomic identification of larvae sequences of this genus. However, even with the great difference between intraspecific average K2P distance (1.9%) and interspecific divergence (21.9%), we were not able to reach a conclusion regarding Paraclinus larvae species in query. These cases of genus level identification reflect the importance of using the most complete reference database available and, if necessary, completing this reference database using identified adult specimens for the sampling region.

In Brazil, marine ichthyoplankton species surveys have been conducted in the north, northeast, and southeast to south regions of the country (Bonecker & Castro, 2018; Katsugawara et al., 2011; Bonecker et al., 2014; Rutkowski et al., 2011). However, none of these studies applied molecular techniques for species identification. Only two studies so far used molecular identification of Brazilian marine ichthyoplankton. Rodrigues et al. (2017) focused on billfish larvae and eggs (Istiophoridae and Xiphiidae) from the southeast region, and this is one of the few studies before ours that also provided photographs of the specimens identified with molecular tools. In their case, images were used to identify larvae of the target families based on prominent characters (four head spines, turned backwards) to select samples for the molecular analyses. However, several specimens, after molecular procedures using COI-5P as marker, were then later identified as Dactylopterus volitans (Dactylopteridae), exemplifying how difficult it is to properly identify fish larvae based on morphology exclusively (Rodrigues et al., 2017). In comparison, Costa et al. (2023) analyzed eggs from lower latitudes, with limited success in the identification at species level, a fact they believe is associated to the lack of reliable reference sequences for the fish fauna of the region.

Limitations of identification

Ko et al. (2013) tested how pronounced misidentification of ichthyoplankton could be, asking five different laboratories to identify fish larvae through morphology and then subsequently conducting molecular procedures. The average accuracy of the identifications, for the five laboratories, were ~75% for genus level and ~43% for species level. Families with commercially targeted species (Scombridae and Serranidae) were among the most misidentified. However, Mene maculata and Microcanthus strigatus were correctly identified by all the laboratories due to their distinctive morphology. Low taxonomic accuracy percentages in ichthyoplankton species identification (30% at genus level) have also been observed when comparing morphological and molecular approaches by Azmir et al. (2017). Mateos-Rivera et al. (2020), in turn, were able to identify almost all fish larvae in their study using anatomical characters, except for seven larvae of the Callionymidae that were identified through DNA barcoding.

Despite good validation of molecular techniques for species identification of ichthyoplankton in the last decade, all these studies stress that the reliability of this approach depends on representative databases with correct taxonomic identifications of the reference data (Ko et al., 2013; Azmir et al., 2017; Mateos-Rivera et al., 2020; Xing et al., 2022). Using two databases to identify species in the similarity analyses in the present study (NCBI and BOLD) helped to achieve reliable identification for almost all specimens (68 specimens presenting >98% of similarity). The four sequences of P. marmoreus with similarities between 97% and 98% were identified based on sequences from the North Atlantic, a situation that might help explaining the relatively lower levels of similarity detected (Table 2). Species molecular identification reliability can be associated with the number of sequences available, for each species, in databases consulted. In the same way, sequences from individuals collected along all the geographic distribution of each species, also improved trustfulness on species identification and the results found here for P. marmoreus are an example of what happen with a lack of a broad coverage of the species distribution and highlights the importance of a continuous improvement of databases.

Specimens with conflicting species identifications were revised using phylogenetic analyses. This helped to confirm the species of Labrisomus, where the different larval sequences could be assigned to species (clear monophyletic clades with high values of bootstrap support; Fig. 2). Larval sequences associated with the genus Paraclinus were not recovered in a monophyletic group with any species amongst the available reference sequences in our analyses, limiting the species level identification of the fourteen larvae sequences we produced within this genus. Incomplete reference sequences in databases are the likely reason for the lack of species definition for these fourteen larvae, since only one of the three species of Paraclinus recognized in the Brazilian coast (Guimarães & Bacellar, 2002) were included in the phylogenetic analyses (P. spectator). The remaining two (P. arcanus and P. rubicundus) were not found in the different databases explored here (no public COI-5P sequences from these species were found). However, the possibility of a cryptic fourth species for Brazilian waters cannot be excluded and the intraspecific divergence observed for P. fasciatus and P. nigripinnis, are evidence that the genus might need a taxonomic revision, where cryptic species can be uncovered and described.

Lack of monophyly for the genus Paraclinus might be due to the fast evolutionary rate of COI-5P and an ancient origin for the genus. Paraclinus has been proposed as one of the first lineages to diverge in the Labrisomidae based on a study that included nuclear molecular markers as well as COI-5P data (Lin & Hastings, 2013). Although COI-5P is probably the molecular marker with the broadest public species sequences libraries, in places with still unexplored species diversity, such as the western South Atlantic, more investment is needed for development of reliable reference sequences libraries that will, in turn, allow more robust molecular species identification analyses.

Morphological characters

Complementary to the development of more broad and inclusive reference sequences databases, the recognition of taxonomic diagnostic anatomical features in fish larvae is of great interest. Here, ventral pigmentation pattern was identified as useful to distinguish among Labrisomus species from the Brazilian coast in the western South Atlantic. For fish larvae, pigmentation patterns are regarded as useful characters for taxonomic identification of closely related species. They are often established during early development, which is interesting when dealing with initial ontogenetic stages, but caution should be taken as some patterns are also highly convergent, with different groups presenting similar patterns (Fuiman et al., 1983; Moser et al., 1984; Moser, 1996; Richards, 2005; Miller & Kendall, 2009).

Moser (1996) described the larvae of two different species of Labrisomus (L. xanti and L. multiporosus) from the Pacific coast of California. The ventral pigmentation patterns described for the Pacific species are similar to the ones observed here for L. cricota and L. conditus. Within these pairs of species, one has fewer ventral spots with one enlarged spot in the mid-section of the body (L. conditus and L. xanti) while the other has more regularly sized and shaped spots, homogeneously distributed along the ventral region of the body of pre-flexion larvae (L. cricota and L. multiporosus). Similarities in the pigmentation pattern observed between species from different regions were not in accordance to phylogenetic proximity, where actually species from the same geographic region (L. cricota + L. conditus vs. L. xanti + L. multiporosus) are more closely related among themselves (Fig. 2). This suggests that evolution of the pigment pattern has followed a similar, apparently convergent, process during the speciation of both the Atlantic and Pacific species pairs, but further phylogenetic analyses including more species should be conducted to evaluate how spot patterns relate to the systematics and biogeography in Labrisomus. In sum, species relationships within Labrisomus have not been properly addressed yet as phylogenetic analyses performed so far did not include a broad ingroup species coverage, including our own presented here.

Regarding the larvae identified as Eucinostomus argenteus, the ventral spot pattern observed here might be common to other species of the genus as well. Moser (1996), for instance, was unable to anatomically distinguish among the three species that occur off California (E. argenteus, E. currani and E. entomelas). The ventral pigmentation pattern described by Moser (1996) is similar to the pattern observed in this study, with 6 to 8 spots on the ventral midline, compared to the 5 to 8 post-anal spots described for E. argenteus here and the 6 to 8 ventral midline spots reported for E. argenteus and E. gula by Richards (2005). The three pre-anal spots observed here correspond to the anterior and posterior pigmented region on the gut, plus one melanophore on the cleithral symphysis described by Moser (1996) and Richards (2005).