Negative binomial regression - which relies on the log-link function - models the expected value of Y (given Xs) as an exponential function. Does this mean that negative binomial regression captures nonlinear (i.e., exponential) relationships?