How many genes you have and which software you are using. Try PAUP to infer phylogenetics

Hello B.A. Padder.

I do not know how many loci are there in the dataset, I know that is a gene family, so I used non-specific primers for the PCR. I am using MEGA for the Neighbour Joining (NJ), Garli for the Maximum Likelihood (ML) and MrBayes 3.2 for Bayesian Inference.

Poor Bootstrap is not related to duplication, it is related to phylogenetic information available in your dataset. If the phylogenetic information is weak, you will not be able to resolve the tree. 450bp is too little information, which might have been complicated due to lack of proper variation. Only choice in this scenario will be to increase your dataset and sequence a longer fragment, especially the one which is know to be phylogenetically informative. You can test your data set by using a software called PUZZLE.

I agree with Prof. Goel, you cannot get high resolution/bootstrap with only 450 bp... I doubt that other methods strongly improve your bootstrap.

Also, you may need to check or improve your alignment, especially if you have high number of sequences, but I guess that with 450bp, your dataset is well-aligned.

If you have several species/genera among the 150 DNA sequence you should check the phylogeny of each to see which species is close to which species, this will give you information to align manually these species.

Thank you very much. I will improve the sequence length then.

Hi Juan - This is an important question, and it looks like you have some good information so far in the answers posted, I would agree in general that duplication is not the most likely problem for lack of support in deeper nodes of your tree reconstruction, and that you likely can improve this situation by adding more characters. But this will help resolve topology and add support only if the additional data increases the phylogenetically informative signal. Adding more data of the same quality (or lack thereof!) will not help. Luckliy you know the divergence time for the lineages in question, as you've stated: 26 MYR. Best approach therefore is to set out to add markers that will likely increase signal at that time scale, i.e. faster markers will approach or reach saturation, so protein coding and other "fast" mtDNA sequences (ND4, COI, II, III, Control Region) are not likely to help. I'd suggest looking to nuclear markers such as 18S, 28S, or even H3, as these tend to retain phylogenetic signal at around the relevant divergence time. But also keep in mind, differences in substitution rate can be hard to predict as different lineages have unique evolutionary histories, differences in replication error correction, generation times, etc. Best of luck!

Up front you stated the taxa divergence time est was 26 mya. It is very likely you have saturation effects. This is the core issue with the bootstrap when it is unable to resolve sister relationships. 1) You can test for sequence saturation using Xia's Saturation Test via his software, DAMBE. 2) He also has a very nice transition and transversion vs divergence graph to visualize the effect. 3) MEGA6 provides a measure of saturation (I've calibrated it with DAMBE; it is consistent) called Transition/Trasnversion bias (R). Workarounds: What you may wish to do is run a fully conservative ML analysis and isolate the most distantly related branches to partition your analysis such that you maximize phylogenetic signal, thus R increases. When this happens bootstrap values will improve. Isolate groups of sequences by how closely related they are to build sub trees. This enables you to set aside the most distantly related sequences and still capture signal to answer questions with more closely related branches. Thus, depending on your question(s) you may not need to add more genes.

You have two separate problems in your dataset. Gene duplication means that you are actually aligning two or more loci that from one point in time onwards evolved independently in the same genome. So before you do any phylogenetic analysis where you want to recover lineage phylogeny, you should separate the paralogs, if that is possible. Ideally, if you have duplicates in each lineage, you would get a tree in which the same taxa or even samples appear twice in two different clades (if you are able to sequence duplicates from the same sample), depending how long ago the duplication happened and how far the duplicates evolved from each other. The more recent the duplication, the more complex the picture, especially if duplication happened only in some lineages but not in others, or you randomly recover one duplicate or another in different samples of the same taxon.

The bootstrap problem is not necessarily related to duplication. Instead, it indicates that your selected gene (dataset) has conflicting information in different portions or sites that result in different phylogenies, due to saturation effects or homoplasies (these might actually be affected by merging duplicated genes in the same dataset). When bootstrapping, portions (sites) of the alignment are randomly over- or underweighted, resulting in conflicting topologies for each bootstrap replicate and hence overall low bootstrap values because in some replicates one topology will be favored and in other replicates another. This problem can either be inherent to the locus you selected or stem from problems with the alignment (especially with ambiguously aligned regions). What locus did you actually use? I don't know of any gene that would give consistently low resolution and support below 50% across the entire tree, unless your taxon sample includes only very recently diverged lineages which apparently is not the case. The length of the locus is not really relevant, you can get very high support and resolution with phylogenies based, for example, on just the ITS1 region (less than 300 bp).

Another problem affecting bootstrap support are chimeras, i.e. sequences assembled from two different taxa or loci (which can also happen when you have gene duplication). In this case, you often get long branches leading to a clade but these have very low support, because the clade includes at least one chimera. This is similar to merging two or more loci into a single dataset and accidentally combining two genes from two different taxa into a single taxon. What happens than is that the best-scoring tree usually recovers the correct topology whereas most bootstrap replicates recover incorrect topologies and as a result the overall support is very low, sometimes close to 0%.