[Question] TopHat aligns more than 1 times

[g781]

I found the accepted_hit.bam has more reads than reads from original fq file. Some articles have discussed that TopHat would hit against genome one time at least so that #reads from accepted_hit.bam are so much bigger than #reads from original fq file.

For my case, I have 42 millions PE-reads in fq files but there are 212 millions reads (106 millions for in average).

1. My question is why TopHat would align more than once?
2. Does people runs cufflinks before filtering out accepted_hit.bam? I pop out this question is that I found there are only ~10% (10.8m/106m) properly PE-reads from accected_hit.bam after filtering by samtools.
3. I know the Cufflinks uses the FPKM as gene's expression based on #fragment (I assume the fragment comes from PE-read). If so, how dose Cufflinks calculate the FPKM using properly SE-reads.
4. I've compared the TopHat with BWA. I know TopHat is good for detecting on splicing isoform. However, the species I aligned is splicing isoform less (~6% of genes have 2 or more isoform). I would expect the number of properly PE-reads from both tools isn't too much different, but the BWA has 6.6 millions PE-reads more than TopHat has.

Those questions are puzzling me a lots. Any answers will be appreciated.

Many thanks in advance.

--
Wei-Ming