I don't recall BITNET being IBM sponsored, it was just many schools had IBM Mainframes. The root nodes (think tree with two main trunks) were PSUVM and CUNYVM. Lose the PSU <--> CUNY link, and things backed up fast with BITNET basically split in two.
The original topology was odd due to telcom leased line costs at the time; for example, it was cheaper for Clarkson University (Potsdam, NY) to go to UCONN (in CT of course) than a third the distance to Syracuse University (also in NY).
Once sites started getting NSFNet (Internet) connections and a RSCS line driver was written to run encapsulated on TCP/IP, the topology of the network was strengthened with sites connecting to whomever they wanted.
Using UREP (Unix RSCS emulation program) from PSU or a VMS based third party package, UNIX and VMS systems could also join the BITNET.
Likewise IBM MVS systems could join, although VM RSCS was more better suited to the task. For example, to allow TSO users to send remote messages, it required a system hack to have users effectively issue a JES2 remote $DM command. MVS systems were also prone to drop things on the floor if they didn't know how to route a source node even if it knew the receiver location; VM was more gracious, with better error messages and later default routes.
The BITNET core application protocols were standard IBM NJE messages and files. Those protocols supported simple user at node addresses, where both were simple eight character tokens. Transmission was NJE oriented store and forward; if you had notify enabled for a file, you would see a message for every hop a file took from your host to the remote receiver.
(Messages were dispatched over the links at a higher priority than files. This drove system programmers nuts, because file transmission notifications, one-on-one chatting and the original RELAY could cause resource starvation for files. I think CUNY hacked their RSCS to eat file notify messages because of the bandwidth, and more than one node disallowed access to RELAY.)
Mail on Bitnet itself was either NOTE (based on SENDFILE) or simple TEXT format, both with simple user at node syntax.
With any BITNET to off network addressing, the user software (such as UCLA's mail client, I think) would send do a sendfile to a batch SMTP (BSMTP) gateway, such as SMTP at WISCVM.
Mail from remote networks such as ARPANET to BITNET generally used (depending on the sender and gateways) user%node.BITNET@gateway.domain.name The gateway of course received it and forwarded it to user at node. Some Internet sites hacked their mailer so a user could specify simply user@site.BITNET, but it still went through a gateway.
Of course, as campuses got the internet locally, they would install a local SMTP gateway and generally talk to the Internet that way. TCP/IP connectivity killed most WAN applications of both NJE and UUCP, although they each took time to fade.
Note: Written this morning in response to a comment about BITNET on a retro software forum.