Hi Giorgia,
Depending on how many samples you have, you might be better off writing out the normalized counts table from DESeq2 and using that for standard GSEA rather than using GSEA Preranked. As for which output of the DESeq2 ranked list to use though, I would probably recommend the Log2 Fold Change column, or the Test Statistic (Stat) column, you should get reasonable (albeit slightly different) results from either of those. I would generally shy away from using the pValue, even transformed to a signed, -log10(pvalue) the correlation between the magnitude of the significance statistic and the (potential) meaningfulness of the change biologically isn't really there.
An important thing to know is that when constructing these ranked lists for input into GSEA you do not want to filter the list to just "significant" or "highly" differentially expressed genes, you'll want to use all the genes that were expressed in the samples.
Generally, we would expect that a pathway which received a negative enrichment score would be comprised mainly of genes which were downregulated. If you're seeing something different from this it would be helpful to see some screenshots at least of the results plots. Also, what did the ranked list you were using for this result look like?
-Anthony
Anthony S. Castanza, PhD
Curator, Molecular Signatures Database
Mesirov Lab, Department of Medicine
University of California, San Diego