There's no technical reason why you can't. Case-control studies often present ORs and cohort RRs, but this can be got round or ignored. The main problem is that it often introduces heterogeneity because the case-control studies often give very different results (presumably because they are more prone to selection and recall bias). So whilst it is possible to do it, and there will be examples where it introduces little heterogneity (the previous responder's paper may be one example of that), it is often wiser not to combine them, or at very least to stratify on design (as another responder suggested) to explore retrospective/prospective data collection as a potential source of heterogeneity. Then you also learn something extra that helps you interpret the literature.