Author + information
- Received January 20, 2014
- Revision received March 8, 2014
- Accepted March 15, 2014
- Published online July 22, 2014.
- Guanglin Cui, MD, PhD,
- Zongzhe Li, MD,
- Rui Li, MD, PhD,
- Jin Huang, MD,
- Haoran Wang, MD,
- Lina Zhang, MD,
- Hu Ding, MD, PhD∗ ( and )
- Dao Wen Wang, MD, PhD∗ ()
- Institute of Hypertension and Department of Internal Medicine, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, People’s Republic of China
- ↵∗Reprint requests and correspondence:
Dr. Dao Wen Wang or Dr. Hu Ding, Departments of Internal Medicine and Institute of Hypertension, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, 1095# Jiefang Avenue, Wuhan 430030, People’s Republic of China.
Background Recent genome-wide association studies identified the APOA5/A4/C3/A1 gene cluster polymorphisms influencing triglyceride level and risk of coronary artery disease (CAD).
Objectives The purposes of this study were to fine-map triglyceride association signals in the APOA5/A4/C3/A1 gene cluster and then explore the clinical relevance in CAD and potential underlying mechanisms.
Methods We resequenced the APOA5/A4/C3/A1 gene cluster in 200 patients with extremely high triglyceride levels (≥10 mm/l) and 200 healthy control subjects who were ethnically matched and genotyped 20 genetic markers among 4,991 participants with Chinese Han ethnicity. Subsequently, 8 risk markers were investigated in 917 early-onset and 1,149 late-onset CAD patients, respectively. The molecular mechanism was explored.
Results By resequencing, a number of newly and potentially functional variants were identified, and both the common and rare variants have remarkable cumulative effects on hypertriglyceridemia risk. Of note, gene dosage of rs2266788 demonstrated a robust association with triglyceride level (p = 1.39 × 10−19), modified Gensini scores (p = 1.67 × 10−3), and numbers of vascular lesions in CAD patients (odds ratio: 1.96, 95% confidence interval: 1.31 to 2.14, p = 8.96 × 10−4). Functional study demonstrated that the rs2266788 C allele destroyed microRNA 3201 binding to the 3′ UTR of APOA5, resulting in prolonging the half-life of APOA5 messenger RNA and increasing its expression levels.
Conclusions Genetic variants in APOA5/A4/C3/A1 gene cluster play an important role in the regulation of plasma triglyceride levels by an increased APOA5 concentration and contribute to the severity of CAD.
Hypertriglyceridemia (HTG), a defining component of the metabolic syndrome (1), is associated with numerous comorbidities, including increased cardiovascular morbidity and mortality risk (2). Recently, a number of genome-wide association studies (GWAS) have successfully identiﬁed multiple genes that have been both known and novel loci associated with plasma triglyceride concentration (3–6). However, genetic variation at these loci explains only ∼10% of overall triglyceride variation within the population, suggesting that many determinants have yet to be clariﬁed (7). Of these genes, a well-known gene cluster, APOA5/A4/C3/A1, located at chromosome 11, had been reported to contribute 38% of the genetic variance of triglycerides, as well as explaining 11% of high-density lipoprotein-cholesterol (HDL-C), 4% of low-density lipoprotein (LDL-C), and 10% of apolipoprotein B genetic variance (8). Increasing evidence has shown that high triglyceride levels and single nucleotide polymorphisms (SNPs) in the APOA5/A4/C3/A1 gene cluster are associated with several risk factors that substantially increase the risk of coronary artery disease (CAD) (9–12). Indeed, a large-scale meta-analysis involving 86,995 individuals of European descent gives reliable evidence that this APOA5/A4/C3/A1 gene cluster is implicated in the pathogenesis of CAD (12). In addition, prospective epidemiological and triglyceride-mediated pathway studies also have found the positive association between circulating triglyceride concentration and the risk of CAD (10,13,14). Despite the evidence, the extent to which triglycerides directly promote cardiovascular disease or represents a biomarker of risk has been debated for 3 decades (15). Other remaining questions are whether and how additional forms of genetic variations, such as rare variants with large individual effects, could contribute to the heritability of a high plasma triglyceride status and whether the APOA5/A4/C3/A1 gene cluster is related to the severity of CAD, specifically in the Chinese Han population. Thus, to fine-map the actual susceptibility variants, we resequenced all the exons of the APOA5/A4/C3/A1 gene cluster in a total of 400 Chinese Han participants: 200 patients with an extremely high plasma triglyceride level (≥10 mmol/l) and another 200 healthy control subjects. We also conducted a comprehensive survey of 20 common SNPs for associations with triglyceride levels in an independent Chinese Han cohort and investigated the molecular mechanism for the SNP to contribute to a high triglyceride level. Finally, we tested the hypothesis that gene dosage of a risk allele in the APOA5/A4/C3/A1 gene cluster predicts CAD severity, which was angiographically documented and stratiﬁed by the degree of vascular lesions.
Study design and eligibility
Details on sample recruitment, inclusion criteria, data collection, and definition of risk factors are provided in the expanded Methods section in the Online Appendix. The clinical characteristics of the samples are shown in Table 1. In brief, the resequencing effort was conducted in 200 affected individuals and 200 controls of Chinese Han descent. Individuals with HTG were unrelated subjects with a diagnosis of Fredrickson HTG phenotypes (16), defined as having a fasting plasma triglyceride level >10 mmol/l, from a the Department of Clinical Chemistry, Tongji Hospital. Ethnically and geographically matched control subjects were recruited from individuals undergoing routine health examinations at the Tongji Hospital in Wuhan, Hubei Province. We chose controls with a maximal recorded fasting plasma triglyceride levels <2.3 mmol/l to exclude undiagnosed HTG (13). A total of 4,991 healthy, community-based Chinese Han individuals were originally recruited from the general population in 2004, with ages ranging from 20 to 85 years. At baseline, a detailed demographic and family history of all study participants was obtained via standardized questionnaires. None of the individuals were using lipid-lowering medication when the blood sample was taken. In fasting venous blood samples, we measured total cholesterol, HDL-C, triglycerides, and LDL-C on the Roche modular DPP system (Roche, Basel, Switzerland), according to standard procedures. The distribution of serum TG levels for control subjects and HTG patients are shown in Online Figure 1. The CAD cohort, including 917 early-onset patients and 1,149 late-onset patients, were enrolled simultaneously from hospitalized patients in the Tongji Hospital and the Institute of Hypertension (Wuhan, China) between May 2004 and October 2011. The 222 angiographic cases with <20% stenosis in any coronary artery were recruited and called 0-vessel disease including 106 early-onset patients and 116 late-onset patients. The modified Gensini scores and the number of lesions were previously reported (17,18). All patients and control subjects were carefully matched by geographic region of recruitment, were of Han Chinese ancestry, and provided written informed consent. This study was approved by the institutional ethics committees of the local participating hospitals.
Genetic variation screening
Sequence data of the APOA5/A4/C3/A1 gene cluster were generated by Sanger sequencing, whereas 3 other triglyceride-modulating genes (APOB, GCKR, LPL) were screened by Ion Torrent semiconductor sequencing (Life Technologies, Carlsbad, California). Details regarding primers, resequencing procedures, and data analysis are given in Online Table 1 and the Methods section in the Online Appendix.
Tagging SNP selection and genotyping
A total of 20 SNPs within the APOA5/A4/C3/A1 gene cluster were genotyped in our study. These SNPs cover a 150-kb genomic region on chromosome 11 (National Centre for Biotechnology Information build GRCh37 from 116165000 to 116215000). TaqMan 5′-nuclease assay on the TaqMan 7900HT Sequence Detection System (Applied Biosystems, Foster City, California) under standard conditions. Details regarding SNP selection as well as TaqMan performances can be found in Online Table 2.
The risk score was calculated on the basis of 8 SNPs that were significantly associated with plasma triglyceride levels in the combined analysis. We modeled the cumulative number of risk alleles carried by each participant. For a detailed description of the calculation methods, see the Methods section in the Online Appendix.
Bioinformatic analysis shows that rs2266788 is located at microRNA 3201 (miR-3201) binding site of human APOA5 3′ UTR, and therefore we tested whether rs2266788*C destroyed the miR-3201 binding site. Two 780-bp fragments containing human APOA5 3′ UTR rs2266788T and rs2266788C, respectively, were cloned into a p-MIR vector downstream of luciferase reporter (Ambion, Austin, Texas), and the resultant plasmids (p-MIR-T and p-MIR-C) were transformed into 293T and HepG2 cells, respectively, with or without miR-3201 to determine effects of the miR-3201 binding site by detecting fluorescence intensity according to the manufacturer’s instructions.
Measurement of genotype-dependent plasma levels of APOA5 was performed in 170 healthy participants to investigate effects of different rs2266788 genotypes on APOA5 levels.
Statistical analyses were performed with SPSS version 13.0 (SPSS, Inc., Chicago, Illinois) for Windows (Microsoft, Redmond, Washington). Haploview version 4.1 (Informer Technologies, Inc.) was used to calculate linkage disequilibrium (LD) and select tagging SNPs. Deviations of genotype frequency from the Hardy-Weinberg assumption were assessed using a chi-square test. All biostatistics calculations were performed using Prism (GraphPad Software, Inc., La Jolla, California). All probability values were 2 sided, and p < 0.05 was considered significant. Detailed descriptions are available in Methods section in the Online Appendix.
The triglyceride-modulating gene APOA5/A4/C3/A1 gene cluster, GCKR, LPL, and APOB were previously reported (14,19). A total of 119 kb (including 20 kb coding and 99 kb noncoding) were sequenced in 200 individuals with severe HTG and 200 control subjects of Chinese Han descent. Cumulatively, we identified 171 DNA variants with an average of 1 variant per 695 bp of the reference human genome sequence (Table 2). Of the variants, 60 were common polymorphisms (minor allele frequency >1%), whereas the remaining 111 were rare variants, including 29 noncoding, 17 synonymous, and 65 nonsynonymous variants. Bioinformatics analysis suggests that 20 of nonsynonymous variants are possibly or probably damaging normal function of corresponding genes (Online Table 3).
To test the hypothesis that these rare coding variant (minor allele frequency <0.02) HTG-associated genes are related to HTG disease cause, we conducted rigorous analyses of these variants found exclusively in either individuals with HTG or control subjects, and the results revealed a significant accumulation of variants in the APOA5/A4/C3/A1 gene cluster in affected individuals (p = 2.0 × 10−4) (Table 3), but not the other triglyceride-modulating genes (p = 0.225).
Given the significant accumulation effect of rare coding variants of the APOA5/A4/C3/A1 gene cluster, we further genotyped and analyzed the effect of 20 common variants in a region of a 150-kb DNA sequence of this gene cluster on triglyceride level in a total of 4,991 participants. No deviation from the Hardy-Weinberg equilibrium was observed for all SNPs in these samples (p > 0.05). Ten of the 20 SNPs within this region were statistically associated with plasma triglyceride levels with Bonferroni correction adjustment (correct p < 0.05) (see details in Online Table 4). It is worth noting that 2 of these 10 SNPs, rs2266788 and rs662799, were identified most significantly as being associated with plasma triglyceride levels in the Chinese Han population after adjusting for traditional risk factors (p = 1.39 × 10−19, p = 4.74 × 10−36, respectively) (Fig. 1⇓). For rs2266788 and rs662799, the proportion of a 1-SD change in standardized log-transformed plasma triglyceride level residual value for each copy of the risk allele was 0.131 and 0.178, respectively, and the contribution of this genetic effect can explain 7.1% and 8.5%, respectively, of the total variance of plasma triglyceride levels in our population. We also found partial polymorphisms in this gene cluster associated with HDL-C, LDL-C, and total plasma cholesterol (Online Table 5).
Given that 3 SNPs (rs2266788, rs662799, and rs964184) were in a strong LD (D′ range, 0.96 to 1.0 [Online Fig. 2], based on the HapMap project Phase II database), rs2266788 can capture rs964184 and rs662799. Therefore, 8 risk SNPs (rs2266788 and the remaining 7 risk SNPs) were selected for the risk score analysis. We divided the general population into tertiles by weighted allelic risk scores and found a significant association between the tertiles of accumulative risk scores and gradually increased triglyceride levels (p = 1.34 × 10−8) (Fig. 2A) and a reverse association with HDL-C (p = 2.37 × 10−3) (Fig. 2B), indicating that these SNPs have cumulative effect on plasma triglyceride and HDL-C levels.
We further assessed the relationship between these 8 risk SNPs and modified Gensini coronary scores and found that these risk SNPs have a cumulative association with the Gensini score (2.25 × 10−3) (Fig. 2C). In addition, 2 SNPs (rs2266788 and rs662799), which showed the greatest association with plasma triglyceride levels, also are associated with modified Gensini scores in combined analysis after multiple corrections (p = 0.003 and p = 1.67 × 10−3, respectively) (Fig. 2D).
Subsequently, we explored whether the gene dosage of these 2 SNPs (rs2266788 and rs662799) is associated with vascular lesions and contributed to CAD severity as well. Results showed that the higher gene dosage carriers of the rs2266788C had a significantly greater chance of having a 3-vessel coronary lesion on combined analysis (Table 4) (p = 8.96 × 10−4, odds ratio: 1.96, 95% confidence interval: 1.31 to 2.14). Similar results can be found in the early- and late-onset studies. However, no significant effect was observed for the rs662799 variant after multiple corrections.
As shown in Figure 3, CAD patients with the CC genotype (meaning both copies of the gene contain the C allele) of rs2266788 have a higher proportion of right coronary artery stenosis (p = 0.005, odds ratio: 1.77, 95% confidence interval: 1.19 to 2.65) (Fig. 3A), but has no effect on left main trunk disease (p = 0.316). With an increasing number of complex manifestations, the minor allele frequency of rs2266788 was significantly and gradually higher (p = 0.019) (Fig. 3C).
Functional analysis of rs2266788 and rs662799
Previous reports have shown that rs662799 had no effect on relative luciferase expression (20), and no transcriptional factors mapped to this site (21). Given that rs2266788 was in strong LD with rs662799 in the Han Chinese population (r2 = 0.92), maintaining important clinical relevance, we focused on rs2266788, which is located in the 3′ UTR of APOA5, for further functional analysis.
Further bioinformatics analysis by a computer alignment program demonstrated that the rs2266788 C allele, not the T allele, destroyed miR-3201 binding to the 3′ UTR of APOA5 (Fig. 4A). To test the prediction model that the miR-3201 can functionally interact with the 3′ UTR of APOA5, 293T and HepG2 cells were used for transfection of APOA5 3′ UTR–based reporter constructs (pMIR-T, with the T allele or pMIR-C, with the C allele). As shown in Figures 4B and 4C, compared with a negative control, luciferase activity decreased with cotransfection of pMIR-T and miR-3201 in both 293T cells (25 ± 1.21%, p = 0.006) and HepG2 cells (43 ± 1.23%, p = 0.01), respectively. On the contrary, miR-3201 had no effect on ﬁrefly luciferase activity when the pMIR-C construct destroying the predicted miR-3201 binding site was transfected (Figs. 4B and 4C). Furthermore, we performed messenger RNA (mRNA) turnover experiments to monitor the influence of rs2266788 on mRNA stability. The estimated half-life of the rs2266788 T transcript was ∼2 h shorter than the rs2266788 C construct (9.8 ± 0.5 vs. 13.7 ± 0.2, p < 0.001) (Online Fig. 3).
We further explored the effects of miR-3201 on APOA5 translation (e.g., protein expression in cultured HepG2 cells in vitro), Western blot results showed that overexpression miR-3201 downregulated APOA5 expression level and the inhibition of miR-3201 expression using its inhibitor significantly upregulated expression of APOA5 in HepG2 cells (Fig. 4D). These results were consistent with the real-time polymerase chain reaction data (Online Fig. 4). To further confirm the effects of miR-3201 and its inhibitor on endogenous APOA5 expression, we sequenced 9 of the human hepatoma cell lines (including HepG2, HuH-7, Hep3B, LM3, HLE, HLF 97L, PLC, L02, and so on), and, fortunately, the cell line HLF was identified to be a heterozygous genotype (CT type) (Online Fig. 5), and then we used this cell line to perform experiments. Results showed that no significant effect of miR-3201 on APOA5 expression was found in HLF cells (Online Fig. 6). Cotransfection of pcMV6-ApoA5 and miR-3201 in HepG2 cells efficiently restored APOA5 expression with enhanced expression of miR-3201 (Fig. 4E). In vivo, plasma concentrations of APOA5 were higher in subjects with the CC genotype (n = 25) than in those with the CT genotype (n = 47, p = 0.032) or the TT genotype (n = 98, p = 0.038) (Fig. 4F). These results are also consistent with the clinical and biochemical profiles of the CC, CT, and TT genotype subgroups (Online Table 6), suggesting that miR-3201 is truly causal with higher triglyceride levels in subjects carrying the C allele.
Taken together, these data indicate an interaction between miR-3201 and a speciﬁc region of the 3′ UTR of APOA5 containing the T, but not the C allele, which subsequently enhances degradation of its mRNA and reduces the protein expression level.
This population-based study on APOA5/A4/C3/A1 gene cluster variations provides clear evidence of associations with the development of HTG (Central Illustration). Based on our large-scale genetic fine-mapping effort for the APOA5/A4/C3/A1 gene cluster, a large number of common polymorphisms and rare mutations were identified, and some of the rare variants are reported for the first time in the Chinese Han population. These common and rare variants have remarkable cumulative effects on HTG risk, and these results contribute to the unexplained genetic component of HTG. Furthermore, we found that the combination of the risk alleles demonstrated a more robust association with the modified Gensini scores than any single 1 after adjustment for the traditional clinical risk factors. Importantly, we confirm that the rs2266788 C allele destroys the miR-3201 binding site at APOA5 3′- UTR, which damages the binding ability of miR-3201 and results in the attenuated degradation of APOA5 mRNA induced by normal miRNA-3201 binding and then increases its translation and plasma APOA5 level. Additionally, a dose response between the number of inherited copies of the risk allele rs2266788 and CAD severity was markedly illuminated, and the functional study partly supports this effect. These findings could have significant clinical implications on evaluating the degree of coronary artery stenosis or the possibility of revascularization strategies in patients with multivessel CAD.
The current findings support previous results that showed that severe HTG subjects were carriers of rare heterozygous loss-of-function mutations in candidate lipoprotein metabolism genes (22,23). Among these polymorphisms, our study was the first to provide reliable evidence of an association between rs2266788 and the triglyceride level in the Chinese Han population. This SNP located at 158-bp downstream of the APOA5 gene was reported to be closely associated with familial combined hyperlipidemia in Hong Kong Chinese (24,25). Another 2 SNPs, rs662799 and rs964184, which were previously reported to be significantly associated with triglyceride level in European and Chinese Han populations (5,26), were successfully replicated in our study. Actually, rs2266788 was in moderate LD (r2 = 0.68) with rs662799 in a European population (25); however, higher LD between these 2 SNPs was observed in the Chinese Han population (r2 = 0.92, 1000 Genomes Pilot 1 Project). Thus, this may provide an explanation that the GWAS-sensitive site rs662799 could well capture rs2266788. Indeed, a recent GWAS from a Chinese Han male cohort confirmed a strong association between serum triglyceride levels and rs651821 (27), which was in a high LD (r2 = 0.917) with rs662799 and (r2 = 1.0) with rs2266788 (1000 Genomes Pilot 1 Project) in the Chinese Han population.
We also found some evidence of a combined effect in common risk alleles. There was a strong association between the triglyceride level and increasing gene dosage of the risk scores. This discovery indicates that the functional properties of multiple variants may act together to influence the development of an atherosclerotic lesion.
Genetic variants in the APOA5/A4/C3/A1 gene cluster have been consistently associated with CAD risk. In the current study, we identified that a gene dosage of 2 SNPs (rs662799 and rs2266788) was significantly related to the modified Gensini scores. The SNP rs662799, a regulatory variant in promoter (-1131T>C) of APOA5, have been shown to be related to the risk of coronary heart disease in an analogous dose-dependent manner, with ∼18% higher risk per C allele (13). A recent study also reported the genotype combinations of common variants including rs662799 causally associated with myocardial infarction with a corresponding odds ratio of 1.87 (95% confidence interval: 1.25 to 2.81) (28). Our data are consistent with this observation and imply that the rs662799 gene dosage response for CAD severity was evident, even among patients undergoing their first catheterization. Our study also demonstrates a significant relationship with modified Gensini scores. However, functional assays of this SNP individually could not provide any explanation for this association (20), and Palmen et al. (20) indicated this allele may act in a cooperative manner (i.e., these SNPs may act cooperatively or there is interaction between them) (21). The variant rs2266788 also functions as a predictor of the disease severity. This effect was evident not only in right coronary artery stenosis but also in complex manifestations of CAD. This finding raises the possibility that patients with a homozygous risk genotype may experience a greater atheromatous burden in the asymptomatic phase of the disease. Thus, screening for rs2266788 may capture an anatomically relevant disease before symptoms.
rs662799 does not functionally affect the relative luciferase expression or transcription factors mapped to this site. On the contrary, the rs2266788 C allele damages the miR-3201 binding site at APOA5–3′ UTR and subsequently decreases APOA5 degradation, while elevating circulating APOA5 levels. We used the HLF cells with heterozygous genotype (rs2266788 CT) to conduct experiments and confirm the effects of miR-3201 on the binding site at APOA5–3′ UTR and APOA5 expression. Given previous knowledge about SNP rs2266788 function and the high LD between rs2266788 and rs662799, we hypothesize that rs2266788, not rs662799, is the causal variant. However, the function of miR-3201 is largely unknown, although it has been reported to be expressed and upregulated in melanoma (29). It should be noted that the expression level of miR-3201 is higher in HepG2 cells than in 293T cells (Online Fig. 7), which may partly explain the difference in the luciferase expression level for allele-specific constructs in different cells.
There are several limitations of our study that must be acknowledged. The first limitations of our fine-mapping study design are that we were underpowered to detect associations with SNPs at MAF <0.05, and we limited sequencing to regions most likely to harbor functional variation. Second, the subjects recruited in our study may not be entirely representative of the general Han Chinese population; the potential false-positive results are still possible even after multiple corrections. It is important to confirm these positive association in prospective cohort studies. Third, we tried to find a mouse or rat model to validate the effect of miR-3201 in vivo. Unfortunately, a suitable animal model match was not available, which is a limitation of our study. Finally, the mechanism by which motif binding to miR-3201 could regulate the plasma levels of APOA5 is not fully understood and requires further elucidation in the future.
Our results suggest that a complex genetic architecture of both common and rare variants in a spectrum of triglyceride-associated genes is responsible for HTG. Our findings also showed that the gene dosage of the APOA5/A4/C3/A1 gene cluster risk allele predicts CAD burden. Furthermore, the rs2266788 C allele destroys the miR-3201 binding site at APOA5 3′ UTR, therefore increases translation of APOA5 and subsequently increases the plasma APOA5 level.
COMPETENCY IN MEDICAL KNOWLEDGE: Genome-wide association studies have identified APOA5/A4/C3/A1 gene cluster polymorphisms associated with hypertriglyceridemia and an increased risk of CAD.
TRANSLATIONAL OUTLOOK: Prospective studies are needed to assess the clinical utility of genotyping for APOA5/A4/C3/A1 polymorphisms and the efficacy of therapeutic interventions, including those that specifically target serum triglycerides to reduce the risk of ischemic events.
The authors thank Drs. Xunna Bao, Kun Miao, Jia Peng, and Luo Zhang for their efforts in recruiting patients and controls for this study. They especially acknowledge all the participants in this study.
This work was supported by grants from the National “973” program (nos. 2012CB518004 and 2014CB541601), the National Natural Science Foundation of Chinahttp://dx.doi.org/10.13039/501100001809 (81100066), and the “863” project (no. 2012AA02A510). The authors have reported that they have no relationships relevant to the contents of this paper to disclose. Drs. Cui and Li contributed equally to this work.
- Abbreviations and Acronyms
- coronary artery disease
- genome-wide association study(ies)
- high-density lipoprotein cholesterol
- linkage disequilibrium
- low-density lipoprotein cholesterol
- microRNA 3201
- messenger RNA
- single nucleotide polymorphism
- Received January 20, 2014.
- Revision received March 8, 2014.
- Accepted March 15, 2014.
- American College of Cardiology Foundation
- Wilson P.W.,
- Grundy S.M.
- Sarwar N.,
- Danesh J.,
- Eiriksdottir G.,
- et al.
- Waterworth D.M.,
- Ricketts S.L.,
- Song K.,
- et al.
- Yuan G.,
- Al-Shali K.Z.,
- Hegele R.A.
- Dandona S.,
- Stewart A.F.,
- Chen L.,
- et al.
- Montorsi P.,
- Ravagnani P.M.,
- Galli S.,
- et al.
- Johansen C.T.,
- Wang J.,
- Lanktree M.B.,
- et al.
- Talmud P.J.,
- Palmen J.,
- Putt W.,
- et al.
- Wang J.,
- Ban M.R.,
- Zou G.Y.,
- et al.
- Wang J.,
- Cao H.,
- Ban M.R.,
- et al.
- Kraja A.T.,
- Vaidya D.,
- Pankow J.S.,
- et al.
- Tan A.,
- Sun J.,
- Xia N.,
- et al.
- Jorgensen A.B.,
- Frikke-Schmidt R.,
- West A.S.,
- et al.