Pega, by default, uses DCAwareRoundRobinPolicy to load balance the queries that it "fires". When no additional settings are provided, DCAwareRoundRobinPolicy auto-discoveres the local datacenter (dc closer to the client) and uses that for all queries.
You can overwrite that behavior by configuring local datacenter manually, by setting a prconfig of key "dnode/cassandra_datacenter" - value of that setting needs to the datacenter that is closer to a particular node. Also, if this setting is put in place, Pega will configure the driver to allow the use of "remote" hosts if no "local" hosts are available.
One last thing that one needs to pay attention to is consistency level (settable via dnode/default_read_consistency and dnode/default_write_consistency prconfig settings). The default is ONE which will allow communication to both DCs. If you want to explicitly require local DC nodes to respond, you can set that to LOCAL_ONE.
In your case, I would suggest Pega nodes in each DC to configure dnode/cassandra_datacenter accordingly.