Evolutionary dynamics of co-segregating gene clusters associated with complex diseases.

Abstract

The distribution of human disease-associated mutations is not random across the human genome. Despite the fact that natural selection continually removes disease-associated mutations, an enrichment of these variants can be observed in regions of low recombination. There are a number of mechanisms by which such a clustering could occur, including genetic perturbations or demographic effects within different populations. Recent genome-wide association studies (GWAS) suggest that single nucleotide polymorphisms (SNPs) associated with complex disease traits are not randomly distributed throughout the genome, but tend to cluster in regions of low recombination.
Here we investigated whether deleterious mutations have accumulated in regions of low recombination due to the impact of recent positive selection and genetic hitchhiking. Using publicly available data on common complex diseases and population demography, we observed an enrichment of hitchhiked disease associations in conserved gene clusters subject to selection pressure. Evolutionary analysis revealed that these conserved gene clusters arose by multiple concerted rearrangements events across the vertebrate lineage. We observed distinct clustering of disease-associated SNPs in evolutionary rearranged regions of low recombination and high gene density, which harbor genes involved in immunity, that is, the interleukin cluster on 5q31 or RhoA on 3p21.
Our results suggest that multiple lineage specific rearrangements led to a physical clustering of functionally related and linked genes exhibiting an enrichment of susceptibility loci for complex traits. This implies that besides recent evolutionary adaptations other evolutionary dynamics have played a role in the formation of linked gene clusters associated with complex disease traits.

Full text

Page 1

Evolutionary Dynamics of Co-Segregating Gene Clusters
Associated with Complex Diseases
Christoph Preuss*, Mona Riemenschneider, David Wiedmann, Monika Stoll
Genetic Epidemiology of Vascular Disorders, Leibniz Institute for Arteriosclerosis Research (LIFA) at the University of Muenster, Muenster, Germany
Abstract
Background: The distribution of human disease-associated mutations is not random across the human genome. Despite
the fact that natural selection continually removes disease-associated mutations, an enrichment of these variants can be
observed in regions of low recombination. There are a number of mechanisms by which such a clustering could occur,
including genetic perturbations or demographic effects within different populations. Recent genome-wide association
studies (GWAS) suggest that single nucleotide polymorphisms (SNPs) associated with complex disease traits are not
randomly distributed throughout the genome, but tend to cluster in regions of low recombination.
Principal Findings: Here we investigated whether deleterious mutations have accumulated in regions of low recombination
due to the impact of recent positive selection and genetic hitchhiking. Using publicly available data on common complex
diseases and population demography, we observed an enrichment of hitchhiked disease associations in conserved gene
clusters subject to selection pressure. Evolutionary analysis revealed that these conserved gene clusters arose by multiple
concerted rearrangements events across the vertebrate lineage. We observed distinct clustering of disease-associated SNPs
in evolutionary rearranged regions of low recombination and high gene density, which harbor genes involved in immunity,
that is, the interleukin cluster on 5q31 or RhoA on 3p21.
Conclusions: Our results suggest that multiple lineage specific rearrangements led to a physical clustering of functionally
related and linked genes exhibiting an enrichment of susceptibility loci for complex traits. This implies that besides recent
evolutionary adaptations other evolutionary dynamics have played a role in the formation of linked gene clusters associated
with complex disease traits.
Citation: Preuss C, Riemenschneider M, Wiedmann D, Stoll M (2012) Evolutionary Dynamics of Co-Segregating Gene Clusters Associated with Complex
Diseases. PLoS ONE 7(5): e36205. doi:10.1371/journal.pone.0036205
Editor: Dirk Steinke, Biodiversity Insitute of Ontario - University of Guelph, Canada
Received December 2, 2011; Accepted April 2, 2012; Published May 14, 2012
Copyright: ? 2012 Preuss 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: The work has been funded by the Joint Initative for Research and Innovation from the Leibniz science association (2008/V2). The funders had no role in
study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: Monika Stoll (senior author) has served as an academic editor for PLoS ONE since October 2011. This does not alter the authors’
adherence to all the PLoS ONE policies on sharing data and materials.
* E-mail: christoph.preuss@lifa-muenster.de
Introduction
Genome-wide association studies (GWAS) have provided proof
of principle and revealed numerous disease loci associated with
common complex diseases in the human genome. Recent
investigations have combined these results with data on genetic
variation in human populations, thus linking disease associations
with recent evolutionary events. For example, Soranzo et al. [1]
showed that a haplotype in a region of long-range linkage
disequilibrium (LD), which contains disease loci for coronary
artery disease, hypertension and type I diabetes, recently spread by
a selective sweep specific to Europeans. Furthermore, evidence
from a recent study on deleterious mutations in the human
genome has shown that linked deleterious mutations can spread
through a population by adaptive selection and cluster in regions
of low recombination [2].
Recently, we reported that a region of low recombination on
chromosome 5q31 associated with dilated cardiomyopathy, which
harbors multiple co-segregating genes, is associated with cardio-
vascular disease [3]. Evolutionary analysis revealed that the
disease-associated genes were clustered along the chromosome in
the course of vertebrate evolution as a result of repeated
chromosomal rearrangements in different species. Interestingly,
these genomic rearrangements coincide with the evolution of heart
anatomy in vertebrates, further pointing towards a common
pattern of traces of evolutionary forces acting upon the genome
and susceptibility to common complex diseases. In the present
study we, therefore, investigated whether the clustering of closely
linked genes enriched for deleterious mutations is an evolutionary
constraint determining complex traits e.g. through clustering of
functionally related genes.
While structural and regulatory factors play an important role in
the formation of linked gene clusters, the underlying evolutionary
dynamics are less understood [4–6]. Although recombination
continually breaks down genetic associations between weakly
related loci, tightly linked gene clusters are a common feature of
eukaryotic genomes. For instance, LD in Caucasians [7] contains
on average blocks of approx. 60 kb (1–100 kb), but several regions
in the genome are characterized by long-range LD (.100 kb).
Some well-known examples of such clusters can be found in the
human genome including the major histocompatibility complex
(MHC) genes on chromosome 6 or the cytokine cluster on
PLoS ONE | www.plosone.org1May 2012 | Volume 7 | Issue 5 | e36205

Page 2

chromosome 5q31 with LD extending 500 kb. From an evolu-
tionary point of view, it appears likely that gene order in long-
range LD blocks in the human genome did not arise simply by
chance but as a result of distinct evolutionary dynamics shaping
the genomic architecture. Therefore, we hypothesized that
recurrent chromosomal rearrangements in combination with
adaptive selection may play a crucial role in the formation of
functional gene clusters associated with complex traits.
To test this hypothesis, we performed an analysis of closely
linked gene clusters enriched for disease-associated variants in the
human genome based on HapMap phase III and 1000 Genomes
data [8,9]. Comparative mapping in multiple vertebrate reference
genomes and multiple tests on genetic hitchhiking were used to
analyze the evolutionary history of linked disease genes in regions
of low recombination. We observed that a large proportion of
disease-associated gene clusters in the human genome originates
from repeated concerted rearrangements in early vertebrate
lineages. Remarkably, these gene clusters overlap with haplotype
blocks associated with a range of complex disease phenotypes and
show traces of recent selective sweeps.
Results
Enrichment of Disease-associated Variants in Regions
Showing Low Recombination
In order to test our initial hypothesis that gene clusters in
regions of low recombination display an enrichment of disease
variants, we screened publicly available data on genotype-
phenotype associations throughout the human genome.
Our dataset consisted of a large meta-analysis of Crohn’s disease
[10] and two publicly available catalogs of GWAS results recently
published by Hindorf et al. and Jonson and O’ Donnell [11,12].
For our analysis, we classified 18869 disease-associated SNPs
across the human genome as those with p-values ,1.0610–4(see
also methods section).
We examined genomic regions using a sliding window approach
for the 4638 non-overlapping windows of 500 kb. The windows
were divided into different bins according to the number of
observed disease association per window, estimates of pair-wise
linkage disequilibrium and recombination rates which were
calculated based on HapMap phase III data for all windows
(Materials and Methods).
After performing pairwise non parametric test statistics on our
dataset, we found a significant difference in recombination rates
between windows enriched for more than 15 disease variants
compared to windows harboring only a limited number of disease
variants (1–15)(Mann-Whitney-Wilcoxon
(Figure 1A). However, we did not find evidence that windows
which show an enrichment of disease variants are over-represented
among regions of low recombination. The most pronounced
enrichment of disease variants is observed for the bin with the
smallest recombination rate (0–0.5 (cm/Mb)) as displayed in
Figure 1B when comparing windows harboring more than 15
disease variants to windows containing less disease variants.
Of note, these regions represent a small proportion of the
human genome with the top 2.6% (119) of all tested windows
showing such enrichment compared to 3553 (76.6%) windows
harboring 1 to 15 disease variants. Among the 119 identified
regions exhibiting significant clustering of disease variants are
several candidate regions which have been independently associ-
ated with common complex disease traits, including the MHC
region on chromosome 6 and the IBD5 gene cluster on
chromosome 5 [11,12].
test,P,0.01)
In the next step, we were interested if we could find a
relationship between the clustering of genes among regions of low
recombination and the enrichment of disease variants. Interest-
ingly, while there is no significant correlation in windows with less
than 15 disease variants, we found a significant difference in the
clustering of 3 or more genes in regions harboring more than 15
disease SNPs as displayed in Figure S1. Also, the gene density
differs with an average of 6.77 genes per window in clusters
enriched with disease variants compared to 3.44 genes in the
remaining windows (Wilcoxon Rank Sum test, P,0.01).
Taken together, these observations support the notion of an
enrichment of disease variants in regions of low recombination
with a high gene density. When comparing the disease-associated
regions we observed distinct differences in clusters of disease SNPs
in LD between the different complex traits listed in the GWA
dataset. For traits with a strong autoimmune component among
the 119 regions enriched for disease variants, such as celiac
disease, Crohn’s disease (CD) or childhood asthma, an extensive
clustering of SNPs in blocks of long-range LD (.100 kb) was
observed, as was for traits that are related to recent adaption to
new environments such as skin pigmentation or height. For these
traits between 25% and 90% of all disease related SNPs are
clustered in blocks of long-range LD. In contrast, common
complex diseases including coronary artery disease and bipolar
disorder show a clustering of less than 10% of SNPs in long-range
LD associated with the disease (Table 1). Table 2 provides an
overview of the 32 gene clusters showing the highest enrichment in
the top 2.6% regions of low recombination rates (0–0.5 cm/Mb)
across the human genome. Among these regions are several loci
which have been previously associated with a remarkable disease
pleiotropy and enrichment of disease variants including a long-
range LD block on chromosome 12q24 harboring variants
associated with multiple disease traits [1] and two regions which
have been previously associated with inflammatory bowel disease
on chromosome 3p21 and 5q31 [13,14].
Genetic Hitchhiking in Regions of Low Recombination
and Long-range LD
Several neutral factors, including genetic drift, population size
and demographic effects, can generate stretches of low recombi-
nation around closely linked loci enriched for disease variants
within the human genome. Here, we were primarily interested in
the impact of positive natural selection on the observed
enrichment of disease variants in regions of low recombination.
Therefore, we compared the distribution of signs of positive
selection between the bins holding a different number of disease-
associated SNPs using the integrated haplotype score developed by
Voight et al. [15]. We observed a significant increase in integrated
haplotype score (iHS) signals in the top 2.6% of all windows
harboring more than 15 SNPs compared to windows without
disease variants and windows containing 1 to 15 disease variants
(Mann-Whitney-Wilcoxon test, P,0.05) (Figure 2A).
Furthermore, regions enriched for disease variants also display a
significant increase in the percentage of strong iHS signals (iHS
.2 | iHS ,2) compared to the remaining windows with less
disease variants as displayed in Figure 2B (Mann-Whitney-
Wilcoxon test, P,0.05). Since the iHS statistic has an increased
power to detect signals of selection in regions of low recombina-
tion, we compared the regions holding a different number of
disease associations with matched recombination rates. A small
but significant effect was observed for the bin with the lowest
recombination rates. This effect could not be observed for regions
with higher recombination rates which might be explained by the
Evolution of Gene Clusters and Complex Diseases
PLoS ONE | www.plosone.org2 May 2012 | Volume 7 | Issue 5 | e36205

Page 3

lack of statistical power to identify strong iHS signals in regions of
high recombination (Figure S2).
Deleterious SNPs Show Population Specific Patterns
To find traces of population specific signs of positive selection,
iHS values were retrieved from the Caucasian Europeans/Utah
(CEU), East Asians (ASN) and Yoruba/Ibidan (YRI) datasets [15].
iHS signals were retrieved for all SNPs associated with Crohn’s
disease (CD) at p,1.061024in the meta-analysis [10] (see
methods section). SNPs with |iHS| .2 were defined as ‘‘iHS
signals’’ while SNPs with |iHS| .2.5 were defined as ‘‘strong iHS
signals’’. The genome-wide frequency of iHS signals for all SNPs
resolves around 4% depending on the population and is lower
compared to the disease-associated SNPs in the different
populations resolving around 6%-11% as displayed in Figure 3A.
The most striking deviation in iHS frequency was observed for the
ASN dataset, in which 11.4% of SNPs associated with CD in the
European population correspond with iHS signals exhibiting
genome-wide significance. In the CEU population, 6.9% of SNPs
within these regions fell into this category and 6.2% of the SNPs in
Figure 1. Enrichment of disease variants in regions of low recombination. (A) Boxplots displaying local recombination rates for sliding
windows of 500 kb harboring a different number of disease variants (0, 1–15,.15). (B) Ratio of windows showing an enrichment of disease variants
(.15 disease variants) compared to windows without such a clustering (,15 disease variants) for different bins of local recombination rates.
doi:10.1371/journal.pone.0036205.g001
Table 1. Clustering of selected disease traits in regions of long-range LD among the 119 sliding windows (500 kb) showing an
enrichment of disease variants.
Complex Trait
SNPs in LD blocks
. .100 kb
All GWAS SNPs
across the genome%Number of associated regions
Serum uric acid levels4244 951
Height7078904
Childhood asthma4958 841
Skin pigmentation63 82771
Celiac disease1864 284
CD 3841407 27 36
HDL cholesterol80 454 187
Type II Diabetes Mellitus 3772345 1654
Type I Diabetes1591158 14 26
Coronary Artery Disease 13016489 21
Hypertension 941115819
Bipolar disorder 80 9868 24
Rheumatoid Arthritis 7611906 20
Legend: Enrichment of disease variants in regions of long-range LD for a number of disease traits obtained from the NIH catalog of GWAS results. The number of LD
blocks corresponds to different genomic regions in the genome with pronounced clustering of disease-associated variants.
doi:10.1371/journal.pone.0036205.t001
Evolution of Gene Clusters and Complex Diseases
PLoS ONE | www.plosone.org3 May 2012 | Volume 7 | Issue 5 | e36205

Page 4

the YRI dataset. Figure 3B highlights that when considering only
iHS signals with a stringent threshold of .2.5, the same
distribution of iHS signals can be observed for the distinct
populations. The deviation in iHS signals between the three
populations points towards population specific differences that
account for the differences in disease allele frequencies between
the populations.
In order to test this hypothesis, we used 1000 Genome Pilot 1
data (www.1000genomes.org) and retrieved allele frequencies for
the distinct populations. We found signs of population specific
differences accompanied by genetic hitchhiking in the two regions
exhibiting the most pronounced enrichment of deleterious SNPs.
The first locus resides on chromosome 5q31, a region including
genes such as SLC22A4, SLC22A5, IL3 and IRF. The LD block in
Figure 4A shows a high iHS signal count in the CEU (8.7% of
SNPs) and YRI (7.7% of SNPs) population. Figure 4B displays the
selective sweep and the differences in allele frequencies around the
IBD5 region, which has been recently associated with genetic
hitchhiking in the European population [16]. Among the disease-
associated SNPs, 49% in the CEU dataset and only 4% in the YRI
dataset display signs of recent selection according to iHS signals
(Figure 4C). This reflects differences in the spatial distribution of
these signals, as most of the YRI iHS SNPs are clustered in a
region without CD association near the gene FNIP1, while CEU
signals are located in the vicinity of IL3 and SLC22A4. Regarding
CD SNP allele frequencies, the European population is very
different from the Asian population (56% of SNP with high allele
frequency difference) and the African population (39%, see
Table 2. Sliding windows encountering the most pronounced clustering of disease variants among regions of low recombination
in the human genome.
Chr. Sliding window start Sliding window end
Number of disease
SNPs
Gene
frequencyAssociated traits
1153500001154000001 36 13Alzheimer, Crohn’s Disease, Hippocampal atrophy
26100000161500001 277Crohn’s Disease, Psoriasis, Rheumatoid arthritis, Ulcerative
colitis
2123500001124000001 170Blood Lipids, Type II Diabetes Mellitus
2198000001198500001 678Crohn’s Disease
3 485000014900000124 13Crohn’s Disease
3 49000001 4950000145 20Crohn’s Disease
3495000015000000194 14Crohn’s Disease
350000001505000013118Crohn’s Disease
5 4050000141000001717Ankylosing spondylitis, Crohn’s Disease
5 99500001100000001 211Rheumatoid Arthritis
5129500001130000001230Crohn’s Disease, Ischemic stroke
5 130000001130500001560Crohn’s Disease
5130500001131000001484 Crohn’s Disease
5131000001 131500001516 Crohn’s Disease
5131500001132000001 133 10Crohn’s Disease
62600000126500001 30 37Alzheimer’s disease, Bipolar disorder, Height
62650000127000001 219 Alzheimer’s disease, Bipolar disorder, Height,
62700000127500001187Schizophrenia
627500001280000012317Type I Diabetes
628000001 28500001 5813Rheumatoid Arthritis, Type I Diabetes
628500001 29000001 316 Amyotrophic Lateral Sclerosis, Crohn’s disease
6 127000001127500001261Crohn’s Disease, Height
7 3300000133500001 200 Crohn’s Disease
103500000135500001 344 Crohn’s Disease
103550000136000001 184 Crohn’s Disease
12110500001111000001 278Celiac disease, Rheumatoid Arthritis, Type I Diabetes
173500000135500001 5319Childhood asthma, Crohn’s Disease
173750000138000001 2416 Crohn’s Disease, Primary biliary cirrhosis
174150000142000001 231 Crohn’s Disease
194200000142500001187 Blood Lipids, Crohn’s Disease, Schizophrenia
203300000133500001 5210Height, Osteoarthritis
222850000129000001 316 Crohn’s Disease
Legend: The 32 windows displaying the highest enrichment of disease variants (.15) among regions of low recombination rates (0–0.5 cM/Mb) across the human
genome.
doi:10.1371/journal.pone.0036205.t002
Evolution of Gene Clusters and Complex Diseases
PLoS ONE | www.plosone.org4 May 2012 | Volume 7 | Issue 5 | e36205

Page 5

Figure 4C). Consistently, the index SNP risk allele rs12521668-T is
very common in Europeans (0.48), while rare in the other
populations (ASN: 0.02, YRI: 0.03). The allele rs1050152-T,
which is a putative causal mutation in the SLC22A4 gene [17],
shows similar allele frequencies (CEU: 0.39, ASN: 0.02, YRI:
0.03). This suggests that selection pressure acted on the European
population and favored CD risk alleles. In the Yoruba population,
the different environment was accompanied by different selection
pressures, which shaped the neighboring FNIP1 gene. A possible
cause of selective sweeps in the 5q31 region might be bacteria, as a
recent study has linked this locus to Mycobacterium tuberculosis
susceptibility [18]. As a study by Huff et al. pointed out that
genetic hitchhiking might have also played a role in increasing CD
risk by driving alleles of IRF1 to high frequency while selection
pressures were acting on SLC22A4 due to changes in nutrition
[16].
The second population specific risk locus is located on
chromosome 3p21 and includes genes such as GPX1, MST1 and
BSN (Figure 5A) [19]. In contrast to the region on chromosome
5q31, the enrichment of disease variants is not accompanied by a
selective sweep within the European population. For the GPX1
gene, a recent selective sweep in the Asian population has already
been established [20] and could be reproduced in this study.
Within LD of the CD SNPs associated in the European
population, a large number of strong iHS signals could be
observed in the ASN dataset (72% of SNPs) and a more moderate,
but still elevated number in the YRI data (11%, see Figure 5B).
Consistently, the allele frequencies show large differences. In the
ASN population, most CD SNPs feature extreme reference allele
frequencies, which are lower than 0.2 or greater than 0.8
(Figure 5C). These differences in allele frequencies due to recent
selection events might have had a profound impact on disease
prevalence between populations. While there is a strong associ-
ation with CD in Europeans, an association signal in individuals of
Asian ancestry could not be replicated. The substantial variation in
the frequency of disease variants due to recent selective events
across human populations may point to differences in disease
prevalence between the populations. In the European and African
populations, allele frequencies resolve around an intermediate
range. This suggests that the strong wide-range selective sweep in
the ASN population also had effects on CD SNPs (Figure 5D) and
might in fact have lowered the disease risk originating from 3p21.
The index SNP rs3197999 was shown to be a non-synonymous
coding SNP in the MST1 gene [21]. Its risk allele A is more
common in EUR (0.28) and YRI (0.24) and less frequent in ASN
(0.08). Thus, assuming that the risk conferred by disease variants is
constant across populations, our data suggest that the common-
disease-common-variant hypothesis does not necessarily extend
across populations since risk alleles discovered in the European
population are found at extremely low or high frequencies in other
populations.
Evolution of Disease-associated Gene Clusters in the
Vertebrate Lineage
As outlined before, the most pronounced clustering of disease-
associated SNPs is found within larger gene clusters where more
than one gene might bare the causal disease association. Among
these gene clusters are various regions which have been
associated with disease pleiotropy including genes on chromo-
some 3p21 and 5q31 and 12q21, which have been associated
with celiac disease, type 1 diabetes, coronary artery disease
(12q24) and osteoarthritis (20q11) [1,13,14]. Inspired by our
previousstudy onthe co-segregation
associated genes during vertebrate evolution [3], we were
interested in the evolutionary dynamics which shaped these
gene clusters. Similar to the approach taken previously, we
performed comparative genome mapping across six vertebrate
genomesinordertotrace
vertebrateevolution. Basedon the
approach we determined shared orthologs between the different
ofcardiomyopathy-
theiremergence
comparative
throughout
mapping
Figure 2. Clustering of iHS signals in regions enriched with disease variants. Boxplots highlighting (A) the distribution of mean iHS signals
in regions enriched with disease variants (.15) compared to regions with a moderate number of disease associations (1–15) and (B) the ratio of
strong iHS signals |iHS .2| for these regions.
doi:10.1371/journal.pone.0036205.g002
Evolution of Gene Clusters and Complex Diseases
PLoS ONE | www.plosone.org5May 2012 | Volume 7 | Issue 5 | e36205