Co-evolutionary analysis accurately predicts details of interactions between the Integrator complex subunits

Solving the structure of large, multi-subunit complexes is difficult despite recent advances in cryoEM, due to remaining challenges to express and purify complex subunits. Computational approaches that predict protein-protein interactions, including Direct Coupling Analysis (DCA), represent an attractive alternative to dissect interactions within protein complexes. However, due to high computational complexity and high false positive rate they are applicable only to small proteins. Here, we present a modified DCA to predict residues and domains involved in interactions of large proteins. To reduce false positive levels and increase accuracy of prediction, we use local Gaussian averaging and predicted secondary structure elements. As a proof-of-concept, we apply our method to two Integrator subunits, INTS9 and INTS11, which form a heterodimeric structure previously solved by crystallography. We accurately predict the domains of INTS9/11 interaction. We then apply this approach to predict the interaction domains of two complexes whose structure is currently unknown: 1) The heterodimer formed by the Cleavage and Polyadenylation Specificity Factor 100-kD (CPSF100) and 73-kD (CPSF73); 2) The heterotrimer formed by INTS4/9/11. Our predictions of interactions within these two complexes are supported by experimental data, demonstrating that our modified DCA is a useful method for predicting interactions and can easily be applied to other complexes.