If you want to detect differentially expressed genes, the minimum is two replicates for each condition. Less than that and you won't be able to compute pvalue associated with the difference of expression. More replicates is better of course. Reference genome is not a control (you'll just map the reads on it) so you'll need healthy samples as controls.

The preferred amount is something like 3 samples of affected and 3 samples of healthy controls if you can possibly swing it, making sure you are sampling from a relevant tissue type of course. However, in rare disease research that isn't always possible. You may only have one affected individual. So the answer becomes "as many as you can reasonably sample" if the numbers are less than 3.