Limitations and Strengths of the Study

In all studies the study subjects were Caucasians and the results cannot be generalized to other populations.

In original communication I, the main reason for not finding additional signals could be the small number of study subjects. In the largest GWASes hundreds of thousands of individuals are required to significantly find the weakest signals. An obvious strength of the study is the replication of the major results in multiple independent cohorts. There were significant differences in sex distributions and mean ages between the cohorts. However, the association of rs676210 with oxLDL concentration remained significant in spite of these and other differences in the background populations.

The association of rs676210 with LDL oxidation is convincing. The oxidation of LDL was, however, measured by structural changes in the apoB protein and does not fully account for the lipid component of these apoB-containing particles.

To confirm that our results are not due to the possibility of the Pro2739Leu substitution altering the binding of the used monoclonal antibody to apoB, we also applied a second, apoB-epitope-structure-independent assay to assess the effect of the SNP on the LDL oxidation, with parallel results. In this independent assay, the SNP rs676210 was significantly associated with LDL diene conjugation. These results put together show that our findings are not due to the oxLDL assay employed and imply a true biological function for rs676210 in LDL oxidation.

The step-wise algorithm and top-SNP-adjusted GWAS strongly suggest that rs676210 is the proxy SNP. We also carried out manual haplotype analyses, arriving at the same conclusion of rs676210 as the implied proxy. This information, together with the predicted damaging change caused in apoB (proline-to-leucine interchange), indicates that out of the available SNPs, this is the most likely functional top-SNP candidate. The SNP in perfect LD with rs676210 was predicted to be benign. Two other nearby missense mutations were also predicted to be damaging but showed no independent effect in the haplotype analysis nor with the forward-selection algorithm. Other associated missense-mutations were predicted not to be damaging. These results strongly suggest that rs676210 is the true functional variant. However, it is important to note that the true proxy cannot be found with 100% certainty by means of bioinformatic methods, and, therefore, we performed the analyses with the most probable one.

In comparison, the cohorts used for the clinical endpoint association assessments were quite similar. The current analyses related to oxLDL did not include the FINCAVAS or ANGES study populations because oxLDL was not measured in them. The clinical implications of rs676210 leading to Pro2739Leu missense mutation require further investigation—however, we did not find significant associations in a meta-analysis of three independent studies.

In original communication II, the obvious limitation of this study is the lack of replication of the association of rs676210 with ischaemic stroke. To verify the association, it needs to be replicated in a sample similar to LURIC. Moreover, the definitions for cerebrovascular disease differ between LURIC and WTCCC2. The cerebrovascular event definition in LURIC includes all TIAs and strokes. The absence of replication of this association in a general ischemic stroke setting means that there is not clear clinical use for the studied SNP per se in, for example, the prediction of stroke in the general population. However, if the result could be replicated in another cohort of angiography patients, there might be some predictive value for this SNP in patients with suspected CAD. LURIC does not have stroke subtype definition available to verify whether the association is with a specific subtype of ischaemic stroke.

In original communication III, there are a number of potential limitations to the study. Not all patients who had IMT measured also had carotid plaque measured. However this would have tended to reduce power to detect any statistical association with plaque, and we found such an association. The CHARGE consortium includes a number of different populations which introduces heterogeneity; therefore we analysed using a meta-analysis approach and

the associations we found were consistent across almost all populations. In the mRNA expression studies there were relatively small sample sizes, although the upregulation of HDAC9 was still detected in atherosclerotic plaque.

In IV, we used imputed data from the Immunochip platform, meaning we only had access to ~40% of the genome across all centres. Secondly, cases were drawn from a number of international centres, meaning that despite efforts to standardize phenotyping, we cannot rule out differences in screening and clinical ascertainment.

In original communication V, even though the control arteries were microscopically free of atherosclerosis, the control subjects have atherosclerosis at least in cardiac arteries since they are undergoing coronary artery bypass surgery.

The bioinformatic method used in the study is only suggestive tool for localization and for example confocal microscopy or double staining immunohistochemistry would be needed to confirm the results. Moreover, the studied plaques are mostly at advanced stage and the role of the studied genes in the early process of atherosclerosis is not known. The advantage of the used method is that in comparison to other methods it a cost effective way to give rough estimation of localization of expression of genes.