Convergence in feeding posture occurs through different genetic loci in independently evolved cave populations of Astyanax mexicanus

Significance

Relatively little is known about the genetic basis of behavioral evolution, in particular for behaviors without any obvious related morphological changes. We have focused on changes in feeding posture that evolved in the small tetra Astyanax mexicanus as it adapted from life in the rivers to the very different ecological conditions found in caves. Using comparative quantitative genetics/genomics, we find that behavioral differences in feeding posture between surface and cave populations arose independently, through different polygenic genetic mechanisms in multiple, independent cave populations. This work provides insights into the genetic architecture of this behavioral trait and shows that this hard-wired behavior can arise through multiple genetic routes.

Abstract

When an organism colonizes a new environment, it needs to adapt both morphologically and behaviorally to survive and thrive. Although recent progress has been made in understanding the genetic architecture underlying morphological evolution, behavioral evolution is poorly understood. Here, we use the Mexican cavefish, Astyanax mexicanus, to study the genetic basis for convergent evolution of feeding posture. When river-dwelling surface fish became entrapped in the caves, they were confronted with dramatic changes in the availability and type of food source and in their ability to perceive it. In this setting, multiple independent populations of cavefish exhibit an altered feeding posture compared with their ancestral surface forms. We determined that this behavioral change in feeding posture is not due to changes in cranial facial morphology, body depth, or to take advantage of the expansion in the number of taste buds. Quantitative genetic analysis demonstrates that two different cave populations have evolved similar feeding postures through a small number of genetic changes, some of which appear to be distinct. This work indicates that independently evolved populations of cavefish can evolve the same behavioral traits to adapt to similar environmental challenges by modifying different sets of genes.

The colonization of caves is an extreme example of a species entering a new environment. Unique attributes of caves relative to the surface environment include darkness, high humidity, relatively constant temperature, absence of predators, and scarcity of food. Under these circumstances, many species of cave animals have evolved a suite of similar traits, including constructive traits such as heightened sensory systems and regressive traits such as loss of pigmentation and reduction in eye morphology (1). To study the evolution of cave-specific traits, we have focused on Astyanax mexicanus, the Mexican cavefish. A. mexicanus exists in two forms, a cave-dwelling form and a river-dwelling surface form. Importantly, these forms are still interfertile (2), allowing one to take a genetic approach using quantitative trait loci (QTL) analysis for the mapping of cave traits. Furthermore, there are multiple, independently evolved cave populations (3–7) that in many cases have evolved similar traits, allowing for the study of convergent evolution.

Populations of cave organisms have often been the subjects of studies on convergence. For example, loss of pigmentation evolved via disruptions in the first step of the melanin synthesis pathway in multiple species of cave organisms (8). Similarly, a decrease in the levels of melanin synthesis arose in multiple cave populations of A. mexicanus through different mutations in the same genes (9, 10). In contrast, crosses between multiple cave populations of A. mexicanus result in embryonic hybrid fish with larger, functional eyes, indicating that evolution of this trait is controlled by different genetic loci in different cave populations (2, 11).

Among the most intriguing and least understood adaptations to caves are behavioral traits. Cave populations of A. mexicanus have evolved an enhanced attraction to a vibration source, loss of schooling behavior, reduction in time spent sleeping, and a loss of aggression (2, 12–16). Little is known about how these behaviors evolve and whether each behavior has evolved through the same or different genes or pathways as an adaptation to cave life.

Here we focus on a convergent aspect of feeding behavior and its potentially related morphological traits found in cave populations of A. mexicanus. Surface fish feed at a steep angle relative to the substrate when feeding in the dark. In contrast, multiple cave populations feed at a much lower angle (17). We examined feeding orientation in three independently evolved cave populations to determine the underlying genetic architecture, including the extent to which this behavior depended on correlated morphological changes, and to determine whether multiple cave populations evolved this alternative feeding posture through changes in the same or different genetic loci.

Results

Characterizing Feeding Posture in Several Independently Evolved Cave Populations.

We developed assays to quantify feeding posture by measuring the feeding angle for individual fish, filming them in the dark, from either above or the side, with an infrared camera (Fig. 1A). We tested whether these two recording methods yield consistent results by measuring feeding angle from the top and from the side simultaneously. These two measurements were highly correlated (R² = 0.944; interclass correlation: r = 0.964, P < 0.0001, n = 24), which allowed us to treat both methods as essentially the same. By using these methods, we measured feeding posture in surface fish, and in three independently evolved populations from different caves, Pachón, Tinaja, and Molino. There were significant differences in feeding angle among groups (ANOVA: F_6,194 = 25.4, P < 0.001). Although surface fish fed at a high angle (average 74°), two populations of cavefish, the Pachón cavefish and the Tinaja cavefish, fed at significantly lower angles on average (38° and 49°; Games–Howell: P < 0.001 and P < 0.001 respectively; Surface n = 43, Pachón n = 29, Tinaja n = 46; Fig. 1 B–D). However, the third population of cavefish, from the Molino cave, which evolved from an evolutionary younger population of surface fish than the Pachón and Tinaja cavefish (6), fed at an angle similar to surface fish (66°, Games–Howell: P = 0.253; Molino n = 24; Fig. 1D). Thus, cavefish populations can differ from each other in feeding posture and a lower feeding posture has likely evolved independently in at least two cave populations.

Fig. 1.

Genetics of feeding angle in multiple, independently evolved cave populations. (A) Schematic of two methods for filming feeding posture: top or side. The surface/Tinaja cross was filmed from the side, whereas the surface/Pachón cross was filmed from above. Feeding posture in surface (B) and a Tinaja cave (C) fish filmed from the side. Feeding posture was measured by drawing a line that bisects the head and measuring the angle relative to the bottom of the tank. (D) Distribution of feeding posture in surface fish (n = 43), surface/Tinaja F₁ hybrid fish (n = 4), surface/Pachón F₁ hybrid fish (n = 30), Tinaja (n = 46), Pachón (n = 29), Molino (n = 24), and Pachón/Tinaja F₁ hybrid fish (n = 25). (E) Distribution of 267 F₂ fish from a surface/Pachón F₁ intercross. (F) Distribution of 226 F₂ fish from a surface/Tinaja F₁ intercross. *P < 0.05, **P < 0.01, ***P < 0.001.

Genetics of Feeding Posture.

To investigate the genetic basis of feeding posture, we performed multiple crosses and measured feeding posture in the resulting progeny. F₁ hybrid fish from both surface/Tinaja and surface/Pachón crosses exhibited average feeding angles similar to surface fish, indicating that a surface-like phenotype is dominant (76° and 71°; Games–Howell: P = 0.998, P = 0.958, respectively; surface/Tinaja F₁ fish n = 4, surface/Pachón F₁ fish n = 30; Fig. 1D). F₁ fish from both of these crosses were significantly different from their parental cave population (Games–Howell: surface/Tinaja F₁ fish P < 0.001, surface/Pachón F₁ fish: P < 0.001; Fig. 1D). In addition, feeding posture was similar in surface/Pachón F₁ hybrid fish regardless of which parent was a surface fish and which parent was a Pachón cavefish (73° and 68°; independent t test, t₂₈ = 0.87, P = 0.39; Pachón-mothered F₁: n = 14, surface-mothered F₁: n = 16).

To determine whether feeding posture evolved by the same genes in these two cave populations, we crossed Pachón and Tinaja fish and measured feeding angle in F₁ hybrid fish. If the genes controlling lower feeding posture in these two populations were the same, hybrid fish should have a phenotype similar to the parental cave populations. If the genes controlling feeding posture were different in the two populations, hybrid fish should show complementation, displaying a surface-like phenotype. Pachón/Tinaja F₁ hybrid fish ranged in their behavior from surface-like to cave-like feeding postures (Fig. 1D). Furthermore, this F₁ population was significantly different from surface (Games–Howell: P < 0.01, Pachón/Tinaja F₁ fish n = 25) and Pachón fish (Games–Howell: P < 0.01), but not significantly different from Tinaja fish (Games–Howell: P = 0.394). Pachón and Tinaja fish populations were significantly different from one another (Games–Howell: P < 0.05). Lack of complete complementation in the Pachón/Tinaja F₁ hybrid fish data suggests that some of the genetic changes controlling evolution of feeding posture in these two caves were shared, or resulted from independent, but additive genetic changes. However, differences between the Pachón and Tinaja fish populations, and between Pachón/Tinaja F₁ hybrid fish and Pachón fish, suggest that at least one genetic change resulting in a lower feeding angle was different between these two populations. These data suggest that feeding posture evolved independently in these two cave populations. Furthermore, these data are consistent with previous results demonstrating that these two cave populations evolved independently (6, 11).

F₂ fish were generated by intercrossing F₁ hybrid fish. F₂ population distributions of feeding angle from the surface/Tinaja and surface/Pachón fish both have peaks weighted toward the mean angle observed in surface fish with tails extending over the angle seen in cave populations (surface/Pachón F₂ fish n = 267, surface/Tinaja F₂ fish n = 226; Fig. 1 E and F). These distributions suggest that this is a multigenic trait in both cave populations.

QTL Mapping of Feeding Posture.

We performed QTL mapping on F₂ fish from a surface/Pachón F₁ intercross and from a surface/Tinaja F₁ intercross. We found two significant QTL at a P value of 0.05 in the surface/Pachón cross, one on linkage group (LG) 6 and one on LG 17 that explained 7 and 5.7% of the variance in feeding angle, respectively (Fig. 2 A–E and Table 1). Homozygous cave genotypes at markers under both QTL resulted in a decrease in feeding angle, with heterozygous genotypes resulting in an intermediate feeding phenotype (Fig. 2 C and E).

Fig. 2.

QTL mapping of feeding angle, jaw angle, eye size, and head depth in a surface/Pachón cross. (A) Genome-wide logarithm of odds (LOD) scores from a scanone QTL for feeding angle (black), jaw angle (red), eye size (dark green), and head depth (light green). Solid horizontal lines indicate a P value cutoff of 0.05 with the color corresponding the color for the trait. The x axis indicates genetic distance in each linkage group. (B and D) LOD traces for feeding angle QTL on LG 6 and LG 17, respectively. LOD scores computed following multiple QTL mapping in these and in F, H, and M. (C and E) Effect plot of phenotypic values of feeding angle against each genotype (mean ± SEM) at the peak locus on LG 6 and 17, respectively. (F) LOD trace for jaw angle QTL on LG 3. (G) Effect plot of jaw angle QTL on LG 3. (H) LOD traces for eye size QTL on LG 2, 3, 17, 19, and 22. (I) Effect plot of eye size QTL on LG 2. (J) Effect plot of eye size QTL on LG 3 and 17 showing epistatic interaction. Homozygote of cavefish alleles at both loci promotes strong eye reduction. (K) Effect plot of eye size QTL on LG 19. (L) Effect plot of eye size QTL on LG 22. (M) LOD traces for head depth QTL on LG 20. (N) Effect plot of head depth QTL on LG 20. For C, E, G, I–L, and N, CC stands for homozygous cave genotypes, SS for homozygous surface genotypes, and SC for heterozygous genotypes. For B, D, F, H, and M, red shaded area indicates a 95% confidence interval using Bayesian credible intervals with probability coverage as 0.95 for the QTL.

Table 1.

QTL analysis

Trait	Cross	LG-P	LG-T	cM	CI	PVE	LOD	alpha	n
Feeding angle	S/P	6	11	35.6	21–42	7.0	4.5	0.05	267
		17	13	49	34–73	5.7	3.7
Eye size	S/P	2	1	5	4–7	7.6	8.7	0.05	357
		3	2	30	24–32	12.1	13.4
		17	13	47	46–48	15.0	16.3
		19	6	43	33–43	3.1	3.7
		22	7	29	18–41	3.8	4.5
		3,17*	2,13*	30,47*	—	9.3	10.5
Jaw angle	S/P	3	2	63	55–72	8.3	5.1	0.05	273
Head depth	S/P	20	24	4	0–20	15.6	6.0	0.05	162
Feeding angle	S/T	20	24	36	20–52	10.8	3.9	0.05	170
Tastebuds – lateral	S/T	25	18	123	114–124	10.9	4.0	0.05	158
Tastebuds – ventral	S/T	2	1	148	76–154	11.6	4.1	0.05	153
Orbit	S/T	11	3	74	70–84	7.8	5.5	0.05	212
		19	6	74	68–75	11.2	7.6
		15	9	35	34–36	7.0
		13	16	88	78–96	5.6
		12	20	12	4–24	6.6
Body depth	S/T	17	13	83	78–89	15.3	5.5	0.05	152
Head depth	S/T	8	10	34	42–122	10.1	3.8	0.15	153
Jaw width	S/T	22	7	61	0–70	7.2	3.4	0.15	211
Maxillary tooth number	S/T	1	21	0	0–46	9.1	3.3	0.15	161

Alpha, significance cutoff for QTL (alpha = 0.05) and suggestive QTL (alpha = 0.15); CI, confidence intervals for the position of the QTL using a 95% Bayesian credible interval; cM, centimorgan at the location of the highest LOD peak; LG-P, linkage groups of the surface/Pachón cross; LG-T, linkage groups of the surface/Tinaja cross; LOD, highest LOD peak at the location from scanone; n, number of individuals used for mapping; PVE, percentage of the phenotypic variance explained; S/P, Surface/Pachón cross; S/T, Surface/Tinaja cross.

^*Epistatic interaction between two QTL.

We found one significant QTL at a P value of 0.05 in the surface/Tinaja cross on LG 24 that explained 10.8% of the variance in feeding angle (Fig. 3 A and C and Table 1). Homozygous cave genotypes at the marker under this QTL resulted in an increase in feeding angle (Fig. 3D). The heterozygous genotype had a similar feeding angle to the homozygous cave genotype.

Fig. 3.

QTL mapping of feeding angle, taste buds, orbit diameter, and body depth in a surface/Tinaja cross. (A) Genome-wide LOD scores from scanone QTL for feeding angle (black), lateral taste bud number residuals (dark blue), and ventral taste bud number residuals (light blue). Lines indicates P value cutoff of 0.05 for each trait derived from 1,000 permutations. (B) Genome-wide LOD score from scanone for feeding angle (black), body depth residuals (light green), and orbit diameter residuals (dark green). Lines indicate P value cutoff of 0.05 for each trait derived from 1,000 permutations. (C) LOD traces for feeding angle QTL on LG 24. (D) Effect plot for feeding angle QTL. (E) LOD traces for lateral taste bud residuals QTL on LG 18. (F) Effect plot for lateral taste bud residuals on LG 18. (G) LOD traces for ventral taste bud residuals QTL on LG 1. (H) Effect plot for ventral taste bud residuals on LG 1. (I) LOD traces for body depth residuals QTL on LG 13. (J) Effect plot for body depth residuals on LG 13. (K) LOD traces for orbital diameter residuals QTL on LGs 3, 6, 9, 16, and 20. (L) Effect plot for orbital diameter residuals on LG 3. (M) Effect plot for orbital diameter residuals on LG 6. (N) Effect plot for orbital diameter residuals on LG 9. (O) Effect plot for orbital diameter residuals on LG 16. (P) Effect plot for orbital diameter residuals on LG 20. For D, F, H, J, and L–P, CC stands for homozygous cave genotypes, SS for homozygous surface genotypes, and SC for heterozygous genotypes. For C, E, G, I, and K, red shaded area indicates the 95% confidence interval. Note that the additional regions are in K are from the stepwise QTL results.

To determine whether the QTL were present in the same or different locations in these two crosses, we aligned the Pachón and Tinaja linkage groups containing QTL by using the latest A. mexicanus genomic assembly (Table S1). Pachón LG 6 displayed synteny with Tinaja LG 11, and Pachón LG 17 displayed synteny with Tinaja LG 13 (Fig. S1). In addition, Tinaja LG 24 displayed synteny with Pachón LG 20 (Fig. S2). None of these linkage groups shared QTL, even at a lenient P value of 0.1. Thus, it appears that we identified distinct QTL, representing different loci regulating feeding angle in each cavefish population.

The Evolution of Feeding Posture and Morphological Evolution.

In addition to behavioral changes, cavefish have evolved a variety of morphological traits to adapt to cave life. Many of these morphological adaptations are found in multiple, independently evolved cave populations. Therefore, it was possible that feeding posture evolved as a result of one or more of these morphological traits, or that an altered feeding posture changes the morphology of the cranial facial skeleton, similar to what is seen in other fish species (18). To test this association, we quantified candidate morphological traits that have evolved in cave populations and compared their inheritance to feeding posture. We quantified taste bud number, orbital opening diameter and eye size, the angle of the jaw, maxillary tooth number, lower jaw width, the depth of the head, and the depth of the body, all of which differed in at least one of the cave populations relative to surface fish (Figs. S3 and S4 and Table 2). For those traits that scaled with body size, we corrected for size by using the residuals from a regression between the trait and a measure of body size (Materials and Methods and Fig. S5). We found that Tinaja and Pachón populations often both differed from surface fish in these traits. Although Molino fish were also occasionally different from surface fish, this population was often more surface-like in phenotype compared with the other two cave populations (Supplemental Figs. S3 and S4 and Table 2). Surface/cave F₁ hybrids were often either intermediate between cave and surface fish, or surface-like in phenotype, indicating that the surface phenotypes were often dominant (Figs. S3 and S4 and Table 2). The distribution of the F₂ fish for all of these traits indicated that these were multigenic traits (Figs. S3 and S4). Finally, when we compared each of these traits to feeding posture in the F₂ population, we did not find significant correlations between the majority of these traits and feeding posture (Figs. S3 and S4 and Table 2). One exception was ventral taste bud number, which was positively correlated with feeding posture and close to significance (P = 0.051; Fig. S3 and Table 2). The strength of this correlation was increased when we included the total number of taste buds (R = 0.27, P < 0.05, n = 75). However, because more taste buds correlated with a higher feeding posture, an increase in number of taste buds cannot explain the lower feeding posture seen in cave populations.

Table 2.

Morphological traits

Trait	Mean S	S/P	S/T	T	P	M	R	P	n
JA	47.1 ± 5	44.5 ± 3.7	—	43.7 ± 3.5	37.4 ± 4	42.9 ± 4.3	0.03†	0.61	231
	n = 38	n = 10 n.s.		n = 22*	n = 21***	n = 10 n.s.
VTB	−155 ± 101	—	−104 ± 87	205 ± 147	165 ± 105	−57 ± 61.5	0.21	0.051	83
	n = 14		n = 4 n.s.	n = 10***	n = 9***	n = 10 n.s.
OD	0.86 ± 0.29	—	0.09 ± 0.23	−1.52 ± 0.33	−0.89 ± 0.33	−0.92 ± 0.42	−0.06	0.54	119
	n = 25		n = 3 n.s.	n = 18***	n = 9***	n = 10***	0.02†‡	0.75	266
JW	−0.52 ± 0.38	—	−0.58 ± 0.8	0.59 ± 0.66	−0.13 ± 0.18	0.25 ± 0.52	0.03	0.77	118
	n = 25		n = 3 n.s.	n = 18***	n = 9**	n = 10**
MT	0.7 ± 0.9	—	2.8 ± 1.6	5.3 ± 1.4	4.3 ± 1.3	4.6 ± 1.1	0.19	0.06	95
	n = 9		n = 5 n.s.	n = 8***	n = 9***	n = 12 ***
BD	−1.33 ± 0.74	—	0.32 ± 0.92	−0.21 ± 0.64	0.24 ± 0.68	−1.19 ± 1.1	0.02	0.84	83
	n = 16		n = 3 n.s.	n = 17**	n = 10***	n = 11 n.s.
HD	0.36 ± 0.29	—	0.18 ± 0.56	−0.85 ± 0.44	−0.47 ± 0.29	−0.008 ± 0.46	0.01	0.93	82
	n = 16		n = 5 n.s.	n = 17***	n = 10***	n = 11 n.s.	−0.09†	0.26	156

BD, body depth residuals; HD, head depth residuals; JA, jaw angle; JW, jaw width residuals; MT, maxillary teeth; OD, orbit diameter residuals; VTB, ventral taste bud residuals. The mean ± SD are given for the number of individuals measured (n) from the surface (S), surface/Pachón F₁ hybrid (S/P), surface/Tinaja F₁ hybrid (S/T), Tinaja (T), Pachón (P), and Molino (M) populations. Asterisks indicate a significant difference from surface fish. *P < 0.05, **P < 0.01, ***P < 0.001, n.s., not significant. Correlations were measured by using a Pearson correlation, in the F₂ populations. Shown are the R, P value, and number of F₂ individuals.

^†The correlation analysis was performed in the surface/Pachón cross. The rest of the correlations were measured in the surface/Tinaja cross.

^‡Eye diameter instead of orbit diameter was analyzed (also see Fig. S4F).

We also mapped the morphological traits described above to determine whether there was overlap with the behavioral QTL we identified. The QTL we identified for morphological traits were all of small effect, and most were in the direction we would expect given the cave and surface phenotypes (Figs. 2 and 3 and Table 1). Most of the QTL for morphological traits did not map to the same location as the QTL for feeding angle in the crosses they were measured in, consistent with the results of our correlation analyses with morphological traits (Figs. 2 and 3 and Table 1). Although one QTL for eye size maps to the same region as a QTL for feeding angle in the surface/Pachón cross, it is unlikely that eye size plays a large role in the evolution of feeding angle in Pachón fish given the lack of correlation between these traits in the F₂ fish from this cross (Fig. S4 and Fig. 2A). These data suggest that evolutionary changes in cranial facial morphology and sensory systems are unlikely to be responsible for the differences in feeding posture observed between some cavefish populations and surface fish.

We compared the QTL from the Tinaja cross with the morphological QTL described here and elsewhere (refs. 19 and 20) and Fig. 3A) from a Pachón cross. We found that the Tinaja taste bud and maxillary tooth QTL were not in the same regions as those in Pachón fish (19, 20). We found that only one eye size QTL, the Tinaja LG 6 QTL for orbit diameter, was located at the syntenic region of the Pachón eye size QTL at LG 19 (Figs. 2 H and K and 3L). These data are consistent with the finding that Pachón and Tinaja fish share only a small number of eye loci (11).

Discussion

Convergence in Feeding Posture in Two Cave Populations Is Controlled by at Least Some Different Genetic Loci.

QTL analysis indicates that evolution of feeding posture is controlled by multiple genetic loci, some of which are not shared between Pachón and Tinaja cave populations. Consistent with these data, hybrid individuals in a Pachón/Tinaja cross have an intermediate phenotype, significantly different from the Pachón parental population.

It is likely, however, that some loci are shared between the two cave populations, which result in some Pachón/Tinaja hybrid fish with a low feeding angle. We may not have identified loci that are similar between these populations in our QTL mapping because they control only a small amount of the variance of this trait. The amount of variance explained by our QTL was small and may be inflated because of the small sample size used for mapping (21). Therefore, there are potentially QTL we are missing in this study. With more individuals one might identify overlapping genetic loci and, hence, detect direct evidence for parallel evolution of this trait.

In this study, we used different surface individuals, from different surface populations, one from Mexico, and one from Texas, as parental fish for our mapping populations. It is possible that differences identified here are due to different alleles between surface fish, rather than differences between cavefish. However, our complementation test indicates that there are some genetic differences controlling feeding posture between Pachón and Tinaja fish. Therefore, it is likely that the differences we identify by QTL analysis are due to these differences between cave populations.

Feeding Posture Is Not Controlled Solely by Evolution of Morphological Traits.

There is evidence in A. mexicanus that some morphological traits evolved through changes in the same genes in multiple cave populations (9, 10). In addition, there is evidence for coevolution of behavioral and morphological traits through the same genetic loci, for example, neuromast number and vibration response (22). Morphological changes can, in principle, affect behavioral traits, either because morphological traits are themselves important for the behavior, or through pleiotropic effects of the underlying genes. Additionally, feeding behavior can alter cranial facial morphological structures in other fish species through phenotypic plasticity (18). However, in the case of feeding posture, we did not find a strong correlation with morphological traits, such as altered craniofacial morphology, body depth, and distribution and number of taste buds, either through correlations in F₂ fish or from QTL mapping.

We found overlapping QTL for eye size and feeding posture in the surface/Pachón cross. Although this congruence may be the result of a single, pleiotropic locus that contributed to the evolution of both feeding posture and eye size, loss of the eyes in cavefish alone cannot explain the change in feeding posture we observe here. First, the feeding posture QTL located at the eye size QTL explains a small amount of the variation in feeding posture (5.7%). Second, we do not see a significant correlation between feeding posture and eye size in the surface/Pachón cross. However, it is possible we do not have the power to detect this correlation given the small effect size of this QTL. Third, we see no correlation or overlap with orbit QTL in feeding posture in the surface/Tinaja cross. Fourth, we had previously measured eye size in the surface/Tinaja cross (23), and we found no correlation with eye size and feeding posture in this cross (R = −0.09, P = 0.27, n = 166), and the eye size QTL identified in this cross lies on LG 3, not overlapping with the feeding posture QTL. Finally, Molino cavefish also have lost eyes, yet feed similarly to surface fish.

We also found that increased number of taste buds was correlated with increased feeding posture. It is unlikely that this change in morphology could explain the change in behavior, because it is in the opposite direction from what we see in cave populations relative to surface populations. Furthermore, a lack of overlapping QTL for these traits makes it unlikely that this correlation is caused by the same gene or genes responsible for both of these traits. However, it would be interesting to investigate the cause of this correlation in future work.

All three cave populations share a similar degree of divergence from surface populations in many of the morphological traits we quantified here. This similarity of divergence provides additional evidence that these morphological traits are not, in and of themselves, sufficient to alter feeding posture. However, because we do not have a large sample size for QTL mapping, the explained variance for the identified QTL could be overestimated (21) and it is possible that we are missing additional QTL. These additional feeding angle QTL, undetected here, could overlap with morphological QTL. Additionally, because we did not survey the same set of morphological traits in both Tinaja and Pachón populations, it is possible that morphological traits measured in one population are correlated with feeding posture in the other population. It is also possible that other morphological changes, not quantified here, played a role in the evolution of feeding posture. It will be interesting to pursue this topic in the future, through the use of morphometric analysis. However, a number of obvious candidate morphological traits had little to no correlation with feeding posture, supporting the alternate hypothesis that feeding posture evolved through changes in the nervous system itself.

Evidence for Adaptive Nature of Feeding Posture in the Cave Environment.

Two of the three independently evolved cave populations we tested fed at a lower angle compared with surface fish. The Molino cave population, which displayed a feeding posture similar to surface fish, is an evolutionarily young population of cavefish (6), which retains some surface-like traits that have subsequently been lost in older cave populations, as shown here and previously (24). Some ecological factors in the Molino cave may play a role in preventing the evolution of feeding posture. However, if there is an adaptive advantage for an altered feeding posture, Molino fish may not have had enough time to evolve this posture.

Although Pachón and Tinaja fish evolved independently, they have converged on similar feeding postures. Moreover, this behavioral change does not appear to be a mere secondary consequence of morphological alterations that arise for some other reason. Together, this evidence suggests some adaptive reason to evolve feeding posture.

Cavefish must forage in caves, many of which have a limited amount of food. Furthermore, they must identify food in an environment independent of sight. That they have successfully adapted to these challenges is indicated by experiments where surface fish are reintroduced to cave conditions. These experiments indicate that cavefish are better at finding food in the dark than surface fish (25). In addition, surface fish found trapped in the cave show signs of starvation (3). Therefore, it is likely that an altered feeding posture evolved as a result of selection for improved foraging. For example, tactile cues resulting from more contact with the substrate may aid in successful foraging in the dark. Other behavioral and morphological traits have been implicated in the evolution of foraging and finding food in the dark (12, 14, 16). Furthermore, differences in feeding posture have been shown to be important for different feeding strategies in other species of fish (for example, refs. 26 and 27). Together, these highlight the complexity of the evolutionary response of A. mexicanus to the entrapment in the extreme cave environment.

Materials and Methods

All animal procedures were in accordance with the guidelines of the National Institutes of Health and were approved by the Institutional Animal Care and Use Committee at Harvard Medical School or the University of Maryland Animal Care and Use Committee. Feeding posture was assayed in the dark, filming from either above or from the side. The effects of body length and sex on feeding posture were measured in the F2 population (Fig. S5). For additional details on the behavioral assay, see SI Materials and Methods. Details of the fish crosses, quantification of morphological traits, statistics, genotyping and QTL analysis, and alignment analysis can be found in SI Materials and Methods.