Virulence mechanisms typically evolve through the continual interaction of a pathogen with its host. In contrast, it is poorly understood how environmentally acquired pathogens are able to cause disease without prior interaction with humans. Here, we provide experimental evidence for the model that Legionella pathogenesis in humans results from the cumulative selective pressures of multiple amoebal hosts in the environment. Using transposon sequencing, we identify Legionella pneumophila genes required for growth in four diverse amoebae, defining universal virulence factors commonly required in all host cell types and amoeba-specific auxiliary genes that determine host range. By comparing genes that promote growth in amoebae and macrophages, we show that adaptation of L. pneumophila to each amoeba causes the accumulation of distinct virulence genes that collectively allow replication in macrophages and, in some cases, leads to redundancy in this host cell type. In contrast, some bacterial proteins that promote replication in amoebae restrict growth in macrophages. Thus, amoebae-imposed selection is a double-edged sword, having both positive and negative impacts on disease. Comparing the genome composition and host range of multiple Legionella species, we demonstrate that their distinct evolutionary trajectories in the environment have led to the convergent evolution of compensatory virulence mechanisms.