Community Structure and Membership

Hi all

I am comparing the gut microbiome structure and membership of fish from two different locations.

After running thetayc and jclass distance metrics, I have analyzed the PCoA plots.

The structure (thetayc) PCoA plots show no significant difference (after unifrac, amova) and that all data points data points cluster tightly (apart from 1 or two outliers).

However, the membership (Jclass) plots show a significant difference (after unifrac,amova) and some slight clustering of data points according to location.

My question is that is it possible to get community structures that are not significantly different, but that memberships can show significant differences?

Thank you

yes, jaccard upweights the importance of rares (because your most abundant OTU and you singletons are both 1). Thetayc does nearly the opposite, abundant nonshared OTUs are strongly penalized. Bray-curtis is somewhere in the middle of those two. If you are seeing differences in membership in different locations you could talk about a seed bank.

Hi kmitchell

Thank you for your helpful reply.

Am I correct in saying that Thetayc is a measure of community structure (i.e. combination of membership and abundance), and Jaccard a measure of membership (i.e. list of OTU’s in a community)?

Is this a good combination of distance matrices to use in my case?

I just want to compare the structure and membership of microbiota between these two fish populations, with the aim of establishing whether they are significantly different.

The fish are of the same species, just raised in different locations.

Thank you so much again

I think looking at both structure (jclass) and membership (thetayc) is a perfectly reasonable thing to do and can lead to ecological insights.

Sorry, just to clarify…

I thought that thetayc is a measure of the dissimilarity between the structures of two communities, as outlined here (

Whereas jclass measures the dissimilarity in membership?

Just want to be clear what each of these is measuring?

Thank you

That is correct. I would add that membership based statistics like Jclass are highly sensitive to incomplete sampling. I tend to avoid those in preference for structure-based metrics like thetaYC.