If you have a secondary that you temporarily take out of the replica set (say for maintenance, index rebuilds, etc.), then when you start up that secondary in standalone mode, the TTL monitor is going to run, possibly deleting some documents while it goes. If this happens, the secondary is going to become out of sync with the primary. In most cases it will just be missing some documents that are being deleted from the primary anyway. However, if any of those missing documents are updated on the primary before they're deleted, this is going to cause replication to stall, and the only way to fix it will involve manually editing the oplog.
Proposed solution is to not start the TTL monitor if mongod is started without the --replSet option but there are documents in local.system.replset.