Path dependence influences macroevolutionary predictability by constraining potential outcomes after stochastic evolutionary events. Although demonstrated in laboratory experiments, the basis of path dependence is difficult to demonstrate in natural systems because of a lack of independent replicates. Here we show two types of complex distributed visual systems, each recently evolved twice within chiton mollusks, demonstrating rapid and path-dependent evolution. The type of visual system a chiton lineage evolves is constrained by the number of openings for sensory nerves in its shell plates: lineages with more openings evolve visual systems with thousands of eyespots, whereas those with fewer evolve visual systems with hundreds of shell eyes. These macroevolutionary outcomes shaped by path dependence are both deterministic and stochastic because possibilities are restricted yet not entirely predictable.

1.1 Taxon sampling and sequencing

To generate sequencing data from specimens that were not preserved in ideal conditions and/or are more than five years post-collection (including a large library of previously sequenced chiton specimens from the personal collections of D. J. Eernisse as part of ongoing systematic revisions), we generated a suite of target capture loci. We designed target capture baits to predict exon regions following a similar approach to Hugall et al., 2016. We identified a set of 891 orthologs from 19 chiton transcriptomes and three reference molluscan genomes (Octopus bimaculoides, Lottia gigantea, and Crassostrea gigas) using the phylogenomics pipeline from Kocot et al., 2017. We identified conserved exons from the three mollusc genomes using EnsemblMetazoa and predicted intron-exon boundaries in the chiton sequences with a custom script using tBLASTx. From this, we compiled a set of predicted chiton exons from 355 genes that had at least four representative chiton sequences from our transcriptome dataset. Arbor Biosciences synthesized 19,980 biotinylated DNA baits that target these regions. The baits have a length of 120 bp and a GC content between 35-55%.

We used the MYbaits Custom DNA-Seq kit v. 3.02 (Arbor Biosciences, Ann Arbor, Michigan, USA) according to the manufacturer’s instructions. We applied our target capture baits to DNA extracted from chiton specimens preserved in ethanol. We extracted total DNA from 87 specimens of preserved chi-tons with a phenol-chloroform extraction. We then produced Illumina sequencing libraries with unique barcodes using the NEBNext® Ultra™ II FS Library Prep Kit (New England Biolabs, Ipswich, Massachusetts, USA) with a 25-minute incubation at the fragmentation step. We used TapeStation (Agilent, Santa Clara, California, USA) automated electrophoresis to confirm that the libraries had an average fragment size of 250 bp with adapters and that adapter dimers were absent. We incubated these libraries with biotinylated DNA baits via the MYbaits Custom DNA-Seq kit, and we pulled down the baits that hybridized with the Illumina library fragments with a streptavidin-coated magnetic bead binding process before washing away the uncaptured library fragments. We amplified the captured fragments with the KAPA HiFi Library Amp Kit (Roche Molecular Systems, Inc., Wilmington, Massachusetts, USA). After the amplification, we confirmed that the libraries maintained an average fragment size of 250 bp and that adapter dimers were absent. We pooled enriched sequencing libraries and sequenced them on either an Illumina MiSeq (11 species) or NextSeq (68 species).

We trimmed Illumina adapters and low-quality sequences (quality score of ¡20) using TrimGalore v0.4.1 (54). We used HybPiper (55) to map reads to reference sequences. To expand our taxon sampling further, we combined target capture data with chiton transcriptomes from recent phylogenies of chitons that included chitons with only aesthetes, chitons with eyespots, and chitons with shell eyes (56). We also included transcriptomes of several aplacophoran mollusks and a few conchiferan mollusks to root the phylogeny (Accession numbers in Supplementary Material, Outgroup and Available Sequences). To recover our target capture loci from chiton transcriptomes, we aligned the sequencing results from each target capture locus across chitons and determined the consensus DNA sequence for each. Then, we created a local blast database of each new chiton transcriptome and used each consensus DNA sequence from a target-capture locus as a query to locate the corresponding sequence within the new transcriptome. For each gene that we included, we retrieved only a single hit at E-value threshold 1e-4, so we are confident in our gene sequence determinations.

1.2 Phylogeny

To generate a robust phylogeny of chitons, we compared coalescence and concatenation-based approaches using Maximum Likelihood and Bayesian approaches. We first aligned all loci with MAFFT v7.305 (58) to produce a single data matrix of 103 chitons (see Table S1) that could be partitioned by locus. We produced a maximum likelihood tree in IQ-TREE2 (57), partitioning by gene and using ModelFinder to determine the best model for each locus. We also calculated gene concordance factors in IQ-TREE2. Separately, we produced a maximum-likelihood phylogeny for each gene individually with IQ-TREE2, again using ModelFinder, and input these trees into ASTRAL-Pro (41), which estimates a species tree from multiple gene trees via the multispecies coalescence (MSC) model and a quartet-based approach.

The placement of Schizochiton has been uncertain across studies of chitons, and our study is limited by the inclusion of only a single representative of the genus Schizochiton, S. incisus. Specimens of the only other accepted congeneric species, S. jousseaumei (Dupuis, 1917), were not available, and this is the only genus in the superfamily, Schizochitonoidea. Some morphological characters, most notably pectinated insertion plates and adanal gills, suggest an affinity of Schizochiton with Chitonidae. However, differences of opinion persist as to whether these traits are useful apomorphies. For example, the insertion plates of Schizochiton were once termed “obsoletely pectinated” (59), although they have strong pectination comparable to members of Chitonidae (DJE, pers. observation). Additionally, the caudal sinus is absent in Chitonidae but present in Schizochiton (65). However, caudal sinuses evolved multiple times across chitons, so the presence or absence of a sinus may not be a useful morphological character. Schizochiton has been recovered as sister to the remaining Chitonina, but with weak support (36). Our placement of Schizochiton requires that pectinated insertion plates evolved convergently in Schizochiton and Chitonidae. To further investigate the placement of Schizochiton in our phylogeny, we used an iterative pruning method to verify that Schizochiton was not an unstable branch (90) and determined that Schizochiton did not have any indicators of being a long branch or a highly unstable branch (Supplemental Data, RogueNaRok).

1.3 Ancestral state reconstruction and morphospace determination

To determine how many times distributed visual systems evolved across the chiton phylogeny, we performed ancestral state reconstruction using RayDISC and corHMM in R, which takes a tree file and a corresponding character matrix for a given trait and estimates transition rates and ancestral states for binary or multistate traits (Beaulieu and O’Meara 2014). We used the consensus tree file from our Mr-Bayes dated phylogeny run with 15 loci from sortadate as our input tree, and inferred ancestral states for shell eyes, eyespots, and the maximum number of insertion slits on the anterior shell plate. We gathered information on slit number from direct observations when possible, then from recent species revisions, and finally from from volumes of Kaas’s monographs (1985-2006). This was critical to accuracy, as some synonymized species in Kaas have dramatically different slit numbers (and often different visual systems). We took the highest recorded slit number for the anterior shell plate to account for the impact of animal size on slit number, where slit number can increase across growth. Whenever possible, we removed anterior shell plates from the specimens included on the tree and counted slits manually.