We study the problem of data collection in Wireless Sensor Networks (WSN). A typical WSN is composed of wireless sensor nodes that periodically sense data and forward it to the base station in a multi-hop fashion. We are interested in designing an efficient data collection tree routing, focusing on three optimization objectives: energy efficiency, transport capacity, and hop-diameter (delay). In this paper we develop single- and multi-hop data collection, which are based on two definitions of node centrality: centroids and balance nodes. We provide theoretical performance analysis for both approaches, present their distributed implementation and discuss the different aspects of using each. Most of our results are for two-dimensional WSNs, however we also show that the centroid-based approach is asymptotically optimal in three-dimensional random node deployments. We also show several simulation results that support our theoretical findings.