Propagation of Poisson Errors

#1
How do I combine the asymmetrical errors when I subtract my samples from the controls?

I know my low counts follow a Poisson distribution and therefor I cannot simply add in quadrature as per Gaussian errors.

Thanks in advance,
N