|Home | About | Journals | Submit | Contact Us | Français|
Although the number of convincingly established genetic associations with systemic lupus erythematosus (SLE) has increased sharply over the last few years, refinement of these associations is required, and their potential roles in gene–gene interactions need to be further investigated. Recent genome-wide association studies (GWAS) in SLE have produced renewed interest in B cell/T cell responses and the NF-κB signaling pathway. The aim of this study was to search for possible gene–gene interactions based on identified single-nucleotide polymorphisms (SNPs), in using an approach based on the role of signaling pathways.
The SNPs in BLK, TNFSF4, TRAF1, TNFAIP3, and REL were replicated in order to evaluate genetic associations with SLE. TaqMan genotyping was conducted in 804 Chinese patients with SLE and 722 matched control subjects. A multiple logistic regression model was used to estimate the multiplicative interaction effect of the SNPs, and additive interactions were analyzed by 2 × 2 factorial designs. Data from a previously published GWAS conducted by the International Consortium on the Genetics of Systemic Lupus Erythematosus were derived for comparison and validation.
Single-marker analysis validated the association of BLK rs2736340 (P = 4.25 × 10–6) as well as TNFSF4 rs2205960 (P = 2.82 × 10–5) and TNFAIP3 rs5029939 (P = 1.92 × 10–3) with SLE susceptibility in Chinese. Multiplicative interaction analysis indicated that BLK had an interactive effect with TNFSF4 in Chinese patients with SLE (P = 6.57 × 10–4). Additive interaction analysis revealed interactions between TRAF1 and TNFAIP3 in both Chinese (P = 2.18 × 10–3) and Caucasians (P = 2.86 × 10–4). In addition, multiple tendencies toward interactions were observed, and an additive effect was observed as the number of risk genotypes increased.
The results of this study provide evidence of the possible gene–gene interactions of BLK, TNFSF4, TRAF1, TNFAIP3, and REL in SLE, which may represent a synergic effect of T cells and B cells through the NF-κB pathway in determining immunologic aberration.
Systemic lupus erythematosus (SLE) is an unusually heterogeneous disease; various combinations of 4 of the 11 clinical criteria are required for case classification. A high sibling risk ratio (ranging from 8 to 29), high heritability (>66%), and higher concordance rates between monozygotic twins (20–40%) relative to dizygotic twins and other full siblings (2–5%) all suggest that SLE has a complex genetic basis. The number of convincingly established genetic associations with SLE has increased sharply over the last few years, with at least 30 robustly associated loci contributing to disease risk. However, these variants jointly explain <15% of the lupus sibling risk ratio (1–8). Thus, refinement of these associations is required, and their potential roles in gene–gene or gene–environment interactions need to be further investigated.
The most compelling reason for concentrating on genes is to generate new hypotheses about disease mechanisms and pathogenesis. The pathogenic mechanisms of SLE are incompletely understood but are postulated to involve multiple diverse aspects of the dysregulated immune response system. Recent genome-wide association studies (GWAS) for SLE have identified new associations with genes in the B cell and T cell signaling pathways (i.e., BLK, BANK1, PTPN22, CTLA4, STAT4, and TNFSF4), producing renewed interest by investigators in the mediation of B cell and T cell responses (2,7,9–11). In terms of signal transmission, one prominent pathway that has been newly emphasized is the NF-κB signaling pathway, refining the perception of the role of NF-κB regulation in disease pathogenesis (“no longer an island, but a piece of a continent,” as was recently reemphasized in the fields of autoimmunity, infection, and cancer) (12–15). The identified elements that were involved in NF-κB regulation, such as TRAF1, TNFAIP3, REL, LYN, and TNIP1, were recently reported to be associated with autoimmune diseases. We thus hypothesized that aberrant activation of T cells and B cells may occur through some common cell signaling pathway, i.e., the NF-κB pathway, in SLE as well as in common autoimmunity (5,6,12,16–27).
To verify such speculation, we chose BLK, TNFSF4, TRAF1, TNFAIP3, and REL as candidates for having a genetic association with SLE (13). BLK is located downstream of the B cell receptor and is a member of the Src family. It has been reported that Src family protein tyrosine kinases play an essential role in the activation of NF-κB during the development of B cells. Interaction between TNFSF4 (also known as OX40 ligand, OX40L, and CD252) and tumor necrosis factor receptor superfamily 4 (TNFRSF4; also known as OX40 and CD134) has a dual effect in promoting T cell responses: it enhances the proliferation of effector T cells and concomitantly blocks the generation of inducible Treg cells (20,27). The intracellular regions of TNFRSF4 associate with TNFR-associated factors (TRAFs), thereby allowing activation of both the canonical and noncanonical NF-κB signaling pathways. TRAF1 is distinct from all other TRAF family members in that it lacks zinc-finger and RING-finger domains that are responsible for mediating downstream signaling directly. Thus, an important function of TRAF1 appears to be the regulation of receptor signaling mediated by other TRAFs. In the mouse, TNFAIP3 encodes a cytoplasmic zinc-finger protein known as A20, and A20 is a major negative regulator of TNF-induced NF-κB signaling pathways (13,15). REL encodes c-Rel, a transcription factor that is a member of the Rel/NF-κB family (p50, p52, RelA, RelB, and c-Rel).
We attempted to validate the association in previously identified single-nucleotide polymorphisms (SNPs) of those candidate genes with SLE and search among them for gene–gene interactions that might account for some of the missing heritability in SLE (Figure 1). Of note, because identifying such interaction effects may require quite a large sample size to guarantee the power, here we do not exclude other genes of interest but rather propose some important genes located from upstream to downstream of the pathway as candidates to verify whether such interaction effects occur.
A total of 1,526 Chinese Han individuals from Beijing were enrolled in the current case–control study. The mean ± SD age of the 804 patients with SLE was 33.6 ± 12.3 years, and 89.1% were women. All of the patients with SLE met the American College of Rheumatology (ACR) revised criteria for the classification of SLE (28,29). The 722 healthy control subjects were matched to the patients geographically and ethnically. The mean ± SD age of the control subjects was 34.6 ± 10.3 years, and 57.0% were women. The study was approved by the medical ethics committee of Peking University. All participants provided informed consent.
SNPs rs2736340 and rs13277113 for BLK (2,5,8,30,31), rs2205960 and rs10489265 for TNFSF4 (1,6,32), rs10818488 for TRAF1 (16,17,23), rs5029939 for TNFAIP3 (24,26), and rs13031237 for REL (18,25), which previously were shown to be most significantly associated with SLE or multiple autoimmune diseases in different studies, were selected for our case–control study in the Chinese population.
In order to compare the current data with previously published data and to validate the current association with that from a second population, data were derived from a GWAS conducted in 720 female patients with SLE and 2,337 female control subjects of European ancestry by the International Consortium for the Genetics of Systemic Lupus Erythematosus (SLEGEN) (4). Because the selected SNPs were somewhat different, both the exact SNP and the SNP that was in high linkage disequilibrium with the selected SNP were chosen for considering the gene–gene interaction. Thus, 7 SNPs including rs2736340 for BLK, rs2205960 and rs10489265 for TNFSF4, rs1468673 (between rs10818488 and rs1468673; D′ = 0.83, r2 = 0.67) for TRAF1, rs5029942 (between rs5029942 and rs5029939; D′ = 1, r2 = 0.84) for TNFAIP3, and rs13031237 and rs6706689 (between rs13031237 and rs6706689; D′ = 1, r2 = 0.40) for REL were investigated.
TaqMan allele discrimination assays (Applied Biosystems) were used according to the manufacturer's instructions to determine the genotypes. Fluorescence was detected using an ABI Prism 7500 Sequence Detection System (Applied Biosystems).
The genotype frequencies of the SNPs were tested for Hardy-Weinberg equilibrium separately in patients and control subjects. Disease associations were analyzed by chi-square tests or by logistic regression analysis. Statistical power was estimated using Power and Sample Size Estimation Software (http://biostat.mc.vanderbilt.edu/PowerSampleSize).
The multiplicative interaction effect of the SNPs was estimated using a multiple logistic regression model. For each individual, key variables were defined as a binary variable indicating case–control status, with SNP variables ranging from 0 to 2 indicating the number of risk alleles in an individual subject. For each SNP pair, a logistic regression model was built to predict case–control status (dependent variable) based on the indicator variables (sex and age) and the 2 SNP variables (independent variable), for a total of 4 variables and an intercept. We tested whether the log likelihood of the model was significantly improved by adding an additional multiplicative pairwise interaction term for those 2 SNPs (5,33).
To test for additive interactions, direct counting and chi-square tests were performed using a 2 × 2 factorial design to calculate the attributable proportion due to interaction (AP) and the relative excess risk due to interaction (RERI) (34,35). Fisher's exact test was used when necessary.
These measures are defined as follows: RERI = RR11 – RR10 – RR01 + 1 and AP = RERI/RR11, where the referent group = group with G(–)/G(–), in which G(–) is the no risk genotype, and RR11 = risk of disease in G(+)/G(+) group/risk in referent group, RR10 = risk of disease in G(+)/G(–) group/risk in referent group, and RR01 = risk of disease in G(–)/G(+) group/risk in referent group (RR indicates relative risk, and G indicates genotype). If there is no biologic interaction, RERI and AP are equal to 0.
Statistical analyses were performed with SPSS12.0 software. Two-tailed P values less than 0.05 were considered significant. For multiple comparisons, Bonferroni adjustment was used.
A total of 7 SNPs in 5 genes involved in the NF-κB pathway, including BLK (rs2736340 and rs13277113), TNFSF4 (rs2205960 and rs10489265), TRAF1 (rs10818488), TNFAIP3 (rs5029939), and REL (rs13031237) were genotyped in 1,526 Chinese Han individuals, including 804 patients with SLE and 722 healthy control subjects. Deviation from Hardy-Weinberg equilibrium was not observed for any of the SNPs in patients with SLE or controls (P > 0.05).
In the current study, the power to detect a 1.5-fold increased risk, assuming an alpha value of 0.05, was >0.999 for rs2736340, rs13277113, rs2205960, rs10489265, and rs10818488, 0.778 for rs5029939, and 0.384 for rs13031237. As shown in Table 1, SNPs within BLK (rs2736340 [P = 4.25 × 10–6, odds ratio (OR) 1.48, 95% confidence interval (95% CI) 1.25–1.75], rs13277113 [P = 7.34 × 10–6, OR 1.45, 95% CI 1.23– 1.71]), TNFSF4 (rs2205960 [P = 2.82 × 10–5, OR 1.39, 95% CI 1.19–1.62], rs10489265 [P = 3.10 × 10–4, OR 1.33, 95% CI 1.14–1.55]), and TNFAIP3 (rs5029939 [P = 1.92 × 10–3, OR 1.58, 95% CI 1.18–2.12]) were associated with SLE at both the allele and genotype levels. This result was compatible with the findings in previous GWAS of SLE, implying a real genetic effect among different populations.
Compared with the SLEGEN GWAS, similar significance as well as ORs can be drawn, although great genetic heterogeneity between the Chinese Han population and European Caucasians was observed. Specifically, the risk allele frequency for BLK SNPs was much higher (72.4% versus 22.9%), and the minor allele frequency for REL SNPs was much lower (<5% versus >30%) in Chinese compared with Caucasians (Table 1), which was consistent with data from the HapMap project (www.hapmap.org) (additional information is available in Supplementary Table 1, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131).
To identify independent risk susceptibility factors as well as to better interpret the possible model of inheritance (36,37) in the multivariate analysis, P values for each SNP under the recessive, codominant, or dominant model were calculated using logistic regression analysis controlled for the 7 SNPs, sex, and age (Table 2). In the Chinese population, the dominant model showed the best fit for BLK rs2736340 (OR 2.78, 95% CI 1.12–6.91, P = 0.03), TNFSF4 rs2205960 (OR 1.69, 95% CI 1.01–2.83, P = 0.05), and TNFAIP3 rs5029939 (OR 1.61, 95% CI 1.15–2.26, P = 5.52 × 10–3). In addition, the codominant model also showed significant associations of TNFAIP3 rs5029939 (OR 1.48, 95% CI 1.05–2.08, P = 0.03) and susceptibility to SLE.
In comparison, in the Caucasian population, the dominant model showed better fit for TNFAIP3 rs5029942 (OR 1.79, 95% CI 1.30–2.47, P = 3.75 × 10–4) and REL rs13031237 (OR 1.34, 95% CI 1.10–1.64, P = 4.64 × 10–3), whereas the recessive model showed better fit for BLK rs2736340 (OR 2.09, 95% CI 1.45– 3.02, P = 8.26 × 10–5) and TRAF1 rs1468673 (OR 1.35, 95% CI 1.06–1.73, P = 0.02).
We tested whether the log likelihood of the logistic model was significantly improved by adding an additional multiplicative pairwise interaction term for the combined 2 SNPs. A total of 21 tests (including different combinations of 7 SNPs, variables including sex and age, any 2 SNPs and their SNP/SNP combination, and an intercept in the model) were conducted in the Chinese population. An interaction term was considered significant only if the P value was less than 2.38 × 10–3 (i.e., 0.05/21). For pairwise tests, we observed that the addition of an interaction test significantly improved the log likelihood in Chinese individuals, especially the interaction between BLK and TNFSF4 (P = 6.57 × 10–4) (Table 3). In comparison, when the same analysis was applied to the SLEGEN GWAS data, no such significant interaction was observed in Caucasians. However, a consistent tendency toward a gene–gene interaction between BLK and TNFAIP3 (P < 0.05), TNFSF4 and TNFAIP3 (P ≤ 0.05), and TNFSF4 and TRAF1 (for rs10818488 × rs2205960, P = 0.05 in Chinese; for rs10818488 × rs10489265, P = 0.08 in Chinese [versus P = 0.10 in Caucasians]) was observed in both Chinese and Caucasians. In addition, a tendency toward interaction (P < 0.05) between TRAF1 and TNFAIP3 in the Chinese population and interactions between TRAF1 and REL and between TNFAIP3 and REL in Caucasians could be observed in a separate data set.
In the current Chinese population, to reduce complexity, we chose rs2736340 and rs2205960 to represent BLK and TNFSF4, respectively, because these SNPs were selected in both studies and showed the better fit from multivariate analysis. Because the numbers for some genotypes were small, and in most cases dominant models showed the better fit for association in both the Chinese and Caucasian populations, we considered the dominant model in a further interaction analysis stage. The hypothesis of “additive interaction” was that the combined effect of 2 risk factors would differ from the sum of the effects of the individual factors.
In the current cohort, differences in risk geno-type counts between patients and controls were high, which was particularly significant when risk genotypes were combined (additional information is available in Supplementary Table 2, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). A significant additive interaction was observed between TRAF1 rs10818488 GA + AA and TNFAIP3 rs5029939 CG + GG (OR 1.84, 95% CI 1.24–2.73 [P = 2.18 × 10–3]); the risk genotype combination contributed the most to the overall interaction, with the remaining combinations within being nonsignificant. The AP was 0.46, and the RERI was 0.85. A similar tendency was observed between TNFSF4 rs2205960 GT + TT and TRAF1 rs10818488 GA + AA (OR 1.43, 95% CI 1.04–1.97 [P = 0.03]).
For the SLEGEN GWAS data set, when the same analysis procedure was employed, the same additive interaction was observed between TRAF1 rs1468673 TC + CC and TNFAIP3 rs5029942 TG + GG (OR 2.09, 95% CI 1.40–3.13 [P = 2.86 × 10–4]; AP 0.34, RERI 0.71) as well as the same tendency between TNFSF4 rs2205960 GT + TT and TRAF1 rs1468673 TC + CC (OR 1.29, 95% CI 1.01–1.65 [P = 0.04]). In addition, a significant additive interaction was observed between BLK and TNFAIP3 (OR 2.97, 95% CI 1.85–4.76 [P = 2.78 × 10–6]; AP 0.48, RERI 1.44).
Furthermore, assuming these genes were members of the same pathway and assuming a possible additive interaction as well as a joint effect, the disease risk was analyzed by studying the effect of carrying multiple risk factors. As the number of risk genotypes increased, the relative risk of susceptibility to SLE increased. For example, using a 2-way–based combination of the above data, the OR for TNFAIP3 was ~0.3-fold higher in the TRAF1 risk genotype group compared with the non–risk genotype group in both populations; a 3-way–based combination of BLK, TNFSF4, and TNFAIP3 indicated a much higher risk than that of any 2-way–based combination (P = 2.76 × 10–6).
Taking all of the genes into consideration, under the dominant model, compared with carriers of 0 or 1 risk factor, the ORs for SLE were 1.13 (95% CI 0.82– 1.54 [P = 0.47]) for carriers of 2 risk factors, 1.61 (95% CI 1.17–2.22 [P = 3.44 × 10–3]) for carriers of 4 risk factors, and 2.456 (95% CI 1.46–4.04 [P = 5.77 × 10–4]) for carriers of 5 risk factors (for the sake of model stability, carriers of 0 and 1 risk factors and carriers of 4 and 5 risk factors were combined). In Caucasians, compared with carriers of 0 or 1 risk factor, the ORs for SLE were 1.43 (95% CI 1.12–1.82 [P = 3.87 × 10–3]) for carriers of 2 risk factors, 1.64 (95% CI 1.27–2.12 [P = 1.38 × 10–4]) for carriers of 4 risk factors, and 2.16 (95% CI 1.51–3.08 [P = 2.02 × 10–5]) for carriers of 5 risk factors. Linear regression analysis showed a multiplicative effect of the risk factors.
In Figure 2, the ORs are shown as a function of number of SLE risk factors. The slope of the line corresponds to a ~1.2-fold (1.24-fold for Chinese and 1.19-fold for Caucasians) increase in the OR for each additional risk factor (38). Finally, the lack of association of TRAF1 and REL in single-marker association was further revealed in the pathway-based study.
In the past few years, tremendous success has been achieved in the identification of SLE susceptibility genes, especially from GWAS. It is believed that an increased understanding of genetics provides the opportunity to investigate potential new therapeutic strategies and to improve diagnostic and prognostic tests for the disease. However, excited optimism gradually yields to sober reflection regarding the fact that genetic variants account for just a small proportion of heritability. Therefore, it is likely that many more remain to be discovered. Among the many components of “missing” heritability, gene–gene interaction is believed to be an important resource (39). To date, gene–gene interactions between HLA–DRB1 and PTPN22 and between TNFRSF14 and TNFRSF6B have been well documented, demonstrating how these genes might work along the same biologic pathway in determining overall susceptibility to rheumatoid arthritis (RA) (33–35,40). However, little evidence for interactions determining SLE susceptibility was documented (38). To integrate the genes that have already been identified, especially those involved in the same biologic pathway (41), the investigation of gene–gene interaction surely will expand our current understanding of genetics and the pathogenesis of SLE or will broadly extend to other diseases regarding shared genetics.
The current study was based on the hypothesis that aberrant T cell/B cell activation occurs through a common cell signaling pathway in SLE. We chose 5 genes that may be key components in this pathway as candidates for replication as well as for gene–gene interaction analysis. BLK is a downstream molecule of the B cell receptor, but its cellular functions are poorly defined (42,43); TNFSF4 provides a late-stage costimulatory signal to sustain the survival of activated T cells (11,27); and TRAF1, TNFAIP3, and REL are critical molecules of the NF-κB pathway, from upstream to downstream (13,15). It was difficult to uncover the susceptibility genes or responsible variants within the associated regions provided by a GWAS (44), and the most significantly associated SNPs were somewhat different among different studies, even in the same population. Therefore, for the case–control study, we selected only those SNPs previously associated with SLE or multiple autoimmune diseases that had the strongest significance in different studies. In a new cohort of individuals from the North of China that was different from the previously reported cohort of Southern Chinese Han individuals, we observed similar results, with similar odds ratios at the single-gene level. The differences in genetic models involving Chinese and Caucasians may be attributable to genetic heterogeneity caused by varied allele frequencies but may also be attributable to noise (i.e., differences between prevalence that are not specific to genetic heterogeneity, such as heterogeneity in the distribution of environmental exposures, age distribution differences between populations, or different classification criteria for case assessments), which we did not address in much detail.
In spite of the recent advances in statistical approaches, investigating genetic interactions has proven difficult. With no “best” statistical approach available, combining several analytical methods may be optimal for detecting epistatic interactions. Gene–gene interactions can be assessed with either additive or multiplicative mathematical models. Additive interactions suggest that the combined effect of 2 factors differs from the sum of the effects of the individual factors, whereas multiplicative interactions suggest synergistic effects greater than those predicted by a model that multiplies the individual effects (45).
Here, we used 2 classic but quite strict models to detect possible gene–gene interactions. In multiplicative interaction effect analysis, we observed significant interactions between BLK and TNFSF4 and a tendency toward interactions between BLK and TNFAIP3, between TNFSF4 and TNFAIP3, and between TNFSF4 and TRAF1 in the Chinese population. When such an analysis was applied to a previously published GWAS data set from a Caucasian population, the same tendency of interactions between BLK and TNFAIP3 and between TNFSF4 and TNFAIP3 was observed. In addition, tendencies toward interactions between TRAF1 and REL and between TNFAIP3 and REL were observed. The different result may be derived from genetic heterogeneity, because great variations in allele frequency were observed with regard to BLK and REL. Genes with low minor allele frequencies would be less sensitive to observations of a positive gene–gene interaction. In addition, SLE that is more severe due to lupus nephritis is more prominent in the Chinese population than in Caucasians; i.e., patients with lupus nephritis account for 68.2% of the current Chinese SLE cohort (46).
To validate the interactions observed in Chinese individuals, we further analyzed SLEGEN GWAS data using multifactor dimensionality reduction software (MDR; http://sourceforge.net/projects/mdr/), which was a nonparametric and genetic model–free alternative to logistic regression for detecting and characterizing nonlinear interactions. Among the 7 SNPs selected, in the 2-way model, interaction between BLK rs2736340 and TNFSF4 rs2205960 revealed the highest training-balanced accuracy (0.54; P = 0.0002). The average training-balanced accuracy across all cross-validation intervals was calculated as sensitivity plus specificity divided by 2. Because MDR software can hardly identify the exact interaction form, and because logistic regression is much stricter, we present only the results from logistic regression. An additive interaction analysis by chi-square test using a 2 × 2 factorial design showed a significant additive interaction between TRAF1 and TNFAIP3, and a tendency toward interaction between BLK and TNFAIP3, both in the Chinese and Caucasian population.
We observed possible interactions between BLK and TNFSF4, BLK and TNFAIP3, TNFSF4 and TRAF1, TNFSF4 and TNFAIP3, TRAF1 and TNFAIP3, TRAF1 and REL, and TNFAIP3 and REL. Thus, the data supported our hypothesis that T cells and B cells cooperate in determining overall immunologic aberration and aberrant T cell/B cell activation through a common cell signaling pathway. The signaling pathway via TNFSF4 may quantitatively enhance B cell proliferation to augment the B cell hyperactivity observed in SLE. Dysregulation in B cell receptor and NF-κB signaling may alter the delicate immune balance and tolerance and predispose to autoimmunity. In addition, TNFSF4 (OX40L) interacts with its unique receptor OX40 to initiate dual regulations in T cell homeostasis.
Finally, the lack of association of TRAF1 and REL in single-marker associations can be further revealed in the pathway-based study. In addition, we observed that the risk of SLE increased with the increased number of risk genotypes. The above 2 phenomena thus can account for some of the missing heritability in GWAS or in simple candidate gene–based case–control studies.
Although the findings of interaction studies could provide insights into the etiology of SLE, it is still premature to apply the results to individual patients. Of note, the data provided from the present SLE study may extend to other autoimmune diseases. In accordance with the “common variants/multiple disease” hypothesis, the single locus–based genetic associations described above have been extended to many other autoimmune diseases, such as RA, multiple sclerosis, and type 1 diabetes mellitus, thus suggesting a shared genetic background in autoimmunity (13,42,47–50). Replication studies including larger cohorts with the same ethnicity and multiple diseases may be indicated. This explorative study represents an attempt to go beyond GWAS in clearly determining the pathogenesis of SLE from a multigene perspective using a gene–gene interaction approach.
We are grateful to the SLEGEN investigators, including John B. Harley, MD, PhD (Oklahoma City, OK), Marta E. Alarcón-Riquelme, MD, PhD (Uppsala, Sweden), Lindsey A. Criswell, MD (San Francisco, CA), Chaim O. Jacob, MD, PhD (Los Angeles, CA), Robert P. Kimberly, MD (Birmingham, AL), Kathy L. Moser, PhD (Minneapolis, MN), Betty P. Tsao, PhD (Los Angeles, CA), Timothy J. Vyse, PhD (London, UK), and Carl D. Langefeld, PhD (Winston-Salem, NC).
Supported by the National Natural Science Foundation of China (grants 30801022 and 30825021) and the Foundation of the Ministry of Health of China (grant 200802052).
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Zhang had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Zhou, Zhang.
Acquisition of data. Lu, Nath, Yang, Qin, Zhao, Su, Shen, Li.
Analysis and interpretation of data. Zhou, Lv, Zhu.