Side chain assignment and assignment analysis

2.3.2 Side chain assignment and assignment analysis

In FastNMR, the assignment of experimental side-chain resonances is performed automatically by comparison with values predicted from protein folds established at stage A in the FastNMR calculation. Assignments are considered only when the difference between predicted and measured chemical shift is less than 0.3 ppm for protons and 1.3 ppm for carbons. To justify the parameters, tests were done, on Ubiquitin, by predicting the chemical shifts from the protein fold obtained at the end of stage A of FastNMR, that are deviating by 2.4 and 3.5 Å from the native structure, 1D3Z. The predicted values were compared to the experimental chemical shifts obtained from 3D HCCONH- and CCONH-TOCSY experiments recorded on the protein. These tests show that the root-mean-square deviation (rmsd) between predicted and experimental chemical shifts is 0.19 ppm for protons and 1.1 ppm for carbons visible in HCCONH- and CCONH-TOCSY spectra [Meiler 2003;Osapay and Case 1994] (Figure 2.4).

In addition, when two experimentally observed ¹H chemical shifts, which belong to the same residue (as established by HCCONH and CCONH-TOCSY spectra), differ by less than 0.3 ppm, any NOE to either of the two shifts is considered

Figure 2.4:Comparison of predicted chemical shifts with experimental values obtained from 3D HC(CO)NH- and C(CO)NH-TOCSY experiments recorded on ubiquitin. (A)

13C chemical shifts were predicted using a artificial neural network as implemented in the software PROSHIFT [Meiler 2003] from a structure that deviates 2.4 Å from the native structure (PDB code: 1d3z [Cornilescu, Marquardt 1998]). Note that the comparison does not include C and C chemical shifts as these are assigned based on the chemical shift information obtained from backbone triple-resonance NMR experiments.

(B) ¹H chemical shifts were predicted using empirical rules as implemented in the software SHIFTS [Osapay and Case 1994] from the same structure as in A). (C) Same as A) but using a structure that deviates 3.5 Å from the 1d3z structure. (D) ¹H chemical shifts were predicted using empirical rules as implemented in the software SHIFTS from the same structure as in C). The straight lines bracket the range ± 0.3 ppm between measured and predicted ¹H chemical shifts and ± 1.3 ppm between measured and predicted ¹³C chemical shifts.

Table 2.3 NMR chemical shift assignments obtained by FastNMR

as ambiguous (indistinguishable assignment) during the automated NOE assignment.

By using this approach, all experimentally observed carbon chemical shifts of Conk-S1, Conk-S2 and ubiquitin were assigned unambiguously. ¹H chemical shifts, however, are often degenerate and about 10% of the measured side chain ¹H chemical shifts could not be assigned unambiguously.

Previously, it was suggested that for successful automated NOE assignment at least 90% of all proton chemical shifts have to be assigned [Jee and Guntert 2003].

FastNMR in its current implementation, however, only uses 3D CCONH- and HCCONH-TOCSY NMR experiments and only ~60 % of all protons were assigned by FastNMR prior to starting the NOE analysis, as shown in Table 2.3. Hence, in an attempt to increase the number of chemical shifts, we used theoretical chemical shifts.

So, for all protons (including side chains) for which no experimental chemical shifts

a Chemical shifts used to establish sequential connectivity.

b Percentage of assigned experimental spin systems. Assignment scores with respect to the total number of assignable residues are given in brackets.

c Percentage of measured side chain chemical shifts that were assigned by FastNMR. Unambiguous (“numerator”) and ambiguous (“denominator”) assignment scores are listed. Assignment scores with respect to the total number of side chain atoms are given in brackets.

d Start: prior to starting the NOE analysis in stage B of FastNMR.

e Final: upon completion of FastNMR. As part of the iterative NOE assignment and structure calculation, the NOE assignment is selected that produces the best-fit between the experimental NOE distance restraints and those back-calculated from the structure. Thus, during stage B the number of unambiguously assigned proton chemical shifts was increased from 89.7 to 93.7% for Conk-S1 and from 88.5 to 96.7% for ubiquitin. The assignments are highly useful for further NMR studies, such as relaxation time measurements to probe backbone and side chain motions.

were available in 3D HCCONH-TOCSY spectra, FastNMR uses predicted chemical shifts, obtained from SHIFTS, for automated NOE assignment. As the predicted chemical shifts are not very accurate, the window size that is used for matching NOEs to ¹H chemical shifts was increased from 0.05 ppm to 0.3 ppm. In case of carbons, missing chemical shifts are not substituted by theoretical values to reduce the risk of wrong NOE assignments. In addition, NOE distance restraints assigned to protons with predicted chemical shifts are used in the final structure refinement only if the same proton (predicted chemical shift) is assigned to two or more NOE peaks and the experimental chemical shift of the two NOE peaks differ by less than 0.1 ppm. In combination with the backbone conformation that is already established prior to the NOE analysis, this allows the determination of high-resolution protein structures.

FastNMR was applied to NMR data of three proteins with all the inherent difficulties of peak overlap, missing backbone resonances, noise peaks and multiple conformations. NOE peaks with multiple chemical shift assignments are fully taken into account by the use of ambiguous distance constraints. Many different tests were performed to highlight the reliability of FastNMR and to show that FastNMR works for real experimental data. In addition, we tested the impact of lower data quality and wrong backbone assignments. These tests are explained in the following section in detail.