Nanopore Amplicon Sequencing Reveals Molecular Convergence and Local Adaptation of Rhodopsin in Great Lakes Salmonids

Abstract

Local adaptation can drive diversification of closely related species across environmental gradients and promote convergence of distantly related taxa that experience similar conditions. We examined a potential case of adaptation to novel visual environments in a species flock (Great Lakes salmonids, genus Coregonus) using a new amplicon genotyping protocol on the Oxford Nanopore Flongle and MinION. We sequenced five visual opsin genes for individuals of Coregonus artedi, Coregonus hoyi, Coregonus kiyi, and Coregonus zenithicus. Comparisons revealed species-specific differences in a key spectral tuning amino acid in rhodopsin (Tyr261Phe substitution), suggesting local adaptation of C. kiyi to the blue-shifted depths of Lake Superior. Ancestral state reconstruction demonstrates that parallel evolution and “toggling” at this amino acid residue has occurred several times across the fish tree of life, resulting in identical changes to the visual systems of distantly related taxa across replicated environmental gradients. Our results suggest that ecological differences and local adaptation to distinct visual environments are strong drivers of both evolutionary parallelism and diversification.

Significance

Previous research has shown parallel evolution of vision genes across the fish tree of life. Two key questions are what role do they play in ecological diversification across light availability gradients and whether these changes can be reversible. This study demonstrates one vision gene in particular, rhodopsin, is involved in local adaptation across a depth gradient in Great Lakes salmonids. Additionally, we provide evidence that several species of salmonids have undergone parallel reversals to an ancestral rhodopsin allele that was last seen in this lineage over 175 Ma, showcasing an extreme example of evolutionary reversal over deep time.

Introduction

Local adaptation to novel environments can drive genetic and phenotypic differentiation among closely related organisms. Diversification may occur as populations become locally adapted to distinct conditions, leading to the evolution of divergent traits that are beneficial in each lineage’s preferred environment. Conversely, a trait may be sufficiently advantageous in a particular environment that multiple distantly related taxa converge upon it, sometimes due to the same mutation or amino acid substitution occurring independently (i.e., parallel evolution; Zhang and Kumar 1997; Futuyma and Kirkpatrick 2017). For example, parallel substitutions have occurred in myoglobin in pinnipeds and cetaceans (Romero-Herrera et al. 1978), lysozyme in ruminants and colobine monkeys (Stewart et al. 1987), and rhodopsin in fishes colonizing brackish or freshwater ecosystems (Hill et al. 2019). In this study, we examined a specific case of local adaptation in the teleost visual system that has led to diversification among similar taxa and parallel evolution among distantly related fishes.

Due to their importance in ecological interactions and their dynamic evolutionary history, the evolution of visual pigment genes (i.e., opsins) in marine and freshwater fishes has received considerable attention (Lin et al. 2017; reviewed in Rennison et al. 2012; Carleton et al. 2020). The vertebrate visual opsin system is divided into five subgroups—one rod opsin (rhodopsin) responsible for vision under low light conditions and four cone opsins (long-wave sensitive, short-wave sensitive 1, short-wave sensitive 2, and rhodopsin 2) responsible for color vision. These subgroups are based in part on differences in their peak absorbance spectra, with each group resulting from a series of gene duplications and subsequent amino acid substitutions, leading to small but measurable functional differentiation among the five opsin types (Okano et al. 1992; Yokoyama 2000). Opsin genes can shape the evolution of vision via several mechanisms. First, single nucleotide polymorphisms (SNPs) can cause nonsynonymous substitutions in key spectral tuning residues, in some cases driving adaptation to different light environments (Terai et al. 2002; Marques et al. 2017). Alternatively, copy number variants (CNVs) may result from tandem duplications and/or whole genome duplications and undergo neofunctionalization to maximally absorb at a new wavelength (e.g., expansions of rod opsins in deep-sea fishes [Musilova et al. 2019] and cone opsins in shallow-water fishes [Weadick and Chang 2007]), although gene conversion may sometimes work to homogenize newly formed paralogs, inhibiting this process (Carleton et al. 2020). Vision can also be modified by variation in opsin expression patterns based on both ontogenetic and environmental cues (Shand et al. 2008; Hofmann et al. 2009).

The cisco species flock (genus Coregonus) of the Laurentian Great Lakes presents a well-suited opportunity to study local adaptation of visual opsins to novel light environments based on depth differences among species (Harrington et al. 2015). The four extant cisco species in Lake Superior show generally low levels of interspecific variation across the genome (Turgeon and Bernatchez 2003; Turgeon et al. 2016; Ackiss et al. 2020) despite considerable differences in depth preferences (Rosinski et al. 2020). Coregonus artedi is typically epilimnetic (10–80 m), whereas Coregonus hoyi and Coregonus zenithicus are both found at intermediate depths (40–160 m), and Coregonus kiyi can be found at depths of 80–>200 m (Rosinski et al. 2020). Based on the observed ecological differentiation, we hypothesize that divergent selection may have acted to fine-tune opsins for maximum absorption of wavelengths of light that penetrate to each species’ preferred depth. Here we assess the evolution of five visual opsin genes in the Coregonus species flock to better understand mechanisms underlying their divergence across a depth gradient.

Materials and Methods

Oxford Nanopore sequencing is contributing to a rapidly expanding toolkit for DNA sequencing, owing to low up-front costs, enhanced ability to detect DNA or RNA base modifications, and read lengths limited only by the quality and quantity of input nucleic acids. Nanopore sequencing also allows for straightforward haplotyping, as whole molecules can be sequenced for each amplicon with no need for assembly. This approach has been successfully applied to microbial metabarcoding and pathogen identification (Shin et al. 2016; Moon et al. 2018; Rames and Macdonald 2018), as well as human genotyping (Cornelis et al. 2017, 2019). As flow cell quality and base-calling algorithms have improved, the accuracy and functionality of nanopore amplicon sequencing have rapidly improved. However, its application to SNP genotyping in nonhuman eukaryotes with large and complex genomes remains relatively unexplored. It is particularly relevant to understand whether accurate genotypes can be obtained, and, if so, what coverage depth is needed to do so.

In the present study, we sequenced amplicons of five teleost opsin genes in a total of 74 samples on the Oxford Nanopore Flongle device. In combination with the PCR Barcoding Expansion 1–12 (Oxford Nanopore Technologies), we sequenced and genotyped 12 individuals simultaneously on a single Flongle flow cell, following the pipeline shown in figure 1 (for a complete protocol, see Supplemental Protocol). This study is one of the first to demonstrate the accuracy and utility of amplicon sequencing with the Oxford Nanopore Flongle for SNP genotyping of eukaryotic samples.

Fig. 1.

Summary of steps for amplicon sequencing and bioinformatic analyses. Boxes on the left represent individual steps, color-coded based on their phase: red represents sample preparation, yellow represents nanopore sequencing, green represents sample demultiplexing, blue represents read mapping, and purple represents genotyping and analysis. Larger boxes to the right show additional information for each of the steps: the simplified mechanisms by which amplicons are generated and barcoded (top); frequency histogram with read length on the x-axis and number of reads in the y-axis (middle); and how reads are mapped to the reference genome (bottom).

A preliminary assembly of the de novo transcriptome of C. artedi (NCBI BioProject #PRJNA659559; Bernal et al. 2020) was used as a reference to extract gene sequences of long-wave sensitive (LWS), short-wave sensitive 1 (SWS1), short-wave sensitive 2 (SWS2), rhodopsin (RH1), and rhodopsin 2 (RH2), representing one gene from each teleost visual opsin subfamily. For each of the five genes of interest, a fragment ∼700–2,100 bp in length was amplified for 18 samples of C. artedi, 19 C. hoyi, 21 C. kiyi, and 16 C. zenithicus (supplementary tables S1–S3, Supplementary Material online). All amplicons from a single individual were assigned a specific barcode and were pooled into a library containing amplicons from 12 samples, which were sequenced simultaneously on a single Flongle flow cell (fig. 1). This process was then repeated until amplicons from all samples were sequenced. Raw reads were deposited in SRA under NCBI BioProject #PRJNA664981. Sample-specific barcodes were detected and trimmed using Guppy v3.2.4 (Oxford Nanopore Technologies) with the command guppy_barcoder, and reads from each sample were mapped with BWA v0.7.17 using the command bwa mem, with version one of the Coregonus sp. “balchen” genome assembly as a reference (De-Kayne et al. 2020; GCA_902810595.1). To verify the accuracy of nanopore amplicon genotyping, we performed a rarefaction analysis in which SNPs were called at various levels of coverage (i.e., maximum possible coverage; 2,000; 1,000; 500; 250; 100; 75; 50; and 25×). An annotated bash script detailing the entire bioinformatic pipeline is available from GitHub (https://github.com/KrabbenhoftLab/rhodopsin; last accessed December 3, 2020).

Results and Discussion

On average, Flongle sequencing runs yielded 206.13 Mb (SD ± 166.64 Mb; min–max: 26.84–471.50 Mb), with an average of 184,958 reads (±154,877 reads; 23,468–435,138 reads), though yield varied based on flow cell quality (flow cells used were earlier versions with low number of active pores). The average sequence N50 was 1,117 bp (±305 bp; 897–1,852 bp), with read length abundances peaking at the approximate lengths of our amplicons (fig. 1). After resequencing genes with low coverage following first-round sequencing, the average coverage was 3,199.58× across all five genes (±4,804.24×, 10.47–31,158.31×; supplementary table S1, Supplementary Material online). Coverage varied slightly by species, but this is likely an artifact of stochastic differences in PCR efficiency and sequencing yield (supplementary table S4, Supplementary Material online). Amplicon reads mapped uniquely (i.e., one genomic region per amplicon) to the C. sp. “balchen” genome, confirming that each amplicon does indeed correspond to a sequence of a single gene. However, based on a BLAST search of our opsins against the C. sp. “balchen” genome, multiple functional copies of LWS, RH2, and SWS1 exist in the Coregonus sp. genome, providing another potential avenue for opsin evolution (supplementary table S5, Supplementary Material online). Additionally, it was determined that the copy of LWS analyzed here is partially pseudogenized and perhaps nonfunctional.

The SNP calls from nanopore data were then compared with Sanger sequences of RH1 for the same individuals. Although accuracy remained high at all sequencing depths (>90%), we found incongruencies in a small proportion of samples between 10× and 75×. Only when reaching 100× coverage were genotypes called with complete accuracy for all individuals, in relation to Sanger sequences. Considering that small errors can impact the results of analyses involving amplicons with few variant sites, we recommend a minimum per-amplicon coverage of 100× for future work.

We used a conservative genotyping approach, with the goal of assessing the coverage needed for accurate genotyping on a Flongle flow cell. Based on our findings, this approach could be used for higher throughput sequencing, which could involve more amplicons, more individuals, or a combination of both. Considering that we generated ∼200 Mb of sequence data per run, the number of individuals and amplicons that can be sequenced simultaneously at 100× coverage can be estimated as follows:

200,000,000 bp=100*A*N*NA,

where A is the amplicon size (in bp), N is the number of samples to be sequenced simultaneously, and N_A is the number of amplicons to be sequenced per sample. To optimize throughput for the maximum number of samples, the PCR Barcoding Expansion 1–96 (EXP-PBC096, Oxford Nanopore Technologies) can be used to generate sequence data for 96 samples simultaneously. Assuming an average amplicon size of 1,000 bp, one could sequence 20 amplicons across 96 samples in a single 24-h Flongle sequencing run for ∼$300, or $0.16 per genotype (supplementary table S6, Supplementary Material online). The use of a MinION flow cell (not analyzed here) would increase output by a factor of ∼16× (based on differences in number of total pores) and further reduce the cost per genotype. With the growth of nanopore sequencing, these conservative cost estimates are expected to drop in upcoming years.

The average fixation index (F_ST) of the detected SNPs in the five opsins analyzed across all four species was 0.11 (supplementary table S7, Supplementary Material online). The only large differences (F_ST > 0.4) were found in four SNPs detected within the coding sequence of rhodopsin, with no highly differentiated SNPs among the four cone opsins. This suggests that differences in dim-light vision and changes in rhodopsin could be driving local adaptation by depth. Of the four high F_ST SNPs, one (F_ST = 0.44) was a synonymous SNP at amino acid residue 78. One SNP (F_ST = 0.44) resulted in a shift from asparagine to histidine at amino acid residue 100, which is located near the C-terminal end of transmembrane helix two, possibly in the extracellular matrix (fig. 2a and b; see also Yokoyama 2000). Another (F_ST = 0.44) resulted in a change from valine to isoleucine at residue 255, which is located in transmembrane helix six, facing away from the retinal binding pocket (fig. 2a and b, see also Baldwin 1993; Hunt et al. 1996). Neither residue 100 nor 255 are known to be key spectral tuning sites in rhodopsin (Yokoyama 2000), although Yokoyama and Jia (2020) recently reported that Ile255Val may have an effect on spectral tuning of RH2. The three aforementioned SNPs possess the exact same F_ST and changes in genotype were completely consistent across all samples, suggesting that these sites are tightly linked.

Fig. 2.

(a) 2D model of Coregonus artedi rhodopsin based on a 3D model generated using PHYRE v2.0 (Kelley et al. 2015) and visualized in UCSF Chimera (Pettersen et al. 2004). Amino acid residues are numbered in order from N-terminus to C-terminus. Amino acid residue 78 (the location of a synonymous, high F_ST SNP) is colored in light blue, residues 100 and 255 (locations of nonsynonymous, high F_ST SNPs) are colored in dark blue, and residue 261 (location of a high F_ST SNP known to be involved in spectral tuning) is colored in red. The 2D model was constructed for the specific model obtained for C. artedi, following Yokoyama (2000) and Musilova et al. (2019). (b) 3D model of bovine rhodopsin, colored blue (N-terminus) to red (C-terminus). Amino acid residues 78, 100, 255, and 261 have been colored in black, and are labeled accordingly.

The most strongly segregating SNP (F_ST = 0.88) occurred at amino acid residue 261 of rhodopsin, which is located in transmembrane helix six, facing the retinal binding pocket of the protein (fig. 2a and b, see also Baldwin 1993; Hunt et al. 1996; Yokoyama 2000). Coregonus artedi, C. hoyi, and C. zenithicus, inhabitants of a relatively shallow, broad-spectrum light environment, were primarily homozygous for tyrosine at this locus (fig. 3). Meanwhile, C. kiyi, which inhabits the blue-shifted deeper waters of Lake Superior, was completely homozygous for phenylalanine. The presence of phenylalanine at this site is known to cause an 8 nm blue-shift in the absorbance spectrum, likely due to the change in polarity associated with this substitution (i.e., polar to nonpolar Y261F) (Yokoyama et al. 1995). Genotypic associations at this locus vary consistently with depth (fig. 3), providing evidence that C. kiyi is adapted to life in deep water after evolving from shallow-water ancestors. This hypothesis is further corroborated by phenotypic data, as C. kiyi have significantly larger eye diameters (as a proportion of total head length) than C. artedi (P < 0.001), C. zenithicus (P < 0.001), and C. hoyi (P < 0.001; supplementary fig. S1, Supplementary Material online). The predictable variation of both genetic and morphological traits provides key evidence that local adaptation to depth accompanies diversification of Great Lakes ciscoes (supplementary fig. S2, Supplementary Material online).

Fig. 3.

(a) The four cisco species included in this study: Coregonus zenithicus (n = 16), Coregonus artedi (n = 18), Coregonus hoyi (n = 19), and Coregonus kiyi (n = 21). For each species, the approximate λ_max of the most common allele for rhodopsin is shown. Pie charts below each photo indicate the allele frequency at residue 261 of rhodopsin, where red represents tyrosine and blue represents phenylalanine. (b) Boxplots indicating the approximate capture depths of samples from each of the four species. (c) An approximation of the visible light spectrum in Lake Superior, from 400 to 700 nm wavelength. The narrowing of the spectrum with increased depth shows how the ability of organisms to perceive certain wavelengths of light diminishes with increasing depth, particularly with red and orange light (following Harrington et al. 2015).

The shift between the two aforementioned amino acids at rhodopsin residue 261 was recently analyzed in a deep phylogenetic context. Hill et al. (2019) showed that fish lineages that have undergone a habitat change from blue-shifted marine waters to red-shifted brackish or freshwater have independently converged on the red-shift-associated 261Tyr phenotype over 20 times across the fish tree of life. We performed an ancestral state reconstruction at this amino acid residue to further examine these evolutionary changes. Our results show that parallel reversals and “toggling” (i.e., switching back and forth several times between two alleles in a single lineage; sensu Delport et al. 2008; Porubsky et al. 2020) are common occurrences at this site across the fish tree of life (fig. 4), consistent with the findings of Musilova et al. (2019). Indeed, it appears that C. kiyi has undergone a reversal to the blue-shifted 261Phe after ∼175 Myr of differentiation from its marine ancestors (fig. 4). Whether toggling is due to true re-evolution of an identical mutation or through re-emergence of a low-frequency allele following incomplete lineage sorting is unknown. Interestingly, this reversal to a deep ancestral state over millions of years has also occurred in several other lineages across the fish tree of life, including in the channel catfish (Ictalurus punctatus), rainbow trout (Oncorhynchus mykiss), Kessler’s sculpin (Leocottus kesslerii), and European flounder (Platichthys flesus), among others, and can occur in both directions (i.e., Phe261Tyr or Tyr261Phe; fig. 4). The Salmoniformes and Esociformes in particular appear to have undergone a high degree of parallel reversal at this site, with multiple taxa (Oncorhynchus spp., Brachymystax lenok, C. kiyi, and Dallia pectoralis) undergoing independent parallel reversals to the deep ancestral blue-shifted allele (fig. 4). These findings indicate that rhodopsin residue 261 may be able to toggle between these two amino acids depending on what is advantageous in a particular light environment, even across incredibly long time scales.

Fig. 4.

The time-calibrated fish tree of life (Rabosky et al. 2018) colorized according to the allele possessed by each species at rhodopsin residue 261, with red indicating the red-shifted tyrosine and blue indicating the blue-shifted phenylalanine (following Hill et al. 2019). Ancestral states were reconstructed using RAxML (Stamatakis 2014) and plotted using the R package ggtree (Yu et al. 2017). Arrows indicate divergence times of clades of interest, in millions of years (Chang et al. 2019). The enhanced clade showcases the Salmoniformes (including Coregonus) and Esociformes; this group shows several parallel reversals to the phylogenetically deep “blue” ancestral state (last seen in the ancestor to this group ∼175 Ma).

Conclusions

Our results indicate that local adaptation to distinct visual environments is associated with genetic and morphological differentiation among the closely related ciscoes of Lake Superior. The identification of several high F_ST SNPs in rhodopsin, including Phe261Tyr, is particularly relevant, as the shifts between these two amino acids at residue 261 are identical to those observed across similar light availability gradients in phylogenetically distant fishes (Hill et al. 2019). Adaptation to similar light conditions therefore seems to lead to phenotypic convergence due to parallel single-nucleotide changes at this site. Additionally, the discovery of independent reversals to the ancient ancestral state in C. kiyi (and other lineages), provides evidence of genetic toggling, whereby organisms transition bidirectionally between different alleles in response to environmental pressures. This result is striking because the genetic background is presumably very different across these taxa after more than ∼175 Myr of divergence. In this case, the conserved nature of rhodopsin protein function likely potentiates evolutionary reversals over deep time periods. The observation of amino acid toggling in Great Lakes Coregonus species and across the fish tree of life is at the extreme end of the time scale of parallel evolution and evolutionary reversals, given the generally held prediction that the likelihood of evolutionary reversal diminishes as time progresses (Storz 2016; Blount et al. 2018).

Long-read amplicon sequencing using the Oxford Nanopore Flongle is highly amenable to genotyping complex eukaryotes. The methodology described here is simple and reliable, and offers the promise of rapid, low-cost genotyping in nonmodel organisms. This protocol has many potential applications—nanopore amplicon genotyping could be used for CRISPR validation, screening for inherited genetic disorders, and eukaryotic eDNA metabarcoding, among others.

Supplementary Material

Supplementary data are available at Genome Biology and Evolution online.