The cichlid oral and pharyngeal jaws are evolutionarily and genetically coupled
Macro- and micro-evolutionary integration between jaw complexesWe examined phenotypic associations between the lower oral and pharyngeal jaws (LOJ and LPJ, respectively) of 88 cichlid species from across Africa, primarily sampling from lakes in the East African Rift Valley: lakes Malawi, Tanganyika, and Victoria (Supplementary Data 1). We characterized jaw shapes based on 107 individuals using 3D geometric morphometrics by placing landmarks in positions that capture functionally (e.g., bony processes, sutures, etc.) and developmentally (e.g., distinct cellular origins) important aspects of morphology, including placing mirrored landmarks across midlines to gain symmetric configurations (Fig. 1e, Supplementary Fig. 1). We conducted a Procrustes superimposition, removed the effects of allometry to account for size differences, and then removed the effects of asymmetry to account for developmental noise. We performed a two-block partial least squares (PLS) analysis on the species mean landmark configurations and corrected for phylogenetic non-independence using a Bayesian time-calibrated tree31. We found the LOJ and LPJ were evolutionarily correlated (r-PLS = 0.482, P = 0.002, effect size (Z) = 2.585), but some taxa, particularly those with unique diets and/or modes of feeding, appeared to deviate from the best-fit line, indicating lower levels (or different patterns) of integration between jaws (Fig. 2a). Indeed, we found numerous taxa, typically from Lake Malawi, whereby covariation between the LOJ and LPJ appeared much different relative to other African cichlids. Taxa placed far from the best-fit line either (1) exhibited a specialized feeding morphology to better exploit an foraging niche shared with many competitors (i.e., Labeotropheus, algae; Copadichromis, zooplankton; Taeniolethrinops, insects), or (2) exhibited a specialized feeding morphology to take advantage of a more challenging food source (i.e., Trematocranus, snails). However, not all taxa that consume specialized prey were far from the best-fit line; Pungu, (primarily a sponge-feeder) and Perissodus (a scale-feeder), while exhibiting specialized feeding apparatuses to consume such prey, exhibited a relationship between their LOJ and LPJ that was in-line with other African cichlids (Supplementary Fig. 2). We also noted, that while Malawi cichlids exhibit a range of LOJ-LPJ relationships (from weak to strong), most Tanganyikan cichlids reside close to the best-fit line. However, when we examine the strength of integration in the Tanganyika group (n = 29, r-PLS = 0.698, P = 0.001, Z = 2.954) and Malawi group (n = 40, r-PLS = 0.541, P = 0.020, Z = 2.155), despite Tangyanika cichlids exhibiting higher Z-scores, consistent with stronger integration, a statistical comparison between groups finds no significant difference (Z pairwise = 1.188, P = 0.235). Comparisons between Tanganyikan and Malawi cichlids should not be influenced by sampling bias, as principal components analyses (PCA) on the LOJ and LPJ landmark data (Supplementary Data 2 and 3) showed that our sampling of Tanganyikan cichlids includes many species with extreme morphologies that reside at the outer edges of LOJ and LPJ morphospace (Supplementary Fig. 3). Indeed, cichlids from Lake Tanganyika exhibited similar LOJ morphological disparity (Malawi Procrustes variance (PV) = 0.074; Tanganyika PV = 0.057, P = 0.253) and greater LPJ morphological disparity (Malawi PV = 0.015; Tanganyika PV = 0.023, P = 0.012), relative to cichlids from Lake Malawi. Taken together, this indicates that while Tanganyikan cichlids exhibit comparable (i.e., LOJ), or greater (i.e., LPJ) morphological variation compared to Malawi cichlids, covariation between LOJ and LPJ shapes was generally similar between groups.Fig. 2: Phylogenetic two-block partial least squares analysis to assess macroevolutionary associations between lower oral and pharyngeal jaws.a Jaw shape associations across a broad sample of African cichlids (n = 88). Taxa from Lake Malawi are placed into two groups based on phylogenetic position: an mbuna ‘rock-dwellers’ group, and a non-mbuna group consisting of the utaka ‘sand-dwellers’ alongside other benthic species88. b Jaw shape associations across the Tropheops sp. species complex from across a depth gradient (n = 22). Oral and pharyngeal jaw wireframes denote morphologies at either end of the correlational axis. Source data are provided as a Source Data file.Full size imageWe next investigated the degree of integration at lower taxonomic levels. First, we analyzed the jaws of cichlids within the Tropheops species complex from Lake Malawi that is diverse and known to partition habitat by depth32,33. While Tropheops exhibited strong integration between jaws in on our macroevolutionary assessment, species within this genus occupy a broader niche space. Investigating integration within such a species complex provided an opportunity to understand whether habitat differences could lead to differences in integration between jaw complexes. Using the same landmarking procedure as described above, we characterized shape variation in the LOJs and LPJs of 22 wild-caught Tropheops taxa from 60 individuals, concentrating on members from localities across the southern portion of Lake Malawi (Supplementary Data 4). We again performed a two-block PLS analysis on the mean landmark configurations and accounted for phylogenetic non-independence using an amplified fragment length polymorphism tree33. Again, we found the LOJ and LPJ were evolutionarily correlated (r-PLS = 0.795, P = 0.006, Z = 2.521), indicating jaw integration does not appear to vary by habitat (Fig. 2b).Finally, we measured and compared integration between a species pair that exhibited relatively strong versus weak covariation between LOJ and LPJ shapes in our macroevolutionary assessments, Tropheops sp. ‘red cheek’ (TRC, relatively stronger integration) and Labeotropheus fuelleborni (LF, relatively weaker integration). Using the same landmarking protocol we performed separate two-block PLS analyses between LOJs and LPJs of LF and TRC (Supplementary Data 5). Notably, we found strong and significant integration between jaw complexes in TRC (n = 11, r-PLS = 0.957, P = 0.001, Z = 3.038; Fig. 3a) relative to LF (n = 17, r-PLS = 0.669, P = 0.22, Z = 0.794; Fig. 3b). Further, we found the effect sizes of jaw integration within TRC and LF to be statistically distinct (Z pairwise = 1.678, P = 0.047). Altogether, our data support the assertion that the LOJ and LPJ are evolutionarily integrated at multiple taxonomic levels, but they also appear to indicate that certain taxa, such as Labeotropheus, can more readily generate adaptive morphological variation in each jaw complex independently.Fig. 3: Two-block partial least squares analysis to assess microevolutionary associations between lower oral and pharyngeal jaws.a Shape associations among Tropheops sp. “red cheek” (TRC) individuals (n = 11). b Shape associations among Labeotropheus fuelleborni (LF) individuals (n = 17). c Shape associations among members of a hybrid cross between TRC and LF (n = 409). Oral and pharyngeal jaw wireframes denote morphologies at either end of the correlational axis. Source data are provided as a Source Data file.Full size imageGenetic basis for oral and pharyngeal jaw shape covariationTo understand whether phenotypic covariation between the LOJ and LPJ is genetically determined we performed a quantitative trait loci (QTL) analysis to identify prospective genomic regions involved in jaw shape variation for both the LOJ and LPJ. Specifically, we extended an existing genetic cross between the more strongly integrated TRC and the more weakly integrated LF to the F5 generation. Details of the pedigree may be found in34 and in the supplement. For this experiment, we genotyped 636 F5 hybrids and produced a genetic map containing 812 single-nucleotide polymorphisms (SNPs) spread across 24 linkage groups (Supplementary Data 6). With a total length of 1431 cM, our high-resolution linkage map contained a marker every 1.83 cM, on average, allowing us to leverage the increased number of recombination events that occurred to reach an F5 population. We then characterized LOJ and LPJ shape in 409 F5 hybrids using the same landmarking scheme described above, and performed a two-block PLS analysis. In concordance with our findings from natural populations, we documented an association between jaw complexes in this laboratory pedigree (r-PLS = 0.491, P = 0.001, effect size = 6.189; Fig. 3c).We next performed a PCA on the hybrid landmark configurations to distill the data down to a series of orthogonal axes that best explain shape variation among individuals. We extracted the first two PCs from the LOJ and LPJ as each axis represented more than 10% of the shape variation (Supplementary Data 7; Supplementary Figs. 4 and 5). The first axis of the LOJ reflected more general variation in depth, width, and length of the element (41.8% of variation), while the second axis reflected more specific variation in the length of the ascending arm of the articular––the process for which jaw closing muscles attach (12.7% of variation). The first axis of the LPJ reflected width, length, and wing process size (33.7% of variation), while the second axis reflected depth and the size of the anterior keel – the process for which the pharyngeal jaw pharyngohyoideus muscle attaches and controls jaw adduction (14.2% of variation). We then utilized these PC scores as traits to run in our QTL analyses to investigate the genetic basis for variation in these structures.QTL mapping implicates pleiotropic control of LOJ and LPJ shape variationIntegration between LOJ and LJP shapes in the F5 predicts that this pattern of covariation will be reflected in the genotype-phenotype map. Specifically, we predict that we will find overlapping QTL for both jaws. We used a multiple QTL mapping (MQM) approach to test this prediction. Specifically, we performed QTL scans for all four traits and quantitatively assessed the evidence for significant QTL marker(s) using a permutation procedure that reshuffles the phenotypic data relative to genotypic data 1000 times to generate a null distribution, disassociating any possible relationship between genotype and phenotype, to then compare the empirical distribution against35. Once candidate QTL markers were identified, we calculated an approximate Bayesian credible interval to determine the region in which a potential candidate locus would reside. We uncovered a total of five QTL for LOJ traits, and four QTL for LPJ traits (Fig. 4a; Supplementary Data 8). While most QTL localize to different linkage groups, we also identified some QTL that colocalized. Two traits (LOJ PC1, LPJ PC1) share a marker on LG4, while three traits (LOJ PC1, LOJ PC2, LPJ PC1) colocalized to the same markers on LG7. These data are consistent with pleiotropy on LG7 and possibly LG4.Fig. 4: Genetic analyses to identify regions of the genome responsible for major changes in jaw shape.All plots are based on 409 LFxTRC F5 hybrids. a QTL analysis to identify positions in the genome most associated with each trait. b Pleiotropy analysis on linkage group seven to determine whether the oral jaw PC1 trait colocalizes to the same region as the pharyngeal jaw PC1 trait. Significance was determined using a likelihood ratio test (LLRT). c Pleiotropy analysis on linkage group seven to determine whether the oral jaw PC2 trait colocalizes to the same region as the pharyngeal jaw PC1 trait. Significance was determined using a LLRT. d Fine mapping all traits across the entirety of LG7. Values furthest from 0 reflect the largest differences between hybrids with LF and TRC genotypes at a given marker. We find peak genotype-phenotype association at ~50 mb that coincides with our Bayes credible interval (grey bar). Intervals that surround the average phenotypic effect line denote standard error of the mean. e Fine mapping all traits across the Bayes credible interval. Population level genetic diversity (FST) data are applied to the map (black dots) with the opacity of each SNP dependent on the degree of segregation between LF and TRC, with those falling above an empirical Z-score threshold of 0.6 determined to be significant, and those above 0.9 deemed highly significant (green lines). Within the credible interval there are four SNPs with FST values of 1.0, but a single SNP that falls within a genotype-phenotype peak residing within an intron of dym (black circle). Source data are provided as a Source Data file.Full size imageWe then quantitatively assessed the evidence for pleiotropy using a likelihood ratio test (LLRT) to compare the null hypothesis of a common pleiotropic QTL to the alternative hypothesis that they are affected by separate QTL36,37. The overlap on LG4 at a single marker (43.57 cM) was deemed significant (LLRT = 1.85, P = 0.03), indicating that we can reject the null hypothesis and that these peaks likely represent separate QTL for each trait (Supplementary Fig. 6). The three traits that overlap on LG7 spanned two markers (19.12 cM–28.04 cM) and were all deemed non-significant (LOJpc1-LPJpc1: LLRT = 0.02, P = 0.66, Fig. 4b; LOJpc2-LPJpc1: LLRT = 0.20, P = 0.41, Fig. 4c), leading us to accept the null hypothesis and conclude that this interval likely contains a single pleiotropic QTL. Whether a single gene, or multiple closely linked genes drive this pleiotropic signal requires a fine-mapping approach.Notably, this locus on LG7 has been implicated previously in underlying LOJ and LPJ shape in another Lake Malawi cichlid cross between LF and Maylandia zebra38,39. Maylandia species, like Tropheops, were generally more integrated in our macroevolutionary analysis (Fig. 2a), and thus another cross between LF and a species with higher integration values point to the same locus. This suggests that the genetic mechanism of integration may be conserved.Fine mapping identifies two candidate genes critical for bone formationTo gain insights into which gene(s) may be pleiotropically regulating LOJ and LPJ jaw shape variation on LG7 we constructed a fine map with greater marker density to investigate genotype-phenotype associations with greater resolution. To that end, we anchored QTL intervals to particular stretches of physical sequence of the Maylandia zebra genome40. We then identified additional RAD-seq SNPs across the linkage group of interest and genotyped them in the F5. Based on this we created two fine maps: the first spanned the entirety of LG7 group with an average spacing of around one marker every 490 kb (Supplementary Data 9), the second matched the QTL marker range revealed by the Bayesian credible interval analysis with an average spacing of around one marker every 180 kb (Supplementary Data 10). We also calculated FST from a panel of wild-caught LF (n = 20) and TRC (n = 20), and primarily focused on FST values of 1.0 that would indicate complete segregation of a SNP between LF and TRC. At every marker on our LG7 fine maps, we calculated the difference in the values of our three colocalized traits between those hybrids homozygous for the LF allele and those homozygous for the TRC allele.We identified a small region on LG7 that exhibited large differences in the average phenotypic effect of those hybrids with either LF or TRC alleles. In our full LG7 map we identified a ~2 mb region (46.7 mb–48.7 mb) that peaked in all three traits (Fig. 4d; Supplementary Data 11). Notably, the traces of all three traits across our LG7 fine maps track together in an almost identical fashion. In our fine map that centered on the Bayes credible interval, we found evidence for both large phenotypic effects among all traits, and the presence of several FST markers approaching or equal to 1.0 (Fig. 4e; Supplementary Data 12). One marker combined an FST score of 1.0, indicating complete segregation of that allele between LF and TRC, with high average phenotypic effects across all traits (Supplementary Fig. 7). This SNP fell within an intron of dymeclin (dym), a gene that is necessary for correct organization of Golgi apparatus and controls endochondral bone formation during early development. Dym is critical for chondrocyte development and previous research using the zebrafish model found an expression pattern that spanned the presumptive mandibular and ceratobranchial regions at larval stages41. Mutations in this gene lead to profound effects on the size and shape of bones due to misregulated chondrocyte development42. Just downstream (8 kb, Supplementary Fig. 7) of dym is mothers against decapentaplegic homolog 7 (smad7), an antagonist of both TGF-β and BMP signaling and a suppressor of bone formation. As an inhibitory Smad, smad7 negatively regulates genes from the BMP and TGF-β signaling pathways (i.e., bmp-2, -4, -7, nodal, etc.) that are known to shape phenotypic differences in the craniofacial skeleton across a wide range of taxa including cichlids25,38,43, Geospiza finches44,45, and Anolis lizards46, primarily because these genes have the capacity to influence size in structures of trophic importance such as the mandible47. Both of these genes represent good candidates for controlling shape variation in the LOJ and the LPJ simultaneously. While two of the three traits peak at the dym SNP, when considering markers just outside the credible interval another peak is visible (especially for LOJ PC1) that sits close to notch1a, a gene involved in skeletal development and homeostasis. Notch1a is flanked by two fully segregated FST markers. The upstream marker is around ~60 kb from the promoter region, while the downstream marker resides around ~52 kb away from the gene within an intron of kcnt1, a gene involved in potassium channel development that appears to regulate brain function48. While kcnt1 reflects a poor candidate gene for our analysis, the intronic SNP could act as a distant enhancer of notch1a. Thus, given the combined results from QTL and fine-mapping, dym and smad7 represent strong candidates, but we cannot rule out notch1a.Correlated expression of key genes between LOJ and LPJWe used quantitative real-time PCR (qPCR) to assess and compare the expression levels of dym, smad7, and notch1a in the LOJ and LPJ of three mbuna genera from lake Malawi (Tropheops n = 6, Labeotropheus n = 8, Maylandia n = 8). We used Labeotropheus and Tropheops to complement our quantitative genetic analysis, and all three taxa were represented in our phenotypic assessments of integration, permitting a comparison between macroevolutionary associations of the LOJ and LPJ with the underlying genetic architecture and expression for jaw complex correlation. We collected tissue samples from young juveniles of these four taxa, taking the LOJ and LPJ, alongside the caudal fin to act as an internal control, and performed a phenol/chloroform RNA extraction. We designed primers with high amplification efficiency ( >90%) for our three genes (Supplementary Data 13), and used β-actin as our control gene. We calculated relative expression of the LOJ and LPJ using the 2-ΔΔCT method49, and compared expression across taxa and between tissues (Supplementary Data 14 and 15).We initially compared tissue level expression levels between Labeotropheus and Tropheops and found small differences in dym expression, with LF typically exhibiting slightly higher levels (t-test LOJ t = 2.863, P = 0.014; LPJ t = 1.212, P = 0.249; Fig. 5a). These results are consistent with previous expression studies that demonstrated how Labeotropheus typically has up-regulated bone and collagen markers and as a consequence has greater bone deposition and a more robust craniofacial skeleton50,51. Expression level differences were also noted for notch1a and smad7 (Fig. 5b-c); both showed reduced expression in LF, which is expected based on each genes role as negative regulators of bone formation52,53. While the differences between species were fairly small in smad7 between taxa (t-test LOJ t = −1.869, P = 0.086; LPJ t = −0.359, P = 0.726), they were more notable in notch1a (t-test LOJ t = −1.947, P = 0.080; LPJ t = −3.221, P = 0.009). Notch1a is involved in skeletal remodeling, previous research has shown LF exhibits a minimal plastic response to environmental stimuli51. Thus, the relatively low expression of notch1a in Labeotropheus compared to Tropheops is consistent with this observation. While only representing a single life-history stage, the expression differences between species suggest that all three genes may underlie the development of species-specific shapes for the LOJ and/or LPJ. However, visualizing the data this way cannot speak to whether one or more of these loci underlie the covariation of the jaws.Fig. 5: Comparing expression levels of dym, smad7, and notch1a via qPCR in the oral and pharyngeal jaws.a dym bar plot; (b) notch1a bar plot; (c), smad7 bar plot; (d), dym scatter plot; (e), notch1a scatter plot; (f), smad7 scatter plot. a–c bar plots depict mean relative expression levels, error bars denote standard error. d–f Scatterplots depict relative expression levels of the LOJ and LPJ, error bounds surrounding the linear regression line denote standard error. e inset, linear regression for each genus. Three cichlid taxa were included: Labeotropheus n = 8, Tropheops n = 6, Maylandia n = 8. Bar plot significance determined via t-tests: ●P More
