All messages they send to the network include a corresponding digital signature verifying the message.
A potential attacker impersonates an existing peer Bob with good reputation and starts flooding the network with fake messages as if coming from Bob. All other peers will have to check the digital signatures of the fake messages before such messages get discarded. What is the mechanism to prevent such flooding and subsequent consumption of CPU resources of those peers who check the signatures?
This is discussed in the last section of the paper.
Spamming of messages across the network
Peers can simply not accept, and more importantly as it is a p2p protocol not relay messages for peers with negative or low ratings. This gives everyone the chance to start from 0 and grow their influence through their rating.