In this paper, we consider a multiuser MISO downlink cognitive network coexisting with a primary network. With the purpose of exploiting the spatial degree of freedom to counteract the inter-network interference and intra-network (inter-user) interference simultaneously, we propose to perform zero-forcing beamforming (ZFBF) at the multi-antenna cognitive base station (BS) based on the instantaneous channel state information (CSI). The challenge of designing ZFBF in cognitive networks lies in how to obtain the interference CSI. To solve it, we introduce a limited inter-network cooperation protocol, namely the quantized CSI conveyance from the primary receiver to the cognitive BS via purchase. Clearly, the more the feedback amount, the better the performance, but the higher the feedback cost. In order to achieve a balance between the performance and feedback cost, we take the maximization of feedback utility function, defined as the difference of average sum rate and feedback cost while satisfying the interference constraint, as the optimization objective, and derive the transmission mode and feedback amount joint optimization scheme. Moreover, we quantitatively investigate the impact of CSI feedback delay and obtain the corresponding optimization scheme. Furthermore, through asymptotic analysis, we present some simple schemes. Finally, numerical results confirm the effectiveness of our theoretical claims.