I have 4 stupid questions.

What is your elongation temperature with the pfu ?

,

Have you checked the purity of your DNA samples?

,

No primers misannealing?

and

What is the amount of DNA used?

answers to Mickael:

Elongation T° : 72°C (is their any improvement at this stage ?) 

Purity not checked.... but we have optimized by adding DMSO and it's working fine now. We have also increased the annealing t° to 60°C (while the method was published with 53-55°C) and we get rid of the 'parasite' bands. 

Thanks for your additional comments.

Concerning the elongation temperature, I don’t think that it can be really “optimized”; but if I’m remember well, my student has already tried to amplify, with a pfu, at 68°C and the results were surprisingly better than the classical 72°C.

As I have read you again, I realized that you have already performed this amplification with a Taq polymerase. Thus, the explanation based on a potential misannealing of your primers can be switch-off. (Sometime, it is so simple)

Explanation not yet ruled out.

The impact of impurities carried by the DNA extracts could be in too high concentrations, pfu are sometimes too sensitive.

Have you tried to decrease or dilute the amount of matrix used, to limit potential impurities?

To assume a perfect amplification without errors, I'm using a Platinum pfx. But it’s not much better, we have to optimize the PCR conditions to obtain a usable amplication.

Dear Pascal,

I used primers with the length of up to 88 bp for CO1 metabarcoding. We had no success using the pfu taq and even other polymerases with higher proof reading capabilities. When we used a good standard Taq (HotmasterTaq from 5prime) we obtained very good results, and the samples did very well on the MiSeq.

Do your primers work with standard Taq without proofreading? They are usually more forgiving and can cope with PCR problems, were proof reading polymerases can have issues (at least from my experience).

Run your samples with a normal Taq. Your sequencing errors are much higher than PCR errors any way. If you are concerned that PCR errors could be an issue in the data set, run several replicas of the same PCR and pool PCR products. This should reduce the abundance of abundant "false" haplotypes generated by PCR errors.

Also if you want to be really fancy, include two replicas of each PCR reaction in the NGS run and only keep reads wich aper in both of the replicas. This way you can exclude PCR errors independent from the used polymerase. 

I hope this helps you a bit, feel free to ask me if you need more details on the setup I used for my metabarcoding runs!

Kind Regards

Vasco