Exchangeability occurs when the unexposed group is a good proxy (i.e., approximation) for the disease experience of the exposed group had they not been exposed. Of course, we can't know what the disease frequency in the exposed group would have been if they had not been exposed (this is the unobservable "counterfactual"), so instead we choose an unexposed group as a substitute for it. In a case-control study, we select subjects on disease status, not exposure status, so while we conceptually want the unexposed group to represent the disease experience of the exposed group (had they not been exposed), we have to think a little differently. Specifically, the control group should represent the exposure distribution in the underlying source population from which the cases arose. If you over-sample one of the following 2 control groups (E+/D-, E-/D-) in the underlying population (i.e., sample dependent on exposure), then you will end up with a control group that does not represent the prevalence of the exposure in that underlying population, and therefore your effect estimate (OR) will be biased. This is why control group selection is a very tricky and important part of the case-control studies.