Phylogenomic Data Analyses Provide Evidence that Xenarthra and Afrotheria Are Sister Groups (original) (raw)

Journal Article

,

Department of Cell and Organism Biology, Division of Evolutionary Molecular Systematics, University of Lund, Lund, Sweden

Search for other works by this author on:

,

Department of Cell and Organism Biology, Division of Evolutionary Molecular Systematics, University of Lund, Lund, Sweden

Search for other works by this author on:

,

Department of Cell and Organism Biology, Division of Evolutionary Molecular Systematics, University of Lund, Lund, Sweden

Search for other works by this author on:

Department of Cell and Organism Biology, Division of Evolutionary Molecular Systematics, University of Lund, Lund, Sweden

Search for other works by this author on:

• PDF
• Split View
• • Article contents
  • Figures & tables
  • Video
  • Audio
  • Supplementary Data
• Cite

Cite

Björn M. Hallström, Morgan Kullberg, Maria A. Nilsson, Axel Janke, Phylogenomic Data Analyses Provide Evidence that Xenarthra and Afrotheria Are Sister Groups, Molecular Biology and Evolution, Volume 24, Issue 9, 1 September 2007, Pages 2059–2068, https://doi.org/10.1093/molbev/msm136
Close

Navbar Search Filter Mobile Enter search term Search

Abstract

The phylogenetic positions of the 4 clades, Euarchontoglires, Laurasiatheria, Afrotheria, and Xenarthra, have been major issues in the recent discussion of basal relationships among placental mammals. However, despite considerable efforts these relationships, crucial to the understanding of eutherian evolution and biogeography, have remained essentially unresolved. Euarchontoglires and Laurasiatheria are generally joined into a common clade (Boreoeutheria), whereas the position of Afrotheria and Xenarthra relative to Boreoeutheria has been equivocal in spite of the use of comprehensive amounts of nuclear encoded sequences or the application of genome-level characters such as retroposons. The probable reason for this uncertainty is that the divergences took place long time ago and within a narrow temporal window, leaving only short common branches. With the aim of further examining basal eutherian relationships, we have collected conserved protein-coding sequences from 11 placental mammals, a marsupial and a bird, whose nuclear genomes have been largely sequenced. The length of the alignment of homologous sequences representing each individual species is 2,168,859 nt. This number of sites, representing 2840 protein-coding genes, exceeds by a considerable margin that of any previous study. The phylogenetic analysis joined Xenarthra and Afrotheria on a common branch, Atlantogenata. This topology was found to fit the data significantly better than the alternative trees.

Introduction

Ongoing genome sequencing and analyses of large expressed sequence tag (EST) libraries have aided in recent progress toward molecular resolution of the most basal parts of the tree of placental mammals, notably the relationship among the 4 clades, Afrotheria, Xenarthra, Laurasiatheria, and Euarchontoglires (as defined by Murphy, Elzirik, Johnson et al. 2001). Laurasiatheria and Euarchontoglires are generally identified as sister groups, forming the so-called Boreoeutheria (Murphy, Elzirik, O'Brien et al. 2001; Springer and de Jong 2001; Waddell et al. 2001). However, the relationship between Boreoeutheria, Afrotheria, and Xenarthra has proven to be difficult to establish. Currently 3 different hypotheses exist, each favoring one of the 3 possible alternative topologies in the rooted eutherian tree.

The traditional morphological hypothesis favors a basal divergence between Xenarthra and remaining eutherians, Epitheria (McKenna 1975; Novacek 1992; Shoshani and McKenna 1998). This hypothesis has also been supported by a recent study employing retroposons as phylogenetic markers (Kriegs et al. 2006), although only 2 relevant markers were found in support of this particular basal split. The authors concluded, however, that while the 2 retroposons provide strong support for the Epitheria hypothesis, their presence might not be sufficient to statistically reject other possible topologies. The commonly favored molecular tree has Afrotheria as a sister group to all remaining placental mammals and places Xenarthra as a sister group to Boreoeutheria (Murphy, Elzirik, O'Brien et al. 2001; Waddell et al. 2001; Amrine-Madsen et al. 2003). This hypothesis was also supported in a recent molecular study based on both protein-coding and noncoding genomic sequences (Nikolaev et al. 2007). The third hypothesis joins Xenarthra and Afrotheria on a common branch (Atlantogenata or Xenafrotheria) with a basal split between this group and Boreoeutheria. This hypothesis received support in some studies of mitochondrial sequences (Waddell et al. 1999; Kjer and Honeycutt 2007) and in 2 recent nuclear studies, one of which was based on long interspersed nuclear element (LINE-1) sequences (Waters et al. 2007) and the other was based on the examination of insertion–deletion (indel) differences and 2 retroposon inserts (Murphy et al. 2007).

The instability of the basal portion of the eutherian molecular tree indicates that the divergences connected to this part of the tree took place within a limited period of time. It is therefore conceivable that resolution of these relationships will require large amounts of sequence data. Only this will overcome the disturbing influence of noise that has accumulated through the long separate evolution of the individual branches involved and that masks the few shared substitutions. The basal eutherian relationships were recently addressed by Nikolaev et al. (2007) in a study based on 205 kb of coding and 430 kb of conserved noncoding nuclear sequences. Despite this amount of data, the study resulted in conflicting hypotheses as the noncoding sequences found it equally likely that Afrotheria or Atlantogenata was the sister to remaining eutherians.

In order to examine basal placental mammal divergences in more detail with the aim to resolve their basal divergences, we have in the current study collected the yet largest amount of data of protein-coding genes from placental mammals that are represented by extensive amounts of genomic sequences. We maximized both the amount of homologous sequences and the number of taxa. The root of the tree was established by using the corresponding sequences of a marsupial and a bird as outgroup.

Methods

The protein-coding data for 13 species were retrieved from the Ensembl FTP site(Release 41), and the RefSeq database of human RNA sequences were downloaded from the National Center for Biotechnology Information FTP site. The species used in this study include: human (Homo sapiens), rat (Rattus norvegicus), mouse (Mus musculus), rabbit (Oryctolagus cuniculus), dog (Canis familiaris), cow (Bos taurus), chimpanzee (Pan troglodytes), rhesus monkey/macaque (Macaca mulatta), armadillo (Dasypus novemcinctus), elephant (Loxodonta africana), small Madagascar hedgehog/tenrec (Echinops telfairi), opossum (Monodelphis domstica), and chicken (Gallus gallus). The last 2 species were used as outgroups to the placental mammals. Putatively orthologous sequences from all species were identified by using the reciprocal best-hit method (Rivera et al. 1998) with the BlastN search algorithm requiring an expected value (E value) no larger than 1 × 10−12, as implemented in the program EST-e-mate (Hallström B, Janke A, in preparation). The sequences were translated to the corresponding amino acids (aa) and multiple sequence alignments (MSA) were created using ClustalW as implemented in EST-e-mate. Columns with gaps in the alignment were removed. Using the aa alignments, the corresponding nucleotide (nt) alignments and their reading frames were deduced. All alignments with an observed nt distance above 30% between any 2 species were discarded in order to compensate for possible inaccuracies of the MSAs at high sequence divergence. This approach also reduces analytical problems related to multiple substitutions and further ensures to include only orthologous sequences in the analysis. All alignments were inspected manually, and obviously erroneous alignments were either corrected or removed from the analysis. The sequences that are included in the alignment were classified using the Protein Analysis THrough Evolutionary Relationships (PANTHER) classification system (Thomas et al. 2003) by submitting a list of accession numbers for the human orthologs to the “Batch search” provided in the Web interface http://www.pantherdb.org (Mi et al. 2007).

From these alignments, 2 main data sets were created for analyses. One maximizes the sequence data for all species and minimizes missing data. This MSA has sequence data for all 13 species and comprises 1,961 aligned sequences (1,569,141 nt). It is named “maxspe.” The other MSA allows missing data in the alignment and maximizes the amount of sequence data instead, allowing for missing sequences. These could involve the chicken and one placental species from a clade for which genomic data for 2 or more species were available. Thus, sequences of cow or dog, elephant or tenrec, and mouse or rat were allowed to be missing. This data set comprises 2,840 (2,168,859 nt) aligned sequences and is named “maxgen”. Accession numbers for human orthologues of the maxgen data set are given in Supplementary Materials online.

These 2 data sets, maxspe and maxgen, were analyzed using maximum likelihood (ML) as implemented in Treefinder (TF; Jobb et al. 2004), Tree-Puzzle 5.2 (TP; Strimmer and von Haeseler 1996), or PAML 3.15 (Yang 1997) with the general time reversible (GTR) model of substitution (Lanave et al. 1984) and rate heterogeneity for nt data (Yang 1994), as suggested by Modeltest (Posada and Crandall 1998) and the Jones-Taylor-Thomton (JTT) model with rate heterogeneity for aa data, as suggested by ProtTest (Abascal et al. 2005). Four discrete categories of gamma distribution and one invariable, 4Γ + I, were used in the analyses using TP and TF. Eight discrete categories of gamma distribution, 8Γ, were used when the analysis was done in PAML because the program package lacks the option for invariable sites. Bootstrap support values were calculated for 100 replicates using TF. Jackknife analysis was done by creating 100 replicates with 99% of the maxspe data set randomly removed. These replicates had lengths of approximately 15 kb and were analyzed with PHYML (Guindon and Gascuel 2003) using the same models as for TP and TF. Different tree topologies were statistically evaluated using the Shimodaira–Hasegawa test (pSH; Shimodaira and Hasegawa 1999) and 1-sided Kishino–Hasegawa test (pKH; Goldman et al. 2000), as well as a confidence value for Expected Likelihood Weight (cELW; Strimmer and Rambaut 2002) as implemented in TP.

A partitioned ML analysis was done in TF. The alignments were partitioned into 3 partitions, based on codon position, for which the GTR and rate heterogeneity parameters were estimated separately. Finally, the sequences were also analyzed using Bayesian inference with MrBayes 3.1.2 (Huelsenbeck and Ronquist 2001), running for 2,000,000 Markov chain Monte Carlo (MCMC) generations with 1 cold and 3 heated chains, discarding the first 200,000 generations as burn-in and then sampling each 100th tree.

In order to determine whether the data conformed to compositional homogeneity, a chi-square test as implemented in TP was performed on the base frequencies for each species. To further investigate if the compositional heterogeneity between different lineages had any effect on the tree reconstruction, we recoded the nt sequences to R (purine) for A and G sites and to Y (pyrimidine) for T and C sites. Although a lot of information is lost when R/Y coding is used, the differences in base composition between lineages are greatly decreased. Thus, this procedure is merely used to detect possible reconstruction artifacts that may be caused by compositional biases rather than to replace the standard analysis. Additionally, the data were analyzed using the PAML program baseml with a nonstationary model (nhomo = 3), which allows for different base composition on different branches (Yang and Roberts 1995).

The 2 data sets from Nikolaev et al. (2007), available as supporting information with the paper, were downloaded and analyzed using the same ML methods as for our own data. This was done to produce information about these data sets that were not explicitly reported in the paper, notably pairwise distances between species. A sliding-window approach was applied to study the distribution of sequence distances throughout the full maxgen alignment and the aa sequence alignment of Nikolaev et al. (2007). A window of 200 aa was used, in which the observed distance were calculated, where after the window was moved 100 positions forward and the same procedure was repeated until the end of the alignment was reached.

Relative branch lengths were calculated against the branch leading to the elephant. This branch was arbitrarily chosen because from the aa ML analysis it appears to be moderately fast evolving.

Divergence time estimates were done with r8s (Sanderson 2002) and the multidivtime (MDT; Thorne and Kishino 2002) program package (http://abacus.gene.ucl.ac.uk/). For optimizations of the r8s divergence time estimations, the nonparametric rate smoothing method and the Powell algorithm were used on the branch lengths of the ML tree from TP (Sanderson 2002). Also, 10 different starting conditions were used to avoid local optima. For calculating standard deviations (SDs), 100 bootstrap replicates of the sequence data were created using SEQBOOT. The branch lengths for each replicate were estimated using TP under the appropriate rate heterogeneity model and the tree given in figure 3. For each of those tree replicates, the divergence times were estimated by the r8s program and SD were calculated from the different results.

A parametric Bayesian analysis for divergence time estimates as implemented in MDT was made on nt and aa sequence data of the maxgen data, using 8 gamma categories of variable sites. A total of 1,000,000 generations of MCMC estimates for dates and rates parameters were sampled after an equally long burn-in. Besides from the calibration points below the prior expected age of the root was set to 135 and its standard deviation to 15. All analyses were performed several times, verifying convergence between the chains. For estimates of the placental mammal divergence times, several constraints have been applied for calibrating the molecular clock. These are marked in figure 6 and are taken from Benton and Donoghue (2007).

Results

Prior to the phylogenetic analyses, the general properties of the data set, such as distances, alignment lengths, and nt and aa compositions, were examined. The distribution of nt distances for the individual human–mouse alignments is shown in figure 1. These distances, which were generally low, were evenly spread around the mean. The average human–mouse nt distance was 10.8% ± 3.3%. Exclusion of sequences with high E values reduces distances and the influence of multiple substitutions, which can become problematic at larger distances. The average length of the individual alignments was 763 ± 612 nt. The distribution of the alignment lengths is shown in figure 2. The great majority of alignments are between 300 and 4000 nt in length. Classification of genes according to the PANTHER system and the number of genes in each class are shown in table 1.

Table 1

Classification of Biological Function of the Human Orthologs, according to PANTHER Database

Table 1

Classification of Biological Function of the Human Orthologs, according to PANTHER Database

FIG. 1.—

Distribution of the observed nt human–mouse pairwise distances for single-nt alignments of the maxgen MSA.

FIG. 2.—

Distribution of the lengths of single alignment of nt data in the maxgen MSA.

The sequences were examined with respect to possible base frequency biases between lineages. The observed base frequencies are shown in table 2. None of the species passed at the 5% significance level of a chi-square test on the homogeneity of base composition, thereby suggesting that the differences in base composition were systematic. When all 3 codon positions were recoded to R and Y, the compositional bias was significantly decreased (table 2). After R/Y recoding, only 4 species (cow, rabbit, tenrec, and chicken) did not pass the chi-square test at the 5% significance level. Comparing results of the phylogenetic analyses of the R/Y recoded with the native sequence data allows examination of how the compositional bias may affect tree reconstruction. The composition of aa sequences was less heterogeneous. Only the outgroup taxa and the tenrec showed significantly deviating aa compositions.

Table 2

The Frequencies of nt Usage for the Individual Species; the Frequencies of R and Y for the R/Y-Recoded Data; and the Probability Values of the chi-square test on Compositional Heterogeneity for Nucleotides, R/Y-Coded nt, and aa

a

Species for which compositional homogeneity could not be rejected at a 5% level of significance, according to the chi-square test, and are therefore assumed to be homogeneous.

Table 2

The Frequencies of nt Usage for the Individual Species; the Frequencies of R and Y for the R/Y-Recoded Data; and the Probability Values of the chi-square test on Compositional Heterogeneity for Nucleotides, R/Y-Coded nt, and aa

a

Species for which compositional homogeneity could not be rejected at a 5% level of significance, according to the chi-square test, and are therefore assumed to be homogeneous.

ML aa trees were constructed under a JTT 4Γ + I model of sequence evolution. The GTR 4Γ + I model was applied to nt sequences, whereas R/Y coded sequences were analyzed using a 2-state (Felsenstein 1981) 4Γ + I model. All ML analyses resulted in the same tree topology irrespective of the use of different data sets or program packages. ML bootstrap values for internal branches were in all instances 100%. The aa ML tree is shown in figure 3. The partitioned likelihood analysis yielded the same tree as shown in figure 3 and the same relative branch lengths.

FIG. 3.—

The phylogenetic tree, with scaled branch lengths, reconstructed for 722,953 aa sites (maxgen MSA) under the JTT 4Γ + I model of sequence evolution.

The 3 possible trees that can be constructed for Xenarthra, Afrotheria, and Boreoeutheria plus outgroup are shown in figure 4. The 3 trees represent the Atlantogenata hypotheses (Xenarthra is the sister group to Afrotheria), the Epitheria hypothesis (Xenarthra is the sister group to all remaining placental mammals), and finally the Afrotheria hypothesis (Afrotheria is the sister group to all remaining placental mammals). The 3 alternatives were compared in an extended ML analysis. Table 3 shows likelihood values, pSH, pKH, and cELW values, for the 3 different topologies, as calculated by TF using nt sequences. The aa sequences were also submitted to the corresponding analyses (table 4). All ML tests of both nt and aa sequences with both TF and TP conclusively identified the Atlantogenata hypothesis as being the most likely, whereas the 2 alternative topologies, Epitheria and Afrotheria, were significantly rejected by all ML tests. Bayesian analysis of both nt and aa sequences using MrBayes identified the Atlantogenata hypothesis with probability (P) 1.0 on all branches.

Table 3

ML Support and Probability Values for the 3 Hypotheses, nt Data Sets

Table 3

ML Support and Probability Values for the 3 Hypotheses, nt Data Sets

Table 4

ML Support and Probability Values for the 3 Hypotheses, aa Data Sets

Table 4

ML Support and Probability Values for the 3 Hypotheses, aa Data Sets

FIG. 4.—

The trees representing the 3 hypotheses on the evolution of the placentals.

ML analysis of RY-coded nt sequences gave rise to the same tree topology as obtained in the analysis of unmodified nt sequences. In this case too, the alternative trees were significantly rejected in all analyses. Also the nonstationary model, as implemented in PAML (nhomo = 3), found the Atlantogenata hypothesis (tree-1) as being the most likely. The nonstationary model uses different frequency parameters for different branches and thus allows analysis of standard (ATGC) nt sequences, even in the case of compositional heterogeneity. The congruent results in both these analyses indicate that the compositional heterogeneity of the nt data has only limited effect on the tree reconstruction. Despite significant deviations in nt composition in many lineages of placental mammals, the ML topology remained the same and differences in relative branch lengths were also limited. This consistency is important in regard to the molecular estimates of divergence times.

Parallel to the ML analyses a jackknife analysis was performed with 100 replicates of 1% of the maxspe data set (∼15 kb). Of the 100 samples, 84 showed the Atlantogenata topology, whereas 11 supported the Afrotheria tree and 5 the Epitheria hypothesis. Other branches remained unchanged with a support of 99–100%. Extended jackknife analysis showed that ≈30% of the nt-coding and ≈55% of the aa-coding maxgen alignments were needed for providing statistically significant support for the best hypothesis (Atlantogenata).

Analysis of nt composition showed that the tenrec deviated distinctly from the nt composition of other placental mammals studied. This bias, which also affected the aa sequence composition, remained after recoding the nt sequences to R and Y. In order to study the effect that this bias might have on the tree reconstruction, the ML analyses were repeated after excluding the tenrec from the data set. The removal of the tenrec had only a minimal influence on the result by slightly increasing the statistical support for the Atlantogenata hypothesis.

The effect of taxon sampling on tree construction was examined in additional ML analyses that included the same species as used in the genomic study of Cannarozzi et al. (2007). Thus, this part of the current study was limited to the sequences of cow, dog, mouse, rat, human, with chicken and/or a marsupial used as outgroup. All analyses performed with rate homogeneity models on the 2.3 mega-base maxgen data set identified rodents as sister to remaining placental mammals, that is, the same tree as in Cannarozzi et al. (2007). However, when a rate heterogeneity (4Γ + I) model was used, a grouping of human and rodents on the same branch became the most strongly supported alternative, although other topologies could not be rejected. The effect of adding more taxa was studied by the stepwise addition of species belonging to the orders Lagomorpha (rabbit), Proboscidea (elephant), Xenarthra (armadillo), and Tenrecidae (tenrec). When the sequences of the rabbit, the elephant, and the armadillo or the tenrec were included, the topology became consistent with the Boreoeutheria hypothesis, that is (primates, [rodents, rabbit], [cow, dog]), with the elephant and the armadillo or the tenrec as sister group to the former. This general topology did not change with the addition of the chimpanzee and macaque, which increased the taxon sampling on the primate branch.

A change of outgroup did not affect the topology of the tree of placental mammals. However, the use of the opossum as a single outgroup yielded stronger support to the Atlantogenata hypothesis than did the analyses with the chicken as the only outgroup.

Reanalysis of the data used by Nikolaev et al. (2007) reproduced topologies and statistical values that were consistent with their original results. However, examination of pairwise distances among the nt sequences of the protein-coding genes of that study showed that these were approximately twice as large as those of the maxgen and maxspe alignments, whereas the noncoding data of Nikolaev et al. (2007) had just slightly lower distance values. Translation of protein-coding sequences to aa sequences resulted in suspiciously high distance values, especially relative to the outgroups. The distances along the aa alignment were analyzed in more detail by a pairwise comparison using a sliding window. Figure 5 shows the distribution of uncorrected human–opossum distances in the maxgen alignment (black line), whereas the gray line shows the corresponding values for the data set used by Nikolaev et al. (2007). As evident in figure 5, the distances within each data set differ to some extent. Thus, the distances within the maxgen alignment have a maximum of about 25% (very few have distances exceeding 15%) as a result of the restrictions for sequence retrieval and alignments. The aa alignment used by Nikolaev et al. (2007) shows a high proportion (26%) of distance values that exceed 30% and a nonnegligible amount (5%) of extremely high distances, 50–71%. Our ML analyses also confirmed that their noncoding data set could not distinguish between the Afrotheria and Atlantogenata hypotheses.

FIG. 5.—

Distribution of aa distances between human and opossum calculated with a sliding window over the full alignments. The black line represents the maxgen MSA and the gray line represents the aa alignment of Nikolaev et al. (2007).

Estimation of divergence times by MDT proved difficult because the program apparently does not allow for more than 1 × 106 characters. This limitation was overcome by splitting the data into alignments with a length of 800,000 characters. Divergence time estimates from r8s and MDT for nt and aa are shown in figure 6 together with the constraints included. Despite the use of different data sets and approaches, the estimates do not differ markedly from each other, except for the basal divergences of Glires and Euarchontoglires. All estimates from MDT fall somewhere within the corresponding constraints. The error values for each estimate are small and are not reported here in order to avoid overconfidence in the estimates. It is probable that the statistics behind the estimates may not be optimized to sets of data corresponding to those of the current study, but this should not invalidate the fact that estimates based on large sets of data should in general be more precise than those based on smaller ones.

FIG. 6.—

Tree with divergence time estimates. The 4 dates correspond to: MDT nt, MDT aa, r8s nt, and r8s aa (top to bottom). Constrained nodes are marked with stars and the upper and lower constrain dates are shown in gray italics.

Discussion

One of the final questions of the relationship of placental mammal orders and higher clades concerns their most basal divergence. These basal relationships are also fundamental to all discussion of mammalian evolution. Recent phylogenetic analyses have commonly recognized 3 basal groups of placental mammals, Afrotheria, Xenarthra, and Boreoeutheria, but the relationships among the 3 groups have tended to be differently and inconclusively resolved (Waddell et al. 1999, 2001; Murphy, Elzirik, Johnson et al. 2001; Murphy, Elzirik, O'Brien et al. 2001; Murphy et al. 2007; Arnason et al. 2002; Amrine-Madsen et al. 2003; Kriegs et al. 2006; Kjer and Honeycutt 2007; Nikolaev et al. 2007). The problem associated with resolving these divergences indicates that they took place within a narrow temporal window, presumably within only a few million years. The current phylogenomic analyses of the 2.2 mega-base maxgen and the 1.6 mega-base maxspe data sets of 11 species of placental mammals unambiguously supported the Atlantogenata hypothesis, that is, a basal split between Afrotheria/Xenarthra and the remaining placentals, by significantly rejecting the 2 competing hypothesis. It appears that sequence data several times larger than that of the phylogenomic analysis of Nikolaev et al. (2007) were needed to conclusively resolve this phylogenetic question.

The effect of compositional heterogeneity of extraordinarily long sequence alignments on the phylogenetic analyses has been a vexed issue in many analyses. In the current study, the base composition was shown to be heterogeneous between several lineages. Although the percentages of A, G, C, and T appeared to be similar at first sight, a chi-square test of the long sequence alignments rejected homogeneity of nt composition for all species. Recoding the nt sequences to R and Y removed the bias for many placental species except for the cow, rabbit, and tenrec. The phylogenetic analysis of the R/Y recoded sequences or analyses using a nonstationary model on the unaltered data resulted in the same and strongly supported topology. Also, excluding species with a compositional bias did not alter the phylogenetic relationships among the basal placental mammals. This indicates that compositional heterogeneity did not affect the tree reconstruction based on this set of data. It could be argued that the chi-square test for rate heterogeneity is unnecessarily stringent for data sets of this size because a difference in base frequency of only a few per mille will cause the test to reject compositional homogeneity. In any case, the results suggested that differences at this level were not likely to adversely affect tree reconstruction in the current study.

Prior to the availability of genome data, the phylogeny of placental mammals was investigated with alignments that were 10–15 kbp long (Murphy, Elzirik, Johnson et al. 2001; Murphy, Elzirik, O'Brien et al. 2001; Arnason et al. 2002). However, jackknife analyses with 15 kbp of the data showed that while the presumably correct tree could be resolved in 84% of the replicates, the support for the deepest divergences was not significant. High Bayesian posterior probabilities are often reported for these short sequence data sets (e.g., Murphy, Elzirik, O'Brien et al. 2001), but they may overestimate the factual support (Simmons et al. 2004). Our analyses showed that far larger amounts of sequences were required to significantly resolve these deep splits.

Somewhat unexpectedly, the phylogenetic results did not agree with those of Nikolaev et al. (2007), despite their use of a total of 635,000 nt of protein-coding sequences and conserved noncoding sequences. The noncoding data set of Nikolaev et al. (2007) yielded equal likelihood values for the Atlantogenata and Afrotheria hypotheses. However, the Atlantogenata hypothesis was rejected in favor of the Afrotheria hypothesis in analyses of the aa data. The data set used by Nikolaev et al. (2007) showed a basal position of Xenartha and Afrotheria relative to the remaining placental mammals, but the support for the relative relationship between Xenartha, Afrotheria, and remaining placentals remained ambiguous. Their alignment contains to some extent missing data, and a considerable portion of the aa sequence shows distances to the outgroup that are almost as high as in random sequences. This may suggest that at least some of these sequences could have been erroneously aligned or may be nonhomologous. In regions showing large sequence distances, the outgroup will function as a random outgroup sequence and make the tree reconstruction prone to artifacts (Sullivan and Swofford, 1997).

Hitherto, analyses of rare genomic events such as insertion of retroposons and indel differences have not provided unambiguous support for any of the 3 hypotheses related to basal relationships among placental mammals. Retroposons are a robust and well-understood tool for studying mammalian evolution and have clarified several phylogenetic issues in mammalian evolution (Schmitz et al. 2005; Kriegs et al. 2006). Yet, only a few retroposons have been identified for examining basal mammalian divergences. The statistics behind retrotransposition has not been fully developed, but it has been suggested that 3 such events are needed to provide “significant” support for a particular phylogenetic hypothesis (Waddell et al. 2001). Other genomic events such as “rare” indel events are still far from being understood for their usefulness for phylogenetic reconstruction. Four such events have been reported in favor of the Atlantogenata hypothesis (Murphy et al. 2007), but 2 of these, ZNF367 exon 5 and LAMC2 exon 13, show signs of homoplasy.

Analyses of large data sets of protein-coding sequences (Jorgensen et al. 2005; Kullberg et al. 2006; Cannarozzi et al. 2007) have suggested that rodents may not group with primates as postulated by the Euarchontoglires hypothesis. Changing the taxon sampling in the 2.2 mega-base “maxseq” alignment led to the same result when rate homogeneity models were used, whereas application of rate heterogeneity models placed rodents and primates on a common branch. The latter topology became further stabilized with the inclusion of additional taxa, for example, rabbit and elephant. In this case, the findings were in contrast to the study of Rokas and Carroll (2005) on yeast phylogenomics, which showed that the inclusion of additional taxa could decrease the support for what was taken as being the correct topology. The results suggest that large amounts of sequence data in conjunction with appropriate taxon sampling are crucial to the recovery of a strongly supported placental mammalian tree. Recently, Hedtke et al. (2006) and Gatesy et al. (2007) addressed the topic of taxon sampling and the amounts of sequence data required in phylogenomics. As discussed by Gatesy et al. (2007), exclusion of “problematic” taxa, could promote the recovery of congruent trees, even with small amounts of data. However, problematic taxa can usually only be identified in hindsight. In the current study, inclusion of the rabbit stabilized the boreoeutherian branch and similarly the elephant neutralized the adverse effect of the fast evolving tenrec, thereby stabilizing Afrotheria and resulting in a more strongly supported tree.

The phylogenetic nomenclature related to the superorders Boreoeutheria, Laurasiatheria, Atlantogenata, and Afrotheria has been derived from the biogeography of extant mammalian orders (Waddell et al. 1999; Murphy, Elzirik, O'Brien et al. 2001). This is problematic, however, partly because the superorders overlap spatially and partly because the names, implicitly and explicitly, imply that some groups (Xenarthra and Afrotheria) originated on the supercontinent Gondwana in the southern hemisphere (Murphy, Elzirik, O'Brien et al. 2001). Assumptions about the origin of living groups based on their current biogeography may not be warranted, however, because they ignore the mammalian fossil record. Two recent studies have pointed out that the most parsimonious assumption for basal divergences among placental mammals is that they took place in Laurasia (Archibald 2003; Hunter and Janis 2006).

The molecular dating of the origin of Xenarthra at ≈97 MYA is consistent with other studies and the fossil record, which have indicated that certain orders were already present in the Cretaceous (Asher et al. 2005; Benton and Donoghue 2007). The stem group Xenarthra has probably not originated in South America but rather migrated during later stages in its evolution. The primary argument for this reasoning is that before the Cretaceous–Tertiary (K/T) boundary at 65 MYA, South America was only the home of archaic mammalian groups, such as multituberculates, gondwanatheres, and sudamericids (Flynn and Wyss 1998; Pascual and Ortiz-Jaureguizar 2007). At the K/T boundary, there was a major turnover in the South American mammalian community. The archaic mammals became extinct and were replaced by therian mammals (Pascual and Ortiz-Jaureguizar 2007). This faunal change correlates well with molecular estimates of the origin and diversification of modern lineages of the 2 major South American mammalian groups, marsupials at 68.5 MYA and xenarthrans at 65 MYA (Delsuc et al. 2004; Nilsson et al. 2004). There is fossil evidence suggesting that stem group marsupials radiated in North America and some migrated to South America using a land bridge (the Aves ridge route) that connected the 2 continents across the Caribbean in the late Cretaceous (Case et al. 2005).

The strongly supported sister group relationship between Afrotheria and Xenarthra may serve to reduce the discussion of the biogeography of early placental mammals to 2 particular topics, Atlantogenata and Boreoeutheria. The name Atlantogenata is connected to the opening up of the Atlantic Ocean, implying that this event led to the divergence of Xenarthra and Afrotheria by vicariance (Waddell et al. 1999). The depauperate mammalian African fossil record covering the period 95–70 MYA does not suggest an existence of Xenarthra in Africa during this period, but does not exclude it either. Similarly, and more strikingly, there are no South American fossils of this age that can be related to Xenarthra or Afrotheria, despite the occurrence of other mammalian fossils on this continent. This circumstance strongly suggests that xenarthrans and other therian mammals colonized South America from the north (Laurasia) in the late Cretaceous. Likewise, the fossil record does not support an African origin of the Afrotheria because the oldest members of crown group Afrotheria, are also of Laurasian origin (Asher et al. 2003; Zack et al. 2005; Tabuce et al. 2007).

The current analyses of the relatively conserved 2.2 mega-base data set of protein-coding sequences have yielded strong support to a basal divergence between Atlantogenata and remaining placental mammals. It would be advantageous to confirm these results by sequence independent data such as retroposon insertion as in Kriegs et al. (2006) or other rare genomic events (Boore 2006). However, the temporal narrowness of the intervals may affect the usability of those characters because incomplete lineage sorting within narrow temporal widows may interfere with these approaches (Nishihara et al. 2006).

Provided the phylogenetic results reflect the true tree and the molecular estimates are reasonably accurate, the findings place the divergence between Atlantogenata and remaining placentals (Boreoeutheria) at ≈99–100 MYA. The split between Xenarthra and Afrotheria is estimated to have taken place ≈2 Myr later (i.e., 97–98 MYA), whereas the basal divergence within Boreoeutheria was estimated to have taken place ≈95–96 MYA and those within Laurasiatheria at 84–88 MYA. These divergence dates are somewhat younger than those estimated earlier on 10–15 kbp of nuclear and mitochondrial data (Arnason and Janke 2002; Murphy et al. 2007). Irrespective of these differences, all these estimates pinpoint to the limited temporal interval within which the basal diversification of placental mammals took place. The time intervals of 2–3 Myr between major branching events correspond to or are only slightly larger than the typical duration of mammalian speciation (Curnoe et al. 2006; van Dam et al. 2006). This provides an explanation to the fact that data sets of lesser sizes may have tended, for stochastic reasons, to resolve these divergences in different ways and illustrates the need of very large sequence data or very specific rare genomic events to resolve such divergences that occurred about 100 MYA.

The Swedish Science Council, the Jörgen Lindström Foundation, and the Royal Physiographic Society (Nilsson-Ehle) supported the study. We thank Torbjön Säll for discussion and advice on the statistical properties of large sequence data and Ulfur Arnason for comments on the manuscript.

References

ProtTest: selection of best-fit models of protein evolution

,

Bioinformatics

,

2005

, vol.

21

(pg.

2104

-

2105

)

A new phylogenetic marker, apolipoprotein B, provides compelling evidence for eutherian relationships

,

Mol Phylogenet Evol

,

2003

, vol.

28

(pg.

225

-

240

)

Timing and biogeography of the eutherin radiation: fossils and molecules compared

,

Mol Phylogenet Evol

,

2003

, vol.

28

(pg.

350

-

359

)

et al. ,

(11 co-authors)

.

Mammalian mitogenomic relationships and the root of the eutherian tree

,

Proc Natl Acad Sci USA

,

2002

, vol.

99

(pg.

8151

-

8156

)

Mitogenomic analyses of eutherian realtionships

,

Cytogenet Genome Res

,

2002

, vol.

96

(pg.

20

-

32

)

The antiquity of Glires

,

Science

,

2005

, vol.

307

(pg.

1091

-

1094

)

Relationships of endemic African mammals and their fossil relatives based on morphological and molecular evidence

,

J Mamm Evol

,

2003

, vol.

10

(pg.

131

-

162

)

Paleontological evidence to date the tree of life

,

Mol Biol Evol

,

2007

, vol.

24

(pg.

26

-

53

)

The use of genome-level characters for phylogenetic reconstruction

,

Trends Ecol Evol

,

2006

, vol.

21

(pg.

439

-

446

)

A phylogenomic study of human, dog and mouse

,

PLoS Comput Biol

,

2007

, vol.

3

pg.

e2

doi:10.1371/journal.pcbi.0030002

“South American” marsupials from the Late Cretaceous of North America and the origins of marsupial cohorts

,

J Mamm Evol

,

2005

, vol.

12

(pg.

461

-

494

)

Timing and tempo of primate speciation

,

J Evol Biol

,

2006

, vol.

19

(pg.

59

-

65

)

Influence of Tertiary paleoenvironmental changes on the diversification of South American mammals: a relaxed molecular clock study within xenarthrans

,

BMC Evol Biol

,

2004

, vol.

4

pg.

11

Evolutionary trees from DNA sequences: a maximum likelihood approach

,

J Mol Evol

,

1981

, vol.

17

(pg.

368

-

376

)

Recent advances in South American mammalian paleontology

,

Trends Ecol Evol

,

1998

, vol.

13

(pg.

449

-

454

)

How many genes should a systematist sample? Conflicting insights from a phylogenomic matrix characterized by replicated incongruence

,

Syst Biol

,

2007

, vol.

56

(pg.

355

-

363

)

Likelihood-based tests of topologies in phylogenetics

,

Syst Biol

,

2000

, vol.

49

(pg.

652

-

670

)

A simple, fast and accurate algorithm to estimate large phylogenies by maximum likelihood

,

Syst Biol

,

2003

, vol.

52

(pg.

696

-

704

)

Resolution of phylogenetic conflict in large data sets by increased taxon sampling

,

Syst Biol

,

2006

, vol.

55

(pg.

522

-

529

)

MrBayes: Bayesian inference of phylogeny

,

Bioinformatics

,

2001

, vol.

17

(pg.

754

-

755

)

Spiny Norman in the garden of eden?

,

J Mamm Evol

,

2006

, vol.

13

(pg.

89

-

123

)

TREEFINDER: a powerful graphical analysis environment for molecular phylogenetics

,

BMC Evol Biol

,

2004

, vol.

4

pg.

18

Comparative analysis of protein coding sequences from human, mouse and the domesticated pig

,

BMC Biol

,

2005

, vol.

3

pg.

2

Site specific rates of mitochondrial genomes and the phylogeny of eutheria

,

BMC Evol Biol

,

2007

, vol.

7

pg.

8

Retroposed elements as archives for the evolutionary history of placental mammals

,

PLoS Biol

,

2006

, vol.

4

pg.

e91

Housekeeping genes for phylogenetic analysis of eutherian relationships

,

Mol Biol Evol

,

2006

, vol.

23

(pg.

1493

-

1503

)

A new method for calculating evolutionary substitution rates

,

J Mol Evol

,

1984

, vol.

20

(pg.

86

-

93

)

Toward a phylogenetic classification of the mammalia

,

Phylogeny of the primates

,

1975

New York

Plenum Press

(pg.

21

-

46

)

PANTHER version 6: protein sequence and function evolution data with expanded representation of biological pathways

,

Nucleic Acids Res

,

2007

, vol.

35

(pg.

D247

-

D252

)

Molecular phylogenetics and the origins of placental mammals

,

Nature

,

2001

, vol.

409

(pg.

614

-

618

)

et al. ,

(11 co-authors)

.

Resolution of the early placental mammal radiation using Bayesian phylogenetics

,

Science

,

2001

, vol.

294

(pg.

2348

-

3235

)

Using genomic data to unravel the root of the placental mammal phylogeny

,

Genome Res

,

2007

doi:10.1101/gr.5918807

Early history of mammals is elucidated with the ENCODE multiple species sequencing data

,

PLoS Genet

,

2007

, vol.

3

pg.

1

Marsupial relationships and a timeline for marsupial radiation in South Gondwana

,

Gene

,

2004

, vol.

340

(pg.

189

-

196

)

Pegasoferae, an unexpected mammalian clade revealed by tracking ancient retroposon insertions

,

Proc Natl Acad Sci USA

,

2006

, vol.

103

(pg.

9929

-

9934

)

Mammalian phylogeny: shaking the tree

,

Nature

,

1992

, vol.

356

(pg.

121

-

125

)

The Gondwanan and South American episodes: two major and unrelated moments in the history of the South American mammals

,

J Mamm Evol

,

2007

doi:10.1007/s10914-007-9039-5

Modeltest: testing the model of DNA substitution

,

Bioinformatics

,

1998

, vol.

14

(pg.

817

-

818

)

Genomic evidence for two functionally distinct gene classes

,

Proc Natl Acad Sci USA

,

1998

, vol.

95

(pg.

6239

-

6244

)

More genes or more taxa? The relative contribution of gene number and taxon number to phylogenetic accuracy

,

Mol Biol Evol

,

2005

, vol.

22

(pg.

1337

-

1344

)

R8s: inferring absolute rates of molecular evolution and divergence times in the absence of a molecular clock

,

Bioinformatics

,

2002

, vol.

19

(pg.

301

-

302

)

Primate phylogeny: molecular evidence from retroposons

,

Cytogenet Genome Res

,

2005

, vol.

108

(pg.

26

-

37

)

Multiple comparisons of log-likelihoods with applications to phylogenetic inference

,

Mol Biol Evol

,

1999

, vol.

16

(pg.

1114

-

1116

)

Higher taxonomic relationships among extant mammals based on morphology, with selected comparisons of results from molecular data

,

Mol Phylogenet Evol

,

1998

, vol.

9

(pg.

572

-

584

)

How meaningful are Bayesian support values?

,

Mol Biol Evol

,

2004

, vol.

21

(pg.

188

-

199

)

Which mammalian supertree to bark up?

,

Science

,

2001

, vol.

291

(pg.

1709

-

1711

)

Inferring confidence sets of possibly misspecified gene trees

,

Proc Biol Sci

,

2002

, vol.

269

(pg.

137

-

142

)

Quartet puzzling: a quartet maximum likelihood method for reconstructing tree topologies

,

Mol Biol Evol

,

1996

, vol.

13

(pg.

964

-

969

)

Are guinea pigs rodents? The importance of adequate models in molecular phylogenetics

,

J Mamm Evol

,

1997

, vol.

4

(pg.

77

-

86

)

Early tertiary mammals from North Africa reinforce the molecular Afrotheria clade

,

Proc Biol Sci

,

2007

, vol.

274

(pg.

1159

-

1166

)

PANTHER: a library of protein families and subfamilies indexed by function

,

Genome Res

,

2003

, vol.

13

(pg.

2129

-

2141

)

Divergence time and evolutionary rate estimation with multilocus data

,

Syst Biol

,

2002

, vol.

51

(pg.

689

-

702

)

Long-period astronomical forcing of mammal turnover

,

Nature

,

2006

, vol.

443

(pg.

687

-

691

)

Assessing the cretaceous superordinal divergence times within birds and placental mammals by using whole mitochondrial protein sequences and an extended statistical framework

,

Syst Biol

,

1999

, vol.

48

(pg.

119

-

137

)

A phylogenetic foundation for comparative mammalian genomics

,

Genome Inform

,

2001

, vol.

12

(pg.

141

-

154

)

Evolutionary history of LINE-1 in the major clades of placental mammals

,

PLoS ONE

,

2007

, vol.

2

pg.

e158

doi:10.1371/journal.pone.0000158

Maximum likelihood phylogenetic estimation from DNA sequences with variable rates over sites: approximate methods

,

J Mol Evol

,

1994

, vol.

39

(pg.

306

-

314

)

PAML: a program package for phylogenetic analysis by maximum likelihood

,

Comput Appl Biosci

,

1997

, vol.

13

(pg.

555

-

556

)

On the use of nucleic acid sequences to infer early branchings in the tree of life

,

Mol Biol Evol

,

1995

, vol.

12

(pg.

451

-

458

)

Affinities of ‘hyopsodontids’ to elephant shrews and a Holarctic origin of Afrotheria

,

Nature

,

2005

, vol.

434

(pg.

497

-

501

)

Author notes

Arndt von Haeseler, Associate Editor

© The Author 2007. Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org

Supplementary data

Citations

Views

Altmetric

Metrics

Total Views 2,328

1,837 Pageviews

491 PDF Downloads

Since 2/1/2017

Citations

99 Web of Science

×

Email alerts

Email alerts

Citing articles via

• Latest

• Most Read

• Most Cited

More from Oxford Academic