According to the book, The domain layer should be isolated. In domain entity, you should avoid adding a property represents database PK (usually identity surrogate column called ID).
There is no problem in identifying a domain entity because by definition it includes a natural key. If this key is the same as PK, then repository will have no problem in persisting the domain entity using PK. Otherwise, the repository will need to construct a SQL command that find the entity based on some column(s) instead of PK.
Allowing PK to be in domain layer is the perfect approach by book, however I cannot see risky practical issues. On the other hand, without this approach, the saving process for an aggregate might lead to a performance issue in saving.
I can see only one practical problem which is "the wrong guidance for other developers". Do you know other practical problems for this approach?