I think the hub concept applies especially well to a protein-protein interaction network, where you have "sticky" proteins, ubiquitins, plymerases, histones, GNGs that are in some cases associated to a lot of other proteins. A co-expression network doesn't necessarily have hubs. 40 connections in particular are not so many for a protein network. In practice, you can plot the distribution of the degree in your network and if you see a local maximum in the high end you can set an empirical threshold there, but it really depends on the kind of biological question you're trying to answer.

I guess you are already familiar with functionally important genes in the context. Look for the number of degrees for those genes. Again it is up to you and data to select these important genes. Take the average of these degrees as your cutoff.

For eg. let us assume that there 180 genes in the list and you know (from literature and validated experiments) 10 genes are important. Look at the degrees they have. Take the average and use that as cut off and be conservative in cut-offs.