Low DESeq2 Sensivity on scRNA Pseudobulk Samples

I'm encountering an unexpected problem when running DESeq2 on pseudobulk scRNAseq datasets.
I have three replicates per condition, and although some genes show evident differential expression, DESeq2 gives them extremely high p values.
These genes do have low expression, but the size effect is quite reproducible.

Example gene:
imgur.com/a/mFsbD0F

All of the standard diagnostic plots look OK (mean - variance fit, heatmaps, PCA).

Any ideas how to increase the sensitivity of the test would are welcome.