Looking at the principal components of our RNASeq data, there is clear separation between the diseased and controlled, however, this separation is in the 5th principal component, which only accounts for 0.45% of variance. There is no clear separation in the lower dimensions, which mostly show batch separation.

How can I statistically leverage the genes associated with this PC when they aren't differentially expressed in DESeq2? I've attached an image of the plot.

The conventional answer is to adjust for the batch and other unobserved variability in the linear model using e.g., a batch factor and likely additional surrogate variables (using svaseq from the sva package), presuming that batch is orthogonal to your variable of interest.