Scaling Number of Coordinator Nodes

Hi,

I'm looking at how to scale out ES clusters and have been thinking about how to go about deciding on the number of dedicated coordinator nodes to be used.

I've noticed that coordinator nodes are useful in shielding data nodes from larger queries and in general absorb some of the CPU/memory that would normally be used.

What I'm not sure on is how to go about deciding on the number of coordinator nodes to be used, or if that too needs to be scaled over time as a cluster grows.

Is there any guidance on the number of coordinator nodes that should be considered when scaling out a cluster? At the moment I use a setup in AWS in which one coordinator node exists in each availability zone of the cluster. Would there be a point at which adding extra coordinator nodes to the pool would make sense, or would it be preferable to vertically scale the coordinator nodes as they come under more load?

Thanks