I suggest to perform a Blast using only the domains but all your four domain together with the other species especially if contained a minor number of domains. Remenber that you could be have only two duplication events (1+1) and (2+2)

As Matteo said, your best bet is to use all domains together, but I think that BLAST would work well only if they happen to be placed one right after another. Otherwise you might get separate hits for the same protein, which you would have to integrate somehow.

An alternative to this bottom-up approach of finding the genes first, is to go top-down using the protein family databases. Check your protein with Pfam or Interpro and identify the domain of your interest. If you're lucky, the single domain unit will be identified and not the four of them. Then grab all the sequences containing that domain from the database and work off from that.

While I generally prefer to go bottom-up myself, because it's more senstitive and because databases do contain errors, I think in your case it should work just fine, if your analysis is not intended to provide 'the complete and ultimate answer' on the evolution of that particular domain.

In addition to the suggestions above, you can also make a phylogenetic tree of the domain, i.e., you align all four copies of the domain and any orthologs/paralogs. This should also give an idea of the closest copies of the domains to each other.

Hi!

in a previous study (Schuman et al. Plant Phys 2011) we had some 5 domain proteins from a plant and from the phylogeny it was not clear if the domains had been gained successively, as from hmmer searches we had predominantly 2 domain proteins existing in all the other compared species.

Our idea, based on previous studies, was that these domains had been gained more recently (by, for instance, replication slippage), which would have ment that the domains should be more similar within the protein and would have diverged more in sequence when compared to the domains of the other proteins, particularly the ones in the same position. To test this, we estimated the genetic distances among domains within the same protein and among different proteins using protdist from the PHYLIP suite (uses ML to estimate protein distances). Then we compared the distance within domains in the same protein to that among domains in the same aminoacid position (from the alignment) using a permutation test with 1 million permutations. The test statistic was the difference of the mean distance, defined as d=mean(distance of the domains within proteins)- mean(distance of the domains among proteins). To estimate the significance, a P-value was calculated as the relative frequency of all random values greater than or equal to our measured value.

Using this approach we found that domains were more similar among proteins (those in the same aminoacid position) than within the same protein. So we actually rejected our hypothesis, that the domains had been successively gained, and confirmed that they had been lost in the other proteins analyzed or had been acquired by blocks separately. After looking at the alignment we went for domain loss, as we found "degenerate domains" at the positions were a domain should have been, but that we couldn't detect using hidden markov models (like Pfam or Interpro).

hopefully this helps...

An alignment of the global sequence including the modules will not help you to deduce the order in which the duplications took place.

As Anurag proposed, I suggest to perform alignments of each domain in a large sample of species including basal clades. Using a ML approach, you will have an idea on the relatve time of divergence for each duplication event. Of course there will be uninformative sequences, too divergent from the rest. It is the approach we used in such a purpose in the case of WIFKKN/GASP evolution in Metazoans. Maybe this paper will help you.

MONESTIER et al, 2012. GASP/WFIKKN proteins: evolutionary aspects of their functions. PLoS ONE 7(8): e43710