A fly in the ointment was in the code for our $s (@stops). The code wouldn't work here with a 'my' declaration. 'our' was necessary.

This is a bug that was corrected in Perl version 5.18 IIRC. With this correction, lexical variables always work as expected in  "(?{ code })" and  "(??{ code })" regex constructs.

The dynamic regex form was necessary because the count of the quantifier changed for each iteration of the 'for' loop ($s-1).

I don't see the necessity here. Except for the fact that aliasing into the  @stop array makes calculating the quantifier a bit tedious, it can all be written normally, given that the  s/// match regex is, by default, re-compiled on each  s/// execution:

c:\@Work\Perl\monks>perl -wMstrict -le
"my @stops = (2,6);
 ;;
 my $tag = '___';
 ;;
 for ('ATCGGATCTGGC', 'A-C-G--CTGGC') {
   my $seq = $_;
   printf qq{'$seq' -> };
   ;;
   for our $s (@stops) {
     local our $q = $s - 1;
     $seq =~ s/
       ((?:[TAGC][^TAGC]*){$q} [TAGC])
       /$1$tag/x;
     }
   print qq{'$seq'};
   }
"
'ATCGGATCTGGC' -> 'AT___CGGA___TCTGGC'
'A-C-G--CTGGC' -> 'A-C___-G--CTG___GC'
[download]

And except for say, it works under Perl version 5.8.9. See also Re: Regex to match range of characters broken by dashes Update 2 for another for-loop example.

Update: I've based my code example on your original code, prior to adding the second  s/// fixup.