Lipid‐like Peptides can Stabilize Integral Membrane Proteins for Biophysical and Structural Studies

Abstract A crucial bottleneck in membrane protein structural biology is the difficulty in identifying a detergent that can maintain the stability and functionality of integral membrane proteins (IMPs). Detergents are poor membrane mimics, and their common use in membrane protein crystallography may be one reason for the challenges in obtaining high‐resolution crystal structures of many IMP families. Lipid‐like peptides (LLPs) have detergent‐like properties and have been proposed as alternatives for the solubilization of G protein‐coupled receptors and other membrane proteins. Here, we systematically analyzed the stabilizing effect of LLPs on integral membrane proteins of different families. We found that LLPs could significantly stabilize detergent‐solubilized IMPs in vitro. This stabilizing effect depended on the chemical nature of the LLP and the intrinsic stability of a particular IMP in the detergent. Our results suggest that screening a subset of LLPs is sufficient to stabilize a particular IMP, which can have a substantial impact on the crystallization and quality of the crystal.


Introduction
Membrane proteins represent about one third of the proteins in living organisms [1] and play central roles in all physiological processes. Integralm embrane proteins (IMPs)a re able to shuttle, pump, exchange, bind, andt ransmit molecules and signals across the membrane on the microsecondt om illisecond timescale. [2] Membrane protein crystallographyh as made tremendous progress in the last decade(s), and recent advances in cryoelectron microscopy now allows structure determination( to nearatomic resolution) of large membrane proteins (and complexes thereof)withoutthe need for crystals. [3] However,the structural biology of integral membranep roteins is still hampered by the instability of many solubilized and purified membrane proteins. [4] The preparation of ap ure, monodisperse, and stable membrane-proteins ample currentlys eems to be the bottleneck in the structuralb iology of membrane proteins. [5] Deter-gents are commonly used to extract IMPs from their native membrane environment by forming am icelle belt around their hydrophobic transmembrane domains (TMs). [6][7][8] Although detergents are able to solubilize IMPs, they are poor membrane mimics, and the loss of membrane pressure can induce conformationald ynamics, which contribute to increased sample heterogeneity,d ramatically reduced thermodynamic stability, and, hence,l ower success rates in crystallization. [8,4,5] For the crystallization of integral membrane proteins, as maller micelle size achieved by reducing the aliphatic chain length has, in general, ap ositive effecto nthe diffraction properties of the crystals owing to easier crystal contact formation. However,s horter chain detergents typically have ah igher denaturing potential than their long-chain counterparts. [9,10] In addition, the activity of many IMPs in detergents is significantly reduced relative to that in their native lipid environment, [11,12] which furthere mphasizes the suboptimal properties of many detergentsi n membrane protein structural biology.
Lipid-like peptides-sometimes also called peptergents [13] have been proposed as alternatives to overcome the detrimental limitations of classical detergents and to expand the toolbox for membrane protein biochemistry. [14][15][16] These lipid-like peptides have detergent-like properties and consist of as hort hydrophobic tail produced by repeating copieso fn onpolar amino acids and ah ydrophilic head group. The head can be neutral or positively( Lys, Arg, His) or negatively (Glu, Asp) charged.I nitial biochemical studies with as mall set of peptide sequences were highly promising in terms of solubilization efficiencies (if used in addition to detergents) and functional longterm stability of variouse xtracted and purified IMPs (e.g.,g lycerol-3-phosphate dehydrogenase, photosystem-I, rhodopsin, and olfactory receptors) compared to classicald etergents. [15,13,[16][17][18][19] Ac rucial bottleneck in membrane protein structural biology is the difficulty in identifying ad etergent that can maintain the stabilitya nd functionality of integral membrane proteins (IMPs). Detergents are poor membrane mimics, and their common use in membrane protein crystallography may be one reasonf or the challenges in obtaining high-resolution crystal structureso fm any IMP families. Lipid-like peptides (LLPs) have detergent-like properties and have been proposed as alternatives for the solubilizationo fGprotein-coupled re-ceptorsa nd other membrane proteins. Here, we systematically analyzedt he stabilizinge ffect of LLPs on integral membrane proteins of differentf amilies. We found that LLPs could significantly stabilize detergent-solubilized IMPs in vitro. This stabilizing effect depended on the chemical nature of the LLP and the intrinsic stability of ap articular IMP in the detergent. Our results suggest that screening as ubset of LLPs is sufficient to stabilizeaparticularI MP,w hich can have as ubstantial impact on the crystallization and quality of the crystal. In this study,w es ystematically evaluated the effect of lipidlike peptides on the stability of integral membrane proteins. Thus, we made use of the change in intrinsic fluorescence upon heat unfolding by using ah igh-throughputd ifferential scanningf luorimetry device. With this setup, the transition midpoint (T m )v alues could be determined with high precision withoutt he need for additional dyes, as used for conventional thermofluor experiments. [20,21] Lipid-like peptides (LLPs) of different compositions (aminoa cids in the hydrophobic tail and the head group) and chain lengths were tested on five different integral membranep roteins ranging from prokaryotic transporters to eukaryotic pumps, and their effects on stability and crystallization were evaluated.

Design of lipid-like peptides
We designed as eries of small, amphiphilic lipid-like peptides possessing detergent-likep roperties andc onsisting of as hort hydrophobic tail and ah ydrophilic head group, [18] and we systematically varied the charges andl ength of the hydrophobic tail. Lipid-like peptides are typicallyb etween four and seven amino acidsl ong. This corresponds to al ength of approximately 2-3 nm, similart ob iological phospholipids. The molecular modelso fa ll the LLPs used in this study are shown in Figure 1. Most LLPs were acetylated at the Nterminus and amidated at the Cterminus (see the table in Figure 1). For those LLPs that contained ap ositively charged amino acid at the Nterminus (e.g.,L LP6) or an egatively charged amino acid at the Cterminus (e.g.,L LP1, LLP7, LLP8, and LLP10), the termini were not modified to maintain the chargec haracteristics ( Figure 1).

Selectiono fintegral membrane proteins as test cases
We chose five different prokaryotic and eukaryotic integral membrane proteins as test cases covering aw ide range of sizes, functions, ands tabilities.S ome of these proteins are very well characterizedi nt erms of structure and function, whereas for others almost no functional or structural data are available. YkoE is at hiamin-specific vitamin-transport protein belonging to the energy-coupling factor (ECF) family of membrane transporters. Its structure was recently determined. [22] GlpG is an intramembrane proteasef rom Escherichia coli belonging to the rhomboid (serine) protease family,a nd its transmembrane domain is structurally very well characterized. [23][24][25] ACA8i sa plasma-membrane Ca 2 + -ATPase from Arabidopsis thaliana belongingt ot he P-type ATPase family of ion pumps. ACA8 is autoinhibited in its restings tate and becomes activated by binding of Ca 2 + -calmodulin to its regulatory domain. [26] ACA8 contains ten transmembrane helicesb ut also three large cytosolic domains. PepT St from Streptococcusthermophilus and the hypothetical sugar transporter from E. coli both belong to the major facilitator superfamily (MFS) transporter family and are involved in nutrient uptake. [27][28][29][30][31] Although PepT St has been extensively studied and its structure hasb een determined in detergenta nd in am ore lipid-like environment by using the lipidic cubic phase method, [32][33][34][35] not much is known about the hypothetical sugar transporter from E. coli.

Solubilization of IMPs by using lipid-like peptides
We first investigated whether the LLPs alone (without additional detergent) could be used to solubilize integral membrane proteins.For this purpose, we used an ACA8-green fluorescent protein (GFP) fusion to follow the solubilization efficiency spectroscopically.W eu sed as ystematic approach and tested various LLP concentrations (< 0.5 mg mL À1 )a nd incubations times (2-24 h), but we could not detect any significant solubilization of ACA8-GFP by the LLPs in the absence of detergents (data not shown). We concluded that although LLPs were previously reported to be able to solubilize GlpD in vitro, [13] they were not suitable to solubilize IMPs from the lipid bilayer in our hands. This discrepancy can be explained by the fact that GlpD is a monotopic membrane protein (not an IMP) that attaches only with asmall fraction of the protein from one side into one leaflet of the lipid bilayer and can therefore be removed much easier from the membrane. [13] The IMPs used in this study, however,c omprise between six and 14 transmembrane helices and are fully integrated in the membrane bilayer.F or this reason,t hey cannotb es olubilized with LLPs alone. The previously observed solubilizationo fG protein-coupled receptors (GPCRs) during cell-freep rotein synthesis( CFPS) [18] cannot be comparedt ot he setup used in this study, as during CFPS the polypeptide chain is produced in the absence of al ipid bilayer, which can result in the formation of an onfunctional protein as ar esult of the lack of tertiary-structure formation. [36] Lipid-like peptides can stabilize detergent-solubilizedI MPs Next, we analyzed whether detergent-solubilized and purified integral membrane proteinsc ould be stabilized against heat denaturation by the LLPs. The thermals tability of ap articular IMP is considered to be one of the key parameters that can be used to determine its protein structure successfully. [37,38] Highthroughput methods to determine the thermal stability of the IMPs were limitedi nt he past and often dependedo nt he use of fluorescent dyes such as 7-diethylamino-3-(4'-maleimidylphenyl)-4-methylcoumarin (CPM) and SYPRO Orange, [21,39,40] which are not alwaysc ompatible with the detergents used for IMPs or display different couplinge fficiencies depending on the pH. Alternative thermalunfolding methods that can usually not be performed in high-throughput processes include far-UV/near-UV circular dichroism (CD)spectroscopy,w hich reports on the secondary/tertiary structures, [41] and differential scanning calorimetry (DSC), which directly measures the absorbed heat upon unfolding but requires as ignificant amount of protein. [42] Here, we measured the thermalu nfolding of variousI MPs in solution by using differential scanning fluorimetry (DSF). We used the intrinsic fluorescenceo ft he aromatic residues and observedt he change at l = 330 and 350 nm after excitation at l = 280 nm ( Figure 2). For all of the investigated IMPs, the DSF traces allowed unambiguous assignment of at ransition midpoint (T m )t hat shifted in response to the addition of the LLPs (Figure 3). For several of the LLPs, this stabilization effect was concentration dependent,w hereas for others no significant stabilization and thus no concentration dependence could be observed ( Figure4).
Subsequently,w es ystematically screened five different IMPs, coveringarange from small prokaryotic membrane transporters to large eukaryotic membrane pumps, with regard to their stabilization effect by the addition of the LLPs. Each membrane protein target was purified in two different detergents to be able to comparet he effects of detergent/LLPc ombinations. We found that PepT St from S. thermophilus and the hypothetical sugar transporter from E. coli were both stabilized by a number of the investigated LLPs, in particular LLP7 and LLP8. The thermals tabilities of thesep roteins were both increased by up to 4.5 8C. The general stabilizing effect was independent of the detergenti nw hich the IMP had been purified;h owever, subtle differences exist with respectt ot he degree of stabilization by the different LLPs. For example, LLP1 stabilized the hypothetical sugar transporter from E. coli in dodecylmaltoside (DDM) by 3.8 8C, but the same protein purified in lauryl maltose neopentyl glycol (LMNG) stabilized it only by 1.3 8C ( Figure 5A and B). For YkoE, we also observed ag eneral stabilizing effect of up to 3 8Cb ym any of the LLPs. As notable exceptions, LLP1 and LLP3 destabilized YkoE in decylmaltoside (DM) by 1.8 and 5.9 8C, respectively (Figure5Gand H). GlpG revealed an interesting stability-response pattern. GlpG purified in the nonylglucoside (NG) detergentw as stabilized by 12.5 and 11.1 8Cb yt he addition of LLP2 and LLP9, respectively (Figure 5E and F). Severalo ther LLPs hadasmalld estabilizing effect in NG. In the DDM detergent, only LLP8, LLP9, and LLP10 had as tabilizing effect ( % 2 8C) on GlpG unfolding (Figure 5F). Notably,G lpG is highly stable in DDM and nonylmaltoside (NM) with at ransition midpoint above 80 8C( data not shown), which probablym akes further stabilization by LLPs unlikely.I nt he NG detergent, the transition midpointi nt he absence of the LLPs was 60 8C, which could be increased by > 10 8Cu pon the addition of LLP2 and LLP9. IMPs with lower thermals tability are typically more strongly stabilized by LLPs. However,f or ACA8 we observed only minor stabilizing or destabilizing effects (AE 1 8C) by the addition of the LLPs independento ft he detergent used ( Figure 5I and J). This could be due to the fact that this calcium pump is composed of three cytoplasmic domains in addition to the transmembrane domain that accounts for approximately 50 %o ft he protein, Figure 2. DSF transition curve and first derivative to illustrate the stabilization of IMPs by LLPs. The maximum of the first derivativeist he unfolding transition midpoint that was used to quantifythe (de)stabilization effect. The data correspond to the hypothetical sugar transporter in the LMNG detergent. Experiments were performed with ap rotein concentrationo f 0.7 mg mL À1 purified IMP in the absence and presenceo f2.5 mm LLP8 at af inal DMSOc oncentrationo f4%b yu sing ah eating rate of 1 8Cmin À1 .

Effect of lipid-like peptides on the crystallization of IMPs
To investigate whether LLPs couldh ave an effect on the crystallization of integral membrane proteins,w es et up crystalliza-tion trials with the hypothetical sugar transporter from E. coli in the presence of the two most-stabilizing LLPs. By using the commercial MemGold2 crystallization screen, [43] we observed the formationo fl arge single crystals, suitable for X-ray diffraction experiments, in 20-30 %o ft he conditions when using LMNG as the detergenta nd LLP7 or LLP8 as the additive ( Figure 6). Without the presence of the LLPs, crystals only appearedi ns even (out of 96) conditions ( Figure 6F). The positive effect of theses tabilizing LLPs on the crystallizability of the E. coli transporter seemedt od epend on the detergent used, as we could not observe this effect with protein purified in DDM. Wes peculate that the stabilizing effect might be due to increased micelles tiffness (see the Discussion below). These initial crystalsc o-crystallized with LLP8 were exposed to X-rays and were diffracted anisotropically beyond5 resolution in the best direction ( Figure S1 in the Supporting Information), which is significantly bettert han the initial crystal hits obtained for protein crystallizedi nt he absence of LLPs (max. 15 diffraction). Therefore, we believe that LLP8 promotes tighter packing of the transporter molecules in the crystal lattice, which results in improved diffraction properties of these crystals.I na ddition, we could also obtain crystals of GlpG in the presence of stabilizing LLP2 and LLP9 ( Figure S2). These crystalsa ppeared under conditions that did not support crystal growth in the absence of the LLPs, which further supports the ability of stabilizing LLPs to enlarge the crystallization space.

Discussion
The use of detergents with all of their detrimental properties is one of the key limitations in structurals tudies of IMPs, because they are poor mimics of the lipid bilayer.N umerous detergents are available and have been tested to solubilize, purify,a nd crystallize IMPs. It is well known that short-chain detergents such as NM and NG are more denaturing than long-chain detergents such as DDM [44] (Figure S3), but at the same time, short-chain detergents are beneficial for crystallization because of their reduced micelle size. Recent molecular dynamics simulations suggested that this destabilizing effect could be explained by ad ecrease in a-helicity in the transmembrane regions (owing to as maller micelle size of the detergenta nd, consequently,e xposure of al argerf raction of the hydrophobic surface), suboptimal a-helicalp acking, and the interpenetration of detergentm olecules between TM-helices. [45] Interestingly,c holesteryl hemisuccinate (CHS) in dodecylmaltoside was found to stabilize selected GPCRs by wedging into hydrophobic crevices and stiffening the detergentm icellea sw ell as by forming ab icelle-like micelle architecture. [45,46] The stabilizing effect of certain LLPs on specific IMP targetp roteins as shown here might be causedb yg eneral micelle stabilization (change in micelle size or dynamics by integration of LLPs), specific LLP-IMP interactions, or ac ombination of both. The fact that variousL LPss howo pposite effects on various IMPs indicates specific interaction sites on the different proteins.I ti st empting to speculate that the extremely pronounced stabilization of GlpG by LLP2a nd LLP9 ( Figure 5E and F) might be caused by those peptides acting as substrates. Binding to the active site of GlpG may lead to at ighter interaction of TM-helices and, thus, might confer higher thermal stability.

Conclusions
In summary,l ipid-like peptidese xtend the toolbox to stabilize integral membrane proteins in solution not only for structural studies but also for functional studies. Our results indicated that screening as ubset of LLPs was sufficient to identify suitable LLPs;t his makes them an additional tool fors uccessful structuralbiology research.

Experimental Section
Materials:T he detergents used for purification were from Anatrace (Maumee, OH, USA), and crystallization reagents were from Qiagen. All other chemicals were of analytical grade and were obtained from Roth, unless otherwise stated. Peptides were purchased from GL Biochem (Shanghai, China). Peptide identity was confirmed by LC-MS. LLP stock solutions at 62.5 mm in DMSO were prepared for DSF and crystallization experiments.
Expression and purification of integral membrane proteins:T he protein PepT St was expressed and purified in the n-dodecyl-b-dmaltopyranoside (DDM) detergent, as described previously. [47] For the protein batch of PepT St in the n-nonyl-b-d-maltopyranoside (NM) detergent, the protein was initially solubilized and purified in the DM detergent up to the immobilized-metal affinity chromatography (IMAC) step, after which the detergent was exchanged to NM in the size-exclusion chromatography (SEC) step.
The hypothetical sugar MFS transporter from E. coli (NP_418 146.4) was identified as as uitable target for structural studies by using our recently described pipeline approach. [47] The gene was cloned into the C-terminally hexahistidine-tagged vector pNIC-CTHF [48] by using ligation-independent cloning. [49] The vectors possessed a tobacco etch virus (TEV) cleavage site for Ta gr emoval. Protein expression in E. coli C41, membrane solubilization, and purification by using IMAC and SEC were performed essentially as described, [47,50,51] except that two different detergents, DDM and LMNG, were used. The protein was eluted from the IMAC column by passing TEV protease over the beads, as the use of high imidazole concentrations for the elution step was found to destabilize the protein strongly. [47] For the gel-filtration step, HEPES (20 mm, pH 7.5), NaCl (150 mm), 5% glycerol, and Tris(2-carboxyethyl)phosphine (TCEP;0 .5 mm)w ith the corresponding detergent (0.4 %N M, 0.03 %D DM, 0.01 %L MNG) were used.
YkoE from Paenibacillus polymyxa M1 was expressed and purified in am anner similar to that previously described for the homologue from Bacillus subtilis. [22] Briefly,t he protein was expressed with an N-terminal His 6 tag in E. coli Lemo21 strain. [52] Membranes were isolated and solubilized in 1% DDM at 4 8Cf or 1h.P rotein was purified by using Ni-NTAr esin, which was followed by TEV cleavage to remove the His 6 tag and re-passage of the digested material over Ni-NTAr esin. The flow through was further purified by using gel-filtration chromatography in Tris·HCl (25 mm,p H7.4), NaCl (300 mm), and 3% glycerol (using 0.4 %N Mor0 .2 %DM).
The transmembrane part of GlpG from E. coli (residues 91-270) was prepared as follows:F ull-length GlpG was expressed with an N-terminal His tag in E. coli Lemo21 cells at 20 8Co vernight. Membranes were isolated and solubilized with DDM. The protein was purified by using Ni-NTAr esin, and the protein was eluted with 0.3 %N G. Subsequently,G lpG was incubated with chymotrypsin in ar atio of 1:50 at 4 8Cf or 36 ht or emove the N-terminal soluble domain. The truncated GlpG was purified by gel filtration in Tris·HCl (25 mm, pH 8.0), NaCl (150 mm), and 0.3 %NG. [23,24] Full-length ACA8 from Arabidopsis thaliana was expressed in Saccharomyces cerevisiae BJ5460 [53] with an N-terminal His 6 tag. Membranes were isolated and solubilized in LMNG. The protein was purified by using Ni-NTAr esin in Tris·HCl (30 mm,p H8.0), NaCl (300 mm), MgCl 2 (1 mm), CaCl 2 (1 mm), b-mercaptoethanol (2 mm), and 0.01 %L MNG and was eluted with imidazole (150 mm). Fractions containing ACA8 were concentrated and were further purified by size-exclusion chromatography.
Thermal stability measurements using nanoDSF:T he stability of the different purified protein preparations in the presence and absence of the LLPs (0-2.5 mm)w as followed by using an anoDSF differential scanning fluorimeter (Prometheus, NanoTemper Te ch-nologies, Munich). Here, the intrinsic fluorescence at l = 330 and 350 nm after excitation at l = 280 nm was used to monitor the fluorescence change upon heat unfolding. Up to 48 samples could be measured in parallel without the addition of ad ye. Typically, protein solution (10 mL) at ac oncentration of 0.5-2 mg mL À1 was loaded in ac apillary,a nd the unfolding was then measured at ah eating rate of 1 8Cmin À1 .T he first derivative of the unfolding curves was used to determine the transition midpoint. Given that the LLPs were prepared in DMSO, respective control experiments of the different IMPs in the presence of DMSO were performed. All analyzed samples contained 4% DMSO. The stability of the analyzed protein was not significantly changed in the presence of 4% DMSO. For GlpG and ACA8, well-established activity assays exist, [26,54,55] which revealed similar activity in 4% DMSO. In addition, the concentration dependence of the transition midpoint was determined in the range of 0.2 to 5mgmL À1 for each protein batch. Resulting transition midpoints were within 1 8C.
Crystallization of IMPs in presence of lipid-like peptides:C rystallization trials with the commercially available MemGold2 screen [43] were performed by vapor diffusioni n9 6-well sitting-drop plates at 293 K. The volume of crystallant added to the reservoir was 50 mL, whereas the drops had at otal volume of 300 nL and were composed of the E. coli transporter sample at ac oncentration of 5mgmL À1 in the absence and presence of LLPs at 2.5 mm and the crystallant in ratios of 1:2, 1:1, and 2:1. After 14 days of incubation, crystal plates were scored, and the results were compared. The crystals were typically flash frozen in liquid nitrogen without prior soaking in cryobuffer and were tested at the synchrotron beamline P13. [56]