Well, what do you want to find out about them specifically? - Generally speaking if you want to do an RNA seq experiment, you'd be carrying out a differential expression test at the gene or transcript level. What you're proposing (by the sounds of it), is to take a particular sample type (brasinosteroids in Arabidopsis) from three different sources..... then what? - You'd need to take the control samples from each of those experiments too.

In short, you're going to have a number of issues to deal with. The only saving grace of this, and that would make it potentially work, is that the samples in each experiment are the same, and prepped in relatively the same way too. If that's the case (you'll have to do some reading to make sure), then you could potentially use an additive model in DESeq2 for gene level differential expression analysis (~ Treatment + Effect). The reason you include a term for "Effect" is because there'll be huge amounts of variation between experiments, as such you need to control for this. You can only control for this effectively if you have samples of the same type within all levels of your "Effect" factor. This may not capture all the technical variation, but it's the best you can really do.

The alternative option is to treat each experiment independently, and treat each of them independently. This means that you can do the differential expression test for each GEO Accession, apply some filters and get a gene list, then look at the overlap between experiments.