**1. Introduction**

Protein sequences provide rich information on structural and functional constraints in the form of residue co-variation in evolution. With the rapid expansion of available sequence data and computational power, and improvements in global statistical approaches [1], the problem of transitive residue correlations (false positive correlations observed for residues that do not actually contact each other in space) could be largely overcome and the analysis of evolutionary couplings (ECs) between protein residues has achieved important breakthroughs [2,3]. Several groups have demonstrated that the analysis of sequence co-variation can be efficiently used for predicting protein structures [2–10], including transmembrane proteins [11,12], defining evolutionary units within proteins [13], and identifying contacting residues of interaction partners [14,15], and interacting subunits of larger complexes [14]. The reason why co-variation analysis does not (yet) provide

the ultimate solution to the sequence-based prediction of protein structures is that it requires large, good-quality alignments of sufficiently diverse sequences [16], restricting its applicability to phylogenetically widespread and reasonably conserved proteins.

Intrinsically disordered proteins/regions (IDPs/IDRs) lack well-defined 3D structures, rather, they exist and function as ensembles of rapidly interconverting conformers [17–20]. The conformational variability and adaptability, extended interaction surface, various embedded interaction motifs [21,22], and post-translational modification sites [23] of IDPs make structural disorder indispensable in regulatory [24], complex-assembly [25], and scaffolding [26,27] functions. IDPs are central to cellular interaction networks [28,29] and are frequently associated with human diseases [30].

IDPs mostly interact with their partners through eukaryotic/short linear motifs (ELMs/SLiMs) comprised of a few specificity-determining residues embedded into a disordered sequence environment that ensures the right positioning of their mostly hydrophobic, crucial interaction residues [21,22,31–33]. SLiM-mediated interactions are frequently switched on and off by alternative splicing [31,34–36], and are frequently rewired in evolution [31]. As IDPs lack a well-defined tertiary structure and their functional modules are restricted to a few critical residues, most of their sequences are under limited structural and/or functional constraints. As compared to folded domains, this inherent freedom leads to increased rates [37] and altered types [38] of residue changes in evolution, hampering both the identification and correct alignment of homologous IDPs. Despite their indisputable importance, true IDPs (with the exception of ribosomal proteins [15]) and their interactions were not subject to high-scale residue co-variation analyses [6,7,9,11,12,14,15] until very recently. Although disordered regions have less co-varying residues than folded domains [39], Toth-Petroczy and colleagues have recently refined their method, EVfold, to predict potential structured states of IDPs through detecting ECs within their chains [40]. The prevalent phenomenon of induced folding or disorder-to-order transition [41] in IDP-partner recognition, however, suggests that the (partly) structured states of IDPs are encoded not only in their own sequence but also in that of the partner. It follows then that co-evolutionary signals in the interface of the IDP and its partner may actually be just as pronounced as those within a protein fold or between two folded interaction partners.

Therefore, to see if functional constraints originating from partner binding resulted in detectable interprotein ECs of (at least for certain phylogenetically wide-spread) IDPs and to characterize the respective protein pairs, interfaces, and their co-evolved residue pairs, we performed a targeted screen to identify such cases.
