Quick answer: you can't. Mainly because of primer bias. Also the different fragment length will introduce bias as well as shorter sequences from the same species will be grouped into different OTUs. Latter could be solved by trimming all your reads to the same region(v4).

It's valid if you're using the same samples and only want to evaluate primer efficiency.

A good alternative to mothur is dada2. It's reference free and avoids clustering into OTUs

here is a good published workflow.

https://f1000research.com/articles/5-1492/v2

best Mina

I settled in using QIIME (closed_reference).