Genetic associations with brain microbleeds

Abstract

Objective:

We performed a systematic review and meta-analyses to assess the evidence for genetic associations with brain microbleeds (BMBs).

Methods:

We sought all published studies of the association between any genetic polymorphism and BMBs studied in a total of >100 people. We critically appraised studies, and calculated pooled odds ratios (ORs) using the generic inverse variance fixed effects method. We used I² and χ² statistics to assess heterogeneity, and fail-safe N estimates to assess the robustness of our results.

Results:

Only the APOE ϵ2/3/4 polymorphism had been studied in >100 people (10 studies, 7,351 participants). Compared with people with the ϵ3/ϵ3 genotype, carriers of the ϵ4 allele (ϵ4+) were statistically significantly more likely to have BMBs in any location (ϵ4+ vs ϵ3/ϵ3: pooled OR 1.22, 95% confidence interval [CI] 1.05–1.41, p = 0.01). For strictly lobar BMBs, this association appeared slightly stronger (ϵ4+ vs ϵ3/ϵ3: pooled OR 1.35, 95% CI 1.10–1.66, p = 0.005). The association of ϵ4+ genotypes with strictly lobar BMBs was reasonably robust to potential publication and reporting biases.

Conclusions:

Given the known associations of APOE alleles with lobar intracerebral hemorrhage and cerebral amyloid angiopathy, these findings support the concept that strictly lobar BMBs may be an imaging biomarker of cerebral amyloid angiopathy.

Brain microbleeds (BMBs) are minute deposits of blood products, seen as focal areas of signal loss typically <10 mm in diameter on haem-sensitive T2*-weighted gradient-recalled echo (GRE) MRI sequences. BMBs occur in 5% to 38% of apparently healthy people, increasing in frequency with age, hypertension, and smoking.^1–3 They are particularly prevalent (>50%) in people with a history of intracerebral hemorrhage (ICH).¹ Lobar BMBs are considered a diagnostic marker of cerebral amyloid angiopathy (CAA),⁴ which appears to be the major cause of lobar, sometimes recurrent ICH in the elderly.⁵

Understanding the vascular pathology underlying BMBs should be enhanced by identifying genetic risk factors. There is particular interest in whether APOE genotype might confer susceptibility to BMBs. The APOE gene on chromosome 19 has 3 common alleles—ϵ2, ϵ3, and ϵ4—giving rise to 6 different genotypes: ϵ2/2, ϵ2/3, ϵ2/4, ϵ3/3 (the wild type, occurring in just under two-thirds of people in most populations⁶), ϵ3/4, and ϵ4/4. Associations of the APOE ϵ2 or ϵ4 alleles with both ICH (particularly lobar ICH) and CAA have been described,^7–11 and strictly lobar BMBs have now been included in the modified Boston diagnostic criteria for CAA.⁴ We would therefore expect an association between these alleles and BMBs, especially those in a strictly lobar location.

We therefore carried out a systematic review of the literature on genetic variants associated with BMBs and, where appropriate, meta-analyses of the relationship between polymorphisms in any gene (including APOE) and BMBs.

METHODS

Search strategy and selection criteria.

We sought articles published in full in any language, obtaining translations where necessary, which studied an association between any genetic polymorphism and BMBs detected on MRI in humans. We used comprehensive electronic search strategies (most recent search date July 2010) including a list of 13 different medical subject heading and text word terms for BMBs and 13 and 17 different medical subject heading and text word terms for genes or polymorphisms in Medline (from 1966) and Embase (from 1980) respectively to identify relevant articles (appendix). Two authors (S.S.M. and L.P.) read the titles, abstracts, and, where necessary, full text of all identified studies to identify potentially relevant studies, resolving any uncertainties with a third author (C.L.M.S.). At least 2 authors (S.S.M. and one or both of C.A.J. and C.L.M.S.) read in full all articles selected as potentially relevant and, for each genetic polymorphism, determined the number of independent studies and the total number of participants included. Two authors (S.S.M. and L.P.) hand-searched the reference lists of all included studies and any related review articles to identify relevant studies missed in the initial search. Where a polymorphism had been studied in a total of >100 participants, we selected the relevant studies for detailed methodologic assessment and meta-analyses. We decided not to review data for polymorphisms studied in ≤100 participants since conclusions drawn on the basis of such small numbers would be likely to be unreliable.

Data extraction.

For each selected study, we extracted information on year of publication; country in which the study was conducted; types of participants studied; number, ethnicity, and mean age of participants; genotyping method; whether genotypes were stated to be in Hardy-Weinberg equilibrium (HWE); and blinding of scanning staff to genotypes and of genotyping staff to scan results. We also extracted information on the quality and features of the brain imaging methods used. These included the criteria specified to identify a BMB (both appearance and size), the awareness of potential BMB mimics, the MRI sequence used, the number of MRI scan readers, and any available data on interobserver and intraobserver agreement of BMB rating. We assessed each study against a checklist of key quality indicators (clearly defined BMB criteria, awareness of ≥2 BMB mimics, ≥2 BMBs raters, raters independent, interobserver and intraobserver agreement recorded, genotypes stated to be in HWE, blinding of BMB raters to genotype and genotyping staff to BMB status, adjustment of results for other risk factors) that we developed from the above methodologic, imaging, and laboratory criteria and with reference to both the Strengthening the Reporting of Genetic Association Studies statement¹² and the ideal characteristics for a study of BMBs.¹ However, we did not include or exclude studies on the basis of these quality indicators.

Two authors (S.S.M. and C.A.J.) independently extracted published data from each study on the number of participants with at least one BMB for each genotype or group of genotypes, or published crude or adjusted odds ratios (ORs) derived from these data. S.S.M. and C.A.J. discussed and resolved any disagreements with a third author (C.L.M.S.). For APOE studies, where possible, we subdivided the participants into those with an ϵ2-containing (ϵ2+), ϵ4-containing (ϵ4+), or ϵ3/3 genotype, and excluded the small proportion with ϵ2/4 genotypes, allowing the most consistent treatment of data across studies. Where available, we extracted data on the relationship between APOE and BMBs in either strictly lobar or not strictly lobar distributions, assigning participants with BMBs restricted to a lobar location to the strictly lobar group and all others with BMBs to the not strictly lobar group. If studies had not published these data, we approached the corresponding author for additional unpublished data on location-specific BMB-genotype associations in an effort to include all available data in our meta-analyses. If the same participants had been included in more than one publication, we only included those from the publication with the larger number of participants.

Statistical analysis.

We performed meta-analyses using Cochrane RevMan (version 4.2) software. We describe here methods used for analyses of the APOE ϵ2/3/4 polymorphism since this turned out to be the only polymorphism studied in >100 participants. Where studies provided the necessary data, we calculated directly study-specific ORs for the presence vs absence of BMBs, comparing participants with an ϵ2+ or ϵ4+ genotype vs the ϵ3/3 genotype, and otherwise used the relevant published crude or adjusted ORs. We used the generic inverse variance fixed effect meta-analysis method to obtain pooled ORs, weighting each study by the inverse of the square of the standard error of its study-specific OR. Where both crude and adjusted ORs were available for a particular study, we used the crude OR for meta-analysis, and compared the crude and adjusted ORs to assess for the possibility of confounding. Unless stated otherwise, we considered a p value of <0.05 to be statistically significant.

We assessed statistical heterogeneity between studies using both the χ² test and the I² statistic.¹³ We performed prespecified subgroup analyses based on study size (above or below the mean study size), ethnicity, and type of study population (healthy individuals compared with those recruited on the basis of existing neurologic conditions), to explore potential modification of effect by these factors, assessing differences between subgroups with χ² statistics. We also assessed the effects of genotype on strictly lobar and not strictly lobar BMBs separately.

To assess the potential effect on our results of unpublished negative studies or studies not reporting or able to provide the data required for our meta-analyses (i.e., the potential for publication and reporting biases), we calculated the “fail-safe N” for any meta-analysis which produced a significant result (p < 0.05). We used a modification of Rosenthal's method,¹⁴ in which we determined the size of a notional study with a null result (OR = 1) required to bring any significant result to a just nonsignificant level (p = 0.05), assuming the overall prevalence of BMBs and the distribution of genotypes in this notional study to be the average of these for the studies included. Since we carried out several tests of significance, we repeated our fail-safe N calculations based on a statistical significance cutoff of p = 0.01.

We prepared this report with reference to the Meta-analysis of Observational Studies in Epidemiology reporting guidelines.¹⁵

RESULTS

We identified 112 articles in our initial search of Medline and Embase, of which 31 were potentially relevant to this systematic review. Only the APOE ϵ2/3/4 polymorphism had been studied in a total of >100 participants. Ten studies assessed the association between APOE ϵ2/3/4 genotypes and BMBs in a total of 7,351 participants (figure 1). Seven studies which included 7,272 participants (99% of the total) presented dichotomous data for BMBs (i.e., presence vs absence per genotype or group of genotypes), enabling their results to be pooled in meta-analyses.^2,3,16–21 The remaining 3 presented continuous data (i.e., mean number of BMBs per genotype) or purely qualitative statements about the association between APOE and BMBs.^22–24

Figure 1. Study selection and exclusion.

BMB = brain microbleed; CADASIL = cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy.

Study characteristics.

Study populations comprised either healthy people from the general population (3 studies, 5,977 participants) or those with neurologic (mainly cerebrovascular) conditions (7 studies, 1,374 participants). Most participants were of European origin, but one study was conducted in Asians.¹⁷ Participants were middle-aged to elderly (mean age per study ranged from 45 to 76 years). Studies had a mean of 735 (range 20 to 3,689) participants. Prevalence of BMBs in the study populations varied from 5% in the Framingham and Framingham offspring population-based cohort to 65% in a population of people hospitalized with intracerebral hemorrhage (table 1).

Table 1.

Characteristics of included studies

Abbreviations: BMB = brain microbleed; CAA = cerebral amyloid angiopathy; CADASIL = cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy; HCHWA-D = Hereditary Cerebral Haemorrhage with Amyloidosis–Dutch type; ICH = intracerebral hemorrhage; NR = not reported.

^aAs described under “types of participants.”

^bEstimated from published % where necessary, as marked.

^cAssumed indicates ethnicity not directly stated.

^dOf number of participants scanned and genotyped, unless stated otherwise.

^eMean age and % male of total study population.

Imaging characteristics varied between the studies. BMBs were typically defined as rounded focal areas of signal loss, but the size criteria varied. All studies but one¹⁷ considered BMB mimics, most commonly calcification in the basal ganglia (although none of the studies appeared to use CT to investigate this) (table e-1 on the Neurology^® Web site at www.neurology.org). The larger studies generally fulfilled more of our quality indicators for a study on BMB genetics than smaller studies, but none fulfilled all 10 (table 2). The studies with data unavailable or unsuitable for our meta-analyses were much smaller in size than the included studies and performed less well against our methodologic quality indicators (table 2).^22–24 The largest study was of high quality and contributed over half of the participants included in the meta-analyses.^3,21 Where both crude and adjusted ORs were available, they were similar.^17,20

Table 2.

Summary of key quality indicators

Abbreviations: BMB = brain microbleed; HWE = Hardy-Weinberg equilibrium.

^aThere were 4 mimics considered: calcification/nonhemorrhagic iron deposits; sulcal flow voids; cavernous malformations; hemorrhagic transformations of cerebral infarcts.

^bOnly had fair to moderate interobserver agreement (others reporting on this found substantial or almost perfect agreement; see table e-1).

Association between APOE ϵ2/3/4 polymorphism and BMBs.

In the 7 studies with dichotomous data on BMBs, pooled results showed that, compared with people with an ϵ3/3 genotype, those with an ϵ4+ genotype had an increased odds of having one or more BMBs (OR 1.22, 95% confidence interval [CI] 1.05–1.41, p = 0.01; figure 2A). There was no difference between the odds of having BMBs among those with an ϵ2+ vs those with an e3/3 genotype (OR 1.14, 95% CI 0.94–1.40, p = 0.2; figure 3A). There was no evidence from χ² or I² statistics of any heterogeneity between the studies' results for either of these genotype group comparisons (figures 2A and 3A), and there were no differences between the pooled results of studies divided into subgroups according to study size, ethnicity, or type of study population (data not shown).

Figure 2. Meta-analysis of the association between ϵ4+ vs ϵ3/3 genotypes and brain microbleeds (BMBs).

(A) BMBs in any location. (B) BMBs in strictly lobar or not strictly lobar locations. Studies are displayed in order of publication date. The squares represent study-specific odds ratios (ORs), with their size proportional to their statistical weight by the generic inverse variance method. Horizontal lines represent 95% confidence intervals (CIs). Diamonds represent pooled ORs, and their width represents the 95% CI for the relevant pooled OR. *Number of participants included in analysis. For some studies, where numbers of participants per genotype were not provided, we estimated the total number of participants with ϵ4+ or ϵ3/3 genotypes, assuming the distribution of the genotypes to be the average of the included studies with available data on each genotype. This provides a rough estimate of numbers of participants included in the comparison, but has no effect on the study-specific OR, standard error, or meta-analysis calculations. **Study used slightly different comparison groups (ϵ4+ vs ϵ4−).

Figure 3. Meta-analysis of the association between ϵ2+ vs ϵ3/3 genotypes and brain microbleeds (BMBs).

(A) BMBs in any location. (B) BMBs in strictly lobar or not strictly lobar locations. Notation as for figure 2. *Number of participants included in analysis. For some studies, where numbers of participants per genotype were not provided, we estimated the total number of participants with ϵ2+ or ϵ3/3 genotypes, assuming the distribution of the genotypes to be the average of the included studies with available data on each genotype. This provides a rough estimate of numbers of participants included in the comparison, but has no effect on the study-specific OR, standard error, or meta-analysis calculations. **Study used slightly different comparison groups (ϵ2+ vs ϵ2−). CI = confidence interval.

Two studies had published sufficient data for inclusion in our meta-analyses of the association between APOE genotypes and strictly lobar vs not strictly lobar BMBs.^3,17,21 We sought additional unpublished data from 3 other studies that had not included them in their publications, of which 2 provided them,^18,20 and one did not.² The 4 studies with location-specific data available included 4,883 (66%) of all 7,351 participants. The association between ϵ4+ genotypes and BMBs was slightly stronger and more significant for participants with BMBs in a strictly lobar location vs those with no BMBs (OR 1.35, 95% CI 1.10–1.66, p = 0.005), while it was slightly weaker and nonsignificant for those with not strictly lobar BMBs vs those with no BMBs (OR 1.16, 95% CI 0.89–1.50) (figure 2B). We could not carry out a formal significance test for the difference between these results for strictly lobar and not strictly lobar BMBs since both used the same no BMBs comparison group. However, we noted the 95% CIs for these results to be overlapping. There was no association of ϵ2+ genotypes with either strictly lobar or not strictly lobar BMBs (OR for strictly lobar BMBs 1.20, 95% CI 0.92–1.57, p = 0.2; OR for not strictly lobar BMBs 1.03, 95 CI 0.73–1.46, p = 0.9) (figure 3B).

Of the studies not contributing data to the meta-analyses of APOE associations with BMBs in strictly lobar vs not strictly lobar locations, 2 studies (between them including 469 participants) mentioned that they did not find an association of APOE genotype with lobar BMBs.^15,18 However, one study including 1,920 participants with data on APOE genotype and BMB location did not comment on the associations of APOE genotypes with BMBs by location and was unable to provide these unpublished data for inclusion in our location-specific meta-analyses.² Fail-safe N calculations showed that it would require null studies of around 7,700 and 4,800 participants respectively to bring the associations of ϵ4+ genotypes with any BMBs and with strictly lobar BMBs to a just statistically nonsignificant level (i.e., p = 0.05). When we set the cutoff for statistical significance at p = 0.01, the association of ϵ4+ genotypes with any BMBs became of borderline significance only while the fail-safe N study size for the association with strictly lobar BMBs was around 940 participants.

DISCUSSION

Our systematic review and meta-analyses of genetic associations with BMBs suggest that ϵ4 APOE allele carriers are at higher risk of having BMBs, particularly in strictly lobar brain locations. Our data would also be compatible with an association between ϵ2+ genotypes and strictly lobar BMBs but this could only be considered speculative, since this association was not significant in our analyses. Given the reported associations of APOE genotype with CAA and lobar intracerebral hemorrhage, our data support strictly lobar BMBs as a possible imaging biomarker of CAA.^7,8,10,25 APOE ϵ4 and ϵ2 alleles may increase the risk of developing cerebral small cortical vessel fragility and consequent leakage of tiny amounts of blood particularly within the lobar regions, because patients who had lobar ICH and carried either of these alleles developed new BMBs at increased frequency compared to ϵ3/3 allele carriers.²⁶ Several animal studies have shown that APOE ϵ4 glycoprotein product enhances deposition of β-amyloid peptide within the vessels causing smooth muscle loss and vessel wall thickening with an increased risk of vessel leakage.²⁷ The ϵ2 allele has been demonstrated to increase fibrinoid necrosis of amyloid laden vessels, and to be associated with an increased incidence of newly developed BMBs on follow-up scans.^28,29

By its restriction to fully published studies, our systematic review is potentially subject to publication bias.³⁰ Furthermore, although we sought some additional unpublished data, only 4 studies in around two-thirds of the total number of participants from relevant studies contributed data on the association between APOE polymorphisms and BMB location. The results may therefore be subject to reporting bias (i.e., those studies which found an association between lobar BMBs and APOE genotype may have been more likely to report their location-specific findings). Our fail-safe N calculations would suggest that a substantial amount of data from a null study or studies would be required to render nonsignificant (p ≥ 0.05) the demonstrated associations of the ϵ4 allele with BMBs and with strictly lobar BMBs. However, with a more stringent significance level of p < 0.01 (accounting for multiple testing), the association of ϵ4+ genotypes with any BMBs was only borderline, while the association with strictly lobar BMBs would become nonsignificant with the addition to our meta-analysis of a further 940 participants with a null result. Since this is not implausible, we conclude that the demonstrated ϵ4+ associations are reasonably robust to the potential effects of publication and reporting bias, but not completely secure.

Although variation in imaging characteristics between studies may have led to a variation in the size of associations detected between genotype and BMBs, we did not detect any such heterogeneity. None of the studies included in our review fulfilled all our methodologic quality indicators for a study on the association between genotype and BMBs. Encouragingly, however, the larger studies which provided the most weight in the meta-analyses performed better against our quality indicators, increasing confidence in the overall pooled OR. Furthermore, a low-quality score does not necessarily indicate a poor quality study but one for which important methodologic quality indicators are incompletely reported. The results of our meta-analyses depend critically on the inclusion of the largest study which contributed over half of the participants. Reassuringly, this study was of high quality.^3,21

A further limitation of our study is that analyses for APOE used a dominant genetic model, while it seems likely that an additive model would be both biologically more accurate and statistically somewhat more powerful, and may have improved our ability to detect a significant association with the APOE ϵ2 allele.^11,21 However, since our analyses relied mainly on the availability of published data, we had to use the dominant model used in the publications. While an individual participant data meta-analysis with inclusion of unpublished data from all studies would have allowed us to analyze the data more flexibly with an additive genetic model, it would have required considerably more time and resources, and may well have been possible only for a more limited dataset.

We have therefore demonstrated an association between ϵ4-containing APOE genotypes and BMBs, and have shown that the association with strictly lobar BMBs is reasonably robust. Future studies in this area would be improved by the use of a consistent definition of BMBs⁵ and a robust method of rating BMBs, such as the Brain Observer Microbleed Scale³¹ or Microbleed Anatomic Rating Scale,³² to improve the comparability of different studies and facilitate future meta-analyses. The statistical power of future studies may be improved by treating BMBs as a continuous or at least ordered categorical variable, rather than as a dichotomous variable, as the majority of studies (and so our meta-analyses) have up to now. While only the APOE genotype has been studied in large numbers of people so far within the published literature, genome-wide association studies may soon be able to identify other, similarly important genes.

Appendix

Electronic database search strategies

^aSearch software used: OvidSP_UI03.02.02.104, SourceID 52336.

AUTHOR CONTRIBUTIONS

S.S. Maxwell: drafting/revising the manuscript, analysis or interpretation of data, acquisition of data. Dr. Jackson: drafting/revising the manuscript, study concept or design, analysis or interpretation of data. Dr. Paternoster: drafting/revising the manuscript, study concept or design. Dr. Cordonnier: drafting/revising the manuscript, study concept or design, acquisition of data. Prof. Thijs: drafting/revising the manuscript, analysis or interpretation of data, acquisition of data. Dr. Al-Shahi Salman: drafting/revising the manuscript, study concept or design, analysis or interpretation of data. Dr. Sudlow: drafting/revising the manuscript, study concept or design, analysis or interpretation of data, acquisition of data, statistical analysis, study supervision.

DISCLOSURE

S.S. Maxwell has received funding for travel from the Centre for Clinical Brain Sciences and scholarship support from the Wellcome Trust. Dr. Jackson receives/has received research support from the MRC UK, the Binks Trust, and the Wellcome Trust. Dr. Paternoster has received support from an MRC PhD studentship. Dr. Cordonnier reports no disclosures. Prof. Thijs serves on scientific advisory boards for SYGNIS Pharma AG, MSD, Shire plc, and Boehringer Ingelheim; has received funding for travel from Shire plc and Boehringer Ingelheim; serves on speakers' bureaus for Shire plc, Abbott, Pfizer Inc., Boehringer Ingelheim, and Medtronic, Inc.; receives research support from Servier, Schering-Plough Corp., Sygnis Pharma AG, CoAxia, Inc., Medtronic, Inc., Novo Nordisk, Bristol-Myers Squibb, Pfizer Inc, Shire plc, AstraZeneca, ThromboGenics NV, Eli Lilly and Company, sanofi-aventis, Boehringer Ingelheim, Daiichi Sankyo, Asubio Pharmaceuticals, Inc., FWO Flanders, and Vlaams Instituut voor Biotechnologie (VIB); and holds stock/stock options in Novo Nordisk. Dr. Al-Shahi Salman serves on the editorial board of Stroke and has received research support from the MRC UK, the UK Stroke Association, the Chief Scientist Office of the Scottish Government Health Department, and Chest Heart & Stroke Scotland. Dr. Sudlow serves on the editorial boards of Stroke and BioMed Central Cardiovascular Disorders; receives publishing royalties for Stroke: A Practical Guide to Management, 3^rd ed. (Blackwell, 2008); and receives/has received research support from the Chief Scientist Office of the Scottish Executive Health Department, the Stroke Association, the Wellcome Trust, and the Binks Trust.