Parallel screening and optimization of protein constructs for structural studies

Abstract

A major challenge in structural biology remains the identification of protein constructs amenable to structural characterization. Here, we present a simple method for parallel expression, labeling, and purification of protein constructs (up to 80 kDa) combined with rapid evaluation by NMR spectroscopy. Our approach, which is equally applicable for manual or automated implementation, offers an efficient way to identify and optimize protein constructs for NMR or X-ray crystallographic investigations.

Introduction

The structural characterization of modular proteins and biomolecular complexes remains a considerable practical challenge. To comprehend the workings of such systems at an atomic resolution requires the production of stable protein samples amenable to structural characterization. Studying biological assemblies often involves optimizing each component separately and reforming the complex in vitro. Likewise, investigating the structure and function of modular proteins can require each domain being excised and studied independently.¹ Although this strategy is commonly applied, it relies on finding autonomously folding constructs of each target domain. The success of a structural biology project is heavily dependent on the design of tractable protein constructs.

One of the major consequences of the structural genomics revolution has been the implementation of high-throughput construct screening.^2–6 In these genome-wide efforts large number of proteins undergo initial screening on a small scale (0.1–0.5 mL) to determine expression level and solubility.^4,5 The folding state and solution behavior of high yield targets are then often evaluated by solution NMR spectroscopy through the acquisition of [¹H,¹⁵N]-correlation spectra.^2–7 This type of spectrum provides a “fingerprint” of the protein which informs on the suitability of a sample for structural studies and also on the degree of folding of the protein. However, obtaining sufficient levels of the target protein for such analyses has historically required purifying [¹⁵N]-labeled protein from 0.1 to 1 L cultures of E. coli.^2,3

Technical and methodological advances in the field of NMR spectroscopy, including the development of cryogenically cooled probeheads and the use of tailored pulse sequences, have pushed the sensitivity threshold of the technique. Consequently, [¹H,¹⁵N]-correlation spectra can now be obtained from microgram quantities of protein.^8–10 Taking advantage of these improvements, we have developed a protocol that allows rapid evaluation of the feasibility of structural studies for multiple protein targets expressed in parallel in E. coli. We have accelerated construct screening by reducing the common pipeline described above to a simple small-scale, fully automatable process. Mindful of the facilities usually available in academic laboratories we also sought to ensure our approach was equally amenable for manual implementation.

Here, we describe our protocol and demonstrate its capability to screen protein targets with molecular weights of up to 80 kDa. Using a multidomain plant protein involved in micro-RNA processing as an example, we show that our approach can quickly define domain composition, assess domain structure, and report on interdomain interactions. We also illustrate the utility of our methodology for optimizing protein constructs for crystallization.

Results and Discussion

Small-scale labeled protein production and spectra acquisition

We devised a procedure, termed “RoBioMol-NMR,” which is outlined schematically in Figure 1(A). Isotope-labeled protein samples are obtained directly from 4 mL cultures of E. coli grown in parallel in minimal medium supplemented by [¹⁵N]-enriched algal extract. The use of a rich medium ensures higher yields than would be obtained from standard minimal medium (e.g., M9). Proteins are expressed with an N-terminal six his-tag and, if present in the soluble fraction, purified using a liquid handling robot and prepacked 96 wells Ni²⁺ sepharose columns. The insoluble fraction is solubilized in 8M urea and denatured protein targets purified using the same prepacked columns. Where possible, purified denatured proteins are refolded by 25-fold dilution into refolding buffer and reloaded on the same column for further purification and concentration. Processing the insoluble fractions allows the opportunity of rescuing potentially valuable targets that would otherwise be discarded.¹² Samples obtained from both soluble and insoluble purification procedures are then buffer exchanged to remove imidazole. To maximize the signal to noise ratio and minimize the effect of salt, 150 μL of protein solution is placed inside a 3-mm NMR tube concentric to a regular 5-mm tube containing D₂O (Supp. Info. Fig. 1). [¹H,¹⁵N]-HMQC spectra of each purified sample are acquired using the SOFAST pulse sequence developed in our laboratory in order to maximize the sensitivity per unit time.⁹ Adopting a parallel protocol allows multiple targets to be conveniently isotopically labeled, purified, and screened by 2D NMR in 3 days, starting from transformed E. coli cells. The use of small-scale preparation greatly lowers costs due to the reduced requirement for isotope-enriched media. Example spectra obtained from several proteins produced using the present protocol are shown in Supporting Information Figure 2. High-quality fingerprint [¹H,¹⁵N] spectra can be obtained in 10–60 min even for proteins up to 20 kDa [Supp. Info. Fig. 2(D)]. The quality of these spectra is impressive considering the small cell culture volumes and short acquisition times used. Furthermore, these spectra clearly identify protein constructs worthy of further analysis and those which require additional optimization. To assess the feasibility of studying larger protein systems, we tried expressing proteins from E. coli cultures grown in [²H,¹⁵N]-enriched algal extract. [¹H,¹⁵N]-TROSY spectra¹¹ recorded of proteins with molecular weights ranging from 28 to 82 kDa [Fig. 1(B) and Supp. Info. Fig. 2(E,F)] purified from [²H,¹⁵N]-enriched 4 mL cultures also yielded high-quality spectra. These results demonstrate that the RoBioMol-NMR parallel screening strategy has the capability to evaluate a large range of protein sizes up to ∼80 kDa.

Figure 1.

A: Schematic illustration of the automated RoBioMol-NMR protocol for protein expression labeling and purification. B: [¹H,¹⁵N]-NMR spectrum of deuterated malate synthase G (82 kDa, 723 residues) produced from a 4 mL culture using the protocol described in panel A. The TROSY spectrum¹¹ was acquired on an 800 MHz spectrometer equipped with a cryogenic probe in 12 h. The inset shows an expansion of the boxed region.

Figure 2.

Analysis of the domain structure of HYL1. A: Schematic representation of the HYL1 constructs. SOFAST-HMQC spectra⁹ of dsRBD1 (B), dsRBD2 (C), Kh domain (D) dsRBD1 + 2 (E), and the triple domain (F) constructs. Spectra for constructs B, C, and E were acquired using samples refolded from the urea-solubilized fraction. All other constructs were purified from the soluble fraction. Spectra in B–E were acquired in 15–60 min on a 600 MHz spectrometer equipped with a cryogenic probe, while the spectrum F was acquired in 12 h. A schematic representation of each construct is given in the top right corner of each spectrum. Insets show an expansion of the same region for all spectra. In (F), a color coded overlay of the individual spectra is shown in the inset. The amount of protein obtained from the 4 mL cultures and used to acquire spectra was 85 μg (B), 70 μg (C), 120 μg (D), 120 μg (E), and 30 μg (F).

Identification of protein domains

We applied the RoBioMol-NMR protocol to expedite the identification and characterization of domains from the A. thaliana protein HYL1, a member of a family of processing proteins that participates in microRNA accumulation.^13,14 Amino acid sequence analysis suggests that HYL1 has two double-stranded RNA binding domains (dsRBDs) and a Kh nucleic acid binding domain. To experimentally determine domain boundaries and find the potential interactions between individual domains, we tested five different constructs: each of three predicted single domains, a double domain, and a triple domain construct [Fig. 2(A)]. The isolated dsRBD-1 and dsRBD-2 as well as the construct spanning both dsRBDs showed high expression levels, but proved difficult to purify in the soluble fraction. Spectra for these constructs were acquired using samples refolded from the urea-solubilized fraction. All other constructs were purified from the soluble fraction. Figure 2(B–D) shows the SOFAST-HMQC spectra obtained from the three single domain constructs. The large dispersion of signals in the proton dimension suggests that the two single dsRBD constructs fold independently [Fig. 2(B,C)] whereas the Kh domain appears unfolded [Fig. 2(D)]. Spectra of longer constructs containing the Kh domain reveal that this domain is also unfolded when expressed in the context of the whole protein [Fig. 2(F)]. The spectra of each dsRBD superimpose perfectly on the top of the spectrum of the full-length construct [inset Fig. 2(E,F)] strongly suggesting that HYL1 is composed of two independent dsRBD domains that do not interact, followed by a long nonstructured tail. These data are consistent with the recent demonstration that the two dsRBDs of HYL1 are sufficient for function in vivo.¹⁵

Crystallizability improvement

As a demonstration of potential applications of the RoBioMol-NMR strategy, we sought to more precisely define the folded core of dsRBD-1 of HYL1. Solution NMR spectroscopy offers a convenient and quick way of determining the presence of unfolded regions in a protein. A series of dsRBD-1 constructs were designed, expressed, and evaluated in parallel using the approach described earlier (Supp. Info. Fig. 3A). Of the five constructs tested only dsRBD-1A to 1C showed promising expression levels. Constructs dsRBD-1A and 1B could be purified from the refolded fraction and yielded good NMR spectra (Supp. Info. Fig. 3B). The spectrum of the longer construct shows a number of signals crowded in the center, which is indicative of the presence of unfolded, flexible residues. Most of these overcrowded signals are absent in the spectrum of the dsRBD-1B construct, while the more dispersed signals of both constructs superimpose well (Supp. Info. Fig. 3B). These results indicate that unfolded regions present in dsRBD-1A have been removed without affecting the overall domain structure. A potential benefit of more precisely defining the minimal folded core of protein domains is an improvement of the crystallizability of the target protein.^16–19 To demonstrate this, dsRBD-1A and 1B were subjected to a parallel crystallization assay using 576 solution conditions. The full-length dsRBD-1A construct showed crystal growth in only nine conditions whereas dsRBD-1B construct produced crystal leads in 56 conditions. This example shows that the RoBioMol-NMR method may be employed to optimize domain crystallization by applying a parallel rational design of the constructs before initiating crystallization trials.

Materials and Methods

Bacterial culture growth

Bacterial cells were grown in M9/H₂O medium supplemented by 10 g/L [¹⁵N]-ISOGRO™ powder (Isotec-Sigma Aldrich), 4 g/L d-glucose, 1 g/L [¹⁵N]-ammonium chloride, 1.8 g/L K₂HPO₄, 1.4 g/L KH₂PO₄, 1 g/L MgSO₄, 0.1 g/L CaCl₂ and 30 mg/L kanamycin or 100 mg/L ampicillin, depending on the plasmid used. Precultures (1 mL) were started from a single colony of freshly transformed E. coli BL21(DE3) Star (Invitrogen) in a 24-well plate and allowed to grow overnight at 37°C, 200 rpm. The following day, 160 μL of the precultures were inoculated in 4 mL of isotopically labeled culture medium in a 24-well plate. Cultures were grown for 2 h at 37°C with shaking, induced with 1 mM IPTG (typically at OD₆₀₀ = 0.7–0.9) and grown further for 4 to 6 h. Then cells were harvested by centrifugation and stored at −20°C. Deuterated proteins were obtained using a similar protocol based on a M9/H₂O medium supplemented with [¹⁵N,²H]-labeled growth supplement in which unlabeled glucose and [¹⁵N]-ISOGRO™ have been replaced by d-glucose-d₇ and [²H, ¹⁵N]-ISOGRO™.

Native protein purification

His-tagged proteins present in the soluble fractions were purified in an automated manner by means of a Hamilton MicroLab Star robotic platform using the protocol described below. Each cell pellet from 4 mL cultures was resuspended in 500 μL of Bug Buster 1× (Novagen), 50 mM Tris pH 7.5, 500 mM NaCl, 0.4 μL Benzonase (Novagen), 40 U from rLysozyme™ solution (a ready-to-use solution of highly purified and stabilized recombinant lysozyme produced by Novagen, reference #71110-3), and incubated with gentle stirring for 20 min to allow the bacterial cell lysis to proceed. The lysates were clarified by centrifugation (10 min at 13,000 rpm on a refrigerated tabletop centrifuge) and the supernatants were loaded on His MultiTrap HP 96 well-plate (GE Healthcare) Columns were pre-equilibrated with native buffer (Tris 50 mM, NaCl 500 mM, pH 8) using a QIAvac 96 (Qiagen). The columns were washed with 500 μL native buffer with 50 mM imidazole and then eluted with 2 × 100 μL of the same buffer with 500 mM imidazole. The excess imidazole was removed using BioRad Micro Bio-Spin 6 columns as it gives interfering signals in the [¹H,¹⁵N] correlation spectrum.

Denaturing purification

Bacterial cell pellets were resuspended in 500 μL of denaturing buffer (8M Urea, 100 mM Phosphate, 10 mM Tris, 5 mM β-mercaptoethanol) at pH 8.0 and lysed by a single freeze-thaw cycle in liquid nitrogen and a 65°C water bath. The lysates were clarified by centrifugation 10 min at 13,000 rpm in a tabletop centrifuge at room temperature. The supernatants were loaded on His MultiTrap HP 96 well-plate (GE Healthcare) pre-equilibrated with denaturing buffer at pH 8. Columns were then washed three times with 500 μL denaturing buffer at pH 6.3 and the denatured proteins were eluted two times with 100 μL denaturing buffer at pH 4.5. Refolding was achieved by dilution of the 200 μL denatured protein sample in 5 mL of refolding buffer (Tris 50 mM, NaCl 500 mM, β-mercaptoethanol 5 mM, pH 8.0). The same columns used for the denaturing purification were equilibrated with four washes of 500 μL refolding buffer, and the refolded protein sample was loaded in the corresponding columns in 500 μL steps. Finally proteins were eluted with 4 × 50 μL refolding buffer with 500 mM imidazole, and excess imidazole was removed using BioRad Micro Bio-Spin 6 columns.

NMR spectra acquisition

Purified protein samples were loaded in 3-mm NMR tubes (2.3 mm inner diameter) placed inside regular 5-mm NMR tubes containing D₂O (Supp. Info. Fig. 1). If required, the samples were concentrated to 150 μL using 3 kDa MWCO Centricon centrifugal concentration units, to optimize the amount of protein present in the active coil volume.

For [¹⁵N]-labeled samples, [¹H-¹⁵N]-HMQC spectra were acquired on a Varian Direct Drive 600 MHz spectrometer equipped with a cryogenic probe using the SOFAST-HMQC pulse sequence^9,20 with typically a 0.5-s recycle delay, 1800 Hz sweep width, 50 increments in the indirect dimension and 4 to 128 scans per increment (depending on sample concentration). Spectra were processed using the nmrPipe software.²¹

For deuterated samples, [¹H,¹⁵N] correlation spectra were acquired on a Varian Inova 800 MHz spectrometer equipped with a cryogenic probe using a TROSY pulse sequence.¹¹ Typical acquisition parameters were 1.6-s recycle delay, 2400 Hz sweep width, and 128 increments in the indirect dimension, with 16 or 96 scans per increment depending on sample concentration. Spectra were processed using maximum entropy reconstruction.^22,23

Crystallization assays

The two constructs from the first domain of HYL1 showing good expression and NMR spectra were expressed in large scale, purified to homogeneity, and concentrated to 1.2 mM. Both proteins were subject to a parallel high-throughput crystallization screening using an automated TECAN crystallization robot on 96 reservoirs/288 wells crystallization plates. A set of 576 conditions based on standard Qiagen kits were assayed in six plates. For each condition, sitting drops of each protein were deposited in the two wells for equilibration with the same reservoir chamber, thereby ensuring exactly the same conditions for both proteins. The plates were inspected for the presence of crystals after 1 week, 2 weeks, and 2 months of the starting day of crystallization.

Conclusions and Perspectives

We have presented here a convenient way to express and screen multiple protein constructs in parallel. We have shown that protein constructs obtained from just 4 mL of culture can be evaluated rapidly by NMR spectroscopy. This reduction in culture size dramatically decreases costs and therefore permits more constructs to be assessed in parallel. We have demonstrated the capability of the RoBioMol-NMR procedure to allow efficient optimization of protein constructs for structural biology. Numerous other applications for our method can be conceived, such as using the analytical power of NMR spectroscopy to assess correct folding or to investigate function of protein constructs following site-directed mutagenesis. The simplicity of the approach makes it amenable for manual implementation in most laboratories equipped for structural biology. Moreover, the strategy is equally suited for high-throughput screening of large number of protein targets by structural genomics initiatives.

Glossary

Abbreviation:

dsRBD

double-stranded RNA binding domain