The dynamic properties of a nuclear coactivator binding domain are evolutionarily conserved

Evolution of proteins is constrained by their structure and function. While there is a consensus that the plasticity of intrinsically disordered proteins relaxes the structural constraints on evolution there is a paucity of data on the molecular details of these processes. The Nuclear Coactivator Binding Domain (NCBD) from CREB-binding protein is a protein interaction domain, which contains a hydrophobic core but is not behaving as a typical globular domain, and has been described as ‘molten-globule like’. The highly dynamic properties of NCBD makes it an interesting model system for evolutionary structure-function investigation of intrinsically disordered proteins. We have here compared the structure and biophysical properties of an ancient version of NCBD present in a bilaterian animal ancestor living around 600 million years ago with extant human NCBD. Using a combination of NMR spectroscopy, circular dichroism and kinetics we show that although NCBD has increased its thermodynamic stability, it has retained its dynamic biophysical properties in the ligand-free state in the evolutionary lineage leading from the last common bilaterian ancestor to humans. Our findings suggest that the dynamic properties of NCBD have been maintained by purifying selection and thus are important for its function, which includes mediating several distinct protein-protein interactions.

E volution has shaped proteins into a wide spectrum of structure, stability, and dynamics, with fully disordered proteins at one end of the scale and well-folded globular proteins at the other one 1 . It is becoming clear that dynamic properties per se are important for protein function 1-3 but it is not trivial to prove that the dynamics are essential for biological function and not merely a general property of proteins. The nuclear coactivator binding domain (NCBD) is a small (~50 residues) domain from CREB-binding protein (CBP, also called CREBBP), which is a transcriptional coactivator with histone acetylase activity and is present in all animals 4 . The NCBD domain has a hydrophobic core, but is very dynamic and even verging on being intrinsically disordered [5][6][7] . It has therefore been described as "molten-globule-like" [8][9][10][11] . The fact that NCBD possesses properties of both intrinsically disordered and globular proteins makes it an interesting system for assessing the role of dynamics in protein structure and how it modulates function. We have previously subjected NCBD to "evolutionary biochemistry" 12,13 . Thus, we resurrected and characterized ancestral versions of NCBD and its protein-ligand CBP-Interacting Domain (CID) from the NCOA1, 2, and 3 protein family (with three members in human called Src1, Tif2, and ACTR, respectively) to address the evolution of affinity in this protein-protein interaction. The oldest "maximum likelihood" (ML) NCBD variant we could resurrect was present before the Cambrian period, some 540-600 million years ago, in the common ancestor of all present-day animals with bilaterian symmetry. These animals are divided into deuterostomes and protostomes (D/P) and the ancestral NCBD domain is denoted NCBD D/P ML . It was shown to bind the CID domain with relatively low affinity (K d ∼1-5 μM) whereas more recent variants of NCBD in the vertebrate lineage (ca. 440 million years ago) had acquired a present-day affinity for CID (K d ∼0.1-0.2 μM). Structural characterization of ancestral and present-day CID/ NCBD complexes showed that the increase in affinity observed for more modern NCBD variants was due to a combination of factors. These include several new interactions driving structural and dynamic changes in the complex, such as the formation of a third α helix in CID upon interaction 13 . In the present paper, we investigate the structural and biophysical properties of the ligandfree state of the ancestral Cambrian-like low-affinity NCBD D/P ML and compare it to that of high-affinity present-day NCBD Human to track the evolution of structure and dynamics and their relation to function. The comparison reveals that the overall properties of the ancestral NCBD D/P ML domain are very similar to those of the present-day NCBD Human . In general, conservation implies function. Thus, this conservation of a highly dynamic structure and molten-globule-like properties suggest that these features are subject to purifying selection and hence that dynamics are indeed important for the biological function of NCBD.

Results
The structures of ancient NCBD D/P and extant NCBD Human are similar but not identical. Because of its dynamic properties, it is challenging to solve the structure of ligand-free NCBD 8 . First, we performed experiments with 15 N labeled NCBD D/P ML . We discovered by serendipity that low pH significantly improved the quality of the 1 H 15 N heteronuclear single quantum coherence (HSQC) spectrum resulting in well-dispersed peaks for NCBD D/P ML (Fig. 1), and allowing for a near-complete assignment of the backbone and side-chain residues from NMR 3D experiments using 13 C and 15 N labeled samples. The NMR structure was thus determined at pH 2.4 by measuring 3D 1H-1H 13 C/ 15 N-resolved NOESY experiments for distance restraints determination and 3 J HNHA for dihedral angle determination. We determined a structural ensemble consisting of 20 structures. A minimum of one long-range and 2-3 medium-range NOEs were used per residue to determine the distance restraints, far fewer than the average for folded domain proteins, but sufficient to define the overall global fold. It is worth noting that the low number of NOEs might be the reason we observe differences between the NCBD variants as detailed below. This is also the reason we searched for conditions (different pH) in our NMR structure analyses that gave minimal conformational heterogeneity, thus enough NOE distance restraints. However, it should be noted that four residues (Ser2079, Gln2082, Ala2098, and Gln2109) could not be resolved under any of the conditions tested, limiting the structural information that could be obtained from these regions.
The structure of NCBD D/P ML was found to be overall consistent with that of the free 8 and bound NCBD Human 6 and bound NCBD D/P ML determined at pH 6.8 13 (Figs. 1, 2). However, a detailed comparison between NCBD D/P ML and NCBD Human shows that the orientation of helix 1 (α1) and helix 3 (α3) differ between the structures. Upon structural alignment of NCBD Human and NCBD D/P ML , we observed that both the α1 and α3 from NCBD D/P ML are slightly tilted away relative to that of NCBD Human (Fig. 1). On the other hand, as shown previously 13 , CID-bound NCBD D/P ML and NCBD Human are structurally more similar (Fig. 2).
The amino acid sequence of NCBD D/P ML was reconstructed using phylogenetic methods 12 . The probability is low that the resulting maximum likelihood sequence is identical to the actual sequence present in the ancestor. Thus, NCBD D/P ML is rather one of a large number of likely ancestral variants with similar properties. We wondered if errors in the sequence of NCBD D/P ML resulted in the differences we observe in helix α1 and α3 compared to NCBD Human . To investigate this further, and to test how structurally robust the dynamic NCBD domain is, we performed a control experiment where we expressed, purified, and characterized an alternative NCBD D/P variant denoted NCBD D/P AltAll . In this NCBD D/P AltAll variant, all residues with a posterior probability lower than around 0.9 12 were replaced with the second most likely residue at that position. For example, residue 2107 is His with 87% probability and Gln with 12% probability in the ancestral NCBD D/P . Thus, NCBD D/P ML has a His in position 2107 and NCBD D/P AltAll a Gln residue. NCBD D/P AltAll can be considered a "worst-case scenario" variant and good control of the robustness of conclusions drawn from resurrection experiments. If ML and AltAll variants display similar properties it is very likely that the actual ancestral protein shares these properties as well. Because of this, an AltAll variant is a convenient alternative to combinations of point mutations 14 . NCBD D/P AltAll differs from NCBD D/P ML by amino acid substitutions at six positions (Fig. 1c). Three of the differences between NCBD D/P ML and NCBD D/P AltAll are in the N-terminus, but only one of these is in a structured region, a Gln2065→Pro at the beginning of α1. The other three differences are Ser2078→Asn in the loop between α1 and α2, a solvent-exposed Gln2088→His in α2, and His2108→Gln at the C-terminus. We solved the NMR structure of NCBD D/P AltAll (Fig. S1) and found that it folds into a similar structure as the other NCBD domains. The overall RMSD between NCBD D/P AltAll and NCBD D/P ML was 7.9 Å. As a comparison, the overall RMSD between NCBD Human and NCBD D/P AltAll or NCBD D/P ML were 4.5 and 7.0 Å, respectively.
Detailed inspection of the structures revealed differences in the backbone between NCBD D/P ML and NCBD D/P AltAll , especially in the orientations of the helices. An alignment of the structures of NCBD D/P AltAll , NCBD D/P ML , and NCBD Human, shows differences between α1, α2, and α3 ( Fig. 1, Fig. S1, and Table S1). One reason for the differences is that there are several long-range NOEs in NCBD D/P AltAll that are not present in NCBD D/P ML , between the following pairs of residues: 2068/2082, 2069/2080, 2070/2083, and 2071/2099. There is also an NOE between Ile2062 Hγ and Pro2065 Hα, two of the substituted residues in NCBD D/P AltAll .
There appears to be a consistency in the hydrophobic core of both NCBD D/P ML and NCBD D/P AltAll since the single Phe residue at position 2100 is packed and buried in a similar fashion in the two variants. Phe2100 is stabilized by Leu2090 in NCBD D/P ML , and both Ile2089 and Leu2090 in NCBD D/P AltAll (Fig. 3). Interestingly, this is the same kind of packing that was observed in the free NCBD Human domain 8 , suggesting conserved features in the hydrophobic core. We also compared the structures by calculating the difference in solvent accessible surface area (ΔSASA) between folded and unfolded proteins using ProtSA (Table S1). We obtained values of ΔSASA of 1264, 1795, and 1825 Å 2 for NCBD D/P ML , NCBD D/P AltAll , and NCBD Human , respectively, indicating an overall similar burial of residues for NCBD D/P AltAll , and NCBD Human . The lower ΔSASA for  Fig. 1 Structures of ancestral and human NCBD domains. a 1 H-15 N HSQC spectrum of NCBD D/P ML at pH 2.4 with assigned residues. All assignments and the structural coordinates have been deposited in the PDB data bank with accession number BMRB ID 34635 and PDB code: 7OSR. b A schematic tree depicting the relationship between the deuterostome/protostome (D/P) ancestor and modern species. The time points on the x-axis are approximate, in particular the divergence of protostomes and deuterostomes. c Sequence alignment for NCBD Human , NCBD D/P ML , and NCBD D/P AltAll . The residues differing between NCBD D/P ML and NCBD D/P AltAll are marked in celery green while additional differences to NCBD Human are marked in gray. d, e Comparison of the NMR structure of ancestral NCBD D/P ML solved at pH 2.4 (teal, PDB code: 7OSR), and the previously determined human NCBD (gold, 2KKJ) 8 . Note that the structure of human NCBD was solved using a slightly longer construct and at pH 6.5. d Structural alignment of NCBD D/P ML (teal) and NCBD Human (gold) with the residues that differ between the two variants represented as stick models. e Structural alignment of the 20 lowest energy structures of the NCBD D/P ML domain with those of NCBD Human . Tilt angles for α1 and α3 from the lowest energy structures. See Table S1 for the determination of the tilt angles.
NCBD D/P ML stems from a higher SASA in the folded state, i.e., a less compact folded structure. This result, which is consistent with the overall higher structural similarity between NCBD D/P AltAll , and NCBD Human , may be due to the lack of NOE constraints as discussed above, but might also reflect the dynamic properties of NCBD, where small changes in sequence could modulate packing of helices. In this respect, errors in the reconstruction could affect the details of the structure.
Despite the differences discussed in the previous paragraphs, the overall biophysical properties of NCBD D/P ML and NCBD D/ P AltAll were found to be similar (Fig. S2) corroborating the overall similarity seen with the NMR structures. Specifically, to test the affinity of NCBD D/P AltAll for CID, we expressed and purified ML and AltAll versions of CID 1R , which was previously reconstructed using sequences from present-day vertebrates. Thus, CID 1R represents the domain present in the ancestor of an early vertebrate some 440 million years ago 12 . K d values of all four combinations (NCBD D/P ML and NCBD D/P AltAll with CID 1R ML and CID 1R AltAll ) were between 2-9 μM (Fig. S2a, b), which is in the same range as previous estimates of affinity of the ancestral complex 12,13 . Furthermore, the stability as monitored by urea denaturation (Fig. S2c) and thermal denaturation (Fig. S2d, e) were similar for NCBD D/P ML and NCBD D/P AltAll . Thus, based on the highly similar biophysical properties of NCBD D/P ML and NCBD D/P AltAll we infer that our conclusions are robust with regard to the uncertainty in the reconstructed sequence.
Ancient NCBD D/P is less thermodynamically stable than extant NCBD Human but displays a similar pH dependence with regard to structure and stability. Since the NCBD D/P ML structure was determined at low pH, we next investigated how pH influences the fold and stability. First, we performed 1 H-15 N HSQC-monitored pH titrations of NCBD D/P ML and compared them to those of NCBD Human at different pH values (Fig. 4). The NMR titration experiments showed that NCBD D/P ML is as folded at low pH as it is at high pH values. However, the peaks ( 1 H-15 N) are much sharper (reduced line-width) at lower pH values indicating less conformational heterogeneity. It should be noted that the sharpness of NMR lines does not depend only on conformational heterogeneity but also on the rate of exchange between these conformational states. With respect to the NMR time regime, fast exchange will result in sharp peaks, intermediate exchange will result in broadening of the spectra while slow exchange will result in distinct conformational peaks. Second, the CD-monitored urea denaturation experiments showed identical stability patterns at two pH values (3.0 and 7.4) for both ancient NCBD D/P ML and extant NCBD Human (Fig. 4c, d, Fig. S3, and Table S2). This indicates that while at lower pH the NCBD D/P ML still exhibits the same stability as at high pH values, there is a shift in conformational heterogeneity towards a single state at low pH. Furthermore, whereas the apparent thermodynamic stability differs bỹ 1 kcal mol −1 for ancient NCBD D/P ML and extant NCBD Human , both are largely unaffected by a drop in pH to 3.0 (Table S2), consistent with few titratable groups in this pH range (one His in NCBD D/P ML and one Asp in NCBD Human ). Previous experiments suggested that the ancestral NCBD D/P ML was slightly less stable than younger and extant NCBD variants as judged by urea denaturation experiments at pH 7.5 12 . However, the broad unfolding transition of NCBD resulting from its small hydrophobic core leads to a large error in the m D-N value, which makes it difficult to unequivocally determine the thermodynamic stability  13 and the structure of free human NCBD (PDB code: 2KKJ) 8 were previously determined. a Overlay of the 1 H-15 N HSQC spectra of free NCBD D/P ML at pH 2.4 (teal) and CID 1R ML -bound NCBD D/P ML at pH 6.8 (gray). b Overlay of the 1 H-15 N HSQC spectra of free NCBD Human at pH 6.5 (gold) and CID Human -bound NCBD Human (gray) at pH 6.8. c Structural alignment of free NCBD D/P ML (teal, 7OSR, pH 2.4) and CID 1R ML -bound NCBD D/P ML (gray, 6ES5, pH 6.8). d Structural alignment of free NCBD Human (gold, 2KKJ, pH 6.5) and CID Human -bound NCBD Human (gray, 6ES7, pH 6.8).
of the domain. However, all present results (including experiments on the stabilized NCBD D/P ML variant NCBD D/P T2073W (Table S2) agree well with the previous experiments and we can therefore conclude that the ancestral NCBD D/P ML was less thermodynamically stable than evolutionarily younger variants. Interestingly, with some exceptions 15,16 , ancestral reconstructions often yield proteins with higher thermostability, which correlates with thermodynamic stability, than extant proteins. This notion has been attributed either to a bias for stability in the reconstruction (an artifact) or to higher temperatures in past times [17][18][19][20][21] , like during Cambrian 22,23 . In any case, ancestral NCBD appears to have populated the native state to a lesser extent than present-day human NCBD.
Ancient NCBD D/P ML and extant NCBD Human display similar temperature-dependent structural changes. Previous experiments demonstrated that NCBD Human shows an apparent noncooperative reversible unfolding behavior when subjected to increasing temperatures, as monitored by CD 6,8 . This behavior may result from a low ΔH D-N value rather than a true noncooperative unfolding. (The urea denaturation experiments are consistent with a cooperative two-state unfolding.) We compared 1 H-15 N HSQC and CD spectra, respectively, of NCBD D/P ML and NCBD Human at different temperatures (Fig. 5). In agreement with the previous data, we observed that both variants unfold noncooperatively with temperature as monitored by the increase in molar ellipticity at 222 nm suggesting loss of α-helix as the temperature is increased. However, for both NCBD D/P ML and NCBD Human the signal is not completely lost even at 363 K (90°C) underscoring that this is not a typical native to a highlydisordered-denatured state transition. Furthermore, 1 H-15 N HSQC spectra of NCBD D/P ML at temperatures up to 333 K (60°C) showed well-dispersed peaks indicating a presumably globular, collapsed architecture (Fig. S4), corroborating the notion that NCBD retains structure at high temperatures.
Rate constants for (un)folding of ancient NCBD D/P ML and extant NCBD Human . The folding of human NCBD is complex,  To further compare the biophysical properties of ancient NCBD D/P ML and extant NCBD Human we performed kinetic folding experiments using stopped-flow as well as temperaturejump fluorescence spectroscopy. To be able to monitor structural rearrangements we used a variant of NCBD D/P ML with a Trp at position 2073, previously used in kinetic binding studies. A Trp in the corresponding position was also previously used in the folding studies of NCBD Human 24 . This Thr→Trp modification resulted in a thermodynamic stabilization of 1.2-1.4 kcal mol −1 for both NCBD D/P T2073W and NCBD Human T2073W (Fig. 6a, b, Supplementary Fig. S3, and Table S2). Previous folding experiments were performed on a slightly longer NCBD Human variant (residues 2058-2116), which was used in the original studies of CID/ NCBD 6 . In our resurrection studies, we used a shorter version of NCBD Human , only containing the evolutionarily conserved region (residues 2062-2109). Therefore, we first repeated the temperature-jump folding experiments for NCBD Human T2073W and showed that the shorter construct used in our evolutionary studies displayed similar kinetics as the longer construct (Fig. 6c). We then performed folding experiments with NCBD D/P T2073W . The kinetic transients of NCBD D/P T2073W from temperaturejump experiments had lower signal-to-noise than those of NCBD Human T2073W , but observed rate constants were in the same range (Fig. 6d). To investigate the folding kinetics further we used stopped-flow fluorimetry, which has a lower time resolution but can be used to compare folding rate constants for NCBD D/P T2073W and NCBD Human T2073W at low temperature (Fig. 6e, f). The stopped-flow folding experiments were conducted by making pH jumps from a 5 mM HCl solution (pH ∼2.3) to a buffer solution of 20 mM sodium phosphate (pH 7.4, 150 mM NaCl, 1 M TMAO) and monitoring the relaxation through a 330 nm band-pass filter at 4°C. The low temperature is necessary to reduce the observed rate constant such that the folding or other conformational transitions take place in the millisecond time window. We observed two kinetic phases for NCBD Human T2073W (one fast phase with a negative amplitude and a second slower phase with a positive amplitude) but only one kinetic phase for NCBD D/P T2073W (with a positive amplitude). The values of the observed rate constants for NCBD Human T2073W (∼300 s −1 and ∼70 s −1 , respectively) were similar to those observed previously for the longer construct 24 . The single value obtained for the NCBD D/P T2073W was ∼140 s −1 . Trp fluorescence is a very sensitive but crude structural probe. Thus, the number of kinetic phases in folding experiments could be dependent on small structural rearrangements around the Trp. The structural basis for the kinetic phases is likely shifts of conformational equilibria between two or more folded states, as mentioned above and suggested by previous studies on human NCBD 10,24,25 . Nevertheless, the observed rate constant for NCBD D/P T2073W in the stopped-flow pH jump experiment is in the same range as the two rate constants observed under the same conditions for human NCBD.

Discussion
The biophysics and molecular evolution of NCBD is interesting for several reasons, which all relate to the dynamic properties of NCBD. Firstly, while it has a hydrophobic core and globular shape, NCBD is a very dynamic protein domain with characteristics of an IDP 5,6,[8][9][10][11]26 . Secondly, NCBD has several binding partners including the three NCOA transcriptional co-regulator paralogs 5,27 , the transcription factors p53 28 , Ets-2 29 , and interferon factor-3 30 , and viral proteins such as Tax 31 from the T-cell leukemia virus. Intriguingly, NCBD displays great conformational plasticity as shown by its complex with interferon factor-3 where the helices of NCBD adopt a distinct conformation as compared to the complex with CID 32 . Thirdly, evolutionary snapshots of the complex between CID and NCBD suggest that several new intermolecular contacts and conformational plasticity increased the affinity when going from the Cambrian-like NCBD D/P ML (K d ∼1-5 μM) to younger variants (K d ∼0.2 μM) 12,13,33 . Fourthly, while affinity and structural order of modern NCBD/CID complexes were increased, the transition state of the folding-induced binding displayed more conformational heterogeneity as compared to the ancestral complex 34 . Our present results, how the free, unbound state of NCBD has evolved during the transition from a low-affinity to a high-affinity CID binder relate to all four points.
The most striking finding is that NCBD D/P ML displays very similar molten-globule-like properties as the previously characterized human NCBD 8,10 . This conservation of properties, the borderline between a globular domain, and an IDP suggests that the dynamics of NCBD is preserved by purifying selection and therefore of functional importance. In general, structural plasticity is regarded as one main functional feature of IDPs allowing for multiple protein-protein interactions 35 . Given the high sequence identity between NCBD D/P ML and human NCBD (Fig. 1), it is perhaps not surprising that the molten-globule-like properties have been conserved. Conservation of sequence is dictated by function, i.e., the affinity toward the binding partners of NCBD. Since NCBD interacts with several protein domains, both folded and disordered ones, we speculate that the dynamic nature of NCBD has been conserved to facilitate multi-partner binding. Moreover, NCBD interactions result in coupled folding and binding. Many such reactions have recently been shown to occur via a mechanism denoted templated folding, where the binding partner influences the folding pathway [36][37][38] . Indeed, templated folding was observed for the interaction between CID and ancestral as well as human NCBD 34,39,40 . Thus, our present findings suggest that templated folding may be a general feature of binding reactions with NCBD where the conserved dynamics facilitate sampling of different free states, heterogeneity of bound states, as well as flexibility in the pathway between free and bound states.
Going into details of the NMR model, we note that while the overall structure is robust to errors in the predicted sequence as  shown by the comparison of the NMR models of free NCBD D/P ML and NCBD D/P AltAll , there are differences in for example helix orientations. Globular domains with a sequence identity as high as these NCBD variants (77-88%) would typically have identical structures due to a highly funneled energy landscape 41,42 , while proteins with larger structural heterogeneity have less funneled landscapes 1 . Indeed, the differences between the structures of NCBD Human , NCBD D/P ML , and NCBD D/P AltAll , in particular regarding the angles of the helices in relation to each other, are consistent with a less funneled energy landscape resulting in  Table S2. c Temperature jump kinetic experiments for a longer human NCBD construct used in previous studies 24  conformational heterogeneity 11 . This heterogeneity is also consistent with the conserved dynamic properties of NCBD. In conclusion, our data suggest evolutionarily conserved dynamics of NCBD that underlie functional plasticity in ground state conformations, templated binding and folding, and a propensity to interact with several binding partners.

Methods
Protein sequences, expression, and purification. The reconstruction of ancestral sequences and posterior probabilities of each amino acid residue were previously published 12 . Briefly, around 180 amino acid sequences of NCBD and CID domains, respectively, from present-day species were collected and aligned. The alignment and a species tree were used to reconstruct ancestral sequences using a maximum likelihood method. Of interest for the present paper, maximum likelihood (ML) ancestral sequences for two nodes were reconstructed: CID (and NCBD) of an early vertebrate at the time of the whole-genome duplications, denoted 1 R, and NCBD from the last common ancestor of deuterostome and protostome (bilaterian) animals, denoted D/P. AltAll versions of reconstructed sequences incorporate the second most likely amino acid in all uncertain positions, with a chosen cutoff value. The AltAll versions of NCBD D/P (Fig. 1) and CID 1R used in the present paper was based on posterior probabilities from the reconstruction with a cutoff of 91%. One position in NCBD D/P (Thr2062) is very uncertain. A Thr residue (15%) was originally chosen at this position in NCBD D/P ML to avoid a hydrophobic residue, although Ile (19%) has a higher posterior probability. In the current NCBD D/P AltAll residue 2062 is Ile. NCBD and CID variants were expressed in E. coli from a pRSETA plasmid and purified as previously described 34 . Concentrations were determined by absorbance at 280 nm for NCBD or 205 nm for CID using calculated extinction coefficients based on amino acid composition. Purity was checked by SDS-PAGE and the identity of the purified proteins was confirmed by MALDI-TOF mass spectrometry.
Circular dichroism. Far-UV CD experiments were performed using a JASCO J-1500 CD spectrophotometer with a Peltier temperature control system. The protein concentration in all CD experiments was 18-20 µM. The buffer solution was either 20 mM sodium phosphate (pH 7.4), 150 mM NaCl or 50 mM potassium formate (pH 3.0), 150 mM NaCl, and CD spectra were recorded at 277 K. The scanning speed was 50 nm/min, bandpass 1 nm and integration time 1 s. Each reported spectrum is an average of 3-4 recorded spectra. In the equilibrium thermal denaturation experiments, the proteins were dissolved in 50 mM potassium formate buffer (pH 3.0), 150 mM NaCl, and the sample was heated from 277 to 368 K. The heating rate was 1 K/min, waiting time 5 s and the denaturation was monitored at 222 nm. During the heating, spectra (average of three measurements) were taken at 283, 303, 323, 343, and 363 K, respectively. In the equilibrium urea denaturation experiments the proteins were denatured in 0-6.6 M urea at 277 K and the unfolding was monitored at 222 nm. The data were analyzed in GraphPad Prism and fitted by nonlinear regression to a two-state equilibrium model 43 .
Because of the broad unfolding transition and lack of a native baseline for several variants, four parameters were shared among all urea denaturation data sets in the curve fitting ( Fig. S3 and Table S2). One of these parameters was the molar ellipticity of the native state, with the reasonable assumption that the secondary structures of the native states are similar for all variants. The other parameters were the slopes of the denatured and native baselines as well as the m D-N value (assuming a similar change in the solvent accessible surface area upon denaturation for all variants).
Temperature-jump experiments. Temperature-jump experiments were conducted using 100-200 µM protein on a TJ-64 temperature-jump system (TgK Scientific). The buffer solution was 20 mM sodium phosphate (pH 7.4), 150 mM NaCl. The proteins were subjected to temperature jumps of 2-8.5 K to different target temperatures. The data were fitted in the Kinetic Studio software (TgK scientific) to a double exponential function, to account both for the fast phase (relaxation time ∼80 µs), which corresponds to the heating time of the instrument, and to the slow phase, which corresponds to the conformational transition in the protein. A Trp variant of NCBD D/P ML (NCBD D/P T2073W ) was chosen for kinetic experiments since a Trp in this position was previously used for human NCBD. We engineered a Trp at other positions in NCBD D/P ML and obtained rate constants in the same range for all variants (Table S2). ITC experiments. The ITC experiments were performed using a MicroCal ITC 200 System in 20 mM sodium phosphate buffer (pH 7.4), 150 mM NaCl at 298 K. The different CID variants (concentrations in the syringe varying from 300-500 µM) were titrated to a 10-fold lower NCBD concentration in the cell, and the data were fitted to a two-state binding model to obtain estimates of the equilibrium dissociation constant (K D ), enthalpy of binding (ΔH), and stoichiometry (n).
NMR experiments. NMR experiments were acquired on Bruker 600, 700, and 900 MHz spectrometers equipped with triple resonance cryogenic temperature probes at 298 K except otherwise stated. The final NMR samples contained 500 µM protein to which 0.01% NaN 3 and 5% D 2 O was added. Experiments for assignment and subsequent structure determination were done at pH 2.4 (HCl adjusted with NaOH) for NCBD D/P ML and pH 6.8 for NCBD D/P AltAll . When we compared the 2D HSQC spectra of both proteins at these two pH values (2.4 and 6.8, respectively), we observed that most of the residues have similar chemical shifts. In addition, the CD spectra of both proteins at these pH values also indicate that they are folded to the same extent (Fig. 4). The reason for using the lower pH for NCBD D/P ML is because, in addition to the mono-dispersed peaks around 8 ppm, additional well-separated peaks appear above 8 ppm, which enabled 3D-type assignment. NCBD D/P AltAll on the other hand, showed a mono-dispersed spectrum at pH 2.4, but the spectrum at pH 6.8 was superior to that at pH 2.4 since most signals were intact and as such, 3D assignment was possible for data collected at pH 6.8. The following NMR experiments were recorded for assignment and subsequent structure determination: standard 2D 1 H- 15  Structure calculation. Structure calculations were done using the CYANA 3.97 47 package as follows: Initially, cross-peaks were converted into upper distance restraints following an automated process in CYANA. These distance restraints together with φ/ψ dihedral angles determined from C α -chemical shifts and 3 J HNH α (measured) were used as input for the initial structure calculations. The structures were calculated with 200,000 torsion angle dynamics steps for 100 conformers starting from random torsion angles by simulated annealing. Twenty conformers with the lowest target function values were selected and analyzed. The structural statistics together with all input data for the structure calculations are presented in Table 1. In the figures, the structures with the lowest CYANA target function are displayed, since these were similar to ensemble average structures. Structures with the lowest target functions were used in all analyses. Assignments and structural coordinates have been deposited in the RCSB protein data bank with PDB code: 7OSR (accession number BMRB ID 34635) (NCBD D/P ML ) and 7OSW (accession number BMRB ID 34636) (NCBD D/P AltAll ). The solvent accessible surface area was calculated using the software ProtSA with a 1.4 Å solvent radius 48,49 .
Statistics and reproducibility. The reproducibility of the type of biophysical experiments reported in this paper is generally high, as described in the reporting summary. All data were presented in the paper or in supplementary files.