Chapter 13RDCs in NMR Structural Biology

This chapter continues our discussion of the residual dipolar coupling method in solution NMR spectroscopy, and then presents some related computational topics.

Residual Dipolar Couplings

The nuclear Overhauser effect (NOE) in NMR spectroscopy provides a local distance constraint, and the scalar coupling provides a local dihedral angle measurement. Both NOEs and scalar couplings have limitations for measuring structural information between atoms far apart. On the other hand, residual dipolar couplings (RDCs) provide global orientational restraints between remote internuclear vectors, and thus give a potential solution to these limitations.

Residual dipolar coupling, measuring anisotropic spin interaction, between spins ii and jj is described as follows:

Dij=μ0γiγjℏ4π2⟨rij3⟩⟨3cos⁡2θ−12⟩,(13.1)D_{ij} = \frac{\mu_0 \gamma_i \gamma_j \hbar}{4\pi^2 \langle r_{ij}^3 \rangle} \left\langle \frac{3 \cos^2 \theta - 1}{2} \right\rangle, \quad (13.1) where γi\gamma_i and γj\gamma_j denote the gyromagnetic ratios of spins ii and jj , and rijr_{ij} denotes the internuclear distance, and θ\theta denotes the angle between the internuclear vector and the axis of the external magnetic field.

When we only consider the coupling between covalently bound spins, the internuclear distance is essentially a constant. Thus, the coupling DijD_{ij} between spin ii and jj depends only on the orientation of the corresponding internuclear vector, that is, the orientational information of the internuclear vector related to the molecular coordinate frame can be obtained from the residual dipolar couplings.

Each experimentally measured RDC can be formulated by the following equation [1]:

d=DmaxvTSv,(13.2)d = D_{\max} v^T S v, \quad (13.2) where DmaxD_{\max} is a constant, and vv is the internuclear vector orientation relative to an arbitrary coordinate frame, and SS is the 3×33 \times 3 Saupe order matrix [1]. Moreover, SS is a real-valued, symmetric, traceless, rank 2 tensor with 5 degrees of freedom.

The tensor SS can be diagonalized according to the following equation:

S=UΔUT,(13.3)S = U \Delta U^T, \quad (13.3) where U∈SO(3)U \in SO(3) is a 3×33 \times 3 rotation matrix, representing the principal order frame (POF), and Δ\Delta is a 3×33 \times 3 diagonal and traceless matrix consisting of the eigenvalues of SS . Note that UT=U−1U^T = U^{-1} .

Computational Topics Related to RDCs

Assignment Problem

Problem Statement Given the experimental RDCs did_i and the internuclear bond vectors vjv_j of a known protein structure, how can we assign each RDC did_i to its corresponding internuclear vector vjv_j ?

We could try to assign each pair of did_i and vjv_j with weight wij=|di−DmaxvjTSvj|w_{ij} = |d_i - D_{\max} v_j^T S v_j| . Then the assignment problem can be formulated into a maximum bipartite matching, which can be solved in O(n3)O(n^3) time by using the Kuhn-Munkres matching algorithm. Here, nn denotes the number of residues.

Notes

Suppose that we know k≥5k \geq 5 pairs of unambiguous assignments between did_i and vjv_j , that is, we are given the following algebraic equation:

(d1⋮⋮dk)=DmaxVTSV,(13.4)\begin{pmatrix} d_1 \\ \vdots \\ \vdots \\ d_k \end{pmatrix} = D_{\max} V^T S V, \quad (13.4) where VV is derived from (v1,…,vk)(v_1, \dots, v_k) . Given (d1,…,dk)T(d_1, \dots, d_k)^T and VV , we can find the tensor SS in equation (13.4) using singular value decomposition (SVD).

Structure Determination Problem

Problem Statement Given an alignment tensor SS and experimental RDC data, how can we find conformation vectors such that the backbone (ϕ,φ)(\phi, \varphi) angles best fit the experimental RDCs [7]?

Suppose that we are given the experimental RDC data in two alignment media, that is, we have two alignment tensors, which give two orientational constraints on each internuclear vector of a certain type, such as NH. Since each RDC in one medium yields a quartic constraint, the corresponding bond vector must lie on the intersection of two conic curves [7].

Let (x,y,z)(x, y, z) denote the coordinates of the unit bond vector in a principal order frame of one medium. Based on two RDC constraints and the rotation transformation between two principal order frames of two media, we can obtain a polynomial in xx in which only the degrees of 0, 2, 4, 8 appear. Thus, we obtain a quartic in x2x^2 , and there are only a finite number of exact solutions for each internuclear bond vector, which generates a finite and discrete set for the final conformation space. This is because univariate quartic polynomials can be solved exactly, in closed form. Based on the computed NH (or other) bond vectors, we can calculate the (ϕ,ψ)(\phi, \psi) torsion angles by using inverse kinematics [7].

Estimation of Alignment Tensor without Assignments

Problem Statement Suppose we know the eigenvalues of the alignment tensor SS . Then given unassigned RDC data and a known protein structure, how can we estimate the alignment tensor without knowing assignments?

When the internuclear vectors are distributed reasonably uniformly in orientation (chapter 6), the magnitude and rhombicity of a molecular alignment tensor can be obtained without assignment from a histogram of experimental RDCs [8]. This immediately yields an estimate of the eigenvalues. This technique is sometimes called the “powder pattern.”

Suppose that the eigenvalues Δ\Delta can be estimated using the powder pattern [1, 3]. Let UU be the rotation matrix of the principal order frame, and DD be the set of unassigned experimental RDC data. Let B(U)B(U) denote the set of back-computed RDCs, that is, B(U)=ETSE=(ET(UTΔU)E)B(U) = E^T S E = (E^T (U^T \Delta U) E) , where EE is the set of internuclear bond vectors. Now we are going to find

argminU∈SO(3)KL(D,B(U)),(13.5)\underset{U \in SO(3)}{\text{argmin}} \quad KL(D, B(U)), \qquad(13.5) where KL(⋅)KL(\cdot) stands for the Kullback-Leibler distance. This approach finds the rotation that brings the two distributions (experimental and predicted) into best agreement.

When UU is obtained, the alignment tensor can be computed by equation (13.3).

Alternative Approach We can also estimate the alignment tensor in an alternative way, that is, we can find the optimal tensor that makes the unassigned bond vectors, computed from the experimental RDCs, best match the geometric pattern of known structure bond vectors. The geometric matching distance can be measured by Hausdorff distance. This approach appears to require more RDCs per residue, or more aligning media.

Structural Homology Detection

Problem Statement Given the unassigned experimental RDC data from a protein of unknown structure, can we apply the algorithm in section 13.2.3 and rank the known structures in a database based on their value of KLKL in equation (13.5)?

Given a known model (structure in PDB format), we can back-compute the RDCs based on the estimated result of the alignment tensor in section 13.2.3. By measuring the Kullback-Leibler distance between simulated RDCs and experimental RDC data, we can compare the distributions of RDCs and see how close the unknown structure is to any known structure.

Alternative Approach We can also measure the structural homology in another way. We first compute the unassigned bond vectors based on the alignment tensor obtained in section 13.2.3. Then we can rank the known structures in the database by measuring the geometric matching distance, such as Hausdorff distance, between the orientational distributions of unassigned bond vectors and known structures. This approach appears to require more RDCs per residue, or more aligning media.

In the next chapters, we will discuss some solutions to these algorithmic problems in more detail.

References

  • [1] J. A. Losonczi, M. Andrec, M. W. F. Fischer, and J. H. Prestegard. Order matrix analysis of residual dipolar couplings using singular value decomposition. Journal of Magnetic Resonance 138 (1999): 334–342.
  • [2] C. J. Langmead and B. R. Donald. An expectation/maximization nuclear vector replacement algorithm for automated NMR resonance assignments. Journal of Biomolecular NMR, 29;2(2004):111–138.
  • [3] C. J. Langmead, A. K. Yan, R. H. Lilien, L. Wang, and B. R. Donald. A polynomial-time nuclear vector replacement algorithm for automated NMR resonance assignments. RECOMB (2003): 176–187; Journal Computational Biology 11;2–3(2004):277–298.
  • [4] C. J. Langmead and B. R. Donald. 3D structural homology detection via unassigned residual dipolar couplings. Proceedings of IEEE Computational Systems Bioinformatics Conference (CSB). Stanford University, Palo Alto (August 10, 2003) pp. 209–217.
  • [5] C. J. Langmead and B. R. Donald. High-throughput 3D structural homology detection via NMR resonance assignment. Proceedings of IEEE Computational Systems Bioinformatics Conference (CSB). Stanford University Palo Alta, (August, 2004) pp. 278–289.
  • [6] C. J. Langmead and B. R. Donald. An improved nuclear vector replacement algorithm for nuclear magnetic resonance assignment. Tech. Rep. TR2004-494, Dartmouth Department of Computer Science, 2004.
  • [7] L. Wang and B. R. Donald. Exact solutions for internuclear vectors and backbone dihedral angles from NH residual dipolar couplings in two media, and their application in a systematic search algorithm for determining protein backbone structure. Journal of Biomolecular NMR, 29;3(2004):223–242.
  • [8] G. M. Clore, A. M. Gronenborn, and A. Bax. A robust method for determining the magnitude of the fully asymmetric alignment tensor of oriented macromolecules in the absence of structural information. Journal of Magnetic Resonance 133(1998):216–221.