Column generation algorithms are typically adopted to address mathematical programming problems defined over a huge number of variables. This approach suffers, however, from several problems that might limit its usability. In this work some of these problems are discussed along with several strategies that take advantage of the use of (meta-)heuristics to help improve the methods performance and reduce the computational effort required to compute an optimal solution. In this work the benefits of using metaheuristic strategies within CG are discussed from the viewpoint of the way they, indirectly, address some of the causes leading to a poor performance. These different methods are tested by solving the maximum network lifetime problem in wireless sensor networks for which a model that naturally leads to column generation is considered. Experimental results show how the use of metaheuristics can generate large improvements on the performance of the basic column generation framework.