Solution NMR: A powerful tool for structural and functional studies of membrane proteins in reconstituted environments

A third of the genes in prokaryotic and eukaryotic genomes encode membrane proteins that are either essential for signal transduction and solute transport or function as scaffold structures. Unlike many of their soluble counterparts, the overall structural and functional organization of membrane proteins is sparingly understood. Recent advances in X-ray crystallography, cryo-EM, and nuclear magnetic resonance (NMR) are closing this gap by enabling an in-depth view of these ever-elusive proteins at atomic resolution. Despite substantial technological advancements, however, the overall proportion of membrane protein entries in the Protein Data Bank (PDB) remains <4%. This paucity is mainly attributed to difficulties associated with their expression and purification, propensity to form large multisubunit complexes, and challenges pertinent to identification of an ideal detergent, lipid, or detergent/lipid mixture that closely mimic their native environment. NMR is a powerful technique to obtain atomic-resolution and dynamic details of a protein in solution. This is accomplished through an assortment of isotopic labeling schemes designed to acquire multiple spectra that facilitate deduction of the final protein structure. In this review, we discuss current approaches and technological developments in the determination of membrane protein structures by solution NMR and highlight recent structural and mechanistic insights gained with this technique. We also discuss strategies for overcoming size limitations in NMR applications, and we explore a plethora of membrane mimetics available for the structural and mechanistic understanding of these essential cellular proteins.

Membrane proteins (MPs) 4 constitute 30% of genomes in both prokaryotes and eukaryotes. Essential for cell homeosta-sis, they are specifically involved in signal transduction, solute transport, post-translational modification and as scaffold proteins maintaining the overall integrity of the outer envelope. MPs are classically divided into subclasses based on their location in the membrane (Fig. 1), viz. (a) integral (or intrinsic) MPs (IMPs), spanning one or both leaflets of a lipid bilayer; (b) lipoproteins, anchored to the membrane through a posttranslationally-modified lipid (myristoyl/palmitoyl/prenyl) or glycolipid (glycosyl-phosphatidylinositol); and (c) peripheral (or extrinsic) MPs, interacting with IMPs at the aqueous/ membrane interface. This review mostly focuses on IMPs (unless specified otherwise), a subclass notoriously recalcitrant to structural investigation. Typically, IMPs are classified into the following two subgroups, largely based on the secondary structure of their transmembrane region: most eukaryotic IMPs are ␣-helical proteins, with one or more transmembrane (TM) ␣-helices interconnected through flexible loops, and Gram-negative bacilli and eukaryotic mitochondrial organelles predominantly house ␤-barrel porins in their outer membrane. Albeit few, there are outliers to this classification: autotransporters, ␤-barrel porins with ␣-helical complexes partially inserted within their cores; secretion pores, for example Wza, responsible for the transport of capsular polysaccharide across the outer membrane. A complete catalogue of available MP structures can be found at the "Membrane Proteins of Known Structure" database curated by Dr. White's lab at the University of California at Irvine.
It is hard to undermine the complexity of basic principles that govern the structure and function of IMPs, especially when one considers the diverse milieu of lipid molecules that surround them. Not surprisingly, IMPs constitute a meager fraction of structures in the PDB (less than 4%), although they constitute more than 50% of therapeutic drug targets (1). G-protein-coupled receptors (GPCRs), with seven TM ␣-heli- The authors declare that they have no conflicts of interest with the contents of this article. 1  cro REVIEWS ces, represent the most elusive yet sought after subgroup in drug development (2). For a complete list of GPCRs, validated as potential therapeutic targets, the database maintained by Dr. Gloriam's group at the University of Copenhagen, in collaboration with the EU COST Action "GLISTEN" (http://gpcrdb. org) 5 serves as a good resource. In this review, we focus on structural data and functional studies carried out by solution NMR of IMPs in general.
In the current era, robust technological developments in the fields of X-ray crystallography, NMR, and cryoEM have enabled an ever-increasing addition of structures to the protein database. X-ray crystallography, with the caveat of obtaining a crystal, remains the gold standard for obtaining atomic resolutions structures. CryoEM, with the advantage of simpler sample preparations, is the current revolution, well on its way toward reaching higher resolutions. Although structures provide a clear snapshot of a protein, dynamic properties govern its functional output. Solution NMR, a technique limited by a protein's size, is potentially the best tool to study protein dynamics along with its structure. We envision a future where a synergy of aforementioned techniques will help quickly delineate the structure and dynamics of a protein to better understand its function within a cell.

NMR, a conceptual summary
NMR is a powerful biophysical tool to ascertain atomic resolution details of a protein. It relies upon the basic quantum mechanical property of nuclear spins. Atoms with nonzero spin numbers, when placed in a magnetic field, are at different energy levels. We can stimulate and follow transitions between these levels at specific resonance frequencies. The most prominent feature of NMR spectroscopy is its ability to differentiate chemically identical atoms, experiencing small variations in their magnetic environment influenced by its neighboring atoms. This property, substantiated by half a century of scientific and technological developments, allows for the assignment of each peak in the spectrum to a specific atom within a molecule of interest.
Delineating a protein's structure can be broadly split into identifying the atoms and their three-dimensional position with respect to each other. Nuclear Overhauser effect (NOE) provides the main source of geometric information utilized for structure determination by NMR (3) and requires the assignment of each proton resonance in a spectrum of the target molecule. For small proteins of a hundred residues or less, this could be accomplished through conventional homonuclear 1 H two-dimensional (2D) experiments developed in the late 1970s and early 1980s (4). For larger proteins, overlapping chemical shifts and increases in linewidth (due to increased rotational correlation time, c ), greatly complicates spectral analysis demanding an improvement of spectral resolution (5). Multidimensional NMR using 15 N and/or 13 C chemical shifts through the introduction of triple-resonance heteronuclear experiments (6 -8) allows for the determination of protein structures in the 25-kDa (9) range, whereas incorporation of deuterium, replacing protons at sites not amenable to exchange, further pushed the limit to 35 kDa (10) by suppression of different relaxation pathways.
Among the different applications of NMR, viz. solution, solid-state, and oriented sample, solution NMR has found wide usage for structural and function studies of IMPs under conditions that closely reflect their native environments. It also allows for real-time mechanistic investigation (11) and makes possible "in-cellulo" applications (12). In addition to structure determination, solution NMR provides information on protein dynamics (13), folding (14), enzymatic reaction rates (15,16), oligomeric states (17), and ligand binding (quantitative in favorable cases of fast exchange (18)), and it has been extensively used to facilitate drug discovery (19,20). The overall size of the molecule determines the success of NMR applications, which in the earlier 1980s was below 10 kDa and later progressed to 25-35 kDa around the mid-1990s. Currently, the size limit has been pushed even further, upward to 100 kDa (21).

Overcoming size limitations
A crucial breakthrough, which increased the susceptibility of the technique to proteins of higher molecular weight, resulted  (1); multipass TM helices (2); and ␤-barrel porin that span both leaflets (3); monotopic amphipathic helix that span a single leaflet (4); lipoproteins (5 and 6); and peripheral extrinsic proteins (7). JBC REVIEWS: Structural studies of MPs using solution NMR from combining the outcomes of the following factors (discussed briefly below): (i) widespread availability of high field (above 700 MHz) super-conducting magnets equipped with cryo-probes and simultaneous four channels decoupling; (ii) application of TROSY-based acquisition protocols; and (iii) universal adoption of novel isotopic-labeling schemes.
The advantages of high-magnetic fields that drove the design and construction of bigger (often synonymous with "better") magnets are based on two general principles: (i) direct proportionality of the spectral resolution to the field strength (␦ ϳ B) and (ii) signal-to-noise, proportional to the field strength in the power of 3/2 (S/N ϳ B 3/2 ). An additional important advantage of high-field magnets was appreciated with the development of transverse relaxation optimized spectroscopy (TROSY)-based protocols. TROSY (22) relies on the notion that the two major relaxation pathways, chemical shift anisotropy and dipole-dipole coupling, are mutually cancelled out in one of the individual multiplet components of the single quantum spectrum, resulting in sharper NMR peaks (even for very large proteins). The efficiency of cancellation depends upon the strength of the magnetic field. Optimal field strengths for TROSY were estimated to be around 900 MHz for the amidedetected TROSY protocols. More recent studies of 15 N-detected TROSYs, however, purport the achievement of maximal peak heights at even a higher field strength of 1.15 GHz (23). Low temperature ("cryo-") probes, which detect signals at lower than ambient temperatures, additionally improved the sensitivity through white noise reduction (N ϳ kT), as reflected in a higher S/N ϳ1/kT. Simultaneous decoupling through four channels ( 1 H, 15 N, 13 C, and 2 H) is essential to take full advantage of data acquisition on highly-deuterated 15 N/ 13 C-labeled samples as it increases the signal-to-noise ratio by consolidating all signal intensities of multiplets into single peaks.
Although 15 N/ 13 C isotopic labeling schemes and triple resonance spectroscopy provide an opportunity to study large macromolecular systems, proteins are still subject to 1 H-1 H dipolar spin relaxation, which adds to the heteronuclear 1 H-X (X ϭ 15 N or 13 C) spin relaxation. Together, these relaxation pathways reduce the efficiency in magnetization transfer during heteronuclear multidimensional experiments, as reflected by the substantial loss of sensitivity and resolution of spectra for proteins above 25 kDa. In remediation, carbon-bonded proton substitutions with deuterium can take advantage of the 6.7-fold lower gyromagnetic ratio of 2 H relative to 1 H, effectively reducing the relaxation pathways as compared with fully-protonated samples. Uniform random 2 H incorporation at varying levels or selective methyl protonation on fully-deuterated samples (24) are methodologies that incorporate such substitutions. One should exercise caution by not eliminating all protons in a sample as they constitute the main source of geometric information essential for structure calculation. Other more exotic examples of novel labeling schemes include "reverse isotope" approaches, where specific protonation of 14 N-labeled residues has been employed in a fully-deuterated 15 N-labeled background (25,26), and segmental labeling, which allows the observance of signals of select N-or C-terminal regions along a peptide chain (27). Employing alternatives to well-established (prokaryotic) Escherichia coli expression systems, like Pichia pastoris (28) or Chinese hamster ovary (CHO) cells (29), may be deemed necessary when post-translational modifications are essential for proper protein folding. Cell-free expression systems (30) are particularly gaining traction as they can successfully produce recombinant IMPs that are otherwise difficult to overexpress due to their endogenous toxicity leading to cell death (31). Dötsch and co-workers (32), for example, have introduced a strategy to generate small isotropic bicelles with IMPs that have been co-translationally embedded into nanodiscs during cellfree expression. The advantage of this method (generating a proper fold while maintaining a small size) becomes clear when we discuss the pros and cons of different membrane mimetics in the next sections.

Membrane mimetics
Stabilizing membrane proteins outside their native membrane environment remains the Achilles heel for proteins of this family. Liposomes, spherical vesicles composed of a lipid bilayer, were the earliest system developed to house membrane proteins. Bangham and co-workers (33,34) successfully employed liposomes around the mid-1960s to study different aspects of membrane biophysics. Although liposomes could routinely vary their lipid composition, net charge, shape, temperature, and lyotropic phases, they remain too large for solution NMR studies. Even the smallest unilamellar vesicles are only best-suited for binding experiments than structure determination. In the following sections, we follow the chronological development of different approaches that were undertaken to increase the susceptibility of solution NMR to larger systems.

Organic solvents
Organic solvents or their mixtures can solubilize complex membrane proteins by mimicking a hydrophobic environment to accommodate the hydrophobic core of an IMP. An obvious advantage is the lack of an additional molecular weight brought about by the system itself with the downside of compromised protein stability. Although organic solvents may preserve secondary structural elements, it could still perturb a protein's tertiary structure, as was observed for DAGK (trimer of 13-kDa subunits, each containing three TM helices) (35). When using organic solvents, it is recommended to obtain additional corroborating data to improve the overall confidence in the structural information on IMPs, as demonstrated by the following two case studies. Girvin and co-workers (36,37) showed the subunit c of F 1 -F 0 -ATPase (two TM helices connected through a short loop) to adopt a stable, native-like fold in at least one organic solvent mixture and determined a high-resolution structure of the dimer (PDB code 1A91/1C0V). Based on their NMR and cross-linking data, they were able to generate a functional model of a1-c12 complex (PDB code 1C17). Yang et al. (38) defined the interface of a major platelet integrin ␣ IIb ␤ 3 transmembrane-cytoplasmic heterodimer (PDB code 2KNC) as supported by numerous mutational and biochemical data, which allowed them to propose a mechanistic model for the activation of this key cell-surface receptor.

Micelles
By far, the most successful method to solubilize IMPs are detergents that spontaneously form a micelle. Their overall smaller size (39), relative to planar bilayers, makes them the best choice of a membrane mimic. A variety of detergents has been used (SDS, dodecyl maltoside, OG, DPC, LDAO, LMPC, LMPG, DHPC, etc.) to access the structural integrity of membrane proteins (40,41), compare their compatibilities with enzymatic functions (39), and determine the quality of corresponding NMR spectra (42). A subset of detergents with a polar or zwitterionic headgroup attached to a medium-chained aliphatic tail has found widespread usage (Fig. 2). From an NMR perspective, DPC (Fos-choline 12) is the most widely used detergent, followed by LDAO, and LMPC, or, in some specific cases, like potassium channels LMPG. Additional lipid supplementation, in a small amount, has shown drastic improvement in the stability and/or functional activity of IMP (43) without a perceptible increase in the overall size (44). The following considerations should be taken into account when planning experiments involving detergents: (i) although popular, SDS, SDS with a negatively charged headgroup, should be sparingly used as it can destabilize a protein's conformation; (ii) the actual size of mixed micelle depends less on the type of detergent and more on the properties of the membrane protein itself, where the extent of molecular size is governed by detergent coverage of the transmembrane region; (iii) the presence of free detergent and a strong hydrophobic environment could substantially impede interaction to ligands and could lead to progressive denaturation of the soluble-binding targets; and (iv) loose packing of detergent headgroups, micellar curvature, and tighter acyl chain interactions at the core may destabilize an IMP conformation and alter dynamics and oligomerization states lead-ing to noticeable differences with respect to these properties in a flat lipid bilayer system (45-48). The above-observed caveats have warranted the need to develop alternative nanoscalephospholipid bilayer systems, as discussed below.

Bicelles
Prestegard's research group (49,50) first introduced bicelles, flat phospholipid bilayer discs decorated by detergents at their edges ( Fig. 3). A bicelle's conception was derived from X-ray studies of a mixture of bile salt (CHAPSO, for example) and phosphatidylcholine (DMPC) that formed discoidal bilayers of uniform size. They further demonstrated spontaneous orientation in a magnetic field (49,50), thus instituting a novel medium for oriented protein samples. Sanders and co-workers (51) further extended the system by substituting bile salts with DHPC, a short-chain phosphatidylcholine with two six-carbon acyl chains. These bicelles, or "binary bi-layered mixed micelles," quickly replaced the DMPC/CHAPSO mixture because phosphatidylcholine is a more abundant natural constituent of biological membranes. Other bicellar compositions of longer chain phospholipids, forming a wider bilayer, or lipid mixtures mimicking eukaryotic membranes, providing a "close to" natural environment, have also been used for NMR studies (52)(53)(54)(55).
In addition to their width and charge, the diameter of a bicelle can also be regulated, offering magnetic properties suitable for alignment. Only systems with magnetic susceptibility anisotropy sufficiently large to overcome Brownian motion, which correlates to a diameter beyond 20 nm, can spontaneously align in a magnetic field. Bicellar diameter (D) depends on the lipid to detergent ratio, often referred to as the "q" number (56). It can be calculated according to Equation 1 originally proposed by Mazer et al. (57) for ideal bicelles and later modi- JBC REVIEWS: Structural studies of MPs using solution NMR fied by Vold and co-worker's (58) to take into account differences between headgroup areas of DHPC and DMPC (k defines this ratio, which was experimentally found to be 0.6; r is the radius of the rim, which equals 2 nm for DHPC): The q parameter was later modified as shown in Equation 2 to accommodate free detergent concentration of a low amphiphilic sample, especially where the total lipid and detergent concentration is below 5% w/v (59). For DMPC/DHPC bicelles with lipid concentrations of 3-40% w/v, in aqueous solutions, and q Ͼ2.3, spontaneously align above the DMPC gel-to-liquid transition temperature (typically within 30°to 50°C) (60). Bicelles are oriented with their bilayer normals perpendicular to the direction of the applied magnetic field. This orientation can be flipped parallel by doping bicelles with paramagnetic ions, such as lanthanides Eu 3ϩ , Er 3ϩ , Tm 3ϩ , and Yb 3ϩ (61), aromatic molecules (62), and even by incorporating membrane proteins, such as gramicidin A (63). This flip could often result in an improved spectrum quality for bilayer constituents even in the absence of its rapid rotation around bilayer normal, and therefore it is particularly useful in determining 15 N-labeled protein amide bond orientations by solid-state NMR. At smaller q ratios of ϳ0.25-0.5, bicelles are typically "lipid-poor" and "detergent-rich," exhibiting a rapid isotropic tumbling ideal for solution NMR. However, their hydrophobic cores are quite similar to that of a lipid/ detergent mixed micelle. At q Ͼ0.6, assemblies are most often too large to yield a high-resolution spectrum, although they do possess fully segregated lipid cores (64). A number of well-defined oligomers, composed of TM helices, have been determined in isotropic bicelles with q ranging between 0.5 and 0.6. They include structures with PDB code 5JYN (65), 6E8W for HIV gp41 TM trimers (66), 2NA7 (human) and 2NA6 (mouse) for TM trimers of Fas/CD95 death receptor (67), and 6NHY (mutant) or 6NHW (WT) for trimer or dimer of trimers, respectively, of DR5 (68). A detailed protocol for samples preparation at these conditions and suggested set of experiments to access oligomerization and TM partitioning has been published recently by Chou and co-workers (69). The authors particularly emphasize the utilization of paramagnetic relaxation enhancement (PRE) experiments and the OG-label method they introduced for these studies (70). This approach was used to confirm the unusual hexameric structure of HCV p7 channel originally determined in DPC micelles (PDB code 2M6X) (71) and  (72). At q Ͻ 0.25, ideally mixed micelles are formed with no observable lipid segregation.
The ability of diluted (Ͻ5% w/v) bicelles to orient in a magnetic field was later proven useful by Tjandra and Bax (73) to generate an alignment medium for measurement of residual dipolar coupling (RDC), an additional resource for distance restraints in determining NMR structures.

Amphipols
Detergents remain the current gold standard for solution NMR studies, but as indicated under "Micelles," they render IMP-ligand interaction studies untenable. Amphiphilic polymers, with a backbone of alternating hydrophilic and hydrophobic chains, represent an alternative solution to the micellar system. Pioneered by Popot and co-workers (74,75), amphipols (APols) have shown the ability to extract and stabilize IMPs in the absence of conventional detergents or denaturants. Amphipol's applicability to solution NMR remains challenging, with no available representation in the PDB database, thus far. A potential explanation may reside in a heterogeneous local environment produced by the interaction of amphipols with residues from the transmembrane region, as it is difficult to control the exact sequence of alternating chains within the polymer. The resonance frequencies may reflect this diversity, due to the extreme sensitivity of NMR to the local environment, and it might be averaged for the same atoms from different molecules, resulting in peak broadenings and reduced spectrum quality. Moreover, polyacrylate-based amphipols, like A8-35, are sensitive to pH Ͻ7 and multivalent cations (76). PMAL polymers (77) containing a sequential pair of alternating hydrophobic and hydrophilic groups might ameliorate the shortcomings of APols. This was clearly demonstrated through higher thermal stability for NavM sodium channels (78). IMPs extracted by amphipols, however, can be transferred to lipidic mesophase, as shown for bacteriorhodopsin (PDB code 4OV0), resolved at 2.0 Å by X-ray crystallography (79). Amphipols do seem suitable for single-particle EM. A 19-Å 3D map of the 1.7-MDa amphipol-solubilized super-complex I(1)III(2)IV(1) (PDB code 2YBB) from bovine heart, obtained by cryo-EM, revealed an amphipol belt replacing the membrane lipid bilayer (80). More recently, a number of amphipol-stabilized TRP channels (PDB codes 6BO9, 5YDZ, and 5YE1) and PMAL-stabilized polycystin 2-l1 (PDB code 6DU8) have been characterized by cryo-EM at high resolutions.

Nanodiscs
Nanodiscs, a detergent-free flat discoidal system, developed by Sligar and co-workers (81)(82)(83)(84), contain a phospholipid bilayer encased within two anti-parallel amphipathic helical membrane scaffold protein (MSP) (Fig. 4). Nanodiscs are derived from apolipoprotein A1 (apoA1), originally involved in cholesterol transport. In the absence of cholesterol, apoA1 forms nascent discoidal particles containing phospholipids. Upon cholesterol ingestion, the proteolipid transforms to a spherical shape, eventually ending up in the liver (85). Fulllength apoA1 is a helical protein with an N-terminal four-helix bundle and two C-terminal helices (PDB code 2A01). A 43amino acid truncation at the N terminus results in a circular belt-like structure made of amphipathic helices (PDB code 1AV1) (86). This deletion, containing 10 ␣-helical repeats, is referred to as MSP1. A larger construct, designed with two MSP1 molecules connected through a stable linker, is termed MSP2. MSP1, MSP2 (82), and other variants are broadly classified as MSPs. The diameter of the disc is determined by the number of helical repeats in the annular MSP. A bilayer of desirable thickness can be achieved by using multifarious lipids like DMPC, dimyristoyl-phosphorylglycerol (DMPG), palmitoyl-oleoyl-phosphocholine (POPC), dipalmitoyl-phosphocholine (DPPC), or lipid mixture like E. coli lipid extracts, etc. (87). Lipid molecules varying in their acyl chain length provide bilayers with altered thickness (Fig. 4), which may be paramount to IMPs that either diverge in the expanse of the bilayer or incur transmembrane reorientations as a result of activation. This nascent promiscuousness of the system for incorporated lipids offers a wide diversity of customizable discs gratifying stability requirements for different membrane proteins. Altogether, nanodiscs provide a competitive edge over other systems for studying IMPs due to their soluble nature, ease of concentration, monodispersity, temperature stability, and compatibility with a cell-free expression system.
The prototypical MSP construct, MSP1D1 (88), with 10 amphipathic helical repeats, forms discs of ϳ10 nm diameter with an overall molecular mass of ϳ150 kDa. Variation in the number of helical repeats within the MSP1D1 construct led to the development of nanodiscs with different diameters. Solution NMR applications are amenable to an upper limit of ϳ100 kDa. Consequently, redesigning the nanodisc to obtain smaller sizes was an obvious alternative for application of nanodiscs in solution NMR. Wagner and co-workers (89) removed three intermediate helices (helices 4 -6), either individually or together, based on the rationale of helical insertion for larger MSP constructs like (MSP1E3D1) while trying to maintain terminal interactions that may potentially stabilize the discs. The smallest disc (⌬H4 -H6), ϳ6.4 nm in diameter, was kinetically unstable, forming larger aggregates of ϳ11 nm over time. The next smaller disc (⌬H4 -H5 construct) was stable overtime with a diameter of 7.3 nm with an overall molecular mass of ϳ70 kDa (Fig. 4) (89). Concurrently, we also independently developed smaller discs by sequential removal of three terminal helices, ⌬H8 -H10. Our approach additionally confirmed the irrelevance of terminal interactions as inconsequential for nanodisc formation. Our smallest disc, called D7, containing the first seven helices (Fig. 4), is 7 nm in diameter and ϳ62 kDa (90). Both ours and Wagner's efforts independently demonstrated the applicability of small discs for NMR studies. Additionally, we also developed an alternative "on-column" method for nanodisc reconstitution on chromatographic resins. From a structural perspective, it is important to note that though Wagner's ⌬H4H5 yielded the smallest stable disc, the ⌬H5 construct, with a slightly larger diameter of about 8.4 nm (Fig. 4), provided the best NMR spectrum and was selected for deducing the structure of OmpX (PDB code 2M06). We find that identifying the oligomeric state of a ␤-barrel porin in a nanodisc may be important before pursual of NMR studies. Image aver-JBC REVIEWS: Structural studies of MPs using solution NMR aging of negatively stained TEM images serve as a good visualization tool for identifying oligomeric states (91).
Larger discs, necessary for incorporation of larger integral membrane complexes, have been investigated as well. Sligar and co-workers (92) fused two MSP1 molecules together into a single construct to yield ϳ17-nm discs. A separate approach for generating larger nanodiscs was demonstrated by Opella and co-workers (93). Multiple copies of an amphipathic 14-residue peptide, mixed with phospholipids at a predefined ratio, produced a macrodisc of ϳ30 nm diameter. The increased diameter generated sufficient magnetic susceptibility anisotropy to overcome tumbling averaging motions allowing them to align in an external magnetic field. Similar to anisotropic bicelles, macrodiscs can be used as an alignment media for RDC measurements of soluble macro-molecules (94).
A notable modification of the nanodisc system has recently come from Wagner and co-workers (95). Discs of controlled diameters (9, 11, 30, or 50 nm) were produced using the sortase A-based system that recognizes a consensus LPGTG sequence near the C terminus and a single Gly residue at the N terminus of MSP. These cNDs exhibit enhanced stability, defined diameter sizes, and tunable shapes. Overall, the improvements in cNDs are manifested through better NMR spectral quality for two tested IMPs, VDAC-1, a ␤-barrel membrane protein, and GPCR, NTR1.
A crucial parameter to consider in lipidic systems is the bilayer thickness. An optimum lipid alkyl chain length should be selected to accommodate the entire span of an IMP's hydrophobic region. Mismatches, potentially occurring when the protein's hydrophobic thickness is less or greater than that of the lipid bilayer, may cause curvature and/or disorder of the bilayer near the protein's core surface (96). In a few certain cases, this condition may seem favorable, for example folding studying on ␤-barrel proteins (97). However, most often, the functional states of channels or receptors, as well as enzyme activity, will be altered due to conformational distortions associated with membrane mismatch, which can be determined by functional assays (98). The anticipated thickness for a protein of A, visual representation of a nanodisc that is either empty or contains IMP. The outer belt protein MSP (green) and lipid molecules are colored by atom type with carbon (gray), oxygen (red), phosphorus (orange), and nitrogen (blue). Also, the approximate hydrocarbon thickness of nanodisc bilayers is shown for various phosphatidylcholines. B, schematic representation of the overall length and disc diameters for several MSP variants. Nanodiscs with smaller diameters (below 8.5 nm) are preferable for solution NMR studies. The largest "MACRODISC," obtained from a 14-amino acid peptide, produces a disc of 30 nm diameter and serves as an alignment medium.

JBC REVIEWS: Structural studies of MPs using solution NMR
interest should be experimentally explored by testing various lipid compositions. A complex lipid composition is often essential to mirror natural bilayer properties and maintain the functional activity of a particular IMP (99). Nanodisc's resilience toward the inclusion of a cornucopia of lipids, divergent in their alkyl chain lengths, degree of saturation, polarity of headgroup, etc., is a prominent advantage of this system.

Saposin A discs
Two saposin A proteins coalesce together around lipid molecules forming a nanoscale saposin A (SapA) disc named "Salipro." SapA disc differs from nanodisc in two general ways: (a) SapA proteins form a discontinuous belt around the lipid bilayer, and (b) the diameter of SapA disc is dynamic and unlike the need for MSP of varying lengths and multiple SapA proteins accommodate IMPs of varying sizes. The apo (lipid/detergent free) form of SapA has a closed architecture that becomes extended upon binding to LDAO molecules (100). Individual detergent-bound SapA proteins are brought together through the hydrophobic interaction of core acyl chains (Fig. 5A). LDAO containing a SapA disc is ϳ3.2 nm and contains 40 detergent molecules in total, arranged asymmetrically with 24 molecules in the upper and 16 in the lower leaflet of the bilayer. Empty SapA discs have a significantly smaller mass of 43 kDa, with respect to nanodiscs. Similar to nanodiscs, SapA discs are also compatible with a variety of lipid molecules and are flexible enough to accommodate large membrane protein complexes with variable molecular weights, as demonstrated by cryo-EM structures of archaeal mechanosensitive channel T2 (32.9 kDa), with four predicted transmembrane helices forming as a putative homopentamer, and bacterial peptide transporter PepT So2 (56 kDa), with 14 transmembrane helices forming a homotetramer (101).
Theoretically, there are certain advantages of employing SapA discs in solution NMR. SapA discs are compatible over a wide range of pH values (100). It has exceptional stability, with- Figure 5. Other potentially useful membrane mimetics. A, saposin-A in its detergent-free form adopts a closed conformation that becomes extended when bound to detergent molecules (PDB code 4DDJ). Saposin-A lipoprotein disc, with a diameter of 3.2 nm, contains two saposin-A proteins brought together by a lipid core. B, SMALP is shown where the synthetic styrene-maleic acid co-polymer forms discs by encapsulating lipid within its central cavity. SMA lipid discs or Lipodisqா has a diameter of 10 nm. Other polymers shown to form lipid particles include methacrylate and DIBMA. C, amphipathic peptides have been recently used for "lipid-free" IMP reconstitution forming "peptidisc": helical peptides wrap around the hydrophobic parts of detergent-purified IMPs eventually displacing detergent molecules.

JBC REVIEWS: Structural studies of MPs using solution NMR
standing several freeze-thaw cycles, and demonstrates high thermostability (0 -95°C), allowing for NMR experiments to be conducted at higher temperatures as is desirable for larger systems rendering better spectral quality. Finally, protein-incorporated discs are homogeneous as confirmed by negativelystained TEM images. However, the system does bring along a few caveats: (a) the propensity of the reconstituted IMP for non-native spontaneous oligomerization; and (b) a potential for spurious interaction between the IMP and SapA, especially because the lipid content around the reconstituted IMP is "very tight." Although the system is not without its challenges, SapA discs do offer a viable alternative to nanodiscs. As of 2018, there are no protein structures available in SapA discs. The 15 N HSQC spectrum of OmpX, incorporated in SapA disc (102), was found similar to the spectrum of OmpX in ⌬5H nanodisc (for which the structure has been determined by solution NMR (89)). The spectrum of SapA-incorporated phototaxis receptor sensory rhodopsin II (pSRII, 24.6 kDa, potentially a homodimer in this setting), was compared with the one in c7-DHPC micelles and appeared to be more heterogeneous and less resolved. Additionally, the functionality of the SapA-incorporated ␤ 1 -adrenergic receptor (␤ 1 AR, 36 kDa, a GPCR) was also tested and confirmed by NMR (102).

Co-polymer discs
SMALPs, trade name Lipodisq, are discoidal lipid-polymer aggregates, where the outer annulus of the lipid bilayer is formed by styrene-maleic acid (SMA) composed of styrene and maleic acid in ratios of 2:1 or 3:1 (103). These discoidal particles assemble spontaneously in aqueous solution and require very low concentrations of SMA, as reflected by extremely low molar ratios of lipid to polymer, forming a disc with 140 lipid molecules and a median diameter of ϳ9 nm (Fig. 5B). SMALP has been used to form bilayer discs with POPC (104) and DMPC (105). Interestingly, the SMA co-polymer can directly solubilize membrane proteins from its native membrane, isolating patches of flat bilayers with endogenous lipids surrounding the IMP. The mitochondrial respiratory complex IV was successfully isolated from its native membrane using SMA, which retained its enzymatic activity (106). The presence of a lighter co-polymer belt, compared with amphipathic protein in nanodiscs, results in an overall lower molecular weight assembly and might offer favorable relaxation advantages for detection using NMR. However, as of writing this review, no structures of IMPs in SMALPs are available. Because SMA has a pK a of 6.5, NMR experiments involving SMALPs need to be performed within a narrow pH range of 6.8 -7.5. Additionally, divalent cations destabilize SMA co-polymer, causing their dissociation, and hence they should be avoided. Also, the major hydrophobic moiety of the polymer, styrene, is purported to nonspecifically interact with the aromatic side chains of proteins and has a propensity to absorb in the UV range, which could impede accurate analysis by various biophysical techniques. We believe a major drawback of SMALPs, akin to amphipols, is its sample heterogeneity as observed by TEM (105), which should translate into an NMR spectrum with compromised quality.
However, there have been successful attempts recently to enhance the stability and homogeneity of lipid-polymer discs by the modification of functional units and/or by exploring other types of styrene-based or styrene-free polymers. Styrene maleimide quaternary ammonium (SMA-QA) derivatization resulted in an improved stability at low pH and discs with controlled size (107). One could envision the use of such polymerbased lipid-discs in solution NMR or as macro-nanodiscs in solid-state NMR applications. In addition to their use as alignment media, these discs are tunable in their size and resistant to the presence of divalent metal ions (107,108). Styrene-free polymer disc with a polymethacrylate framework (107) has also been designed as a cheaper alternative to SMA. These are amphiphilic and can easily be derivatized with a variety of side chains. The charge on the polymer plays a vital role in the functional reconstitution of membrane proteins: a high charge density around the disc could produce unexpected and undesirable interactions (109). Polyacrylic acid (PAA), with systematically varied hydrophobic groups through a robust functionalization method, was utilized to probe how alkyl-PAA affects the formation, stability, and other disc properties (110). It was indeed confirmed that the choice of hydrophobic group can have a noticeable effect on the polymer solubilization properties. Another aliphatic co-polymer with alternating diisobutylene/ maleic acid, DIBMA, was recently shown to form nanoscale discs (111). DIBMALPs are compatible with both long-and short-chain phospholipids, unfazed by the presence of cations, and do not absorb at 280 nm. Although the lack of potentially interfering hydrophobic groups makes them an attractive alternative, DIBMALPs form slightly larger heterogeneous discs than SMALPs. Further rigorous experimentation will determine the merit of the application of co-polymer systems with NMR.

Peptidiscs
Amphipathic peptides have been recently used for "lipidfree" IMP reconstitution forming a "peptidisc" (112). Carlson and co-workers have shown that bi-helical peptides wrap around the hydrophobic parts of detergent-purified IMPs, eventually displacing detergent molecules (Fig. 5C). Being a lipid-free reconstitution, we are inclined to believe that the replacement of detergents by peptides should result in their close association to IMP and thereby become an obvious concern for a direct interaction between the two components. Future studies comparing NMR spectra of a known protein in detergent micelle and peptidisc should address this query. Peptidisc can be reconstituted in solution using a mixture of peptides and IMPs or directly on a chromatographic resin bound to target IMP. Reconstitution on resin provides a single step enrichment of the reconstituted protein similar to our on-column method, developed for nanodiscs (90). Being a fairly new system, it remains to be seen whether their stability and homogeneity will allow for high-resolution structure determination of reconstituted IMPs.

Structural gallery
Although there is an increase in the overall number of IMP structures determined by solution NMR, it remains signifi-JBC REVIEWS: Structural studies of MPs using solution NMR cantly lower than X-ray crystallography. Here, we present a comprehensive overview of the structures in PDB as of mid-2019 (Fig. 6). These are predominantly smaller IMPs with one to two transmembrane ␣-helices, although several porins composed of membrane-spanning ␤-barrels have been defined as well. Additional strategies for resonance assignments, which JBC REVIEWS: Structural studies of MPs using solution NMR include trypsinization and use of synthetic peptides, have shown promising potential toward structural characterization of ␤-barrels such as Opa 60 (113) with long and flexible extracellular loops.
The majority of IMP structures has been determined in micelles, followed by isotropic bicelles. Nanodiscs, in our opinion, are yet to become a true choice of a mimetic with only structures of two outer membrane porins and a monomeric BclxL TM helix (PDB code 6F46) available thus far. The first structure of a ␤-barrel protein, E. coli OmpX (PDB code 2M06), solved in nanodiscs, was obtained using small discs formed by the ⌬H5 MSP deletion construct (89). A later study, using RDC measurements of OmpX in nanodiscs with Pf1 phage as an alignment medium (PDB code 2MNH), helped refine the structure of the classical eight-stranded ␤-barrel fold with betterdefined orientation of N-H bonds (114). A second structure of the Yersinia pestis outer membrane protein Ail (PDB code 5VJ8) in the ⌬H5 discs composed of DMPC/DMPG mixture, was achieved by solid-state NMR data acquired with membranes containing lipopolysaccharide (LPS) (115). The membrane composition had a marked effect on protein dynamics, with LPS enhancing conformational order and slowing down the 15 N transverse relaxation rate.
Two dimeric conformations, one of F 0 -F 1 -ATPase subunit C (PDB code 1A91/1C0V) and major platelet integrin ␣ IIb ␤ 3 transmembrane-cytoplasmic heterodimer (PDB code 2KNC), may be considered as reliable representatives of IMPs in organic solvent mixtures. Other membrane mimetics (amphipols, Salipro, and SMALPs), discussed in this review, have yet to prove their applicability toward structural studies using solution NMR.

Mechanistic studies
HSQC chemical shift perturbations, observed upon addition of potential ligands, are generally the core experiment to study binding. The utility of nanodisc for monitoring resonance frequencies (or chemical shifts) of two membrane proteins was demonstrated early on as follows: (i) CD4 mutant, containing a single transmembrane and cytoplasmic tail, where the aliphatic 1 H-13 C HSQC chemical shifts were compared with ones in DPC micelles (116); and (ii) VDAC-1, human anion channel protein, where 1 H-15 N TROSY HSQC spectra in LDAO isotropic bicelles and MSP1D1 nanodiscs were compared in the presence and absence of its native ligand NADH (117), later followed by a similar investigation of VDAC-2 (118). As expected, the spectra in nanodiscs show broader peaks than in micelles. Most peaks overlap, although some demonstrate differences in chemical shifts reflecting the variations arising from a micellar to a bilayer packing. Other examples of HSQC-based IMP studies in nanodiscs include the following: (i) the voltage-sensing domain (4TM) of KvAP channel, which was characterized and shown to maintain a proper conformation only in a zwitterionic environment (119); (ii) effect of different membrane mimetics on the structure of bacteriorhodopsin (7TM) (120); and (iii) importance of lipid deuteration shown for YgaP (2TM and a cytoplasmic Rhodanese domain) in d 54 -DMPC containing MSP1D1 nanodiscs (121).
A 22-residue-long amphipathic peptide nanodisc was used to study protein-protein interactions within the 70-kDa cytochrome P450 complex (CYP2B4-Cytb 5 ), defining the binding surface of Cytb 5 (122). NMR data-driven model of GTPase KRas-GDP, tethered to a lipid-bilayer nanodisc (PDB codes 2MSC/2MSD/2MSE), was published in 2015 (123) and later refined in 2018 (PDB codes 6CC9/6CCX/ 6CCH). In this combined computational approach, support was derived from experimental distance restraints obtained from PRE and additional NOE experiments in the refined version. HADDOCK (124) simulations allowed us to distinguish the reorientation of the effector-binding site on K-RAS4B GTPase anchored to (MSP1D1) nanodisc previously occluded by the anionic membrane.
Lipid bilayer interactions of the substrate-binding hemopexinlike domain of membrane type-1 matrix metalloproteinase (MT1-MMP) were investigated in MSP1D1 nanodiscs by PRE experiments, fluorescence, and mutagenesis, followed by docking and MD simulations (125). They helped define the insertion mode of hemopexin-like, which appears to happen through the blades II (PDB code 6CM1) and IV (PDB code 6CLZ) from the opposite sides of its ␤-propeller fold.
It is our contention that real-time in vitro monitoring of functional activity of IMPs by solution NMR methods should be further explored in nanodiscs. The accessibility of either side of a receptor, surrounded by a lipid membrane, could enhance the interaction of a potential ligand(s). Phosphorylation, a classic example for cell-surface receptor activation, can be monitored by NMR. A proof of concept demonstration of phosphorylation has been reported in nanodiscs under tightly-controlled conditions for neurotensin receptor 1, a GPCR class member, through the addition of G-protein-coupled receptor kinases (126). We also provided the first demonstration (Fig. 7) of in vitro phosphorylation by mixing SRC-kinase and nanodiscs containing the transmembrane and cytoplasmic domains of recombinant integrin ␤ 3 subunit in an NMR tube (127).
Attempts to incorporate individually purified IMPs into a nanodisc while forming a heteromeric complex bring to light problems associated with proper stoichiometry and relative mutual orientation. Wagner and co-workers (128) demonstrate a generalized approach for co-reconstitution of a membrane protein-DNA adduct where complementary DNA strands attached to different proteins initiate oligomerization. Fluorescently-labeled voltage-dependent anion channel was used to demonstrate the formation of heterodimeric versus heterotrimeric complexes.
Because GPCRs constitute an essential subclass of cell-surface receptors and serve as potential drug targets for intervention in various diseases, it is important to understand its structure-activity relationships in native environments. It is also well-recognized that GPCRs are flexible and highly-dynamic receptors that can adopt various shapes and oligomeric states. From the perspective of drug development, an investigation on how ligand binding affects GPCR's overall motility and the dynamics of its side chains under the conditions most closely mimicking native membrane environment is warranted. Solution NMR is uniquely capable of providing this type of information even for very large macromolecular complexes with the extensive use of 13 C-methyl TROSY-based spectroscopy on perdeuterated samples (129). However, specific 13 C labeling of aliphatic methyl groups in deuterated proteins has been primarily limited to recombinant proteins overexpressed in E. coli. Many of the most interesting targets, including several GPCRS, require expression and purification of the receptors from mammalian cells to ensure proper folding and posttranslational modifications. This calls for novel schemes of effective isotopic labeling in eukaryotes. Clark and co-workers from the Rosenbaum and Gardner labs have demonstrated the feasibility of efficient [ 13 C]isoleucine ␦1-methyl labeling in a deuterated background in P. pastoris. This was first demonstrated for maltose-binding protein, as a proof of concept, and compared with the recombinant protein overexpressed in E. coli. The authors have also show that this method can be used to label eukaryotic proteins, such as actin (130). They later extended their approach to WT human A 2A R GPCR, resolving 20 out of 29 expected peaks in the Ile-1␦ region of the 13 C TROSY spectrum (131). Their data further indicate that low Na ϩ concentration is necessary to allow large agonist-induced structural changes, and how the pattern of side-chain dynamics is quite different between agonist and inverse agonist-bound receptors, with the inverse agonist suppressing fast picosecondnanosecond time-scale motions at the G-protein-binding site. On a separate note, with the help of a thermostabilized mutant of the turkey ␤ 1 AR, 15 N-labeled at valine residues, Grzesiek and co-workers (132) have shown that GPCR motions can be followed at virtually any backbone site as well. They also provide a thorough protocol for cost-effective isotopic labeling of proteins expressed in insect cells, using ␤ 1 AR and CCR5 as examples (133). Altogether, these impressive studies illuminate the unique capabilities of solution NMR for studying dynamics that enable a better understanding of GPCR's functionality.

Hybrid techniques
A combination of different experimental techniques, coupling their individual advantages, has proved beneficial for structural studies of IMPs. Park et al. (134) have explored membrane-bound forms of the major coat protein of Pf1 bacteriophage (PDB code 2KSJ) through orientation restraints derived both from solid-state and solution NMR experiments. The spectra were obtained in glass-aligned planar lipid bilayers, magnetically-aligned bicelles, and isotropic bicelles. Isolated resonances only from the mobile N-terminal helix were observed in the solution NMR spectra, whereas resonances exclusive to the immobile transmembrane helix were observed in solid-state 1 H/ 15 N-separated spectra in magnetically aligned bicelles. Thus, using a combination of techniques the dynamic properties of Pf1 were addressed, allowing for a mechanistic view of the protein's rearrangement during bacteriophage assembly. Veglia and co-workers (135) also used a similar approach to study phospholamban in monomeric (PDB code 2KB7), pentameric nonphosphorylated (PDB code 2KYV) (136), and serine-phosphorylated (PDB code 2M3B) (137) forms, outlining a detailed protocol to determine structure, topology, and depth of insertion of membrane proteins using a hybrid, solution, and solid-state NMR restraints (138).

Structural discrepancies
Application of different biophysical techniques to structure determination of an IMP may result in inconsistencies, especially when different membrane mimetics are used for the study (139). One should exercise extreme caution analyzing functional consequences of these structural deviations, taking into account known limitations for each method. Validating structural data in question under a native lipid environment becomes imperative, even though high-resolution structural determination under such conditions might be impractical. In Fig. 8 we present examples of a few such cases. DAGK has been studied by solution NMR in DPC micelles (140) (PDB code 2KDC), X-ray crystallography in lipidic cubic phase (141) (PDB codes 3ZE3, 3ZE4, and 3ZE5), oriented sample solid-state NMR in liquid crystalline lipid bilayer (45), and magic angle-spinning solid-state NMR in native E. coli lipid membranes (142). Although all the methods agree on a trimeric architecture composed of three TM (plus one amphipathic) helices from each monomer, their arrangements, secondary elements, and conformations are quite different. Two major differences between NMR and X-ray structures (Fig. 8A) that stand out are as follows: (i) in solution NMR model helices 1 and 3 are domain-swapped, such that these helices primarily interact with helices from different monomers, and (ii) all TM helices have an outward curvature, producing a barrel-shaped structure. This significantly deviates from strictly cylindrical arrangement obtained from X-rays and might illustrate a potential artifact arising from detergent micelles. The unusual domain swappings, however, might reflect problems with structure calculation due to the dynamic state of the system and lack of multiple alignments followed by over-interpretation of Figure 7. Applicability of nanodisc systems for studying signaling pathways. The cytoplasmic tail of integrin ␤ 3 subunit is phosphorylated by Src kinase (in vitro) in an NMR tube: 15 N-labeled ␤ 3 incorporated discs were mixed with the kinase domain of Src kinase in the presence of ATP. Phosphorylation is manifested through chemical shift perturbations observed in the overlay of the 1 H-15 N TROSY HSQC spectra obtained from the unphosphorylated (black) and bi-phosphorylated (red) ␤ 3 were collected on a 600-MHz magnet. Also shown is the ␤ 3 tail sequence indicating phosphorylation sites at the two tyrosine residues. JBC REVIEWS: Structural studies of MPs using solution NMR PRE and RDC data. Low-resolution solid-state NMR data do not support either model, but are more consistent with X-ray structure in terms of secondary elements and topology.
Phospholamban (PNL), a homopentamer expressed in sarcoplasmic reticulum that controls intracellular Ca 2ϩ levels through its phosphorylation state, is another example where solution NMR structure determined in DPC micelles (PDB code 1ZLL) (143) differs significantly from the one obtained in lipid bilayer by a combination of solution and solid-state NMR methods (PDB code 2KYV) (136). The "bellflower" model originally proposed (Fig. 8B), with amphipathic helices of the cytoplasmic domain sticking out into the cytoplasm, could reflect the influence of dynamic micellar environment combined with the effect of artificially introduced surface curvature. This conformation was not observed in a flat lipid bilayer where amphipathic helices were found interacting with the surface, consistent with an overall L-shaped architecture. Importantly, the ion channel activity of PNL, conceptualized on the idea of a bellflower pore, has not been substantiated in later experiments and theoretical calculations under a more natural lipid surrounding (for details see Ref. 139).
Another case is a model of p7 (Fig. 8C), a hepatitis C virus (HCV) small membrane protein (PDB code 2M6X), presented by Chou and co-workers in DPC micelles (71). This model suggests an unusual mode of hexameric assembly, where the individual p7 monomers interact not only with their immediate neighbors but also with farther ones, forming a sophisticated, funnel-like architecture. Zitzmann, Schell, and co-workers (72,144) challenged this model with the evidence from their own solution NMR and size-exclusion chromatography-multiangle light-scattering studies claiming that p7 is monomeric in DPC micelles. It is worthwhile to note that DPC to protein ratios are very different between the two studies, with the latter using 40 times higher excess (DPC/p7 ϭ 250 versus 10,000). It has been known that high detergent to protein ratios may cause small membrane proteins' denaturation and mis-folding (145); hence, proper caution should be taken during reconstitution. Chou and co-workers (70) recently confirmed the original hexameric arrangement of p7 in isotropic bicelles.
Bax and co-workers (146) investigated the HIV-1 gp41 viral coat protein in bicelles (Fig. 8D). They highlighted a deviation in gp41's oligomeric state from the classical trimeric state as determined by multiple methods like cryoEM (of full-length gp160), X-ray crystallography, and solution NMR (in isotropic bicelles) (PDB codes 5JYN and 6E8W) (65,66). Using a combination of various biophysical techniques, including measurements made under different alignment conditions for RDC restraints, PRE, analytical ultracentrifugation, and doubleelectron-electron resonance experiments, Chiliveri et al. (146) found that gp41 TM domain is monomeric (PDB code 6B3U), highly-ordered, and uninterrupted for a total length of 32 residues, extending well into the membrane-proximal region. This contradicts the trimeric architecture of the gp41 construct, containing the membrane-proximal external region folded into a three-folded cluster, from Chou's laboratory. It is possible that the observed monomeric form represents a conformation that preludes trimerization. Nevertheless, it remains to be determined how construct selections, reconstitution methods, and/or lipid conditions factor into such observed differences.
Out of ␤-barrel membrane proteins, OmpX (Fig. 8E) probably represents the best example, as it was extensively studied by solution NMR in DHPC (PDB code 1Q9F) (147) and DPC (PDB code 2M07) (89) micelles, in nanodiscs (PDB codes 2M06 and 2MNH), and by X-ray crystallography (PDB code 1QJ8) (148). The most significant differences between micellar and lipid conditions were found at the ␤-strands edges, with nanodisc structures having a few additional residues at the ends. It is best seen at the top of strands 1, 3, and 8 as well as at the bottom of JBC REVIEWS: Structural studies of MPs using solution NMR strands 1, 2, 4, 5, and 8. Thus, as one may easily imagine, the stability at the edges of the barrel depends strongly upon the reconstitution environment. Varying lengths of lipids might accommodate residues of the TM region differently. Functional assays in different mimetics should levy credence toward selecting the best mimetic for structural studies.

Conclusions and future directions
In this review, we discuss technological developments that allowed for the investigation of membrane proteins using solution NMR techniques. We also analyze the advantages and challenges associated with different membrane mimetic systems utilized for studying IMPs. Of all the different newlyemerging mimetics, it is interesting to note that there is not a single preferred mimetic for all IMPs. This strongly reflects on the complex architecture and dynamics of different IMPs. Although X-ray crystallography and, more recently, cryoEM demonstrate higher efficiency in determining structures of large macromolecular complexes, solution NMR remains the most compelling technique to investigate binding kinetics, conformational diversity, and dynamic properties of membrane proteins, especially for small proteins recalcitrant to crystallization. In addition to attempts at delineating protein structure, understanding dynamic modalities of a protein that regulate its function could become the core strength of future NMR applications. Selective labeling and a combination of different experimental techniques will lead in the future to answering specific questions regarding a protein's modus operandi, taking advantage of both in vitro and in vivo applications of solution NMR. Altogether, we should be able to glean enough information to expedite novel strategies for therapeutic interventions of numerous diseases associated with IMP malfunctions.