Universally distributed single-copy genes indicate a constant rate of horizontal transfer.

Abstract

Single copy genes, universally distributed across the three domains of life and encoding mostly ancient parts of the translation machinery, are thought to be only rarely subjected to horizontal gene transfer (HGT). Indeed it has been proposed to have occurred in only a few genes and implies a rare, probably not advantageous event in which an ortholog displaces the original gene and has to function in a foreign context (orthologous gene displacement, OGD). Here, we have utilised an automatic method to identify HGT based on a conservative statistical approach capable of robustly assigning both donors and acceptors. Applied to 40 universally single copy genes we found that as many as 68 HGTs (implying OGDs) have occurred in these genes with a rate of 1.7 per family since the last universal common ancestor (LUCA). We examined a number of factors that have been claimed to be fundamental to HGT in general and tested their validity in the subset of universally distributed single copy genes. We found that differing functional constraints impact rates of OGD and the more evolutionarily distant the donor and acceptor, the less likely an OGD is to occur. Furthermore, species with larger genomes are more likely to be subjected to OGD. Most importantly, regardless of the trends above, the number of OGDs increases linearly with time, indicating a neutral, constant rate. This suggests that levels of HGT above this rate may be indicative of positively selected transfers that may allow niche adaptation or bestow other benefits to the recipient organism.

Full text

Page 1

Universally Distributed Single-Copy Genes Indicate a
Constant Rate of Horizontal Transfer
Christopher J. Creevey1, Tobias Doerks2, David A. Fitzpatrick3, Jeroen Raes4, Peer Bork2*
1Animal and Bioscience Research Department, Teagasc, Grange, Dunsany, Ireland, 2European Molecular Biology Laboratory, Heidelberg, Germany, 3Department of
Biology, National University of Ireland Maynooth, Maynooth, Ireland, 4VIB Department of Molecular and Cellular Interactions, Vrije Universiteit Brussels, Brussels, Belgium
Abstract
Single copy genes, universally distributed across the three domains of life and encoding mostly ancient parts of the
translation machinery, are thought to be only rarely subjected to horizontal gene transfer (HGT). Indeed it has been
proposed to have occurred in only a few genes and implies a rare, probably not advantageous event in which an ortholog
displaces the original gene and has to function in a foreign context (orthologous gene displacement, OGD). Here, we have
utilised an automatic method to identify HGT based on a conservative statistical approach capable of robustly assigning
both donors and acceptors. Applied to 40 universally single copy genes we found that as many as 68 HGTs (implying OGDs)
have occurred in these genes with a rate of 1.7 per family since the last universal common ancestor (LUCA). We examined a
number of factors that have been claimed to be fundamental to HGT in general and tested their validity in the subset of
universally distributed single copy genes. We found that differing functional constraints impact rates of OGD and the more
evolutionarily distant the donor and acceptor, the less likely an OGD is to occur. Furthermore, species with larger genomes
are more likely to be subjected to OGD. Most importantly, regardless of the trends above, the number of OGDs increases
linearly with time, indicating a neutral, constant rate. This suggests that levels of HGT above this rate may be indicative of
positively selected transfers that may allow niche adaptation or bestow other benefits to the recipient organism.
Citation: Creevey CJ, Doerks T, Fitzpatrick DA, Raes J, Bork P (2011) Universally Distributed Single-Copy Genes Indicate a Constant Rate of Horizontal
Transfer. PLoS ONE 6(8): e22099. doi:10.1371/journal.pone.0022099
Editor: David Liberles, University of Wyoming, United States of America
Received November 17, 2010; Accepted June 17, 2011; Published August 5, 2011
Copyright: ? 2011 Creevey et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This project was funded under the EU 7th Framework Programme (number: HEALTH-F4-2007-201052). CJC is funded under the Science Foundation
Ireland (SFI) Stokes Lecturer Scheme (number: 07/SK/B1236A). The funders had no role in study design, data collection and analysis, decision to publish, or
preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: bork@embl.de
Introduction
From the earliest comparative genomic studies it was obvious
that horizontal gene transfer (HGT) occurred frequently [1–3] and
would impinge upon our efforts to understand the evolutionary
history of all life [4–8]. HGT has been shown to occur between
both closely and distantly related organisms, in both fast and
slowly evolving gene families [9–12]. While proteins with multiple
interactions are not immune to transfer [13,14], they seem to
undergo fewer HGT events providing evidence for the ‘‘complex-
ity hypothesis’’ [1,15]. Efforts have been made to quantify the
effect of HGT in completely sequenced organisms [7], including
estimating the rates of HGT across all organisms [16,17].
Although barriers for HGT have been revealed [15,18–20]
fundamental factors that influence the rate of HGT remain to
be identified, proven and quantified [21].
The vast majority of gene families shared by multiple organisms
have undergone horizontal transfer events at some point in their
evolutionary history [22]. However, as the majority of these
families are present in multiple copies in at least some organisms,
factors influencing the correct identification of orthologs, or the
occurrence of duplication and multiple loss events [23,24] can mar
the identification of fundamental factors influencing the rate of
HGT in these genes. In order to address this, selection of gene
families for analysis must minimise the potential for inclusion of
these kinds of events. To that end, we focussed on the
approximately 1% [25] of gene families that are universally single
copy [14], and likely have been functionally preserved since the
emergence of the three domains of life. When these genes are
successfully subjected to HGT, they should maintain the
interactions of the original copy that they displace (orthologous
gene displacement (OGD) [26]). This should be an extremely rare
process as two copies of genes that are part of large multi subunit
complexes (like the genes studies here) will not be tolerated due to
dosage effects [20]. Furthermore, it has been suggested that even
marginal differences in sequence identity between the displaced
copy and it’s replacement is enough to cause a marked decrease in
fitness of the acceptor organism [27], requiring compensatory
evolutionary change to occur. This suggests that the successful
fixation of an OGD in these genes requires overcoming the most
stringent barriers of any horizontal transfer event [20]. These
characteristics make HGTs in these genes potentially important
for the elucidation of constraints and promoting factors of HGT in
general.
Here we introduce an automated approach of detecting OGD
events in universal single gene families that i) is based on a
statistical framework, ii) has the ability to detect ancient events and
iii) can determine not only the recipient but also the donor
organism. Applied to 40 universal single-copy genes in 191 species
with completely sequenced genomes, we explored parameter-
space to identify optimum settings. The surprisingly large number
of robust orthologous gene displacements identified allowed us to
PLoS ONE | www.plosone.org1August 2011 | Volume 6 | Issue 8 | e22099

Page 2

quantify factors that have governed the occurrence of horizontally
transferred orthologous gene replacements since the last universal
common ancestor (LUCA).
Results and Discussion
Detection of OGDs
Our approach of automatically detecting OGDs is based on the
comparison of the phylogenetic signal of each individual gene
family (gene tree) to the combined phylogenetic signal of all the
genes used in the study (combined tree). The theory behind this
approach is that phylogenetic signal is cumulative as opposed to
homoplastic noise which is dispersive [28], therefore strong
disagreement between the combined phylogenetic signal of all
the gene families and that of any one gene may be representative
of a homoplastic event like HGT [29]. Our method depends on
the genes studied sharing a core phylogenetic history, such as was
demonstrated previously for the informational genes used in this
analysis [14]. This is a similar idea to the commonly used
approach of comparing a species tree with a gene tree to identify
HGT, but allows the identification of HGT in sets of functionally
related genes with a shared core phylogenetic history (like
information processing genes, or genes in operons) in organisms
where a species tree concept may not apply or be difficult to
reconstruct (like in prokaryotes).
We identified 40 gene families that are universally distributed in
single copy across all life and used their combined phylogenetic
signal to construct a tree. This tree was then used in an exhaustive
maximum likelihood procedure where the sequence data for each
individual gene family was used to determine the best phylogenetic
placement of every branch of the tree (inspired by [30] where the
concept had been used to infer species compositions from
metagenomics samples), identifying when this indicated a possible
orthologous gene displacement (Figure 1).
To identify the ‘‘best’’ parameters for our analysis and to
examine the robustness of results to different parameter selection,
a wide range of parameters were explored for the number of
OGDs identified (Figure 2). Depending on the settings used the
number of HGTs detected ranged from 0 to 80, however
interestingly the parameter exploration identified a range of
settings where the number of OGDs plateaued at around 65
(Figure 2). Other parameter settings existed that increased the
number of OGDs detected above that observed in the plateau in
Figure 2 and may represent false positives. For instance we used
the expected likelihood weight (ELW) to test the support for an
alternative position for a branch (and hence a possible OGD
event). In likelihood weighting each sampled tree is weighted by
the likelihood that it accords with the evidence (the alignment).
Trees for which the alignment is unlikely are given less weight.
The sum of the weights calculated for all the trees tested equals 1.
This allows the quantification of the weight of evidence supporting
each of the un-rejected trees and their ranking according to their
support of the evidence in the alignment [31] (see the methods for
more detail). Using an ELW cut-off of 0.55 and minimum path
length distance of 0.3, increased the number of OGDs observed to
80 (Figure 2). However we feel that the parameters at the plateau
at around 65 OGDs represent an optimum, given the shape and
branch-lengths of the combined tree (Figure 3). To further
minimise the possibility of false positives we chose one of the most
conservative of the settings in the plateau to identify putative
HGTs for further analysis: an expected likelihood weight (ELW)
[31] of 0.65 (representing a single un-rejected tree that contributes
65% of the total weighted evidence in support of the alignment
from all the un-rejected trees – see methods for more details) and a
minimum the path length distance across the tree from the best
placement of the branch to its placement on the combined tree of
0.40 substitutions per site, [32] (Figure 2).
Using these settings, 68 orthologous gene displacement events
were identified in the 40 gene families analysed (Figure 3)
consisting of 38 OGDs from the subset of 31 gene families
previously processed using a manual approach [14] (including, all
7 OGDs found in that analysis) and a further 30 OGDs from the
additional 9 gene families analysed here (Table 1).
The rate of orthologous gene displacement should represent
the lower bound of HGT in all gene families as transfers of
informational genes should be already rare and displacement
represents yet another barrier. Therefore, it came as a surprise that
even with the limited species set used here we found up to 50% of
the ribosomal proteins analysed had undergone OGD according
to our stringent method (Table 1). While in the ribosomal genes a
Figure 1. The assessment of possible OGDs on the tree. The
calculation of OGDs involves the assessment of the likelihood of the
placement of every branch of the unrooted tree at every possible
position for each of the gene families separately. 1) The branch leading
to Species A and B is to be assessed and is pruned. It is then replaced at
every possible position on the remaining tree (I to V above) and the
likelihood of each is assessed. 2) When all likelihoods have been
assessed the expected likelihood weight (ELW) of each is calculated. If
any placement except the original position receives an ELW of at least
0.65 (implying high support) this is considered a putative transfer event
(as in IV above). The branch with the best placement is considered the
donor and the recipient is the branch that was assessed at each
position (A and B above). 3) To be considered an OGD the path-length
distance of the original position of the clade to its best placement is
calculated. If this distance is greater than 0.4 (thereby avoiding the
possibility of wrongly identifying phylogenetic uncertainty as an OGD),
the placement is considered a putative OGD.
doi:10.1371/journal.pone.0022099.g001
Constant Rate of HGT
PLoS ONE | www.plosone.org2August 2011 | Volume 6 | Issue 8 | e22099

Page 3

rate of 0.76 OGDs per family in the gene’s life span is observed
across the 191 species examined (Table 1), proteins with fewer
predicted interactions [33] (e.g. tRNA synthethases and GTPases)
have a 4-fold higher rate of 3.4 OGDs per gene family (Figure 4a)
across the same set of 191 species. For all 40 informational genes
from 191 species the average rate is 1.7 OGDs per gene family,
which is similar to recent estimates of a lower bound of 1.1 HGT
per gene family and gene family life span [16].
Factors associated with HGT
Given the considerable amount of OGD discovered, 5 different
possible factors that may promote or inhibit the rate of horizontal
transfer in these universally distributed single-copy genes were
investigated.
1) GC content.
Using the GC content of the 191 extant
species we estimated the GC content of predicted ancestors at each
of the internal branches of the tree in Figure 3. When we
compared the GC content of the donors and acceptors however
we found no correlation between the GC content of the donor and
acceptors of the OGDs detected. Even though having a similar
GC content between the donor and acceptor has been suggested to
influence the successful fixation of HGTs [34] our results suggest
that it doesn’t influence the rate of OGD (Figure S1).
2) Habitat.
Using habitat information from cultivated strains
in culture collections we estimated the habitat specificity (how
specialised the organism’s niche was) of predicted ancestors at the
internal branches of the tree in Figure 1 (see methods). We then
examined the habitat specificity of the donor and acceptor of each
OGD detected. We found that the habitats were not more similar
than would be expected by chance (Figure S2); however this result
may be undermined by the amount of time since the OGD events
occurred. In this timeframe, it is entirely likely that the present-day
descendents of the donors or acceptors have different habitats to
their ancestors.
3) Genome size.
Using the genomes of the extant 191
species, the genome sizes were estimated for the predicted
ancestors at the internal branches of the tree in Figure 3 (see
methods). Comparing the genome size of the species involved in
the OGDs detected we found that the genomes of the acceptors
were significantly bigger than the donors (p=0.02 using wilcoxon
rank sum test with continuity correction), indicating that species
with larger genomes are more likely to undergo OGD (Figure 4b).
4) Biological Function.
For the 40 genes analysed we found
a significant negative correlation between the number of predicted
interactions in which a gene was involved and the number of
OGDs found (Table 1 and Figure S3). Furthermore we found that
ribosomal proteins underwent significantly fewer OGDs than
genes from other functional categories (Figure 4a) supporting
biological function as a barrier to successful displacement (i.e. the
complexity hypothesis [1]). However, on a more fine-grained level
there was no difference within the ribosomal proteins based on
their number of interactions, assembly order, or whether they
bound to the large or small subunit (Figure S4). Only anecdotal
evidence supported any discrimination, that of whether the
proteins bind directly to the RNA core or not (Figure S4).
5) Evolutionary distance.
the last common ancestor (LCA) of the donor and acceptor was
calculated for each of the OGDs detected (see methods). These
distances were compared to distribution of distances between all
possible donor and acceptors given the process of evolution
described in the tree in Figure 3. We found that the distribution of
distances for the OGDs detected were significantly shorter than
would be expected from the shape of the tree (p=0.001 using the
Kolmogorov-Smirnov test) confirming [27] that evolutionary
The evolutionary distance since
Figure 2. Parameter exploration. The surface illustrates the effect on the number of OGDs detected of all combinations of ELW cut-offs from 0.55
to 1.0 (in steps of 0.5) with all combinations of branch length distance cut-offs from 0.1 to 1.0 (in steps of 0.1). Highlighted in white around its edges is
the range of parameters that converge on the same level of HGT detected. The black dot indicates the settings chosen for the purposes of
investigating the underlying factors influencing orthologous gene displacement in these genes.
doi:10.1371/journal.pone.0022099.g002
Constant Rate of HGT
PLoS ONE | www.plosone.org3 August 2011 | Volume 6 | Issue 8 | e22099

Page 4

Constant Rate of HGT
PLoS ONE | www.plosone.org4August 2011 | Volume 6 | Issue 8 | e22099

Page 5

Figure 3. Predicted OGDs. In the main figure the total predicted OGDs for all 40 gene families (as visualised using the iTOL web server [45]) project
onto the tree constructed using the combined phylogenetic information from all 40 gene families. The colours of the arrows represent the statistical
certainty of the transfer, from yellow (0.65) to red (1.00). The Eukaryotes are marked in reds, the Archaea are in greens and the Bacteria in Blues. The
minor figures represent I) the total OGDs predicted for all the Ribosomal proteins (26 gene families) and II) the total OGDs predicted for all the non-
Ribosomal proteins (14 gene families).
doi:10.1371/journal.pone.0022099.g003
Table 1. The number of OGDs predicted for each of the 40 universal single-copy gene families used in this study.
OGDsInteractions Gene family ID Description
0 111COG0048Ribosomal protein S12
0 146COG0052Ribosomal protein S2
0115 COG0080Ribosomal protein L11
0158COG0085 DNA-directed RNA polymerase
0119COG0087Ribosomal protein L3
0111 COG0091 Ribosomal protein L22
0106 COG0092Ribosomal protein S3
0 108 COG0093Ribosomal protein L14
0 121 COG0094Ribosomal protein L5
0119COG0096Ribosomal protein S8
0 110COG0097 Ribosomal protein L6P/L9E
0126COG0100Ribosomal protein S11
0123 COG0184Ribosomal protein S15P/S13E
0 114COG0200Ribosomal protein L15
080 COG0201Preprotein translocase subunit SecY
09 COG0552Signal recognition particle GTPase
1 122COG0088Ribosomal protein L4
1124 COG0098Ribosomal protein S5
1118COG0099Ribosomal protein S13
1114COG0103Ribosomal protein S9
1111 COG0185Ribosomal protein S19
1 116COG0186Ribosomal protein S17
1 122COG0197Ribosomal protein L16/L10E
1 147COG0522 Ribosomal protein S4
2 115COG0081 Ribosomal protein L1
2 111 COG0102Ribosomal protein L13
2 24COG0172Seryl-tRNA synthetase
2 11COG0215 Cysteinyl-tRNA synthetase
2 111COG0256Ribosomal protein L18
2 25 COG0495Leucyl-tRNA synthetase
2 19COG0541 Signal recognition particle GTPase
312 COG0016 Phenylalanyl-tRNA synthetase alpha subunit
3 128COG0049Ribosomal protein S7
3 129COG0090Ribosomal protein L2
3 153 COG0202DNA-directed RNA polymerase
38 COG0533 Metal-dependent protease
4 31COG0525Valyl-tRNA synthetase
6 22COG0012 GTP Binding Protein
10 17COG0018 Arginyl-tRNA synthetase
1142COG0124Histidyl-tRNA synthetase
The gene families in bold are the 9 gene families analysed here in addition to the 31 previously analysed using a manual approach [14]. ‘‘Interactions’’ are the number of
interactions with other gene families predicted for this gene family using the String 7 database [33] with a cut-off of 0.7. A significant negative correlation was found to
exist between the number of Interactions predicted and the number of OGDs detected for each of the genes (see Figure S3). There was also a significant correlation
between the number of OGDs detected and whether the gene coded for a ribosomal protein or not (Figure 4a).
doi:10.1371/journal.pone.0022099.t001
Constant Rate of HGT
PLoS ONE | www.plosone.org5August 2011 | Volume 6 | Issue 8 | e22099