If one group is one population this will not work with Arlequin AMOVA. There must be more than one population (I might have misread the question however).
There could several reasons for this output. The most likely is there are insufficient SNPs within the cox1 locus for MANOVA to identify a population structure with sufficient statistical robustness. What it requires are a minimum of 3 independent alleles (SNPs), really 3 genotypes. If the SNPs are linked on the same bit of DNA that doesn't count - thats just one genotype. What I suspect is happening is there is a dearth (not enough) genetic information within cox1 to perform MANOVA.
There could be other explanations, but this is the first issue to address. This isn't just Arlequin it is standard theory for any allele frequency based calculations and why this is important is more complicated to explain.
Please note, it is the run-up to Christmas .. okay put Christmas on hold. To respond to the comments ...
Firstly, if you hold the data set it is very easy to identify the number of genotypes. There are past posts on this herein. I am not inclined to describe alternative approaches to an algorithm I've still not yet published :-) ... however it is really easy and needs an empirical test. It is good to assess number of genotypes per se Arlequin will do this analysis ... finding that within the output I dunno.
Extending loci is always good and for mtDNA I would not recommend with 16S nor cox2. rRNA should be avoided in allele frequency calculations, its okay for taxonomic assignment (bacteria rather than mtDNA). cox2 is a small locus. I would recommend for mtDNA
- cobB, formerly called cytB
- intergenic region of mtDNA
cobB is a good target, it has a lot more selection on it for within species analysis than cox1. The intergenic region is rarely used, its a good locus because of the higher rates of mutation therein.