Ah, I see! Yes, that explains the difference.
If you check the heatmap from one of your former experiments, you'll see that it uses the phenotype labels (from your CLS file) as the column headers. That is, the heatmap is showing the difference in gene expression across the samples. Since the Preranked function omits any phenotypic information, there's no way to create such a heatmap in the context of this type of analysis.
The best advice I could give in this case is to try to produce these heatmaps - with something like R or Python or even Excel - from the upstream data at a point where the samples are still distinct from one another. It doesn't really solve your problem exactly, however, since the GSEA heatmaps show only the genes that map to the given gene set. You would also need to "slice" just those rows out of the dataset as well.
For reference, this particular heatmap uses the input dataset directly as given (unmodified) and generates the heatmap for just those matching rows.
Regards,
David