A) I think you pointed out the biggest weakens to this method yourself, 18s will not be good enough for what you want to do, meaning different otus based ITS1/2  will have the same 18S seq, this will be especially  problematic if you are not working with phylotypes but with actuall otus (which in my opinion at least for bacterial work is more ecologically relevant, well...as relevant as working with marker genes can be anyway) how will you work around that? 

B) What you can do (what I'm doig at least) is generating a distance matrix by pairwaise alignment of the its seqs against each other (if you are working with mothur the command is pairwise.seqs). That would allow you to at least cluster them to OTUs and work with otu based b-diversity estimators like theat, jaclass or Bray-Curtis.

Hope this helps

Ido. 

Hi Ido

Thanks for your reply... Here are my humble thoughts on what you suggest:

a) When designing an experiment people decide what is more important to them: i) to be able to get a measure of phylogenetic distance such as UniFrac or Faith's Phylogenetic Diversity for a whole community (in which case most choose 18S), or ii) to maximise the ability to discriminate between populations (in which case most choose ITS1 or 2).

I know that populations with different ITS sequences may share the same 18S sequence but this is not necessarily a problem if all you want to do is generate an 18S dataset that corresponds to your ITS sequences so that you can estimate community level phylogenetic metrics. In this case it is almost advantageous that 18S is less specific as it means that there are fewer errors in estimating the 18S sequence that corresponds to an ITS sequence. This approach would allow people with ITS sequences to access the benefits of using 18S i.e. you can produce a global alignment.

b) I have used Robert Edgar's USEARCH algorithm to generate OTUs by pairwise comparisons. Typically I then perform a Hellinger transformation and investigate drivers of beta diversity using PERMANOVA or RDA etc... This is all fine BUT it assumes that the evolutionary distance between all OTUs is equal (i.e. a star phylogeny), which is why I want to generate UniFrac or similar.

I hope this makes more sense... 

One solution is to generate a species "profile" tree using 18S and then within the closely-related members of the species generate additional subtrees using ITS.  Alternately create a synthetic alignment containing both 18S and ITS - a bit messier, and certainly computationally more intense, and at the end of the day probably not more reliable than the first approach.

I do not know the scope of the primers you are talking about, but yeast's systematics relies strongly on phylogenies built using the D1/D2 domain of the LSU (also called 26s rDNA). Using this portion of the gene, relatively broad phylogenetic analysis have been performed on yeasts, such as Kurzman & Robnett (1998) and Fell et al. (2000), whose links I attach.

I have been working recently isolating yeasts and fungi from different resources, and the 26s rDNA also allows very precise identifications, as well as the construction of phylogenies, for filamentous fungi up to the phylum level.

I hope this helps.

Nicolás

http://link.springer.com/article/10.1023/A:1001761008817#page-1

http://ijs.sgmjournals.org/content/50/3/1351.short