1. Evolutionary Biology
Download icon

Multifactorial processes underlie parallel opsin loss in neotropical bats

  1. Alexa Sadier
  2. Kalina TJ Davies
  3. Laurel R Yohe
  4. Kun Yun
  5. Paul Donat
  6. Brandon P Hedrick
  7. Elizabeth R Dumont
  8. Liliana M Dávalos  Is a corresponding author
  9. Stephen J Rossiter  Is a corresponding author
  10. Karen E Sears  Is a corresponding author
  1. University of California, United States
  2. Queen Mary University of London, United Kingdom
  3. Stony Brook University, United States
  4. Yale University, United States
  5. University of Illinois, United States
  6. Harvard University, United States
Research Article
  • Cited 4
  • Views 1,258
  • Annotations
Cite this article as: eLife 2018;7:e37412 doi: 10.7554/eLife.37412

Abstract

The loss of previously adaptive traits is typically linked to relaxation in selection, yet the molecular steps leading to such repeated losses are rarely known. Molecular studies of loss have tended to focus on gene sequences alone, but overlooking other aspects of protein expression might underestimate phenotypic diversity. Insights based almost solely on opsin gene evolution, for instance, have made mammalian color vision a textbook example of phenotypic loss. We address this gap by investigating retention and loss of opsin genes, transcripts, and proteins across ecologically diverse noctilionoid bats. We find multiple, independent losses of short-wave-sensitive opsins. Mismatches between putatively functional DNA sequences, mRNA transcripts, and proteins implicate transcriptional and post-transcriptional processes in the ongoing loss of S-opsins in some noctilionoid bats. Our results provide a snapshot of evolution in progress during phenotypic trait loss, and suggest vertebrate visual phenotypes cannot always be predicted from genotypes alone.

https://doi.org/10.7554/eLife.37412.001

eLife digest

Bats are famous for using their hearing to explore their environments, yet fewer people are aware that these flying mammals have both good night and daylight vision. Some bats can even see in color thanks to two light-sensitive proteins at the back of their eyes: S-opsin which detects blue and ultraviolet light and L-opsin which detects green and red light. Many species of bat, however, are missing one of these proteins and cannot distinguish any colors; in other words, they are completely color-blind.

Some bat species found in Central and South America have independently lost their ability to see blue-ultraviolet light and have thus also lost their color vision. These bats have diverse diets – ranging from insects to fruits and even blood – and being able to distinguish color may offer an advantage in many of their activities, including hunting or foraging. The vision genes in these bats, therefore, give scientists an opportunity to explore how a seemingly important trait can be lost at the molecular level.

Sadier, Davies et al. now report that S-opsin has been lost more than a dozen times during the evolutionary history of these Central and South American bats. The analysis used samples from 55 species, including animals caught from the wild and specimens from museums. As with other proteins, the instructions encoded in the gene sequence for S opsin need to be copied into a molecule of RNA before they can be translated into protein. As expected, S-opsin was lost several times because of changes in the gene sequence that disrupted the formation of the protein. However, at several points in these bats’ evolutionary history, additional changes have taken place that affected the production of the RNA or the protein, without an obvious change to the gene itself. This finding suggests that other studies that rely purely on DNA to understand evolution may underestimate how often traits may be lost. By capturing ‘evolution in action’, these results also provide a more complete picture of the molecular targets of evolution in a diverse set of bats.

https://doi.org/10.7554/eLife.37412.002

Introduction

The reduction and eventual loss of previously adaptive traits can be seen across the Tree of Life, and is typically linked to relaxation in selection. Within vertebrates, examples of losses include flight in birds, armor plates in sticklebacks, and the ability to synthesize vitamin C in bats (Burga et al., 2017; Cui et al., 2011; Le Rouzic et al., 2011). Strikingly, many instances of trait loss occur in parallel across multiple independent lineages (e.g. Colosimo et al., 2004, Drouin et al., 2011, and Harshman et al., 2008). There have been attempts to relate parallel trait losses to shared ecological conditions such as salinity tolerance or switches in diet, but the precise causal links are not always clear (Marchinko and Schluter, 2007). In contrast, the genetic bases of parallel trait loss are often known, with pseudogenization – whereby a non-essential gene loses some functionality – being a frequently invoked mechanism (e.g. Cui et al., 2011 and Protas et al., 2006).

One of the best known examples of parallel phenotypic loss via pseudogenization, which can often be directly related to shifts in ecology, is that of color vision in vertebrates. Opsins encode the photoreceptor proteins of rod cells that are responsible for dim-light and cone cells responsible for color vision. Most mammals possess three visual opsins: rhodopsin (RHO) in rods, and opsin one long-wave sensitive (OPN1LW) found in L-cones, and opsin one short-wave sensitive (OPN1SW) found in S-cones. Reconstructions of the highly complex evolutionary history of mammalian vision suggest that there have been >20 independent losses of cone-opsins, with associated reduction in color sensitivity (e.g. Bowmaker, 1998, Emerling et al., 2015, Lucas et al., 2003, Porter et al., 2012 and Yokoyama et al., 2008). This is exemplified in some cetacean and xenarthran lineages, which appear to have lost both of their cone-opsins (Emerling and Springer, 2015; Meredith et al., 2013).

Evolutionary reconstructions of color vision have nearly all been based solely on opsin gene sequences, with gene expression and protein data limited or missing for most mammalian species, including cetaceans and primates (but see Kraus et al., 2014, Peichl et al., 2017, Schweikert et al., 2016 and Wikler and Rakic, 1990). To date, no large-scale comparative study of color vision in mammals has considered each of the steps in protein production (e.g., transcription, translation). Thus, the extent to which visual phenotypes are expressed or masked due to the modulation of protein production is currently unknown, raising the possibility of underestimating the true complexity of the evolutionary history of vertebrate color vision. This represents a major gap in our understanding of visual evolution, as mounting evidence from a range of systems reveals that complex post-transcriptional and post-translational routes shape phenotypic variation and complicate genotype-to-phenotype mapping (Blount et al., 2012; Csárdi et al., 2015; Schwanhäusser et al., 2011). Such incomplete information might also lead to erroneous conclusions surrounding the adaptive significance of particular genotypes.

The potential for selection to act on phenotypes at different stages of protein production may be particularly important during rapid functional trait diversification, as is often the case in visual systems. In sticklebacks, for example, the repeated colonization of lakes with different photopic environments has driven shifts in spectral sensitivity via recurrent selective sweeps in short-wave opsin genes, and changes in opsin expression (Marques et al., 2017; Rennison et al., 2016). Similarly, rapid shifts in the visual ecology of cichlid fishes have involved a combination of coding sequence evolution and changes in expression (O'Quin et al., 2010; Spady et al., 2005). However, in contrast to fishes, much less is known about the changes underpinning rapid visual adaptations in mammals and reptiles, for which relevant studies have tended to focus on ancient transitions to nocturnal, aquatic or subterranean niches (Emerling, 2017; Emerling et al., 2017; Jacobs et al., 1993).

In this study, we investigate the molecular signatures of the repeated loss of S-opsins, and associated dichromatic and UV-vision capabilities, in bats of the superfamily Noctilionoidea (~200 species of New World leaf-nosed bats and allies within the suborder Yangochiroptera). These bats underwent ecological diversification approximately 40 million years ago (Rojas et al., 2012; Rossoni et al., 2017), and show marked morphological and sensory adaptations linked to their unparalleled dietary specializations (Davies et al., 2013; Dumont et al., 2012; Hayden et al., 2014; Monteiro and Nogueira, 2010; Yohe et al., 2017). Switches in feeding ecology from generalized insectivory to blood-, insect-, vertebrate-, nectar- or fruit-based diets have occurred multiple times among closely related species, making noctilionoid bats an outstanding group in which to examine the genetic basis of visual adaptations.

Until recently it was thought that S-opsin, encoded by the OPN1SW gene, was likely functional across the suborder Yangochiroptera (e.g. Butz et al., 2015, Feller et al., 2009, Marcos Gorresen et al., 2015, Gutierrez et al., 2018, Müller et al., 2009, Winter et al., 2003 and Zhao et al., 2009a). However, with increased taxonomic sampling of neotropical bat species this has been shown to not be the case, and multiple independent lineages with diverse ecologies (e.g. blood feeding, plant-visiting species) show evidence of OPN1SW pseudogenization (Kries et al., 2018; Li et al., 2018; Wu et al., 2018). Notably, lineages shown to have lost their S-opsins – and thus by association UV-sensitivity – are from the Noctilionoidea superfamily. In contrast, within the other bat suborder – the Yinpterochiroptera – multiple losses of S-opsin function have previously been documented in lineages of Old World fruit bats as well as horseshoe bats and Old World leaf-nosed bats that have evolved a derived form of laryngeal echolocation (Zhao et al., 2009a). The loss of S-opsins could have profound impacts on bat visual acuity, as inferences from amino acid sequence analyses and action spectra suggest that bat short-wave opsins are sensitive to UV, and their retention is possibly related to the demands of visual processing in mesopic, or low-light, conditions (Zhao et al., 2009a), and/or plant visiting (Butz et al., 2015; Feller et al., 2009; Kim et al., 2008; Müller et al., 2009; Müller et al., 2007). However, the limited taxonomic sampling to date has precluded clear conclusions. Similarly, while the taxonomic sampling of the recent studies of neotropical bat vision (e.g. Gutierrez et al., 2018, Kries et al., 2018, Li et al., 2018, and Wu et al., 2018) is considerably more extensive than previous work, it remains limited, and the functionality of OPN1SW has been based on analyses of DNA sequences and a few transcriptome samples.

To determine whether patterns of adaptations and loss in cone opsins (OPN1SW and OPN1LW) in noctilionoid bats are associated with ecological factors such as diet shifts, we applied analyses of sequence evolution, gene expression, and immunohistochemistry across the taxonomic and ecological breadth of the clade and outgroup taxa. For the first time in mammals, our findings reveal that extensive losses of S-opsin gene function can result from disruption at all three levels of protein synthesis (i.e. DNA open-reading frame, mRNA, and protein). Furthermore, we identify three putative molecular routes that may lead to disruptions of protein synthesis leading to the loss of S-opsins in key lineages. In each instance, the specific route to loss of function was seen in multiple independent lineages. Thus, across the noctilionoids we find evidence that parallel losses leading to identical phenotypes have arisen by both the same and different failures of translation. Hence, current studies both underestimate the extent of parallel losses, and might lead to an incomplete picture based on genes alone.

Results

Our comprehensive analyses of visual opsins that combined information from DNA, mRNA transcripts, and proteins across noctilionoid bats revealed unexpected variation, as well as evidence of extensive parallel losses in S-opsins that have arisen from failures at multiple stages of protein synthesis. First, we used immunohistochemistry to characterize and quantify S- and L-opsin proteins in the retinas of adult bats. Second, for a subset of these taxa, we performed RNA-Seq to assess the presence or absence of transcripts for OPN1SW, OPN1LW, and RHO, and estimated the mode and the strength of selection in coding sequences. Finally, we modeled the presence or absence of S-opsin intact ORF, mRNA or protein presence, as a function of dietary and roosting ecology.

Pervasive parallel losses of shortwave opsin pigments in neotropical bats

To assess the presence or absence of OPN1SW and OPN1LW proteins we applied immunohistochemistry (IHC) to whole, flat-mounted retinas of adult bats (neyes = 218, nindividuals = 187, nspecies = 56). While the presence of a given protein does not guarantee its functionality, here we interpret the detection of protein as indicating a functional cone in the absence of contradictory evidence. Since the absence of protein is difficult to assess, we applied quality control (see Materials and methods for the criteria to accept or reject a retina based on its condition or number of replicates). Surprisingly, OPN1SW was only detected in just over half of the species assayed (n = 32), including all members of the primarily frugivorous subfamily Stenodermatinae, which invariably retain their S-cones (Figure 1 and Figure 1—figure supplements 1 and 2). In contrast, OPN1SW was found to be absent in approximately one third of species assayed (n = 18), including representative species from five bat families. Thus, we not only find evidence of widespread loss of S-opsins within the noctilionoids but also find the first evidence of S-opsin loss in non-noctilionoid Yangochiroptera (Chilonatalus micropus, Eptesicus fuscus and Molossus molossus).

Figure 1 with 5 supplements see all
Distribution of an intact open reading frame (ORF), mRNA transcript, and protein for the OPN1SW, OPN1LW, and RHO photopigments in ecologically diverse noctilionoid bats.

The composition of species diet follows Rojas et al. (2018), dietary types are indicated with the following symbols: invertebrates – moth, vertebrates – frog, fruit – fruit and nectar/pollen – flower. The species phylogeny follows Rojas et al. (2016) and Shi and Rabosky (2015). Vertical black bars, from left to right, indicate: (1) Noctilionoidea, (2) Phyllostomidae, (3) and Stenodermatinae. RNA-Seq data was generated to both infer the presence of an intact ORF (in combination with genomic and PCR sequence data) and to determine the presence of an expressed mRNA transcript. The presence of an intact ORF and mRNA transcript for RHO was verified across all transcriptomes. The presence/absence of a protein product for S- and L-opsins was assayed by IHC on flat mounted retinas. The presence of an intact ORF, mRNA, and protein are indicated by a filled color marker (OPN1SW – purple, OPN1LW – green and RHO – blue), and its absence by a white marker. Missing data (i.e. species for which we were unable to obtain tissue) are indicated with a grey marker with grey outline. Mismatches between intact ORFs and transcripts, or between transcripts and protein data are indicated by an inequality symbol. Note: OPN1SW protein assays for P. quadridens revealed polymorphisms within the sample, and we recorded positive OPN1SW assays in some P. poeyi individuals despite an apparent disrupted ORF. Finally, a grey marker with no outline indicates the failure of protein assay for some species represented by museum specimens (Tadarida brasiliensis, Phyllostomus hastatus, Sturnira tildae, Sturnira ludovici, Platyrrhinus dorsalis and Chiroderma villosum).

https://doi.org/10.7554/eLife.37412.003

To test whether a lack of signal corresponds to a loss of opsin, we aligned the opsin gene sequences among species and confirmed that the epitope-binding site was relatively conserved and showed no correspondence with loss (Figure 1—figure supplement 3, Figure 1—figure supplement 4). For immunohistochemistry, five bat species had replicates that were both wild-caught and from museum collections and exhibited the same phenotype, highlighting the robustness of the experiments (Figure 1—figure supplement 5). We also verified absences using multiple replicates of slides from the same individuals, and, when available, from multiple individuals from the same species (with a minimum of two individuals; see Supplementary file 1). In one species, P. quadridens, we detected evidence of polymorphism in the presence of S-opsin cones among fresh specimens, with three of the 17 specimens lacking S-cones. Samples derived from museum specimens of six species had low signal-to-background ratios in OPN1SW protein labeling (e.g. opsin-specific staining was seemingly detected, but non-specific background staining was high, making specific staining difficult to distinguish from background), generating inconclusive results, and these specimens were therefore excluded from our models. In contrast to our OPN1SW results, we detected OPN1LW protein in all species examined (Figure 1 and Figure 1—figure supplements 1 and 2).

We then analyzed patterns of OPN1SW and OPN1LW protein localization among cells. Consistent with cone-specific roles, we found that almost all cones expressed either OPN1SW or OPN1LW protein, with no strong evidence of co-localization of both proteins (Figure 2 and Figure 1—figure supplements 1 and 2).

L and S opsin cone distribution in 14 representative noctilionoid bat species.

Density maps of L and S opsin cone topography in 14 noctilionoid bat species. For each species, a representative dissected flat-mounted retina is shown. Insets are representative IHC magnifications of flat mounted retinas immune-stained for either L- or S-opsins in the highlighted region. Dietary types are indicated with the following symbols: invertebrates – moth, vertebrates – frog, fruit – fruit and nectar/pollen – flower. Measured opsin densities (0–8000 cones/mm2) are represented by the following color scales: L-opsins – green and S-opsins – purple.

https://doi.org/10.7554/eLife.37412.009

Multiple parallel losses in transcripts across taxa

To investigate the underlying molecular causes of the above-detected losses of OPN1SW, we began by sequencing total mRNA isolated from the eye tissue of 39 species, assembled the short-read data and used a BLAST approach to annotate visual pigments. We found evidence of at least partial OPN1SW mRNA transcripts in a total of 34 bat species; thus, expression was only absent in five of the species assayed. Absences of OPN1SW transcripts were phylogenetically widespread, and included divergent species from three families (Natalidae: Chilonatalus micropus; Mormoopidae: Mormoops blainvillei and Phyllostomidae: Macrotus waterhousii, Brachyphylla (nana) pumila and Lionycteris spurrelli) (see Figure 1 and Supplementary file 1). In addition to losses observed in C. micropus and M. blainvillei, the three phyllostomid species each belong to separate subfamilies, and therefore, likely represent independent losses of OPN1SW expression. In comparison, we were able to recover the complete RHO and OPN1LW transcript from all taxa assessed (n = 39; Figure 1).

Mismatches in opsin transcript and protein suggest alternate parallel failures of translation

While IHC revealed pervasive loss of S-cones across our study sample, we only detected loss of the OPN1SW transcript in a few species. Comparison of our OPN1SW transcript and protein data revealed numerous conflicts in species-specific absences, with a total of nine lineages found to possess OPN1SW transcripts but lack OPN1SW protein (see Figures 1, 3 and 4). These species included Molossus molossus, Pteronotus parnellii, Desmodus rotundus, Trachops cirrhosus, Tonatia saurophila, Gardnerycteris crenulatum, Monophyllus redmani, Erophylla bombifrons, and Carollia brevicauda. This may also be the case in additional species, for example Tadarida brasiliensis, Eptesicus fuscus, Pteronotus davyi and Diaemus youngi, but we currently lack the complementary data to confirm this.

The putative routes explaining variation in S-cone presence in noctilionoid bats.

In each panel, upper images (left and right) show the gross phenotype of the eye in representative bat species, and lower images (numbered) show IHC magnifications of their respective flat mounted retinas immune-stained for S-opsin. Diets are depicted with the following symbols: invertebrates – moth, vertebrates – frog, fruit – fruit and nectar/pollen – flower. (i): Information in the intact DNA Open Reading Frame (dark purple) is transcribed to form mRNA (light purple), which is then translated into OPN1SW. Codon analyses reveal purifying selection. Example species: (L + 1) Artibeus jamaicensis; (R + 2) Noctilio leporinus. (ii): The DNA ORF is disrupted (grey) by the presence of STOP codons (*) and indels (black triangles). Neither OPN1SW mRNA (dashed boxes) nor OPN1SW are detected. Codon analyses reveal relaxed selection. Example species: (L + 3) Mormoops blainvillei; (R + 4) Macrotus waterhousii. (iii): Although the DNA ORF (dark purple) appears to be intact, information is not transcribed to mRNA (dashed boxes), and no OPN1SW is detected. Example species: (L + 5) Brachyphylla pumila (R + 6) Chilonatalus micropus. (iv): Information in the intact DNA ORF (dark purple) is transcribed to form mRNA (light purple); however, the OPN1SW is not detected. Codon analyses reveal purifying selection. Example species: (L + 7) Monophyllus redmani (R + 8) Pteronotus parnellii..

https://doi.org/10.7554/eLife.37412.010
Inferred parallel losses of S-opsins mapped on to the species phylogeny and exonic content of reconstructed mRNA.

(i) Taxa and branches are colored as follows: presence of protein – black; presence of intact ORF but no protein – green; presence of mRNA but absence of protein – blue; evidence of pseudogenization (disrupted ORF) – red; protein status not determined – grey. Weight of branches indicates: inferred presence of protein – heavy; inferred absence of protein – light; protein absence based on evidence of gene loss but not confirmed by IHC – dashed light. We were either not able to recover mRNA, or preserved material was not available, for species marked with ‘*’, evidence for ORF status for Diaemus youngi taken from Kries et al., 2018. The species phylogeny follows Rojas et al., 2016 and Shi and Rabosky, 2015. (ii) Reconstructed mRNA transcript variants of seven species (M. molossus, T. cirrhosus, T. saurophila, G. crenulatum, P. parnellii, M. redmani, and C. brevicauda) with OPN1SW mRNA present but no detected protein, and P. quadridens for which presence of detected protein varied across individuals. The four biological replicates of P. parnellii are numbered 1–4. Sections of the mRNA are indicated as follows: exons 1–5 – purple filled boxes; introns 1–4 – black lines; the 3’UTR – white filled triangle; and missing regions – white regions. (iii) Reconstructed mRNA transcript variants of four species (P. elongatus, A. geoffroyi, G. soricina and A. jamaicensis) with OPN1SW mRNA present and detected protein. The four biological replicates of A. jamaicensis are numbered 1–4. Sections of the mRNA are indicated as above. (iv) Reconstructed mRNA transcript variants of two species (M. nigricans and T. brasiliensis) with OPN1SW mRNA present but protein status not determined. Sections of the mRNA are indicated as above.

https://doi.org/10.7554/eLife.37412.011

Of the nine species lacking S-cones, but in which the presence of mRNA transcripts was detected, we further examined the nucleotide sequence of the assembled transcripts for both an intact open-reading frame (ORF) and completeness of transcript. Only a single partial mRNA fragment was recovered each for D. rotundus and M. redmani. The partial D. rotundus fragment (~300 base-pairs of exons 2–4) contained a premature stop codon – confirmed by PCR and the recently published common vampire bat genome. The partial M. redmani fragment (~240 base-pairs of exons 2–3) did not contain premature stop codons or indels; however, we cannot rule out the possibility that these may be present in the remaining exons not sequenced. We recovered a total of three OPN1SW transcripts for E. bombifrons from our transcriptome assembly, and the longest of these transcripts contained a putative four base-pair deletion and retained a portion of the intron between exons 2 and 3. Therefore, D. rotundus and E. bombifrons appear to have transcribed OPN1SW pseudogenes.

For the remaining six species (M. molossus, P. parnellii, T. cirrhosus, T. saurophila, G. crenulatum and C. brevicauda) for which the OPN1SW transcript was present but the S-cone protein was absent several alternative scenarios emerged. First, for two species, C. brevicauda and T. saurophila, our RNA-Seq assemblies recovered a single and complete OPN1SW mRNA transcript containing all five exons and the 3’ UTR, albeit with ~5 codons missing at the 5’ end in one of these taxa (C. brevicauda). The individuals sequenced for M. molossus revealed five OPN1SW transcript isoforms, two for T. cirrhosus and four for G. crenulatum. While the complete transcript (i.e. exons 1–5, and no intronic sequences) was detected in each of these species, we also found evidence of alternative splice variants characterized by either missing exons (M. molossus), or retained introns. As a result, the reason for the apparent failure of the S-opsin translation is unclear. Finally, for Pteronotus parnellii (n = 4), we detected the most splice variation, with 2, 4, 5, or 18 variants assembled per individual. Furthermore, in P. parnellii, we were unable to recover any intact mRNA isoforms among these variants. Many of the variants were repeated across individuals; for example, an entirely missing exon five (despite the presence of the up-/down-stream sequences), and partial deletion of exon 1, was seen in two individuals (see Figure 4). We thus speculate that these splice variants explain the observed failure of translation that leads to the absence of OPN1SW protein in P. parnellii. We also detected an in-frame three base-pair deletion (Y190del) in three of four P. parnellii individuals sequenced. A similar pattern of isoform variants, as detected in P. parnellii, was also seen in P. quadridens (n = 1). Therefore, given the polymorphic status of S-cones in P. quadridens, we speculate that the individual sequenced may not have had functional S-cones.

The retention of intronic sequence, while unexpected, does not necessarily indicate a non-functional gene by itself. For example, we detected limited evidence of some intron retention in species for which the protein data suggest S-opsin presence, for example, Artibeus jamaicensis and Phyllops falcatus. Furthermore, across the 39 species, we found some evidence of OPN1LW transcript variation in 15 individuals (14 species), with instances of retained introns, missing exons and, in one case (Artibeus planirostris) an indel, although this latter case may arise from assembly error. For RHO, we also found some evidence of retained introns for three species (Anoura geoffroyi, T. brasiliensis and T. cirrhosus). However, summing across these 16 species, and except for T. cirrhosus, we always recovered an isoform with all exons and no intronic sequences for both RHO and OPN1LW.

Among the four species lacking the OPN1SW transcript and protein, we found the ORF recovered by manual PCR of exons 3–4 was intact in C. micropus (the single sampled Natalidae species), as well as the unrelated phyllostomid B. (nana) pumila. In contrast, the genomic sequences recovered by blastn was disrupted by a number of insertions and deletions resulting in premature stop codons in Mormoops blainvillei, Macrotus waterhousii, and Lionycteris spurrelli.

In the 17 species for which the S-opsin protein was detected and for which we were also able to test the presence of the transcript, we found a 1-to-1 correspondence in all but one case. The exception was Phyllonycteris poeyi, for which data from four individuals showed S-cone presence, yet the transcript of one other individual was inferred to be non-functional based on a four base-pair insertion (confirmed via PCR). We also note that within the OPN1SW sequences of most species examined, the ATG start codon was found to be three codons downstream relative to that of the human orthologue (Figure 1—figure supplement 2). In contrast to the mismatches between OPN1SW mRNA and OPN1SW protein, we found complete correlation between the presence of the transcript and protein for OPN1LW in these species.

Molecular evolution of opsin genes

To gain further insights into the molecular evolution of opsins, we performed tests of divergent selection in alignments for each of the three opsin genes (OPN1SW, OPN1LW and RHO) among three types of lineages: (1) those with S-cones; (2) those without S-cones but with OPN1SW transcripts and an intact OPN1SW ORF; and (3) those without either S-cones or OPN1SW transcripts (see Figure 5—figure supplement 1). We found a significantly higher ω in lineages with a pseudogenized OPN1SW in both the OPN1SW gene (ωbackground = 0.13; ωOPN1SW.intact=0.24; ωOPN1SW.pseudo=0.78; χ2(2)=70.99, p=3.84e-16), and the OPN1LW gene (ωbackground = 0.08; ωOPN1SW1.intact=0.08; ωOPN1SW.pseudo=0.19; χ2(2)=9.18, p=0.01). In contrast, we found no differences in ω across the different lineages for RHO, indicating strong and negative selection in this gene (Table S2 in Supplementary file 2). To test the influence of diet on rates of molecular evolution, we compared ω for all three opsin genes between frugivorous and non-frugivorous lineages. We found no differences in rates for OPN1SW or OPN1LW, however background branches (non-frugivorous) had a significant and slightly higher ω for RHO (ωbackground = 0.04; ωfugivory= 0.01; χ2(1)=13.77, p=2.07e-4; Table S3 in Supplementary file 2).

Ecological correlates of opsin presence and density

We compared the locations and densities of long-wavelength-sensitive cones (or L-cones expressing OPN1LW protein) and short-wavelength-sensitive (or S-cones expressing OPN1SW protein) in the whole, flat-mounted retinas of adult bats for 14 species for which we had sufficient specimen replicates (Table S4 in Supplementary file 2). We found photoreceptor density varied among examined species (Figure 2, Table S4 in Supplementary file 2), with mean cone densities ranging from 2500 to 7500 cones/mm2 for L-cones, and 327 to 5747 cones/mm2 for S-cones (Figure 2, Table S4 in Supplementary file 2). In every case, the density of S-cones was lower than that of L-cones. Densities of both cone types tended to be highest near the center of the retina in all species. Bat species with S-cones had significantly higher densities of L-cones, with the presence of S-cones increasing the ln-transformed density of L-cones by 43%, explaining on average ~24% of the variance in density between species (Table 1).

Table 1
Summary of Bayesian regression models of the presence of S-cones or the ln-transformed density of L-cones as a function of predictor variables.

R2, variance explained by sample-wide factors; Σ, species-specific phylogenetic effect for species.

https://doi.org/10.7554/eLife.37412.012
FormulaR2ParameterMedian2.5%97.5%
Presencei ~ a +  b. fruit_prevalenti + Σ0.503intercept (a)−1.10−3.190.52
b3.661.856.76
Σ1.940.028.64
 ln(density)i ~ a +  b. fruit_prevalentj + Σj0.000intercept (a)8.428.008.84
b0.22−0.210.66
Σj0.140.040.46
 ln(density)i ~ a +  b. S-cones_presentj + Σj0.240intercept (a)8.287.958.63
b0.430.090.78
Σj0.070.0000.51

We tested the influence of ecology on the presence of the OPN1SW ORF, mRNA, or S-cones using Bayesian hierarchical models in which the observations corresponded to species (OPN1SW ORF nspecies = nobservations = 45, mRNA nspecies = 39, nspecies = 50), and a phylogenetic structure of the data was included as a species-specific effect. Two types of predictor variables were analyzed: three variables for diet and one for roosting. Comparisons of the coefficients, which are all at the same scale because they are multipliers of the presence of a particular ecology, showed frugivory was the best factor for explaining the presence of S-cones (Figure 5, Tables S5-S7 in Supplementary file 2). The predominance of fruit in the diet increases the odds of having S-cones roughly 39 times, and explains about 50% of the between-species variance in the presence of the S-cone (Table S7 in Supplementary file 2, Figure 5).

Figure 5 with 1 supplement see all
S-opsin presence is correlated with diet.

(A) Representative IHC magnifications of flat mounted retinas immune-stained for both L and S opsin for four species representatives of the diversity of phenotypes observed. Fruit-based diet: Artibeus jamaicensis, pollen/nectar-based diet: Monophyllus redmani, and insect-based diet: Mormoops blainvillei and Pteronotus quadridens. (B) Violin-plots of the posterior estimates of the coefficients for the presence of an OPN1SW ORF, mRNA, or cone as a function of ecological covariates. The gray horizontal lines indicate a coefficient of 0, or no effect of the covariate on the response. The high-probability density estimates for all coefficients are given in Tables S5-S7, and for fruit in Table 1.

https://doi.org/10.7554/eLife.37412.013

Discussion

By studying opsin gene sequences, transcripts and proteins across a radiation of neotropical bats (suborder Yangochiroptera, superfamily Noctilionoidea), we discovered remarkable diversity in visual genotypes and phenotypes, including parallel loss of function in OPN1SW and the associated S-cones. Although parallel losses have been reported before, the multiple steps involved in functional loss were hitherto unsuspected. We evaluate both the diversity of paths to acquiring a similar phenotype, and the ecological covariates that may explain evolution in this system.

Complex independent routes of phenotypic loss

Across our study taxa, we documented molecular signatures consistent with as many as 17 instances of parallel loss in S-opsins, and equated these with a minimum of three putative routes leading to the failure of the formation of the S-opsin cones (Figures 1, 3 and 4). First, we found evidence of multiple independent instances of pseudogenization, associated with either absent or fragmentary mRNA transcripts. Second, we found evidence of an apparently intact gene sequence that did not result in a mRNA transcript. Finally, we recovered putatively intact OPN1SW transcripts that did not result in the corresponding S-cone protein, which in some species appears to arise from aberrant isoforms.

The first route of parallel loss in OPN1SW in noctilionoid bats involved disruption in opsin reading frames, a finding that has been documented in other mammals (e.g. Emerling et al., 2015Emerling et al., 2017, Hunt and Peichl, 2014, Kraus et al., 2014 and Zhao et al., 2009a), including bats from both suborders (e.g. Emerling et al., 2015, Kim et al., 2008, Kries et al., 2018, Müller et al., 2007, Wu et al., 2018 and Zhao et al., 2009a). In general, pseudogenization is thought to occur relatively frequently within mammalian genomes, and previous estimates suggest several thousand pseudogenes may be present per genome (e.g. Torrents et al., 2003). The recent proliferation in published genomes has also led to increased efficiency in the detection of this form of gene function loss, and typically this is one of the most frequently cited mechanisms of gene loss (e.g. Emerling et al., 2018 and Jebb and Hiller, 2018).

The second inferred form of parallel OPN1SW loss in which a putatively intact open reading frame exists but appears not to be transcribed was found in two highly divergent bat species. To our knowledge such mismatches between opsin coding DNA and mRNA have not previously been documented in mammals, including bats, although this largely reflects a shortage of suitable datasets. Indeed, obtaining material with intact mRNA is challenging, and few studies have been able to test the commonly held assumption that an intact ORF equates to functionality. Additional research is thus necessary to confirm the existence, extent and underlying mechanism of ORF-transcript mismatched in bats and other groups. Possible explanations for our observed mismatches include regulatory elements and epigenetic modifications, but a lack of genomic resources for these species precludes more detailed investigations at the present time.

The most widely detected form of parallel loss of S-opsins, seen in six species, was associated with the apparent failure of the expressed OPN1SW transcript to be translated into protein. Of these species, protein data for three were obtained from field specimens (<4 years old), two from museum samples, and one from both field (<3 years old) and museum samples. Our inspection of the transcript repertoires of these affected species suggests that mismatches might arise from multiple molecular routes. Across the sampled bats, we found evidence of the expression of multiple mRNA isoforms, that in many cases contained either retained introns or skipped exons, both of which are likely to impede translation. Four individuals of P. parnellii exemplify this transcript variation, as none of many mRNA transcripts among these individuals were complete. Reports of readthrough of introns, and the skipping of exons, are becoming increasingly common (e.g. Gaidatzis et al., 2015, Wen et al., 2018 and Wong et al., 2016), and these have previously been linked to loss of gene function (e.g. Lopes-Marques et al., 2018). Indeed it is particularly noteworthy that the OPN1SW mRNA of the blind mole-rat Spalax ehrenbergi has also been found to contain introns (David-Gray et al., 2002; Esquiva et al., 2016). Underlying mechanisms for these cases could potentially include mutations leading to loss of splice sites, the evolution of novel cryptic splice sites or a reduction of spliceosome efficiency (David-Gray et al., 2002; O'Neill et al., 1998). We note, however, that in species other than P. parnellii either a single, complete mRNA transcript was recovered, or at least one of the alternative assemblies represented the complete transcript – therefore, the ultimate molecular cause of the failure of the protein to be synthesized is unclear in these cases.

Although the causal mutations or mechanisms underpinning losses of function in OPN1SW are currently not known, the observed absence of expression with putatively intact ORFs in some species, alongside the converse condition in D. rotundus, strongly indicates independent routes. Similar diversity is seen in the Mormoopidae in which we detected a disrupted ORF and no OPN1SW mRNA expression in the Mormoops lineage, but an intact ORF and mRNA expression in the Pteronotus lineages, as well as evidence of S-cones in some Pteronotus species. Given that the Mormoops and Pteronotus lineages diverged ~30 million years ago, and the taxa sampled within the Pteronotus lineages diverged ~16 million years ago (Pavan and Marroig, 2017), these patterns do not support a disruption in the common ancestor of Mormoopidae as this would have to be followed by different trajectories that led to complete gene loss in one lineage, and partial retained function in the other. The alternative scenario in which each of these cases of loss involved the same mechanism seems highly unlikely given that it would have had to have taken place independently at least four times within the family, with each of the sampled taxa from our study being at a different stage of the gene loss process.

Differing degree of retention of opsins

In strong contrast to the results from OPN1SW, data from proteins and transcripts revealed complete retention of OPN1LW across our study species. Such extreme differences in the conservation of color vision genes have previously been reported in other vertebrates (e.g. Zhao et al., 2009a and Zhao et al., 2009b). Our IHC assays also revealed little evidence of co-localization of both proteins, which is consistent with cone-specific roles. This contrasts with a previous study of two noctilionoid bat species that found that almost all L-cones expressed some S-opsin (Müller et al., 2009), although this discrepancy could have arisen from methodological differences. To document fluorescence, the previous study used epifluorescence microscopy, while our study used confocal microscopy. Through its generation of serial optical sections, confocal microscopy typically provides better resolution for co-localization studies. We also found that bat species with S-opsin cones tend to have more L-opsin cones, consistent with both types of cones serving a common functional role.

We also found that S-cone retention varied among conspecifics. In Pteronotus quadridens, three of 17 individuals were found to lack S-cones. This heterogeneity could indicate the ongoing degradation of protein synthesis. Indeed, allelic variation has been reported to contribute to opsin variation in diurnal lemurs (Jacobs et al., 2017) and has previously been detected in OPN1SW in Pteronotus mesoamericanus (Wu et al., 2018).

Ecological determinants and agents of selection

Alongside parallel losses of shortwave-sensitive opsins in some noctilionoids lineages, we found strong conservation of S-cones, OPN1SW transcription, and protein-coding sequences in around 20 of the species studied. Thus, S-cones appear to still play an important function in these bats. Although the pseudogenization of OPN1SW, or loss of transcription had both been previously explained by the use of caves as roosts (Gutierrez et al., 2018; Wu et al., 2018), our phylogenetic regressions estimated the coefficient for this factor to include 0 (Figure 5). Instead, we identified the predominance of fruit consumption as the single most powerful explanatory factor explaining the variation in S-cone presence across the clade, with a similarly positive but not statistically significant effect for OPN1SW transcription and protein-coding sequences. Surprisingly, the result for plant-visiting (which includes flower-visiting bats) was not similarly strong and no such result was found for insectivory or cave-roosting. Diet therefore appears to be the primary selective agent for maintaining S-opsin function. We also found that while some predominantly nectarivorous species from both independent nectar-feeding lineages, have lost their S-cones, others have retained them (e.g. Anoura geoffroyi). While the loss of S-opsins in flower visiting bats may seem maladaptive, behavioral assays have previously been used to infer that some nectarivorous phyllostomid species appear to be color blind, and thus may be able to utilize UV reflectance to locate flowers via an alternative rod-based mechanism (Winter et al., 2003). This suggests that either more than one strategy to locating flowers has evolved among New World leaf-nosed bats, or other non-visual cues are used (e.g. Gonzalez-Terrazas et al., 2016).

Since fruit consumption arose as an evolutionary innovation within the Yangochiroptera, selection for this novel niche cannot explain the ancestral or present-day persistence of S-cones in non-frugivorous species. A role in light capture rather than in detecting novel visual cues might explain the density of S- and L-opsin cones in the non-frugivorous lineages sampled, as well as in ancestral bats. The signals of strong purifying selection of all three visual opsin sequences to conserve ancestral function in both frugivorous and non-frugivorous lineages further buttresses interpretation, as it implies there is no detectable relaxation of selection on non-frugivorous lineages, at least among species for which sequences were available. At the same time, divergent selection in frugivorous and non-frugivorous lineages in RHO may further support the importance of light capture, and dim light vision, in relation to novel diets in frugivorous species.

As expected once pseudogenization has occurred, the main difference in molecular selection was between species with an OPN1SW ORF and a pseudogene at this locus. Instead of directly reflecting ecological covariates, the process of pseudogenization appears to represent the culmination of a longer term process that starts earlier with cone loss. This highlights post-transcriptional regulation as a more direct response to ecology than pseudogenization of the relevant opsin. Therefore, protein composition should more closely reflect visual ecology than high rates of sequence evolution and pseudogenization in the relevant opsin, as the latter only responds to long-term functional loss. We further tested this inference by modeling the presence of an OPN1SW ORF, mRNA, or protein as a function of ecological covariates, finding the strongest ecological association—estimated by higher coefficients—with the presence of S-cones (rather than with earlier steps in protein production).

Recent studies of color vision evolution in New World leaf nosed bats have begun to explore the complex picture of opsin gene loss in the context of selection and ecological factors (Gutierrez et al., 2018; Kries et al., 2018; Li et al., 2018; Wu et al., 2018). The detected pseudogenization of OPN1SW in infrared sensing vampire bats and in high-duty cycle (HDC) echolocating bats such as Pteronotus mesoamericanus (formerly P. parnellii mesoamericanus) have led researchers to invoke evolutionary sensory trade-offs as one factor behind the loss of color vision (Kries et al., 2018; Li et al., 2018; Wu et al., 2018). An additional loss was detected in Lonchophylla mordax (Kries et al., 2018), a nectar bat that roosts in caves, with the roosting preference taken to be driver of color vision loss in this species. In cases in which either no S-opsin losses were inferred in Yangochiroptera, or selection analyses were performed across both Old and New World species simultaneously (Gutierrez et al., 2018), it is not possible to interpret the results solely in the context of noctilionoids. Sensory trade-offs, foraging strategy and obligate cave roosting are hypotheses that have previously been applied to loss of S-opsins in Old World bat lineages (e.g. Zhao et al., 2009a), however these traits often co-vary within species so the signal may be difficult to disentangle. By allowing us to detect previously ‘hidden’ opsin phenotypes across noctilionoid species, our approach has allowed us to identify previously undetected ecological factors, that is, fruit consumption, as an explanatory variable of S-opsin retention. Furthermore, the discovery of loss of gene function in non-HDC Mormoopidae, e.g. Mormoops blainvillei and Pteronotus davyi, also call into question the sensory trade-off hypothesis within this family.

Study limitations and alternative interpretations

The surprising diversity in S-opsin retention recorded in this study was seen across divergent species, congeners, and even conspecifics. Although such patterns could also arise from methodological issues, some of our findings appear consistent with emerging trends. For example, within Pteronotus and the Mormoopidae family as a whole, there is increasing evidence to support an extremely complex evolutionary history of S-opsins (Gutierrez et al., 2018; Simões et al., 2018; Wu et al., 2018). In comparison, methodological artifacts are less easy to rule out as causes of variation among Carollia spp. given that S-cone presence in two species was inferred from either recently collected field specimens, or a mixture of both field and museum specimens, while S-cone absence in a third species was based on museum specimens collected in 1968 and 1972.

Despite this, the utility of long-term fixed specimens for immunohistochemical staining of proteins (e.g. vimentin and GFAP) has been described previously (Hühns et al., 2015; Thewissen et al., 2006), including visual opsins in some museum specimens (~20 years old and using the same S-opsin antibody as used in the current study) (Nießner et al., 2016). In line with this, we were able to recover good IHC staining for both S- and L-cones from our oldest sampled museum specimen with a confident date, an Artibeus fraterculus collected in 1921 (see Figure supplement S3). We note, however, that our S-opsin assays were inconclusive for six species (T. brasiliensis, P. hastatus, S. tildae, S. ludovici, P. dorsalis and C. villosum) represented by museum samples due to low signal-to-background ratios, but L-opsin assays were successful in these species. The ability to detect protein in these, and other, museum specimens is a function of the condition of the retina and the density of the cones in question. Because L-cones are present at higher densities than S-cones (this study and Müller et al., 2009), we were able to more readily detect L-cones even in more poorly preserved retina. Thus, while museum collections are rarely used in protein studies, relative to their use, for example, in genetic and genomic studies (e.g. Bailey et al., 2016; Nießner et al., 2016), they offer great potential for a range of comparative studies provided that caution is exercised (e.g. Hedrick et al., 2018). These benefits apply particularly to groups that cannot be sampled in the wild for ethical, conservation and logistic reasons (e.g. Russo et al., 2017). In our study, restrictions on sampling necessitated our comparisons of conspecifics collected from different countries for genetic and protein assays. For this reason, we cannot rule out geography as a source of variation, and it is noteworthy that one focal taxon, Pteronotus parnellii was recently recognized as a species complex with a strong phylogeographic divergence (Pavan and Marroig, 2017). Although other focal bat species have not been split in this way, many have wide ranges and their genetic diversity may be underestimated (Clare et al., 2011). In another instance, our fresh specimen of M. molossus from Belize and two older museum specimens from 1968 from Uruguay all lacked S-cones, whereas a published record of an individual from an unknown geographical locality showed S-cone presence (Nießner et al., 2016). These patterns of S-cone presence, and indeed those across the entire noctilionoid tree, suggest that losses may arise across populations of the same species.

We must also consider whether our results might arise from methodological artifacts related to the short read data. For example, low gene expression can limit the number of representative reads in RNA-Seq datasets for transcript assembly (e.g. Zhao et al., 2011) and this caveat likely applies to shortwave-sensitive opsins that show low S-cone densities, inferred loss of function in some taxa, and also low expression levels based preliminary expression analyses (data not shown). To cross-validate our assembled transcripts, whenever possible we compared our sequences to published data from PCR amplicons, RNA-Seq, and/or genome datasets. For example, for Monophyllus redmani, Trachops cirrhosus and Pteronotus parnellii we were able to confirm that our assembled transcripts matched published assembled mRNA contigs, PCR amplicons and genomic sequences, respectively (Gutierrez et al., 2018; Tsagkogeorga et al., 2013; Wu et al., 2018). For Mormoops blainvillei and Macrotus waterhousii, the absence of OPN1SW mRNA transcripts and S-cones was supported by disrupted ORFs in published genome datasets, as well as the highly divergent and fragmented transcripts recovered by a recent study, which our visual inspections suggest may be due to cross contamination or misassembly (Gutierrez et al., 2018). Therefore, several lines of evidence support loss of function of OPN1SW in these taxa. For C. micropus, our inferred intact ORF is based on PCR of ~100 codons (also seen in B. pumila) and was supported by sequence data from two closely related species from the Natalidae (Emerling et al., 2015; Simões et al., 2018).

For several species, our transcriptomic analysis detected multiple OPN1SW mRNA transcripts variants, characterized by retained introns and missing exons. We are able to confirm the species-specific intronic sequences of several of the species due to recently available genomes and gDNA PCRs (Kries et al., 2018; Li et al., 2018; Tsagkogeorga et al., 2013; Wu et al., 2018). The observed retention of introns in OPN1SW mRNA as well as the expression of pseudogenized opsin mRNA are both supported by earlier studies (David-Gray et al., 2002; Schweikert et al., 2016); however, alternative scenarios for these findings could include gDNA contamination, sequencing of immature mRNA or low-level cross contamination resulting in the assembly of highly divergent transcripts. Finally, for most species our mRNA evidence is based on one individual. However, the lack of a clear relationship between RIN score or sequencing depth and presence of OPN1SW mRNA (data not shown), together with the presence of the two other visual opsins in all samples, suggests this should be sufficient.

Other studies of parallel loss and associated mechanisms

Our findings provide important insights into how parallel losses occur in response to diverse ecological demands, as well as how several alternative molecular routes may lead to the same phenotype. There are other examples of parallel loss from pelvic reduction in sticklebacks (via repeated changes in a Pitx1 enhancer), color and vision in Astyanax cavefish (via loss of function of Oca2), trichomes in Drosophila spp., and floral pigments in Iochrominae (Chan et al., 2010; Larter et al., 2018; McGregor et al., 2007; Protas et al., 2006), but few of these have examined as many species across as many steps of phenotype production. Our data and those from other recent studies on bat opsins associate independent losses of S-cones with diverse adaptations (e.g. shifts in diet, roosting ecology and sensory traits), and are therefore consistent with multiple, distinct ecological demands leading to the same phenotype. Hence, our findings are also consistent with the hypothesis that UV vision represents a genetic ‘hot spot’ of evolution (Hoekstra and Coyne, 2007; Martin and Orgogozo, 2013; Stern and Orgogozo, 2008), along an evolutionary line of least resistance (Schluter, 1996). Therefore, by documenting a range of molecular routes to functional degradation, this study supports the hypothesis that vision is a highly evolvable trait that repeatedly and rapidly changes in response to diverse selective demands.

In conclusion, our findings reveal that assessments of visual perception based purely on genotypic analyses of either opsin sequences or RNA transcripts can be misleading, and may even obscure the evolutionary processes and ecological agents of selection. Although variation in the complement of photoreceptors across vertebrates is usually explained by disruptions to the protein-coding sequence (e.g. Mundy et al., 2016 and Zhao et al., 2009a), findings of mismatches between genotype and phenotype also indicate a role for transcriptional and even translational control in this process. It follows that because routes of gene loss are mainly studied at the genetic level or, in fewer cases, at the transcriptomic level, the input of changes in translation and other connections between the genetic, transcriptomic, and proteomic levels may be being underestimated. More broadly, our results highlight the importance of rapid trait loss in evolution, with apparent shifts in translation and transcription that precede pseudogenizing changes in ORFs. As genotype-centered analyses would miss important functional changes, our study also illustrates the importance of probing multiple levels of protein synthesis.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Antibodysc-14363, goat anti-OPN1SWSanta-Cruz BiotechnologiesRRID: AB_2158332(1∶1000)
Antibodyab5405, rabbit anti-opsin red/greenMillipore Ibérica, MadridRRID: AB_177456(1∶750)
Antibodydonkey anti-goat Alexa Fluor 568ThermofisherRRID:AB_2534104(1∶500)
Antibodydonkey anti-rabbit Alexa Fluor 647ThermofisherRRID: AB_10891079(1∶500)
Software,
algorithm
FIJIhttps://fiji.sc/
Commercial assay or kitRNeasy Mini kitsQiagen
Commercial assay or kitQiagen DNeasy Blood and Tissue KitQiagen

Species sampling and tissue preparation 

Request a detailed protocol

We obtained eye tissue from 59 New World bat species, of which 49 were collected from the wild and 34 from the American Museum of Natural History (AMNH), with 24 species common to both sources (Supplementary file 1). Our sampling was designed to maximize taxonomic coverage and include as many replicates as possible within ethical and regulatory limits. Unlike lab animals such as mice or rats, most bat species have just one offspring per year (Wilkinson and South, 2002), limiting the rate of recovery from adult mortality. All wild bats were captured with traps set in forests and/or at cave entrances, were handled, and then euthanized by isoflurane overdose, under appropriate research and ethical permits (see Appendix 1).

RNA sequence analysis

Request a detailed protocol

Intact eyes were placed in RNAlater and incubated at 4°C overnight and then frozen. Total RNA was isolated using Qiagen RNeasy Mini kits with the addition of DTT and homogenization using a Qiagen TissueLyser. Following QC, total RNA from each individual was used to construct a cDNA library using the Illumina TruSeq RNA v2 kit. Pooled libraries were sequenced (NextSeq 500). Eye transcriptomes were generated for 46 individuals (39 species) including biological replicates of Pteronotus parnellii (n = 4), Artibeus jamaicensis (n = 4) and Phyllops falcatus (n = 2). Raw reads were trimmed, and clean reads were assembled with Trinity v.2.2.0 (Grabherr et al., 2011) (see Appendix 1).

We tested for the presence of the three focal gene transcripts (RHO, OPN1SW and OPN1LW) in each bat transcriptome using a reciprocal best hit blast approach against the full set (n = 22,285) of human protein-coding genes from Ensembl 86 (Yates et al., 2016). To confirm the absence of OPN1SW sequence, we performed additional steps in several species. First, we cut, trans-chimeras, which can prevent detection by reciprocal blast (Yang and Smith, 2013), and repeated the reciprocal blast. Second, we manually screened sequences that were initially identified as matching OPN1SW, but did not pass initial blast filtering (see Supplementary Information). Recovered opsin gene sequences have been submitted to GenBank (accession numbers MK209460 - MK209505 [RHO]; MK209506 - MK209551 [OPN1LW]; and MK209552 - MK209592 [OPN1SW]).

Additionally, for each individual RNA dataset we manually aligned all assembled transcripts, that passed the tblastn step of the reciprocal blast for OPN1SW, together with individual exons and introns obtained from the Myotis lucifugus structural annotation downloaded from Ensembl. Finally, we obtained all OPN1SW DNA and mRNA sequences currently available for our study species from GenBank, produced by recently published studies or genomes (Gutierrez et al., 2018; Kries et al., 2018; Li et al., 2018; Wu et al., 2018; Zepeda Mendoza et al., 2018). This data was used to confirm either our mRNA assemblies or the intronic sequences, and also to infer ORF status for species in which we had protein data for but were not able to obtain tissue for RNA-seq (e.g. Eptesicus fuscus, Pteronotus davyi, Diaemus youngi, Phyllostomus discolor, Sturnira lilium)

Immunohistochemistry (IHC) and photoreceptor quantification

IHC assays

Request a detailed protocol

Specimens for IHC were obtained from the wild (fresh) and from collections of the AMNH (preserved). Given the variability of the age of the preserved museum specimens, the initial fixation method is not always known. However, it is most likely they would have been initially fixed with formaldehyde/formalin and then stored in 70% ethanol. Fresh eyes were fixed overnight at 4°C in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS), transferred into 1X PBS, and then stored in 1% PFA in 1X PBS at 4°C until further processing. Preserved eyes had been collected previously and stored in 70% ethanol at room temperature for varying lengths of time (see Supplementary file 1). Preserved eyes were rehydrated through an ethanol series (70% 20 min, 50% 20 min, 20% 20 min) in 1X PBS and then stored in 1% PFA at 4°C in 1X PBS. Prior to processing, retinas were dissected from eyeballs and flattened by making three or four radial incisions from the outside of the retina inwards, with the deepest cut in the nasal pole. Immunodetection was carried out following standard procedures described in Ortín-Martínez et al., 2014. Briefly, retinas were permeated with two washes in PBS 0.5% Triton X-100 (Tx) and frozen for 15 min at −70°C in 100% methanol. Retinas were then thawed at room temperature, rinsed twice in PBS-0.5%Tx and incubated overnight at 4°C in the appropriate mixture of primary antibodies diluted in blocking buffer (PBS, 2% normal donkey serum, 2%Tx). The next day, retinas were washed four times in PBS-0.5Tx and incubated for 2 hr at room temperature in secondary antibodies diluted in PBS-2%Tx. Finally, retinas were thoroughly washed four times in PBS-0.5%Tx and, after a last rinse in PBS, mounted scleral side up on slides in anti-fading solution (Prolong Gold Antifade, Thermofisher). For each species, IHC was performed on at least three retinas from two individuals (for details see Supplementary file 1).

IHC - antibodies and working dilutions 

Request a detailed protocol

The following primary antibodies were used: goat anti-OPN1SW, 1∶1000 (RRID: AB_2158332, sc-14363, Santa-Cruz Biotechnologies, Heidelberg, Germany; detects S-opsin protein) and rabbit anti-opsin red/green, 1∶750 (RRID: AB_177456, ab5405, Millipore Ibérica, Madrid, Spain; detects L-opsin protein). Sc-14363 is an affinity-purified goat polyclonal antibody raised against a 20-amino-acid synthetic peptide mapping within amino acids 1 to 50 of human blue-sensitive opsin, and AB5405 was raised in rabbit against the last 42 amino acids of the C-terminus of recombinant human red/green opsin (Gaillard et al., 2009). These antibodies have been used successfully in many groups, including rodents, artiodactyls, bats, and birds (e.g. Gaillard et al., 2009, Müller et al., 2007 and Nießner et al., 2016). The following secondary antibodies were used at a 1:500 dilution: donkey anti-goat Alexa Fluor 568 (RRID: AB_2534104) and donkey anti-rabbit Alexa Fluor 647 (RRID: AB_10891079) (Thermofisher). In addition, we created amino acid alignments of the peptide regions thought to correspond to the antibody epitopes across the bat species studied to assess sequence variation.

Quality control of retinal IHC

Request a detailed protocol

We used a strong quality control protocol to ensure that we could interpret an absence of labelling as a true loss of S-opsin protein. Given the variable age and preservation of museum specimens, we evaluated the anatomical preservation of the retina during dissection and excluded specimens (data not shown) if the retina was: (1) attached to the crystalline lens, poorly preserved, impossible to dissect/damaged, (2) highly fragile, poorly preserved, disintegrated/damaged upon dissection, and (3) intact or preserved in large pieces but exhibited shrinkage and/or an orange color characteristic of tissue degradation. When necessary, we also slightly modified the IHC protocol for some museum samples. Specifically, since museum samples where already permeabilized by their storage in ethanol, we reduced the number of PBS-0.5%Tx washes and removed the methanol permeabilization step at −70C. With these quality-control measures in place, and given the consistency of the detection of our chosen antibodies across all bats and other mammals (Müller et al., 2009; Müller et al., 2007; Ortín-Martínez et al., 2014), and the number of replicates and individuals we examined, we are confident in our interpretation that no labeling indicate a true loss of the respective cone type. In addition, we created amino acid alignments of the peptide regions thought to correspond to the antibody epitopes across the bat species studied to gain a measure of the sequence variation at these points.

IHC - photoreceptor quantification 

Request a detailed protocol

Flat-mounted retinas were photographed using a 20X objective on a confocal microscope (LSM710; Zeiss Microscopy). 564 and 633 lasers were used to excited Alexa 568 and Alexa 647 dyes, labelling S- and L- opsins, respectively. Each entire retina was completely imaged using 512 × 512 pixel tiles. For each retina, each tile was then Z-stacked and automatically counted using a 3D object counter plugin using Fiji (ImageJ). The accuracy of this automatic approach was verified by manually counting three biological replicates of five bat species, by two different people. For each retina quantified, the density was calculated for each tile and then averaged for each individual (total count was average over three individuals) and for each species (by averaging the average of the three individuals). The spatial distribution of L- and S-cone density was visualized for the following 14 species: Artibeus jamaicensis, Artibeus phaeotis, Carollia sowelli, Sturnira lilium, Monophyllus redmani, Erophylla sezekorni, Glossophaga soricina, Brachyphylla nana pumila, Desmodus rotundus, Pteronotus quadridens, Mormoops blainvillei, Macrotus waterhousii, Gardnerycteris crenulatum and Phyllops falcatus (see Figure 2, Table S2 in Supplementary file 2).

Opsin gene evolution

Request a detailed protocol

We used aligned sequences from the transcriptomes of 38 species together with those from six noctilionoid genomes (Zepeda Mendoza et al., 2018) to estimate rates of molecular evolution of visual opsin genes (OPN1SW, OPN1LW, and RHO) in focal bats. First, we tested for divergent selection modes among species that had S-opsin cones, lacked the S-opsin cones but had an intact mRNA sequence, and those that lacked the S-opsin cones but either did not have OPN1SW transcripts or had a pseudogenized OPN1SW sequence (Figure 5—figure supplement 1) using the Branch Model 2 of codeml in PAML 4.8a (Yang, 2007). Second, we applied the same approach to test divergent selection modes between frugivorous and non-frugivorous bat species (Figure 5—figure supplement 1; gene alignments have been submitted to DRYAD http://dx.doi.org/10.5061/dryad.456569k).

Ecological correlates of cone presence and density

Request a detailed protocol

To determine whether cone phenotypes are explained by dietary specialization, we applied the hierarchical Bayesian approach implemented in the R packages MCMCglmm and mulTree (Guillerme and Healy, 2014; Hadfield, 2010), using a sample from the posterior distribution of phylogenies of New World noctilionoids grafted onto the phylogeny of bats (Rojas et al., 2016; Shi and Rabosky, 2015). We modeled S-cone presence with species as observations as function of diet represented by four variables (nspecies = nobservations = 50). Since all predictor variables correspond to the presence or absence of a given diet or roosting habit, the coefficients of the resulting models were used to compare the strength of the association between the OPN1SW genotype or phenotype and the ecological covariate. Since this modeling approach neither tests against a null hypothesis of no effect, nor assumes the point estimates –in this case the mans by ecological group– are stationary, there is no requirement to adjust for multiple comparisons (Gelman et al., 2012). Using the predictor variable from the best model for presence/absence of S-opsin cones, or the presence of S-cones as a factor, we then repeated this approach to explain L-cone density across individuals within species (nspecies = 14, nobservations = 33, see Supplementary Information). To normalize the response data (density), we transformed by taking the natural logarithm of the cone density estimate. These analyses took advantage of the hierarchical structure of observations of density replicates clustered within species, with estimates of variance between species (corresponding to the phylogenetic regression), and residual variance remaining between observations. Given this data design, the estimate of the mean density per-group (i.e. frugivory/non, or presence/absence of S-cones) accounts for both between and within species variance. The R code for all regression models is available from DRYAD http://dx.doi.org/10.5061/dryad.456569k

Appendix 1

Species sampling 

We obtained eye tissue samples from a total of 263 eyes from 232 individuals, representing 59 bat species from seven families. This includes three families from the focal Noctilionoidea superfamily and four closely related outgroup families. Specimens used for this study were either wild-caught animals or obtained from museum collections (see Supplementary file 1 for full species and permit information). Wild-caught specimens from 49 bat species were collected following the approved IACUC protocols and site-specific permits. Bats were sampled from wild populations, and caught using mist nests set in forests and/or at cave entrances. Animals were handled following IACUC and site-specific protocols to minimize stress and were euthanized using an excess of isoflurane. Fresh bat specimens used for RNA-Seq analyses (nRNA-Seq = 45) were sampled in the Dominican Republic, Peru and Costa Rica, and those for immunohistochemistry were sampled in the Dominican Republic, Puerto Rico, Belize, and Trinidad. Additional eye samples from 34 species were dissected from specimens from the American Museum of Natural History (AMNH). For immunohistochemistry, five bat species had replicates that were both wild-caught and from museum collections and exhibited the same phenotype, highlighting the robustness of the experiments. Due to preservation methods, the AMNH samples were only suitable for immunohistochemistry (IHC) and so were not included in the transcriptomic study (see Supplementary file 1 for AMNH specimen identification numbers).

RNA sequence analysis 

Shortly after death, intact eyes were excised and placed in RNAlater and incubated at 4°C overnight before being stored at −180°C in vapor-phase liquid nitrogen, or at −80°C in a freezer. Frozen eye tissues were placed in Buffer RLT with added Dithiothreitol (DTT) then homogenized using a Qiagen TissueLyser. Total RNA was then extracted using Qiagen RNeasy Mini kits following the manufacturer’s protocol. Following extraction, RNA integrity was assessed using an Agilent 2100 Bioanalyzer and RNA concentration was measured using a Qubit Fluorometer. Library preparation was performed using Illumina TruSeq RNA Sample Preparation v2, with 500 ng of total RNA used for each sample. Constructed libraries were pooled and sequenced on the NextSeq500 High Output Run (150 cycles) to give 2 × 75 base-pair (bp) paired-end (PE) reads at the Genome Centre, Queen Mary University of London. Using the above approach, we sequenced eye transcriptomes from 45 individuals, representing 38 bat species; including biological replicates for three species (Pteronotus parnellii: n = 4; Artibeus jamaicensis: n = 4 and Phyllops falcatus: n = 2).

We assessed the quality of the short-read data with FastQC v.0.11.5 (http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc). Raw reads were trimmed with Trimmomatic-0.35 (Bolger et al., 2014), with the following settings LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36. Cleaned reads were assembled into de novo transcriptomes with Trinity v.2.2.0 (Grabherr et al., 2011), using default parameters.

Opsin gene annotation and examination of CDS

We used a combination of transcriptomic and genomic data to establish if the coding sequences (CDSs) of the three focal genes (OPN1SW, OPN1LW and RHO) were intact, and whether or not the mRNA of these genes was expressed in the eyes of the bats under study. We used a reciprocal best hit blast approach to establish whether or not transcripts corresponding to the three visual pigments (OPN1SW, OPN1LW and RHO) were present in each of the assembled transcriptomes. Sequences representing the protein products encoded by 22,285 human protein coding genes were downloaded from Ensembl 86 (Yates et al., 2016), for each gene product only the longest protein sequence was retained. These sequences were then used as tblastn (blast +v0.2.2.29) queries against each of the 45 bat transcriptome databases, the top hit was kept with an e-value cut-off <1e−6. Reciprocal blasts were then carried out using blastx (blast +v0.2.2.29), with bat transcripts as queries against the human protein database, again only keeping the top hit and e-value <1e−6. Percentage coverage of each bat hit against the human protein was calculated with the perl script analyze_blastPlus_topHit_coverage.pl available in Trinity utils. Candidate coding sequences (CDSs) were then extracted from the transcriptome assembly based on the blast coordinates using a custom perl script.

For transcriptome assemblies in which the OPN1SW sequence was not initially recovered, we undertook a number of additional steps to confirm that the sequence was not present. Firstly, as de novo transcriptome assemblies can create erroneous chimeric transcripts that may affect reciprocal blast results we followed the approach of (Yang and Smith, 2013) to reduce the number of trans-self-chimeras. Briefly, this involves performing a blastx search of the bat queries against the human protein database with an e-value cutoff of 0.01. Hits that met the default parameters of identity ≥30% and length ≥100 base-pairs are used to then search for either self-chimeras or multi-gene chimeras. Detected putative chimeras are then cut into segments, and retained if >100 base-pairs. This approach is only able to screen for trans-chimeras. Following chimera detection, the initial reciprocal blast described above was repeated. Lastly, we manually re-blasted all sequences that were initially identified as matching OPN1SW, but did then not pass the stringent reciprocal blast procedure.

Verification of OPN1SW sequence with PCR

For three species we amplified the genomic region spanning exons 3–4 of OPN1SW, from DNA extracted from the same individuals used to generate the RNA-Seq data. DNA extractions were carried out using Qiagen DNeasy Blood and Tissue Kit (69504). We used previously published primers (Zhao et al., 2009a). PCR mixtures consisted of 12.5 μl EconoTAQ Master Mix 2x, 3 μl of each primer (10 μM), 4 μl of genomic DNA (>100 ng). On an Eppendorf Mastercycler ProS, PCR was carried out with a single cycle at 94°C for 2 min followed by 35 cycles of 94°C for 30 s, annealing temperature of 55°C for 40 s, 72°C for 1 min 30 s, and finally a single cycle of 72°C for 10 min. PCR products were visualized on a 1% TBE-agarose gel. PCR product was cleaned up using Agencourt AMPure XP and submitted for cycle sequencing. Sequences have been submitted to GenBank MK248618 - MK248630.

Opsin gene evolution

We obtained protein-coding sequences for OPN1LW and RHO from the transcriptomes of all 39 species and included sequence data for Eptesicus fuscus (from GenBank) and Phyllostomus discolor (mPhyDis1_v1.p.fasta, available from GenomeArk vgp.github.io), for a total of 41 species. For OPN1SW, we obtained sequence data for 31 species from the transcriptome data sets. In a number of cases coding sequences from multiple contigs were manually joined together to produce a full-length CDS. We then supplemented the OPN1SW nucleotide sequences with those extracted from genome data using a combination of blastn and bl2seq on five noctilionoid bat genomes (Artibeus jamaicensis, Desmodus rotundus (Zepeda Mendoza et al., 2018), Lionycteris spurrelli, Macrotus waterhousii, Mormoops blainvillei, Phyllostomus discolor and Noctilio leporinus) using the Miniopterus natalensis OPN1SW (XM_016213323.1) sequence as a query. The L. spurrelli and M. waterhousii genomes were sequenced by the Rossiter Lab, and the A. jamaicensis, M. blainvillei and N. leporinus genomes were made available by the Broad Institute, and P. discolor as above. We also obtained the OPN1SW sequences from E. fuscus from GenBank. The extracted and aligned sequences are available from DRYAD (http://dx.doi.org/10.5061/dryad.456569k).

Sequences for OPN1LW and RHO were aligned using MUSCLE v3.8.425 (Edgar, 2004) as translated amino acids to keep the sequences in frame. The software implementation of our model requires the alignment to be in frame without any stop codons, therefore, stop codons at the end of the reading frame were removed from the alignment. Due to stop codons and indels, nucleotide sequences for OPN1SW were aligned by eye, with columns containing disruptions to the reading frame being removed to keep the remaining sequences in frame. Hypervariable regions at the beginning or end of sequences that may be caused assembly errors were masked by ‘Ns’ in two cases (Phyllops falcatus and Phyllonycteris poeyi). Additionally, premature stop codons were masked with ‘Ns’ and columns containing insertions that shifted the translation frame were deleted to keep codons in frame.

We tested for whether there were differences in rates of molecular evolution in the three opsin genes by estimating the ratio of the rates of nonsynonymous to synonymous substitutions (ω) for different branch classes. We set up two frameworks: S-cone presence and diet. When looking at S-cone presence variation, in our 2-branch class test, we estimated differences for branches that lacked the S-cone protein (ωS-cone.absent), and those that had the S-cone protein present (ωbackground). We also designed a three branch-class test in which we estimated different rates for bats with S-cones (ωbackground), bats that lack S-cones but have an intact reading frame for the OPN1SW transcripts (ωOPN1SW.intact), and bats that lack S-cones and OPN1SW is a pseudogene (ωOPN1SW.pseudo). If bats that lack the S-cone experience relaxed selection, we expect higher rates of ω in bats without the S-cone in the OPN1SW gene, but no differences among groups in the other two opsins. The sequences from lineages for which no S-cone data was available were removed from this analysis (n = 8). Finally, we tested if there were differences in rates in frugivorous lineages (ωfrugivore) and all other bats (ωbackground). Figure 5—figure supplement 1 depicts branches labeled with respective branch classes. Frugivory data was available for all lineages, and thus all available sequences were used in this analysis.

These analyses were performed using the branch model implemented in the codeml routine of PAML 4.8a (Yang, 2007). Differences among branches were compared against estimates for a single ω for all branches (ωbackground). We used a likelihood ratio test to compare the best-fit model for each opsin gene. The analysis used the species topology that merged a recently published phylogeny of all bats (Shi and Rabosky, 2015) with that of a recently published noctilionoid tree (Rojas et al., 2016). The tree was trimmed using the geiger v. 2.0.6 package in R (Harmon et al., 2008).

Ecological correlates of cone presence and density 

A hierarchical Bayesian approach was used to relate the ecological factors to the presence of an OPN1SW ORF, mRNA, or S-cone while accounting for the phylogenetic correlation between observations from different species. A hierarchical approach is often called a mixed model in the literature, with cluster-specific effects called ‘random’, and sample-wide effects called ‘fixed’. As different fields apply ‘random’ and ‘fixed’ to different levels of the hierarchy, here we adopt the language of cluster-specific and sample-wide effects (Gelman, 2005). The effect of species was quantified by including species as a cluster-specific, or random effect in the R package MCMCglmm (Hadfield, 2010). Additionally, to address variation among different estimates of phylogeny, we used the R package mulTree (Guillerme & Healy, 2014) to run the Bayesian models across a sample of trees obtained from the posterior of phylogenies.

The first set of models explained presence or absence of an OPN1SW ORF, mRNA, or S-cone as a function of one in a series of ecological dummy variables coded as prevalence or non-prevalence of plant materials, fruit, or other (insects or small vertebrates) items in diet, or whether known roosts included caves or not. In the sample-wide or fixed portion of these models, observations y for each species from one to i for each ecological category one to j correspond to a single-trial binomial response of the probability of observing the OPN1SW genotype or phenotype given by pri such that:

yidbern(pri)
logit(pri)=a+b.ecology[ecologyj]

Models for each of the dietary categorizations were then compared using the estimated coefficients. For dummy predicators indicating presence/absence, the absolute coefficient on the presence indicates the strength of the association with the response (in this case the genotype or phenotype). The predictor variable identified as the most strongly associated with the presence/absence of S-cones was then used in subsequent analyses of cone density.

Analyses of the sample-wide or fixed portion of cone density modeled the natural logarithm of the cone density y for each observation i as a function of dummy predictor variables defined by an diet group j, or S-cone group k. ln(yi) was modeled as a random, normally distributed variable with mean mu and variance, as below:

ln(yi)dnorm(mu,variance)
ln(yi)=a+b.diet[dietj]
ln(yi)=a+b.S.cone[S.conek]

Unlike the presence/absence analyses, these response variables were normally distributed, with the sample-wide portion of the model accounting for the effect of diet or S-cone presence on L-cone density. The cluster-specific or random effect accounted for both the relationships between species and the clustering of observations when more than one measurement was taken for each species.

To estimate the covariation arising from phylogeny for all species analyzed, we used the phylogeny of Shi and Rabosky (2015) for non-noctilionoids and as a base tree. For New World noctilionoids, a posterior sample of 100 trees from the phylogenies of Rojas et al., 2016 was grafted onto the base tree and, after rate smoothing using the chronopl routine in ape Paradis et al., 2004, used as input in mulTree analyses. Each regression ran for 20M generations, sampling every 1000 generations, with a burn-in of 100,000. Each regression ran two separate chains, assessing post-burnin convergence by comparing posteriors and reaching estimated sampling sizes (ESS) of at least 200 for all model parameters.

Data on the dietary ecology of the species sampled were obtained from a recent study of trophic ecology (Rojas et al., 2018) for noctilionoids. Non noctilionoids in the subfamily Yangochiroptera are predominantly insectivorous and were coded as such. To evaluate the influence of roosting ecology on the presence of an OPN1SW ORF, mRNA, or S-cone, we used data from two recent studies (Garbino and Tavares, 2018, Voss et al., 2016), in addition to unpublished data for neotropical species made available by Danny Rojas.

MCMCglmm uses inverse Wishart distributions for priors on sample-wide variance and cluster-specific or phylogenetic variance. For uncorrelated predictor variables, these functions collapse into inverse gamma distributions for residuals. The choice of the residual prior is particularly important for phylogenetic logistic regressions (Ives and Garland, 2014), for which this variance is not identified in the likelihood (Hadfield, 2010). A problem arises when estimating residual variance from binary data such as presence/absence with a single trial (in this case, most species have only one observation). While there may be heterogeneity in the underlying probability of the trial, this heterogeneity is unobserved and cannot be estimated with one trial only. For this reason, many logistic regression implementations set the residual variance at zero, but this is an arbitrary choice (Hadfield, 2016). Instead, here we use an approach to the prior variances validated by previous comparisons of Bayesian logistic regressions (Yohe and Dávalos, 2018). Briefly, we set the residual variance at one (1), and allow for a flexible prior on the variance of the phylogenetic effect. In MCMCglmm notation, this prior is given by:

list(R=list(v=1fix=1),G=list(G1=list(V=1,nu=1000,alpha.mu=0,alpha.V=1)))

As the phylogenetic structure of the data corresponds to a matrix structure of correlations between observations, it becomes necessary to expand the parameters by specifying both the mean and covariance matrix on the prior, in addition to the shape and scale parameters given by nu/2. For the logistic regressions, the prior on the phylogenetic structure was given by nu = 1000 and V = 1 (generating a very long-tailed distribution), with prior mean of 0 and covariance of 1.

Unlike most standard applications of MCMCglmm, our analyses of binary responses (i.e., logistic regressions of OPN1SW ORF, mRNA, or S-cone on various predictors) had a fixed residual variance. This produces errors in mulTree, which uses ESS and potential scale reduction factors to determine convergence. Therefore, we modified the mulTree_fun.R R script in mulTree to overlook residual variation in its evaluation. We include this modified function in DRYAD.

The prior above, however, does not account for boundaries on coefficients. These become necessary when there is complete separation in the response variable. In our analyses, no fruit-eating species lacked mRNA, and no species roosting outside caves lacked mRNA. This complete separation yielded high coefficients for both predictors, but with very high standard errors, generating non-convergent Bayesian chains, as there is no way to calibrate the model. For this reason, we placed prior boundaries on the coefficients by adding another element to the prior for these two regressions. This element centered the prior coefficient on zero (no effect), with a variance of nine or three standard deviations. Hence the prior in these two cases was given by:

list(R=list(v=1,fix=1),G=list(G1=list(V=1,nu=1000,alpha.mu=,alpha.V=1)),B=list(mu=rep(0,2),V=diag(9,2)))

In contrast to binary responses, it is feasible to estimate the residual variance of normally distributed responses such as density. Hence, for Gaussian regressions, the prior on the residuals was given by nu = 1 and V = 1 (Hadfield, 2016), or:

list(R=list(V=1,nu=1),G=list(G1=list(v=1,nu=1000,alpha.mu=0,alpha.V=1)))

Finally, we modified a script by Smith et al. (2016) to estimate the variance explained by the sample-wide (or fixed) effects in the models. This is also known as the conditional R2, and is reported in model tables.

RNA quality and sequencing statistics 

Integrity of extracted RNA varied across samples (RIN 4.1 to 10), the majority of samples obtained RINs greater than the recommended value of 8. Following sequencing, the number of raw reads ranged from 13,221.073 to 47,639,831 per sample. Cleaned reads ranged from 12,682,257 to 45,280,391 per sample, which resulted in assemblies of 67,459 to 131,925 across species. For several species the OPN1SW transcript was recovered following chimera removal as the transcript was initially joined to that of CALU.

Immunohistochemistry

We used IHC to assess the presence or absence of OPN1SW and OPN1LW proteins in whole, flat-mounted retinas of adult bats (Figure 2 and Figure 1—figure supplements 1 and 2). We detected OPN1LW proteins in the adult eyes of all sampled bat species (n = 56 species total), whereas we detected the OPN1SW protein in the eyes of only 32 bat species. The following species lack the OPN1SW protein: Molossus molossus, Eptesicus fuscus, Chilonatalus micropus, Pteronotus davyi, P. parnellii, Mormoops blainvillei, Macrotus waterhousii, Diaemus youngi, Desmodus rotundus, Trachops cirrhosus, Tonatia saurophila, Gardnerycteris crenulatum, Monophyllus redmani, Erophylla sezekorni, E. bombifrons, Brachyphylla pumila, Lonchophylla robusta, and Carollia brevicauda. Specimens assayed for Phyllostomus hastatus, Sturnira tildae, S. ludovici, and Platyrrhinus dorsalis, were obtained from museum specimens. While we detected the OPN1LW protein in these samples, they were characterized by a low signal-to-background ratio in the OPN1SW protein labeling, which prevented us from determining OPN1SW presence or absence.

In most species examined either both the OPN1SW transcript and protein were detected (n = 32), or neither were detected (n = 5). For example, all members of the Stenodermatinae clade of fruit-eating bats examined were found to have both the OPN1SW transcript and protein. Outside of this clade, we also detected the OPN1SW protein, and in most cases confirmed transcript presence with RNA-Seq, in species distributed widely throughout the phylogeny. Including within Emballonuridae (Saccopteryx bilineata and S. leptura), Noctilionidae (Noctilio leporinus), Mormoopidae (Pteronotus quadridens), and other Phyllostomidae (Phyllostomus elongatus, Anoura geoffroyi, Glossophaga soricina, Carollia perspicillata, Rhinophylla fischerae, and R. pumilio).

We detected the OPN1SW transcript, but no protein, in fewer species. These species are also widely distributed throughout the phylogeny and include the following: Molossus molossus, Pteronotus parnellii, Tonatia saurophila, Gardnerycteris crenulatum, Monophyllus redmani, Erophylla bombifrons, and Carollia brevicauda.

Opsin gene evolution

The alignments used for input for the PAML analyses resulted in 349 codons for OPN1SW, 364 for OPN1LW, and 348 for RHO. For the analysis testing the difference in ω rates between species that do and do not express the S-cone protein, there was no difference in ω estimates between branch classes for the RHO2(1)=0.34, p=0.56; χ2(2)=0.63, p=0.73). However, there was a difference favoring the three-branch class model for OPN1SW (ωbackground = 0.13; ωOPN1SW.intact=0.24; ωOPN1SW.pseudo=0.78; χ2(2)=70.99, p=3.84e-16) and OPN1LW (ωbackground = 0.08; ωOPN1SW.intact=0.08; ωOPN1SW.pseudo=0.19; χ2(2)=9.18, p=0.01) gene (Table S2 in Supplementary file 2). For the analysis testing for difference in ω rates between frugivorous species and background branches, there was no difference in ω estimates between branch classes for the OPN1SW2(1)=0.88, p=0.35) and OPN1LW2(1)=0.38, p=0.54) genes (Table S3 in Supplementary file 2). However, there was a difference in ω for the RHO gene (Table S3 in Supplementary file 2), in frugivorous lineages showed significantly lower rates than background branches (ωbackground = 0.04; ωfugivory= 0.01; χ2(1)=13.770.0, p=2.07e-4).

Ecological correlates of cone presence and density

Tables S5-S8 and Figure 4 summarize the results from Bayesian regressions. The frequency distributions of the coefficients on ecological covariates for 12 phylogenetic regressions of OPN1SW ORF, mRNA, or S-cone presence against diet prevalence or cave roosting reveal the prevalence of fruit as a predictor of cone presence is the strongest covariate of any model (Figure 4). The posterior estimates of parameters for various models model reveal the prevalence of fruit in diet increases the odds of having the S-cone by a factor of 39 (given by e^coefficient of fruit prevalence, or ~39x higher odds of having S-cones for lineages that eat fruit than those that do not, Table S7 in Supplementary file 2). Analyses of the density of long-wave cones revealed no statistically meaningful effect of fruit prevalence in diet on this density (Table S8 in Supplementary file 2). Instead, analyses of the density of long-wave cones as a function of the presence of S-cones revealed having the S-cones increases the density of L-cones by 0.43 in the natural logarithm scale (or a factor of ~1.54 in the linear scale) compared to species without S-cones (or from a baseline of ~3944 to~6063, Table S8 in Supplementary file 2).

https://doi.org/10.7554/eLife.37412.018

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
    Non-image Forming Light Detection by Melanopsin, Rhodopsin, and Long-Middlewave (L/W) Cone Opsin in the Subterranean Blind Mole Rat, Spalax ehrenbergi: Immunohistochemical Characterization, Distribution, and Connectivity
    1. G Esquiva
    2. A Avivi
    3. J Hannibal
    (2016)
    Frontiers in Neuroanatomy, 10, 10.3389/fnana.2016.00061, 27375437.
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
    Parameterization and Bayesian Modeling
    1. A Gelman
    (2004)
    Journal of the American Statistical Association 99:537–545.
    https://doi.org/10.1198/016214504000000458
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
    mulTree: a package for running MCMCglmm analysis on multiple trees
    1. K Guillerme
    2. T Healy
    (2014)
    Zenodo, 10.5281/zenodo.12902.
  33. 33
  34. 34
    MCMC methods for multi-response generalized linear mixed models: The MCMCglmm R Package
    1. JD Hadfield
    (2010)
    Journal of Statistical Software, 33, 10.18637/jss.v033.i02.
  35. 35
    MCMCglmm Course Notes. Unpublished Manuscript Edinburgh
    1. JD Hadfield
    (2016)
    University of Edinburgh.
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
    Phylogenetic Regression for Binary Dependent Variables
    1. AR Ives
    2. T Garland
    (2014)
    Modern phylogenetic comparative methods and their application in evolutionary biology: concepts and practice, Phylogenetic Regression for Binary Dependent Variables, Heidelberg, Springer Berlin Heidelberg.
  44. 44
  45. 45
  46. 46
  47. 47
    The photoreceptor populations in the retina of the greater horseshoe bat Rhinolophus ferrumequinum
    1. TJ Kim
    2. YK Jeon
    3. JY Lee
    4. ES Lee
    5. CJ Jeon
    (2008)
    Molecules and Cells 26:373–379.
  48. 48
    Enhancing the visibility of fishing ropes to reduce right whale entanglements
    1. S Kraus
    2. J Fasick
    3. T Werner
    4. P McFarron
    (2014)
    Report to the Bycatch Reduction Engineering Program (BREP), National Marine Fisheries Service, Office of Sustainable Fisheries.
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
    Diversity of photoreceptor arrangements in nocturnal, cathemeral and diurnal malagasy lemurs
    1. L Peichl
    2. A Kaiser
    3. F Rakotondraparany
    4. RR Dubielzig
    5. SM Goodman
    6. PM Kappeler
    (2017)
    Journal of Comparative Neurology, 19, 10.1002/cne.24167.
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
    As blind as a bat? opsin phylogenetics illuminates the evolution of color vision in bats
    1. BF Simões
    2. NM Foley
    3. GM Hughes
    4. H Zhao
    5. S Zhang
    6. SJ Rossiter
    7. EC Teeling
    (2018)
    Molecular Biology and Evolution, 10.1093/molbev/msy192, 30476197.
  86. 86
  87. 87
  88. 88
  89. 89
  90. 90
  91. 91
  92. 92
  93. 93
    Exploring the genome and transcriptome of the cave nectar bat Eonycteris spelaea with PacBio long-read sequencing
    1. M Wen
    2. JHJ Ng
    3. F Zhu
    4. YT Chionh
    5. WN Chia
    6. IH Mendenhall
    7. BP Lee
    8. AT Irving
    9. LF Wang
    (2018)
    GigaScience, 7, 10.1093/gigascience/giy116, 30247613.
  94. 94
  95. 95
  96. 96
  97. 97
  98. 98
  99. 99
  100. 100
  101. 101
  102. 102
  103. 103
  104. 104
  105. 105
  106. 106
  107. 107
  108. 108

Decision letter

  1. Patricia J Wittkopp
    Senior and Reviewing Editor; University of Michigan, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: this article was originally rejected after discussions between the reviewers, but the authors were invited to resubmit after an appeal against the decision.]

Thank you for submitting your work entitled "Evidence for multifactorial processes underlying phenotypic variation in bat visual opsins" for consideration by eLife. Your article has been reviewed by Senior Editor, a Reviewing Editor, and two reviewers.

The first reviewer indicated the manuscript failed to provide sufficient new insight into relevant aspects of evolutionary biology. This reviewer also thought that the overall claim of association between S-opsin presence and frugivory is of interest but is overly speculative. The second reviewer criticised the extent of gene and protein expression data sets required to make these claims and questioned the validity of the ecological correlation. In the light of these comments we have decided to decline the submission.

Reviewer #1:

A frame disruption represents only one mutation that disables the functional consequence of a protein-coding gene. Other mutations, including promoter and regulatory element mutations, will be equally effective in disabling the gene. Sadier et al. use RNA-sequencing and immunohistochemistry to study the mRNA and protein products of two cone opsin genes (OPN1SW and OPN1LW) and rhodopsin among 38 noctilionoid bats. OPN1SW disruptions have been known within the suborder Yinpterochiroptera for some time. From Figure 4, DNA sequence predicts OPN1SW protein absence 6 times, RNA transcript predicts protein absence 5 times, and that for 5 species neither DNA sequence nor RNA transcript predicted protein absence. These 5 instances (Molossus molossus, Tonatia saurophila, Gardnerycteris crenulatum, Monophyllus redmani and Pteronotus parnellii) and a sixth, the discordance between DNA or RNA and protein for P. poeyi, form the basis to the authors' claim in the final sentence of the Abstract. As expected, the authors show significant relaxation of constraint for lineages in which OPN1SW has become a pseudogene. Data shown in Figure 5 are proposed to indicate that S-opsin presence is partially predictive of whether a bat species is frugivorous.

The authors have clearly provided new evidence that independent OPN1SW gene loss has occurred more widely in bats, outside of the Yinpterochiroptera. Such independent losses are, however, already known to occur throughout the mammals. They also make a valid point that their "findings reveal assessments of visual perception based purely on genotypic analyses of opsin sequences, or RNA transcripts, can be misleading, and may even obscure the evolutionary processes and ecological agents of selection". They argue that variation in S-cone presence can be explained, in part, by fruit consumption. It was unclear to me how many similar hypotheses (beyond plant or flower-visiting or insectivory) that had been considered in statistical tests by the authors, and thus whether a multiple testing correction needed to have been employed.

Comparing against the aims of the journal, I came to a conclusion that this manuscript did not provide a substantially new insight into this aspect of evolutionary biology. The speculative association between S-opsin presence and frugivory is of interest, but it will remain unclear whether other (multiple?) traits could equally well explain the authors' findings.

Abstract "maintenance [of S-opsin] relates to frugivory". Clarify that this correlation is not necessarily reflective of causality.

Abstract "Discordance between DNA, RNA, and protein". In essence there is no discordance at the level of these molecules, only in our interpretation of their sequences.

Introduction "we identify specific and diverse molecular mechanisms by which selection has acted". Really?

Subsection “Variation in opsin transcripts across taxa” "Finally, we detected multiple introns in the assembled transcripts of the sister taxa Erophylla bombifrons and Phyllonycteris poeyi, suggesting transcriptional readthrough." Transcriptional readthrough is due to aberrant polyadenylation of transcripts rather than splicing out of introns, so I did not understand this statement. From the two references provided (Gaidatzis et al., 2015; Vilborg andand Steitz, 2017), I think intergenic and intronic transcription are being confused. If there is intron retention, then is there evidence that this is due to mutations at the splice site consensus sequences?

Merge Figure 1, Figure 2 and Figure 4. OPN1LW and RHO data can be presented elsewhere, for example in a Supplemental Table or Figure.

The Discussion section is over-long and could be condensed in length.

Formally, the presence of protein is not always indicative of functionality. This should be stated.

Reviewer #2:

In the paper the authors set out to provide "evidence for multifactorial processes underlying phenotypic variation in bat visual opsins". More specifically focusing on the correlation between DNA, RNA and protein levels of opsin genes in the noctilionoid bats. Lastly, they link such changes to diverse feeding ecologies.

It is indeed known and accepted that the relationship between the expression level of transcripts and that of the downstream proteins is not fully understood and that studies so far have shown to vary across tissues and cell types. It is indeed an important question.

My main problem with the study presented here is that the authors have not provided any tangible evidence to actually understand further the comparison between sequences, gene expression and protein levels.

Transcript presence and gene expression:

There is no analysis of gene expression, just an analysis of transcript presence (supported by the Trinity assemblies). Of course, it would not be possible to analyse gene expression when only few of the species transcriptomes (3 out of 38 if I understand correctly) have been carried out in replicates. In addition, the lack of replicate data does hamper the extant transcript analysis as well. Furthermore, the integrity of some RNA used was RIN<7 (the lowest being 4.1) – if this is the case for some samples with the addition of the lack of replicates, this is a great concern. The lower end of reads per sample was 8.6 million – again raising concerns on power.

Protein Levels: I also find the use of the Histochemistry to assess the protein level correlation to transcript very weak. I would have like to see a mass spec/targeted proteomic approach and more rigorous analysis underpinning the correlation between transcript abundance (not quantified) and protein levels.

Ecological correlation: Again, this analysis is not possible with the current datasets and methodology. Number of replicates and their usage is highly confusing. Out of 14 species there were biological replicates for 11 however, in their table S4, three of those do not appear to have biological replicates (M. waterhousii, G. crenulatum and P. falcatus), also they mention two replicates for P. falcatus (n=2) transcriptomes but in Table S4 I only see one.

There could be a bias towards a fruit diet based on the number of species (and their diet) used for measuring correlation; 14 species used, made up of 33 bats total – some of these had double (2 species) or triple (1 species) diets. In total, 18 individuals had a fruit/partial fruit diet, 15 insect, 3 vertebrate, and 11 plants – could the correlation biased towards fruit? there are no details on normalisation.

Museum samples: I could not understand to what extent the museum samples were used to infer the above correlations. Some clarity is necessary.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for submitting your article "Evidence for multifactorial processes underlying phenotypic variation in bat visual opsins" for consideration by eLife. Your article has been reviewed by Patricia Wittkopp as the Senior Editor, a Reviewing Editor, and two reviewers. The following individual involved in review of your submission has agreed to reveal his identity: Todd Oakley (Reviewer #4).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This is an interesting, and timely, investigation of opsin expression in a diverse group of bats. It provides a more in-depth study of underlying mechanisms mediating Sws1 opsin function, and loss, across a large group of bats, than many other investigations to date. Moreover, it is an unusual study in that it combines not only gene and transcriptome sequencing, with immunohistochemistry, but that it uses both freshly sampled tissue as well as museum specimens. This gives their investigation an impressive sampling across a large and extremely diverse group of bats, which has allowed them to investigate interesting questions such as the association between diet and visual abilities.

Formal reviews of this manuscript were provided by two experts in the field, but I also consulted two additional experts informally. All four of these scholars were impressed by the scope of the dataset and agreed that the work addressed an important problem. They were split, however, on how well the novelty of the findings comes through in the manuscript. In preparing a revised version, please pay special attention to this issue. Suggestions for ways to consider modifying the manuscript to address this concern are provided below. Closer integration with the literature (including some recent unpublished studies) is expected to help address this point.

In addition, the second major issue raised by the reviewers dealt with the reader's ability to assess the reliability of the data, as detailed below. Related to this the authors should think critically – and write clearly – about absence of evidence versus evidence of absence, and what that means for their overall conclusions. Basically, their main conclusions relate to absence of expression – in what cases can you be confident in your inferences; and if you cannot be 100% certain, how does that impact your conclusions?

Essential revisions:

1) Immunohistochemistry on museum specimens. Some background and information is needed, in order to better assess their results. For example, how were these museum samples fixed and stored? For how long? References to other studies that have performed similar experiments successfully are required, in order to provide context for this study, and to assess the likelihood of degradation due to storage methods.

The authors mention that they compared museum specimens with fresh sample for a few species. "For immunohistochemistry, five bat species had replicates that were both wild-caught and from museum collections and exhibited the same phenotype, highlighting the robustness of the experiments." This is important data that needs to be included in the manuscript, with figures in the supplement. Is there any evidence of degradation in older samples? On a related issue, are some of the mismatches between DNA-mRNA-protein data as shown in Figure S3 attributable to degraded museum specimens? How was a negative result determined (i.e. "Assay -ve"), as opposed to a "failure" of the assay?

2) Antibody information. Despite a central role in the immunohistochemistry experiments, almost no information is provided concerning the antibodies used to detect the S- and L-cones. Is anything known about the antibody epitopes? In which species have they been used successfully in other published studies? Is it possible that protein sequence variation resulted in a lack of S-cone antibody binding in some of the cases? Background, and references to other studies in which these antibodies were used are necessary in order to provide context, and to assess how robust antibody detection might be across different species. Citing the original papers in which the antibodies were created, and the methods used to create them are necessary. Also providing information concerning how these antibodies were obtained (from a company, from another laboratory, etc.) is essential.

3) Redundancy of the data in the figures. The data presented in Figure 1, Figure 2 and Figure 4 are redundant. The same data for OPN1SW DNA, mRNA and protein is shown twice, and the information for diet is shown three times. I agree that these three figures should be collapsed to one, to avoid this redundancy.

4) Lack of caveats and alternate interpretations of their data. In general, there is not enough discussion of caveats and limitations of their experimental approaches, nor possible alternate interpretations of their data. A few examples of this follow.

Discussion section. "In the case of opsin 1, short wave sensitive (OPN1SW) gene, while the presence (or absence) of the transcript and protein was consistent across most species, there were also multiple exceptions." This is simply a statement of the results. More interpretation of the results needs to be provided, along with caveats and limitations.

Figure 6C. Intact ORF, but no mRNA nor protein detected. Was there any evidence for mutations in regulatory regions? Figure 6D. Intact ORF and mRNA but no protein detected. Is there any evidence for protein degradation if it is a museum specimen, or sequence variation leading to substitutions in the antibody epitope region? Any evidence for abnormal post-translational modifications of the protein, leading to its degradation within the photoreceptor? Some discussion of these possible alternate interpretations is warranted.

5) Update references. This is an area of extremely active research, with several studies of bat Opn1SW published this year that were not cited. These would include Kries et al., 2018 and Gutierrez et al., 2018, which expanded opsin sampling across Neotropical lineages, and Li et al., 2018, which provides the first in vitro experimental evidence of UV sensitivity in bat Sws1 opsin. Interpreting the results presented in the MS in the context of recently published papers would help to increase the significance of this study.

6) Although the dataset is impressive and represents a major accomplishment, my major concern is the importance of the results is not clearly articulated for the broad interdisciplinary readership of eLife. My critique is based on my own reading but concurs with critiques of two previous reviewers.

As an attempt at synthesis, I do see the authors articulate novelty mainly along two different lines: (1) they show transcripts may not always yield protein and (2) frugivory predicts presence of SW-sensitivity in bats. At some points in the manuscript, the authors argue for novelty of their study because they are studying a new group of vertebrates (they study recently diverged mammals, whereas others studied fishes or reptiles, or anciently diverged mammals). I find the last claim (new taxon) to be uncompelling because it does not articulate what general feature about evolution they learn.

Importantly I believe the authors did not well-synthesize their first two claims for generality. First, they make a methodological critique: In order to understand links between phenotype and genotype, we must study both transcripts and proteins. This story is not synthesized with their second claim, that bat frugivory predicts color vision. Because of lack of synthesis, and because we already know that transcripts don't always lead to proteins, I'm left feeling that the authors have not spent enough time to distill and communicate clearly their results.

Still, I could imagine some general syntheses, but these would require massive re-writing, synthesis, and distillation. Alternatively, a more discipline-specific journal might also be better. If the authors do believe the best story is a methodological warning that transcripts don't predict protein – they would need to show how ignoring protein data leads to wrong conclusions. For example, what if by ignoring their protein data, they arrive at a different conclusion – maybe using transcript data alone leads to a failure to find the correlation between frugivory and color vision? If so, this provides a clear example and cautionary tale of why we cannot simply assume the protein is there – at least for cases when loss of expression is involved. It would also synthesize the main general claims.

However, my own feeling is that methodological themes are usually less broadly interesting than learning something general about biology. So, what is general here? Well, the best I can come up with along those lines is this might be an example of convergent molecular pathways leading to convergent trait loss. The general evolutionary question is – are convergent losses underpinned by the same or different genetic mechanisms? They have some hints that different changes (stop codons, intron read-through, etc.) might interrupt opsin expression in different species (I didn't look carefully at the distribution of losses on the tree though). A paper that does a good job of setting up the general question of parallel loss (although morphology) is (Sumner-Rooney et al. 2016). A similar topic is also studied for flower color (Zufall and Rausher, 2004). Although parallel loss is fairly well studied, I don't think different failures of translation are known in parallel. I do see two main challenges to taking this approach though. (1) the authors will not know which mutations caused the failure to express proteins versus those that came later (2) I'd like to be convinced that the sequences found with RNA-seq can be replicated with PCR because transcriptome assembly can be error-prone.

Of course, the authors know the data far better than I do and so I am not really trying to dictate what the synthesis is (also my suggestions don't incorporate their dn/ds analyses). Rather, I provided two examples to try to articulate more clearly what I feel is lacking from the current manuscript.

This is a very valuable data set in its breadth. It is indeed rare to have transcripts and protein expression in broad comparisons. However, current descriptions of the importance of the work go in multiple directions, in my opinion leaving any one direction insufficient. I am open to the possibility that the writing (namely the synthesis and communication of the importance; sometimes called "novelty") could be improved, providing a story of very general interest across disciplines for eLife. But this would take a rather major re-write.

References

Sumner-Rooney L., Sigwart J.D., McAfee J., Smith L., Williams S.T. 2016. Repeated eye reduction events reveal multiple pathways to degeneration in a family of marine snails. Evolution. 70:2268-2295.

Zufall R.A., Rausher M.D. 2004. Genetic changes associated with floral adaptation restrict future evolutionary potential. Nature. 428:847-850.

https://doi.org/10.7554/eLife.37412.022

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

The first reviewer indicated the manuscript failed to provide sufficient new insight into relevant aspects of evolutionary biology. This reviewer also thought that the overall claim of association between S-opsin presence and frugivory is of interest but is overly speculative. The second reviewer criticised the extent of gene and protein expression data sets required to make these claims and questioned the validity of the ecological correlation. In the light of these comments we have decided to decline the submission.

Thank you for the opportunity to provide a rebuttal of the reviews on our paper ‘Evidence for multifactorial processes underlying phenotypic variation in bat visual opsins’. Between them the two reviewers raised three issues: (1) the novelty of the findings, (2) the extent of sampling and (3) the ecological models. We address these in turn below.

1) Novelty of findings

To our knowledge, while there have been numerous studies of opsin gene evolution in mammals, there has not yet been a single study that has attempted to relate coding sequences to the presence of transcripts and proteins in a comparative framework. Moreover, almost all studies of mammals have focused at deeper evolutionary timeframes. By revealing that opsin phenotypes are highly variable among even closely related bat species (sometimes even between congeners) within a single monophyletic clade, and that this variation stems from diverse molecular routes involving each step of the central dogma, we feel we are making a significant contribution to our understanding of the evolution of color vision across mammals.

Furthermore, it has not previously been shown that mammalian species that possess seemingly intact opsin gene sequences may still lack the corresponding cone proteins. Therefore, this finding suggests we may be underestimating phenotypic variation in mammalian color vision across species, which could have wider implications for divergent taxonomic groups such as primates.

2) The extent of sampling

Reviewer 2 raised concerns regarding the extent of sampling, in terms of biological replicates, for the gene expression and protein components of the study. While we agree it would be desirable to sample larger numbers of biological replicates to estimate variance, as is routine in studies of insects and lab animals, this was not possible due the nature of our study (i.e. on wild mammals – some rare – that required numerous ethical and research permits in order to sample). Additionally, there is a hard ceiling on the number of offspring most bats can have limiting their population recovery. For this reason, as well as the many external threats to wild populations, permitting agencies are loath to grant collecting permits for many individuals unless a population is known to be large, common, and unthreatened. We suspect that with the possible exception of some fishes, the same restrictions would apply to most other wild vertebrates. For this reason, we feel that enforcing comparable standards is not practical and would inevitably limit knowledge to restricted groups of model taxa.

Our study group of bats are arguably unparalleled among mammal in their ecological diversity within a single superfamily, and thus are highly suited for the broad comparative approach we opted to take. Increasing technical and biological replicates would have come with the cost of reduced taxonomic sampling which would ultimately have limited the ecological component of our study and thus also the significance of our findings. We are confident that our findings will have implications for other key groups that have been the focus of opsin studies (including primates, which would be even more difficult to sample for RNA).

3) Ecological models

To address these questions, we have added additional information and explanation to the text detailing these methods. The hierarchical Bayesian approach we adopted to analyse the relationship between diet and opsin phenotype is widely used in evolutionary ecological studies of trait evolution (e.g. Wagner et al., (2012); Lukas andand Clutton-Brock, (2013).

The evolutionary models we use are based on a Bayesian hierarchical format (and not a classical, frequentist approach), therefore, we are not testing a null hypothesis of no effect but instead modelling the distribution of an effect, whatever its size. Therefore, there is not the same requirements for adjustments following the analyses of multiple hypotheses, we detail this further below.

Reviewer #1:

A frame disruption represents only one mutation that disables the functional consequence of a protein-coding gene. Other mutations, including promoter and regulatory element mutations, will be equally effective in disabling the gene. Sadier et al. use RNA-sequencing and immunohistochemistry to study the mRNA and protein products of two cone opsin genes (OPN1SW and OPN1LW) and rhodopsin among 38 noctilionoid bats. OPN1SW disruptions have been known within the suborder Yinpterochiroptera for some time. From Figure 4, DNA sequence predicts OPN1SW protein absence 6 times, RNA transcript predicts protein absence 5 times, and that for 5 species neither DNA sequence nor RNA transcript predicted protein absence. These 5 instances (Molossus molossus, Tonatia saurophila, Gardnerycteris crenulatum, Monophyllus redmani and Pteronotus parnellii) and a sixth, the discordance between DNA or RNA and protein for P. poeyi, form the basis to the authors' claim in the final sentence of the Abstract. As expected, the authors show significant relaxation of constraint for lineages in which OPN1SW has become a pseudogene. Data shown in Figure 5 are proposed to indicate that S-opsin presence is partially predictive of whether a bat species is frugivorous.

The authors have clearly provided new evidence that independent OPN1SW gene loss has occurred more widely in bats, outside of the Yinpterochiroptera. Such independent losses are, however, already known to occur throughout the mammals. They also make a valid point that their "findings reveal assessments of visual perception based purely on genotypic analyses of opsin sequences, or RNA transcripts, can be misleading, and may even obscure the evolutionary processes and ecological agents of selection".

We are pleased reviewer 1 recognises that our findings are valid. We further clarify the novelty and significance of our findings in terms of both bat and mammalian evolution below.

They argue that variation in S-cone presence can be explained, in part, by fruit consumption. It was unclear to me how many similar hypotheses (beyond plant or flower-visiting or insectivory) that had been considered in statistical tests by the authors, and thus whether a multiple testing correction needed to have been employed.

This would make sense if we were using a classical, frequentist approach, but we are using Bayesian hierarchical models. Unlike classical statistics, the Bayesian approach we use has two advantages. First, we are not testing a null hypothesis of no effect but instead modelling the distribution of an effect, whatever its size. Second, a hierarchical model shifts points estimates (i.e., the mean by group) and their confidence intervals toward each other and thus does not suffer from the stationarity of point estimates and the need to widen intervals in multiple comparisons that emerges from classical statistics. A direct quote from Gelman et al., (2012) elaborates on these points:

“First, we are typically not terribly concerned with Type 1 error because we rarely believe that it is possible for the null hypothesis to be strictly true. Second, we believe that the problem is not multiple testing but rather insufficient modeling of the relationship between the corresponding parameters of the model. Once we work within a Bayesian multilevel modeling framework and model these phenomena appropriately, we are actually able to get more reliable point estimates. A multilevel model shifts point estimates and their corresponding intervals toward each other (by a process often referred to as “shrinkage” or “partial pooling”), whereas classical procedures typically keep the point estimates stationary, adjusting for multiple comparisons by making the intervals wider (or, equivalently, adjusting the p values corresponding to intervals of fixed width).” From Gelman et al. (2012). Why We (Usually) Don't Have to Worry About Multiple Comparisons. Journal of Research on Educational Effectiveness 5(2): 189-211.”

We have also clarified how many models were applied in the Materials and methods section and Results section.

Comparing against the aims of the journal, I came to a conclusion that this manuscript did not provide a substantially new insight into this aspect of evolutionary biology. The speculative association between S-opsin presence and frugivory is of interest, but it will remain unclear whether other (multiple?) traits could equally well explain the authors' findings.

As highlighted in our paper much work has indeed been performed on assessing OPN1SW gene loss both in bats and other mammals (e.g. Emerling et al., 2015; Emerling et al., 2017; Zhao et al., 2009). Within bats, it has only previously been shown that some species possess pseudogenized OPN1SW these have all been inferred based on the presence of disrupted ORFs and the presence of STOP codons. Indeed, at the same time as our paper was submitted for review high impact studies are being published (e.g. Gutierrez et al., (2018) that are continuing to infer functional S-cones based on the presence of mRNA. Furthermore, we are able to show that several species in which OPN1SW has been inferred to be functional (e.g. Pteronotus parnellii as published in Emerling et al., 2015 and Gutierrez et al., 2018) is in fact not expressed as a protein.

Therefore, while it has been suggested that OPN1SW gene loss has occurred across mammals we are able to show that the prevalence of this is actually much higher than has previously been considered and is still happening (in the process of been lost?). Therefore, we are presenting data which provides a major new insight into the true patterns of short-wave sensitive vision across mammals, tnat could also be applied to many other genes and taxa. Together, these data are within the aims of the journal.

Our model systems of ecologically diverse, yet phylogenetically related, neotropical bat species is able to exemplify this pattern by highlighting the short evolutionary time periods that the diversity in colour vision has evolved. Previous accounts of S-opsins in mammals have often been linked to changes in photic environment e.g. shifts to aquatic/subterranean environments, therefore, our study also provides new insights into the potential drivers of OPN1SW gene loss as we see differences in its retention in closely related species that inhabitant identical terrestrial habitats.

Abstract "maintenance [of S-opsin] relates to frugivory". Clarify that this correlation is not necessarily reflective of causality.

This sentence has been removed from the current version, and we have reworded other sections.

Abstract "Discordance between DNA, RNA, and protein". In essence there is no discordance at the level of these molecules, only in our interpretation of their sequences.

We are unsure of what this means. There is discordance when the putatively functional sequence is present and yet its protein product is not. We now clarify what is meant in the Abstract.

Introduction "we identify specific and diverse molecular mechanisms by which selection has acted". Really?

To avoid confusion, we have rephrased the sections of the manuscript where the term ‘mechanism’ was used.

Subsection “Variation in opsin transcripts across taxa” "Finally, we detected multiple introns in the assembled transcripts of the sister taxa Erophylla bombifrons and Phyllonycteris poeyi, suggesting transcriptional readthrough." Transcriptional readthrough is due to aberrant polyadenylation of transcripts rather than splicing out of introns, so I did not understand this statement. From the two references provided (Gaidatzis et al., 2015; Vilborg and Steitz, 2017), I think intergenic and intronic transcription are being confused. If there is intron retention, then is there evidence that this is due to mutations at the splice site consensus sequences?

We thank the reviewer for highlighting our error here, we had indeed through an oversight used the incorrect term and references at this point.

Merge Figure 1, Figure 2 and Figure 4. OPN1LW and RHO data can be presented elsewhere, for example in a Supplemental Table or Figure.

We have opted to present the datasets separately in this way to may it clearer exactly which data points correspond to each particular part of the study. However, if the consensus agreement across reviewers and editors is to merge the figures, we are happy to do so.

The Discussion section is over-long and could be condensed in length.

We have shortened the Discussion section by two paragraphs.

Formally, the presence of protein is not always indicative of functionality. This should be stated.

We now state this in the results of IHC.

Reviewer #2:

In the paper the authors set out to provide "evidence for multifactorial processes underlying phenotypic variation in bat visual opsins". More specifically focusing on the correlation between DNA, RNA and protein levels of opsin genes in the noctilionoid bats. Lastly, they link such changes to diverse feeding ecologies.

It is indeed known and accepted that the relationship between the expression level of transcripts and that of the downstream proteins is not fully understood and that studies so far have shown to vary across tissues and cell types. It is indeed an important question.

We are pleased that the reviewer recognises that the motivation for our study is as an important issue.

My main problem with the study presented here is that the authors have not provided any tangible evidence to actually understand further the comparison between sequences, gene expression and protein levels.

Levels of opsin transcripts are known to fluctuate with circadian rhythm and exposure to light around the time of sampling, and thus assessing variation in expression levels would require biological replicates to account for sample variance. While this is desirable experimentally, it would mean euthanizing many more individuals and is thus not feasible for wild mammals, either ethically or in terms of obtaining permissions. By measuring the presence or absence of opsin transcripts we provide a more robust result given these limitations. See below for more details.

Transcript presence and gene expression:

There is no analysis of gene expression, just an analysis of transcript presence (supported by the Trinity assemblies). Of course, it would not be possible to analyse gene expression when only few of the species transcriptomes (3 out of 38 if I understand correctly) have been carried out in replicates. In addition, the lack of replicate data does hamper the extant transcript analysis as well. Furthermore, the integrity of some RNA used was RIN<7 (the lowest being 4.1) – if this is the case for some samples with the addition of the lack of replicates, this is a great concern. The lower end of reads per sample was 8.6 million – again raising concerns on power.

RIN scores vary across samples due the challenging nature of sample collection, but if the quality of the RNA determined the detection of OPN1SW transcripts, then species lacking transcripts would also tend to have low RIN. Instead, we find no relationship between RIN and presence of OPN1SW. For example, the species with the lowest RIN (Pteronotus quadridens 4.1) was found to contain a nearly intact OPN1SW mRNA transcript. The same is true for reads per sample. Additionally, in all bats sampled we were able to recover complete or nearly complete OPN1LW and RHODOPSIN sequences regardless of variation in RIN or sequencing depth. For the three species that we do have biological replicates we have consistent results in terms of mRNA presence/absence across all three visual pigments. We can add measures of transcriptome completeness (e.g. BUSCO scores) to the paper if required.

Protein Levels: I also find the use of the Histochemistry to assess the protein level correlation to transcript very weak. I would have like to see a mass spec/targeted proteomic approach and more rigorous analysis underpinning the correlation between transcript abundance (not quantified) and protein levels.

While documenting the presence or absence of either transcript or protein implies some level of quantitative correspondence, correlating the levels of mRNA and protein is not within the scope of this study. We did not attempt to estimate protein level – we primarily aimed at documenting presence or absence of OPN1SW and OPN1LW. Given the contrasting methods necessary for preserving either mRNA or protein (RNAlater vs. 4% paraformaldehyde), and as it was only possible to assay protein and not mRNA from museum specimens, we did not use the same individual for the RNA-Seq and IHC studies. As different individuals were used for RNA/protein detection assessing correlations in terms of mRNA expression/protein level are beyond the scope of the study. Furthermore, a number of studies suggest that there should not be a direct correlation between mRNA and protein level, for example, sometimes mRNA levels regulate gene expression etc.

Our estimates of S/L-cone density measures the distribution of the cones across the retina, so they correspond to protein levels only to the extent that having more cones indicates having more protein. For our purposes, density is more informative of how the distribution varies across the surface. A simple quantification in terms of protein level may be difficult to adjust given the size differences of the eyes.

Ecological correlation: Again, this analysis is not possible with the current datasets and methodology. Number of replicates and their usage is highly confusing. Out of 14 species there were biological replicates for 11 however, in their table S4, three of those do not appear to have biological replicates (M. waterhousii, G. crenulatum and P. falcatus), also they mention two replicates for P. falcatus (n=2) transcriptomes but in Table S4 I only see one.

There could be a bias towards a fruit diet based on the number of species (and their diet) used for measuring correlation; 14 species used, made up of 33 bats total – some of these had double (2 species) or triple (1 species) diets. In total, 18 individuals had a fruit/partial fruit diet, 15 insect, 3 vertebrate, and 11 plants – could the correlation biased towards fruit? there are no details on normalisation.

We have clarified the language in the text and the tables so it is clear when there are and there aren't biological replicates. Two kinds of regressions were conducted. The first one related the presence of cones to the diet of species. In this case each species was represented by one observation. There were 49 species, 29 of which included fruit in their diet.

The other type of regression modelled the density from 1-several individuals within species. In this case, we take advantage of the hierarchical structure of the data in which observations (individuals) cluster within species. The variance within and between species is both included in the model and modelled separately. Hence the model simultaneously captures variance between replicates (when present) and between species, when estimating the means corresponding to the groups of interest. We now indicate the natural log transformation, which served to normalize the density data.

For the 14 bat species that were used for the ecological correlates of opsin density models we state in the Material and methods section that “The accuracy of this automatic approach was verified by manually counting three biological replicates of five bat species, by two different people.”

Specifically, regarding the confusion surrounding the P. falcatus samples, two individuals were sampled for the transcriptome analyses (DR171 and DR003, as detailed in the RNAseq data tab of Supplementary table S1). Whereas, Table S4 presents the cone densities in the 14 species representative of different diet types and is based on IHC data and inferred dietary data only.

Museum samples: I could not understand to what extent the museum samples were used to infer the above correlations. Some clarity is necessary.

We have provided complete sample information in Supplementary file 1 and Figure 1—Figure Supplement 1, to clearly state which samples were wild caught and which were museum samples.

[Editors’ note: the author responses to the re-review follow.]

Summary:

This is an interesting, and timely, investigation of opsin expression in a diverse group of bats. It provides a more in-depth study of underlying mechanisms mediating Sws1 opsin function, and loss, across a large group of bats, than many other investigations to date. Moreover, it is an unusual study in that it combines not only gene and transcriptome sequencing, with immunohistochemistry, but that it uses both freshly sampled tissue as well as museum specimens. This gives their investigation an impressive sampling across a large and extremely diverse group of bats, which has allowed them to investigate interesting questions such as the association between diet and visual abilities.

Formal reviews of this manuscript were provided by two experts in the field, but I also consulted two additional experts informally. All four of these scholars were impressed by the scope of the dataset and agreed that the work addressed an important problem. They were split, however, on how well the novelty of the findings comes through in the manuscript. In preparing a revised version, please pay special attention to this issue. Suggestions for ways to consider modifying the manuscript to address this concern are provided below. Closer integration with the literature (including some recent unpublished studies) is expected to help address this point.

In addition, the second major issue raised by the reviewers dealt with the reader's ability to assess the reliability of the data, as detailed below. Related to this the authors should think critically – and write clearly – about absence of evidence versus evidence of absence, and what that means for their overall conclusions. Basically, their main conclusions relate to absence of expression – in what cases can you be confident in your inferences; and if you cannot be 100% certain, how does that impact your conclusions?

We thank the Reviewing Editor and the four reviewers for these constructive comments concerning our manuscript. We have made extensive revisions to our study following these recommendations, these include the addition of new figures to more clearly display our data and protocol, and considerable restructuring of the focus of the study to make it more accessible to readers from a wider range of subject. Please see our detailed response to the specific comments raised below.

Essential revisions:

1) Immunohistochemistry on museum specimens. Some background and information is needed, in order to better assess their results. For example, how were these museum samples fixed and stored? For how long? References to other studies that have performed similar experiments successfully are required, in order to provide context for this study, and to assess the likelihood of degradation due to storage methods.

We agree that more information regarding the museum specimens was needed in our paper. We have now added the following references: Hühns et al., 2015 and Nießner et al., 2016. The former describes the utility of museum samples for a range of molecular approaches, and the latter presents information on the use of the same S-opsin antibody across a range of mammal species, preserved by a range of methods. In addition, we added the collection date of the museum and field specimens used for the IHC assays to supplementary table S1, we also present data in Figure 1—figure supplement 4 to show that the staining is consistent across museum specimens (including the oldest specimen collected in 1921), and the IHC protocol is published in Ortín-Martínez et al., 2015, which we cite.

Museum samples were originally fixed in 70% ethanol and stored at room temperature. After dissecting out tissues from museum samples, we re-fixed in PFA and processed. All of these details have now been added to the manuscript.

The authors mention that they compared museum specimens with fresh sample for a few species. "For immunohistochemistry, five bat species had replicates that were both wild-caught and from museum collections and exhibited the same phenotype, highlighting the robustness of the experiments." This is important data that needs to be included in the manuscript, with figures in the supplement. Is there any evidence of degradation in older samples? On a related issue, are some of the mismatches between DNA-mRNA-protein data as shown in Figure S3 attributable to degraded museum specimens? How was a negative result determined (i.e. "Assay -ve"), as opposed to a "failure" of the assay?

We have added supplementary Figure 1—figure supplement 4 to display representative staining of both field caught and museum specimens from the same species and therefore, compare the consistency of the opsins detection.

We have added information regarding degradation of older samples to the manuscript. Basically, the oldest specimen we looked at had solid L- and S- cone staining so age in and of itself was not an issue. We did have instances in which we could detect L- but not S- cones however, that were probably due to preservation issues. To distinguish between real and false (non-)signals, we developed a stringent set of criteria based on tissue preservation state, and also completed staining on multiple samples and individuals whenever possible. We also now address the relationship between mismatches and the possible preservation/methodological issues explicitly in the manuscript. Given our sample sizes, methods, and tissue evaluation, we are confident that any non-signals (for species included in our analyses) represent real non-signals. However, we also explicitly discuss the possible issues with our approach and alternative interpretations in the manuscript.

2) Antibody information. Despite a central role in the immunohistochemistry experiments, almost no information is provided concerning the antibodies used to detect the S- and L-cones. Is anything known about the antibody epitopes? In which species have they been used successfully in other published studies? Is it possible that protein sequence variation resulted in a lack of S-cone antibody binding in some of the cases? Background, and references to other studies in which these antibodies were used are necessary in order to provide context, and to assess how robust antibody detection might be across different species. Citing the original papers in which the antibodies were created, and the methods used to create them are necessary. Also providing information concerning how these antibodies were obtained (from a company, from another laboratory, etc.) is essential.

The following primary antibodies were used: goat anti-OPN1SW, 1∶1000 (sc-14363, Santa-Cruz Biotechnologies, Heidelberg, Germany; detects S-opsin protein) and rabbit anti-opsin red/green, 1∶750 (ab5405, Millipore Ibérica, Madrid, Spain; detects L-opsin protein). AB5405 was raised in rabbit against the last 42 amino acids at the C-terminus of recombinant human red/green opsins. Sc-14363 is an affinity-purified goat polyclonal antibody raised against a 20-amino-acid synthetic peptide mapping within amino acids 1 to 50 of human blue-sensitive opsin. The antibodies were designed by commercial companies, and not for a specific study or manuscript. Both of these antibodies have been used successfully in many species, including but not limited to bats, rodents, artiodactyls, etc. This information has now been added to the manuscript. We also provide figures of the amino acid alignments for all the bats included in our study for which we have sequence data for in these regions so that the degree of conservation can be accessed by readers. Based on these alignments we are confident that lack of the S-opsin binding in the species which gave a negative result are not due to increased sequence variation.

3) Redundancy of the data in the figures. The data presented in Figure 1, Figure 2 and Figure 4 are redundant. The same data for OPN1SW DNA, mRNA and protein is shown twice, and the information for diet is shown three times. I agree that these three figures should be collapsed to one, to avoid this redundancy.

As suggested, we now present this data as a single figure.

4) Lack of caveats and alternate interpretations of their data. In general, there is not enough discussion of caveats and limitations of their experimental approaches, nor possible alternate interpretations of their data. A few examples of this follow.

Discussion section. "In the case of opsin 1, short wave sensitive (OPN1SW) gene, while the presence (or absence) of the transcript and protein was consistent across most species, there were also multiple exceptions." This is simply a statement of the results. More interpretation of the results needs to be provided, along with caveats and limitations.

We agree with the reviewers and have now included a new subsection “Study limitations and alternative interpretations”. In this section we have discussed whether some of our findings might have arisen as artifacts due to problems with museum specimens, data assembly, sample sizes and cryptic diversity. Although we acknowledge there are a small number of cases where artifactual results cannot be completely ruled out, many of the results are supported by other recent findings, and the overarching finding (of multiple routes and steps underpinning parallel degradation) does not change. In this section we have now included additional references to papers that support our data and interpretations.

Figure 6C. Intact ORF, but no mRNA nor protein detected. Was there any evidence for mutations in regulatory regions? Figure 6D. Intact ORF and mRNA but no protein detected. Is there any evidence for protein degradation if it is a museum specimen, or sequence variation leading to substitutions in the antibody epitope region? Any evidence for abnormal post-translational modifications of the protein, leading to its degradation within the photoreceptor? Some discussion of these possible alternate interpretations is warranted.

For the two species with putatively intact ORFs but with no mRNA or protein, genomic resources are currently unavailable so are not yet able to study the regulatory regions. This information has been added to the manuscript. For the cases where we find mRNA and no protein, we have now included additional evidence that suggests that in some species the ORF is in fact disrupted at the level of mRNA. We have also included information relating to the origin of the samples used for the protein assay and sequence conservation of the protein in the area of the epitope (see new Figure 1—figure supplement 2 and Figure 1—figure supplement 3). We agree that we needed to explain our quality control measures that we used to ensure the quality of our retinas before staining. We added a section in the Material and methods section which describe the different steps of our quality control to keep or discard a retina from our dataset. While protein degradation was not proven per se, we carefully inspected the anatomy of the retina to ensure its integrity before staining. In addition, we checked the morphology of the cones after staining. We have not found any evidence of mutation leading to protein degradation within the photoreceptor.

5) Update references. This is an area of extremely active research, with several studies of bat Opn1SW published this year that were not cited. These would include Kries et al., 2018 and Gutierrez et al., 2018, which expanded opsin sampling across Neotropical lineages, and Li et al., 2018, which provides the first in vitro experimental evidence of UV sensitivity in bat Sws1 opsin. Interpreting the results presented in the MS in the context of recently published papers would help to increase the significance of this study.

We thank the reviewers for highlighting these studies that have been published since we initially submitted our paper. These studies have now been referenced in our paper, and we have also been able to incorporate some of the data produced by these to confirm some of our findings, e.g. the sequences that we recovered based on transcriptome assembly match those produced by PCR in these studies (also see below). As pointed out due to the area of active research we have also included reference to Wu et al., 2018 and Simoes et al., 2018.

6) Although the dataset is impressive and represents a major accomplishment, my major concern is the importance of the results is not clearly articulated for the broad interdisciplinary readership of eLife. My critique is based on my own reading but concurs with critiques of two previous reviewers.

As an attempt at synthesis, I do see the authors articulate novelty mainly along two different lines: (1) they show transcripts may not always yield protein and (2) frugivory predicts presence of SW-sensitivity in bats. At some points in the manuscript, the authors argue for novelty of their study because they are studying a new group of vertebrates (they study recently diverged mammals, whereas others studied fishes or reptiles, or anciently diverged mammals). I find the last claim (new taxon) to be uncompelling because it does not articulate what general feature about evolution they learn.

Importantly I believe the authors did not well-synthesize their first two claims for generality. First, they make a methodological critique: In order to understand links between phenotype and genotype, we must study both transcripts and proteins. This story is not synthesized with their second claim, that bat frugivory predicts color vision. Because of lack of synthesis, and because we already know that transcripts don't always lead to proteins, I'm left feeling that the authors have not spent enough time to distill and communicate clearly their results.

Still, I could imagine some general syntheses, but these would require massive re-writing, synthesis, and distillation. Alternatively, a more discipline-specific journal might also be better. If the authors do believe the best story is a methodological warning that transcripts don't predict protein – they would need to show how ignoring protein data leads to wrong conclusions. For example, what if by ignoring their protein data, they arrive at a different conclusion – maybe using transcript data alone leads to a failure to find the correlation between frugivory and color vision? If so, this provides a clear example and cautionary tale of why we cannot simply assume the protein is there – at least for cases when loss of expression is involved. It would also synthesize the main general claims.

However, my own feeling is that methodological themes are usually less broadly interesting than learning something general about biology. So, what is general here? Well, the best I can come up with along those lines is this might be an example of convergent molecular pathways leading to convergent trait loss. The general evolutionary question is – are convergent losses underpinned by the same or different genetic mechanisms? They have some hints that different changes (stop codons, intron read-through, etc.) might interrupt opsin expression in different species (I didn't look carefully at the distribution of losses on the tree though). A paper that does a good job of setting up the general question of parallel loss (although morphology) is (Sumner-Rooney et al. 2016). A similar topic is also studied for flower color (Zufall and Rausher 2004). Although parallel loss is fairly well studied, I don't think different failures of translation are known in parallel. I do see two main challenges to taking this approach though. (1) the authors will not know which mutations caused the failure to express proteins versus those that came later (2) I'd like to be convinced that the sequences found with RNA-seq can be replicated with PCR because transcriptome assembly can be error-prone.

Of course, the authors know the data far better than I do and so I am not really trying to dictate what the synthesis is (also my suggestions don't incorporate their dn/ds analyses). Rather, I provided two examples to try to articulate more clearly what I feel is lacking from the current manuscript.

This is a very valuable data set in its breadth. It is indeed rare to have transcripts and protein expression in broad comparisons. However, current descriptions of the importance of the work go in multiple directions, in my opinion leaving any one direction insufficient. I am open to the possibility that the writing (namely the synthesis and communication of the importance; sometimes called "novelty") could be improved, providing a story of very general interest across disciplines for eLife. But this would take a rather major re-write.

The reviewers acknowledged the value of our dataset in terms of its breadth, but also had a number of recommendations for improving the paper’s angle to highlight the novelty of the findings. We found these comments to be particularly constructive and have taken them onboard. We hope that the reviewers will agree with us that the paper is much improved as a result.

Briefly, to make our results of wider interest, and more accessible to the readers of eLife, we have restructured the manuscript. Our revision contains two new analyses, references to new supporting literature, and highlights our most salient findings more clearly.

Briefly, we have reworked the paper to emphasize the parallel and independent nature of losses of OPN1SW, in line with recommendations of the reviewer (“convergent molecular pathways leading to convergent trait loss…are convergent losses underpinned by the same or different genetic mechanisms?”). To obtain further insights into the pathways of loss, for our revision we performed reconstructions of mRNA transcripts for the three genes for each taxon for which we have RNA-seq data. We then examined the integrity and diversity of mRNA transcripts in relation to the presence of the ospins proteins. The findings showed that complete RHO and OPN1LW mRNA transcripts were almost always present, whereas OPN1SW transcripts were highly variable among taxa with respect to the number of isoforms, and, in some cases, exons. We validated our findings with PCRs and other published data whenever possible and we discuss our results in the light of similar isolated reports of splice variation in ospin sequences. We show these mRNA transcripts in a new figure, Figure 4. An important finding of this analysis was the potentially high number of independent pathways of degradation.

The reviewer also made the point that we could make more of our ecological models (“if the authors do believe the best story is a methodological warning that transcripts don't predict protein – they would need to show how ignoring protein data leads to wrong conclusions”). We agree with this too and have now addressed this comment by expanding the models to each step of the synthesis of S-opsin: DNA, mRNA and protein. Although several new manuscripts have pointed out that cave roosting ecology may play a role in the evolution of parallel losses, none have formally tested this relationship using quantitative models. Therefore, we have included cave roosting as a factor and now discuss the results. Our results hold: the protein data have the strongest links to the ecology of the lineages, as evidenced by the coefficients (which are directly comparable because they all correspond to the multiplier on the presence of a given ecology). These results are now presented in Tables S5-S7 in Supplementary file 2, as well as Figure 5. Using odds ratios based on the coefficients we show that frugivory is the best predictor for the presence of mRNA and protein, and that protein presence has the strongest relationship to ecology of all the data sets. To our knowledge this is the first time a study has attempted to formally quantify these relationships across a broad comparative sample. Finally, ecological models are not mutually exclusive with a focus on parallel routes of loss. Instead, the full knowledge of the routes of loss has a bearing on the ecological model results.

To summarise, combining our new findings with our original analyses, we believe that our study is stronger and has the potential to become a textbook example of how parallel losses can shape trait evolution among closely related species. While this idea is not new, our study is important because it shows that losses of a trait (in our case OPN1SW protein synthesis in neotropical bats) can arise in parallel by multiple different genetic routes. Furthermore, consideration of each these stages must be considered in order to obtain a full picture of evolutionary loss, which would otherwise be underestimated. Parallel losses are often investigated in relatively divergent species and are never studied at all the main three levels of regulation. We are able to show that these independent losses occur through different mechanisms over evolutionary time. Finally, we reveal that the complex patterns of parallel losses occur in response to ecological pressures. We have been able to confirm via a number of sources (i.e. published RNA-Seq transcripts, PCRs and genomes) that the DNA ORFs recovered by our RNA-Seq assemblies are correct. At the same time, we acknowledge that we are unable to verify our alternate isoforms using PCR or genomic assemblies. Finally, while our findings could apply to any trait, the fact that these results come from opsin genes is likely to be of very wide interest. There have been many recent papers on opsins, published in leading journals, but nearly all studies consider genes alone, and relate observations of mutations to ecological habits without statistical support.

https://doi.org/10.7554/eLife.37412.023

Article and author information

Author details

  1. Alexa Sadier

    Department of Ecology and Evolutionary Biology, University of California, Los Angeles, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Kalina TJ Davies
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9907-3714
  2. Kalina TJ Davies

    School of Biological and Chemical Sciences, Queen Mary University of London, London, United Kingdom
    Contribution
    Conceptualization, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Alexa Sadier
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4258-4775
  3. Laurel R Yohe

    1. Department of Ecology and Evolution, Stony Brook University, New York, United States
    2. Geology & Geophysics, Yale University, New Haven, United States
    Present address
    Department of Geology & Geophysics, Yale University, New Haven, United States
    Contribution
    Conceptualization, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1567-8749
  4. Kun Yun

    Department of Animal Biology, University of Illinois, Urbana, United States
    Contribution
    Investigation, Visualization
    Competing interests
    No competing interests declared
  5. Paul Donat

    Department of Ecology and Evolution, Stony Brook University, New York, United States
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  6. Brandon P Hedrick

    Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, United States
    Contribution
    Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4446-3405
  7. Elizabeth R Dumont

    School of Natural Sciences, University of California, Merced, United States
    Contribution
    Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7809-388X
  8. Liliana M Dávalos

    1. Department of Ecology and Evolution, Stony Brook University, New York, United States
    2. Consortium for Inter-Disciplinary Environmental Research, School of Marine and Atmospheric Sciences, Stony Brook University, New York, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    liliana.davalos@stonybrook.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4327-7697
  9. Stephen J Rossiter

    School of Biological and Chemical Sciences, Queen Mary University of London, London, United Kingdom
    Contribution
    Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    s.j.rossiter@qmul.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3881-4515
  10. Karen E Sears

    Department of Ecology and Evolutionary Biology, University of California, Los Angeles, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    ksears@ucla.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9744-9602

Funding

National Science Foundation (1442314)

  • Karen E Sears
  • Alexa Sadier
  • Kun Yun

National Science Foundation (1442142)

  • Kalina TJ Davies
  • Laurel R Yohe
  • Paul Donat
  • Liliana M Dávalos
  • Stephen J Rossiter

European Research Council (310482)

  • Stephen J Rossiter
  • Kalina TJ Davies

National Science Foundation (1442278)

  • Elizabeth R Dumont
  • Brandon P Hedrick

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank M Agudo-Barriuso, PK Ahnelt, L Peichl, G Tsagkogeorga and staff at the Queen Mary Genome Centre for advice, lab assistance and protocols. For help with permits and field support in the Dominican Republic, we thank J Almonthe, ME Lauterbur, YM León, MS Nuñez, and J Salazar; in Peru, F Cornejo, J Pacheco, J Potter, H Portocarrero, MK Ramos, E Rengifo, JN Ruiz, C Tello; and in Puerto Rico, A Rodriguez-Duran, and N Ann. For access to museum specimens, we thank N Simmons (American Museum of Natural History). For help with permits and field support in Belize, we thank M Howells and the Lamanai outpost lodge staff, N Simmons and B Fenton. For help with permits and field support in Costa Rica, we thank B Matarrita, M Porras, LB Miller, A Kaliszewska, S Santana and the La Selva Biological Station staff. For providing bat images we thank E Clare, and D Rojas for providing roosting ecology data. Results in this paper were obtained using the high-performance LI-RED computing system at the Institute for Advanced Computational Science at Stony Brook University, the Indiana University Mason server funded by NSF-DBI 1458641, and Queen Mary's MidPlus computational facilities supported by QMUL Research-IT and funded by EPSRC grant EP/K000128/1. This research was conducted under research permits VAPB-01436 in the Dominican Republic, 0002287 in Peru, and R-018-2013-OT-CONAGEBIO in Costa Rica.

Ethics

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols 14199 at UIUC, 2017-093 at UCLA, and 2013-2034-NF-4.15.16-BAT and 2014-2090-R1-1.20.17-Bat at SBU. Every effort was made to minimize suffering.

Senior and Reviewing Editor

  1. Patricia J Wittkopp, University of Michigan, United States

Publication history

  1. Received: April 10, 2018
  2. Accepted: December 4, 2018
  3. Accepted Manuscript published: December 18, 2018 (version 1)
  4. Version of Record published: January 15, 2019 (version 2)

Copyright

© 2018, Sadier et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,258
    Page views
  • 211
    Downloads
  • 4
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Evolutionary Biology
    2. Genetics and Genomics
    Tim A Crombie et al.
    Research Article
    1. Evolutionary Biology
    2. Microbiology and Infectious Disease
    Camilo Barbosa et al.
    Research Article Updated