*2.4. Constraints and Nb23 Structure Calculation*

Given the lack of assignment for a number of Nb23 sidechain resonances, an alternative strategy was employed to collect necessary constraints for restrained modeling. The CS-Rosetta server was used to provide a model for Nb23 in order to facilitate the search for experimental constraints. CS-Rosetta uses chemical-shift-constrained homology modeling to outline a 3D protein structure, based on the prediction of backbone and side-chain dihedral angles from the amino-acid sequence and the analogy of the experimental chemical shifts with those of a characterized model ensemble derived from PDB and BMRB [21]. The CS-Rosetta run generated 40,000 models of Nb23. The Cα-Root Mean Square Deviation (Cα-RMSD) was calculated for all of the models with respect to the lowest energy structure, yielding an averaged Cα-RMSD of 1.53 ± 0.99 Å for the ten best structures, calculated over the fragments 1–102, 117–122. Residues 103–116, coinciding with the tentative location of CDR3 loop, were considered as a flexible region. The CS-Rosetta run was deemed as successful as it achieved a Cα-RMSD below 2 Å for non-flexible regions for the ten lowest energy structures and the run converged towards a single structure.

The average β-structure content of the CS-Rosetta models was 49.2%, comparable to the β-structure content of TALOS-N and CD. The β-strand positions also coincided well with the TALOS-N β-strand positions except between residues 57 and 60, where β-secondary structure was consistently absent in the models.

Given the good agreement between the TALOS-N estimates, CD spectroscopy results, and the CS-Rosetta models regarding the β-secondary structure content, as well as the satisfactory Cα-RMSD for the ten best structures, the CS-Rosetta models were deemed as representative of Nb23 for the residues 1–102 and 117–122, and used as prior knowledge for NOE-constraint identification. The conformation of the CDR3 (residues ~101–116) was however not defined for the CS-Rosetta models and was not used for the same purpose.

A 3D <sup>15</sup>N-1H NOESY HSQC spectrum, and aliphatic and aromatic 3D <sup>13</sup>C-1H NOESY HSQC spectra, were acquired in order to extract NOE constraints for structure determination. Complementary 2D <sup>1</sup>H-1H NOESY spectra were also acquired using unlabeled protein samples. Besides the attribution difficulties deriving from the missing sidechain assignments, the NOE identification was also hampered by resonance overlap and critical signal-to-noise ratio due to progressive decrease of protein concentration. The total number of NOE constraints extracted from the spectra using automated and manual assignments, handled by means of the software PONDEROSA [22,23], with prior knowledge from CS-Rosetta models was limited (619), first because of the lack of extensive assignment for the aliphatic and aromatic sidechains, and second because of selection of only unequivocal correlations. This apparently "minimalist" approach was adopted because the structural restraining was already based on the experimentally constrained models of CS-Rosetta, that included 734 chemical shift values constraining 353 dihedral angles. Nonetheless, very characteristic NOE patterns for β-secondary structure types [24] concerning backbone atoms were identified for most residues expected to be found in β-strands as per the chemical shift indexing analysis. Hydrogen bonded amides were also identified by recording a <sup>15</sup>N-1H HSQC spectrum one week after transferring the protein to D2O. This allowed for identification of slowly exchanging amide protons which are involved in secondary structure formation or are otherwise hydrogen bonded [25]. In that spectrum, the backbone NHs of 18 residues were characterized as slowly exchanging, all of which were expected to occur in secondary structure elements as per the chemical shift indexing analysis. The corresponding H-bonds were thus added as distance restraints (the relative list is reported in Supplementary Materials, Table S1). The 20 best NOE-restrained structures were validated with the tools of the PDB Validation Service [26–28] (see Supplementary Materials) and subjected to energy minimization as described in the Materials and Methods section. The ensemble of the ten lowest energy and most similar structures was retained. The relative validation report can be found in the Supplementary Materials.

A summary of the structural features and violations of the CS-Rosetta ensemble, the 20 NOE-restrained structures, and the ten NOE-restrained energy-minimized ensemble is shown in Table 2.

**Table 2.** Summary of features and violations for the CS-Rosetta ensemble, the NOE-restrained ensemble, and final NOE-restrained and energy minimized ensemble.


**Table 2.** *Cont.*


\* The pariwise Cα-RMSD for the respective ensembles, as well as the pairwise Cα-RMSD between the CS-Rosetta ensemble and the final NOE-restrained and energy minimized ensemble, are reported in Table S2 in the Supplementary Materials.
