In this paper http://www.ncbi.nlm.nih.gov/pubmed/25732064, the aligned the protein sequences with distant homology using hmmalign, then converted to nucleotide  sequences, and constructed the tree using modified clearcut. Maybe it helps.

If you use a method like hhalign for the HMM-HMM alignment the output can be converted into Nexus format and put into MrBayes. If you don't have HMMs I've found Clustal Omega + hmmbuild handy to create initial core HMMs of more closely related sequences, which can be aligned HMM-HMM style or extended with distantly related sequences. - The latter can be done with one or more rounds of hmmalign + hmmbuild, using hmmalign to yield the gapped/aligned sequences for tree generation

 Great! Thanks Tim and Yunzeng. 

I also thought of hhalign. I even tried downloading the hhsuite but it seems to ned Linux. I will try the NEXUS conversion + MrBayes. 

I also found this MRFalign (Markov Random Fields http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003500)

as alternative to HMM. What do you think of MRF?

Many thanks again!