Salps are marine pelagic urochordates with a complex life cycle including a solitary and colonial stage composed of asexually-budded individuals. These colonies develop into species-specific architectures with distinct zooid orientations, including transversal, oblique, linear, helical, and bipinnate chains; as well as whorls, and clusters. The evolutionary history of salp colony architecture has remained obscured due to the lack of a homology-based ontology to characterize architectures, as well as a lack of phylogenetic taxon sampling and resolution of critical nodes. We (1) collected and first-time sequenced eight species of salps, (2) inferred the phylogenetic relationships among salps, and (3) reconstructed the evolutionary history of salp colony architecture. We collected salp specimens via offshore SCUBA diving, dissected tissue samples, extracted their DNA, amplified their 18S gene, and sequenced them using Sanger technology. We inferred a new molecular phylogeny using both Maximum Likelihood and Bayesian approaches. Using this phylogeny, we reconstructed the ancestral states of colony architecture using a Bayesian ordered Markov model informed by the presence and absence of specific developmental mechanisms that lead to each architecture. We find that the ancestral salp architecture is either oblique or linear, with every other state being derived. Moreover, linear chains have evolved independently at least three times. While transversal chains are developmentally basal and hypothesized to be ancestral, our phylogenetic topology and reconstructions strongly indicate that they are evolutionarily derived through the loss of zooid torsion. These traits are likely critical to multijet locomotory performance and evolving under natural selection. Our work showcases the need to study the broader diversity of salp species in order to gain a comprehensive understanding of their organismal biology, evolutionary history, and ecological roles in pelagic ecosystems.

To build a well-resolved molecular phylogeny, we primarily used the 18S gene accession list for salps and outgroups from Govindarajan et al. 2011 with a few modifications. We expanded taxon sampling by collecting tissue samples from understudied salp species (Metcalfina hexagona, Ihlea punctata, Helicosalpa virgula, Helicosalpa younti, Cyclosalpa bakeri, Cyclosalpa pinnata, and Ritteriella amboinensis) using tissue samples from specimens we collected while bluewater SCUBA diving (Haddock & Heine, 2005) from a small vessel off the coast of Kailua-Kona (Hawai’i Big Island, 19°42'38.7" N 156°06'15.8" W), over 2000m of offshore water. When possible, we sampled a variety of tissues from the zooid excluding the gut to avoid contamination from food particles, as well as the tunic to avoid clogging the DNA extraction columns. These samples were preserved in ethanol at room temperature until the point of DNA extraction in the lab. We aligned these sequences using MUSCLE 5.1 (Edgar 2004) with default settings. As a sensitivity analysis, we alternatively aligned them with MAFFT 7.419 (Katoh et al. 2009) with default settings. In addition, we experimented with post-processing these alignments with GBLOCKS 0.91b (Castresana 2000) with default settings except for allowing half-gap positions (as used in Govindarajan et al. 2011). The alignment contained every sequence except for Cyclosalpa bakeri specimen D27-Cbak-B-1, which appeared truncated and non-comparable after post-processing. GBLOCKS retained 43% of sites when aligning with MUSCLE and 45% of sites when aligning with MAFFT. To make an ML inference from these alignments, we used IQTree 1.6.12 (Nguyen et al. 2015) with 1000 bootstrap replicates (SM Fig. 1). Node support was reported using bootstrap support (BS). The consensus trees obtained using the model selected by the best Bayesian Information Criterion (TIM3e+R5: transition model with equal base frequency, with 5 rate categories) and using GTR+I+Gamma are congruent, regardless of whether MAFFT or MUSCLE was used for alignment. However, the consensus trees with GBLOCKS were not congruent with these trees by several nodes which had low support, due to many trimmed sequences appearing identical. We suspect that GBLOCKS is removing critical phylogenetic signal from the data, and therefore decided not to use it for downstream analyses. We collected between one and five specimens of adult blastozooid colonies from each target species via bluewater SCUBA diving. We took photographs of these colonies using a Nikon DSLR camera with a 75mm lens facing downwards on a tripod with the colonies fully submerged in glass dishes and a ruler for scale. We anesthetized the salp specimens using 0.2% MS-222 prior to photographing them in order to avoid swimming motion in the dishes. We coded the colony architecture for each species we encountered in the field based on our photographs and observations. In addition, we complemented these observations with published records such as Madin (1990) for species we did not encounter, such as Pegea confoederata, Pegea bicaudata, Thalia democratica, Thalia orientalis, Thetys vagina, Cyclosalpa floridiana, Salpa younti, and Salpa thompsoni. In addition to categorizing each architecture as transversal, helical, whorl, cluster, oblique, linear, or bipinnate; we also measured the dorsoventral zooid-stolon (zooid oral-aboral axis to stolon axis) angle. To do so, we photographed salp colonies from three homologous orthogonal planes of observation defined in Damian-Serrano & Sutherland (2023) as the oral-aboral-normal (normal sensu perpendicular) plane, the dorsoventral-normal plane, and the stolon-normal plane. Using the photographs taken from the dorsoventral-normal plane, we measured the zooid-stolon angle in ImageJ using the aligned endostyle and gill bar as a proxy for the zooid oral-aboral angle, and the line connecting the opaque guts of serially neighboring zooids as a proxy for the stolon angle. We measured at least three zooids per colony and between one and three individual colonies per species. 

We used the Bayesian time tree to reconstruct the ancestral states of colonial architectures as a categorical character. We ran the ancestral reconstruction using a Bayesian ordered Markov model (OMM) that constrains the rate matrix to allow only transitions between states that are adjacent in the developmental ontology. To do this, we hard-coded the transition rates between non-adjacent states (e.g. between helical and linear architectures) to be zero, thus requiring states changes across developmental pathways to transition back to a transversal architecture (representing the loss of specific developmental mechanisms) and then shift towards a different pathway following the required order of underlying mechanism gains and losses. This model estimated twelve rate parameters allowing for asymmetrical rates of gain and loss for each transition between architecture states. Alternatively, we repeated this analysis estimating a single rate for all transitions, while still constraining non-adjacent transitions (SM Figure 4). We used RevBayes for this analysis, adapting the categorical Markov model ancestral state reconstruction protocol described in the “morph_ase” tutorial on the RevBayes website (scripts available in the Dryad repository). Alternatively, we reconstructed the ancestral states using stochastic mapping with simpler “equal rates” (single rate parameter for all state transitions with 100 simulations, SM Figure 5) and “all rates different” (42 independent rate parameters with 25 simulations, one for each rate transition in each direction, SM Figure 6) models in the R package phytools (Revell 2012). While a myriad of alternative models could be estimated and compared, this is beyond the scope of this study.

While colonial architecture as a whole can be conceptualized as a categorical trait, some architectures differ from each other across a gradient in a continuous trait that drives their structural differences. The dorsoventral zooid-stolon angle is one such trait, which drives the streamlining of salp chains. This continuous trait ranges from one extreme with zooids arranged perpendicularly (90°) to the stem of the colony (transversal, helical, whorl, and cluster architectures), all the way to linear chains with zooids arranged in parallel to the stem (as in the linear chains of Soestia zonaria), with a gradation of more or less oblique intermediate forms. We reconstructed the evolutionary history of the dorsoventral zooid-stolon angle of salp colonies on the Bayesian time tree using tip values measured from our photographs, and a Brownian Motion model in the R package phytools (Revell 2012). We used an ML ancestral state reconstruction with 95% confidence intervals in the R package ape 5.6 (Paradis & Schliep 2019).

We matched the sequences that form the tips of the molecular phylogeny and the species of the specimens from which we took the morphological and kinematic data. In the case of Pegea, the species we observe off Hawaii has intermediate traits between P. confoederata (blastozooid morphology) and P. socia (oozoid morphology), possibly representing either a new species, a hybrid, or phenotypic variation within either species. Since we are confident that this is a Pegea species and that the genus Pegea is likely monophyletic, the branch length for this specimen should be congruent to that of any other Pegea species as long as it is the only Pegea species in the tree. Thus, we mapped the morphological and speed data to the species tip of Pegea confoederata.

R, RStudio, RevBayes, IQTree, GBLOCKS, FigTree