I do not think it is acceptable to publish while intentionally leaving out the bootstrap supports... Doing so is (IMO) essentially misrepresenting the certainty of your tree.

Bootstrapping generates pseudoreplicates of your data (by randomly sampling with replacement, to create new datasets of the same size), which you then re-run the analysis on. Having a low bootstrapped support for a node in your tree means that the node was found in a low proportion of your pseudoreplicates... And thus your tree is very weakly supported by the data.

Without considering the support of an individual node, how can we be sure that the relationship it represents is not due to random chance? Does your data in fact support this relationship (e.g. is there enough information), or is the final topology merely an artifact of forcing a bifurcating relationship where the data does not in fact support one?

By what threshold are you collapsing nodes, following bootstrapping? 50% support? What criterion are you using (strict consensus, majority rule consensus)? If you are using a strict consensus than that might explain why all resolution is collapsed...

If you are using a realistic threshold though, and nodes are collapsed, than perhaps it is that your data does not support further resolution..

Just my take

If your data do not support a tree topology, then summarising them as one is clearly misleading, especially if the tree is presented without support values. You might be better off showing a splits network for the data, or perhaps even an ordination plot.

Support values are essential. As I understand it, bootstrap analysis does not collapse nodes per say. It is constructing a consensus tree that collapses nodes. Nodes collapse when there is conflict (poor support) at a node. When it comes to morphology, bare in mind that characters can be coded as additive or non-additive (which can influence the result). Also, choice of outgroup is important to polarise the character states (determine which are plesio- or apomorphic). Then, exclude all non-informative characters because they artificially bump-up Retention and Consistency Index values, etc.  I used WinClada & NONA for parsimony analyses on a Windows pc.

http://www.cladistics.com/wincDownload.htm

http://www.cladistics.com/aboutNona.htm

Thanks Tyler, Thomas and Matts for your suggestions. the nodes may be collapsing due to polytomy. I will look for more characters and try to revise the data matrix.