Abstract
Genome-wide scans for positive selection have become important for
genomic medicine, and many studies aim to find genomic regions affected
by positive selection that are associated with risk allele variations
among populations. Most such studies are designed to detect recent
positive selection. However, we hypothesize that ancient positive
selection is also important for adaptation to pathogens, and has
affected current immune-mediated common diseases. Based on this
hypothesis, we developed a novel linkage disequilibrium-based pipeline,
which aims to detect regions associated with ancient positive selection
across populations from single nucleotide polymorphism (SNP) data. By
applying this pipeline to the genotypes in the International HapMap
project database, we show that genes in the detected regions are
enriched in pathways related to the immune system and infectious
diseases. The detected regions also contain SNPs reported to be
associated with cancers and metabolic diseases, obesity-related traits,
type 2 diabetes, and allergic sensitization. These SNPs were further
mapped to biological pathways to determine the associations between
phenotypes and molecular functions. Assessments of candidate regions to
identify functions associated with variations in incidence rates of
these diseases are needed in the future.
Introduction
Genome-wide scans of positive selection are a recent advance in genomic
medicine, and have become an important way to infer risk allele
variations across populations and elucidate genetic mechanisms of human
evolutionary adaptation to local environments, dietary patterns, and
infectious diseases [[32]1]. Because detection of positive selection
will help improve population-specific disease prevention strategies and
treatments, many previous studies revealed that risk alleles for common
complex diseases show substantial variation across human populations
and contribute to disease risk variation among populations
[[33]2–[34]8]. For example, risk alleles for type 2 diabetes (T2D) show
high frequencies in African populations and low frequencies in Asian
populations [[35]8]. The patterns of risk allele frequencies are shown
to be consistent with the disparity in T2D risk across populations of
different ancestries, which is thought to be due to adaptations to
different agricultural developments across continents. If we know
populations have a higher T2D risk (e.g., African ancestry), we can
take population-specific preventive actions for T2D based on the
genetic background of individuals. Another well-known example is
cytochrome P450 (CYP) genes [[36]9]. The allele of an SNP in CYP3A5, a
member of the CYP3A subfamily, shows large frequency differences
between African Americans and non-Africans [[37]9–[38]11]; and the
region that contains this gene also shows a high degree of linkage
disequilibrium (LD) that was affected by positive selection in
Europeans [[39]9, [40]12]. Because this allele is involved in CYP3A5
expression and metabolism of clinically important drugs (e.g., the
immunosuppressant tacrolimus [[41]13] and the HIV protease inhibitor
saquinavir [[42]14]), differences in genetic background may be
associated with differential drug responses among populations
[[43]9–[44]11]. Other common complex diseases with risk allele
frequencies that differ across human populations include cancers (e.g.,
breast cancer and prostate cancer), cardiovascular diseases, metabolic
diseases (e.g., hypertension), neurodegenerative diseases (e.g.,
Alzheimer’s disease), and systemic autoimmune diseases (e.g., systemic
lupus erythematosus and rheumatoid arthritis) [[45]3, [46]15].
Whereas most studies have focused on recent positive selection, ancient
human adaptation to pathogens is known to have affected the immune
system and is also associated with risk allele frequency variation for
common diseases, such as autoimmune and metabolic disorders among
populations [[47]16]. It was reported that ancient local adaptation to
pathogens affected celiac disease, type I diabetes, and multiple
sclerosis susceptibility loci [[48]17]. It was also reported that
ancient selection in response to a sleeping sickness pathogen in Africa
contributed to the high rate of renal disease in African Americans
[[49]18]. Another example is adaptation to malaria pathogens,
Plasmodium spp., which appeared more than 100,000 years ago (100 kya)
in Africa. Most malaria resistance alleles occur in African
populations, and the LD segments associated with the alleles are short
and highly variable between populations [[50]16]; however, whether
variation among populations affects the incidence of recent common
diseases has not been well documented [[51]19]. Therefore, in addition
to recent positive selection, ancient positive selection is important
for detecting immune-mediated common diseases.
Approaches to finding positively selected regions in the human genome
are classified into four groups [[52]20]: summary statistics, LD-based
statistics [[53]21–[54]26], comparative genomics, and neutrality tests.
These approaches are mainly applied to detect recent positive
selection. For example, positive selection signals of the lactase
persistence allele at the LCT locus were detected by long haplotype
tests (i.e., LD-based approaches such as LRH, iHS, and XP-EHH) [[55]27,
[56]28]. XP-EHH [[57]28] also detected positive selection of SLC24A5
that is associated with skin pigment differences among populations.
Significant variations in T2D risk alleles across populations have been
revealed using iHS and XP-EHH [[58]8, [59]29, [60]30]. These methods
aim to identify positive selection that occurred after dispersal out of
Africa (< 30 kya) [[61]27, [62]28], and the mean lengths of detected
regions are more than 400 kb. Recently, selection events have been
detected in the ancestral population of all present-day humans
[[63]31–[64]33], and 3P-CLP [[65]34] was developed to detect ancient
selection events that occurred before the split of Yoruba and Eurasians
but after their split from Neanderthals.
In this study, we develop a pipeline to detect ancient positive
selection events. We use the term ‘ancient’ to describe the period
before the human migrations out of Africa (~100 kya). We hypothesize
that haplotype blocks, i.e., conserved regions, that contain variants
that were selected in ancient times have spread with human migration,
and some mutations occurred for adaptation to each local environment
([66]Fig 1). This pipeline first identifies ancient haplotype blocks by
screening common blocks after extracting those within each population.
The pipeline then scans the identified ancient haplotype blocks to
check whether they have haplotype frequency variation among
populations.
Fig 1. Signatures of ancient haplotype blocks with population-specific
positive selection.
[67]Fig 1
[68]Open in a new tab
(A) Some important loci adapted to ancient African environment arose
(red triangle) and formed haplotype blocks. The haplotype blocks spread
during human migration, and some mutations may have occurred for
adaptation to each environment (blue and green triangles). This change
is a signature of an ancient haplotype block with population-specific
positive selection. (B) A proposed network model to represent the
positive selection signature. Each node represents the population in a
region. Throughout this paper, red, blue, and green nodes represent
populations in Africa, Europe, and Asia, respectively. Arrows represent
migration routes. Edges represent relationships between populations. In
this work, relationships were evaluated using t-statistic scores that
represent degrees of difference between populations. Asterisks
represent mutations.
After extracting ancient haplotype blocks with haplotype frequency
variation across populations by applying the pipeline to HapMap2
genotype data [[69]35], we annotated the genes in the extracted blocks
using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway
database [[70]36], and identified genes associated with immune
system-related functions that are potentially related to common
diseases. We also analyzed SNPs in the blocks using the NHGRI GWAS
catalog [[71]37] to infer the relationships among SNPs, diseases, and
genes whose biological functions are described by functional categories
in the KEGG pathway database.
Materials and methods
HapMap data for genome-wide scan
We downloaded unphased diplotype data sets of 22 autosomal chromosomes
from release 24 of the HapMap database [[72]35]. The data sets
consisted of unphased diplotypes of 270 individuals: 90 Yoruba from
Ibadan, Nigeria (YRI); 90 Utah residents with ancestry from northern
and western Europe (CEU, from the CEPH diversity panel); and 90
Japanese from Tokyo and Japan, and Han Chinese from Beijing, China
(ASN). All markers in the data set were diallelic. We selected
3,619,226 SNPs that were common to the three populations ([73]Fig 2);
among these, 879,657 SNPs had no missing data. The genotypes of these
879,657 SNPs were used to identify ancient haplotype blocks that were
present in African populations and spread with migrating populations.
Fig 2. HapMap SNPs from three populations.
[74]Fig 2
[75]Open in a new tab
The relationships between the numbers of SNPs in 22 autosomal
chromosomes from three populations, YRI, CEU, and ASN, in the HapMap
database are shown. A total of 3,619,226 SNPs were found in all three
populations. Among them, 879,657 SNPs were selected under the condition
that all of the SNPs could be attributed to the genotypes of all 270
individuals.
The Entrez SNP search tool ([76]https://www.ncbi.nlm.nih.gov/snp) was
used to retrieve nonsynonymous SNPs (nsSNPs) from dbSNP build 132. We
downloaded all three kinds of nsSNPs: 173,911 missense, 6,838 nonsense,
and 24,296 frame-shift SNPs, among which 4,316 nsSNPs were included in
the HapMap data sets. CCDS [[77]38] build 36.3 was further used to
evaluate the location of each SNP in terms of protein-coding genes. In
total, 3,298 nsSNPs were mapped to 2,467 genes across the 22 autosomal
chromosomes.
KEGG for functional annotation
KEGG is a suite of databases that includes molecular interaction
networks (PATHWAY database) and information about genes and proteins
(GENES/SSDB/KO databases), and biochemical compounds and reactions
(COMPOUND/GLYCAN/REACTION databases) [[78]36]. We used KEGG PATHWAY,
which includes 430 reference pathway maps (downloaded on 25 February
2015), among which 74 are of human diseases. The human disease maps
contain 12 cancer maps.
KEGG mapper is a web-based interface that accepts gene lists as input,
and outputs lists of KEGG pathway maps that contain the genes in the
input list. We used KEGG mapper to identify the functions of the genes
obtained by our scans. We also used KEGG pathway maps for a Monte Carlo
test that showed to which pathway maps the genes were likely to belong.
Inter-diplotype distance
In our previous work [[79]39], we defined an inter-diplotype distance
called Haplotype Inference Technique (HIT) Hidden Markov Model-based
Distance (HHD). Unlike the allele sharing distance (ASD) [[80]40], HHD
reflects the founder (or ancestral) haplotypes well. HHD assumes
multiple founder haplotypes [[81]39] and calculates the distance
between founder and present-day haplotypes. The distances between
founder and present-day haplotypes were used to calculate the distance
between individual SNP genotypes. If we hypothesize the existence of
common founder haplotypes in several populations, HHD performs better
than ASD. When specific haplotypes are conserved in populations, both
HHD and ASD produce small values, but when they are not conserved, HHD
produces much larger values than ASD. Thus, for blocks that have both
common founder and population-specific haplotypes, it is highly
possible that the inter-population HHD would be larger than ASD.
Therefore, we implemented a pipeline that utilizes HHD ([82]Fig 3).
Fig 3. Pipeline for ancient haplotype block scan and functional annotation.
[83]Fig 3
[84]Open in a new tab
(A) Novel procedure for ancient haplotype block scan using HHDs. (B)
Functional annotation procedure based on biological pathways. Each box
shows materials or tools used in that step.
Briefly, the difference between HHD and ASD in terms of their
algorithms is as follows. The algorithm for ASD between genotypes first
counts allele differences at each SNP site; then, the total allele
differences are normalized. The HHD algorithm first infers candidate
haplotypes and their frequencies in populations for each genotype.
Second, it calculates distances between candidate haplotypes of two
genotypes. The distances between candidate haplotypes are weighted by
their frequencies in the populations. Finally, for HHD, the distances
between candidate haplotypes are added and normalized. Unlike ASD, HHD
identifies differences between common founder and present-day
haplotypes. When haplotype composition of two populations are similar,
HHD between the genotypes is small like ASD. If two populations have
different haplotype composition, HHD calculates the distance between
genotypes more accurately and becomes larger than ASD. If of the
difference between average HHD values between two populations is large,
we infer that the region has haplotype variation and it is possible
that there are population-specific haplotypes.
Genome-wide scan of ancient haplotype blocks ([85]Fig 3A)
1. Identification of ancient haplotype blocks
We assumed that functionally important conserved regions in African
populations spread with other populations during human migration.
Currently, such conserved regions differ by population but may have
shared regions [[86]41]. We defined the shared regions as ancient
haplotype blocks.
We first identified haplotype blocks for each population with Haploview
4.2 [[87]42]. Haploview estimates Hedrick’s multiallelic D′ [[88]43,
[89]44] between a pair of SNPs, and 95% confidence bounds on D′ are
used to evaluate the strength of LD between the SNP pair. The default
setting of Haploview ignores pair-wise comparisons of SNPs further than
500 kb apart.
Next, we extracted the haplotype blocks of the YRI population that
overlapped with the haplotype blocks of both the CEU and ASN
populations. For a haplotype, let H[i..j] denote the haplotype, where
positions of the first and last SNPs are i (bp) and j (bp) in the
genome. Two haplotypes, H1[i..j] and H2[k..l], are thought to overlap
with each other in any of the following: i ≤ k ≤ j ≤ l, k ≤ i ≤ l ≤ j,
i ≤ k ≤ l ≤ j or k ≤ i ≤ j ≤ l. We considered the extracted haplotype
blocks of the YRI population as ancient positive selection candidates
that spread with population migration.
To identify the shared regions of the haplotype blocks, we detected
common haplotype blocks. Here, the common haplotype blocks were defined
as the haplotype blocks obtained from genotype data of all three
populations. To evaluate whether the identified common haplotype blocks
were affected by ancient positive selection and really exist for each
population, we further searched the common haplotype blocks that
overlapped with the previously extracted candidates to identify ancient
positive selection events. We defined the extracted final set of
haplotype blocks as ancient haplotype blocks.
[90]Fig 4 shows an example of ancient haplotype blocks that were
identified from the 879,657 genotypes. The 14-kb haplotype block was
identified in 270 individuals, already existed in the YRI population,
and overlapped with the haplotype blocks of the CEU and ASN
populations. Although recent studies analyzed population-specific
features of LD distribution [[91]45], we identified haplotype blocks
common to all of the populations for ancient haplotype block regions.
Fig 4. Example of ancient haplotype blocks identified in this work.
[92]Fig 4
[93]Open in a new tab
Four haplotype blocks identified in all three populations (YRI, CEU,
and ASN) are shown. The region of overlap between the dashed lines is
defined as the ancient haplotype block.
2. Calculation of inter-population distances for ancient haplotype blocks
For the k-th ancient haplotype block, we calculated HHD between two
individuals i and j, d[ijk] (1 ≤ i < j ≤ 270), across all three
populations and constructed a 270 × 270 HHD matrix for each ancient
haplotype block ([94]S1 Text, [95]S1 Fig). To identify ancient
haplotype blocks that differed between populations (i.e., ancient
haplotype blocks with common founder haplotypes and population-specific
haplotypes), we used a t-statistic score based on inter-population
distance X[k] and intra-population distance Y[k] for each haplotype
block k:
[MATH:
tk=Xk¯−Yk¯s<
/mi>XYk(1m+1n), :MATH]
(1)
where
[MATH:
sXYk=<
mo>(m−1)sXk
+(n−1)sYk
m+n−2,
:MATH]
m is the total number of inter-population pairs of individuals that
belong to different populations, and n is the total number of
intra-population pairs of individuals that belong to the same
population ([96]S1 Fig).
[MATH: Xk¯ :MATH]
and
[MATH: Yk¯ :MATH]
are the sample means of the inter- and intra-population distances, and
[MATH:
sX<
mi>k :MATH]
and
[MATH:
sY<
mi>k :MATH]
are the unbiased variances of the inter- and intra-population
distances. This score measures the difference between the mean HHD
value for pairs of people that belong to different populations
(inter-population distance) and pairs of people that belong to the same
population (intra-population distance); if the score is high, the
haplotype block is considered to represent a difference between
populations. We ranked the ancient haplotype blocks with this score for
the three populations. We considered that blocks in the upper tail of
the score distribution (i.e., top 1% of blocks) were likely to have
common founder and population-specific haplotypes that were created by
ancient positive selection and population-specific mutations. In the
present work, top 1% of blocks were considered to show population
differentiations and further validated by the following steps (see
“Relationship between the top 1% of blocks and Fst” for additional
detail).
3. Ancient haplotype block characterization
We used networks that represented differences between the three
populations evaluated using t-statistic scores ([97]Fig 1B) to classify
the ancient haplotype blocks. Each node of the network represented a
population (i.e., YRI, CEU or ASN), and the weight of each edge
represented the sample mean of t-statistic scores between the two
populations. k-means clustering was applied to all the ancient
haplotype blocks based on the weights of the three edges, CEU–YRI,
CEU–ASN, and ASN–YRI.
Functional annotation of candidate regions ([98]Fig 3B)
1. Monte Carlo test for enrichment analysis
We performed KEGG pathway enrichment analysis using the genes in the
detected ancient haplotype blocks, and evaluated the result by Monte
Carlo test using the genes obtained from 10,000 random samples of 310
ancient haplotype blocks (1% of all ancient haplotype blocks). The
Jaccard index was used as a measure of the overlap between all genes in
a KEGG pathway and the genes in the ancient haplotype blocks. For each
pathway, p-values were calculated based on the distribution of the
Jaccard index of random samples.
2. Annotation of genes and SNPs by pathway mapping and GWAS catalog
We mapped genes in the detected regions to biological pathways in the
KEGG database. We also investigated known phenotypes associated with
SNPs in the regions using the NHGRI GWAS catalog [[99]37], which
collects relationships between SNPs and human phenotypes. The SNPs that
have known phenotypes were then mapped to biological pathways through
reported genes. KEGG Mapper was used to identify associated biological
pathways and their functional categories.
Results
Identification of ancient haplotype blocks
In the 22 autosomal chromosomes, Haploview [[100]42] identified 62,123,
56,597, and 56,325 haplotype blocks in the YRI, CEU, and ASN
populations, respectively. We also identified 76,119 haplotype blocks
in all three populations, 39,228 of which were defined as ancient
haplotype blocks. Of these, we used 30,966 ancient haplotype blocks
that consisted of more than two SNPs. The maximum, minimum, and average
lengths of the identified ancient haplotype blocks were 499,794, 42,
and 24,584.36 bp, respectively. The average length of 24,584.36 bp is
much shorter than that of the regions identified by studies based on
previous LD-based methods, such as the long-range haplotype test
[[101]27, [102]28], which focuses on recent positive selection
([103]Table 1). The number of SNPs and genes in the blocks varied from
3 to 97 and 0 to 6, respectively. The total number of SNPs and genes in
the identified ancient haplotype blocks were 240,752 and 5,577,
respectively.
Table 1. Average length of regions identified by representative methods.
Method Average lengths (bp)
LRH, iHS [[104]21] 310,049.59
LRH, iHS, XP-EHH [[105]22] 151,579.03
EHHS [[106]23] 336,811.55
CMS [[107]24] 86,178.84
XP-CLR [[108]25] 1,280,084.33
HaploPS [[109]26] 449,043.75
Ancient haplotype blocks by the present study 24,584.36
Top 1% t-score of the ancient haplotype blocks 35,803.89
[110]Open in a new tab
Inter-population distances
To find haplotype blocks that represent differences among the three
populations, we calculated the t-statistic score, t[k], which was
defined in Eq ([111]1), for each ancient haplotype block. [112]Fig 5
shows the distribution of the calculated scores. The distribution can
be fitted to the generalized extreme value (GEV) distribution. Larger
scores represent greater disparity between inter-population and
intra-population distances. In the top 5% of sorted haplotype blocks,
there was a set of 1,548 haplotype blocks that includes 592 genes and
13,955 SNPs. When we examined the top 1% of sorted haplotype blocks, we
identified a set of 310 haplotype blocks. The 310 haplotype blocks
included 130 genes ([113]S1 Table, [114]S2 Table) and 2,803 SNPs. The
average length of the 310 ancient haplotype blocks was 35,803.89 bp
([115]Table 1). Additionally, 35% and 49% of the SNPs had Fst [[116]2]
values larger than 0.2 in the top 5% and 1% of blocks, respectively.
The average Fst values for the SNPs in the top 5% and 1% of blocks are
0.162 and 0.187, which are significantly different based on the
two-tailed Welch’s t-test (p-value < 0.05). (see “Relationship between
the top 1% of blocks and Fst” for additional detail).
Fig 5. Distribution of calculated scores.
[117]Fig 5
[118]Open in a new tab
The x-axis shows the t-statistic score, and the y-axis shows the number
of ancient haplotype blocks.
Characterization of ancient haplotype blocks
We classified all ancient haplotype blocks into eight clusters (i.e., k
= 8 for k-means clustering) based on the network of populations and
their t-statistic score profiles ([119]Fig 6, [120]S3 Table). We used k
= 8, because the network with three edges can be classified into eight
patterns if we classify each edge as either long or short. Using this
setting, we could not find Cluster 8 that corresponds to a network with
all three edges long. Instead, Cluster 5′, which was similar to Cluster
5, was obtained. However, the degrees of the differences for the YRI
population pairs were much smaller for Cluster 5′. The largest portion
(~30%) of the ancient haplotype blocks was classified in Cluster 1
([121]Table 2). Clusters 2, 3, 4, and 5 had almost the same number of
cluster members. Clusters 6 and 7 had almost twice as many cluster
members as Clusters 2, 3, 4 and 5.
Fig 6. Classification of ancient haplotype blocks.
[122]Fig 6
[123]Open in a new tab
Eight clusters of ancient haplotype blocks obtained by clustering based
on the network of populations and their t-statistic score profiles. The
number on each edge represents the average t-statistic score; smaller
scores reflect shorter edges.
Table 2. Summary of screening results.
Cluster 1 2 3 4 5 6 7 5’ Total
Top 1% 0 76 39 35 160 0 0 0 310
(0%) (24.52%) (12.58%) (11.29%) (51.61%) (0%) (0%) (0%)
Total 9,459 1,772 1,657 2,121 2,094 3,682 4,237 5,944 30,966
(30.55%) (5.72%) (5.35%) (6.85%) (6.76%) (11.89%) (13.68%) (19.20%)
[124]Open in a new tab
Each element in the table shows the number of obtained haplotype
blocks. The numbers in parentheses are percentages of the total pool of
haplotype blocks.
Association between clustering results and t-statistic score
Based on the score distribution for each cluster shown in [125]Fig 5,
the clusters can be classified into three groups: group I, which
consists of Cluster 1; group II, which consists of Clusters 2, 3, 4,
and 5; and group III, which consists of Clusters 6, 7, and 5′ ([126]Fig
7). The largest portion of the ancient haplotype blocks was classified
in group I, with scores below 18, and showed no large differences
across the three populations. The scores of groups III and II ranged
from 11 to 39 and 23 to 86, respectively.
Fig 7. Score distributions for each cluster.
[127]Fig 7
[128]Open in a new tab
The score distribution of ancient haplotype blocks is shown for each
cluster. The clusters can be classified into three groups: I, II, and
III. Group I consists of Cluster 1 (blue). Group II consists of
Clusters 2, 3, 4, and 5 (red). Group III consists of Clusters 6, 7, and
5′ (green).
The top 1% of the sorted ancient haplotype blocks contained
significantly higher proportions of Clusters 2 and 5 than the total
pool of ancient haplotype blocks (p-value < 0.05) ([129]Table 2). This
result for Cluster 5 is consistent with the previous results, which
indicates that the genetic distance between the African population and
the other populations is large [[130]46, [131]47]. Our results also
showed that twice as many members of Cluster 2 are in the top 1% that
of Cluster 4.
Functional annotation of blocks in the top 1% of t-statistic scores
The Monte Carlo test for enrichment of genes in the top 1% of ancient
haplotype blocks (310 haplotype blocks) showed that the 130 genes were
enriched for 22 pathways categorized in “Metabolism,” “Genetic
Information Processing,” “Cellular Processes,” “Organismal Systems,”
and “Human Diseases” ([132]Table 3). In the “Human Diseases” pathways,
we found several diseases already known to have some differences
between populations: hepatitis C, non-alcoholic fatty liver disease
(NAFLD), and some cancers.
Table 3. Pathways for which the genes in the top 1% of ancient haplotype
blocks are enriched.
Category Pathway
Genes[133]^* p-value
Cluster 2 Cluster 3 Cluster 4 Cluster 5
Organismal Systems
T cell receptor signaling pathway
GSK3B
IL10,
PAK7
0.029
Immune system
Nervous system Neurotrophin signaling pathway GSK3B, SH2B3 BRAF,
RPS6KA2 0.007
Endocrine system Progesterone-mediated oocyte maturation BRAF,
GNAI1,
MAD1L1
RPS6KA2 0.005
Metabolism
beta-Alanine metabolism
GADL1
ACADM
0.016
Metabolism of other amino acids
Genetic Information
Processing
Ribosome biogenesis in eukaryotes
EFTUD1,
RBM28
0.039
Translation
Environmental Information
Processing
Neuroactive ligand receptor interaction
GLP2R,
ADRA1A,
CHRNB4,
PARD3
GRID2
GRIK1
GRIK2,
0.012
Signaling molecules and
interaction
Signal transduction Hippo signaling pathway GSK3B APC,
DLG2,
PARD3 0.048
Cellular Processes
Focal adhesion
GSK3B,
LAMA3
MYLK
ACTN1,
BRAF,
PAK7
0.018
Cellular community
Signaling pathways regulating pluripotency of stem cells GSK3B, APC,
JAK1 0.019
Tight junction ACTN1,
JAM2, GNAI1,
PARD3,
PRKCH 0.038
Cell motility Regulation of actin cytoskeleton MYLK, ACTN1, APC,
BRAF,
PAK7,
PIP5K1B
SSH2 0.001
Human Diseases
Toxoplasmosis
LAMA3
IL10,
GNAI1,
JAK1,
0.003
Infectious diseases
Hepatitis C GSK3B, BRAF,
JAK1 0.022
Pertussis IL10 GNAI1, 0.023
Leishmaniasis IL10, JAK1 0.025
Cancers Colorectal cancer GSK3B APC,
BRAF,
DCC, 0.001
Renal cell carcinoma ARNT2, BRAF,
PAK7 0.008
Endometrial cancer GSK3B APC,
BRAF, 0.018
Basal cell carcinoma GSK3B APC, 0.023
Viral carcinogenesis ACTN1, JAK1,
MAD1L1 0.046
Endocrine and metabolic diseases Non-alcoholic fatty liver disease
(NAFLD) GSK3B
NDUFS6 NDUFA8 0.016
Neurodegenerative diseases Parkinson's disease NDUFS6 GNAI1,
NDUFA8, 0.013
[134]Open in a new tab
* Enriched genes in each cluster.
Hepatitis C varies (HCV) in incidence rate and treatment response
across populations [[135]48]. The chronic HCV infection rate is higher
in African Americans than in people of European ancestry in the United
States. It has also been reported that histologic progression of HCV
infection is less rapid among African American patients than among
those of European ancestry. Rates of adverse events are higher among
patients of European ancestry. The rate of sustained virologic response
in African Americans is significantly lower than for patients of
European ancestry. In our results, BRAF (Cluster 5), GSK3B (Cluster 2),
and JAK1 (Cluster 5) were mapped to “Hepatitis C.” BRAF and JAK1 have
not previously been found to be affected by positive selection, but
GSK3B was reported to be affected by positive selection in people of
Mexican ancestry in Los Angeles, California, USA [[136]26].
Differences in HCV-specific CD4 T cell responses between African
Americans and people of European ancestry have been previously
discussed, and may explain some of these differences across populations
[[137]48]. Previous haplotype analyses have also suggested that
variants of the immunomodulatory IL10 and IL19/20 genes play a role in
the spontaneous clearance of HCV in African American patients but not
in patients of European ancestry [[138]49]. The “T cell receptor
signaling pathway” appeared in our results, and IL10 (Cluster 3) GSK3B
(Cluster 2) and PAK7 (Cluster 5) were mapped to this pathway.
NAFLD, an endocrine and metabolic disease, has been suggested to have
pathophysiological differences among populations [[139]50]. Latinos
(45%) show the highest prevalence of hepatic steatosis and African
Americans show the lowest prevalence; people of European ancestry
showed an intermediate prevalence of 33% [[140]50]. There might be
differences in metabolic responses related to NAFLD in different
populations. NDUFA8 (Cluster 5), NDUFS6, and GSK3B (Cluster 2) were
mapped to “Non-alcoholic fatty liver disease (NAFLD)”. NDUFA8 has been
reported to be affected by positive selection in European populations
[[141]23], but NDUFS6 has not previously been found to be affected by
positive selection.
Regarding cancers, higher renal cell carcinoma incidence rates have
been identified in men of African ancestry [[142]51]. Endometrial
cancer is reported to have higher incidence rates in women of European
ancestry than in any other population [[143]52, [144]53]. Basal cell
carcinoma is known to be common in fair-skinned individuals [[145]54].
ARNT2 (Cluster 5), BRAF (Cluster 5), and PAK7 (Cluster 5) were mapped
to “Renal cell carcinoma;” APC (Cluster 5), BRAF (Cluster 5), and GSK3B
(Cluster 2) were mapped to “Endometrial cancer;” and APC (Cluster 5)
and GSK3B (Cluster 2) were mapped to “Basal cell carcinoma” in our
results. APC has been reported to be a positive selection candidate in
European and Asian populations [[146]24, [147]26], and the others have
not previously been reported to be affected by positive selection.
Functional annotation of genes and SNPs in each cluster
To check the functional annotation details of the top 1% of regions,
which included only members of Clusters 2, 3, 4, and 5, as previously
discussed, we mapped the genes and SNPs in each cluster to pathways and
the GWAS catalog, respectively.
Cluster 2
The 76 ancient haplotype blocks in Cluster 2 included 34 genes ([148]S2
Table). Nine genes had previously been reported as being affected by
positive selection ([149]S4 Table) [[150]21, [151]23–[152]26]. ARHGAP30
and USF1 in Cluster 2 have been reported to show especially strong
signals of positive selection in African populations [[153]24].
Ten genes were mapped to 58 pathway maps (i.e., five “Metabolism”, nine
“Environmental Information Processing,” five “Cellular Processes,” 21
“Organismal Systems,” and 18 “Human Diseases” pathways. In addition to
the pathways that appeared in the enrichment analysis, GSK3B was mapped
to the “Immune System” pathways “B cell receptor signaling pathway” and
“Chemokine signaling pathway,” and MYLK was mapped to “Platelet
receptor signaling pathway.” Regarding infectious diseases, GSK3B was
mapped to “Amoebiasis,” “Epstein–Barr virus infection,” “HTLV-I
infection,” “Influenza A,” and “Measles.”
In the NHGRI GWAS catalog, five SNPs in 76 haplotype blocks were
previously reported [[154]55–[155]58]. These five SNPs in Cluster 2
were associated with bone mineral density, prostate-specific antigen
levels, hair morphology, and breast cancer ([156]S5 Table). Only one
SNP, rs9383951, which was associated with breast cancer, was mapped to
a KEGG pathway through ESR1.
Cluster 3
The ancient haplotype blocks in Cluster 3 included 17 genes ([157]S2
Table). Eight were previously reported as candidates of positive
selection ([158]S4 Table) [[159]26]. SH2B, known to be associated with
celiac disease, is in Cluster 3 and has been reported to be under
convergent evolution in Asia and Europe [[160]26].
Eight genes were mapped to 40 pathway maps, which included one “Genetic
Information Processing,” eight “Environmental Information Processing,”
five “Cellular Processes,” eight “Organismal Systems,” and 18 “Human
Diseases” pathways. In addition to the pathways that appeared in the
enrichment analysis, IL10 was mapped to immune system-related pathways
such as the “Jak-STAT signaling pathway,” and immune system-related
diseases such as “Asthma,” “Inflammatory bowel disease (IBD),”
“Systemic lupus erythematosus,” “Epstein–Barr virus infection,” and
“Malaria.” IL10 has been reported to be associated with pathogen
diversity and susceptibility to autoimmune diseases [[161]17].
In the NHGRI GWAS catalog, two SNPs in 39 haplotype blocks were
previously reported [[162]59, [163]60]. We found two SNPs, rs1194289
and rs7101446, in Cluster 3 associated with response to anti-depressant
treatment in major depressive disorder, and economic and political
preferences ([164]S5 Table). These two SNPs were not mapped to any KEGG