Colour vision variation in leaf-nosed bats (Phyllostomidae): Links to cave roosting and dietary specialization

2.1 Study species and sample collection

We isolated genomic DNA from liver and/or muscle samples from 21 species (20 phyllostomid and one vespertilionid species, Myotis lavali) (Table 1). We classified species into broad trophic categories—frugivore, nectarivore, insectivore, omnivores or haematophagous—based on the predominant food type in their diets (Table 1, Aguirre, Herrel, Van Damme, & Matthysen, 2002), although we acknowledge that while most phyllostomids specialize on a particular type of diet, they may also feed occasionally on other food items (Gardner, 1977).

Bat tissue samples were obtained from specimens collected as vouchers during studies on bat occurrence, biology and ecology, and permission to use tissue samples for the present research was granted by the Brazilian Institute of the Environment and Renewable Natural Resources (IBAMA) (export permits # 13BR010191/DF and 16BR022149/DF). Bats were collected under Brazilian federal permissions given by the Chico Mendes Institute for Biodiversity Conservation (ICMBio) through the System for Authorization and Information on Biodiversity (SISBIO), between 2009 and 2015, at eleven Brazilian municipalities: four located in the state of Rio Grande do Norte (Natal, Nísia Floresta, Caraúbas and João Câmara; SISBIO Permits # 25233, 30730, 48325 and 53169), two in the state of Paraíba (Rio Tinto and João Pessoa; SISBIO Permits # 35846 and 45168), one in the state of Bahia (Igrapiúna; SISBIO Permit # 26934), one in the state of Minas Gerais (Januária; SISBIO Permit # 14875), one in the state of Santa Catarina (Florianópolis; SISBIO Permit # 17131) and two in the state of Rio Grande do Sul (Maquiné and Barracão; SISBIO Permit # 21438). Bats were captured in mist nets at the ground level and were euthanized using sulphuric ether or cervical dislocation following established protocols (Kunz & Parsons, 2009; Sikes & Gannon, 2011). Liver or muscle samples were preserved in absolute ethanol and kept frozen at ca. −18°C.

2.2 Isolation, amplification and sequencing of opsin genes

We isolated genomic DNA from tissue samples using a Qiagen DNeasy Blood and Tissue Kit, following the manufacturer's protocol. We amplified and sequenced partial exons 1 to 4, along with introns 1 and 3, of the short-wavelength-sensitive opsin gene (SWS1; OPN1SW). We also amplified and sequenced partial exons 1 to 5 of the long–wavelength-sensitive opsin gene (OPN1LW). Five sites of the OPN1LW are known to impact spectral tuning and are found on exon 3 (positions 180), exon 4 (position 197) and exon 5 of the OPN1LW (positions 277, 285, 308); the ten tuning sites of the OPN1SW gene are located on exon 1 (positions 46, 49, 52, 86, 90, 93, 97, 114, 116, 118; Yokoyama et al., 2006, 2008; Hunt et al., 2007; Carvalho et al., 2012). We note that OPN1LW amino acid numbering follows nomenclature for the human opsin and OPN1SW sequences follow bovine nomenclature.

We designed primer pairs for OPN1SW and OPN1LW opsin genes based on an alignment of sequences published in GenBank for related bat species using Primer3 (Supporting Information Tables S1 and S2). In addition, we used previously described OPN1SW (exons 1–3) and OPN1LW primers (Zhao, Rossiter, et al., 2009). In several cases, some species failed to amplify with our original primers and we designed new ones as needed (Table 1, Supporting Information Table S1). We amplified the target sequence by PCR using a BioRad C1000 Touch Thermocycler, as follows: 2 μl of template DNA, 2.5 μl of 10× DreamTaq buffer, 1.25 μl of each primer (10 μM), 2.5 μl dNTPs (2 mM each), 2 μl MgCl₂ (25 mM) 13.38 μl ddH₂0 and 0.125 μl DreamTaq (5 U/μl). PCR conditions were optimized for each primer set (Supporting Information Table S1).

We visualized PCR products on 1% agarose gels against a 100-bp ladder (GeneRuler) to verify PCR success. PCR products that showed a single band in the gel were purified using ExoSAP-IT (Applied Biosystems) protocol. If multiple bands were present, we used the Promega Wizard SV Gel and PCR Clean-Up kit to isolate the band of interest, following the manufacturer's protocol. Purified products were directly Sanger-sequenced (ABI platform via Eurofins’ SimpleSeq service) with the same primer sequences used during amplification, with one exception. For OPN1SW exons 3–4 (primer set g), we designed new sequencing primers slightly offset from the PCR primers to increase specificity (Forward: 5′ TGCARTGTTCCTGTGGCCCCG 3′; Reverse 5′ GTATGGGATTGTAGACAC 3′).

We assembled the forward and reverse sequences into contigs using geneious version 10.0.2. The consensus sequences were aligned against other species from our experiment as well as additional reference sequences obtained from GenBank (Supporting Information Table S2) using the ClustalW function in Geneious. Alignments were manually adjusted as needed.

2.3 Phylogeny and selection analyses

For tests of selection, we aligned the opsin gene sequences we generated in addition to those from six additional species of interest (Microcebus murinus, Pteropus vampyrus, Pteropus alecto, Miniopterus natalensis, Rousettus aegyptiacus and Artibeus jamaicensis; Figure 2, accession numbers listed in Supporting Information Table S2). Relationships among species were inferred using divergence dates calculated with TimeTree ( www.timetree.org, accessed February 2018).

We tested for evidence of purifying or relaxed selection acting on OPN1SW and OPN1LW using codeml branch models implemented in paml (Yang, 2007) and the program relax (Wertheim, Murrell, Smith, Kosakovsky Pond, & Scheffler, 2015). In both programs, we tested whether there was a difference in selection acting on branches with putative pseudogenes vs. functional genes (OPN1SW) or with regard to roosting or diet style (OPN1LW) and compared branches. In both paml and relax analyses of OPN1SW, we compared the evolutionary scenarios of pseudogenization as one ancestral event vs. convergent events against null models in an effort to illuminate the evolutionary history of colour vision loss in the lineage. We removed premature stop codons present in the OPN1SW alignment by replacing the deletions leading to them with ambiguous characters (Ns). In addition to branch models, for OPN1LW, we also used codeml site models to test whether any of the opsin tuning sites are under positive selection. When interspecific variation at spectral tuning sites was found, we inferred the ancestral state using maximum likelihood (JTT model) in MEGA v7.0 (Jones, Taylor, & Thornton, 1992; Kumar, Stecher, & Tamura, 2016; Nei & Kumar, 2000). For ancestral state reconstruction analyses, we included OPN1LW sequences from 18 additional outgroups within Chiroptera to resolve ambiguities at some nodes (accession numbers listed in Supporting Information Table S2).

3.1 Short-wavelength-sensitive opsin gene (OPN1SW)

We sequenced partial exons 1–3 of the OPN1SW opsin gene for 20 of the 21 species newly examined in this study. Our repeated attempts to amplify exons 1–3 of the OPN1SW opsin gene for Mimon bennettii were unsuccessful, which we attribute to low-quality DNA as this sample was difficult to work with for all analyses. We successfully obtained sequences for partial exons 3–4 of the OPN1SW opsin gene for all 21 species examined. We found variation in the persistence of a functional OPN1SW gene across phyllostomid species (Figure 2).

The OPN1SW gene for each of the three extant species of vampire bats, Desmodus rotundus, Diaemus youngi and Diphylla ecaudata, showed strong evidence of pseudogenization. Each of the three species possessed at least one deletion leading to a premature stop codon in exon 1. Diaemus youngi and Desmodus rotundus share one of the deletions in exon 1, suggesting that this is a shared mutation that occurred in the common ancestor of these two species. Several unique deletions were also present in exon 4 of D. youngi and in D. rotundus, resulting in other premature stop codons (Figure 3; Supporting Information Figure S1). None of the indels leading to a premature stop codon that we identified were shared between all three vampire bat species; however, the large number of single nucleotide polymorphisms and deletions that were present in each species may be obscuring a previously shared mutation. In addition, a mutation to the start codon is evident in the whole-genome sequence of Desmodus rotundus (Zepeda Mendoza et al., 2018). However, we were missing this sequence in D. youngi and D. ecaudata, as our primer design necessarily included the conserved regions leading up to the start codon and we cannot know whether this mutation is shared. We found evidence of a comparatively recent and independent pseudogenization of the OPN1SW gene of Lonchophylla mordax. A single nucleotide deletion has occurred in this species, leading to a premature stop codon in exon 1 (Supporting Information Figure S1). In addition, a unique _Phe46_Cys substitution in L. mordax was found; this site is known to affect the spectral tuning of the opsin, however, the substitution may have occurred after the loss of gene function (Supporting Information Figure S1).

The OPN1SW gene appeared to be intact and presumed functional for the remaining 16 phyllostomid bats and M. lavali, as we observed no indel, frameshift or nonsense mutations in the opsin coding regions (Figure 2, Supporting Information Figure S1). For all species with putatively functional OPN1SW genes, amino acids at the 10 sites known to impact the spectral tuning of the cone photopigment were invariant in the species we sequenced. This suggests that all species with a functional OPN1SW gene maintain a UV-sensitive OPN1SW opsin.

Within the OPN1SW gene of Phyllostomidae, we found a total of 101 variable sites of 310 examined. If we exclude vampire bats and L. mordax and include only the 16 species with seemingly intact OPN1SW genes, 52 amino acids of 310 positions are variable. Results of the paml analysis strongly favoured the two-branch models (M2) over the null model of a single ω value for all branches in the tree (M2 Hyp. 1: p < 0.00001, LR = 47.69; M2 Hyp. 2: p < 0.00001, LR = 45.14), indicating that branches with putative pseudogenes are under relaxed selection (ω = 0.80–0.81), while branches with functional genes are under purifying selection (ω = 0.13). We attempted to elucidate the evolutionary history of OPN1SW pseudogenization in the vampire bats by modelling the two hypotheses (ancestral vs. convergent pseudogenization) separately; however, the log likelihood scores for both models were very similar and could not be distinguished statistically.

Results of the relax analyses likewise supported the hypothesis that lineages with putative pseudogenes are under relaxed selection compared to those with functional genes (p < 0.0001, LR = 49.27) but could not distinguish between the two evolutionary scenarios of OPN1SW pseudogenization in vampire bats. In both relax models, ω values for all sites along pseudogenes were significantly closer to neutrality than ω values for sites along functional genes (Figure 4a). Estimates of the selection intensity parameter k further indicated relaxed selection in extant vampire bats, as well as in the ancestral branch of Desmodontinae (Figure 4c).

3.2 Long-wavelength-sensitive opsin gene (OPN1LW)

We successfully sequenced partial exons 1–5 of the OPN1LW opsin gene for all but one of the 21 sampled species. We failed to amplify exons 2, 3 and 4 of the OPN1LW of M. minuta and thus excluded this species from paml and relax analyses. The OPN1LW gene was intact and presumed to be functional for all species examined as we found no frameshift, indel or nonsense mutations (Supporting Information Figure S2). Within the 20 phyllostomid species, we found a total of 37 variable sites of 328 examined. Using the relax test, we found no difference in the intensity of selection acting on the different branches (p = 0.5, LR = 0.45; Figure 4b), indicating purifying selection is present across species, regardless of diet or roosting category (Supporting Information Figure S3). Results of the paml analysis of OPN1LW did not find the tuning sites to be under positive selection, but did indicate that three other sites (4, 20, 171) are under positive selection (p < 0.001, LR = 20.59). Despite this, we noticed substantial amino acid variation at one of critical tuning sites of the OPN1LW.

Tuning site 180 (exon 3) was dynamic among chiropteran species. To assign the ancestral state at this site for Phyllostomidae, and track patterns of substitutions occurring across the radiation of species, we examined exon 3 of our OPN1LW sequences and those of 18 additional bat species downloaded from GenBank (Supporting Information Figure S4). Across the Order Chiroptera, we found 18 species along four separate taxonomic branches possessed serine at site 180, while 21 species along six branches retained alanine as the ancestral character state (Figure 2, Supporting Information Figure S4). In addition, we found a substitution at a different tuning site, _Ala285_Tyr, in one of the reference species, Myotis ricketti (EU912345.1).

4.1 OPN1SW variation

The lack of missense or nonsense mutations in sixteen of the twenty leaf-nosed bat species we sequenced, along with low substitution rates, suggests that the OPN1SW gene is under purifying selection in most phyllostomid bats. We did, however, find that Lonchophylla mordax has a premature stop codon in exon 1. Although we did not directly examine opsin function, a premature stop codon this early in the opsin gene—together with the absence of a resurrecting start codon nearby—strongly suggests lost function of the OPN1SW gene. Bats from the genus Lonchophylla are well known to use caves as roosts (Coelho & Marinho-Filho, 2002; Graham, 1988), and some species are considered highly dependent on cave habitats (e.g., Lonchophylla dekeyseri, Aguiar & Bernard, 2016). Although the roosting preferences of Lonchophylla mordax have not been studied extensively, the sparse available records indicate that this species, like other members of its genus, roosts mainly in caves (Gregorin & Ditchfield, 2005; Gregorin & Mendes, 1999; McCarthy, Albuja, & Manzano, 2000) or cavelike structures (Auler et al., 2006). Our results may therefore provide support for the hypothesis that cave-roosting relaxes selection on OPN1SW. Multiple cases of colour vision loss in the suborder Yinpterochiroptera linked to cave roosting (e.g., Genera Dobsonia and Rousettus), led Zhao, Rossiter, et al., (2009) to hypothesize that the behavioural innovation of cave roosting released OPN1SW from selective constraint due to the scotopic light environment. S-cones are involved in nonimage forming biological roles under mesopic conditions, and constraints linked with photoentrainment to day–night cycles may explain the persistence of the S opsin among mammals that are active in the canopy (e.g., tree roosting) at dawn and dusk (Allen, Brown, & Lucas, 2011). Additional evidence from other species of a link between scotopic roosting environments and evolutionary loss of colour vision would be instructive. For example, Lonchorhina aurita, thought to be closely related to the genus Lonchophylla, is reported to be strictly cave dwelling (Baker, Solari, Cirranello, & Simmons, 2016; Nowak, 1994; Rojas, Warsi, & Dávalos, 2016). Future work examining the opsin genes of Lonchorhina aurita and other neotropical cave specialists (e.g., families Natalidae, Furiperidae and Mormoopidae) as well as optionally cave-roosting bats (e.g., families Phyllostomidae, Molossidae and Vespertilionidae; Eisenberg & Redford, 1999) would add to our growing understanding of the link between roosting environment and colour vision evolution.

4.2 Loss of OPN1SW opsin function in Desmodontinae

All three species within Desmodontinae have a highly specialized diet—they feed exclusively on blood from mammals and birds (Greenhall, Joermann, Schmidt, & Seidel, 1983; Greenhall & Schutt, 1996; Ito, Bernard, & Torres, 2016). Desmodus rotundus shows a preference for mammalian prey (Bobrowiec, Lemes, & Gribel, 2015; Voigt & Kelm, 2006), while Diphylla ecaudata and Diaemus youngi preferentially prey on birds (Costa, Oliveira, Fernandes, & Esberard, 2008; Greenhall, Schmidt, & Joermann, 1984). Our results suggest OPN1SW pseudogenization and monochromacy occurred either: 1) in the last common ancestor of all vampire bats prior to their divergence ca. 23 mya, or 2) in the common ancestor of Desmodus rotundus and Diaemus youngi before their split ca. 14 mya, and independently in Diphylla ecaudata. The former explanation is more parsimonious, but could not conclusively be demonstrated. Either scenario suggests that changes linked to a haematophagous diet correlate with the loss of colour vision in the evolutionary radiation of this clade.

In contrast to many bats, which prefer to be more active under dim (mesopic) conditions early in the evening when colour cues may still be salient (Aguiar & Marinho-Filho, 2004; La Val, 1970; Melin et al., 2012; Roth et al., 2008), vampire bats are usually more active during the darkest (scotopic) hours and on evenings with reduced moonlight (Flores Crespo, Linhart, Burns, & Mitchell, 1972; Mitchell, Burns, & Kolz, 1973; Uieda, 1992). For example, common vampire bats (Desmodus rotundus) forage only after dark (Wimsatt 1969), and emerge from their roosts about 75–90 min after sunset, which is later than other species roosting in the same cave (Trajano, 1984). The change in diet and shift to more scotopic foraging may have rendered UV sensitivity and colour vision obsolete.

Vampire bats are known to rely heavily on nonvisual senses to locate prey. Desmodus rotundus has highly acute passive sound localization and the most sensitive hearing among bat species so far tested (Heffner, Koay, & Heffner, 2013). Due to specialized noise-sensitive neurons, common vampire bats are even able to detect and recognize breathing sounds of individual animals (Gröger & Wiegrebe, 2006; Schmidt, Schlegel, Schweizer, & Neuweiler, 1991). In addition to hearing, D. rotundus uses olfaction to locate prey (Bahlman & Kelt, 2007), and over short distances, it also relies on a thermoreception system that is unique among mammals and enables them to detect more vascularized areas on the prey's body surface (Gracheva et al., 2011; Jones et al., 2013; Kürten & Schmidt, 1982). Due to the high metabolic energy costs of sensory systems (Moran, Softley, & Warrant, 2015), an energetic prioritization may have been placed on the development of the nonvisual senses at the expense of the OPN1SW opsin protein. Loss of function in UV-sensitive opsins among bats has previously been linked to this form of sensory trade-off in Yinpterochiropteran bats (Zhao, Rossiter, et al., 2009). Vision is costly, and instances of cost–performance trade-offs for sensory information processing have been implicated in a wide variety of systems (Lan, Sartori, Neumann, Sourjik, & Tu, 2012; Moran et al., 2015; Niven & Laughlin, 2008). Deciphering whether the loss of UV vision among vampire bats was driven by an energetic trade-off or simply due to neutral evolution due to disuse in leaf-nosed bats (Yangochiroptera) would shed light on the mechanisms shaping the sensory evolution of these unique mammals.

4.3 Persistence of colour vision in phyllostomid bats, links with echolocation mechanisms

Setting vampire bats and L. mordax aside, other leaf-nosed bats with diverse ecological habits (Table 1) appear to have retained a functional OPN1SW gene, indicating that variation in diet (fruit, nectar or insects) does not affect UV perception significantly. This is perhaps surprising given the loss of the OPN1SW opsin functionality in insectivorous species, including rhinolophoid bats (in Figure 2, represented by Hipposideros armiger and Rhinolophus affinis) (Zhao, Rossiter, et al., 2009); however, there is at least one fundamental difference between these bats and leaf-nosed bats. Rhinolophoids have a highly acute, derived form of echolocation, known as “high-duty cycle echolocation” (Ho, Fang, Chou, Cheng, & Chang, 2013). Ultraviolet light sensitivity may be of relatively higher importance for the phyllostomids, which rely on low-duty cycle echolocation (Shen, Fang, Dai, Jones, & Zhang, 2013). It is plausible that colour vision and other senses are more important to species with less acute sonic perceptions. Most bats that use high-duty cycle echolocation are insectivorous (Fenton et al., 2012), and acute sonic perception appears to be sufficient to locate prey. Several of the leaf-nosed bat species in this study are also insectivorous, yet all possess the capacity for UV light perception. Our results therefore are also consistent with the hypothesis that sensory trade-off has occurred between high-duty echolocation and functional ultraviolet perception (Zhao, Rossiter, et al. 2009).

4.4 Functionality of OPN1LW

Our data indicate that the gene for the long-wavelength-sensitive opsin (OPN1LW) is functional and under purifying selection in all the species we investigated in our study. This suggests that sensitivity to longer wavelengths of light is largely maintained across the family Phyllostomidae. This retention of a functional long-wavelength-sensitive opsin is consistent with patterns observed across bats and all other nonmarine mammals examined to date (Jacobs, 2009; Meredith et al., 2013; Zhao, Rossiter, et al., 2009).

4.5 Spectral tuning of opsin genes

Regarding the hypothesis that diet or roosting behaviour will impact the tuning of opsins to different wavelengths of light, we do not find clear support. We found amino acid consistency at all known tuning sites of the OPN1SW, indicating uniform UV vision in phyllostomid bats with a functional OPN1SW regardless of the species-specific ecology. However, we report for the first time considerable variation in tuning of the OPN1LW opsin of Phyllostomids and other Yangochiropterans (Figure 2). To our knowledge, OPN1LW variation among closely related mammalian species (within the same family) is only known among treeshrews and primates, the latter group well known for polymorphic colour vision and high levels of inter- and intraspecific variation (Jacobs, 2009; Jacobs et al. 2017, Kawamura & Melin, 2018). Here, we report numerous shifts in amino acids at site 180 among taxa (Figure 2, Supporting Information Figure S4). Bat species with alanine at site 180 are predicted to be maximally sensitive to wavelengths of light ca. 553–558 nm; the λ_max for species with serine at site 180 is red-shifted and predicted to be sensitive to light ca. 560–563 nm (Neitz, Neitz, & Jacobs, 1991; Yokoyama et al., 2008). While the spectral tuning of OPN1LW at site 180 is not clearly correlated with any broad dietary category we consider here (classification as fruit, nectar or insect-based diets, Figure 2), our results underscore the evolutionary plasticity of opsins among nocturnal mammals and may reflect habitat or finer-scale diet heterogeneity. Intriguingly, we also noted a shift at site 285 from Tyr to Ala in one of the reference sequences, Myotis ricketti. This substitution is predicted to reduce the spectral sensitivity of this species by 16 nm (Yokoyama et al., 2008). Taken together with previously reported OPN1LW tuning variation for Myotis velifer (Wang et al., 2004), the visual ecology of this genus may be of special interest in the future. Future study of the reflectance properties of foods together with the irradiance spectra of nocturnal illumination in the habitats of these species may provide insight into the pressures shaping variation in long-wavelength colour perception among bats.