Geographic variation in opsin expression does not align with opsin genotype in Lake Victoria cichlid populations

Abstract Sensory adaptation to the local environment can contribute to speciation. Aquatic environments are well suited for studying this process: The natural attenuation of light through water results in heterogeneous light environments, to which vision‐dependent species must adapt for communication and survival. Here, we study visual adaptation in sympatric Pundamilia cichlids from southeastern Lake Victoria. Species with blue or red male nuptial coloration co‐occur at many rocky islands but tend to be depth‐differentiated, entailing different visual habitats, more strongly at some islands than others. Divergent visual adaptation to these environments has been implicated as a major factor in the divergence of P. pundamilia and P. nyererei, as they show consistent differentiation in the long‐wavelength‐sensitive visual pigment gene sequence (LWS opsin). In addition to sequence variation, variation in the opsin gene expression levels may contribute to visual adaptation. We characterized opsin gene expression and LWS genotype across Pundamilia populations inhabiting turbid and clear waters, to examine how different mechanisms of visual tuning contribute to visual adaptation. As predicted, the short‐wavelength‐sensitive opsin (SWS2b) was expressed exclusively in a population from clear water. Contrary to prediction however, expression levels of the other opsins were species‐ and island‐dependent and did not align with species differences in LWS genotype. Specifically, in two locations with turbid water, the shallow‐water dwelling blue species expressed more LWS and less RH2A than the deeper‐dwelling red species, while the opposite pattern occurred in the two locations with clear water. Visual modeling suggests that the observed distribution of opsin expression profiles and LWS genotypes does not maximize visual performance, implying the involvement of additional visual tuning mechanisms and/or incomplete adaptation. OPEN RESEARCH BADGE This article has earned an Open Data Badge for making publicly available the digitally‐shareable data necessary to reproduce the reported results. The data is available at https://hdl.handle.net/10411/I1IUUQ.

In fish (and vertebrates in general), visual sensitivity is determined by photosensory pigments in the retina, comprised of a lightsensitive chromophore bound to an opsin protein (Bowmaker, 1990).
Pundamilia pundamilia (Seehausen, Lippitsch, Bouton, & Heleen, 1998) and Pundamilia nyererei (Witte-Maas & Witte, 1985) form one of the best-studied pairs of closely related rock-dwelling haplochromine cichlids. They occur at rocky islands in southeastern Lake Victoria. Similar sympatric Pundamilia species pairs (P. sp. "pundamilia-like" & P. sp. "nyererei-like") occur further south in the Mwanza Gulf (Meier et al., 2017;Meier, Marques, Wagner, Excoffier, & Seehausen, 2018; see Figure 1a). Males of the sympatric species are distinguished by their nuptial coloration; P. pundamilia and P. sp. "pundamilia-like" are blue/gray, whereas P. nyererei and P. sp. "nyererei-like" are bright orange or red dorsally and yellow on the flanks; all males have black vertical bars on the flanks. Females of both species are yellow/gray (Seehausen, 1996). At each location, the sympatric species tend to have different depth distributions-the blue species occur in shallow waters while the red species extends to greater depths (Seehausen, 1996;Seehausen et al., 2008). High turbidity in Lake Victoria results in a shift of the light spectrum toward longer wavelengths with increasing depth and, as such, the red species tend to inhabit an environment largely devoid of short-wavelength light (Maan, Hofker, Alphen, & Seehausen, 2006;Seehausen et al., 2008;Castillo Cajas, Selz, Ripmeester, Seehausen, & Maan, 2012; see also Figure 1b). Previous work has shown that, in comparison with P. pundamilia, P. nyererei has greater behavioral sensitivity to long-wavelength light (Maan et al., 2006). In line with this, both red species, P. nyererei and P. sp. "nyererei-like," carry LWS alleles that confer a more red-shifted sensitivity, compared to the allele that dominates in P. pundamilia and P. sp. "pundamilia-like" Seehausen et al., 2008).
In this study, we characterize the opsin expression profiles of wild caught blue and red Pundamilia from multiple islands in southeastern Lake Victoria (Figure 1a). Based on the findings of Hofmann et al. (2009; higher SWS2b expression in Lake Victoria cichlids from short-wavelength rich environments), we predict that SWS2b expression will be highest at locations with relatively clear waters, where short-wavelength light penetrates deeper than in more turbid locations (Hofmann et al., 2009;Maan et al., 2006;Seehausen et al., 1997). Within island locations, we predict that LWS expression will be highest in the red species and SWS2b/SWS2a expression will be highest in the blue species, in line with their respective visual habitats and with observations in laboratory fish .
In cichlids, the cone opsins are arranged in a retinal mosaic: single cones expressing SWS1 or SWS2 are surrounded by double cones expressing RH2 and/or LWS (Carleton et al., 2016). As RH2A and LWS are both expressed in double cones, an increase in one means a decrease in the other. Thus, we also predict that RH2A expression will be higher in the blue species. Finally, we quantify whether variation in opsin expression, within and among islands, covaries with variation in the visual environment and LWS genotype, and we use visual modeling to evaluate whether the observed patterns of opsin expression and genotype are adaptive.

| Fish
In 2014 (September-November), we sampled Pundamilia of both sympatric species/morphs from five rocky islands in the open lake and Mwanza Gulf of Lake Victoria (Figure 1a). We included only males, as they could be easily identified from physical characteristics (morphology, color), whereas females are cryptically colored and hard to identify (Seehausen, 1996). For clarity, we refer to the differently colored males as species at all locations except Luanso Island (more details below), island denotes our different sampling locations, and population indicates species-island combinations.
The blue species at Anchor Island has not previously been studied. It is referred to as Pundamilia "red chest" (Seehausen, 1996) and resembles the other blue species in ecology (occupying shallow habitat around ~1-2 m depth) and morphology, but males have orange-red coloration on the operculum and behind the F I G U R E 1 Sampling locations. (a) Blue and red Pundamilia males were sampled from five island locations in southeastern Lake Victoria. (b) Irradiance spectra at four of the sampling locations (irradiance was not measured at Anchor Island). Vertical lines indicate the spectral midpoint at 1 m depth: The wavelength at which the total light intensity of shorter wavelengths is equal to that of longer wavelengths pelvic fin (Seehausen, 1996). At Luanso, finally, blue and red phenotypes show no detectable genetic differentiation (Meier et al., 2018;Seehausen et al., 2008) and we categorized individuals as blue, intermediate, or red by visually scoring coloration. As in previous studies, we used the mean color scores of multiple observers (Dijkstra, Hekman, Hekman, Schulz, & Groothuis, 2007;Dijkstra, Seehausen, Seehausen, Pierotti, & Groothuis, 2007;Seehausen et al., 2008). the University of Groningen, the Netherlands for analyses. To maximize RNA yield and minimize differences due to circadian variation in opsin expression (Halstenberg et al., 2005), all fish were euthanized in the early evening on the day of capture (~17:00-20:00

| Opsin mRNA expression
We used real-time polymerase chain reaction (qPCR) to determine the relative amount of each cone opsin gene expressed . From preserved eyes, we removed the retina and isolated total RNA using Trizol (Ambion). We reverse transcribed one microgram of total RNA using Oligo(dT) 18 primer (Thermo Scientific) and RevertAid H Minus Reverse Transcriptase (Thermo Scientific) at 45°C to create retinal cDNA. qPCR reactions were set up for each cone opsin (SWS2b, SWS2a, RH2A, LWS) using TaqMan chemistry (Applied Biosystems) and gene-specific primers and probes (Table S1). As in previous studies, we collectively measured the functionally and genetically similar RH2Aα and RH2Aβ as RH2A Hofmann et al., 2009;Spady et al., 2006). Fluorescence was monitored with a CFX96 Real-Time PCR Detection System (Bio-Rad) over 50 cycles (95°C for 2 min; 95°C for 15 s; 60°C for 1 min).
We used LinRegPCR (Ramakers, Ruijter, Deprez, & Moorman, 2003) to determine the critical threshold cycle numbers (C t ) for all four opsin genes. This approach examines the log-linear part of the PCR curve for each sample, determining the upper and lower limits of a "window-of-linearity" (Ramakers et al., 2003). Linear regression analysis can then be used to calculate the individual PCR efficiency and to estimate the initial concentration (N 0 ) from a line that best fits the data (Ramakers et al., 2003). In this way, N 0 values can be estimated without having to assume equal PCR efficiencies between amplicons (Ramakers et al., 2003). All samples were run in duplicate and for consistency, we applied specific quality control parameters: PCR efficiency 75%-125% and Ct standard deviation ≤ 0.5. We used the mean of the duplicate N 0 estimates to calculate relative expression levels for each sample (described below).
On each plate, we included a serially diluted construct containing one fragment of each of the four opsin genes ligated together.
From this, we used linear regression to examine the relationship between Log (concentration) and Ct values of the construct, enabling us to calculate the slope (m) and intercept (b) of the regression. Using these values, we calculated relative cone opsin expression as: where N 0i /N 0all is the expression for a given opsin gene relative to the total expression of all measured opsin genes, Ct i is the critical threshold value for the focal sample, and b and m are the intercepts and slope values derived from the construct linear regression (as detailed in Gallup, 2011). This approach differs from previous work on Pundamilia opsin expression, where only the slope (efficiency) of the construct was considered : see Supplemental Information for details of both approaches).

| LWS sequence variation
LWS is the most variable visual pigment among Lake Victoria cichlids Terai et al., 2002), and there is evidence for strong parallel divergent selection on LWS, both between P. pundamilia and P. nyererei and between P. sp. "pundamilia-like" and P. sp. "nyererei-like" (Meier et al., 2018;Seehausen et al., 2008).
From fin clips, we isolated DNA (Meeker, Hutchinson, Ho, & Trede, 2007) and sequenced the LWS gene of 90 fish that were also measured for opsin expression (Sanger sequencing, GATC Biotech).
We sequenced exons 4 and 5 (498 bp, including the 91 bp intron; Forward primer: GTTTGGTGTGCTCCTCCCAT; Reverse primer: CAGAGCCATCGTCCACCTGT: Figure S1) and categorized individuals as "H" if: 216Y, 230A, 275C, "P" if: 216F, 230T, 275I, and "M3" if: 216Y, 230T, 275I (as in Seehausen et al., 2008). Isolation of genomic DNA was unsuccessful for 21 fish, so we sequenced their cDNA generated for qPCR (see above). For 11 individuals (nonheterozygous, see below), we sequenced both gDNA and cDNA to establish that sequencing results of both methods were identical (see Table S2). All fish were sequenced twice, in forward and reverse directions, and alignments were performed in Mega 7 (Kumar, Stecher, & Tamura, 2016), using the LWS coding sequences reported in Seehausen et al. (2008) as reference. For 17 fish, we observed multiple peaks at one or more of the polymorphic nucleotide sites (see Figure S2). We had gDNA and cDNA for 11 of these individuals; heterozygosity was confirmed by sequencing both sample types (Table S2). In total, we sequenced 111 of 112 fish measured for opsin expression (we failed to sequence one red male from Makobe Island).

| Light measurements
We used the light measurements reported in Castillo Cajas et al.  Note: We examined how opsin expression was influenced by the interaction of species and island, as well as the light environment and LWS genotype. Orange ratio (OR), based on either individual capture depth or population-level depth distribution, did not covary with opsin expression. The only exception was SWS2b, which displayed a significant, negative relationship with population-level OR. However, this can be attributed to the fact that SWS2b was expressed at Makobe Island only (lowest OR; see Figure 3); at all other locations, SWS2b expression was essentially zero (with higher OR scores; see also Figure 4). Together, these results indicate that the light environment alone does not predict the observed variation in opsin expression. Significant effects are indicated in bold.
locations (as reported in (b) the shape of the light spectra ( Figure 3 of Seehausen et al. (1997) is consistent with Castillo Cajas et al. (2012) and Figure 1 of the present study). Thus, the differences in visual conditions between depths (at every location) is highly consistent and the direction of this difference does not change (i.e., deeper habitats will always have less short-wavelength light).
To quantify depth-associated changes in visual conditions, we calculated the orange ratio (OR)-the ratio of light transmitted in the 550-700 nm range over the transmittance in the 400-549 nm range (Endler & Houde, 1995). Pundamilia do not express the SWS1 opsin and the peak sensitivity of the LWS opsin is ~565 nm (Carleton, 2009;Carleton et al., 2005), so visual sensitivity <400 nm or >700 nm is unlikely. We assigned OR values based on the capture depth of each fish. Since spectral measurements at Anchor Island were unavailable, we estimated OR as the median of the orange ratios observed at each depth for Makobe and Python Islands (Figure 3). This estimate was based on prior work showing that the water transparency (Bouton, Seehausen, & Alphen, 1997;Mrosso, Msuku, & Seehausen, 2004) and spectral width (Bouton, Visser, & Barel, 2002) at Anchor Island are intermediate to the other two locations. Additionally, the light spectrum at Anchor Island has a similar shape as that in other locations in the Mwanza Gulf (see Figure 3 in Seehausen et al., 1997). Spectral measurements at Luanso Island were only available down to 4 m depth (light intensity is too low in deeper waters), so we used linear regression to estimate OR experienced by fish caught deeper (~5 m).

| Quantum catch
To evaluate whether population-specific combinations of opsin expression and LWS genotype optimized light capture, we compared the estimated visual performance of local fish with that of hypothetical immigrants from all other locations. We calculated quantum catchthe amount of light captured by the visual system in a given light environment (Kelber, Vorobyev, & Osorio, 2003)-for both sympatric species at each location, except Anchor Island (no spectral measurements, see above). Quantum catch estimates were obtained for each opsin by multiplying relative expression of that opsin with the depth-specific irradiance spectrum, assigned from the capture depth of each fish. Visual sensitivity in cichlids is also affected by differential chromophore usage (vitamin A1 vs. A2; Torres-Dowdall et al., 2017) but this has never been measured in Pundamilia. Thus, our quantum catch estimates incorporated only LWS genotype and opsin expression, using previously reported peak sensitivity values for each opsin, from A1-derived retinal (Carleton, 2009;Seehausen et al., 2008;Spady et al., 2006). We calculated quantum catch (Q c ) as follows: We excluded the two Kissenda Island fish with mismatched LWS genotypes (i.e., blue fish with the "HH" genotype). To compare the visual performance of residents to hypothetical immigrants, we calculated the frequency-weighted mean depth for each genotype/ island combination and "transplanted" the immigrants to that depth.

| Statistical analysis
Prior to analyses, data were filtered for outliers, calculated as 1.5 * the interquartile range (IQR

| RE SULTS
Patterns of opsin expression differed significantly between islands and between species. Species differences were found at most islands, but the direction of differences between the blue and red phenotypes was not consistent between islands. We first present between-island variation in expression patterns, then highlight species differences, both within and between islands.

| Geographic variation in opsin expression
In support of our first prediction, both species expressed more SWS2b at Makobe Island (clear water) than at locations with higher turbidity (blue: p < 0.01 for all comparisons; red: p < 0.001; Figure 4). SWS2b expression at Anchor Island (also relatively clear water) was similar to that at the turbid locations (essentially zero). SWS2a expression did not differ between locations (p > 0.28). Geographic variation in RH2A and LWS expression was different between the two species; both opsins were influenced by significant island by species interactions (RH2A: χ 2 (4) = 39.96, p < 0.001; LWS: χ 2 (4) = 21.62, p = 0.0002).
For the blue species, opsin expression profiles were similar between locations, except at clear water Makobe (Figure 4)

| Light environment
To evaluate whether opsin expression profiles could be predicted by the specific light environment that the fish experience, our model included OR as an individual effect. OR had no influence in full models, nor when the species and island variables were removed (p > 0.41 for all opsins, see Table 1). It is possible that individual capture depth may not adequately represent the populations' light environment, so we also calculated population-level OR scores using depth distribution data from a larger sample of fish (as reported in Seehausen et al., 2008) and reexamined the relationship between expression and OR. The results were quantitatively similar (see Table 1) and together indicate that the local light environment alone does not adequately predict variation in opsin expression. This is supported by the fact that similar light conditions occur in multiple habitats (island-depth combinations), yet we observed highly different opsin expression profiles ( Figure S3).

| Distribution of LWS genotypes
Consistent with prior work (Seehausen et al., 2008), color phenotype matched LWS genotype at Makobe and Python Islands: at these locations, most blue fish were "PP" genotypes and most red fish were "HH" genotypes ( Figure 5b). We also observed a small number of heterozygotes (two red fish at Makobe; four blue fish at Python). A similar pattern was present at Kissenda, though we did observe one blue fish with the "HH" genotype, one red fish with the "PP" genotype, and a considerable number of heterozygotes (eight blue; two red). All fish at Luanso Island were "PP" genotypes ( Figure 5b). LWS genotype had never been assessed at Anchor Island; we found that the blue species-P. "red chest"-had exclusively "PP" genotypes, while the red species-P. nyererei-had both "PP" and "M3" genotypes ( Figure 5b). One red fish at Anchor was F I G U R E 5 Island-and species-specific opsin expression and LWS genotype. (a) Species differences in opsin expression varied across islands. Makobe Island: LWS expression did not differ between Pundamilia pundamilia and Pundamilia nyererei (p = 0.3) but RH2A expression was higher in P. pundamilia (Z = 3.02, p = 0.016). SWS2a did not differ (p > 0.9) and SWS2b was slightly higher in P. nyererei (Z = 2.63, p = 0.055). Anchor Island: all comparisons were nonsignificant (p > 0.86). Python Island: P. sp. "pundamilia-like" expressed more LWS than P. sp. "nyererei-like" (Z = 3.68, p < 0.01), while P. sp. "nyererei-like" expressed more RH2A than P. sp. "pundamilia-like" (Z = 5.00, p < 0.001). SWS2a or SWS2b expression did not differ (p > 0.83). Kissenda Island: LWS expression was slightly higher in P. sp. "pundamilia-like" (Z = 2.64, p = 0.053) while P. sp. "nyererei-like" expressed significantly more RH2A (Z = 3.32, p < 0.01). SWS2a and SWS2b expression did not differ (p > 0.9). Luanso Island: there were no differences in opsin expression (p > 0.9). Sample sizes are indicated above each bar and error bars represent ± standard error. ***indicates p < 0.001, **indicates p < 0.01, *indicates p < 0.05, • indicates p < 0.1. (b) Consistent with previously reported patterns (Seehausen et al., 2008), the blue species were generally "PP" genotypes and the red species were "HH" genotypes. Anchor Island had not been previously investigated: the "H" allele was absent, but the "M3" allele was present in the red phenotypes. All fish at Luanso Island were "PP" genotypes heterozygous. For LWS allele frequencies per species and location, see Table S3.

| Relationship between LWS genotype and opsin expression
The data presented above suggest that there is no consistent relationship between LWS genotype and opsin expression across populations. Indeed, opsin expression significantly covaried with the interaction between LWS genotype and location (model: expression ~ genotype*island), for RH2A and LWS (p < 0.001 for both). As seen in Figure 6, in turbid waters (Python and Kissenda), individuals with LWS genotype "PP" had lower RH2A and higher LWS expression than individuals with "HH" genotypes. In clear water (Makobe), this pattern was reversed. For opsin expression patterns for each island and LWS genotype, see Figure S4.
F I G U R E 6 Relationship between LWS genotype and opsin expression differs between islands. Opsin expression is both genotype-and location-dependent, as evidenced by the significant interaction of LWS genotype and island for RH2A (χ 2 (5) = 41.24, p < 0.001) and LWS expression (χ 2 (5) = 27.53, p < 0.001). In turbid waters (Python and Kissenda), individuals with LWS genotype "PP" had (a) lower RH2A expression and (b) higher LWS expression than individuals with "HH" genotypes. This pattern was reversed in clear waters (Makobe). All fish at Luanso Island were "PP" genotypes. Error bars represent ± 95% C.I

| Visual performance
If the variation reported above is adaptive, we predict that the observed combinations of opsin genotype and opsin expression maximize visual performance in the local environment. To test this prediction, we calculated the quantum catch (Q c ) of the different genotypes from all locations (except Anchor-spectral measurements were unavailable), considering their opsin expression profiles. We then compared the Q c of the local fish to the Q c that would be achieved by hypothetical immigrants from other islands.
Despite geographic variation in opsin expression, visual performance did not consistently differ between the residents and hypothetical immigrants for either species (Figure 7). Only in one out of 18 comparisons did residents achieve significantly higher total Q c : the blue species at Luanso had higher total Q c than hypothetical immigrants from Makobe (all "PP" genotypes; Z = 3.76, p = 0.014). Within the red species (only "HH" genotypes), the resident populations never achieved higher total Q c than the transplants (Makobe: p > 0.9; Python: p > 0.12; Kissenda: p > 0.9). Q c values for the red species at Makobe were generally low, but all red fish from Makobe came from the lower end of the previously reported depth distribution (Seehausen et al., 2008), and thus from a narrow-spectrum light environment. We therefore recalculated Q c for a more representative depth range, but this did not generate differences between residents and immigrants either (see Figure S6).
We also compared the visual performance of the heterozygous genotypes with the "HH" or "PP" genotypes (of the same color), within islands. We found no evidence of increased or decreased light capture in the heterozygotes ( Figure S7). Thus, we cannot explain the presence or absence of heterozygous genotypes by visual system performance. Taken together, these results suggest that the observed combinations of opsin expression and LWS gene sequence do not maximize local visual performance.

| D ISCUSS I ON
Divergent adaptation to alternative visual habitats has been implicated in cichlid speciation. Previous work in Pundamilia has revealed correlations between the local light environment and LWS genotype across populations (Seehausen et al., 2008). The contribution of differential opsin expression to visual adaptation remained to be addressed: Haplochromine species and populations (in Pundamilia and other genera) differ in opsin expression (Carleton, 2009;Carleton et al., 2005;Hofmann et al., 2009Hofmann et al., , 2010Parry et al., 2005;Smith et al., 2011;Spady et al., 2006) but a systematic exploration of this variation in Pundamilia was lacking.
Here, we report that opsin expression profiles differ markedly between populations and do not covary with LWS genotype. Visual modeling suggests that the observed variation does not contribute to local visual performance.

| Geographic variation in opsin expression
We predicted higher SWS2b expression at clear water locations (following: Hofmann et al., 2009). Our results conform to this pattern.
At Makobe Island, where short-wavelength light is more abundant than at the more turbid locations further south, SWS2b expression was high in both phenotypes. The waters at Anchor Island are also relatively clear but, in contrast to Makobe, SWS2b expression was low. The evolutionary history of this population has not been explored but, perhaps, it is similar to other Mwanza Gulf populations (Python/Kissenda), characterized by recent hybridization between the blue and red species (Meier et al., 2017(Meier et al., , 2018 and low SWS2b expression. The other opsins did not show clinal variation in expression levels. SWS2a expression did not differ among locations, while RH2A and LWS were both species-and island-specific. In the blue species, LWS expression was the lowest at clear water Makobe Island and RH2A expression was the highest. In the red species, LWS expression was less variable between islands but RH2A expression tended to be higher at more turbid locations (Python and Kissenda).
These patterns could not be explained by variation in the local light environment alone-orange ratio (OR) did not predict opsin expression-but may be due to (a) the different evolutionary histories of the populations and/or (b) different modes of visual adaptation at different locations. We address both explanations in more detail below.

| Species differences in opsin expression
Within each island, we predicted that the local red species would express more LWS and the blue species would express more SWS. We found inconsistent support for this prediction. SWS2a expression did not differ between the phenotypes at any of the five locations and SWS2b was slightly higher in P. nyererei at Makobe. Patterns of LWS and RH2A expression were more variable: at the two locations with clearer waters, Makobe and Anchor, the blue species (P. pundamilia and P. "red chest") tended to express more RH2A and less LWS than the red species (P. nyererei). At locations with higher turbidity-Python and Kissenda Islands-the difference in expression pattern reversed: LWS expression was higher in P. sp. "pundamilialike" (significantly so at Python, trending at Kissenda) and RH2A expression was higher in P. sp. "nyererei-like." This is in opposition to our prediction but in agreement with results of Hofmann et al.
(2009), who also observed higher LWS expression in the red types of the species pair P. pundamilia and P. nyererei from Senga Point (clear water) but higher LWS expression in the blue types for P. sp. "pundamilia-like" and P. sp. "nyererei-like" sampled at Kissenda and Python Island, respectively. Finally, at the most turbid location in our study, Luanso Island, we found no differentiation in opsin expression. This is consistent with the lack of genetic differentiation and overlapping depth ranges at this location, as documented previously (Meier et al., 2017;Seehausen et al., 2008).
Taken together, we find patterns of differentiation in opsin expression profiles at all studied locations where blue and red species are genetically differentiated, but not where they are not. The direction of differentiation between blue and red species, however, was opposite between the two sites with relatively clear waters versus the two sites with relatively turbid waters. This discrepancy may be related to the evolutionary histories of the populations. Meier et al. (2017) found that the most likely scenario for Pundamilia speciation involves divergence of P. pundamilia and P. nyererei outside the Mwanza Gulf, with settlement of both species at Makobe Island. P. pundamilia then colonized the Mwanza Gulf (including Python Island). Many generations later, this population received gene flow from P. nyererei leading to a renewed speciation event in which a "nyererei-like" species with red males and a "pundamilia-like" species with blue males emerged from the original P. pundamilia population at Python, within the past 500 years (Meier et al., 2017(Meier et al., , 2018. This distinct evolutionary history (as well as possible mixing with other species in the Mwanza Gulf) may have resulted in different, better gulf-adapted expression profiles. Inconsistent with this scenario, however, is that we did not find evidence for superior visual performance for the resident fish at Python and Kissenda-we discuss this more below.

| Distribution of LWS genotypes
Pundamilia harbors three versions of the LWS opsin gene: The "H," "P," and recombinant "M3" alleles. At locations where the blue and red species are genetically differentiated-Makobe, Python, and Kissenda-the allele types were previously found to be nearly fixed in each species (Seehausen et al., 2008). We observed similar patterns; at Makobe and Python Islands, the "PP" genotype occurred only in the blue species and the "HH" genotype only in the red species. This pattern was also present at Kissenda, although we did observe two fish with mismatched genotypes (see Figure 5b).
Heterozygous genotypes occurred at all three locations: Makobe (two red), Python (four blue), and Kissenda (eight blue, two red). At Anchor Island, which had not been previously investigated, all P. "red chest" were "PP" genotypes, and P. nyererei were "PP" or "M3" genotypes, plus one heterozygous individual. The "M3" allele also occurs at low frequency in P. nyererei at Makobe Island (Seehausen et al., 2008; not observed in the present sample) but our results suggest it may have replaced the "H" allele in red types at Anchor. This may reflect the evolutionary history of the Anchor population-perhaps Makobe P. nyererei with the "M3" allele colonized it-and/or the "M3" allele is selected for in the light environment at Anchor Island. Both of these scenarios are speculative and will require further study.
Finally, at Luanso Island, all fish were "PP" genotypes, again consistent with earlier results and with the lack of genetic differentiation at this location (Meier et al., 2017;Seehausen et al., 2008). Our results confirm that LWS genotype is under divergent selection in blue versus red Pundamilia and that gene flow at this locus is more common at more turbid locations, where the species are less strongly isolated (both in space and in genome-wide genetic variation).

| Relationship between opsin expression and LWS genotype
Patterns of opsin expression did not consistently covary with species differentiation in LWS genotype. This was highlighted by significant interactions between genotype and island for both RH2A and LWS expression: the direction of differentiation in opsin expression profile between "HH" and "PP" genotypes at turbid Python and Kissenda Islands was the reverse of that at clear water Makobe (Figure 6).
Estimates of visual performance, using Quantum catch calculations, suggest that this variation does not increase local visual performance: for both species, residents did not consistently achieve higher total light capture than hypothetical immigrants from other islands.
Comparisons of the heterozygotes and "HH" or "PP" genotypes also revealed no difference in total light capture (see Figure S7), suggesting neither selection for nor against heterozygous genotypes.
Together, our results indicate that the observed combinations of opsin expression and LWS genotype do not maximize local visual performance. However, our estimates of visual performance may be inadequate. First, quantum catch is a relatively crude measure of visual perception, that may not reflect actual performance at relevant visual tasks in nature, such as object-background discrimination (Guthrie, 1986). Second, our visual model did not incorporate all aspects of visual perception. For example, cichlids can use either Vitamin A1-or Vitamin A2-based chromophores (Torres-Dowdall et al., 2017), which influences visual sensitivity (Dartnall & Lythgoe, 1965;Hárosi, 1994;Toyama et al., 2008). Chromophore usage has never been measured in Pundamilia but it might differ between populations and may contribute to visual adaptation. Prior work has estimated mixed chromophore usage in the red species  but precise measurements are required to assess differences in visual performance across environments.

| CON CLUS ION
We analyzed opsin expression patterns of wild-caught Pundamilia cichlids from several locations and depth ranges in Lake Victoria.
Opsin expression differed between species and islands, and replicate populations of species pairs from clear waters were similar to each other but distinct from species pairs inhabiting turbid waters. These patterns could not be explained by variation in visual environments alone and did not consistently correlate with species differences in LWS opsin genotype. Visual modeling suggests that the observed combinations of opsin expression and LWS genotype do not maximize local visual performance. Our results highlight the need to explore other visual tuning mechanisms, as well as more sophisticated ways of measuring visual performance, to understand how different components of the visual system adapt and co-evolve.

ACK N OWLED G M ENTS
We thank the Tanzanian

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
MEM and DSW designed the study; DSW collected eye samples; RvE designed the qPCR protocol and created the standard construct; DSW and RM completed laboratory work, with assistance from RvE; WV established the sequencing protocol; OS confirmed fish identities from photographs and contributed field data. DSW performed the analyses, with assistance from MEM; DSW and MEM wrote the manuscript, with contributions from OS. All authors approved the contents of this manuscript.