TopHat "junction" error

[sudarynja3]

Hi I have some wey strange error.
I run ~200 samples with tophat2 and 7 do not run with following error:
___________________________________________________________________
tophat2 -p 8 --read-mismatches 4 -o thaln_HV ~/HV_consensus.fa unmapped.fq

[2012-10-09 12:01:43] Beginning TopHat run (v2.0.0)
-----------------------------------------------
[2012-10-09 12:01:43] Checking for Bowtie
Bowtie version: 2.0.0.6
[2012-10-09 12:01:43] Checking for Samtools
Samtools version: 0.1.18.0
[2012-10-09 12:01:43] Checking for Bowtie index files
[2012-10-09 12:01:43] Checking for reference FASTA file
[2012-10-09 12:01:43] Generating SAM header for HV_consensus.fa
format: fastq
quality scale: phred33 (default)
[2012-10-09 12:01:43] Preparing reads
left reads: min. length=18, count=515457
Warning: short reads (<20bp) will make TopHat quite slow and take large amount of memory because they are likely to be mapped to too many places
[2012-10-09 12:01:58] Mapping left_kept_reads against HV_consensus.fa with Bowtie2
[2012-10-09 12:02:40] Mapping left_kept_reads_seg1 against HV_consensus.fa with Bowtie2 (1/3)
[2012-10-09 12:02:47] Mapping left_kept_reads_seg2 against HV_consensus.fa with Bowtie2 (2/3)
[2012-10-09 12:02:54] Mapping left_kept_reads_seg3 against HV_consensus.fa with Bowtie2 (3/3)
[2012-10-09 12:02:59] Searching for junctions via segment mapping
Warning: junction database is empty!
[2012-10-09 12:03:07] Retrieving sequences for splices
[2012-10-09 12:03:07] Indexing splices
Warning: Empty fasta file: 'thaln_HV/tmp/segment_juncs.fa'
Warning: All fasta inputs were empty
Error: Encountered internal Bowtie 2 exception (#1)
Command: /bin/bowtie2-2.0.0-beta6/bowtie2-build thaln_HV/tmp/segment_juncs.fa thaln_HV/tmp/segment_juncs
[FAILED]
Error: Splice sequence indexing failed with err =1
_____________________________________________

strange thing is if i use --no-novel-junc option it runs ... even though i actually do not specify GTF file .....
my reference is mixed fasta file for metagenomic study and it works fine for other samples. So i assume something is wrong with my *fq file, but they look fine, and reads phred quality is >= 30

anyone can help with the mystery?