This seemed to work. Just need to ask for the original read format.

~/tools/sratoolkit.2.4.2-centos_linux64/bin/fastq-dump --origfmt -I --split-files --gzip SRR1514950/SRR1514950.sra

it's probably because this isn't the default way paired reads are usually named so bwa is confused. Try a quick sed:

sed -i 's,.1,/1,g' file1 and sed -i 's,.2,/2,g' file2. You will however need to manually fix the first read in file1 and second read in file2 to

SRR1514950.1/1 and SRR1514950.2/2

Hope this works.

The following command worked for me:

cat sra.fq | perl -ne 's/\.([12]) /\/$1 /; print $_' > sra.fix.fq

I had something similar. The reads I got from SRA look like so:

@SRR1531517.4.1 D3NH4HQ1:58:D091WACXX:7:1101:1448:2140 length=75
AACTTCCAGTGGAAATGAGATTCTGATTCTACCAAAAATGGCCCTCCGAATAGTCAGCATGTAGTTTGTTTGCCC
+SRR1531517.4.1 D3NH4HQ1:58:D091WACXX:7:1101:1448:2140 length=75
CCCFFFFFHHHHGIJIJIJJJJJJJJJJIJJJJJJIJJIJJIGIGIJJIIJIIIIIIJJJJIGIJJJIIJJJHHH

I tried something like this to make it compatible with BWA. It works with both forward and reverse files. I prefer to pipe (and zip) it to another file to keep the original as a backup.

sed 's;@SRR1531517\.\([0-9.]*\)\([0-9]\) \([a-zA-Z:0-9]*\) length=[0-9]*;@\3/\2;' sra.fq | gzip > sra.fix.fq.gz

Which gives me:

@D3NH4HQ1:58:D091WACXX:7:1101:1448:2140/1
AACTTCCAGTGGAAATGAGATTCTGATTCTACCAAAAATGGCCCTCCGAATAGTCAGCATGTAGTTTGTTTGCCC
+SRR1531517.4.1 D3NH4HQ1:58:D091WACXX:7:1101:1448:2140 length=75
CCCFFFFFHHHHGIJIJIJJJJJJJJJJIJJJJJJIJJIJJIGIGIJJIIJIIIIIIJJJJIGIJJJIIJJJHHH