Self-assembly is central to the formation of biological structures and has generated a number of synthetic structures, such as biomaterials and photonic devices with important physical properties. Self assembly is of fundamental importance to soft matter science and synthetic biology. For most colloidal and nanoscale particles, the self-assembly properties are defined by the isotropic interaction energy between the particles. However for biological macromolecules such as proteins, anisotropic interaction energies and multiple degrees of freedom dominate the behaviour, making predictability of solution behaviour and crystallization extremely complex. To probe the impact of anisotropic interactions on the self-assembly of proteins, we have created three double mutants of human γD-crystallin (an eye lens protein) for which the phase diagrams for the single mutants are known. In effect, these double mutants establish a competition between the influences of each individual amino acid substitution, each of which are known to significantly influence the phase behaviour of the protein. Hence, the degree to which the molecular anisotropy of the protein contributes to the behaviour of the double mutant can be robustly examined by measuring phase diagrams for the double mutant proteins. Protein phase diagrams establish the solution conditions in which protein condensed phases such as the dense liquid phase due to liquid-liquid phase separation, protein crystals, protein gels and physical states such as protein aggregates occur. In a series of double mutants, two distinct behaviours associated with each of the two single amino acids substitutions were incorporated into the same protein (1) propensity to crystallize and (2) inverted solubility. Each single mutant differs in either the position of equilibrium phase boundary or the kinetics associated with either aggregate or crystal formation. The first double mutant was the P23VR36S variant of HGD. The phase diagram for this double mutant revealed that the protein formed crystals with normal solubility at very low protein concentrations (less than 1mg/ml). These crystals had an equilibrium phase boundary identical to the R36S single mutant. Given that the inverted solubility associated with the P23V substitution only occurs for aggregates formed at significantly higher concentrations (greater than 10mg/ml), the behaviour of the double mutant in this case is consistent with the behaviour expected from these substitutions. The second double mutant, P23TR36S, again incorporating an amino acid substitution associated with crystallization (R36S) and another associated with inverted solubility (P23T) had a more dramatic and complex phase behaviour. It formed two different crystal types with different morphologies and very significantly, each crystal type displayed different solubility behaviour. One crystal had normal solubility (rod-shaped crystals) and the second crystal had inverted solubility (rhombic-shaped crystals). This type of polymorphism in proteins is a rare occurrence, where the only solution condition change required to interchange between the two crystal forms is temperature. Since the P23T single mutant occurs at significantly lower concentrations than for the P23V single mutant, it was possible to obtain a mutant protein which displayed the two single mutant behaviours within the same protein. The phase diagram also revealed that the phase boundary lines for the two different crystal types were qualitatively consistent with the phase lines for the individual single mutants. Furthermore, the point at which the solubility lines for the two single mutant proteins overlap, co-existence of the two crystal forms is observed. High resolution X-ray crystallography data (2.2Ǻ and 1.2 Ǻ respectively) confirmed that the two crystal forms are polymorphs. Finally, the third double mutant, P23VR58H, exhibits condensed phases associated with both single mutant proteins. The R58H mutant is associated with increased propensity to crystallize, but the kinetics of crystal growth is much slower than for the R36S mutant. Therefore, it was possible to form large aggregates (like those for the P23V single mutant) before crystallization occurred. When crystallization did occur, it was on the surface of the aggregates and not in the bulk solution. The equilibrium solubility line measured for this protein represents the coexistence between protein monomers and the protein condensed phases (crystals + aggregates). This suggests, contrary to the accepted view, that the protein aggregates are as thermodynamically stable as the protein crystal. Kinetic studies on the growth of the protein aggregates and crystals show that the growth of the aggregates occurs by first order growth kinetics and that crystallization occurs independently of the initial aggregate growth. The formation of both condensed phases is independent, but kinetically controlled. These observations, using three double mutants of HGD are striking and are the first time that this approach has been used to examine the role that molecular anisotropy of the protein surface has on the formation of condensed protein phases. While HGD is an effective model to examine these interactions, the work has important implications for protein crystallization efforts in general and for industrial processes where high concentration protein solutions are required such as in the food and biopharmaceutical industries, since anisotropic protein-protein interactions are particularly important in understanding these systems.