Diet evolution of carnivorous and herbivorous mammals in Laurasiatheria

Background Laurasiatheria contains taxa with diverse diets, while the molecular basis and evolutionary history underlying their dietary diversification are less clear. Results In this study, we used the recently developed molecular phyloecological approach to examine the adaptive evolution of digestive system-related genes across both carnivorous and herbivorous mammals within Laurasiatheria. Our results show an intensified selection of fat and/or protein utilization across all examined carnivorous lineages, which is consistent with their high-protein and high-fat diets. Intriguingly, for herbivorous lineages (ungulates), which have a high-carbohydrate diet, they show a similar selection pattern as that of carnivorous lineages. Our results suggest that for the ungulates, which have a specialized digestive system, the selection intensity of their digestive system-related genes does not necessarily reflect loads of the nutrient components in their diets but appears to be positively related to the loads of the nutrient components that are capable of being directly utilized by the herbivores themselves. Based on these findings, we reconstructed the dietary evolution within Laurasiatheria, and our results reveal the dominant carnivory during the early diversification of Laurasiatheria. In particular, our results suggest that the ancestral bats and the common ancestor of ruminants and cetaceans may be carnivorous as well. We also found evidence of the convergent evolution of one fat utilization-related gene, APOB, across carnivorous taxa. Conclusions Our molecular phyloecological results suggest that digestive system-related genes can be used to determine the molecular basis of diet differentiations and to reconstruct ancestral diets. Supplementary Information The online version contains supplementary material available at 10.1186/s12862-022-02033-6.

ancestral diets is of importance to understanding the evolutionary origin of the dietarily specialized taxa.
The recent development of a molecular phyloecological (MPE) approach provides a new opportunity to investigate ancestral traits using molecular data [4][5][6][7]. The MPE approach mainly uses the adaptive evolutionary analyses of the molecular markers indicative of trait states to determine the molecular basis of phenotypic differentiation and to infer ancestral traits given a phylogeny, and it has been used to infer the diel activity patterns and diets of ancestral taxa [4][5][6][7][8]. Regarding diet reconstruction, the MPE approach employs digestive system-related genes as the molecular markers indicative of diets to infer ancestral diets. Accordingly, carnivores are characterized by the selection intensification of protein and fat utilization, while herbivores are normally characterized by the selection intensification of carbohydrates [7][8][9][10] because carnivore diets are high in proteins and fats and herbivore diets are normally high in carbohydrates [7,[9][10][11][12][13]. The MPE method has been used to infer ancestral diets in birds [7,8], and its fitness to mammals remains to be explored. In this study, we used Laurasiatheria as an ideal clade to examine the molecular basis underlying the diet differentiations and to reconstruct their ancestral diets as it contains the taxa with highly specialized diets. Our results provide new insights into understanding their evolutionary history of diets.

Results
Following the MPE approach to examine the molecular evolution of diets [7,8], we examined the adaptive evolution of 119 digestive system-related genes (Additional file 2: Table S1) in the given Laurasiatheria phylogeny (Fig. 1). These genes are involved in three KEGG pathways, and play important roles in the digestion and absorption of carbohydrates, proteins, and fats [7] (Fig. 2). The positive selection of these genes along particular branches (A-L in Fig. 1) was analyzed using branch and branch-site models implemented in PAML software [14]. Positively selected genes (PSGs) were mainly detected based on the branch-site model (Additional file 3: Table S2). We initially analyzed the positive selection along the lineages with highly specialized diets, including three primarily carnivorous lineages (Eulipotyphla, Pholidota, and Cetacea) and one typically herbivorous lineage (Ruminantia). Intriguingly, our results reveal a highly similar selection pattern across the four lineages and show the predominant selection of fat and protein utilization with relatively the weakest selection of carbohydrate utilization in terms of both the p values and the number of PSGs found. These results remain unchanged even after the Bonferroni multiple testing correction of the p values of PSGs (Additional file 3: Table S2, Additional file 4: Table S3, Additional file 5: Table S4).
We also analyzed the positive selection of the digestive system-related genes in Chiroptera and Carnivora, both of which harbor dietary diverse species. Chiroptera contains both carnivores (e.g., insect-eaters) and herbivores (e.g., fruit-eaters), and our positive selection analyses along the ancestral Chiroptera branch (branch K, Fig. 1) revealed eight PSGs (Additional file 3: Table S2, Additional file 4: Table S3, Additional file 5: Table S4). These eight PSGs include only protein utilization-related genes (SLC3A2, SLC1A5, CELA3B, and DPP4) and fat utilization-related genes (APOB, CD36, ABCG8, and PLPP2). CELA3B is a pancreatic serine proteinase that digests dietary protein substrates [53]. CD36 is mainly involved in the uptake and processing of fatty acids [54]. ABCG8 functions in the excretion of neutral sterols in the liver and intestines [55]. Like Chiroptera, for the ancestral branch of Carnivora (branch H, Fig. 1), only fat utilization-related genes (APOB and PIK3CD) and protein utilization-related genes (CPB2 and KCNK5) were found to be under positive selection (Additional file 3: Table S2). PIK3CD encodes phosphatidylinositol 3-kinase with a broad phosphoinositide lipid substrate specificity [56]. CPB2 encodes carboxypeptidase B2, cleaving C-terminal residues from peptides [57]. KCNK5 is considered to play an important role in potassium transport [58].
To determine the selection characterization of ancestral taxa, we subsequently examined the positive selection of the digestive system-related genes along other early branches of Laurasiatheria (branches A, B, C, D, E and I, Fig. 1) (Additional file 3: Table S2, Additional file 4: Table S3, Additional file 5: Table S4). For the ancestral Laurasiatheria branch (branch A, Fig. 1), we found two fat utilization-related genes (APOB and NPC1L1), one protein utilization-related gene (MEP1B), and one glucose metabolism-related gene (HKDC1) [59] to be under positive selection. For branch B, only one fat utilizationrelated gene, AGPAT2, was found to be under positive selection. It plays a role in converting lysophosphatidic acid into phosphatidic acid [60]. For branch C, three PSGs (SLC7A8, MGAM2, and ATP1B3) were found. For branch I, one positively selected fat utilization-related gene, APOB, was detected. For branch D, two fat utilization-related genes (APOB and PLA2G2D) were found to be under positive selection, and notably, PLA2G2D is a member of lipolytic enzyme [61]. For branch E, two fat utilization-related genes (LIPF and PNLIP) and two protein utilization-related genes (SLC3A2 and MEP1B) were found to be under positive selection, of which PNLIP encodes a pancreatic lipase, also known as pancreatic triacylglycerol lipase. This pancreatic lipase hydrolyzes dietary triglycerides to free fatty acids and monoacylglycerols and is critical for the efficient digestion of dietary triglycerides in the intestines [62,63].
Among the PSGs found, one gene APOB showed a particularly strong positive selection with hundreds of positively selected amino acid sites found in most taxa (See figure on next page.) Fig. 2 Digestive system pathways and positively selected genes found in ruminants (green) and cetaceans (red). Three digestive system pathways (A, B, and C) were modified from that of KEGG, including carbohydrate digestion and absorption (map04973), protein digestion and absorption (map04974), and fat digestion and absorption (map04975). Positively selected genes are shown in parentheses, and their corresponding proteins are highlighted in red (cetaceans) and green (ruminants) examined, including those carnivorous taxa (Additional file 3: Table S2). To test whether the APOB gene was subject to convergent evolution among the carnivorous taxa, we subsequently examined the convergent and/or parallel amino acid substitutions along the branches related to those carnivores by reconstructing ancestral sequences using PAML [14], and many parallel amino acid substitutions were detected among them with high statistical significance (Additional file 1: Fig. S1; Additional file 6: Table S5). For instance, nine parallel substitutions each were found between the branches of Eulipotyphla and Pholidota, and between the branches of Chiroptera and Pholidota. Eight parallel substitutions were found between the branch of Carnivora and two other branches of Eulipotyphla and Chiroptera. These parallel amino acid substitutions may have led to their sequence convergence and thus to their phylogenetic affinity. To test this, we then constructed the maximum likelihood phylogeny based on the protein sequence of the gene APOB. Intriguingly, our results showed that the APOB tree ( Fig. 3) was largely different from their species tree (Fig. 1). In particular, we found that four carnivory-dominant taxa (Carnivora, Pholidota, Chiroptera, and Eulipotyphla) were grouped into the same clade with bootstrap values ranging from 39 to 43 upon three independent runs, indicating their sequence convergence.

Discussion
We in this study examined the adaptive evolution of digestive system-related genes to determine the diet evolution within Laurasiatheria. Consistent with previous studies that demonstrate the evolutionary enhancement of protein and fat utilization in carnivores [7,9,10], all three primarily carnivorous mammal taxa (Eulipotyphla, Pholidota, and Cetacea) examined in this study showed a particularly intensified selection of fat and protein utilization with relatively the weakest positive selection of carbohydrate utilization (Additional file 3: Table S2, Additional file 4: Table S3, Additional file 5: Table S4). This is consistent with their high-protein and high-fat diets. Unexpectedly, for the typical herbivores, the ruminants, which have a high-carbohydrate diet, we detected an intensified selection of fat and protein utilization rather than carbohydrate utilization (Additional file 3: Table S2), resembling that of carnivores. This may suggest that convergent evolution may occur between the carnivores and the herbivores studied. The convergent evolution of diet-related genes is often considered to be resulted from the utilization of similar food [12,[64][65][66], while its occurrence in the ruminants, as evidenced previously [67][68][69], may largely attribute to their specialized digestive system rather than similar food. As we know, ruminants primarily consume plant materials rich in carbohydrates, but they have no enzymes to digest the refractory materials (e.g., cellulose) in their diets. These refractory materials are transferred through microbial fermentation in their guts to volatile fatty acids and microbe proteins, constituting the major sources of energy and amino acids for subsequent utilization by ruminants [70][71][72][73]. This suggests that though their diets are rich in carbohydrates, the main nutritional substrates that ruminants are capable of directly utilizing are actually fats and proteins that are generated through microbial fermentation. Therefore, the intensified selection of fat and protein utilization found in the ruminants may be mainly due to their specialized digestive system.
Our results show that the carnivorous mammals studied are consistently characterized by an intensified selection of fat and protein utilization, which is in line with their high-protein and high-fat diets. For herbivorous mammals, because fermentation through their gut microbes can transfer dietary carbohydrates to other nutritional substrates, such as volatile fatty acids and microbial proteins, for their subsequent use [70][71][72][73], the selection characterization of their digestive systemrelated genes do not necessarily reflect the amounts of nutritional substrates in their diets. Previous studies suggest that microbial fermentation widely occurs in animals, while the contribution of microbial fermentation to energy production is largely different among animals. Its importance seems to be limited to particular taxa (e.g., ungulates), possibly due to high amounts of refractory materials (e.g., cellulose) in their diets, but is relatively trivial to other animals [74]. Consequently, the adaptive evolution of digestive system-related genes of animals is considered to be generally positively related to loads of their dietary substrates [10,12,13,[74][75][76]. Accordingly, carnivores are characterized by the selection intensification of protein and fat utilization, while herbivores are normally characterized by the selection intensification of carbohydrates [7,9,10]. Thus, we could reconstruct the diets of ancestral taxa based on the selection characterization of digestive system-related genes [7,8].
To determine ancestral diets, we analyzed the positive selection of the digestive system-related genes along the ancestral branches of the living animals studied (Fig. 1). For the ancestral branches of bats (branch K) and of carnivorans (branch H), we detected their evolutionary enhancements mainly in fat and protein utilization, as found in the ancestral branches leading to other primarily carnivorous mammals (Eulipotyphla, Pholidota, and Cetacea) (Additional file 3: Table S2, Additional file 4: Table S3, Additional file 5: Table S4). This may suggest that all the ancestral taxa including ancestral bats and ancestral carnivorans were largely carnivorous (Fig. 1). Similarly, an evolutionary enhancement of fat and/or protein utilization was also found in other ancestral branches (branches A, B, and I) (Additional file 3: Table S2). This may suggest that the early evolutionary diversification of Laurasiatheria was mainly characterized by carnivory (Fig. 1), which is largely consistent with one previous study [2]. Nonetheless, for the ungulates examined in this study, we unexpectedly found an intensified selection of fat utilization along branch D, leading to ancestral ungulates, and an evolutionary enhancement of fat and protein utilization along branch G, leading to  Table S2). This may reflect their high fat and/or protein nutrition generated by nutritional transformation through microbial fermentation, and hence it may suggest that the ancestral ungulate and the ancestral ruminant may be herbivorous (Fig. 1). For branch C, we detected the positive selection of three PSGs involved in carbohydrate and protein utilization (Additional file 3: Table S2, Additional file 4: Table S3, Additional file 5: Table S4), suggesting a high-carbohydrate and high-protein diet. A high-carbohydrate and high-protein diet may suggest a combination of herbivory and carnivory, hence implying that the common ancestor of carnivorans and ungulates was possibly omnivorous (Fig. 1), which may be derived from its carnivorous progenitor (branch B). If this is the case, it may suggest that the herbivory of ungulates and the carnivory of carnivorans were secondarily evolved. This is consistent with fossil evidence showing that the earliest stem carnivorans, such as Ravenictis and Pristinictis, exhibit relatively unspecialized molars, indicating an omnivorous diet with only limited specialization to true carnivory [77]. Fossil evidence indicates the resemblance of some primitive ungulates to carnivores. For instance, Phenacodus, which is considered a stem Perissodactyla [78], lived during late Palaeocene and early Eocene, and was a planteater yet shows some characteristics (e.g., large canine teeth) resembling a primitive carnivore [79]. These lines of evidence may suggest the herbivory of the ungulates, and the pure carnivory found in modern carnivorans may be secondarily evolved from an omnivorous ancestor.
For the branch leading to the common ancestor of ruminants and cetaceans (CARC), we detected the enhanced selection of fat and protein utilization (Additional file 3: Table S2), which is similar to that found in both of its two derived taxa, the ruminants and the cetaceans. This seems to make the reconstruction of the diet of the CARC unresolved; however, our finding of the positive selection of the two fat digestion-related genes (PNLIP and LIPF) along the branch leading to the CARC (Additional file 3: Table S2) may suggest that the CARC was more likely carnivorous (Fig. 1). This is because: i) PNLIP and LIPF are both critical lipases mainly involved in digesting dietary triglycerides in the digestive system [17,62,63,80], and the selection enhancement of the digestion of dietary triglycerides may suggest a lipid-rich diet of the CARC. A lipid-rich diet most often characterizes carnivores rather than herbivores because carnivore diets are relatively rich in lipids, while herbivore diets are normally rich in carbohydrates [9][10][11][12][13]. ii) The evolutionary enhancement of digesting dietary triglycerides found in the CARC may suggest that the CARC itself may have the capability to digest dietary fats (e.g., triglycerides). This is consistent with carnivorous mammals (e.g., cetaceans), which normally use their own lipases to digest fats [81], but is substantially different from ruminants, from which their dietary lipids (e.g., triglycerides) are predominantly hydrolyzed by the lipases of rumen bacteria in their guts [82][83][84][85]. iii) The detected positive selection of PNLIP and LIPF in the CARC has been found in carnivores (e.g., cetaceans) [68,76] but not in ruminants, which is evidenced in this study and one previous study [68], suggesting the resemblance of the digestion ability of the CARC with that of carnivores (cetaceans). In addition to the molecular evidence, fossil evidence shows that early ruminant ancestors were omnivores and did not ruminate until about 40 Ma based on dental morphology [86]. This indicates that the herbivory and rumination observed in modern ruminants may be secondarily evolved, which is consistent with the possible carnivory of the CARC. These four lines of evidence may suggest that the CARC was more likely a carnivore, though its existence in fossils remains to be explored. Considering that the CARC is phylogenetically deeply nested within the ungulates, it is thus more likely a carnivorous ungulate closely related to cetaceans and/or ruminants. Among carnivorous mammals known, previous studies have long considered one extinct carnivorous ungulate, mesonychians, as early members of cetaceans or Cetartiodactyla [87][88][89][90][91], though there is uncertainty regarding the phylogenetic position of mesonychians [77,92]. Mesonychians are considered secondary carnivores derived from archaic ungulates (Condylarthra) [88]. If this is the case, mesonychians might be the candidate of the CARC from which cetaceans and ruminants derived, though this requires further investigation.

Conclusions
Our molecular phyloecological results show that the carnivorous mammals consistently exhibit the evolutionary enhancement of fat and protein utilization, which is in line with their high-protein and high-fat diets. This is previously found in birds and crabs as well. For herbivores, previous studies on birds and crabs suggest that they tend to show an evolutionary enhancement of carbohydrates; however, the ungulates with a high-carbohydrate diet examined in this study present an evolutionary enhancement of fat and protein utilization, resembling that of carnivores. Apparently, this is largely due to their specialized digestive system that transfers abundant carbohydrates to volatile fatty acids and microbial proteins for their use. Our results suggest that the adaptive evolution of digestive system-related genes do not necessarily reflect the nutritional loads in the diets of the herbivorous animals (e.g., ungulates) that mainly rely on nutritional transformation before utilization but appear to be positively related to the loads of the nutrient substrates that can be directly utilized by the herbivores themselves. Based on these findings, we reconstructed ancestral diets, and our results revealed predominant carnivory during the early diversification of Laurasiatheria. More importantly, our reconstructed results suggest that the ungulates and carnivorans may have been derived from an omnivorous ancestor, and ancestral bats and the common ancestor of ruminants and cetaceans may be carnivorous. We also found evidence of the convergent evolution of one fat utilization-related gene, APOB, across carnivorous lineages, suggesting the resemblance of nutritional utilization in carnivorous mammals. Further studies incorporating information about the gene duplications and losses besides positive selection may be helpful to understand the molecular bases underlying the diet evolution of the herbivorous and carnivorous mammals.

Taxa used
Ninety species within Laurasiatheria were included (Fig. 1). These 90 species covered all known main clades of Laurasiatheria, including four species of Eulipotyphla, two pangolin species of Pholidota, three species of oddtoed ungulates (Perissodactyla), and 14 species of eventoed ungulates (Artiodactyla), of which nine species belong to Ruminantia. We also included 25 bat species from both two suborders (Yangochiroptera and Yinpterochiroptera) of Chiroptera. For Carnivora, 28 species from its two suborders (Feliformia and Caniformia) were included. For Cetacea, 14 species from its two suborders (Mysticeti and Odontoceti) were included. In addition to these 90 Laurasiatheria species, we included two species from the sister taxa (Euarchontoglires) of Laurasiatheria as outgroups. For the two outgroup species, Homo sapiens and Mus musculus were primarily used, while the two relatives (Rattus rattus and Rattus norvegicus) of Mus musculus were also considered if some gene sequences of Mus musculus were unavailable.

Diet data
The dietary categories of each species used in this study were based on one previously published dataset, Elton-Traits 1.0 [93], in which the dietary information of a total of 5400 extant mammal species from diverse published literature is compiled and the dietary composition of each species is recorded in 10% dietary categories. To determine the dietary categories of the species used in our study, we converted EltonTraits' 10% dietary categories into our three dietary categories (carnivore, herbivore, and omnivore). Carnivore = Diet-Inv + Diet-Vend + Diet-Vect + Diet-Vfish + Diet-Vunk + Diet-Scav, herbivore = Diet-Fruit + Diet-Nect + Diet-Seed + Diet-PlantO, and omnivores were referred to as the animals that contain a percentage of dietary categories of both the carnivore and the herbivore.

Genes and sequence alignment
We included the digestive system-related genes that have been recently used to determine the diet evolution in birds [7]. These genes were from three KEGG pathways, including carbohydrate digestion and absorption (map04973), protein digestion and absorption (map04974), and fat digestion and absorption (map04975). For these genes, we downloaded their coding sequences of our focal species from GenBank (Additional file 2: Table S1). We excluded genes with sequences unavailable or available for only a few species from our analyses, and ultimately, 119 genes were retained for subsequent analyses. We aligned gene sequences using webPRANK with default parameters (http:// www. ebi. ac. uk/ goldm an-srv/ webpr ank/), and individual species sequences with lengths that were too short were removed. The sequence alignments were checked by eye and the sequence gaps that lead to incorrect protein translation were cut. After this pruning, we blasted the translated protein sequences of these genes against the non-redundant protein sequence database to confirm the correctness of the sequence cutting.

Positive selection analyses
For the positive selection analyses, we initially constructed a Laurasiatheria phylogeny of the 90 species used in this study, as shown in Fig. 1. Our Laurasiatheria phylogeny was based on published studies [94][95][96]. In particular, the phylogenetic relationships used among taxonomical orders within Laurasiatheria, which have received increasing support for the past 20 years [95][96][97][98], are the same as those used in one previous diet study of mammals [2]. Based on the Laurasiatheria phylogeny, we analyzed the positive selection of our target genes using branch and branch-site models implemented in the Codeml program of PAML [14]. The ratio of nonsynonymous to synonymous substitutions per site (dN/ dS or ω) was evaluated, and likelihood ratio tests (LRT) were employed to determine the statistical significance. Positive selection is determined by the value of ω > 1 with statistical significance. The Bonferroni multiple testing correction was used to adjust the p values.
Branch model Branch model allows for the variation of the ω ratio among branches in a given phylogeny, and it is used to detect the positive selection of genes on a particular branch. For the branch model analyses, we used a two-rate branch model by labeling our focal branches as foreground branches and the others as background branches. During the analyses, the goodness of fit of the two-rate branch model relative to the null model-that is one-rate branch model-was analyzed using the LRT. When a statistically significant value of ω > 1 was found in our foreground branches, to determine whether the value of ω > 1 of the foreground branch was further supported with statistical significance, we then compared the tworatio branch model with the two-ratio branch model with ω = 1 fixed in our foreground branches.
Branch-site model The branch-site model allows for the variation of ω among sites in the protein and across phylogenetic branches, and it is used to detect positive selection affecting some sites along a particular branch. For the branch-site model analyses, we employed a branchsite test of positive selection (Test 2), which compares a modified model A with its corresponding null model with ω = 1 fixed. The modified model A assumes four classes of sites, and site class 0 and site class 1, respectively, represent evolutionarily conserved (0 < ω 0 < 1) and neutral codons (ω 1 = 1) for both background and foreground branches. Site classes 2a and 2b, respectively, represent evolutionarily conserved (0 < ω 0 < 1) and neutral (ω 1 = 1) codons for background branches, but allowed to be under positive selection (ω 2 > 1) for the foreground branches. The goodness of fit of the modified model A was evaluated using the LRT by comparing it with a null model with ω = 1 fixed. Positively selected sites were identified by employing an empirical Bayes method.

Ancestral sequence reconstruction and convergent evolution analyses
Amino acid-based marginal reconstruction implemented in the empirical Bayes approach in PAML [14] was used to reconstruct the ancestral sequence. For the marginal reconstruction, we employed two different substitution models (JTT and Poisson) of amino acids to examine the consistency of our results. For the model JTT, different substitution rates of different amino acids were assumed, and for the Poisson model, the same substitution rate of all amino acids was assumed. The analyses based on the two substitution models generated similar results, and for convenience, we only showed the results based on the JTT model. Based on the reconstructed ancestral sequences of internal nodes, convergent and/or parallel amino acid substitutions along branches could then be identified. To further estimate the probabilities that the observed convergent and/or parallel substitutions are attributable to random chance, the program converg2 implemented in the software Convergent and Parallel Evolution at the Amino Acid Sequence Level (CAPE) [99] was used.

Phylogenetic analyses
Phylogenetic analyses were conducted using the IQ-TREE, a fast and effective stochastic algorithm for inferring maximum-likelihood (ML) phylogeny [100,101]. The IQ-TREE is considered to have high performance for ML inference compared to other popular software, such as RAxML [102] and PhyML [103]. This is considered to result from its efficient integration of fast model selection, an effective tree search algorithm, and a novel ultrafast bootstrap approximation [100]. Especially, the effective tree search algorithm was believed to overcome the problem of local optima and thus to help to achieve ML phylogeny with higher likelihoods than RAxML or PhyML. For our phylogenetic analyses, 4550 amino-acid sites of the gene APOB were used. Among 546 protein models examined by ModelFinder implemented in the IQ-TREE, two models, HIVb+F+R4 and HIVb+F+R5, were recommended as the best-fit substitution models according to Bayesian and Akaike information criteria, respectively. For result robustness, the two substitution models were both used and almost identical results were obtained. For convenience, only the ML phylogeny based on the substitution model of the HIVb+F+R4 was presented with the bootstrap value of 10,000 used.