MEGA4: Molecular Evolutionary Genetics Analysis (MEGA)
Software Version 4.0
Koichiro Tamura,*
Joel Dudley,* Masatoshi Nei, and Sudhir Kumar§*
*Center for Evolutionary Functional Genomics, The Biodesign Institute, Arizona State University;
Department of Biological
Sciences, Tokyo Metropolitan University, Tokyo, Japan; Department of Biology and the Institute of Molecular Evolutionary
Genetics, The Pennsylvania State University; and §School of Life Sciences, Arizona State University
We announce the release of the fourth version of MEGA software, which expands on the existing facilities for editing
DNA sequence data from autosequencers, mining Web-databases, performing automatic and manual sequence
alignment, analyzing sequence alignments to estimate evolutionary distances, inferring phylogenetic trees, and testing
evolutionary hypotheses. Version 4 includes a unique facility to generate captions, written in figure legend format, in
order to provide natural language descriptions of the models and methods used in the analyses. This facility aims to
promote a better understanding of the underlying assumptions used in analyses, and of the results generated. Another
new feature is the Maximum Composite Likelihood (MCL) method for estimating evolutionary distances between all
pairs of sequences simultaneously, with and without incorporating rate variation among sites and substitution pattern
heterogeneities among lineages. This MCL method also can be used to estimate transition/transversion bias and
nucleotide substitution pattern without knowledge of the phylogenetic tree. This new version is a native 32-bit Windows
application with multi-threading and multi-user supports, and it is also available to run in a Linux desktop environment
(via the Wine compatibility layer) and on Intel-based Macintosh computers under the Parallels program. The current
version of MEGA is available free of charge at http://www.megasoftware.net.
Since the early 1990s, MEGA software functionality
has evolved to include the creation and exploration of
sequence alignments, the estimation of sequence diver-
gence, the reconstruction and visualization of phylogenetic
trees, and the testing of molecular evolutionary hypotheses.
The three versions of MEGA have been released, and they
integrate Web-based sequence data acquisition and align-
ment capabilities (fig. 1) with the evolutionary analyses
(fig. 2), making it much easier to conduct comparative anal-
yses in a single computing environment (Kumar, Tamura,
and Nei 2004). Over time, MEGA has come to enhance
the classroom learning experience as its use by researchers,
educators, and students in diverse disciplines has expanded
(Kumar and Dudley 2007). The fourth version (MEGA4)
contains three distinct newly developed functionalities,
which are outlined below.
First, we have developed a Caption Expert software
module that generates descriptions for every result obtained
by MEGA4. This description informs the user of all of
the options used in the analysis, including the data subset
used (e.g., codon positions included), the chosen option for
the handling of sites with gaps or missing data, the evolu-
tionary model of substitution (e.g., DNA substitution pat-
tern, uniformity of evolutionary rates among sites, and
homogeneity assumption among lineages), and the methods
applied for estimating pairwise distances and for inferring
and testing phylogeny. The caption also includes specific
citations for any method, algorithm, and software used in
the given analysis. Two examples of descriptions generated
by the Caption Expert are shown in figure 3.
The availability of these descriptions is intended to
promote a better understanding of the underlying assump-
tions used in analyses, and of the results produced. This is
needed because MEGA’s intuitive graphical interface
makes it easy for both new and expert users to
conduct a variety of computational and statistical analyses.
However, some users may not immediately realize the
underlying assumptions and data-handling options in-
volved in each analysis. Even expert molecular and popu-
lation geneticists may not be able to discern all of the
assumptions implied. In general, we expect a written de-
scription of methods and results to be useful for students
and researchers when preparing tables and figures for pre-
sentation and publication.
Second, we have now added a Maximum Composite
Likelihood (MCL) method for estimating evolutionary
distances (dij) between DNA sequences, which MEGA
users frequently employ for inferring phylogenetic trees,
divergence times, and average sequence divergences
between and within groups of sequences. In this approach,
the Composite Log Likelihood (CL) obtained as the sum
of log likelihood for all sequence pairs in an alignment
is maximized by fitting the common parameters for nucle-
otide substitution pattern (h) to every sequence pair (i,j):
CL5
P
i;j ln lðh; dijÞ (Tamura, Nei, and Kumar 2004). This
approach was previously referred to as the ‘‘Simultaneous
Estimation’’ (SE) method, because all dij’s are simul-
taneously estimated (Tamura, Nei, and Kumar 2004).
The MCL approach differs from current approaches for
evolutionary distance estimation, wherein each distance
is estimated independently of others, either by analytical
formulas or by likelihood methods (independent estimation
[IE] approach).
The MCL method has many advantages over the IE
approach. To begin with, the IE method for estimating evo-
lutionary distance for each pair of sequences will often
cause rather large errors unless very long sequences are
used. The use of the MCL method reduces these errors con-
siderably, as a single set of parameters estimated from all-
sequence pairs is applied to each distance estimation. When
distances are estimated with lower errors, distance-based
methods for inferring phylogenies are expected to
be more accurate. This is indeed the case for the
Key words: selection, genomics, phylogenetics, software, cross-
platform.
E-mail: s.kumar@asu.edu
Mol. Biol. Evol. 24(8):1596–1599. 2007
doi:10.1093/molbev/msm092
Advance Access publication May 7, 2007

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

Neighbor-Joining method (Saitou and Nei 1987), as the use
of the MCL distances leads to a much higher accuracy
(Tamura, Nei, and Kumar 2004). Even when the topologies
estimated are the same, the use of the MCL distances
often gives higher bootstrap values for the estimated phy-
logenetic tree compared to the use of IE distances, which is
evident from the example given in figure 4 A (MCL: bold,
IE: italics).
In addition, the IE distances are not always estimable
when pairwise distances are calculated between very dis-
tantly related sequences, because the arguments of loga-
rithms in the analytical formulas may become negative
by chance. The probability of occurrence of such inappli-
cable cases increases as the number of sequences in the
data increases, the evolutionary distances become larger,
and the substitution pattern becomes more complex
(Tamura, Nei, and Kumar 2004). The use of the MCL
method eliminates this problem effectively and allows
for the use of sophisticated models in inferring phylogenies
from an increasingly larger number of diverse sequences.
MEGA4 implements the MCL approach for estimat-
ing distances between sequence pairs, average distances
between and within groups, and average pairs overall with
their variances estimated by a bootstrap approach. Our
implementation of the MCL method allows for the consid-
eration of substitution rate variation from site to site, using
an approximation of the gamma distribution of evolutionary
rates, and the incorporation of heterogeneity of base com-
position in different species/sequences. The user also has
the flexibility to estimate the numbers of transition and
transversion type substitutions per site separately. Natu-
rally, the MCL distances can be used for inferring phylog-
enies by the distance-based methods, along with the
bootstrap tests of phylogenies.
MEGA4 implements the MCL approach under the
Tamura-Nei (1993) substitution model, in which the rates
of two types of transitional substitutions (between purines
[a1] and between pyrimidines [a2]) and the rate of trans-
versional substitutions (b) are considered separately by
taking into account the unequal frequencies of four nucleo-
tides (base composition bias). The MCL estimates of the
transition/transversion rate ratio have been found to be
close to the true values in previous simulation experiments
(Tamura, Nei, and Kumar 2004). We have employed this
feature to provide users with a facility to compute the rel-
ative rates of substitutions between nucleotides based on the
MCL estimates of a1, a2, b, and on the observed frequencies
of the four nucleotides under the Tamura-Nei (1993) model
(fig. 3C). For ease of comparison, we have expressed these
substitution rates as relative frequencies of substitutions
FIG. 1.—Sequence alignment editor and Web-data mining features in MEGA4. In the Alignment Explorer (A), the integrated web browser (B)
permits downloading sequences from online databases directly into the current alignment, without the need for manual cutting-and-pasting and
reformatting. The DNA sequences can be translated to the corresponding protein sequences by a single mouse click (D), and the protein sequences can
be aligned by ClustalW (E) (Thompson, Higgins, and Gibson 1994) and adjusted manually by eye. Returning to the nucleotide view automatically
aligns the nucleotide sequences according to the protein alignments, and DNA and protein sequence alignments can be exported in a variety of formats
for use with other programs. Alignment Editor also contains facilities for editing and importing of trace data files output from DNA sequencers (C).
MEGA4 software 1597

between nucleotides such that the sum of all frequencies is
100 (see also Gojobori, Li, and Graur 1982).
Third, we have now programmed MEGA4 to run on
some versions of Linux through the Wine software com-
patibility layer (www.winehq.org). The first advancement
alleviates the problem of performance degradation (and
the need to purchase Windows emulation software) when
using MEGA on Linux. Wine is neither a hardware nor
a software emulator, but an open source tool that allows
for the native execution of Windows applications on Linux.
Our tests of MEGA4 running on Linux show the display,
stability, and performance to be highly satisfactory and
comparable to the native Windows system (fig. 4B). Fur-
thermore, investigators now report MEGA4 running on
FIG. 2.—A collection of menus that provide access to many different data analysis options in MEGA4, including exploration of input data set (A),
estimation of evolutionary distances (B), inferring and testing phylogenetic trees (C), tests of homogeneity of substitution patterns and its estimation
(D), tests of selection (E), alignment of DNA and protein sequences (F), and the dialog box that provides users with options to select model of
substitution and data sub-setting options (G).
FIG. 3.—The Tree-Explorer displaying a Neighbor-Joining tree of mitochondrial 16S rRNA sequences (A), and the description generated by the
Caption Expert (B). Estimates of the relative probabilities of nucleotide substitutions for 70 control-region sequences of human mitochondrial DNA
sequences are shown in (C). The gamma shape parameter (a5 0.35) was estimated using the Yang and Kumar (1996) method, and the rest of the
analysis details are given in (B). It is worth noting that the Tree Explorer shown in (A) includes a high-resolution tree drawing facility that includes
displaying trees in a variety of formats, with options to display/hide branch lengths as well as clade confidence labels, and re-rooting and rearranging
trees, among other functionalities. MEGA4 can export the drawings to graphics programs, and can export trees in Newick format for use by other
programs. Furthermore, MEGA can import and draw trees from Newick format files that have been estimated by other programs (see fig. 2C).
1598 Tamura et al.

Intel-based Macintosh computers under the Parallels pro-
gram as well as it does on Windows-native personal com-
puters (see Hall 2007). The Parallels program is a native
solution for Macintosh computers that permits them to
simultaneously run Windows and Macintosh software.
We have also built support for a multi-user environ-
ment, which will allow each user of the same computer
to keep his/her customized settings, including file locations,
window sizes, choice of genetic code table, and previously
used analysis options. This feature will facilitate educa-
tional and laboratory usage, where a single computer is
often shared by multiple users.
In conclusion, MEGA4 now contains a wide array of
functionalities for the molecular evolutionary analysis
of data (http://www.megasoftware.net/features.html). It is
useful to note that while we are continuously adding
new methods and functions to MEGA, we do not intend
to make it a catalog of all evolutionary analysis methods
available. Rather, it is anticipated to become a workbench
for the exploration of sequence data from evolutionary
perspectives.
Acknowledgments
We thank the colleagues, students, and volunteers
who spent countless hours testing the early release versions
of MEGA; almost all facets of MEGA’s design and imple-
mentation benefited from their comments. We thank
Ms. Linwei Wu for assistance with MEGA Web site and
for handling bugs, and Ms. Kristi Garboushian for edito-
rial support. We thank the two reviewers for suggesting
many useful text additions, which have been included
in the figure 1 legend and in the text. We also thank
Drs. Masafumi Nozawa and Barry Hall for comments on
an earlier version of this manuscript. The MEGA software
project is supported by research grants from National
Institutes of Health (S.K. and M.N.) and from Japan Society
for Promotion of Sciences (K.T.).
Literature Cited
Gojobori T, Li WH, Graur D. 1982. Patterns of nucleotide
substitution in pseudogenes and functional genes. J Mol Evol.
18:360–369.
Hall BG. Phylogenetic trees made easy: A how-to manual.
Sunderland (MA): Sinauer Associates.
Kumar S, Dudley J. 2007. Bioinformatics for biologists in
the genomics era. Bioinformatics. 10.1093/bioinformatics/
btm239.
Kumar S, Tamura K, Nei M. 2004. MEGA3: an integrated
software for Molecular Evolutionary Genetics Analysis and
sequence alignment. Brief Bioinform. 5:150–163.
Saitou N, Nei M. 1987. The Neighbor-Joining method—a new
method for reconstructing phylogenetic trees. Mol Biol Evol.
4:406–425.
Tamura K, Nei M. 1993. Estimation of the number of nucleo-
tide substitutions in the control region of mitochondrial-
DNA in humans and chimpanzees. Mol Biol Evol. 10:
512–526.
Tamura K, Nei M, Kumar S. 2004. Prospects for inferring very
large phylogenies by using the Neighbor-Joining method.
Proc Natl Acad Sci USA. 101:11030–11035.
Thompson JD, Higgins DG, Gibson TJ. 1994. ClustalW—
improving the sensitivity of progressive multiple sequence
alignment through sequence weighting, position-specific gap
penalties and weight matrix choice. Nucleic Acids Res.
22:4673–4680.
Yang Z, Kumar S. 1996. Approximate methods for estimating
the pattern of nucleotide substitution and the variation of
substitution rates among sites. Mol Biol Evol. 13:650–659.
William Martin, Associate Editor
Accepted May 2, 2007
FIG. 4.—(A) Bootstrap support for the branching order of 16 Laurasiatheria species reconstructed with MCL approach (bold) and without MCL
approach (italics) under the Tamura-Nei (1993) model (see figure 3B for rest of the analysis details). The 16S rRNA sequences used were downloaded
from GenBank and were aligned in MEGA4 using CLUSTALW (accession numbers: AJ428578, NC004029, X72004, AF303109, NC008093,
DQ480502, X97336, X79547, DQ534707, AJ554051, AJ554061, NC000889, NC007704, AB074968, NC005044, and NC001941). (B) Comparison of
MEGA4 performance benchmarks on Windows and Linux (with Wine application compatibility layer). Identical hardware configuration was used, and
example data sets included in the MEGA4 installation were employed. The results show that computations executed under Wine are penalized by about
2 s, which is attributable to the need for Wine’s initialization.
MEGA4 software 1599