Absence of light drives the evolution of cave animals towards a suite of characteristic, cave-related (troglomorphic) phenotypes. In the dark, eyes and pigmentation lose their functions, and tend over the generations to regress or disappear. Without light there is no photosynthesis, and the trophic base of many cave communities is narrow. Cave animals typically cope with the scarcity of food by evolving more sensitive tactile and chemical senses and slower or more efficient metabolisms. Compensatory changes like these probably evolve because of strong selection, but what causes the regression of eyes and pigmentation? The three modern competing hypotheses for eye regression are natural selection, recurrent mutation/genetic drift, and pleiotropy [2
is an ideal species to study the genetics of troglomorphy because it has both eyed surface and cave adapted populations, all of which are interfertile. Cave fish were collected from Pachón cave in NE Mexico [locality map in 4] and surface fish were collected from nearby streams (Supplemental Fig. 1
). We hybridized Pachón cave and surface fish, creating a mapping progeny of 539 F2
siblings. We mapped 178 loci in the cross (2191 cM) for an average distance between adjacent markers of 14.7 cM. We phenotyped the F2
fish by measuring eye size, lens size, counting the density of melanophores in four places on the bodies, measuring the lengths of the dentary and maxillary bones in the jaw apparatus, and counting maxillary teeth and taste buds ( lists sample sizes for the different traits). This gave us a set of standardized phenotypes that could be correlated with genotypes. Phenotypic and genotypic data (see Supplemental Online Material) were used to identify chromosome regions where genes affecting the traits were located. Quantitative trait loci (QTL) were detected in two phases, first by simple interval mapping (SIM) of putative QTL, followed by a refinement phase using multiple interval mapping algorithms (MIM). We used MultiQTL software (www.multiqtl.com
), with P < 0.05 and a false detection rate < 0.10. (Supplemental Online Material details Material and Methods.)
Table 1 Phenotypic correlations among the traits assessed in this study. Correlation coefficients above the diagonal and case-wise sample sizes below the diagonal. The six correlations significant at P ≤ 0.01 are shown in italicized bold face and the (more ...)
With few exceptions, phenotypic correlations among traits in the F2 are weak or non-existent (). Not all correlations could be determined because some traits (notably lens sizes) were assessed in different siblings but, of the 26 correlations we calculated, only six were significant at the P = 0.01 level, and three others at the P = 0.05 level. Eye size was significantly negatively correlated with three melanophore traits and the number of maxillary teeth and positively correlated with lens size. MelE and MelD were strongly correlated, and the length of the maxillary bone was significantly correlated with the length of the dentary and the number of taste buds. It is notable that eye size was not significantly correlated with the lengths of the dentary or maxillary or number of taste buds.
We detected 48 QTL for these traits: eight affecting eye size, six affecting lens size, 18 affecting pigmentation, seven affecting lengths of the jaw bones, six affecting the number of maxillary teeth, and three affecting the number of taste buds (). The total proportions of variance explained by QTL for each trait ranged from 0.11 to 0.77 (mean = 0.44) and the total proportions of additive variance explained ranged from 0.03 to 0.52 (mean = 0.28).
Table 2 QTL for cave related traits detected in the analysis of a cross between Pachon cave and surface Astyanax mexicanus. Trait and trait means are listed in column 1. LOD scores and P values for SIM and MIM analyses are listed separately. Also listed are explained (more ...)
Some of the QTL co-mapped and might represent the effects of single genes or tightly clustered genes (Supplemental Fig. 2
). On LgP7, LgP9, and LgP15, there were two, four and two co-mapping QTL for different melanophore traits. On LgP8 and LgP20, QTL for eye and lens size co-map. Because of the possibility that these sets of co-mapping QTL each represent single loci, for statistical comparisons of eye and melanophore QTL we counted each region only once. On LgP13 QTL for maxillary size and number of maxillary teeth co-map. While these may represent one gene, the polarities of substitution effects are in disagreement, with smaller maxillae associated with more teeth. On LgP27, QTL for eye size and number of maxillary teeth co-map. On LgP5 and LgP25, QTL for eye or lens size co-map with QTL for taste-buds. Other examples of co-mapping traits can be seen in Supplemental Figure 2
Trait means (μ), and estimates of allelic substitution (d) and heterozygous effects (h) are given in . Expected trait values for cave and surface homozygotes and heterozygotes are μ + d, μ − d, and μ + h, respectively. We calculated trait values for all 48 QTL and identified loci at which the heterozygote fell between the two homozygotes as intermediate in dominance. Based on this criterion, 36 of the cave alleles are of intermediate dominance. The remaining 12 loci cannot be classified unambiguously because the standard errors of estimate sometimes exceeded the differences in trait values among genotypes, but at four of the loci the cave allele seems recessive, at two it seems dominant, and at two more it seems clearly overdominant. We calculated a measure of dominance as the absolute value of the ratio of h/d and found the median value to be 0.44, or semi-dominant.
In order to compare patterns of substitution between eye/lens and melanophore QTL, the two regressive trait classes, we calculated trait values for all three genotypes at each QTL using estimates of d, h, and μ. To standardize the scales, we divided expected trait values by their trait means. In the three cases in which two or more melanophore QTL co-mapped and it was possible that single genes were affecting multiple traits, the scaled trait values were averaged for each genotypic class. This reduced the number of melanophore QTL to 13 for statistical testing. In the two cases where eye and lens QTL co-mapped, we chose the one with the higher LOD score to represent the QTL.
The patterns of substitution effects differ radically between QTL for eye/lens size and melanophore numbers. Cave alleles at all 12 eye/lens QTL effect relatively modest, but steady decreases of eye/lens size (). In contrast, cave alleles at QTL affecting melanophore number have positive (n = 5), as well as negative slopes (n = 8), and their substitution affects are much larger (). The distributions of polarities differ significantly between the two classes of traits (12:0 vs. 8:5, 2-tail P = 0.039, Fisher’s exact test). Comparison of the slopes for the two trait classes () also reveals an obvious difference in dispersion (Wilcoxon two-sample statistic for testing homogeneity of variances, R11,13 = 186, P = 0.005).
Standardized trait values for surface homozygotes (SS), heterozygotes (SC) and cave homozygotes (CC), for eye/lens and Melanophore QTL.
Our interpretation of these differences in effects between the two classes is that regression of eyes came about primarily through selection, while decreases in numbers of melanophores resulted mainly from recurrent mutation/genetic drift or indirectly through pleiotropy. If there were strong direct selection against melanophores, it is unlikely that five QTL, all with major effects, would have cave alleles increasing the numbers of melanophores. If eye/lens reduction were accomplished through genetic drift, it is unlikely that the pattern of effects would contrast so radically with that for melanophores.
If eyes regressed through selection, was the selection directed against the eye itself or was it indirect, through negative pleiotropy of alleles selected for affects on other traits? Hedgehog signaling pathways direct the development of midline structures, including jaws, teeth and tastebuds (reviewed in 5
). Hedgehog activities also have important affects on eye development, in part, because Hh
expression is antagonistic to that of PAX6
and alters patterns of expression of PAX2
. Yamamoto et al.
] have shown through experimental alteration of gene activity in A. mexicanus
embryos that hedgehog activity is a strong determinant of eye size. Increased unilateral expression of sonic hedgehog (shh)
and tiggy-winkle hedgehog
) in surface fish suppresses the development of the treated eye. Thus, one hypothesis is that increased feeding efficiency may be an important adaptation in cave fish, accomplished through up-regulation of hedgehog signaling but at the expense of eye development [6
hypothesis has two parts. The first is that up-regulation of hedgehog activity suppresses development of the eyes; the second is that hedgehog activity was up-regulated during cave fish evolution by selection to improve feeding efficiency and that this was the primary cause of eye regression. The evidence linking hedgehog activity to eye development seems compelling, but our data do not yet provide a definitive test of the second part of the hypothesis, although they suggest it cannot be the sole explanation of eye regression. Six QTL for eye/lens size co-map with QTL affecting feeding traits (jaw bone sizes, numbers of teeth and taste-buds), but six others do not, and the QTL in the latter group control a much greater proportion of explained additive variance than those in the former (not co-map vs
. co-map groups: Eye: 0.233 vs
. 0.087; LensE: 0.364 vs
. 0.070; LensL: 0.014 vs
0.015). Furthermore, it is not just feeding trait QTL and eye/lens QTL that co-map. Feeding trait QTL co-map with QTL for melanophore numbers three times and QTL for eye/lens size and melanophore number co-map four times. We attribute this co-mapping to a general tendency towards pleiotropy with these traits [7
] rather than to any specific relationship between feeding efficiency and eye loss. In addition, if the QTL affecting feeding traits were major contributors to eye regression, we might expect to see strong negative phenotypic correlations between these traits and eye size in the F2
. Such correlations are weak or non-existent (). In sum, definitive tests of the generality of the second part of the Hh hypothesis await the molecular identification of the genes underlying eye loss and feeding morphology, and characterization of the fitness effects of their alleles.
We also mapped candidate genes shh (LgP28), twhh (LgP15) and PAX6 (LgP10). No eye QTL are located near these loci, making it unlikely that mutations in any of them are directly responsible for eye regression. One eye QTL maps to a point near the gene for ocular and cutaneous albinism (OCA2, LgP5).
Is it possible that Darwin’s premise was simply incorrect? Are eyes in a cave disadvantageous, and if so, why? In essence, the argument against selection is that the cost of making an eye is trivial compared to the cost of its replacement tissue in the socket [2
], or that the developmental cost is paid by cave fish anyway because the eyes start developing and only degenerate after many cell cycles of tissue growth and replacement [4
]. However, modern physiology and molecular biology suggest these arguments might address the wrong costs. The vertebrate retina is one of the most energetically expensive tissues, with a metabolism surpassing even that of the brain [8
]. Underscoring this high metabolic demand is the observation that one manifestation of genetic defects decreasing the efficiency of mitochondria is blindness (e.g
., Leber's Hereditary Optical Neuropathy [9
]). Thus, maintenance of eyes might pose a significant burden in the cave environment. Increasing this burden, the vertebrate retina uses more energy in the dark than in the light, because the membranes of the photoreceptor disks must be maintained in the hyperpolarized state until depolarized in response to light [10
]. Oxygen consumption by the vertebrate retina is approximately 50% higher in the dark than in the light [8
]. Adding further to the retina’s cost is its structural maintenance. Ten percent of the photoreceptor outer disks in vertebrates are shed and renewed each day, and the structure may be completely replaced over 35 times yearly [12
Thus, while the energetic cost of making an eye may be trivial, the expense of maintaining one is much greater. In the dark, it may be costly enough to create effective selection for eye regression. In contrast, the argument of metabolic cost cannot be made for regression of pigmentation, and the QTL trait value data () show that the two traits have regressed through different mechanisms.
This study shows that regression may be effected by active selection as well as by the passive accumulation and fixation of damaging mutations, and that the various possibilities can be distinguished by the patterns of allelic substitutions involved. Thus, regression, an integral part of the progress of evolutionary change, can be accomplished in a variety of ways.