grep ">" a1.fna  > a1.txt #get descriptions from 1st file
grep ">" a2.fna > a2.txt #get descriptions from 2nd file
comm -13 <(sort a1.txt) <(sort a2.txt) > IDS.txt #suppress lines unique to a1 and lines common in a1.txt and a2.txt

Now you have the list of IDs that are present in a2 only (IDS.txt). This Ids can be used to extract the sequences from a2.fna through various methods (standlaone blast: blastdbcmd, QIIME: http://qiime.org/scripts/extract_seqs_by_sample_id.html, cdbyank http://nebc.nox.ac.uk/bioinformatics/docs/cdbyank.html) or using simple perl or awk scripts. Hope this helps. EDIT: my assumption was that you have fasta sequences of reasonable size(and not reads). Obviously it won't work on large files efficiently.

Here's one way to do it, using Biopython. I'm assuming the IDs itself are enough to identify whether two sequences are the same or not (i.e. two identical sequences always share the same ID and records with different IDs always have different sequences).

from Bio import SeqIO

# get the IDs in a1.fna
a1_ids = set([rec.id for rec in SeqIO.parse('a1.fna', 'fasta')])

# get entries unique to a2.fna
a2_unique = (rec for rec in SeqIO.parse('a2.fna', 'fasta') if rec.id not in a1_ids)

# write unique entries to new file: a2_unique.fna
with open('a2_unique.fna', 'w') as target:
    SeqIO.write(a2_unique, target, 'fasta')

With Biopieces www.biopieces.org) its a variant of this HowTo - the difference being the -i in grab:

read_fasta -i a1.fna | write_tab -k SEQ_NAME -o a1.tab -x
read_fasta -i a2.fna | grab -i -E a1.tab -k SEQ_NAME | write_fasta -o a2_without_a1.fna -x

That doesn't sound like something SAMtools would do. You should be able to write a simple shell or perl script for that though or possibly even simply use a series of UNIX commands. The -v option in JOIN could come in handy, although you'd have to be careful of sequence lines that aren't unique to one particular part of the file.

Hope this helps.