DSE 2009 Question 39: Linear regression and OLS

Suppose least squares is used to fit $y_{i}=\beta_{1}+\beta_{2}x_{i}+\varepsilon_{i}$, with not all $x$'s identical. Assumptions: (1) $E(\varepsilon_{i})=0$; (2) $\text{Cov}(x_{i},\varepsilon_{i})=0$; (3) homoskedasticity $\text{Var}(\varepsilon_{i})=\sigma^{2}$; (4) no autocorrelation $\text{Cov}(\varepsilon_{i},\varepsilon_{j})=0$ for $i\ne j$. When are the least-squares estimators unbiased?