This paper presents a general framework for distributed estimation of a dynamical process in a Wireless Sensor Network (WSN) in which the Sensor Nodes (SNs) communicate their measurements to the Fusion Center (FC) via a single hop wireless channel. The SNs adapt their sensing/transmission strategy based on a minimal feedback message provided by the FC, which informs them on the estimation quality achieved. Intuitively, when the estimation quality is poor, the SNs react by sending accurate measurements, at higher cost. When the estimation quality is good, the SNs remain idle to preserve energy. The sensing/transmission strategy of each SN, the channel access, and the feedback from the FC are jointly optimized, with the objective of minimizing the mean squared error at the FC, given a cost constraint for each SN. It is shown that a performance gain can be achieved by exploiting the estimation quality feedback, over policies that do not exploit such information, while achieving scalability to large WSNs. Moreover, a low complexity myopic policy is provided, which is shown to achieve near-optimal performance.