It doesn't say in the legend (or the one for Fig. 2) From the provided branch support values (please note: boostrap resampling and PP provide branch support values, not "nodal support": one counts the number of taxon bipartitions, i.e. internodes, in the pseudoreplicate or Bayesian tree sample – common semantic error), I suppose "-" stands for < 50/0.5? If so, why not just give the actual values?
Also, they are missing for all root-proximal branches.
With respect to the occassional topological differences in comparison to Fig. 4 using the same gene sample, one could have looked into the reason for ambiguous support using consensus networks: is it low-amplitude signal (all other alternatives with very low support) or internal signal conflict (two/three competing alternatives)?