Background
Currently, over 200 million people worldwide are chronically infected with hepatitis C virus (HCV) [
1]. HCV is an important cause of acute and chronic hepatitis, with only 20% of patients have spontaneous clearance (SC) of the virus. Unfortunately, around a quarter of all patients will progress to develop cirrhosis. Other complications such as co-infection with HIV are common among patients infected with HCV, because both viruses share the same routes of transmission [
2].
To date, the mechanisms underlying HCV infection have not been completely defined. Over the past few decades, several host and viral factors have been found to be associated with differences in HCV clearance or persistence. However, an unexplained variability in treatment outcome still exists, suggesting that the genetic background of the host plays an important role [
1].
Nowadays, the standard of care for chronic HCV infection consists of pegylated interferon -α plus ribavirin (PEG-IFN/RBV). However, this treatment produces sustained virologic response (SVR) rates in only about 40 to 50% of patients with HCV genotype 1 and approximately 60% in those infected with genotype 4, whereas over 80% of patients with genotypes 2 or 3 achieve SVR [
2]. PEG-IFN/RBV treatment is prolonged and costly, and is associated with dose-limiting side-effects, highlighting the need for accurate prediction of treatment failure.
Recently, three independent groups discovered several strongly correlated single genetic polymorphisms (SNPs) that, together with standard therapy, seem to play a role in clearing the virus [
3‐
5]. These polymorphisms are located close to the
interleukin 28B (
IL28B) gene on chromosome 19.
IL28B encodes IFN-λ3, which induces antiviral activity by itself and through the Janus kinase-signal transducer and activator of transcription (JAK-STAT) complex, which induces IFN-stimulated genes (ISGs) that also have antiviral activity against HCV [
6]. Beyond their identification, little is known about the mechanisms involved between these genomic variants and viral clearance. However, it has been shown that unfavorable
IL28B genetic variations are associated with higher pre-activated levels of ISGs, which could explain the poor response in these patients [
7,
8].
Interest in this gene has led to a large number of publications showing
IL28B polymorphisms as having an influence on HCV clearance. However, conflicting conclusions have been reached in some cases, with some studies reporting significant associations, and others reporting no such associations [
9‐
11].
The aim of this study was to elucidate the pooled estimated effect of IL28B polymorphisms on PEG-IFN/RBV treatment response and SC, by conducting a meta-analysis of all eligible studies published up to 9 July 2012.
Methods
Search strategy and selection criteria
Relevant studies were identified by a literature search of PubMed without imposing study-period restrictions, using the following terms: 'hepatitis C', '
IL28B', 'SNP', 'spontaneous clearance', 'treatment', 'ribavirin' and 'interferon' (for the complete electronic search strategy, see Additional file
1). The information contained in this report is based on articles published before 9 July 2012 in any language. The meta-analysis was conducted following the guidelines published by Sutton
et al. [
12], and the data are reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [
13].
We developed strict inclusion and exclusion criteria before reviewing the studies and extracting the data in order to ensure maximum possible homogeneity between studies. Only articles satisfying the following criteria were further evaluated by two independent reviewers: 1) patients infected with HCV or HCV/HIV; 2) any SNP located within or near the IL28B gene; and 3) SVR data available for patients receiving standard treatment or d) SC data available for patients with acute HCV infections. Exclusion criteria were: 1) treatment duration of less than 24 weeks; 2) co-infection with HBV; 3) studies that included patients with organ transplantation; 4) studies with sample sizes of less than 40 subjects; 5) SVR data reported by methods other than PCR; 6) HCV genotypes other than 1, 2, 3 or 4.
To select the candidate studies, we carefully screened the title and abstract of each citation. When articles fulfilled the inclusion criteria, we examined the full text and extracted data from the study. When studies included several subgroups that did not all fulfill the inclusion criteria, we only incorporated into the meta-analysis those subgroups that did meet the inclusion criteria.
Two investigators independently assessed the selected papers and extracted all data. When data were unclear or required assumptions to be made, another investigator was consulted, so that a consensus could be reached before recording an entry in the database. When more than one paper studying the same cohort was found, only the study with the most extensive cohort was reviewed, excluding the remaining overlapping studies or data (for the extracted data from each study, see Additional file
2). When incongruent data were detected for one particular variable, they were discarded. We only included those data that could be confirmed by their appearance in different parts of the text (for instance in the main text and the tables).
When articles provided data from patients of different ethnicities, HCV genotypes, and/or types of infection (HCV and HCV/HIV), these data were divided into subgroups, with each subgroup identified by a sequential letter. Regarding ethnicity, subjects were grouped into the main racial group of the study population based on their geographical origin or ancestry [
14]. Studies with more than 99% of the study population described as Caucasian were considered to be Caucasian. For data presentation, genotypes 1 and 4 were considered to be in the same subgroup because of their similarity in response and the same applied for genotypes 2 and 3 [
15].
To compare results between studies on fibrosis, we used 4 to 6 Ishak levels for the scale of fibrosis corresponding to a METAVIR score of 3 to 4. HCV viral load data were collected at pre-treatment.
Only patients who had an absence of detectable serum HCV RNA by PCR 24 weeks after treatment cessation were considered to have achieved SVR. For SC, patients with acute HCV infection who had undetectable levels HCV RNA on repeated examination without prior IFN therapy were considered to have spontaneously resolved the infection.
In an attempt to include the majority of studies and to complete any missing data, we contacted some authors of individual studies.
Quality appraisal
To evaluate the quality of the included studies, two investigators appraised them independently using a checklist based on the Graphic Appraisal Tool for Epidemiological studies (GATE) [
16]. Each item was rated as +1 (well reported and reliable), 0 (unclear, insufficient detail provided) or -1 (poorly reported, not useful or reliable). The overall validity of each study was also rated by a similar system: +1 (most of the quality items were fulfilled), 0 (some criteria were not fulfilled), or -1 (few or none of the items were fulfilled).
Statistical analysis
We calculated whether all studied SNPs in the present meta-analysis concurred with Hardy-Weinberg equilibrium (HWE) [
17] using the χ
2 test, with equilibrium considered at
P > 0.05.
Overall, meta-analyses were performed only when two or more papers studying the same SNP were available. In all analyses, pooled odds ratios (ORs) and 95% confidence intervals (CIs) were calculated (favorable homozygous versus heterozygous plus unfavorable homozygous). The significance of the pooled OR was calculated by the
Z-test, and was considered significant at
P < 0.05. A fixed effect model (the traditional Mantel-Haenszel test) was used for homogeneous studies [
18]. The heterogeneity of each group of ORs was assessed by the χ
2 test, which suggests the presence of heterogeneity when
P < 0.1. Heterogeneity was quantified with the
I
2 metric, which provides a measure of the degree of inconsistency in the studies' results (
I
2 > 50% indicated considerable heterogeneity). When significant heterogeneity existed, a random-effects model (the DerSimonian and Laird method [
19]), was used, and a subgroup meta-analysis and forest plot based on ethnicity, HCV genotype, and co-infection data (HCV or HCV/HIV) were performed to identify the effect modifiers [
20]. Moreover, the Galbraith plot was used to detect possible outliers of the heterogeneity, which could have biased the combined estimate. This graphical method allowed those studies that had a strong influence on the pooled results to be checked [
21,
22]. Trials outside the Galbraith limits were trials where the 95% CI did not contain the pooled estimate.
In addition, when heterogeneity was detected, meta-regression analysis was also performed, with the aim of defining the potential effect of the covariates on spontaneous or treatment-related clearance. The regression coefficient obtained describes how SC or SVR changes with each unit increase in the covariate. Significance of the linearity relationship between SC or SVR and the covariate was identified by the P-value; the adjusted R2 term indicates the proportion of between-study variance explained by the included variable. The evaluated variables were as follows: genotype (1/4 versus 2/3), ethnicity (Caucasian versus other, and Caucasian versus East Asian), type of viral infection (HCV mono-infection versus HCV/HIV co-infection), viral load (HCV RNA < 6 log10 versus HCV RNA ≥ 6 log10), advanced fibrosis F3/F4 (≤ 30% versus > 30%), and prior HCV treatment (naive versus previous HCV treatment failures).
Publication bias was assessed by funnel plot and the Egger linear regression test [
23,
24], which detects funnel plot asymmetry. When the Egger test reported a
P < 0.05, publication bias was assumed to exist.
Sensitivity analyses [
24] were also conducted to assess the consistency of results and to investigate the influence of one study on the overall meta-analysis. It was carried out by sequential omission of individual studies.
All analyses were performed using Stata software (version 11.0; Stata Corporation, College Station, TX, USA).
Discussion
There is growing evidence that polymorphisms can contribute to differences in complex disease traits between individuals. Since 2009, several studies have shown that there is an important association between IL28B polymorphisms and HCV clearance. However, the mechanism of this association remains unclear, and is still under study.
To our knowledge, a total of five reviews containing data from meta-analyses have been published to date about the relationship between
IL28B polymorphisms and SVR [
96‐
100]. The reports of Romero-Gomez
et al. and Li
et al. are limited because they involved literature searches only up to January and May 2010, respectively, leading to the selection of a low number of studies in both cases (only seven studies). In addition, all meta-analyses were performed for rs12979860 and rs8099917 only. The other three meta-analyses were broader, ranging from 17 to 36 studies. However, all of them analyzed only the effect of the
IL28B polymorphisms rs12979860 and rs8099917 on SVR by ethnicity and HCV genotype, while the study by Scheiber
et al. was limited to HCV genotypes 2 and 3. In addition, the literature searches of these meta-analyses included publications only up to the end of 2011. Consequently, our meta-analysis provides the most up-to-date compilation of studies, including 74 articles, a larger number of SNPs, and association analysis with other clinical situations such as SC. In addition, this is the first meta-analysis investigating the association between
IL28B polymorphisms and SC, to our knowledge. Therefore, because there has been a very large increase in the number of papers, especially in 2012, it is necessary to assess all data and elucidate controversial or inconclusive results. This meta-analysis has allowed us to estimate the overall OR of all studies, and functions as a robust tool to investigate discrepant results.
Based on the global analysis, there was a significant association between the favorable genotype of seven studied SNPs (rs12979860, rs8099917, rs12980275, rs8105790, rs11881222, rs8103142, and rs7248668) and SVR. In most of the cases, the probability of achieving SVR in patients with a favorable genotype was more than double that in patients with an unfavorable genotype. Although the majority of results had similarities, the magnitude of the association was different in many cases. This could be caused by the different criteria considered in each individual study. For this reason, our meta-analysis focused on displaying general conclusions about the trend of this association. Moreover, we investigated several variables that might contribute to the different magnitude of associations found in different studies.
IL28B and ethnicity
The significant genetic association of all
IL28B SNPs seems to be due to the high linkage disequilibrium (LD) of this genomic region [
3‐
5], which varies across ethnicities. Regarding the Asian population, some authors have described strong LDs between rs12980275, rs8105790, rs11881222, rs8099917, rs7248668, rs10853728, and rs12979860 [
5,
37]. For Caucasian populations, the results are slightly more diverse depending on the genotyped platform used. Thus, Ge
et al. [
3] reported a strong LD between rs12979860 and rs12980275, whereas Suppiah
et al. [
4] reported that rs12980275 is strictly linked to rs8105790, rs8103142, rs8109886, and rs8099917. Finally, there was a low LD between rs12979860 and rs12980275 for African Americans (r
2 = 0.56), whereas for Hispanics, a higher LD (r
2 = 0.88) was obtained [
3]. Our results have shown that in respect to the favorable
IL28B genotype frequency of rs12979860, there was a marked differential distribution between racial groups (in order from highest to lowest frequency): Asian, Caucasian, North African, Hispanic, African, and African American. Regarding rs8099917, the order was similar, except for African populations, which showed frequencies that were intermediate between Asians and Caucasians, as shown by Thompson
et al. [
25]. This differential distribution seems to explain much of the observed clinical differences between ethnic groups in response to treatment [
101].
We found similar association for rs12979860 for Asians (OR = 3.27) and Caucasians (OR = 3.63). The strength of the association in Asians was almost double that for Caucasians for rs8099917. Similarly, in the case of rs12980275, a larger OR was seen for Asians than for Caucasians. However, rs12980275 was only represented by two studies in Asians and four in Caucasians, therefore the most reliable results and the most robust conclusions were obtained for rs12979860 and rs8099917.
It is also important to note that in our study, the significant association between favorable genotypes and SVR was lost in several instances, such as for rs12979860 in African American and Hispanic patients. However, these data should also be interpreted with caution, because only two studies were included for each ethnicity [
9,
25,
59]. Regarding African Americans, the results were right at the limit of significance (OR = 3.19;
P = 0.052), which may be related to the low LD described for African Americans. Owing to the scarcity of available results, new studies in these populations are necessary, especially ones that investigate the effects of different SNPs. For Hispanics, both studies individually showed a significant association, but this significance was lost upon performing the meta-analysis. This could be due to the extremely wide confidence interval of the Venegas
et al. study [
9]. On the other hand, rs8099917 and rs12980275 seem to be strongly associated for Hispanic populations; however, these results corresponded to only one study, which prevents us from drawing any firm conclusions.
IL28Band hepatitis C virus genotype
To date, a broad association between favorable
IL28B genotypes and SVR has been described in patients infected with HCV genotype 1 [
50,
102], with a similar association described for genotype 4, although this has been less studied. However, conflicting results have been published about HCV genotypes 2/3 [
33,
41]. One of our goals was to discern the pooled significance of such an association, which would have relevance to the decision of initiating therapy. As expected, we found that the favorable genotypes of polymorphisms rs12979860, rs8099917, and rs12980275 were positively associated with SVR for HCV genotypes 1 and 4. Regarding the HCV genotypes 2 and 3, the polymorphisms rs12979860 and rs8099917 showed significant associations. However, the strength of this association was almost three times lower than for genotypes 1 and 4, and in addition, we found that the Asian population was solely responsible for this association in rs8099917. The generally reduced association for patients with HCV genotypes 2/3 could be related to the high rate of SVR present in these IFN-sensitive genotypes, for which larger sample sizes are required to find significant differences [
7]. In summary, our findings show that
IL28B polymorphisms are a strong pre-treatment predictor for SVR in patients with HCV genotypes 1 and 4, but its usefulness is limited for other genotypes.
IL28B and type of viral infection
The predictive value of
IL28B polymorphisms has been extensively studied in patients with HCV mono-infection, but only seven eligible studies included patients with HIV/HCV co-infection [
28,
29,
35,
42,
61,
63,
85]. After stratifying by type of infection, we found that in patients co-infected with HIV/HCV, the strength of association between rs12979860 and SVR was similar to that for patients with HCV mono-infection. For rs8099917, only the study by Aparicio
et al. [
29] provided data for patients with HIV/HCV co-infection, which was also divided by HCV genotypes (1, 3, and 4). This study did not show any overall significant association, but when we analyzed the data in more depth, we found differences related to HCV genotype, with only HCV genotype 1 being significantly associated with SVR. Therefore, the benefit of
IL28B genotyping seems to apply to both patients with HCV genotype 1 mono-infection and co-infection. HIV/HCV co-infection could play a significant role in treatment response, but further studies are necessary to confirm this. Again, results should be interpreted with caution.
IL28Bpolymorphisms and spontaneous clearance of hepatitis C virus
The identification of markers predicting the persistence of HCV infection is very important to distinguish between patients whose acute hepatitis C resolves and those who develop a chronic hepatitis C infection. Roughly, 20% of patients infected with HCV have SC of the virus. The mechanism of this is not clear, but epidemiological, viral, and host factors have all been associated with the difference in HCV clearance likelihood. Thomas
et al. [
103] showed that rs12979860 strongly enhances the likelihood of clearance of HCV among individuals of either European or African ancestry. However, to date, few articles about the
IL28B polymorphisms involved in SC have been written. Owing to the low number of published articles, we could perform meta-analyses only for rs12979860 and rs8099917. In both cases, a clear association was detected. These results seem completely plausible because all individual studies analyzed showed significant associations for rs12979860 and rs8099917, except for Asian populations, which were represented by only one study [
94]. Stratification was possible only for rs12979860. Regarding ethnicity, significant results were obtained only for Caucasians, with similar pooled OR as for SVR.
With respect to HCV genotype, a strong association was detected for rs12979860 in HCV genotype 1. It has been shown that HCV genotype influences hepatitis C chronicity, as patients infected with genotype 1/4 who harbor favorable
IL28B genotypes are less likely to have chronic HCV infection. By contrast, this protective effect is not seen for infections with either genotypes 2 or 3 [
1]. Because of insufficient data, we could not perform a stratified study on other HCV genotypes or ethnicity. Additional studies on these variables are needed to clarify this association.
IL28B polymorphisms as clinical predictors
During the past number of years, the main focus in HCV infection has been the identification of markers or factors predicting the likelihood of achieving SVR. Recently, some countries have incorporated
IL28B genotyping as a diagnostic criterion in clinical practice [
7]. In those patients with unfavorable
IL28B genotypes, which result in response rates of less than 40%, clinicians may consider deferral of treatment until novel therapies are licensed, something likely to occur soon [
7]. However,
IL28B genotype is not solely responsible for therapy response. Indeed, our meta-regression data suggested that several other factors such as ethnicity, HCV genotype, and stage of fibrosis might have a significant effect on SVR and/or SC. Recently, a model including
IL28B genotype (rs12979860) and four clinical variables (pre-treatment viral load, ratio of alanine and aspartate transaminases, Ishak fibrosis score, and previous treatment with ribavirin) has been developed [
51]. This model predicts SVR in patients of European ancestry with HCV genotype 1 who have failed to respond to previous treatment [
51]. This algorithm has shown a high predictive ability, but as the authors pointed out, it could be improved in future studies by including other relevant variables such as ethnicity and HCV genotype. Another predictive model has been described for patients with HCV/HIV co-infection, including two host-related variables (the
IL28B SNP rs12979860 and the level of liver stiffness) and two HCV-related variables (genotype and viral load) [
104]. This model was found to have an adequate predictive index, but it could also be enhanced by incorporating HIV variables such as viral load. Apart from the aforementioned variables, it would be of great value for clinical practice if future algorithms could be designed that were applicable to different circumstances, such as naive patients or patients with SC, for instance.
Future directions
The number of studies on new antiviral therapies have risen in the past few years. Therefore, it would have been interesting to investigate whether
IL28B polymorphisms also play a predictive role in novel therapies such as direct-acting antivirals (DAAs). However, it was not been possible here because only a few studies have been published that include data on novel therapies. Our literature search returned five studies involving triple therapy. Three of them involved standard of care (PEG-IFN/RBV) with inclusion of telaprevir in the same cohort [
105‐
107], while the fourth studied the inclusion of danoprevir [
108], and the fifth studied the inclusion of boceprevir [
109]. As we could not perform meta-analysis on these, all of these were excluded. These therapies, which are based on protease inhibitors, are the most advanced DAAs in clinical development. However, any influence of
IL28B polymorphisms on the outcome of these novel therapies is not clear. These few studies have shown that, regardless of treatment history,
IL28B SNPs seem to enhance rapid, early, and SVR when combined with PEG-IFN/RBV in patients with chronic HCV genotype 1 infection [
109,
110]. Further studies are needed to clarify this association.
Finally, to properly interpret our results, some considerations have to be taken into account. Our meta-analysis was performed by using the unadjusted raw data provided from each study, whereas most of the results given by the authors were previously adjusted by age, fibrosis stage, HCV viral load, and/or other factors. For this reason, our ORs may differ slightly from those cited by the original articles. For rs8099917, we identified publication bias, which could indicate that smaller studies dealing with this SNP could have been more likely to be published if their results were significant than if their results were negative or inconclusive. When heterogeneity was evaluated, studies were stratified by ethnicity, genotype, and type of infection, but in some cases heterogeneity remained, indicating the possibility that different causes of heterogeneity may exist. As we have previously mentioned, the number of studies in some subgroup analyses was too small, which led to weak results. As for the five least studied polymorphisms (rs11881222, rs7248668, rs8103142, rs8105790, rs10853728), results are limited and new studies are still needed. Consequently, these results should be interpreted with caution.
Conclusions
IL28B polymorphisms influence both IFN treatment outcomes and the natural clearance of HCV infection. However, although we cannot provide a biological explanation, our findings indicate the most adequate genetic marker seems to vary depending on ethnicity, HCV genotype, and type of viral infection. Taking into account the most robust analyses, the SNP most associated with SVR in Caucasians was rs12979860, whereas in Asians it seemed to be rs8099917. However, for Africans, African Americans, North Africans, and Hispanics, extensive studies are still needed. After analyzing the conflicting results described above for genotypes 2 and 3, our data seem to indicate that rs8099917 is apparently the most adequate predictive marker for SVR with these genotypes. For patients with HCV/HIV co-infection, the most studied SNP (rs12979860) gave similar results to those seen for patients with HCV mono-infection patients. Finally, both rs12979860 and rs8099917 were clearly associated with SC. Moreover, because IL28B genotyping needs be performed only once in a patient's life, it is relatively cheap and provides high predictive value. IL28B polymorphisms could thus be used to perform personalized treatment in clinical practice, which could play a substantial role in the selection of candidates for standard treatment versus triple therapy with DAAs.
Acknowledgements
We thank those authors of the studies included in this review for providing additional information upon request. We also are grateful to Cristina Asensio and Rocío Carmona for valuable help with the Stata software. We thank D. Aurelio Tobías for his valuable suggestions and help with heterogeneity analyses. We thank the reviewers for their suggestions and useful comments. This work was supported by grants from the 'Health Institute Carlos III (ISCIII)' [PI08/0738 and PI11/00245]. MAJS, AFR MGF and MGA were supported by the after grants [CM10/00105, PI08/0738, CM09/00031, and CM08/00101 respectively].
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
MAJS, AF, and SR designed the study. MAJS and AF collected all data, performed the statistical analysis, and drafted the report. MAJS, AFR, MGA, MGF, and SR interpreted the data and critically reviewed the report. All authors have approved the final version of the manuscript.