Combining Evolutionary Covariance and NMR Data for Protein Structure Determination

Modeling Alternative Conformational States of Pseudo-Symmetric Solute Carrier Transporters using Methods from Machine Learning

The Solute Carrier (SLC) superfamily of integral membrane proteins function to transport a wide array of solutes across the plasma and organelle membranes. SLC proteins also function as important drug transporters and as viral receptors. Despite being classified as a single superfamily, SLC proteins do not share a single common fold classification; however, most belong to multi-pass transmembrane helical protein fold families. SLC proteins populate different conformational states during the solute transport process, including outward open, intermediate (occluded), and inward open conformational states. For some SLC fold families this structural "flipping" corresponds to swapping between conformations of their N-terminal and C-terminal symmetry-related sub-structures. Conventional AlphaFold2 or Evolutionary Scale Modeling methods typically generate models for only one of these multiple conformational states of SLC proteins. Here we describe a fast and simple approach for modeling multiple conformational states of SLC proteins using a combined ESM - AF2 process. The resulting multi-state models are validated by comparison with sequence-based evolutionary co-variance data (ECs) that encode information about contacts present in the various conformational states adopted by the protein. We also explored the impact of mutations on conformational distributions of SLC proteins modeled by AlphaFold2 using both conventional and enhanced sampling methods. This approach for modeling conformational landscapes of pseudo-symmetric SLC proteins is demonstrated for several integral membrane protein transporters, including SLC35F2 the receptor of a feline leukemia virus envelope protein required for viral entry into eukaryotic cells.

Exploring the Alternative Conformation of a Known Protein Structure Based on Contact Map Prediction

The rapid development of deep learning-based methods has considerably advanced the field of protein structure prediction. The accuracy of predicting the 3D structures of simple proteins is comparable to that of experimentally determined structures, providing broad possibilities for structure-based biological studies. Another critical question is whether and how multistate structures can be predicted from a given protein sequence. In this study, analysis of tens of two-state proteins demonstrated that deep learning-based contact map predictions contain structural information on both states, which suggests that it is probably appropriate to change the target of deep learning-based protein structure prediction from one specific structure to multiple likely structures. Furthermore, by combining deep learning- and physics-based computational methods, we developed a protocol for exploring alternative conformations from a known structure of a given protein, by which we successfully approached the holo-state conformations of multiple representative proteins from their apo-state structures.

MIM-ML: A Novel Quantum Chemical Fragment-Based Random Forest Model for Accurate Prediction of NMR Chemical Shifts of Nucleic Acids

We developed a random forest machine learning (ML) model for the prediction of 1H and 13C NMR chemical shifts of nucleic acids. Our ML model is trained entirely on reproducing computed chemical shifts obtained previously on 10 nucleic acids using a Molecules-in-Molecules (MIM) fragment-based density functional theory (DFT) protocol including microsolvation effects. Our ML model includes structural descriptors as well as electronic descriptors from an inexpensive low-level semiempirical calculation (GFN2-xTB) and trained on a relatively small number of DFT chemical shifts (2080 1H chemical shifts and 1780 13C chemical shifts on the 10 nucleic acids). The ML model is then used to make chemical shift predictions on 8 new nucleic acids ranging in size from 600 to 900 atoms and compared directly to experimental data. Though no experimental data was used in the training, the performance of our model is excellent (mean absolute deviation of 0.34 ppm for 1H chemical shifts and 2.52 ppm for 13C chemical shifts for the test set), despite having some nonstandard structures. A simple analysis suggests that both structural and electronic descriptors are critical for achieving reliable predictions. This is the first attempt to combine ML from fragment-based DFT calculations to predict experimental chemical shifts accurately, making the MIM-ML model a valuable tool for NMR predictions of nucleic acids.

A coarse-grained approach to NMR-data-assisted modeling of protein structures

The ESCASA algorithm for analytical estimation of proton positions from coarse-grained geometry developed in our recent work has been implemented in modeling protein structures with the highly coarse-grained UNRES model of polypeptide chains (two sites per residue) and nuclear magnetic resonance (NMR) data. A penalty function with the shape of intersecting gorges was applied to treat ambiguous distance restraints, which automatically selects consistent restraints. Hamiltonian replica exchange molecular dynamics was used to carry out the conformational search. The method was tested with both unambiguous and ambiguous restraints producing good-quality models with GDT_TS from 7.4 units higher to 14.4 units lower than those obtained with the CYANA or MELD software for protein-structure determination from NMR data at the all-atom resolution. The method can thus be applied in modeling the structures of flexible proteins, for which extensive conformational search enabled by coarse-graining is more important than high modeling accuracy.

Simultaneous Assignment and Structure Determination of Proteins From Sparsely Labeled NMR Datasets

Sparsely labeled NMR samples provide opportunities to study larger biomolecular assemblies than is traditionally done by NMR. This requires new computational tools that can handle the sparsity and ambiguity in the NMR datasets. The MELD (modeling employing limited data) Bayesian approach was assessed to be the best performing in predicting structures from sparsely labeled NMR data in the 13th edition of the Critical Assessment of Structure Prediction (CASP) event—and limitations of the methodology were also noted. In this report, we evaluate the nature and difficulty in modeling unassigned sparsely labeled NMR datasets and report on an improved methodological pipeline leading to higher-accuracy predictions. We benchmark our methodology against the NMR datasets provided by CASP 13.

Improving integrative 3D modeling into low- to medium-resolution electron microscopy structures with evolutionary couplings

Electron microscopy (EM) continues to provide near-atomic resolution structures for well-behaved proteins and protein complexes. Unfortunately, structures of some complexes are limited to low- to medium-resolution due to biochemical or conformational heterogeneity. Thus, the application of unbiased systematic methods for fitting individual structures into EM maps is important. A method that employs co-evolutionary information obtained solely from sequence data could prove invaluable for quick, confident localization of subunits within these structures. Here, we incorporate the co-evolution of intermolecular amino acids as a new type of distance restraint in the integrative modeling platform in order to build three-dimensional models of atomic structures into EM maps ranging from 10-14 Å in resolution. We validate this method using four complexes of known structure, where we highlight the conservation of intermolecular couplings despite dynamic conformational changes using the BAM complex. Finally, we use this method to assemble the subunits of the bacterial holo-translocon into a model that agrees with previous biochemical data. The use of evolutionary couplings in integrative modeling improves systematic, unbiased fitting of atomic models into medium- to low-resolution EM maps, providing additional information to integrative models lacking in spatial data.

Accurate and cost-effective NMR chemical shift predictions for proteins using a molecules-in-molecules fragmentation-based method

We have developed an efficient protocol using our two-layer Molecules-in-Molecules (MIM2) fragmentation-based quantum chemical method for the prediction of NMR chemical shifts of large biomolecules. To investigate the performance of our fragmentation approach and demonstrate its applicability, MIM-NMR calculations are first calibrated on a test set of six proteins. The MIM2-NMR method yields a mean absolute deviation (MAD) from unfragmented full molecule calculations of 0.01 ppm for 1H and 0.06 ppm for 13C chemical shifts. Thus, the errors from fragmentation are only about 3% of our target accuracy of ∼0.3 ppm for 1H and 2-3 ppm for 13C chemical shifts. To compare with experimental chemical shifts, a standard protocol is first derived using two smaller proteins 2LHY (176 atoms) and 2LI1 (146 atoms) for obtaining an appropriate protein structure for NMR chemical shift calculations. The effect of the solvent environment on the calculated NMR chemical shifts is incorporated through implicit, explicit, or explicit-implicit solvation models. The expensive first solvation shell calculations are replaced by a micro-solvation model in which only the immediate interaction between the protein and the explicit solvation environment is considered. A single explicit water molecule for each amine and amide proton is found to be sufficient to yield accurate results for 1H chemical shifts. The 1H and 13C NMR chemical shifts calculated using our protocol give excellent agreement with experiments for two larger proteins, 2MC5 (the helical part with 265 atoms) and 3UMK (33 residue slice with 547 atoms). Overall, our target accuracy of ∼0.3 ppm for 1H and ∼2-3 ppm for 13C has been achieved for the larger proteins. The proposed MIM-NMR method is accurate and computationally cost-effective and should be applicable to study a wide range of large proteins.

Principles for Integrative Structural Biology Studies

Integrative structure determination is a powerful approach to modeling the structures of biological systems based on data produced by multiple experimental and theoretical methods, with implications for our understanding of cellular biology and drug discovery. This Primer introduces the theory and methods of integrative approaches, emphasizing the kinds of data that can be effectively included in developing models and using the nuclear pore complex as an example to illustrate the practice and challenges involved. These guidelines are intended to aid the researcher in understanding and applying integrative structural methods to systems of their interest and thus take advantage of this rapidly evolving field.