Meta Gene 16 (2018) 39–49

Contents lists available at ScienceDirect

Meta Gene
journal homepage: www.elsevier.com/locate/mgene

A multifactor dimensionality reduction model of gene polymorphisms and T

an environmental interaction analysis in type 2 diabetes mellitus study
among Punjabi, a North India population
⁎
Basanti Barna, Badaruddoza , Manpreet Kaur, A.J.S. Bhanwer
Department of Human Genetics, Guru Nanak Dev University, Amritsar, Punjab, India

A R T I C L E I N F O A B S T R A C T

Keywords: The effects of any single genetic variation for a common complex disease such as T2DM may be dependent on
Type 2 diabetes mellitus other genetic variations (gene-gene interactions) and environmental factors (gene-environment interactions).
Multifactor dimensionality reduction (MDR) Multifactor Dimensionality Reduction (MDR) method helps in detection and characterization of susceptibility in
analysis common complex multifactor disorders like Type 2 Diabetes Mellitus (T2DM). The negligible studies are
Punjab
available with this model approach to detect the status of T2DM in north Indian population. Hence, the major
North-West India
objectives of the present study were to investigate the association of ENPP1 K121Q (rs1044498), TCF7L2 G > T
Gene-gene interactions
Gene-environment interactions (rs12255372) and GYS1 XbaI (A1 > A2) (rs8103451) gene variants with T2DM in the north Indian population;
and to determine whether significant gene-gene and gene-environment (risk factors related to obesity and
cardiovascular diseases) interactions exist between these selected genes in affecting type 2 diabetes mellitus
using MDR analysis. A total of 500 participants consisting of 250 type 2 diabetes mellitus (T2DM) and 250
healthy subjects were recruited for this study. Genotyping was performed by PCR-RFLP method. Anthropometric
and physiometric variables such as height, weight, waist circumference (WC), hip circumference (HC), SBP and
DBP, were measured using standard protocol. The odds ratio and Hardy- Weinberg equilibrium deviation ana-
lyses were performed. The gene-gene and gene-environment interactions were performed by multifactor di-
mensionality reduction (MDR) analysis. In results it was observed that two genes ENPP1 and TCF7L2 are as-
sociated with T2DM. However, an insignificant association of the XbaI (A1 > A2) polymorphism in GYS1 gene
with T2DM was demonstrated. The gene-gene interactions revealed that all the three SNPs have a synergistic
effect with each other. The MDR method for gene-environment interactions showed all interaction models first to
ninth order interactions for T2DM patients as significant for susceptibility of obesity. The results showed that
both the genes ENPP1 and TCF7L2 interacting with WHR and WC increase the susceptibility of obesity many
folds among T2DM patients and non-diabetic controls. In conclusion, it is suggested that pathogenesis of T2DM,
obesity and hypertension involves interplay of a variety of susceptibility alleles and environment. The gene-gene
and gene-environment interactions are not only possible, but, are probably ubiquitous in determining the sus-
ceptibility of complex human diseases. Further studies on epistatic interactions are warranted to elucidate their
possible underlying role in pathogenesis of T2DM.

1. Introduction prevalence of diabetes and pre-diabetes in Punjab representing north

Diabetes mellitus is a worldwide epidemic, causing serious physical
harm and economic burden. This disease has emerged as a major public
health problem of 21st century and is the fifth leading cause of death
globally (Sanghera et al., 2011; Sanghera and Blackett, 2012). The
India is 13.6% and 14.6%, respectively which indicates that there are
0.12 million diabetic and 0.13 million pre-diabetic individuals in
Punjab (Anjana et al., 2011).
Despite numerous literature and evidences, the genetics of T2DM is
still a puzzle and it became evident that the disease is much more

Abbreviations: χ2, chi-square; ADA, American Diabetes Association; BMI, body mass index; DNA, deoxyribonucleic acid; dNTP, deoxynucleotide triphophates; HC, hip circumference;
HWE, Hardy-Weinberg equilibrium; LADA, latent autoimmune diabetes in adults; MODY, maturity-onset diabetes in the young; MDR, multifactor dimensionality reduction (MDR)
analysis; PCR, polymerase chain reaction; RFLP, restriction fragment length polymorphism; SD, standard deviation; T2DM, type 2 diabetes mellitus; WC, waist circumference; WHR, waist
to hip ratio
⁎
Corresponding author.
E-mail address: doza13@yahoo.co.in (Badaruddoza).

https://doi.org/10.1016/j.mgene.2018.01.009
Received 25 July 2017; Received in revised form 25 January 2018; Accepted 27 January 2018
2214-5400/ © 2018 Elsevier B.V. All rights reserved.
B. Barna et al.
heterogeneous and has a strong genetic interaction between type 1
diabetes and type 2 diabetes with an affluent environment. Evidence
has been accumulated that multiple genetic environmental factors may
play important role in determining susceptibility to T2DM. Although
the different variants of candidate genes have become prime targets of
genetic analysis, few studies have considered their interactions
(Neuman et al., 2010; Ramu et al., 2011; Zhou et al., 2012). The gene-
gene interaction which is called epistasis might explain a large portion
of genetic variability for such complex diseases. Most of the studies for
genetic dissection of T2DM have focused on estimating effects of in-
dividual genes and excluded potential epistasis effects in their analytic
models.
The effects of any single genetic variation for a common complex
disease such as T2DM may be dependent on other genetic variations
(gene-gene interactions) and environmental factors (gene-environment
interactions). To address this issue, in recent years, a rapid progress has
been made in the development of statistical methods for analyzing
gene-gene and gene-environment interactions such as logistic regres-
sion models, stratified analysis, cross over analysis, general relative risk
models, composite linkage and disequilibrium methods (Basu et al.,
2011; Wu et al., 2011). However, most of these methods require a large
sample size to model high order interactions (Moore and Williams,
2002; Moore et al., 2006). Hence, for moderate sample size data, one
method of detecting and characterizing susceptibility in common
complex multifactor disorders is the Multifactor Dimensionality Re-
duction (MDR) method (Ritchie et al., 2001). This method detects and
characterizes a high order gene-gene and gene-environment interaction
in case-control studies. Using this method, multi-locus genotype is
classified into high-risk and low-risk and it effectively reduces genotype
predictors from n dimensions to one dimension. The new one dimen-
sional multi-locus genotype variable is evaluated for its ability to
classify and predict the disease status through cross-validation. The
MDR method is model free and it does not assume any particular ge-
netic model.
Almost no study is available with this model approach to detect the
status of T2DM in north Indian population. Hence, the present study has
adopted this highly important statistical tool to assess both the main
effects of single locus and multi-locus interactions to test the hypothesis
that T2DM related genes may contribute to etiology of obesity and
cardiovascular disease. The major objectives of the present study were
to investigate the association of ENPP1 K121Q (rs1044498), TCF7L2
G > T (rs12255372) and GYS1 XbaI (A1 > A2) (rs8103451) gene
variants with type 2 diabetes mellitus in the north Indian population;
and to determine whether significant gene-gene and gene-environment
(risk factors related to obesity and cardiovascular diseases) interactions
exist between these selected genes in affecting type 2 diabetes mellitus
using MDR analysis.
2. Materials and methodology
2.1. Subjects
The detailed protocol of the study was approved by the Ethical
Committee of Guru Nanak Dev University. A total of 500 subjects (250
T2DM patients, 250 controls) were recruited for this study. T2DM
subjects were recruited from different clinical centres in Amritsar dis-
trict in Punjab, whereas unrelated healthy individuals were collected
randomly. All participants provided written informed consent. Diabetes
was diagnosed based on the American Diabetes Association's protocol
(ADA, 2009).
2.2. Inclusion/exclusion criteria
Inclusion: During the sample collection, the individuals over
40 years old were recruited for the study and only one member from
one family was taken. Individuals with diabetes in whom the diagnosis
40
Meta Gene 16 (2018) 39–49
was made by a physician, who had fasting (10−12h) plasma glucose of
7 mmol/L or higher 126 mg/dL were included in the study. The control
group of the subjects who has fasting plasma glucose of < 5.6 mmol/L
(100 mg/dL) and 2 h plasma glucose of 7.8 mmol/L or less (140 mg/dL)
were included in the study.
Exclusion: The subjects with following conditions were excluded for
the study: pregnant or nursing mother, subjects with any illness/
chronic disease interfering with patients' ability to comply with the
diagnostic protocol, individuals exposed to any medication and pos-
sessing rare forms of diabetes such as maturity-onset diabetes in the
young (MODY) and latent autoimmune diabetes in adults (LADA), pa-
tients having hypertension or undergoing immunosuppressive trans-
plant therapy and those who were unwilling and/or not able to give
written informed consent.
2.3. Anthropometric measurements
Actual age and age on the onset of the disease were recorded from
the subject's health card provided by the clinical centres. Height,
weight, circumferences of waist (WC) and hip (HC) were taken from
each individual using standard anthropometric techniques and tools
(Singh and Bhasin, 1968). Height and weight were measured to the
nearest 0.5 cm and 0.1 kg, respectively. Body mass index (BMI) was
calculated for an estimate of overall adiposity using the formula:
BMI = weight (kg)/height (m2). Waist and hip circumference (WC and
HC) for an estimate of central obesity were measured to the nearest
0.5 cm with a steel tape. Waist to hip ratio was calculated using the
standard formula: WHR = WC (cm)/HC (cm). Two subsequent mea-
surements were taken and averages were used in the analysis.
2.4. Physiometric measurements
Left arm blood pressures (first-phase systolic and fifth-phase dia-
stolic) were taken from each participant with standard mercury
sphygmomanometer after a 5-min rest. The average of the two sub-
sequent measurements was used for analysis. All efforts were made to
minimize the factors which affect the blood pressure like anxiety, fear,
stress, laughing and recent activity (Badaruddoza and Afzal, 1999). The
radial artery at the wrist was used to count the pulse. It was counted
over 1 min. The difference of SBP and DBP was used as pulse pressure.
2.5. Molecular analysis and reason for selection of the polymorphisms
The genes involved in the pathways controlling insulin secretion
and regulating pancreatic β-cell mass in the progression of diabetes
mellitus have been studied using different advanced techniques. The
current approaches to root out the role of some potential genetic var-
iants present in the pathway of glucose metabolism have reached at the
genome-wide association study (GWAS) level. However, there are many
approaches through which the variants of many genes can be validated
for their role in causing the disease. In the present investigation case-
control design was followed for analysis of single nucleotide poly-
morphisms (SNPs) in the ENPP1, GYS1 and TCF7L genes to assess the
association of these markers with T2DM. The SNPs were analyzed and
were selected after critical perusal and evaluation of the literature and
reports presented in other population.
The three SNPs selected were: (i) ENPP1 K121Q (rs1044498)
polymorphism, Lysine-K to Glutamine-Q substitution of A > C base in
exon 4. It is located on chromosome 6q22–23; (ii) TCF7L2 G > T
(rs12255372) which involves G > T change in intron 4. It is located on
chromosome 10q25.3; (iii) GYS1 XbaI (A1 > A2) (rs8103451) poly-
morphism, change of a single base (C > T) in intron 14. It is located on
chromosome 19q13.3.
Several human studies have been performed to find out the asso-
ciation between these polymorphisms and T2DM in different ethnic
populations. In many studies, the investigators found strong association
B. Barna et al.
of these polymorphisms and T2DM (Kuroyama et al., 1994; Weedon
et al., 2006; McAteer et al., 2008), whereas, many studies have failed to
detect any association (Babadjanova et al., 1997; Vasudevan et al.,
2009; Saberi et al., 2010) creating controversies as to whether this
polymorphism is correlated with T2DM and suggested genetics and
heterogeneity or population/ethnicity specific association.
Some of the TCF7L2, ENPP1 and GYS1 variants have been found to
be strongly associated with T2DM in Asian Indian subjects from south,
south-west India and also north Indian populations (Kadowaki et al.,
1993; Bochenski et al., 2006). Recently, researchers have reported
strong association between TCF7L2 gene and T2DM in many popula-
tions (Fredriksson et al., 2007; Gupta et al., 2013; Jyothi et al., 2013).
Based on these backgrounds, the present investigation selected three
variants of ENPP1, TCF7L2 and GYS1 genes respectively, to investigate
the association with T2DM in north Indian population.
Genotyping of ENPP1, TCF7L2 and GYS1 gene polymorphisms: DNA
was isolated from peripheral blood of the controls and patients with
type 2 diabetes using salting out method with minor modification ac-
cording to the laboratory conditions (Miller et al., 1988). A master mix
volume of 15 μL for PCR reaction was prepared consisting of 20 ng
genomic DNA, 10 pmol of each primer, 1.5 μLMgCl2 Taq buffer
(25 mmol/L), 0.12 μL of Taq DNA polymerase enzyme (5 U/μL) and
0.3 μL dNTPs (10 mmol/L). The PCR was carried on thermal cycler
(Eppendorf, USA) under the conditions given in Table 1. To ensure the
genotyping quality and accuracy, we also performed random duplicates
in 10% of the samples, and there was 99% concordance in the geno-
typing.
2.6. Statistical analysis
All the statistical analyses were done by using Statistical Package for
Social Sciences (SPSS version 17.0, Chicago, IL, USA) and Multifactor
Dimensionality Reduction (MDR) software. The probability values less
than or equal (two-tailed) to 0.05 were considered to be significant.
Departure from Hardy-Weinberg equilibrium (HWE) in cases and con-
trols was tested by the Pearson chi-square test. Genotypic frequency at a
biallelic locus was calculated through simple gene counting technique.
Genotype and allele frequencies of ENPP1 (K121Q), TCF7L2 (G > T)
and GYS1 (A1 > A2) variants in cases and control groups were com-
pared and tested using a Pearson Chi-square (χ2) statistics. The analysis
was performed assuming dominant, co-dominant and recessive genetic
inheritance models.
The case-control studies with relatively small samples have been
detected for the high order gene-interactions using Multifactor-di-
mensionality reduction (MDR). Interaction dendrograms were em-
ployed to detect the interaction between gene-gene (G1 = ENPP1
K121Q; G2 = TCF7L2 G > T and G3 = GYS1 A1 > A2) and gene-
environment interaction (E1 = body mass index; E2 = waist hip ratio;
E3 = waist circumferance; E4 = systolic blood pressure; E5 = diastolic
blood pressure and E6 = Pulse Rate). The attributes (SNPs/factors) that
Table 1
Primer pairs, annealing temperature, PCR product size and restriction enzymes used for the amplification of selected SNPs.
Locus Primer pair
ENPP1 K121Q Forward primer 5′-GCA ATT CTG TGT
TCA CTT TGG A − 3′
Reverse primer 5′-GAG CAC CTG ACC
TTG ACA CA-3′
TCF7L2 G > T Forward primer 5′-CTG GAA ACT AAG
GCG TGA GG −3′
Reverse primer 5′-GGG TCG ATG TTG
TTG AGC TT −3′
GYS1 XbaI (A1 > A2) Forward primer 5′-CTC CTT CCT CTA
CAG TTT CTG −3′
Reverse primer 5′-GTG AGT CTC CTC
TTT GGC CA −3′
41
Meta Gene 16 (2018) 39–49
are strongly interacting appeared close together at the leaves of the tree
whereas, those that do not interact appeared far from one another. The
colour scheme used in the dendrogram comprises a spectrum of colors
representing continuum from synergy (strong interaction) to re-
dundancy (weak interaction). The colors range from red representing a
high degree of synergy, orange- a lesser degree and gold- the midpoint
between synergy and redundancy. On the redundancy end of the
spectrum, the highest degree was represented by the blue colour with
the lesser degree represented by the green colour. From the colour in
the diagram one can understand the interaction of genes and factors.
3. Results and discussion
The entire result was based on case-control study which is an ana-
lytical retrospective observation study. The genotypic analysis were
performed with respect to allelic and genotypic frequencies, test of
association between T2DM patients and non-diabetic controls with
different genetic models for three single nucleotide polymorphisms
(SNPs) including ENPP1 (K121Q), TCF7L2 (G > T) and GYS1
(A1 > A2). The analysis was further categorised with respect to obese
and non-obese; hypertensive and normotensive with respect to three
different SNPs of respective genes.
3.1. ENPP1 K121Q polymorphism
The ectoenzyme phosphodiesterase pyrophosphatase 1(ENPP1)
gene also known as plasma cell glycoprotein-1 (PC-1) encodes for a
transmembrane glycoprotein which inhibits insulin resistance and
subsequent insulin signaling when over expressed in peripheral insulin
target tissues, it is involved in human insulin resistance. Genetic asso-
ciation analysis by many authors have supported and suggested that the
polymorphisms of ENPP1 are significantly associated with the risk of
T2DM, obesity and a series of metabolic syndromes (Costanzo et al.,
2001; Abate et al., 2005; Bacci et al., 2005; Bouhaha et al., 2008; Moore
et al., 2009).
In the present study, the frequencies of KK, KQ and QQ genotypes in
T2DM patients and non-diabetic controls were 62.4%, 27.6%, 10% and
71.2%, 23.6%, 5.2%, respectively. Whereas, the distributions of K and
Q allele were 76.2%, 23.8% and 83%, 17% in T2DM patients and non-
diabetic control group, respectively. The genotypic and allelic dis-
tributions for this variant in pooled data among T2DM patients and
non-diabetic control group were significantly different (p = 0.049 for
genotypes and p = 0.008 for alleles, respectively) and both the alleles K
and Q of ENPP1 have shown a significant association with T2DM (OR:
1.52, 95% CI: 1.12–2.08, p = 0.009). It showed significant association
of QQ genotype with T2DM (OR: 2.19, 95% CI: 1.08–4.43, p = 0.027).
A significant association of T2DM were seen under all the genetic
models such as dominant (OR: 1.49, 95% CI: 1.02–2.17, p = 0.036), co-
dominant (OR: 1.42, 95%CI: 1.07–1.89, p = 0.015) and recessive (OR:
2.03, 95% CI: 1.01–4.06, p = 0.041) (Table 2).
Annealing temperature PCR product Restriction enzyme
50 °C 208 bp AvaII
60 °C 346 bp Tsp5091
61 °C 600 bp XbaI
B. Barna et al. Meta Gene 16 (2018) 39–49

Table 2
Allelic and genotypic frequencies with estimates of relative risks for ENPP1 (K121Q) polymorphism in T2DM patients (n = 250) and non-diabetic controls (n = 250).

Genotypes/alleles T2DM Patients n (%) Control n (%) p value Test of Association OR (95% CI) Genetic model OR (95% CI)

KK 156 (62.4) 178 (71.2) 0.049 1.00 Dominant model 1.49(1.02–2.17)

p=1 (KQ/QQ vs KK) p = 0.036
KQ 69 (27.6) 59 (23.6) 1.33 (0.89–2.00) Co-dominant model 1.42(1.07–1.89)
p = 0.177 (QQvsKQ) = (KQvsKK) p = 0.015
QQ 25 (10.0) 13 (5.2) 2.19 (1.08–4.43) Recessive model 2.03(1.01–4.06)
p = 0.027 (QQ vs KK/KQ) p = 0.041
Total genotypes 250 250
K allele (%) 381 (76.2) 415 (83.0) 0.008 1.52 (1.12–2.08)
Q allele (%) 119 (23.8) 85 (17.0) p = 0.009
Total alleles 500 500

n = number of individuals; values in parentheses indicate percentage; OR = Odds Ratio; CI=Confidence Interval; p = probability of significance level (bold figures indicate corrected
significant values).

The positive association between ENPP1 K121Q variant and T2DM
has previously been reported in US Caucasians and South Asians (Abate
et al., 2005), European (McAteer et al., 2008), French (Meyre et al.,
2005), African-American (Chandalia et al., 2007) and Indian
(Badaruddoza et al., 2014) populations and many other studies. In
contrast no significant association was found in many populations such
as Sicilian (Pizzuti et al., 1999), Finnish (Kubaszek et al., 2004), Italian
and United States Caucasian (Bacci et al., 2005), Spanish (Gonzalez-
Sanchez et al., 2003) populations.
3.2. TCF7L2 G > T polymorphism
The polymorphisms of TCF7L2 (rs12255372 and rs7903146) are the
strongest genetic variants associated with T2DM and have been con-
vincingly replicated in multiple populations (Gracia, 2013; Jyothi et al.,
2013). The present study aimed to explore the effect of TCF7L2
(rs12255372) polymorphism in the north Indian Punjabi population.
The results of this study suggested significant association of this poly-
morphism with T2DM. The frequencies of GG, GT and TT genotypes in
T2DM patients and non-diabetic controls were 66.8%, 20.8%, 12.4%
and 67.2%, 26.4%, 6.4%, respectively. Whereas, the distributions of G
and T allele were 77.2%, 22.8% and 80.4%, 19.6% in T2DM patients
and non-diabetic control groups, respectively among pooled data. The
genotypic distributions for this variant among T2DM patients and non-
diabetic control group have shown significant differences (p = 0.05),
however, allelic frequencies were not found to differ between T2DM
patients and control group (p = 0.191). The Web-Assotest program
analysis of the data showed significant association of TT genotype with
T2DM (OR: 1.95, 95% CI: 1.03–3.69, p = 0.043). A significant asso-
ciation with T2DM were also seen under the recessive model (OR: 2.07,
95% CI: 1.10–3.89, p = 0.021) (Table 3).
In many populations a significant association of the T allele of
rs12255372 (G > T) polymorphism with susceptibility to T2DM has
been established which were similar to the present study (Gupta et al.,
2010; Wang et al., 2013). Besides that some Indian studies have also
reported similar kind of association with this polymorphism (Gupta
et al., 2010; Alami et al., 2012). All subsequent studies conducted on
other populations (Nemr et al., 2012; Ciccacci et al., 2012; Turki et al.,
2013) have replicated the findings for strong association between
TCF7L2 variants and T2DM. Therefore, the selection of this genetic
variant has been done for its possible association in the north Indian
Punjabi population who are highly prone to diabetes.
Moreover, the present results confirmed that the genetic variant
rs12255372 in TCF7L2 is strongly associated with T2DM as previously
described in European populations (Grant et al., 2006; Groves et al.,
2006). The variation in TCF7L2 is the most significant genetic factor for
diabetes mellitus described in Indian population to date (Chandak et al.,
2007; Sanghera et al., 2008a; Jyothi et al., 2013).
Hence, the importance of TCF7L2 as a diabetic gene is now estab-
lished across the ethnicities but many more genes for diabetes espe-
cially in non-white population still remain to be identified. Sanghera
et al. (2008b) observed that the risk alleles found to be associated with
T2DM in the Asian Indian population in four significant SNPs (IGF2BP2,
TCF7L2 and FTO) were consistent with those reported by the genome-
wide association studies in Caucasians and Japanese (Scott et al., 2007;
Zeggini and McCarthy, 2007).
3.3. GYS1 XbaI (A1 > A2) polymorphism
It has been suggested that impaired glycogen synthesis could play a
central role in the pathogenesis of T2DM. A common variant XbaI (A1/
A2) on intron 14 in GYS1 gene has been associated with T2DM and
insulin resistance, as well as increased risk of cardiovascular morbidity
and mortality (Fredriksson et al., 2007; Groop and Orho-Melander,
2008). The carriers of this variant allele of GYS1 polymorphism are not
able to increase their glycogen synthesis in the response to exercise.
There seems to be an interaction between this polymorphism and ex-
ercise in the prevention of cardiovascular disease (Fredriksson et al.,

Table 3
Allelic and genotypic frequencies with estimates of relative risks for TCF7L2 (G > T) polymorphism in T2DM patients (n = 250) and non-diabetic controls (n = 250).

Genotypes T2DM patients n (%) Control n (%) p value Test of association OR (95% CI) Genetic model OR (95% CI)

GG 167 (66.8) 168 (67.2) 0.050 1.00 Dominant model 1.02 (0.70–1.48)
p=1 (GT/TT vs GG) p = 0.924
GT 52 (20.8) 66 (26.4) 0.792(0.52–1.21) Co-dominant model 1.16 (0.89–1.52)
p = 0.286 (TTvsGT) = (GTvsGG) p = 0.276
TT 31 (12.4) 16 (6.4) 1.95 (1.03–3.69) Recessive model 2.07 (1.10–3.89)
p = 0.043 (TT vs GG/GT) p = 0.021
Total genotypes 250 250
G allele (%) 386 (77.2) 402 (80.4) 0.191 1.15 (0.85–1.56)
T allele (%) 114 (22.8) 98 (19.6) p = 0.394
Total alleles 500 500

n = number of individuals; values in parentheses indicate percentage; OR = Odds Ratio; CI=Confidence Interval; p = probability of significance level (bold figures indicate corrected
significant values).

42
B. Barna et al. Meta Gene 16 (2018) 39–49

Table 4
Allelic and genotypic frequencies with estimates of relative risks for GYS1 (A1 > A2) polymorphism in pooled T2DM patients (n = 250) and non-diabetic controls (n = 250).

Genotypes T2DM patients n (%) Control n (%) p value Test of association OR (95% CI) Genetic model OR (95% CI)

A1A1 237 (94.8) 236 (94.4) 0.577 1.00 Dominant model 0.92 (0.43–2.01)
p=1 (A1A2/A2A2 vs A1A1) p = 0.843
A1A2 9 (3.6) 12 (4.8) 0.746 (0.31–1.81) Co-dominant model 1.05 (0.58–1.91)
p = 0.656 (A2A2vsA1A2) = (A1A2vsA1A1) p = 0.879
A2A2 4 (1.6) 2 (0.8) 1.99 (0.36–10.97) Recessive model 2.02 (0.37–11.11)
p = 0.685 (A2A2 vs A1A1/A1A2) p = 0.407
Total genotypes 250 250
A1 allele (%) 483 (96.6) 484 (96.8) 0.859 1.06 (0.53–2.13)
A2 allele (%) 17 (3.4) 16 (3.2) p = 0.990
Total alleles 500 500

n = number of individuals; values in parentheses indicate percentage; OR = Odds Ratio; CI=Confidence Interval; p = probability of significance level.

2007). The frequencies of A1A1, A1A2 and A2A2 genotypes in T2DM Table 5
patients and non-diabetic controls were 94.8%, 3.6%, 1.6% and 94.4%, Test for Hardy-Weinberg Equilibrium departure of three Single Nucleotide polymorph-
isms for study subjects.
4.8%, 0.8%, respectively. Whereas, the distributions of A1 and A2 allele
were 96.6%, 3.4% and 96.8%, 3.2% in T2DM patients and non-diabetic Study group χ2 p value
controls, respectively among pooled data. The genotypic and allelic
distributions for this variant were not found to be statistically sig- ENPP1(K121Q)
nificant between T2DM patients and non-diabetic controls (p = 0.577 T2DM patients 14.288 0.000
Non-diabetic controls 6.701 0.010
for genotypes and p = 0.859 for alleles). However, the genotype asso- Both 20.989 0.000
ciation under recessive model was higher as compared to other models
TCF7L2 (G > T)
(OR: 2.02, 95% CI: 0.37–11.11, p = 0.407) but, it did not show any
T2DM patients 41.850 0.000
statistical significance. The association of A2A2 genotype with T2DM Non-diabetic controls 6.590 0.010
also showed higher but insignificant association (OR: 1.99, 95% CI: Both 48.439 0.000
0.36–10.97, p = 0.685) (Table 4). GYS1 (A1 > A2)
Although several authors have reported an association between T2DM patients 51.066 0.000
GYS1 gene and T2DM (Majer et al., 1996; Shimomura et al., 1997). A Non-diabetic controls 12.680 0.000
number of studies with a negative association have also been reported Both 63.745 0.000

(Kadowaki et al., 1993; Rissanen et al., 1997). These discrepancies in

p = probability of significance level (bold figures indicate corrected significant values).
results highlight the problems with population-based association stu-
dies. Differences in the genetic backgrounds between the patients and
interactions have differential epidemiological significances for etiology
control subjects can contribute to a false positive or negative result.
of disease and have public health significance. Understanding the gene-
However, the samples included in the present study are almost homo-
gene and gene-environment interaction have also important contribu-
geneous in nature as because controls are also collected from the same
tion for public health planning because it helps in predicting the sus-
communities and surroundings where the cases collection centre was
ceptibility of the disease which provides a significant basis for re-
present. Therefore, the question remains unanswered that how
commendation of disease prevention (Ottman, 1996; Wu et al., 2011).
A1 > A2 polymorphism would confirmed increased susceptibility to
The disease like T2DM is caused by multi-factorial, multistep and
hypertension and T2DM or both.
complex pathological changes. It has the internal genetic factors and
The present study demonstrated an insignificant association of the
external environmental factors like lifestyle, food habits, alcohol con-
XbaI (A1 > A2) polymorphism in GYS1 gene with T2DM. This finding
sumption, smoking, physical activity, body mass index, waist-hip ratio,
can be compatible with the previous demonstrations that already have
waist circumference and elevated blood pressure. Therefore, this dis-
been mentioned. Thus, these findings re-emphasize the need to consider
ease is especially affected by genetic and environmental lifestyle fac-
the effects of the genetic variants in complex diseases in concern with
tors. The analysis of interactions has been performed by multifactor
their environmental triggers.
dimensionality reduction (MDR) analysis.
Table 5 describes the test for Hardy-Weinberg Equilibrium (HWE)
MDR is a methodology that was recently discovered for analyzing
departure of three single nucleotide polymorphisms for the study sub-
the interactions. It has an ability to detect and characterize simulta-
jects. The Hardy-Weinberg Equilibrium (HWE) of the three poly-
neously the impact of combined effects of multiple disease risk factors.
morphisms [ENPP1 (K121Q), TCF7L2 (G > T) and GYS1 (A1 > A2)]
The other advantage is that when the main effects are not statistically
in T2DM patients and non-diabetic controls was examined using Co-
significant, it still has high level of interactions. Using a dendrogram
chran-Armitage trend test (3 × 2 contingency table) based on linear
and interaction entropy graph, researcher can visually determine the
regression model. A significant deviation from the HWE within T2DM
effects and strengths of interactions to determine basis for future study.
patient group and non-diabetic control group has been observed. The
deviation from the HWE among all three polymorphisms can be pointed
to hospital-based sampling bias. 3.4.1. Gene-gene interaction
The multifactor dimensionality reduction (MDR) method was used,
followed by conventional statistical analysis. This study effectively
3.4. Gene-gene and gene-environment interaction by multifactor found evidence for a significant gene-gene interaction among the three
dimensionality reduction (MDR) analysis loci (ENPP1-K121Q, TCF7L2- G > T and GYS1-A1 > A2) for T2DM
patients and non-diabetic controls. (Table 6). There was a significant 3-
The present study has explored the interactions among risk factors loci model involving ENPP1-K121Q, TCF7L2-G > T and GYS1-
through gene-gene and gene-environment (factors affecting obesity and A1 > A2, indicating a potential gene-gene interaction between these
cardiovascular disease) interaction for complex diseases like T2DM, three genes. This interaction showed maximum cross validation con-
which is important for the epidemiological research. It is because these sistency (CVC) of 10/10 with the highest testing balance accuracy of

43
B. Barna et al. Meta Gene 16 (2018) 39–49

Table 6 GYS1 and TCF7L2 belonged to one cluster and showed significant in-
Best gene-gene interaction models identified by Multifactor Dimensionality Reduction teraction with the risk to develop T2DM. The gene ENPP1 had also
(MDR) method for T2DM patients and non-diabetic controls.
shown a synergistic interaction to some extent with T2DM. The inter-
No. of Interacting SNPs Training Testing CVC Sign test action entropy graphical model described the percentage of entropy
loci in balance balance (p value) prepared using MDR. In this investigation, 3-loci interaction model
model accuracy accuracy revealed differences of the three genes (ENPP1-K121Q, TCF7L2-G > T,
GYS1-A1 > A2) in interaction estimate. The schematic colouration
1 ENPP1-K121Q 0.5464 0.5392 10/10 7(0.172)
2 ENPP1-K121Q 0.5525 0.5287 10/10 7(0.172) used in the visualization tools representing a continuum of synergy to
TCF7L2-G > T redundancy. The present entropy interaction graphical model (Fig. 2)
3 ENPP1-K121Q 0.5661 0.5430 10/10 8(0.050) revealed that TCF7L2 and GYS1 have significant synergistic interaction
TCF7L2-G > T i.e. at least 0.48% sharing the positive information gain with respect to
GYS1-A1 > A2
T2DM, whereas, TCF7L2 contributed 0.51% and GYS1 alone con-
CVC: cross-validation consistency; p = probability of significance level. tributed 0.10% towards susceptibility to T2DM. Therefore, TCF7L2-
GYS1 pair had a higher degree of entropy percentage (0.48%) as
compared to other pair of combinations. This showed MDR method to
be effective in detecting multi-genetic interaction among different loci.
Evidence of these interactions between the genes have also been ob-
served and confirmed from the dendrograms and interaction entropy
models.
Using candidate SNPs several studies have shown gene-gene inter-
actions in obesity and insulin resistance, however, no studies are
available to see the gene-gene interaction with the present SNPs
(ENPP1-K121Q, TCF7L2- G > T and GYS1-A1 > A2) in T2DM from
north Indian population. Therefore, it is difficult to compare the present
results with others. However, many other genes have been frequently
Fig. 1. The dendrogram representing the nature of interaction between the three SNPs examined and reported a connection with gene-gene interactions in
(ENPP1-K121Q, TCF7L2- G > T and GYS1- A1 > A2) for T2DM patients and non-dia- many populations such as Danish, Finnish, White, Japanese and mix
betic controls. American population (Mentuccia et al., 2002; Cho et al., 2004).
Obesity is a complex disease which is influenced by genetic and
environmental factors. In recent years, many studies have been focused
on gene-gene interactions which are partly due to the appearance of
new statistical theories. Therefore, multifactor dimensionality reduc-
tion analysis (MDR) is provided to be a useful statistical tool for de-
tecting gene-gene interaction. Among T2DM patients (Table 7), it was
observed that cross validation consistencies (CVC) of 3-loci (ENPP1-
K121Q, TCF7L2-G > T, GYS1-A1 > A2) were highest (10/10). How-
ever, the accuracies of testing sample were not reaching 50% in any of
the models and no interaction was found significant with respect to
sign-test (p values). Therefore, the 3-loci interaction model may be
considered the best model with respect to cross validation consistency
(CVC) values which showed that there was an interaction between the

Table 7
Best gene-environment interaction models identified by Multifactor Dimensionality
Reduction (MDR) method for T2DM patients and non-diabetic controls.

Factor Factors Training Testing CVC Sign test (p

number balance balance value)
accuracy accuracy

1 E4 0.6116 0.6119 10/10 10(0.001)

2 E4 G1 0.6254 0.6091 9/10 10(0.001)
3 E4 G1 G2 0.6395 0.5872 6/10 10(0.001)
4 E2 E4 G1 G2 0.6562 0.5694 5/10 8(0.05)
5 E2 E3 E4 G1 0.6804 0.5616 6/10 7(0.172)
G2
6 E2 E3 E4 E5 0.7068 0.5657 3/10 8(0.05)
Fig. 2. Interaction entropy model analyzed by multifactor dimensionality reduction G1 G2
(MDR) for T2DM patients and non-diabetic controls. 7 E2 E3 E4 E5 0.7362 0.5653 9/10 8(0.05)
Note: Colour scheme used - Red = High synergistic interaction (Synergy), E6 G1 G2
Orange = Lesser degree of interaction, Green = Weak interaction, Blue = No interaction 8 E1 E2 E3 E4 0.7660 0.5627 10/10 7(0.172)
(redundancy), Gold = Midway interaction between synergy and redundancy. (For inter- E5 E6 G1 G2
pretation of the references to colour in this figure legend, the reader is referred to the web 9 E1 E2 E3 E4 0.7776 0.5707 10/10 8(0.05)
version of this article.) E5 E6 G1 G2
G3

54.30% and 8 for sign-test (p = 0.05). The dendrogram provided by CVC: Cross-validation consistency; E1 = body mass index; E2 = waist hip ratio;
MDR analysis was examined to assist the visualization and interpreta- E3 = waist circumference; E4 = Systolic blood pressure; E5 = Diastolic blood pressure;
tion of potential interaction. As examined in the dendrogram (Fig. 1), E6 = Pulse Rate; G1 = ENPP1-K121Q; G2 = TCF7L2-G > T; G3 = GYS1-A1 > A2;
p = probability of significance level.

44
B. Barna et al. Meta Gene 16 (2018) 39–49

Table 8
Best gene-gene interaction models identified by Multifactor Dimensionality Reduction
(MDR) method for obese and non-obese T2DM patients.

No. of Interacting SNPs Training Testing CVC Sign test (p

loci in balance balance value)
model accuracy accuracy

1 TCF7L2-G > T 0.5248 0.4694 8/10 3(0.945)

2 TCF7L2-G > T 0.5422 0.4519 9/10 3(0.945)
GYS1-A1 > A2
3 ENPP1-K121Q 0.5496 0.3906 10/10 0(1.00)
TCF7L2-G > T
GYS1-A1 > A2

Fig. 3. The dendrogram representing the nature of interaction between the three SNPs CVC: Cross-validation consistency; p = probability of significance level.
(ENPP1-K121Q, TCF7L2- G > T and GYS1- A1 > A2) and environmental factors for
T2DM patients and non-diabetic controls.

Fig. 5. The dendrogram representing the nature of interaction between the three SNPs
(ENPP1-K121Q, TCF7L2- G > T and GYS1- A1 > A2) for obese and non-obese pooled
T2DM patients.

Fig. 4. Interaction entropy model analyzed by multifactor dimensionality reduction

(MDR) for T2DM patients and non-diabetic controls.
Note: Colour scheme used - Red = High synergistic interaction (Synergy),
Orange = Lesser degree of interaction, Green = Weak interaction, Blue = No interaction
(redundancy), Gold = Midway interaction between synergy and redundancy. (For inter-
pretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)

three genes. As observed in the dendrogram (Fig. 3), ENPP1 and GYS1
belonged to one cluster and TCF7L2 belonged to another cluster and
showed redundancy which referred to the situation in which the en-
tropy-based interaction between the two SNPs provided less informa-
tion. However, TCF7L2 showed an intermediate interaction with other
two genes. As observed in the entropy interaction graph (Fig. 4), GYS1-
TCF7L2 pair showed some kind of association and contributed 0.16%
towards the susceptibility to T2DM. However, GYS1 and TCF7L2 con-
tributed independently only 0.26% and 0.28%, respectively.
The results of gene-gene interaction of MDR model analysis among
hypertensive and normotensive T2DM patients are summarized in
Table 8. The cross validation consistency (CVC) was maximal (10/10)
in 1-locus model (ENPP1) and 3-loci models (ENPP1-K121Q, TCF7L2-
G > T, GYS1-A1 > A2). The accuracy of the test samples was highest
in 1-locus model (54.93%) which was followed by 3-loci model
(53.60%). The training balance accuracy was found maximum in 3-loci
model (58.73%). However, no loci interaction model was found to be
significant. Therefore, 1-locus interaction model was the best model
followed by 3-loci interaction model towards contribution to the
etiology of hypertension in T2DM patients. As seen in the dendrogram
Fig. 6. Interaction entropy model analyzed by multifactor dimensionality reduction
(MDR) for obese and non-obese pooled T2DM patients.
Note: Colour scheme used - Red = High synergistic interaction (Synergy),
Orange = Lesser degree of interaction, Green = Weak interaction, Blue = No interaction
(redundancy), Gold = Midway interaction between synergy and redundancy. (For inter-
pretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
(Fig. 5), ENPP1 and TCF7L2 formed one cluster and GYS1 formed an-
other cluster. The genes ENPP1 and TCF7L2 showed a strong synergistic
interaction while GYS1 showed a weak synergistic association with
other genes. As observed in the interaction entropy graph (Fig. 6),
ENPP1-TCF7L2 pair had the highest degree of entropy percentage
(1.43%) when compared to other pair of SNP combinations. However,
individual contribution of ENPP1, TCF7L2 and GYS1 towards prediction
of the susceptibility to hypertension in T2DM patients was 1.14%,
0.55% and 0.59%, respective.

45
B. Barna et al. Meta Gene 16 (2018) 39–49

Table 9
Best gene-environment interaction models identified by Multifactor Dimensionality
Reduction (MDR) method for obese and non-obese T2DM patients.

Factor Factors Training Testing CVC Sign test (p

number balance balance value)
accuracy accuracy

1 E2 0.7530 0.7496 10/10 9(0.010)

2 E2 G1 0.7582 0.7367 8/10 9(0.010)
3 E2 E5 G2 0.7721 0.7207 5/10 9(0.010)
4 E2 E5 G1 G2 0.7861 0.7166 8/10 9(0.010)
5 E1 E2 E5 0.8053 0.6954 8/10 9(0.010)
G1 G2
6 E1 E2 E3 E5 0.8210 0.6174 4/10 8(0.05)
G1 G2
Fig. 7. The dendrogram representing the nature of interaction between the three SNPs
7 E1 E2 E3 E4 0.8395 0.6324 10/10 9(0.010)
E5 G1 G2 (ENPP1-K121Q, TCF7L2- G > T and GYS1- A1 > A2) and environmental factors for
8 E1 E2 E3 E4 0.8469 0.6198 9/10 9(0.010) obese and non-obese T2DM patients.
E5 G1 G2 G3
9 E1 E2 E3 E4 0.8469 0.6198 10/10 9(0.010)
E5 E6
G1 G2 G3

CVC: Cross-validation consistency; E1 = body mass index; E2 = waist hip ratio;

E3 = waist circumference; E4 = Systolic blood pressure; E5 = Diastolic blood pressure;
E6 = Pulse Rate; G1 = ENPP1-K121Q; G2 = TCF7L2-G > T; G3 = GYS1-A1 > A2;
p = probability of significance level.

3.4.2. Gene-environment interaction

Epidemiological studies have predominant source of literature on
gene-environment interaction (lifestyle) in T2DM and CVD but rarely a
handful of studies have examined the gene-environment interaction on
the susceptibility of T2DM and CVD.
The present study has presented a comparative gene-environment
interactions for three genes (ENPP1-K121Q, TCF7L2-G > T and GYS1
A1 > A2) and environment (obesity and cardiovascular related risk
factors) for the detection of susceptibility for T2DM and CVD and in-
terpretation of epistasis involved in genetic studies of disease suscept-
ibility. (Table 9). It was observed that the cross validation consistencies
(CVC) of 1 factor E4 (SBP), 8 factor [E1 (BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E4
(SBP) ∗ E5 (DBP) ∗ E6 (PR) ∗ G1 (ENPP1) ∗ G2 (TCF7L2)] and 9 factor
Fig. 8. Interaction entropy model analyzed by multifactor dimensionality reduction
[E1 (BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E4 (SBP) ∗ E5 (DBP) ∗ E6 (PR) ∗ G1 (MDR) for obese and non-obese T2DM patients.
(ENPP1) ∗ G2 (TCF7L2) ∗ G3 (GYS1)] models were maximal (10/10). Note: Colour scheme used - Red = High synergistic interaction (Synergy),
Whereas, the accuracy of test sample was highest in 1 factor interaction Orange = Lesser degree of interaction, Green = Weak interaction, Blue = No interaction
model E4 (SBP) (61.19%) and the accuracy of training sample was (redundancy), Gold = Midway interaction between synergy and redundancy. (For inter-
77.76% for 9 factor interaction model with respect to sign-test. All the pretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
models except 5 factor interaction model [E2 (WHR) ∗ E3 (WC) ∗ E4
(SBP) ∗ G1 (ENPP1) ∗ G2 (TCF7L2)] and 8 factor interaction model have
been found significant at 5% level. Therefore, from the present analysis, Table 10
it may be assumed that 1 factor E4 (SBP) and 2 factor [E4 (SBP) ∗ G1 Best gene-gene interaction models identified by Multifactor Dimensionality Reduction
(MDR) method for hypertensive and normotensive T2DM patients.
(ENPP1)] interaction models were the best models which showed that
there was an interaction between the SBP and ENPP1 for the prediction No. of Interacting SNPs Training Testing CVC Sign test (p
of T2DM. As observed in the dendrogram (Fig. 7), of interaction system loci in balance balance value)
for genes and factors contributing to obesity and cardiovascular diseases, model accuracy accuracy
E2 (WHR) and G1 (ENPP1) belonged to one cluster and E4 (SBP) and G2
1 ENPP1-K121Q 0.5594 0.5493 10/10 7(0.172)
(TCF7L2) belonged to another cluster. Therefore, interaction of E2 2 ENPP1-K121Q 0.5783 0.4898 6/10 4(0.828)
(WHR) and G1 (ENPP1) showed a strong synergistic effect in predicting TCf7L2-G > T
susceptibility to T2DM. However, combination of E4 (SBP) and G2 3 ENPP1-K121Q 0.5873 0.5360 10/10 6(0.377)
(TCF7L2) showed a weak interaction. The two clusters E2 (WHR), G1 TCF7L2-G > T
GYS1-A1 > A2
(ENPP1) and E4 (SBP), G2 (TCF7L2) showed a midway interaction be-
tween synergy and redundancy. As indicated in interaction entropy CVC: Cross-validation consistency; p = probability of significance level.
graphs (Fig. 8) the pooled data showed that E2 (WHR)-G1 (ENPP1) pair
had highest degree of entropy percentage (0.93%) when compared to Among obese and non-obese T2DM patients (Table 10), it has been
other combinations. Whereas, G1 (ENPP1), G2 (TCF7L2) and E4 (SBP) observed that cross validation consistencies (CVC) of 1 factor E2
individually contributed 0.72%, 0.51% and 3.75%, respectively in pre- (WHR), 7 factor [E1 (BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E4 (SBP) ∗ E5
dicting T2DM. Therefore, E4 (SBP) single factor interaction model itself (DBP) ∗ G1 (ENPP1) ∗ G2 (TCF7L2)] and 9 factor [E1 (BMI) ∗ E2
contributed the highest percentage towards susceptibility to T2DM. The (WHR) ∗ E3 (WC) ∗ E4 (SBP) ∗ E5 (DBP) ∗ E6 (PR) ∗ G1 (ENPP1) ∗ G2
interaction of E4 (SBP) and G1 (ENPP1) also showed a lesser degree of (TCF7L2) ∗ G3 (GYS1)] were maximal (10/10). The accuracy of test
synergy (0.31%). samples was highest in 1 factor interaction model E2 (WHR) (74.96%)

46
B. Barna et al. Meta Gene 16 (2018) 39–49

Table 11
Best gene-environment interaction models identified by Multifactor Dimensionality
Reduction (MDR) method for hypertensive and normotensive T2DM patients.

Factor Factors Training Testing CVC Sign test (p

number balance balance value)
accuracy accuracy

1 E4 0.7653 0.7552 10/10 10(0.001)

2 E1 E4 0.7564 0.7456 8/10 10(0.001)
3 E1 E4 G1 0.7719 0.7663 10/10 10(0.001)
4 E1 E4 G1 G3 0.7804 0.7119 4/10 10(0.001)
5 E3 E4 E5 G1 0.7956 0.6740 6/10 9(0.010)
Fig. 9. The dendrogram representing the nature of interaction between the three SNPs
G2
(ENPP1-K121Q, TCF7L2- G > T and GYS1- A1 > A2) for hypertensive and normoten-
6 E2 E3 E4 E5 0.8105 0.6676 7/10 9(0.010)
sive T2DM patients.
G1 G2
7 E1 E2 E3 E4 0.8253 0.6477 6/10 8(0.05)
E5 G1 G2
8 E1 E2 E3 E4 0.8329 0.6560 10/10 8(0.05)
E5 E6 G1 G2
9 E1 E2 E3 E4 0.8329 0.6560 10/10 8(0.05)
E5 E6 G1 G2
G3

CVC: Cross-validation consistency; E1 = body mass index; E2 = waist hip ratio;

E3 = waist circumference; E4 = Systolic blood pressure; E5 = Diastolic blood pressure;
E6 = Pulse Rate; G1 = ENPP1-K121Q; G2 = TCF7L2-G > T; G3 = GYS1-A1 > A2;
p = probability of significance level.

(ENPP1)], 8 factor [E1 (BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E5 (DBP) ∗ E6

Fig. 10. Interaction entropy model analyzed by multifactor dimensionality reduction
(MDR) for hypertensive and normotensive T2DM patients.
Note: Colour scheme used - Red = High synergistic interaction (Synergy),
Orange = Lesser degree of interaction, Green = Weak interaction, Blue = No interaction
(redundancy), Gold = Midway interaction between synergy and redundancy. (For inter-
pretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
which was followed by 2 factor interaction model [E2 (WHR) ∗ G1
(ENPP1)] (73.67%). The accuracy of the training samples was highest
84.69% in factor 8 [E1 (BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E4 (SBP) ∗ E5
(DBP) ∗ G1 (ENPP1) ∗ G2 (TCF7L2) ∗ G3 (GYS1)] and factor 9 [E1
(BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E4 (SBP) ∗ E5 (DBP) ∗ E6 (PR) ∗ G1
(ENPP1) ∗ G2 (TCF7L2) ∗ G3 (GYS1)]. All interaction models have been
found significant with respect to sign-test (p < 0.05). Therefore, 1
factor E2 (WHR) interaction model was the best model which was
followed by 2 factor [E2 (WHR) ∗ G1 (ENPP1)] which showed that there
was an interaction between WHR and ENPP1. Factors 7 and 9 have also
shown a strong interaction between obesity and CVD related factors and
genetic sites 1, 2 and 3. As observed in the dendrogram (Fig. 9), E2
(WHR) and G2 (TCF7L2) belonged to one cluster and E5 (DBP) and G1
(ENPP1) belonged to another cluster. The E2 (WHR) and G2 (TCF7L2)
showed a strong synergistic association and both the clusters showed an
intermediate interaction midway between synergy and redundancy. In
entropy interaction graph (Fig. 10), G2 (TCF7L2)-E2 (WHR) pair
showed highest degree of entropy percentage (0.97%). However, G1
(ENPP1), G2 (TCF7L2), E2 (WHR) and E5 (DBP) independently con-
tributed to predict the susceptibility to T2DM with respect to obesity
with 0.06%, 0.28%, 16.27% and 0.04%, respectively. Therefore, the
contribution of E2 (WHR) alone had maximum contribution towards
susceptibility to T2DM.
The results of the best gene-environment interaction model identi-
fied by MDR have been summarized for hypertensive and normotensive
T2DM patients (Table 11). It was observed that cross validation con-
sistencies (CVC) of 1 factor E4 (SBP), 3 factor [E1 (BMI) ∗ E4 (SBP) ∗ G1
(PR) ∗ G1 (ENPP1) ∗ G2 (TCF7L2) ∗ G3 (GYS1) and 9 factor [E1
(BMI) ∗ E2 (WHR) ∗ E3 (WC) ∗ E4 (SBP) ∗ E5 (DBP) ∗ E6 (PR) ∗ G1
(ENPP1) ∗ G2 (TCF7L2) ∗ G3 (GYS1)] models have been found maximal
(10/10). The accuracy of the test sample was highest (76.63%) in 3
factor interaction model [E1 (BMI) ∗ E2 (WHR) ∗ G1 (ENPP1)] followed
by 2 factor interaction model [E1 (BMI) ∗ E4 (SBP)] which was 74.56%.
The training balance accuracies were highest in 8 and 9 factor inter-
action models (83.29% each). However, all these interaction models
(1–9) were found to be significant (p < 0.05). Therefore, the 3 factor
interaction model was the best model which showed there was an in-
teraction between hypertension related factors such as BMI and SBP
with genetic site G1 (ENPP1). As observed in the dendrogram (Fig. 11),
G1 (ENPP1) and G2 (TCF7L2) showed significant synergistic association
and formed one cluster while E1 (BMI), E4 (SBP) showed redundancy
and formed separate cluster. The two clusters showed a weak sy-
nergistic interaction in predicting susceptibility of T2DM patients to
hypertension. In interaction entropy graph (Fig. 12), G1 (ENPP1)-G2
(TCF7L2) pair had the highest degree of entropy percentage (1.43%)
when compared to the individual genes or pair-wise genes and hy-
pertension related factors. However an independent contribution of G1
(ENPP1), G2 (TCF7L2) and E4 (SBP) was found as 1.14%, 0.55% and
20.21%, respectively towards etiology of hypertension among T2DM
patients.
In this analysis, it has been noticed that entropy based measures
Fig. 11. The dendrogram representing the nature of interaction between the three SNPs
(ENPP1-K121Q, TCF7L2- G > T and GYS1- A1 > A2) and environmental factors for
hypertensive and normotensive T2DM patients.

47
B. Barna et al.
Fig. 12. Interaction entropy model analyzed by multifactor dimensionality reduction
(MDR) for hypertensive and normotensive T2DM patients.
Note: Colour scheme used - Red = High synergistic interaction (Synergy),
Orange = Lesser degree of interaction, Green = Weak interaction, Blue = No interaction
(redundancy), Gold = Midway interaction between synergy and redundancy. (For inter-
pretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
were useful for selecting the subsets of polymorphisms that have shown
in the interactions graph and interaction dendrograms (Jakulin and
Bratko, 2003). Jakulin et al. (2003) constructed using these measures
which are useful for model interpretation. All these interactions and
approaches were carried out using the methods of MDR. Therefore,
MDR is one of the many novel methods that have been developed to
combine attributes.
MDR analysis showed the interaction between environmental fac-
tors SBP genetic factor (ENPP1) for T2DM patients and SBP and TCF7L2
have significant contribution on susceptibility to T2DM.
The results obtained from the gene-gene interaction model are also
supported by dendrogram and interaction entropy graph model for
contribution and combination of interaction for susceptibility to T2DM.
Therefore, exploring the interaction among risk factors (gene-environ-
ment) for complex T2DM is important for disease epidemiology because
these interactions exist and various interaction patterns have differ-
ential epidemiological significances for management of the disease.
Therefore, an understanding of gene-interaction also has important
implication for public health. The analysis showed that environmental
factors (BMI, WHR, WC, SBP, DBP and PR and genetic factors (ENPP1-
K121Q, TCF7L2 -G > T and GYS1 A1 > A2) have identified risk fac-
tors and their interaction. All these interactions were observed to be
significant.
The MDR method showed all the interaction models first to ninth
order interactions for T2DM patients as significant for susceptibility of
obesity. The results have also indicated that both the genes ENPP1 and
TCF7L2 interacting with WHR and WC and increase the susceptibility of
obesity of many folds among T2DM patients and non-diabetic controls.
These results were also supported by dendrogram and interaction en-
tropy model.
The three factor interaction model (BMI ∗ SBP ∗ ENPP1) for T2DM pa-
tients have been found significant for predicting hypertension in T2DM pa-
tients. However, third order model (SBP ∗ TCF7L2 ∗ GYS1) have been found
to be a strong predictor for hypertension. In non-diabetic control group, nine
factor model (BMI ∗ WHR ∗ WC ∗ SBP ∗ SBP ∗ PR ∗ ENPP1 ∗ TCF7L2 ∗ GYS1)
were strong predictor for hypertension.
4. Conclusions
In the entire gene-gene interaction analysis, it was observed that
two genes ENPP1 and TCF7L2 out of three play a crucial role in the
48
Meta Gene 16 (2018) 39–49
pathogenesis of obesity, T2DM and hypertension. It was also observed
that all the three SNPs have a synergistic effect with each other.
Therefore, this further corroborates that pathogenesis of T2DM, obesity
and hypertension involves interplay of a variety of susceptibility alleles.
Thus, more genes need to be identified to clarify the nature of inter-
actions among the SNPs, genes studied and in better understanding of
the pathway implicated. It is suggested that gene-gene interaction is not
only possible but is probably ubiquitous in determining the suscept-
ibility of complex human diseases. Further studies on epistatic inter-
actions are warranted to elucidate their possible underlying role in
pathogenesis of T2DM. However, these results presented here should be
treated with caution, since this is the first epidemiological evidence
identifying the complex relationship between gene-gene interactions in
the north Indian Punjabi population. Further, studies with large sample
in independent populations are required to validate the findings of the
present study.
Acknowledgement
The work was supported by University Grants Commission, New
Delhi [DRS I (UGC-SAP)]. The first author would like to extend her
sincere gratitude to Prof. AJS. Bhanwer for his utmost help to carry out
this research work.
Declaration of interest
The authors report no conflicts of interest.
References
Abate, N., Chandalia, M., Satija, P., Adams-Huet, B., Grundy, S.M., Sandeep, S., Radha, V.,
Deepa, R., Mohan, V., 2005. ENPP1/PC-1 K121Q polymorphism and genetic sus-
ceptibility to type 2 diabetes. Diabetes 54, 1207–1213.
Alami, M.F., Ahmadi, M., Bazrafshan, H., Tabarraei, A., Khosravi, A., Tabatabaiefar, M.A.
Samaei, M, N., 2012. Association of the TCF7L2 rs12255372 (G/T) variantwith type 2
diabetes mellitus in an Iranian population. Genet. Mol. 35, 413–417.
American Diabetes Association, 2009. Diagnosis and Classification of Diabetes Mellitus.
Diabetes Care 29 (Supple 1), S62–S67.
Anjana, R.M., Pradeepa, R., Deepa, M., Datta, M., Sudha, V., Unnikrishnan, et al., 2011.
Prevalence of diabetes and prediabetes (impaired fasting glucose and/or impaired
glucose tolerance) in urban and rural India: phase I results of the Indian Council of
Medical Research–India diabetes (ICMR–INDIAB) study. Diabetologia 54,
3022–3027.
Babadjanova, G., Allolio, B., Beuschlein, F., Chuchalin, A., Reincke, M., 1997.
Polymorphism of glycogen synthase gene and non-insulin-dependent diabetes mel-
litus in the Russian population. Metabolism 46, 121–122.
Bacci, S., Ludovico, O., Prudente, S., Zhang, Y.Y., Di Paola, R., Mangiacotti, D., Rauseo,
A., Nolan, D., Duffy, J., Fini, G., Salvemini, L., Amico, C., Vigna, C., Pellegrini, F.,
Menzaghi, C., Doria, A., Trischitta, V., 2005. The K121Q polymorphism of the
ENPP1/PC-1 gene is associated with insulin resistance/atherogenic phenotypes, in-
cluding earlier onset of type 2 diabetes and myocardial infarction. Diabetes 54,
3021–3025.
Badaruddoza, Afzal, M., 1999. Age specific differences in blood pressure among inbred
and non-inbred North Indian children. J. Biol. Sci. 24, 177–184.
Badaruddoza, Barna, B., Matharoo, K., Bhanwer, A.J.S., 2014. Role of ENPP1 K121Q
polymorphism and susceptibility to type 2 diabetes in north Indian Punjabi popula-
tion. J. Diabetes Metab. Disord. 5, 5–10.
Basu, S., Wei Pan, W., Oetting, W.S., 2011. A dimension reduction approach for modeling
multi-locus interaction in case-control studies. Hum. Hered. 71, 234–245.
Bochenski, J., Placha, G., Wanic, K., Malecki, M., Sieradzki, J., Warram, J.H., Krolewski,
A.S., 2006. New polymorphism of ENPP1 (PC-1) is associated with increased risk of
type 2 diabetes among obese individuals. Diabetes 55, 2626–2630.
Bouhaha, R., Meyre, D., Kamoun, H.A., Ennafaa, H., Vaillant, E., Sassi, R., Baroudi, T.,
Vatin, V., Froguel, P., Elgaaied, A., Vaxillaire, M., 2008. Effect of ENPP1/PC-1-K121Q
and PPARgamma-Pro12Ala polymorphisms on the genetic susceptibility to T2D in the
Tunisian population. Diabetes Res. Clin. Pract. 81, 278–283.
Chandak, G.R., Janipalli, C.S., Bhaskar, S., Kulkarni, S.R., Mohankrishna, P., Hattersley,
A.T., Frayling, T.M., Yajnik, C.S., 2007. Common variants in the TCF7L2 gene are
strongly associated with type 2 diabetes mellitus in the Indian population.
Diabetologia 50, 63–67.
Chandalia, M., Grundya, S.M., Adams-Huet, B., Abate, N., 2007. Ethnic differences in the
frequency of ENPP1/PC1 121Q genetic variant in the Dallas heart study cohort. J.
Diabetes Complicat. 21, 143–148.
Cho, Y.M., Ritchie, M.D., Moore, J.H., Park, J.Y., Lee, K.-U., Shin, H.D., Lee, H.K., Park,
K.S., 2004. Multifactor-dimensionality reduction shows a two-locus interaction as-
sociated with type 2 diabetes mellitus. Diabetologia 47, 549–554.
B. Barna et al.
Ciccacci, C., Di Fusco, D., Cacciotti, L., Morganti, R., D'Amato, C., Novelli, G., Sangiuolo,
F., Spallone, V., Borgiani, P., 2012. TCF7L2 gene polymorphisms and type 2 diabetes:
association with diabetic retinopathy and cardiovascular autonomic neuropathy.
Acta Diabetol. http://dx.doi.org/10.1007/s00592-012-0418-x.
Costanzo, B.V., Trischitta, V., Di Paola, R., Spampinato, D., Pizzuti, A., Vigneri, R.,
Frittitta, L., 2001. The Q allele variant (GLN121) of membrane glycoprotein PC-1
interacts with the insulin receptor and inhibits insulin signalling more effectively
than the common K allele variant (LYS121). Diabetes 50, 831–836.
Fredriksson, J., Anevski, D., Almgren, P., Sjogren, M., Lyssenko, V., Carlson, J., Isomaa,
B., Taskinen, M.R., Groop, I., Orho-Mehlander, M., 2007. Variation in GYS1 interacts
with exercise and gender to predict cardiovascular mortality. PLoS One 2, e285.
Gonzalez-Sanchez, J.L., Martınez-Larrad, M.T., Fernandez-Perez, C., Kubaszek, A.,
Laakso, M., Serrano-Ros, M., 2003. K121Q PC-1 gene polymorphism is not associated
with insulin resistance in a Spanish population. Obes. Res. 11, 603–605.
Gracia, M.L., 2013. A case study and meta-analysis of type 2 diabetes research. Adv.
Anthrop. 3, 96–100.
Grant, S.F., Thorleifsson, G., Reynisdottir, I., Benediktsson, R., Manolescu, A., Sainz,
et al., 2006. Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type
2 diabetes. Nat. Genet. 38, 320–323.
Groop, L., Orho-Melander, M., 2008. New insights into impaired muscle glycogen
synthesis. More a problem of muscle contractility than of glucose tolerance. PLoS
Med. 5 (e25).
Groves, C.J., Zeggini, E., Minton, J., Frayling, T.M., Weedon, M.N., Rayner, N.W., Hitman,
G.A., Walker, M., Wiltshire, S., Hattersley, A.T., McCarthy, M.I., 2006. Association
analysis of 6736 U.K. subjects provides replication and confirms TCF7L2 as a type 2
diabetes susceptibility gene with a substantial effect on individual risk. Diabetes 55,
2640–2644.
Gupta, V., Khadgawat, R., Ng, H.K., Kumar, S., Aggarwal, A., Rao, V.R., Sachdeva, M.P.,
2010. A validation study of type 2 diabetes-related variants of the TCF7L2, HHEX,
KCNJ11, and ADIPOQ genes in one endogamous ethnic group of north India. Ann.
Hum. Genet. 74, 361–368.
Gupta, V., Vinay, D.G., Sovio, U., Rafiq, S., Kumar, M.V.K., Janipalli, C.S., et al., 2013.
Association study of 25 type 2 diabetes related loci with measures of obesity in Indian
sib pairs. PLoS One 8, e53944.
Jakulin, A., Bratko, I., 2003. Analyzing attribute interactions. Lect. Notes Artif. Int. 2838,
229–240.
Jakulin, A., Bratko, I., Smrke, D., Demsar, J., Zupan, B., 2003. Attribute interactions in
medical data analysis. Lect. Notes Artif. Int. 2780, 229–238.
Jyothi, K.U., Jayaraj, M., Subburaj, K.S., Prasad, K.J., Kumuda, I., Lakshmi, V., Reddy,
B.M., 2013. Association of TCF7L2 gene polymorphisms with T2DM in the population
of Hyderabad, India. PLoS One 8, e60212.
Kadowaki, T., Kadowaki, H., Yazaki, Y., 1993. Polymorphism of glycogen synthase gene
and non-insulin dependent diabetes mellitus. N. Engl. J. Med. 328, 1568–1569.
Kubaszek, A., Markkanen, A., Eriksson, J.G., Forsen, T., Osmond, C., Barker, D.J., Laakso,
M., 2004. The association of the K121Q polymorphism of the plasma cell glycopro-
tein-1 gene with type 2 diabetes and hypertension depends on size at birth. J. Clin.
Endocrinol. Metab. 89, 2044–2047.
Kuroyama, H., Sanke, T., Ohagi, S., Furuta, M., Furuta, H., Nanjo, K., 1994. Simple
tandem repeat DNA polymorphism in the human glycogen synthase gene is asso-
ciated with NIDDM in Japanese subjects. Diabetologia 37, 536–539.
Majer, M., Mott, D.M., Mochizuki, H., Rowles, J.C., Pedersen, O., Knowler, W.C.,
Bogardus, C., Prochazka, M., 1996. Association of the glycogen synthase locus on
19q13 with NIDDM in pima Indians. Diabetologia 39, 314–321.
McAteer, J.B., Prudente, S., Bacci, S., Lyon, H.N., Hirschhorn, J.N., Trischitta, V., Florez,
J.C., ENPP1 Consortium, 2008. The ENPP1 K121Q polymorphism is associated with
type 2 diabetes in European populations: evidence from an updated meta-analysis in
42,042 subjects. Diabetes 57, 1125–1130.
Mentuccia, D., Proietti-Pannunzi, L., Tanner, K., Bacci, V., Pollin, T.I., Poehlman, E.T.,
Shuldiner, A.R., Celi, F.S., 2002. Association between a novel variant of the human
type 2 deiodinase gene Thr92Ala and insulin resistance: evidence of interaction with
the Trp64Arg variant of the beta-3- adrenergic receptor. Diabetes 51, 880–883.
Meyre, D., Bouatia-Naji, N., Tounian, A., Samson, C., Lecoeur, C., Vatin, et al., 2005.
Variants of ENPP1 are associated with childhood and adult obesity and increase the
risk of glucose intolerance and type 2 diabetes. Nat. Genet. 37, 863–867.
Miller, S.A., Dykes, D.D., Olesky, H.F., 1988. A simple salting out procedure for extracting
DNA from human nucleated cells. Nucleic Acids Res. 16, 1215.
Moore, J.H., Williams, S.W., 2002. New strategies for identifying gene-gene interactions
in hypertension. Ann. Med. 34, 88–95.
Moore, J.H., Gilbert, J.C., Tsai, C.T., Chiang, F.T., Holden, T., Barney, N., White, B.C.,
2006. A flexible computational framework for detecting, characterizing, and inter-
preting statistical patterns of epistasis in genetic studies of human disease suscept-
ibility. J. Theor. Biol. 241, 252–261.
Moore, A.F., Jablonski, K.A., Mason, C.C., McAteer, J.B., Arakaki, R.F., Goldstein, B.J.,
Kahn, S.E., Kitabchi, A.E., Hanson, R.L., Knowler, W.C., Florez, J.C., Diabetes
Prevention Program Research Group, 2009. The association of ENPP1 K121Q with
diabetes incidence is abolished by lifestyle modification in the diabetes prevention
49
Meta Gene 16 (2018) 39–49
program. J. Clin. Endocrinol. Metab. 94, 449–455.
Nemr, R., Turki, A., Echtav, A., Al-Zaben, G.S., Daher, H.S., Irani-Hakime, N.A.,
Keleshian, S.H., Almawi, W.Y., 2012. Transcription factor-7-like 2 gene variants are
strongly associated with type 2 diabetes in Lebanese subjects. Diabetes Res. Clin.
Pract. 98, e23-e27.
Neuman, R.J., Wasson, J., Atzmon, G., Wainstein, J., Yerushalmi, Y., Cohen, J., Barzilai,
N., Blech, I., Glaser, B., Permutt, M.A., 2010. Gene-gene interactions lead to higher
risk for development of type 2 diabetes in an Ashkenazi Jewish population. PLoS One
5, e9903.
Ottman, R., 1996. Theoretical epidemiology. Gene-environment interaction: definitions
and study designs. Prev. Med. 25, 764–770.
Pizzuti, A., Frittitta, L., Argiolas, A., Baratta, R., Goldfine, I.D., Bozzali, M., Ercolino, T.,
Scarlato, G., Iacoviello, L., Vigneri, R., Tassi, V., Trischitta, V., 1999. A polymorphism
(K121Q) of the human glycoprotein PC-1 gene coding region is strongly associated
with insulin resistance. Diabetes 48, 1881–1884.
Ramu, P., Umamaheswaran, G., Shewade, D.G., Swaminathan, R.P., Dutta, T.K.,
Balachander, J., Adithan, C., 2011. Candidate gene polymorphisms of renin angio-
tensin system and essential hypertension in a south Indian Tamilian population. Int.
J. Hum. Genet. 11, 31–40.
Rissanen, J., Pihlajamaki, J., Hekkinen, S., Kekalainen, P., Mykkanen, L., Kuusisto, J.,
Kolle, A., Laasko, M., 1997. New variants in the glycogen synthase gene (Gln71His,
Met416Val) in patients with NIDDM from eastern Finland. Diabetologia 40,
1313–1319.
Ritchie, M.D., Hahn, L.W., Roodi, N., Bailey, L.R., Dupont, W.D., Parl, F.F., Moore, J.H.,
2001. Multifactor-dimensionality reduction reveals high-order interactions among
estrogen-metabolism genes in sporadic breast cancer. Am. J. Hum. Genet. 69,
138–147.
Saberi, H., Mohammadtaghvaei, N., Gulkho, S., Bakhtiyari, S., Mohammadi, M., Hanachi,
P., Gerayesh-nejad, S., Zargari, M., Ataei, F., Parvaneh, L., Larijani, B., Meshkani, R.,
2010. The ENPP1 K121Q polymorphism is not associated with type 2 diabetes and
related metabolic traits in an Iranian population. Mol. Cell. Biochem. 350, 113–118.
Sanghera, D.K., Blackett, P.R., 2012. Type 2 diabetes genetics: beyond GWAS. J. Diabetes
Metab. Disord. 3, 6948–6971.
Sanghera, D.K., Nath, S.K., Orgeta, L., Gambarelli, M., Kim-Howard, X., Singh, J.R.,
Ralhan, S.K., Wander, G.S., Mehra, N.K., Mulvihill, J.J., Kamboh, M.I., 2008a.
TCF7L2 polymorphisms are associated with type 2 diabetes in Khatri Sikhs from
North India: genetic variation affects lipid levels. Hum. Genet. 72, 499–509.
Sanghera, D.K., Orgeta, L., Han, S., Singh, J., Ralhan, S.K., Wander, G.S., Mehra, N.K.,
Mulvihill, J.J., Ferrell, R.E., Nath, S.K., Kamboh, M.I., 2008b. Impact of nine common
type 2 diabetes risk polymorpdisms in Asian Indian Sikhs: PPARG2 (Prop12Ala).
IGF2BP2, TCF7L2 and FTO variants confer a significant risk. BMC Med. Genet. 9,
59–68.
Sanghera, D.K., Been, F.B., Ralhan, S., Wander, G.S., Mehra, N.K., Singh, J.R., Ferrell,
R.E., Kamboh, M.I., Aston, C.E., 2011. Genome-wide linkage scan to identify loci
associated with type 2 diabetes and blood lipid phenotypes in the Sikh diabetes study.
PLoS One 6, e21188.
Scott, L.J., Mohlke, K.L., Bonnycastle, L.L., Willer, C.J., Li, Y., Duren, et al., 2007. A
genome-wide association study of type 2 diabetes in Finns detects multiple suscept-
ibility variants. Science 316, 1341–1345.
Shimomura, H., Sanke, T., Ueda, K., Hanabusa, T., Sakagashira, S.S., Nanjo, K., 1997. A
missense mutation of the glycogen synthase gene (M416V) is associated with insulin
resistance in Japanese population. Diabetologia 40, 947–952.
Singh, I.P., Bhasin, M.K., 1968. Anthropometry. Kamla Raj Enterprises, Delhi.
Turki, A., Al-Zaben, G.S., Mtiraoui, N., Marmmuoch, H., Mahjoub, T., Almawi, W.Y.,
2013. Transcription factor 7-like 2 gene variants are strongly associated with type 2
diabetes in Tunisian Arab subjects. Gene 25, 244–248.
Vasudevan, R., Ismail, P., Ali, A., Mansor, M.S., 2009. Association of TCF7L2 and ENPP1
gene polymorphisms in Malaysian type 2 diabetes mellitus with or without hy-
pertension. Curr. Res. J. Biol. Sci. 4, 703–709.
Wang, J., Zhang, J., Li, L., Wang, Y., Wang, Q., Zhai, Y., You, Z., Hu, D., 2013. Association
of rs12255372 in the TCF7L2 gene with type 2 diabetes mellitus: a meta-analysis.
Braz. J. Med. Biol. Res. 46, 382–393.
Weedon, M.N., Shields, B., Hitman, G., Walker, M., McCarthy, M.I., Hattersley, A.T.,
Frayling, T.M., 2006. No evidence of association of ENPP1 variants with type 2
diabetes or obesity in a study of 8,089 UK Caucasians. Diabetes 55, 3175–3179.
Wu, Y., Zhang, L., Liu, L., Zhang, Y., Zhao, Z., Liu, X., Yi, D., 2011. A multifactor di-
mensionality reduction-logistic regression model of gene polymorphisms and an
environmental interaction analysis in cancer research. Asian Pac. J. Cancer Prev. 12,
2887–2892.
Zeggini, E., McCarthy, M.I., 2007. TCF7L2: the biggest story in diabetes genetics since
HLA? Diabetologia 50, 1–4.
Zhou, Y., Zhang, E., Berggreen, C., Jing, X., Osmark, P., Lang, S., Cilio, C.M., Goransson,
O., Groop, L., Renstrom, E., Hansson, O., 2012. Survival of pancreatic beta cells is
partly controlled by a TCF7L2-p53-p53INP1-dependent pathway. Hum. Mol. Genet.
21, 196–207.