Population-Specific Susceptibility to Crohn's Disease and Ulcerative Colitis; Dominant and Recessive Relative Risks in the Japanese Population


*Corresponding author: Hiroki Oota, Ph.D., Department of Integrated Biosciences, Graduate School of Frontier Sciences, University of Tokyo, Kashiwanoha 5-1-5, Kashiwa, Chiba 277-8562, Japan. E-mail: hiroki_oota@k.u-tokyo.ac.jp


Crohn's disease (CD), a type of chronic inflammatory bowel disease (IBD), is commonly found in European and East Asian countries. The calculated heritability of CD appears to be higher than that of ulcerative colitis (UC), another type of IBD. Recent genome-wide association studies (GWAS) have identified more than thirty CD-associated genes/regions in the European population. In the East Asian population, however, a clear association between CD and an associated gene has only been detected with TNFSF15. In order to determine if CD susceptibility differs geographically, nine SNPs from seven of the European CD-associated genomic regions were selected for analysis. The genotype frequencies for these SNPs were compared among the 380 collected Japanese samples, which consisted of 212 IBD cases and 168 controls. We detected a significant association of both CD and UC with only the TNFSF15 gene. Analysis by the modified genotype relative risk test (mGRR) indicated that the risk allele of TNFSF15 is dominant for CD, but is recessive for UC. These results suggest that CD and UC susceptibility differs between the Japanese and European populations. Furthermore, it is also likely that CD and UC share a causative factor which exhibits a different dominant/recessive relative risk in the Japanese population.


Inflammatory bowel disease (IBD), a generic term for Crohn's disease (CD) and ulcerative colitis (UC), is characterised by inflammation of the gastrointestinal tract. The typical age of onset for both CD and UC is between 20–40 years of age (Xavier & Podolsky, 2007). CD and UC share many diagnostic features, but also have distinguishing characteristics that differentiate the two diseases from each other. For example, CD is able to affect any part of the intestine, while UC is confined to the colon and rectum (Xavier & Podolsky, 2007; Mathew, 2008; Zhang et al., 2008). The calculated sibling recurrence risk ratio (λs: the ratio of the risk for siblings of a patient to develop the disease compared to the risk for a general member of the same population) is higher for CD (λs: 20–35) than for UC (λs: 8–15) (Zhang et al., 2008), indicating a greater risk of CD inheritance (Tysk et al., 1988; Orholm et al., 1991; Podolsky, 2002; Gaya et al., 2006). Neither CD nor UC is inherited in a simple Mendelian fashion, and both diseases appear to involve multiple genes. The incidence of both CD and UC has continued to increase within the European population for the past 50 years, suggesting that environmental factors associated with a rapid change of lifestyle may also be important in the rise of IBD (Loftus, 2004; Mathew, 2008). Hence, IBD can be categorized as a multifactorial disease (Podolsky, 2002; Xavier & Podolsky, 2007).

Recent genome-wide association studies (GWAS) and candidate gene analyses have identified more than 30 susceptible genomic regions of CD in European cases and controls (Hugot et al., 2001; Ogura et al., 2001; Peltekova et al., 2004; Yamazaki et al., 2005; Duerr et al., 2006; Hampe et al., 2007; Libioulle et al., 2007; Parkes et al., 2007; Raelson et al., 2007; Rioux et al., 2007; The Wellcome Trust Case Control Consortium, 2007; Barrett et al., 2008; Kugathasan et al., 2008). From the 30 candidate genomic regions that were surveyed three genes, IL23R, NKX2-3, and MST1, were identified which are also significantly associated with UC in Europeans (Dubois & van Heel, 2008; Fisher et al., 2008; Franke et al., 2008a; Franke et al., 2008b; Silverberg et al., 2009). However, a significant association between CD and these genomic regions has not been detected in non-European populations (Yamazaki et al., 2002; Croucher et al., 2003; Guo et al., 2004; Yamazaki et al., 2004; Zaahl et al., 2005; Ozen et al., 2006; Yamazaki et al., 2007; Yamazaki et al., 2009). For example, while linkage analysis (Hugot et al., 2001; Ogura et al., 2001) and GWAS (Hampe et al., 2007; Barrett et al., 2008) have identified three risk alleles for NOD2 in Europeans, these SNP sites have not been identified as polymorphic in the East Asian population (Yamazaki et al., 2002; Croucher et al., 2003; Guo et al., 2004). Only the TNFSF15 gene has been identified as a CD susceptibility gene in Japanese, as well as in Europeans (Yamazaki et al., 2005). The incidence of CD has been increasing substantially within the Japanese population. Furthermore, the prevalence rate has risen seven times between 1985 and 2006 (Hilmi et al., 2006). These data suggest that the causative genetic and environmental factors of CD differ between geographical and/or ethnic populations.

In order to determine whether the CD risk alleles that have been identified in Europeans also exist in the Japanese population, and are associated with CD, we selected nine SNPs located in seven susceptible genomic regions (i.e. NOD2, IL23R, ATG16L1, TNFSF15, SLC22A4, IRGM, 10q21). Each of these regions, which have been confirmed by multiple independent GWAS and meta-analyses, is known to be associated with CD in the European population. Previous studies of CD in the Japanese population have examined subjects from the Kanto region, an eastern area of Japan's main-island (Honshu). A recent genome-wide SNP study, however, has shown stratification in the Japanese population that consists of the Honshu and the Kyushu-Okinawa (the southwestern islands of the Japanese archipelago) cluster (Yamaguchi-Kabata et al., 2008). Therefore, we examined the Kyushu population (130 CD and 82 UC cases and 168 controls), and compared the genotype data of the nine SNPs with those of the Europeans and Honshu Japanese populations. This study provides confirmation of TNFSF15 susceptibility for CD in the Japanese population. Furthermore, we are also the first to identify TNFSF15 susceptibility for UC, and evaluate its genotypic relative risk for the development of CD and UC.

Material and Methods


Blood samples were collected from 130 CD patients (94 males and 36 females; 25+/−11 years old at the time of diagnosis) and 82 UC patients (36 males and 46 females; 32+/−14 years old at the time of diagnosis) at Fukuoka University Faculty of Medicine (Yoshitake et al., 1999). The diagnosis of CD or UC was made on the basis of clinical symptoms as well as by radiologic, colonoscopic, and histopathologic examinations (Friedman & Podolsky, 1997). Each patient group consisted of the following clinical subtypes: three CD types (35% ileitis, 14% colitis, and 51% ileocolitis) and three UC types (54% pan-colitis, 32% left-sided colitis, and 14% proctitis). Unaffected controls consisted of 168 unrelated Japanese samples. All of the patient and control samples were obtained from Kyushu, a southwestern island of the Japanese archipelago. All individuals included in this study have provided written informed consent. This project was approved by the ethical committees at the Fukuoka University Faculty of Medicine, and Graduate School of Frontier Sciences, University of Tokyo.

Selection of Susceptibility Genes for CD and SNP Typing

We first selected genomic regions which two or more independent research groups have identified as being susceptible to CD. These previous GWAS, linkage analyses, and candidate gene analyses used distinct sample sets and platforms, and also included the subsequent replication with different case-control sample sets. We next selected the genomic regions for which CD association was confirmed by a meta-analysis, using three large data sets of European GWAS (Barrett et al., 2008). Finally, we selected seven genomic regions and targeted nine SNPs (six non-synonymous, two synonymous, and one non-coding) (Table 1).

Table 1.  Susceptibility genes for Crohn's disease
Susceptibility genesNo. of previous studiesap-values in meta-analysisbSNP IDs genotyped in this studySNP position in the geneSNP typeAmino acid changes
  1. aThe numbers are corresponding to previous GWAS referred in this study: [1]Hugot et al. (2001); [2]Ogura et al. (2001); [3]Peltekova et al. (2004); [4]Yamazaki et al. (2005); [5]Duerr et al. (2006); [6]Hampe et al. (2007); [7]Libioulle et al. (2007); [8]Parkes et al. (2007); [9]Raelson et al. (2007); [10]Rioux et al. (2007); [11]The Wellcome Trust Case Control Consortium (2007); [12]Kugathasan et al. (2008).

  2. bThe p-values show the significant level of association with CD reffered from Barret et al. (2008).

NOD2[1], [2], [5], [6], [9], [10], [11], and [12]4.62E-08 ∼ 1.49E-24rs2066844Exon 4NonsynonymousArg > Trp
  rs2066845Exon 7NonsynonymousGly > Arg
  rs2066847Exon 11Insetion (C allele)frame shift
IL23R[5], [7], [9], [10], [11], and [12]1.82E-39 ∼ 2.15E-68rs11209026Exon 9NonsynonymousAlg > Gln
ATG16L1[6], [9], [11], and [12]4.61E-25 ∼ 1.18E-32rs2241880Exon 8NonsynonymousThr > Ala
TNFSF15[4], [11], and [12]3.67E-09 ∼ 1.3E-10rs3810936Exon 4Synonymous-
SLC22A4[3], [6], [8], [10], and [11]1.54E-06 ∼ 1.16E-18rs1050152Exon 9NonsynonymousLeu > Phe
IRGM[8] and [11]7.33E-16 ∼ 1.7E-16rs10065172Exon 1Synonymous-
10q21[9] and [11]2.23E-20rs10761659Intergenic region -

PCR primers were designed to amplify 400–1500 bp products that encompassed one of the targeted SNP sites (Table S1). All of the PCR products were purified by precipitation using 30% polyethylene glycol 6000 and the Montage-PCR cleanup kit (Millipore, Tokyo, Japan). DNA sequencing was performed using the BigDye terminator cycle-sequencing kit version 3.1 according to the standard protocol (Applied Biosystems, Tokyo, Japan). The samples were then purified by ethanol precipitation with 3.4 mM EDTA (pH 8.1)/81.1 mM sodium acetate, and subjected to DNA sequencing using the PCR primers included with the ABI 3130 or ABI 3730 DNA Analyzer (Applied Biosystems, Tokyo, Japan).

Statistical Analysis

The SNP genotyping data for the control samples were analysed for deviation from Hardy-Weinberg equilibrium before being included in the subsequent analysis (χ2 test, p > 0.05). Association of allelic states with CD or UC was examined by the χ2 test (the degree of freedom (df) = 1), and the probabilities (p-values) were corrected by 1000 random permutations for each SNP implemented in the program using PLINK version 1.05 (http://pngu.mgh.harvard.edu/purcell/plink/) (Purcell et al., 2007). In addition to our Kyushu subjects, the previously reported data from the Honshu Japanese studies (Yamazaki et al., 2002, 2004, 2005, 2007, 2009) were included in the analysis. The difference in allele frequency between the Kyushu and Honshu samples was examined by the χ2 tests (df = 1).

The genotypic and phenotypic associations of CD and UC were examined using χ2 tests with permutation correction (p-values) by PLINK 1.05 (Purcell et al., 2007). The genotype, recessive, and dominant tests were also performed. The genotype test provides a general assessment of association in the 2 × 3 table of disease-by-genotype (df = 2). The recessive test detects associations of the homozygous risk genotype (“AA”) with the indicated disease (df = 1). The dominant test detects association of the heterozygous and homozygous risk genotypes (“Aa” and “AA”) with the indicated disease (df = 1).

Modified Genotype Relative risk (mGRR) Test

The genotype relative risk (GRR) test (Schaid & Sommer, 1993) was conducted for each SNP site that was found to be significantly associated with CD or UC. This test was adopted to determine whether the dominant, recessive, or additive model best fit the observed genotype frequency using the likelihood ratio test. Based on the likelihood method, we modified the GRR test to include unrelated individuals and also low penetrance variants.

Let A be the disease-associated allele, and let a be the non disease-associated allele. We then defined the probability of disease, conditional on the genotype at the particular SNP site as:


where D is the event that an individual has the disease. The multifactorial disease allele is retained not only in cases, but also in controls. We next defined the probability of non-disease, conditional on the genotype at the particular SNP site as:


Assuming that a population is in Hardy-Weinberg equilibrium, by the Bayes’ rule we have:




p is the population frequency of A, q is the frequency of a, and N is the non-disease state. Note that the likelihood depends on four independent parameters, ψ2case, ψ1case, p, f0case. A standard numerical maximization procedure can then be used to find maximum-likelihood estimates of ψ2case, ψ1case, p, f0case with the likelihood


where nAAcase, nAacase, naacase or nAAcontrol, nAacontrol, naacontrol are the observed numbers of cases or controls exhibiting each genotype.

We considered the following four hypotheses:

  • (I) No association of the marker with disease, H01case2case= 1,
  • (II) Dominant disease expression, HD1case2case,
  • (III) Recessive disease expression, HR1case= 1,
  • (IV) Additive disease expression, HA1case=ψ, ψ2case= 2ψ.

The maximum likelihood estimate can be found by maximizing the likelihood function under each hypothesis with the condition: 0 < p < 1, 1 < ψ1case, 1 < ψ2case, and 0 < f0case < 1. As described by Schaid & Sommer (1993), we next adopted the likelihood-ratio test (LRT). The LRT statistics for testing the hypothesis of no association is:


the statistics for testing the hypothesis of a dominant model are:


the statistics for testing the hypothesis of a recessive model are:


and the statistics for testing the hypothesis of an additive model are:


Based on these statistics, we determined which model best explained the observed data.

A Simple Moment Estimator for mGRR

To calculate mGRR based on the prevalence (K) of CD and UC in the Japanese population, we developed the moment estimator. Therefore,


where E[ni] (E[nNi]) is the expected number of affected (non-disease) individuals with i risk alleles (i= 0, 1, 2) and ncase=n0+n1+n2, ncontrol=nN0+nN1+nN2. Using the prevalence (K=f0caseRcase), where the prevalence rates of CD and UC in the Japanese population are 2.12 × 10−4 and 6.36 × 10−4, according to Asakura et al., (2009), the moment estimators satisfy,


Consequently, we define:



Associations of Susceptible Alleles with CD and UC

The samples acquired from each of the Kyushu Japanese subjects were first analysed to determine the genotypes of the nine SNPs, which were previously identified from the CD-associated genomic regions. These samples were also examined to determine any association of CD or UC with the risk alleles that had been identified in Europeans. The overall success rate of genotyping was greater than 98.6%. Table 2 illustrates the genotype data as well as the allele frequencies. The p-values calculated from the χ2 test and permutation test are shown in Table 3. In the Kyushu Japanese subjects, the risk alleles for NOD2 (rs2066844, rs2066845, rs2066847) and SLC22A4 (rs1050152) were not found to be present. Furthermore, the sequence data did not demonstrate any other variants around these SNP sites. However, analysis of the IL23R (rs11209026) gene showed that the risk allele was fixed in the two cases and the control. In subsequent analyses, SNPs with a fixed allele were excluded. Therefore, these SNPs are not polymorphic in the Japanese population, as reported in previous studies (Yamazaki et al., 2002, 2004, 2007). However, the remaining four SNP sites (rs2241880, rs3810936, rs10065172, and rs10761659) were shown to be polymorphic. Furthermore, three of the SNPs including rs2241880 in ATG16L1, rs10065172 in IRGM, and rs10761659 in 10q21 did not show significant association with either CD or UC (Table 3). The only SNP site found to be significantly associated with CD (corrected p-value = 0.047) and UC (corrected p-value = 0.050) was TNFSF15 (rs3810936). The odds ratio (OR) of the risk allele (“G” allele in rs3810936) was 1.551 (95% CI: 1.090–2.207) for CD and 1.692 (95% CI: 1.117–2.562) for UC (Table 3). Thus, the association test showed that TNFSF15 is the susceptibility gene for both UC and CD in the Kyushu Japanese subjects.

Table 2.  Number of genotypes and alleles in the Kyushu subjects
Susceptibility genesSNP IDsAllelesRisk allele (A)aNumber of genotypesb (Genotype frequency,%)cNumber of chromosomes (Allele frequency,%)c
CDUCControlsCD A/aUC A/aControls A/a
  1. aAssociated alleles were referred from the previous studies and annotated as “A”.

  2. bThe genotypes were shown as “AA” (homozygote of risk alleles), “Aa” (heterozygote of risk and non-risk alleles), and “aa” (homozygote of non-risk alleles).

  3. cGenotype or allele frequency (percentage) is given in parentheses.

   (0)(0)(100) (0)(0)(100) (0)(0)(100) (0/100)(0/100)(0/100)
   (0)(0)(100) (0)(0)(100) (0)(0)(100) (0/100)(0/100)(0/100)
rs2066847-/CC (insertion)00130130008282001631630/2600/1640/326
   (0)(0)(100) (0)(0)(100) (0)(0)(100) (0/100)(0/100)(0/100)
   (100)(0)(0) (100)(0)(0) (100)(0)(0) (100/0)(100/0)(100/0)
   (8)(40)(53) (7)(32)(61) (7)(39)(55) (28/72)(23/77)(26/74)
   (49)(47)(4) (57)(34)(9) (39)(49)(12) (73/27)(74/26)(63/37)
   (0)(0)(100) (0)(0)(100) (0)(0)(100) (0/100)(0/100)(0/100)
   (36)(47)(16) (40)(44)(16) (37)(51)(12) (60/40)(62/38)(63/38)
   (54)(36)(10) (46)(44)(10) (47)(42)(10) (72/28)(68/32)(68/32)
Table 3.  Associations of alleles and genotypes with Crohn's disease or ulcerative colitis in the Kyushu subjects
Susceptibility genesSNP IDsAllele (A/a) p-values (corrected p-values)aOdds ratio (95% CI)Genotype (AA/Aa/aa) p-values (corrected p-values)aRecessive (AA vs others) p-values (corrected p-values)aOdds ratio (95% CI)Dominant (AA+Aa vs others) p-values (corrected p-values)aOdds ratio (95% CI)
CD vs ControlUC vs ControlCD vs ControlUC vs ControlCD vs ControlUC vs ControlCD vs ControlUC vs ControlCD vs ControlUC vs ControlCD vs ControlUC vs ControlCD vs ControlUC vs Control
  1. aCorrected p-values were calculated by permutation test after performing 1000 random permutations for multiple testing adjustment.

  2. *p-values < 0.05.

  3. **p-values < 0.01.

 (0.986)0.948(0.744 – 1.557)(0.551 – 1.327)(0.999)(0.933)(0.997)(0.767)(0.477 – 2.828)(0.388 – 3.068)(1.000)(0.994)(0.678 – 1.716)(0.448 – 1.322)
 (0.047*)(0.050*)(1.090 – 2.207)(1.117 – 2.562)(0.064)(0.083)(0.232)(0.019*)(0.966 – 2.454)(1.247 – 3.644)(0.025*)(0.744)(1.341 – 9.117)(0.608 – 3.693)
 (0.921)1.000(0.634 – 1.237)(0.670 – 1.454)(0.942)(0.920)(0.999)(0.974)(0.592 – 1.543)(0.660 – 1.960)(0.710)(0.826)(0.355 – 1.311)(0.347 – 1.568)
 (0.849)(1.000)(0.818 – 1.674)(0.664 – 1.490)(0.957)(1.000)(0.741)(0.999)(0.808 – 2.041)(0.566 – 1.644)(1.000)(0.994)(0.484 – 2.229)(0.443 – 2.616)

In order to compare the differences in allele frequency between the Kyushu and Honshu subjects, we performed a χ2 test that included the Kyushu (K-) CD, Honshu (H-) CD, and each control (2-by-2 pairs: K-CD and H-CD, K-controls and H-controls, K-CD and H-controls, K-controls and H-CD) (Fig. 1 and Table S2). Of the eight SNPs genotyped in previous studies using the Honshu subjects, seven did not demonstrate a significant difference in allele frequency between any pairwise comparisons. Meanwhile, a significant association between TNFSF15 (rs3810936) and CD was detected in all pairs of CD/controls from both the Kyushu and Honshu subjects (Fig. 1 and Table S2).

Figure 1.

The distributions of risk allele frequency for three SNPs (rs2241880, rs3810936, rs10761659) in the Kyushu and the Honshu subjects are constructed from the values shown in Table 3 and Table S2. The abbreviations, K-CD, K-controls, H-CD, and H-controls indicate Kyushu CD, Kyushu controls, Honshu CD, and Honshu controls. Asterisks above the columns indicate a significant allele frequency difference between cases and controls in the same region (p-values for the Kyushu subject are referred from Table 3 and those for the Honshu subject are referred from Yamazaki et al. 2005). Asterisks above the brackets indicate a significant difference between Kyushu CD and Honshu controls, or Honshu CD and Kyushu controls (see Table S2). A single asterisk denotes p < 0.05, and double asterisks denote p < 0.01.

Genotype Associations with CD and UC

We next examined the genotype association of the four polymorphic SNP sites with either CD or UC. No significant association of ATG16L1 (rs2241880), IRGM (rs10065172), or 10q21 (rs10761659) was observed with either CD or UC. However, a significant association was detected between TNFSF15 (rs3810936) and UC (p= 0.021), as well as between TNFSF15 (rs3810936) and CD (p= 0.019). However, this significance was not confirmed by the 1000 permutation test (UC: p= 0.083, CD: p= 0.064) (Table 3). In order to clarify this association, we separated the genotypes into two classes according to the recessive model (i.e. AA vs. Aa+aa) and the dominant model (i.e. AA+Aa vs. aa) (Table 3). In the recessive model, the permutation tests showed significant association of the homozygote for the risk allele with UC (corrected p-value = 0.019). The OR of the risk allele homozygote was determined to be 2.132 (95% CI: 1.247 – 3.644). In the dominant model, both the heterozygote and the homozygote for the risk allele were found to be significantly associated with CD (corrected p-value = 0.025 and OR = 3.497, 95% CI: 1.341 – 9.117).

mGRR Test for CD and UC

In order to determine whether the dominant or recessive model best fit the data, an mGRR test was performed (Table 4). Four parameters including ψ2case, ψ1case, pcase, and f0case were inferred from the general likelihood equation so that the likelihood ratio test could be applied (7). Based on these hypotheses, the CD likelihood ratio test rejected the null (LRT p-value = 0.005), recessive (LRT p-value = 0.024) and additive models (LRT p-value = 0.021). However, the dominant model was not rejected (LRT p-value = 0.361). In contrast, the UC likelihood ratio test was not found to reject any of the models, most likely due to a lack of power. Nevertheless, the recessive model demonstrated a better fit with the observed genotype frequency than did the dominant or additive models (LRT p-values are not assessed because the same LogL value is in both the recessive and full model). These results are also supported by Akaike information criterion (AIC) values, which show the minimum values of the CD dominant model (AIC = 540.562) and the UC recessive model (AIC = 470.516). In addition, the moment estimators of the mGRR values were calculated to be ψ2case= 4.063 and ψ1case= 3.050 for CD, and ψ2case= 2.131 and ψ1case= 1.000 for UC. These mGRR values are very similar to the ψ2case and ψ1case values of the full model for each disease: moment estimates and likelihood estimates complementarily supported each other (Table 4). Thus, the statistical test for the genotype risk of rs3810936-G showed that the CD mGRR data best fits the dominant model, while the UC mGRR data best fits the recessive model.

Table 4.  Modified genotype relative risk (mGRR) test for Crohn's disease and ulcerative colitis in the Kyushu subjects
ModelCrohn's diseaseUlcerative colitis
LogLψ1caseψ2casepf0caseAICaLRT p-valuesLogLψ1caseψ2casepf0caseAICaLRT p-values
  1. aThe AIC is −2log L+ 2k, where k is the number of estimated parameters.

  2. bN.A. means “not assessed” (there is no LRT p-values, because LogL value in Recessive model is equal to that in Full model).

  3. *p-values < 0.05.

  4. **p-values < 0.01.

Null (no association)−272.188--0.670.881548.3760.005**−236.100--0.830.669476.2000.146
mGRR values from moment estimators 3.0504.06     1.0002.13    


We investigated whether the CD risk alleles found in Europeans are also present in the Japanese population. Analysis of the data has confirmed that none of these alleles, except for TNFSF15, are present and associated with CD in the Japanese population. Previous studies of the Honshu Japanese subjects have examined a greater number of SNPs, compared to our study, but did not detect an association between CD and the SNPs studied (Yamazaki et al., 2002, 2004, 2007, 2009). As an exception, Yamazaki et al. (2005) identified TNFSF15 as a CD susceptibility gene in the Honshu subjects, due to significant association (p < 0.0001) between 20 SNP sites and CD. Using the Kyushu subjects, our study supports previous results demonstrating that the CD susceptibility genes identified in Europeans are not associated with CD in the Japanese population. These data also imply a positive association between CD and the TNFSF15 gene in the Japanese population (Fig. 1 and Table S2). Thus, these results strongly support TNFSF15 as a CD susceptibility gene in the Japanese population.

Our study is the first to report a significant association of the TNFSF15 risk allele with UC, as previously shown for CD in the Japanese population. The similarities and differences between CD and UC were thought to be important in understanding the pathogenesis of each disease (Dubois & van Heel, 2008). The results of our χ2 and mGRR tests clearly show that the genotype of the rs3810936-G allele in TNFSF15 exhibited a different effect in CD compared to UC (Tables 3 and 4). This result was also supported by the moment estimator of the mGRR values, which were calculated from the CD and UC prevalence rates acquired from the Japanese samples (Table 4). The risk of the rs3810936-G allele was determined to be comparable between CD and UC (1.551 fold in CD and 1.692 fold in UC) (Table 3). However, the genotype relative risk between CD and UC was found to be greatly different (the risk of “GA” or “GG”: 3.050 fold or 4.063 fold in CD; no risk or 2.131 fold in UC) (Table 4 and Supplementary Fig. 1). Thus, it is likely that the risk variant of TNFSF15 functions as a “dominant” allele in CD, whereas it functions as a “recessive” allele in UC. Since the sample size of our subjects is smaller than that used in previous studies, it is necessary to apply the mGRR test to a larger sample size of Japanese subjects, in order to confirm a difference in genotype-phenotype correlation between CD and UC.

Considering that CD and UC are not inherited in typical Mendelian fashion, it is possible that the dominant or recessive effect of the risk allele, as suggested by the mGRR test above, is shaped by multiple genetic and environmental factors that are involved in each disease mechanism. Recently, other investigators have argued that the reciprocal interaction between multiple genetic and environmental factors is important for multifactorial diseases (Emison et al., 2005). Our studies confirm that the CD risk alleles identified in Europeans are absent, and are not associated with CD in the Japanese population, except for TNFSF15. Ethnic group specific susceptibility is not a novel idea, and has been generally observed in common diseases (Altshuler et al., 2008). Furthermore, this study is also the first to report that the CD risk allele is associated with UC. However, a previous study found no association with UC at the same SNP in TNFSF15, using non-Kyushu Japanese subjects (Kakuta et al., 2006). These results imply that UC, as well as CD, may exhibit a population-specific susceptibility between the various Japanese subjects, similar to the susceptibility shown to exist between Japanese and European subjects. Using world-wide population samples, Myles et al. (2008) examined the allele frequencies of common diseases, including CD, and indicated the importance of geographic variation in disease-associated alleles (Myles et al., 2008). Specifically, allele frequencies were observed to commonly vary among human populations (Tishkoff et al., 1996; Oota et al., 2004; Tishkoff & Kidd, 2004; The International HapMap Consortium, 2005, 2007; Li et al., 2008), including the Japanese population (Yamaguchi-Kabata et al., 2008). Hence, allele frequencies fluctuate in each geographic region where environmental factors are different, and the population-specific susceptibility of multifactorial disease might be reflected by subdivisions among human populations. We speculate that population-specific genetic and environmental factors cause dominant or recessive genotype risks in each subpopulation. Therefore, it is necessary to investigate a greater number of local cases and controls in order to reveal mechanisms of multifactorial diseases.


We are very grateful to the anonymous donors of Fukuoka University and Chikushi Hospital for their vast cooperation with this study. We thank Dr. Kenneth K. Kidd, Dr. Jun Ohashi, and Dr. Ryosuke Kimura for their helpful comments on the earlier version of this manuscript, and also thank Dr. Amy Russel for technical advice (http://www.amyrussell.net/research.html). We are also thankful for the technical assistance in sequencing that was provided by Ms. Rie Nishizawa, Dr. Yutaka Suzuki, and Dr. Sumio Sugano. Finally, we would like to thank all of the anonymous reviewers for their useful comments and suggestions. This manuscript has been proofread by a professional English editor from BioMed Proofreading, LLC. This work was supported by a Grant-in-Aid for Scientific Research (A) from the Japan Society for the Promotion of Science (JSPS) (19207018) to SK, by a Grant-in-Aid for Scientific Research (C) from JSPS (19570226) to HO, by a Grant-in-Aid for Scientific Research (B) from JSPS (21370108) to HO, and by a Grant-in-Aid for JSPS Research fellow to SN. We have deposited the nucleotide sequences into the international DNA database, Genebank/EMBL/DDBJ. The accession numbers of these sequences are AB542699-AB542710.