Open Journal of Genetics, 2012, 2, 131-135 OJGen
http://dx.doi.org/10.4236/ojgen.2012.23018 Published Online September 2012 (http://www.SciRP.org/journal/ojgen/)
Exploring correlations among copy number variants
Joseph Abraham1*, Thomas LaFramboise2
1Programa do Pós Graduação em Genética, Faculdade de Medicina de Ribeirão Preto, Universidade de São Paulo, Ribeirão Preto,
Brazil
2Department of Genetics, Case Western Reserve University, Cleveland, USA
Received 16 May 2012; revised 16 June 2012; accepted 6 July 2012
ABSTRACT
There have been a great many recent studies investi-
gating the extent of Copy Number Variation in the
genomes of various species such as human, cattle,
dogs and many others. The results from these studies
indicate that the extent of the Copy Number Varia-
tion in the genome is considerable, and that in hu-
mans and in cattle, frequencies of different Copy
Number Variants may differ in different breeds/eth-
nicities. This is not entirely unexpected as allele fre-
quencies of certain loci vary with different breeds/
ethnicities/species and many known Copy Number
Variants behave similarly to ordinary markers as
regards Mendelian segregation. It is also well known
in many instances, species/breeds/ethnicities show
variation not only in marker allele frequencies, but
also in the extent of Linkage Disequilibrium between
markers. Thus it is worth investigating the extent of
association between Copy Number Variants in dif-
ferent populations. In this paper we will investigate
the extent of correlations between selected Copy
Number Variants in different human populations and
show that statistically significant correlations exist
and are strongly population dependent.
Keywords: Copy Number Variant; Pathway; Selection
1. INTRODUCTION
Over the past several years the genome sequences of
many species have been investigated in increasing detail
and with increasing precision. As a result much is now
known about the extent of variation in the genome se-
quences between different individuals belonging to the
same species. In particular it is known that there is, in
addition to point variation, also considerable structural
variation in the genome. One particular structural variant
which has attracted a lot of attention is Copy Number
Variation (CNV) [1-3] which can be interpreted as DNA
mutations arising from a gain or loss of a certain number
of contiguous base pairs during meioses. The distribution
of CNVs in the human genome is not random, and at-
tempts have been made to understand the evolutionary
events which account for the locations of Copy Number
Variants [4] along with the population genetics aspects of
Copy Number Variants [5,6]. In addition in humans the
association between Copy Number Variants and gene
expression level [7] has been studied, as well as the as-
sociation between Copy Number Variation and a number
of diseases of interest [8]. Apart from humans, Copy
Number Variation has been studied in a number of other
species such as cattle [9], chimpanzee [10], fruitfly [11]
among others. These studies have not only elucidated the
range and extent of Copy Number Variation, studies in
cattle [9] have also shown a clear association between
certain Copy Number Variants and breeds. Based on
these studies it is worth taking a closer look at the con-
nection between Copy Number Variation and ancestry in
humans. In this regard it is worth recalling that not only
marker allele frequencies but also the extent of Linkage
Disequilibrium between markers varies with ancestry, for
example Linkage Disequilibrium in African populations
typically extends over a shorter range than in European
and Asian populations. This suggests that the extent of
statistical correlations between Copy Number Variants
could vary in different populations. It is this issue which
is the main focus of this paper.
In order to investigate this question, what is needed is a
catalogue of Copy Number Variants observed in multiple
unrelated individuals in different populations. The copy
number catalogue we will use is in [12] where different
Copy Number Variants are typed in a number of indi-
viduals in various HapMap populations. The catalogue in
[12] contains information on 1320 Copy Number Variants
with a minor allele frequency larger than 1%, referred to
from now on as Copy Number Polymorhpisms (CNPs) to
analyze differences between populations.
2. MATERIALS AND METHODS
The data in [12] consist of records of over a thousand
*Corresponding author.

J. Abraham, T. LaFramboise / Open Journal of Genetics 2 (2012) 131-135 132
CNPs in three HapMap populations, Yoruba YRI), Euro-
pean (CEU) and Japanese combined with Han Chinese
(CHBJPT). In [12], associated with each CNP is a nu-
merical identifier and also the different levels in which
that CNP appears in the populations. For example, a
CNP with levels (2,3,4) would be interpreted as appear-
ing with no insertions or deletions, one insertion or two
insertions. For each individual there is of course a unique
level; (2,3,4) would indicate that all the individuals con-
sidered have either 0, 1 or 2 insertions.
In each population there are 90 individuals, 30 parent
offspring trios in the case of Yoruba and European
populations and unrelated individuals in the case of the
Japanese Chinese population. In order to get statistics on
a sample of unrelated individuals, offspring data are re-
moved from the Yoruba and European populations. Fur-
thermore, we omit all 27 non autosomal Copy Number
Variants. Summary Statistics of the remaining 1292
CNVs is presented in the Table 1 below.
From Table 1 we see that there are 407 CNPs which
appear in just one level among all the unrelated YRI
parents. Of these 407, 163 of these are common to both
the unrelated YRI parents and the unrelated CEU parents
and 164 are common to YRI parents and the members of
the CHBJPT sample; the other entries can be interpreted
in a similar manner. In all three populations we observe
that CNPs with two levels are relatively common. Fur-
thermore, among CNPs which appear in two levels we
notice that there is relatively little overlap between the
three populations. For example, significantly less than
half of the CNPs which appear in two levels in the
CHBJPT sample also appear in two levels among the
unrelated parents of the YRI sample. Analyzing the sta-
tistics for CNPs which appear in three levels in at least
one of the populations, we see that there is much more
overlap between the different populations. For example
over half of the 238 polymorphisms which appear in two
levels in the CEU populations also appear in two levels
in the YRI population. Indeed the number of these
polymorphisms common to more than one population is
rather larger, even though the actual number of poly-
morphisms which appear in three levels is considerably
smaller than those which appear in two levels. The num-
ber of CNPs which appear in four or five levels is much
Table 1. Distribution of CNP levels.
No. of
Levels YRI CEU CHBJT YRI & CEU
YRI &
CHBJPT
CEU &
CHBJPT
1 407 686 676 163 164 438
2 580 359 327 100 84 88
3 273 238 254 125 128 154
4 23 17 25 7 9 10
5 9 10 10 3 2 3
smaller and will be neglected.
As we are interested in comparing statistical correla-
tions between different polymorphisms we ignore those
polymorphisms which appear in a single level among all
unrelated individuals. Based on Table 1 it appears that
the polymorphisms which occur in two levels not only
occur in large number but also show the least overlap
between the three populations. From now on we will
focus from now exclusively on CNPs which appear in
two levels in at least one of the three populations under
consideration, in order to understand the difference in
statistical correlations between CNPs in different popula-
tions. As we will see, the corresponding χ2 test has a sin-
gle degree of freedom, which is useful when the sample
size is somewhat small affecting the power of the test.
As all CNPs under consideration appear in just two
levels, they will be treated from now on as binary geno-
types. As we would like to study differences in popula-
tions over and above those due to CNP frequencies, we
impose some restrictions on which CNPs we retain. For
CNPs where the sample frequency of the less frequent
level is less than 5%, the distributions in the three popu-
lations are very different. If we remove these CNPs from
the discussion, the frequency distributions become quite
similar as is seen in the Figure 1. In addition we we re-
move all CNPs where the missingness is larger than 5%.
This procedure is analogous to filtering SNPs based on
missingness and minor allele frequency. With this selec-
tion criterion, the mean and median minor level frequen-
cies in the three samples are very similar and not statis-
tically significantly different at the 5% significance level.
The CNP selection criteria can be summarized as fol-
lows:
 Choose only CNPs with two levels (based on the re-
sults of Table 1).
 Remove 2 level CNPs where the frequency of the less
frequent levels is less than 5%.
 From the surviving CNPs remove those with a miss-
ingness of larger than 5%.
The number of CNPs which survive the selection cri-
teria in the YRI, CEU and CHBJPT populations are 203,
119 and 77 respectively. The surviving CNP data for
each population in [12] can be recast in the form of a
matrix as shown below:
1 1 0 1 ...
0 1 1 0 ...
1 1 1 0 ...
. . . . ...
. . . . ...
0 0 0 1 ...

J. Abraham, T. LaFramboise / Open Journal of Genetics 2 (2012) 131-135 133
Figure 1. Distribution of level frequencies.
In this matrix the rows, correspond to different CNPs
which survive the selection criteria and each column
corresponds to a distinct unrelated invividual. The 0 and
1 entries arise since we consider only those CNPs which
are present in just two levels in a given population. In a
given row, 0 might correspond to a single insertion and 1
to no insertions or deletions while in some other row 1
might correspond to an addition and 0 to neither insertion
or deletion. 1 & 0 in this data matrix are purely categori-
cal; there is no numerical significance attached to these
values. As each column corresponds to a distinct unre-
lated individual, the data matrices obtained from the
Yoruba and European individuals have 60 columns while
that from the combined Japanese Han-Chinese sample
has 90 columns. The number of rows is different in each
population, corresponding to the fact that the binary
CNPs are different in different populations.
where n11 counts the number of times where the same
individual has the CNP corresponding to 1 at both loci,
n10 the number of times the same individual has the CNP
corresponding to 1 at one locus and 0 at the other etc. If
the CNPs are on an average uncorrelated, then one
should find, after performing a Fisher’s Exact Test using
the counts in the contingency table, p-values that are not
particularly small. In particular this analysis can be ex-
tended to CNPs on different chromosomes, shedding
light on very long range correlations between CNPs.
In the absence of statistically significant correlations
between CNPs, any small p-values observed should be
artefacts of multiple testing, and not indicative of any
deeper structure underlying CNPs. To check that this is
not the case, we created 1000 permuted data matrices by
shuffling each row of the original data matrix independ-
ently. As the individuals are arranged in columns, in each
permutation any correlations between CNPs in the same
individual are broken up. In each permuted data matrix
we can compute the strength of the correlations between
distinct rows using Fishers Exact Test and the corre-
sponding p-values. The range of p-values obtained in this
manner can be used to decide which p-values are the
Considering an arbitrary pair of rows in any of the
three data matrices, it is possible to construct a (2 × 2)
contingency table of the form
n11 n10
n01 n00

J. Abraham, T. LaFramboise / Open Journal of Genetics 2 (2012) 131-135 134
consequence of multiple testing.
3. RESULTS
The analysis described in the previous section was per-
formed on all three data sets. For a preliminary analysis
of the difference between the populations we focused on
p-values less than 0.005 and on correlations between
CNPs in different chromosomes. If the chromosomes are
different then the CNPs may be considered to be truly
independent and any correlation found is an indication of
the nonrandom nature of Copy Number Variants. It was
found that there were 33 such p-values in the YRI popu-
lation, 5 p-values in the CEU population and just 3 in the
CHBJPT population. The range of p-values is also dif-
ferent, in the YRI population the lowest p-value is 1.678
× 10−6, in the CEU population it is 1.127 × 10−3 and in
the CHBJPT population it is 1.169 × 10−3. It is also
noteworthy that the 7 lowest p-values in the YRI analysis
are smaller than the smallest p-values in the other popu-
lations. This is suggestive of the possibility that the cor-
relation structure between CNPs differs in different
populations. However, this might just be due to the fact
that the number of tests performed is much larger in the
YRI population than in the other two populations.
To rule out this possibility, the permutation test as de-
scribed in the previous section was carried out; to be
conservative in the permuted tests correlations between
CNPs both in the same chromosome and in different
chromosomes were included to decide which small
p-values could possibly be the consequence of multiple
testing. Assuming a significance level of 0.05, none of
the associations found in the CEU and CHBJPT sample
were, significant after compensating for multiple testing.
However, the most significant association found in the
YRI sample (p-value 1.678 × 10−6) remained significant
with a p-value of < 0.02 even after taking into account
multiple testing. This association is between polymor-
phisms on chromosome 6 (hg17 coordinates 202,353 to
326,149) and on chromosome 16 (hg 17 coordinates
33,208,395 to 33,618,281) with minor level frequencies
of 0.1 and 0.15. Such minor level frequencies are not
atypical of the other two populations, what distinguishes
these two CNPs in the YRI sample is the extent to which
Copy Numbers at these locations are correlated. These
CNPs have identifiers 902 & 2172 in [12]. Using 2 as the
baseline for defining no extra copies in [12] all the YRI
individuals could be considered have either one or two
extra copies at these locations. In the YRI population
unrelated individuals with two extra copies at one loca-
tion tend to have two extra copies at the other location.
4. DISCUSSION
Based on the discussion of the previous section we have
evidence for correlations in Copy Number Variants
which are statistically significant, and whose statistical
significance varies from population to population. This
represents a novel approach for analyzing Copy Number
Variants in the same population as well as for the con-
trasting the patterns of copy number variation in different
populations. Our methodology is conceptually similar to
using the structure of observed Linkage Disequilibrium
between markers and not just marker allele frequencies
in order to compare different populations. Furthermore,
the only significant long range correlations between
CNPs were found in the population where LD between
markers has the shortest range. It is also interesting to
note that no correlations of any significance were found
in the largest of all the samples, the CHBJPT sample.
This is not what one would expect if significant p-values
were determined only by sample sizes. Thus the differ-
ences observed between populations cannot be due to
different sample sizes, but may have their origins in the
differences in population histories. In Drosophila mela-
nogaster for example the pattern of Copy Number Varia-
tion is influenced by natural selection [11]. Selection can
also give rise to long range correlations between markers
[14]; this suggests that the strong correlations observed
in the YRI population could be driven by population ge-
netic events unique to that population. In this regard, it is
noteworthy that the region on chromosome 6 that we
identified overlaps with the location of the DUSP22 gene
which participates in the JnK signalling pathway [15]
whose role in cancer proliferation [16] is well docu-
mented. Any possible signals of selection in this region
region would be of considerable interest and worthy of
further study.
5. ACKNOWLEDGEMENTS
KJA was supported during the course of this investigation by the
United States Department of Agriculture, National Research Iniative
Grant USDA NRI-2009-03924 and also by the program Professor
Visitante do Exterior of Coordenação de Aperfeiçoamento de Pessoal
de Nível Superior (CAPES), Brasil. In addition, KJA wishes to thank
Prof. Cheryl Thompson for valuable and encouraging discussions.
REFERENCES
[1] Sebat, J., Lakshmi, B., Troge, J., et al. (2004) Large-scale
copy number polymorphism in the human genome. Sci-
ence, 316, 445-449. doi:10.1126/science.1138659
[2] Iafrate, A.J., Feuk, L., Rivera, M.N., et al. (2004) Detec-
tion of large-scale variation in the human genome. Nature
Genetics, 36, 949-951. doi:10.1038/ng1416
[3] Feuk, L., Carson, A.R. and Scherer, S.W. (2006) Struc-
tural variation in the human genome. Nature Review Ge-
netics, 7, 85-97.
[4] Cooper, G.M., Nickerson, D.A. and Eichler, E.E. (2007)

J. Abraham, T. LaFramboise / Open Journal of Genetics 2 (2012) 131-135 135
Mutational and selective effects on copy-number variants
in the human genome. Nature Genetics, 39, S22-S29.
doi:10.1038/ng2054
[5] Campbell, C.D., Sampas, N., Tsalenko, A., et al. (2011)
Population-genetic properties of differentiated human copy-
number polymorphisms. American Journal of Human Ge-
netics, 88, 317-332. doi:10.1016/j.ajhg.2011.02.004
[6] Mills, R.E., Walter, K., Stewart, C., et al. (2011) Map-
ping copy number variation by population-scale genome
sequencing. Nature, 470, 59-65. doi:10.1038/nature09708
[7] Stranger, B.E., Forrest, M.S. and Dunning, M. (2007)
Relative impact of nucleotide and copy number variation
on gene expression phenotypes. Science, 315, 849-853.
doi:10.1126/science.1136678
[8] McCarroll, S.A. and Altshuler, D.M. (2007) Copy num-
ber variation and association studies of human disease.
Nature Genetics, 39, S37-S42. doi:10.1038/ng2080
[9] George, E.L., Hou, Y.L., Zhu, B., et al. (2010) Analysis
of copy number variations among diverse cattle breeds.
Genome Research, 20, 693-703.
doi:10.1101/gr.105403.110
[10] Perry, G.H., Yang, F., Marques-Bonet, T., et al. (2008)
Copy number variation and evolution in humans and
chimpanzees. Genome Research, 18, 1698-1710.
doi:10.1101/gr.082016.108
[11] Emerson, J.J., Cardoso-Moreira, M., Borevitz, J.O. and
Long, M. (2008) Natural selection shapes genome-wide
patterns of copy-number polymorphism in Drosophila
melanogaster. Science, 320, 1629-1631.
doi:10.1126/science.1158078
[12] McCarroll, S.A., Kuruvilla, F.G., Korn, J.M., et al. (2008)
Integrated detection and population-genetic analysis of
SNPs and copy number variation. Nature Genetics, 40,
1166-1174. doi:10.1038/ng.238
[13] Redon, R., Ishikawa, S. and Fitch, K.R. (2006) Global
variation in copy number in the human genome. Nature,
444, 444-454. doi:10.1038/nature05329
[14] Walsh, B. (2003) Population and quantitative-genetic
models of selection limits. In: Janick, J., Ed., Plant Breed-
ing Reviews: Long Term Selection Maize: Maize, Vol. 24,
Purdue University, West Lafayette.
[15] Shen, Y., Luche, R., Wei, B., Gordon, M.L., Diltz, C.D.
and Tonks, N.K. (2001) Activation of the Jnk signaling
pathway by a dual-specificity phosphatase, JSP-1. Pro-
ceedings of the National Academy of Sciences, 98, 13613-
13618. doi:10.1073/pnas.231499098
[16] Wagner, E.F. and Nebreda, A.R. (2009) Signal integra-
tion by JNK and p38 MAPK pathways in cancer devel-
opment. Nature Genetics Reviews, 9, 537-549.
doi:10.1038/nrc2694