I am fairly sure it doesn't rely on the DRM itself, as you can have books without DRM. Plus I use DeDRMed only books (with plugboards applied) and it still gets synced. But it probably needs you to use the file Amazon provides since they don't use the same conversion calibre does, plus it will
for sure need to have a matching ASIN -- which you are already doing, right?
Conversion adds in all kinds of changes, many in an unpredictable manner. One thing I know is that synced annotations data relies on location, which is calculated in some obscure manner that takes formatting into account. So when calibre converts and organizes the book's css formatting, it changes the precise location calculations. It is possible that throws off your Kindle enough that Amazon cannot recognize the book as being the same book (even if, for the purposes of collections, it is recognized) causing it to fail.
Or perhaps your Kindle recognizes the book when it comes to collections, but not when matching the book for the sake of collections.
This is totally a guess, though, and good luck getting Amazon to confirm anything!
Someday I may reverse-engineer the whole of the Kindle and answer your question. (And many others, no doubt!)
I cannot remember why you needed to convert the books, but as far as the metadata/plugboards applied by calibre, it is not necessary to convert the book first, as calibre should apply those changes when sending Amazon's (imported) files to your Kindle, and according to my experiences, neither the Kindle nor Amazon should notice the difference, but of course if you need to edit the books to fix typos, as many people do, you need to convert in order to edit. Once the book is converted, it is no longer Amazon's book, and we had to use the ASIN trick just to get collections working, butI have no idea where we go from there to getting any other similarities recognized.