De novo assembly, as sdriscoll mentioned, is the typical solution when no decent reference genome exists.
Given that you're dealing with gene expression, I assume you have transcriptome reads, in which case you could look into Trans-ABySS, which is the transcriptome-specific version of ABySS. It is a single software pipeline that aims to assemble reads into transcripts and quantify transcript abundance, all without a reference genome. In theory you would end up with two sets of transcripts and expression levels, after which standard DE analysis could be conducted. Although finding corresponding transcripts between the two sets could be tricky...
Software link is
here and paper is
here