My preference for input to PCA is definitely normalised data that follows a Gaussian / binomial distribution. In RNA-seq, normalised data follows a negative binomial distribution, so, it 'requires' a transformation to logged (e.g. logCPM (EdgeR) or regularised log (DESeq2)) data prior to running PCA. The additional scaling step can still be used, as it helps to 'iron out' any extra creases in the data.
As you have qPCR data, you should check its distribution and then make a decision. Logging it is certainly feasible. If you log it and additionally convert it to Z-scores outside of PCA, then you should switch off the further scaling step in
For checking the the relationship of genes to each PC, you need to look at the
rotation object that is returned by
pca <- prcomp(t(x))