RNA crystal improvement with definitive screening designs

Abstract

When one or more crystallization leads have been obtained from prior knowledge or sparse matrix screening, the next step is determining which experimental factors are essential for crystal growth. This task is often done by varying one or two factors with evenly spaced factor levels, often at great expense in time and material. The Design of Experiments (DOE) approach offers experimental designs that can simultaneously vary from three to many factors with a relatively small number of samples. However, the interpretation of the results requires the fitting of linear models. Traditional DOE screening designs include two - level fractional factorials (introduced to protein crystallography by Carter and Carter in 1979) and optimal experimental designs where three or more factors are varied (introduced to protein crystallography by Carter and Yin in 1994). Most factors in vapor diffusion experim ents cause a non-linear response in crystal quality and size. The non -linear respo nse requires three factor levels to be detected. The newer Definitive Screening Designs (DSDs) have three factor levels (Jones and Nachtsheim 2011). We were attracted to DSDs because t hey require roughly half the samples the corresponding optimal designs require. Here, we apply these designs to identify essential factors in the crystallization of several RNAs to optimize crystal size fo r single-crystal diffraction studies with synchrotron radiation. We used the hanging drop method for crystallization by vapor diffusion. We used the longest dimension of the largest crystal in a drop as our response variable. We used Response Surface Methodology (R SM) to identify the active factors in DSD experiments with 3 to 8 factors. The DSD experiments enabled us to elim inate the unimportant factors from downstream crystal size optimization experiments, saving us time and material. We envision an efficient workflow in which we screen experimental factors by using a DSD after sparse matrix screening and before optimizing the factors' levels with an optimal design or grid screens. After optimization, we replicate