Electronic Supplementary Material
Methods
(a) Sequence assembly and orthology assignment
Sequenced results were quality filtered accordingly to a threshold average quality
Phred score of 30 and adaptors trimmed using Trimgalore v 0.3.3 [1]. Ribosomal
RNA (rRNA) was filtered out using Bowtie v.1.0.0 [2]. De novo strand-specific
assemblies were made using Trinity [3,4] using paired read files, with an enforced
path reinforcement distance of 75. Resulting assemblies were processed in
TransDecoder [4] in order to identify candidate open reading frames (ORFs) within
the transcripts. Redundancy reduction was done with CD-HIT [5] for all Trinity
assemblies (95% similarity). Predicted peptides were then processed with a further
filter to select only one peptide per putative unigene, by choosing the longest ORF
per Trinity subcomponent, thus removing variation in coding regions attributable to
alternative splicing, closely-related paralogs, and allelic diversity. Predicted ORFs
were assigned to orthologous groups using the Orthologous MAtrix (OMA)
algorithm, (OMA stand-alone v.0.99u; [6,7]), which has been shown to outperform
alternative approaches (e.g., reciprocal best hit; OrthoDB) toward identification of
true orthologs and to minimize Type I error in orthology assignment [8].
Additional taxa not sequenced by us (table S1) were obtained from GenBank and
processed in TransDecoder. For Sanger-sequenced EST and 454 libraries,
redundancy reduction was done with CD-HIT as described above. Due to the small
size of additional data sets and/or the quality of the genome of Mesobuthus
martensii, predicted ORFs were assigned to orthologous groups using OMA in two
separate runs, one for Buthidae (Iurus, Chaerilus, and the pseudochactids used as
outgroups; taxon occupancy criterion set to representation in at least 19 taxa) and a
second for “Chactoidea” and Scorpionoidea sensu stricto (Iurus and Bothriurus used
as outgroups; taxon occupancy criterion set to representation in at least 16 taxa).
This was done for computational expediency as well as to ensure representation of
the smallest libraries in supermatrices.
(b) Heuristics in phylogenomic analysis
Maximum likelihood (ML) analyses were conducted using RAxML v.7.7.5 [9] and
Bayesian inference (BI) using PhyloBayes MPI 1.4e [10]. For RAxML v.7.7.5, we
implemented a unique CAT + LG4X + F model for each gene [11], with 100
independent starts for each analysis. Bootstrap resampling frequencies were
estimated with 250-500 replicates using a rapid bootstrapping algorithm [12].
Analyses with PhyloBayes MPI 1.4e were limited to smaller datasets (Matrices 1-2),
as implementation of PhyloBayes for large matrices requires intractable amounts of
time for convergence. We implemented PhyloBayes MPI 1.4e using the siteheterogeneous CAT + GTR model of evolution [13]. Four independent chains were
run for 10,216-59,229 cycles, depending on the data set analyzed. The initial 5,000
cycles were discarded as burnin, with convergence assessed using the maximum
bipartition discrepancies across chains. For both matrices, the chains did not
converge (maximum bipartition discrepancies approximately equal to 0.7), but
ingroup topologies were identical and well supported (posterior probability ≥ 0.97);
failure to converge was driven by uncertainty in the relative placement of two
outgroup terminals (alternating positions of Eremobates sp. and Ricinoides atewa as
sister group to Opiliones).
We incorporated a branch length mixture model of heterotachy, a model that has
been shown to outperform covarion and homotachous counterparts when
heterotachy is incident [14]. A maximum likelihood approach was implemented
using the PhyML v.3 extension software M3L [14-16]. We analyzed our two most
complete matrices (Matrix 1 and Matrix 11; both >90% occupancy) using a branch
length mixtures model (+B) with four mixture categories in conjunction with the LG
substitution model. A best-of-SPR-and-NNI branch-swapping algorithm was
selected to optimize tree topology, starting with a BioNJ tree. For both analyses, a
single mixture was selected (weight: 1), and all other mixture classes received a
weight of 0. Basal scorpion relationships in both analyses were consistent with their
CAT-GTR and/or CAT+LG4X counterpart topologies.
(c) Statistical tests of non-random distribution of missing data
To infer the impact of non-randomly distributed missing data, we treated the
absence or presence of genes as cells in a matrix and quantified the degree of
phylogenetic structure in the unoccupied cells for each of the 5,025 genes. Using
custom R and Python scripts, we scored the parsimony length of each column (gene)
on the tree topology shown in figure 2c. For a given gene, random distribution of
missing data is expected to yield a high parsimony score, whereas non-random
(clade-specific) distribution of missing data is expected to yield a low parsimony
score.
We then compared this to a null distribution of parsimony scores obtained for the
corresponding column upon randomly reshuffling terminals 1,000 times. If a
column’s parsimony length was below the lower bound of the 95% confidence
interval of parsimony scores in the null distribution, we scored the orthogroup as
having significantly non-randomly distributed missing data.
This strategy is a modification of an older approach (cladistic permutation tail
probability or PTP test; [17]), but inverts the interpretation of the result. As applied
here, a significant result means that the distribution of a gene’s absences in the
matrix is non-random and should therefore be discarded from analysis. As the
original PTP test was shown to have a high type I error, the test we apply is highly
conservative.
(d) Bayesian relative rates test
To assess the possibility of long-branch artifacts affecting the ingroup, a Bayesian
relative rates test (sensu [18]) was conducted as follows: from the Bayesian
inference analysis of Matrix 2, 500 post-burnin trees from each of the four
PhyloBayes chains were randomly selected. All 2,000 trees were rooted with
Limulus polyphemus, and all chelicerate orders represented by a single terminal
were culled (Limulus polyphemus, Eremobates sp., and Ricinoides atewa). The
patristic distances from the root to each terminal was calculated across the set of
2,000 trees. The resulting set of distances for each terminal was used to conduct
statistical comparison of branch length distributions [18].
References
1.
Wu ZP, Wang X, Zhang XG. 2011. Using non-uniform read distribution models to improve isoform
expression inference in RNA-Seq. Bioinformatics 27:502-508.
2.
Langmead B, Trapnell C, Pop M, Salzberg SL. 2009 Ultrafast and memory-efficient alignment of
short DNA sequences to the human genome. Genome Biology 10, R25.
3.
Grabherr MG et al. 2011 Full-length transcriptome assembly from RNA-seq data without a
reference genome. Nature Biotechnology 29, 644-652. (doi:10. 1038/nbt.1883)
4.
Haas BJ, Papanicolaou A, Yassour M, Grabherr M, Blood PD, Bowden J, Couger MB, Eccles D, Li B,
Lieber M, et al. 2013 De novo transcript sequence reconstruction from RNA-seq using the Trinity
platform for reference generation and analysis. Nature Protocols 8, 1494-1512.
5.
Fu LM, Niu BF, Zhu ZW, Wu ST, Li WZ. 2012 CD-HIT: accelerated for clustering the nextgeneration sequencing data. Bioinformatics 28, 3150-3152.
6.
Altenhoff AM, Schneider A, Gonnet GH, Dessimoz C. 2011 OMA 2011: Orthology inference among
1000 complete genomes. Nucleic Acids Research 39, D289-D294.
7.
Altenhoff AM, Gil M, Gonnet GH, Dessimoz C. 2013 Inferring hierarchical orthologous groups
from orthologous gene pairs. PLoS ONE 8, e53786.
8.
Altenhoff AM, Dessimoz C. 2009 Phylogenetic and Functional Assessment of Orthologs Inference
Projects and Methods. PLoS Computational Biology 5, e1000262.
9.
Berger SA, Krompass D, Stamatakis A. 2011 Performance, accuracy, and Web server for
evolutionary placement of short sequence reads under maximum likelihood. Systematic Biology
60, 291-302.
10. Lartillot N, Rodrigue N, Stubbs D, Richer J. 2013 PhyloBayes MPI: Phylogenetic reconstruction
with infinite mixtures of profiles in a parallel environment. Systematic Biology 62, 611-615.
11. Le SQ, Dang CC, Gascuel O. 2012 Modeling protein evolution with several amino acid replacement
matrices depending on site rates. Molecular Biology and Evolution 29, 2921-2936.
12. Stamatakis A, Hoover P, Rougemont J. 2008. A rapid bootstrap algorithm for the RAxML Web
servers. Systematic Biology 57, 758-771.
13. Lartillot N, Philippe H. 2004 A Bayesian mixture model for across-site heterogeneities in the
amino-acid replacement process. Molecular Biology and Evolution 21, 1095-1109.
14. Kolaczkowski B, Thornton JW. 2008 A Mixed Branch Length Model of Heterotachy Improves
Phylogenetic Accuracy. Molecular Biology and Evolution 25, 1054-1066.
15. Guindon S, Dufayard JF, Lefort V, Anisimova M, Hordijk W, Gascuel O. 2010 New Algorithms and
Methods to Estimate Maximum-Likelihood Phylogenies: Assessing the Performance of PhyML
3.0. Systematic Biology 59, 307-21.
16. Hanson-Smith V. 2013 M3L. Google Code Repository (https://code.google.com/p/m3l/)
17. Faith DP. 1991. Cladistic permutation tests for monophyly and nonmonophyly. Systematic
Zoology 40, 366-375.
18. Wilcox TP, García de Léon FJ, Hendrickson DA, Hillis DM. 2004. Convergence among cave
catfishes: long-branch attraction and a Bayesian relative rates test. Molecular Phylogenetics and
Evolution 31, 1101-1103.