Rapid, robotic, small-scale protein production for NMR screening and structure determination

Abstract

Three-dimensional protein structure determination is a costly process due in part to the low success rate within groups of potential targets. Conventional validation methods eliminate the vast majority of proteins from further consideration through a time-consuming succession of screens for expression, solubility, purification, and folding. False negatives at each stage incur unwarranted reductions in the overall success rate. We developed a semi-automated protocol for isotopically-labeled protein production using the Maxwell-16, a commercially available bench top robot, that allows for single-step target screening by 2D NMR. In the span of a week, one person can express, purify, and screen 48 different ¹⁵N-labeled proteins, accelerating the validation process by more than 10-fold. The yield from a single channel of the Maxwell-16 is sufficient for acquisition of a high-quality 2D ¹H-¹⁵N-HSQC spectrum using a 3-mm sample cell and 5-mm cryogenic NMR probe. Maxwell-16 screening of a control group of proteins reproduced previous validation results from conventional small-scale expression screening and large-scale production approaches currently employed by our structural genomics pipeline. Analysis of 18 new protein constructs identified two potential structure targets that included the second PDZ domain of human Par-3. To further demonstrate the broad utility of this production strategy, we solved the PDZ2 NMR structure using [U-¹⁵N,¹³C] protein prepared using the Maxwell-16. This novel semi-automated protein production protocol reduces the time and cost associated with NMR structure determination by eliminating unnecessary screening and scale-up steps.

Statement for Broader Audience

Using a compact commercial bench-top robot, we accelerated the process of ¹⁵N-labeled protein production for structural genomics NMR screening by more than 10-fold. Moreover, this automated purification strategy easily yielded sufficient material to support 3D structure determination, and we speculate that it could replace most conventional large-scale protein production pipelines.

A persistent bottleneck in structural genomics is the identification of soluble, folded domains suitable for 3D structure determination. Some of the major centers for structural genomics employ NMR spectroscopy in target selection¹ and to augment X-ray crystallography as a complementary approach for 3D structure determination.^2–5 To meet their annual production goals, thousands of proteins or domains must be screened,⁶ and this demands that traditional methods for molecular cloning and protein purification be adapted to parallel operation.⁷ Automated processing of nucleic acids is now routinely performed using commercial instruments, but parallelized robotic protein purification has typically been achieved with costly customized systems⁸ or at production scales that are insufficient for screening by NMR spectroscopy.

After target selection, cloning and transformation into an expression host, a small-scale expression study is conducted and analyzed by SDS-PAGE to select targets expressing high levels of soluble protein.⁹ Additional screening at this stage using 1D ¹H NMR can identify folded proteins¹⁰ and predict success in crystallization trials,¹ however the 2D ¹H-¹⁵N HSQC spectrum is essential for selection of candidates for structure determination by NMR. Production of ¹⁵N-labeled proteins for NMR is typically a secondary screen requiring large-scale expression and purification at a substantial cost per target. For abundantly expressed, soluble target proteins, a large-scale (0.5–1 L) culture of ¹⁵N-labeled protein is subjected to immobilized metal affinity chromatography (IMAC) purification and the dispersion, number, and intensity of signals in the 2D ¹H-¹⁵N HSQC spectrum are evaluated before proceeding to production of the ¹³C/¹⁵N-enriched material needed to solve the structure. Production of ¹⁵N-labeled proteins is not easily parallelized and the entire screening process often requires two weeks to complete. Poorly behaved eukaryotic proteins often contain domains that would be amenable to structure determination if optimal sequence boundaries can be identified, but empirical analysis of multiple constructs is typically necessary. Protein production is thus a significant barrier to NMR-based screening of large numbers of domain constructs.

Cryogenic probe technology can dramatically increase the sensitivity of NMR measurements, but the enhancement is highly dependent on sample geometry. Improved mass sensitivity can be obtained with RF coils optimized for sample volumes smaller than the typical 5-mm diameter format.¹¹ Microcoil probes using 1 or 1.7 mm sample cells have been used for 1D¹² or 2D NMR screening,¹³ and this format is particularly valuable when protein yields are severely limiting. However, in our testing, the signal-to-noise ratio of an HSQC spectrum acquired on a 1.7-mm cryoprobe was 10-fold lower than one acquired on a 3-mm sample with a 5-mm cryoprobe using identical sample and acquisition parameters. Surprisingly, we observed that HSQC measurements using as little as 100 μg of ¹⁵N-labeled protein in a 3-mm sample cell yielded high quality spectra in 90 min or less, with little or no reduction in signal when compared with a sample of equal concentration in a 5 mm tube. Consequently, we investigated the possibility that small-scale expression testing and 2D NMR screening could be combined into a single high-throughput process. The Maxwell-16 (Promega, Madison, WI) is a small, relatively inexpensive bench top robot that enables the simultaneous lysis and IMAC purification of up to sixteen protein samples in ∼45 min. Cell lysis, resin binding, and four wash steps are performed in a seven-well cartridge followed by elution in a single cuvette [Fig. 1(a)]. An earlier study by Frederick et al. used the Maxwell-16 for expression testing of eukaryotic proteins but screening was limited to SDS-PAGE analysis.¹⁴ We speculated that an optimized approach to robotic purification could supply ¹⁵N-labled proteins in quantities sufficient for evaluation by 2D NMR.

Figure 1.

Optimization of the Maxwell-16 buffer conditions and workflow. (a) IMAC purification is performed in a seven-well cartridge. Magnetic IMAC beads are conveyed by a magnetic plunger that also serves as a mixing device. Cell lysis is accomplished by the addition of FastBreak cell lysis buffer and gentle agitation by the plunger in well 1. The plunger retrieves the MagneHis resin from well 2 and returns to well one to adsorb the His-tagged protein. The bound protein undergoes four washes in wells 3–6. The purified protein is deposited in a single cuvette at the end of the procedure. The binding, wash and elution steps are repeated once in every Maxwell-16 purification. (b) Flowchart comparing conventional and automated protein purification workflows. In a conventional two-stage scheme for target screening, a workgroup of 24 targets is transformed, cultured, and evaluated for soluble expression by SDS-PAGE during week 1. In week 2, up to 8 targets may be selected for large-scale expression in ¹⁵N-enriched minimal medium, purification and 2D ¹H-¹⁵N HSQC analysis. The optimized Maxwell-16 protocol parallelizes the protein purification process using 60 mL cultures, enabling a single technician to process up to 48 targets/week, reducing the total time for NMR screening by half, and increasing overall throughput by ∼10-fold. (c) Protein yields were enhanced by optimizing the volume of beads, lysis buffer, and elution buffer used. Additionally, we investigated how the elution buffer pH and imidazole concentration influenced release of protein from the MagneHis Ni-resin. SDS-PAGE analysis of the three test proteins revealed target-to-target variations. However, increasing the volume of MagneHis Ni-resin to 150 μL and lowering the pH of the elution buffer to 6.5 yielded the most consistent results. All studies were conducted with culture volumes corresponding to a total OD₆₀₀ = 60.

Results and Discussion

Evaluation of more than 10,000 recombinant protein constructs by structural genomics projects over the last 10 years has led to a consensus strategy for expression screening and purification of structure targets.⁶ In the most common approaches to target validation, the scale of production and stringency of the selection criteria increase in a stepwise manner [Fig. 1(b)]. After transformation of the expression plasmids and initial bacterial cell culture of a group of targets, expression levels and protein solubility are analyzed by SDS-PAGE. Based on these results, a smaller number of targets (∼8) are selected for large-scale (0.5–1 L) ¹⁵N-labeled production, IMAC purification, and screening by 2D NMR. Targets that satisfy three spectral criteria (high chemical shift dispersion; peak count similar to the number of amino acid residues; uniform peak intensities) are evaluated as “HSQC+” and proceed to stability testing before finally being selected for ¹⁵N/¹³C labeling and 3D structure determination by NMR.

Automated parallel purification of ¹⁵N-labeled proteins for 2D NMR screening could significantly increase throughput relative to the conventional pipeline strategy if each cartridge of the Maxwell-16 generates enough material for acquisition of a ¹H-¹⁵N HSQC spectrum. First, we sought to improve the yield of pure protein by maximizing the quantities of MagneHis beads (Promega, Madison, WI) and bacterial cell paste that could be reproducibly processed in a single cartridge (Supporting Information Fig. 1). The elution buffer was also varied to obtain the best combination of yield and purity for a set of three test proteins. As illustrated in Figure 1(c), optimal results were obtained using 0.15 mL of MagneHis beads, 50–60 mL of cell culture (OD₆₀₀ ∼ 1), and an elution buffer consisting of 50 mM sodium phosphate, 300 mM sodium chloride, 500 mM imidazole, and 0.02% azide at pH 6.5. Target proteins for structural genomics are often produced as fusions linked by cleavage sites for thrombin, TEV protease, or other proteolytic enzymes. By modifying the standard Maxwell-16 protocol to incorporate an additional incubation period, we also found that it is possible to automate TEV digestion and separation of a target protein from its fusion partner (Supporting Information Fig. 2).

We tested the optimized Maxwell-16 protocol on a control workgroup of eight proteins previously characterized at the Center for Eukaryotic Structural Genomics (CESG), including five successful NMR structure targets^15–19 and three unfolded proteins (Table I). Proteins were expressed in ¹⁵N-enriched medium and purified using the optimized Maxwell-16 protocol from a culture volume corresponding to a total OD₆₀₀ = 60. Purified proteins were concentrated to a final volume of 0.2 mL in NMR buffer (20 mM Na₂PO₄, pH 6.5, 50 mM NaCl) and evaluated by acquiring 1D ¹H and 2D ¹H-¹⁵N HSQC NMR spectra (Fig. 2). Each protein from the control group was expressed at levels (0.2–0.5 mg) sufficient to record an HSQC spectrum except for one (At2g20490) that failed to purify using the optimized Maxwell-16 protocol. Two proteins, ZNF24 and At1g16640, were purified in quantities sufficient to collect HSQC spectra but precipitated during exchange into the selected NMR buffer.

Table I.

Maxwell-16 Screening Results for 8 Control Proteins

Construct	Expression	Purification	HSQC
CoB	+++	+++	+
At3g17210	+++	+++	+
ZNF24	++	++	−
At5g22580	+++	+++	+
At1g16640	+++	+++	−
At3g29075	++	++	unfolded
At3g05570	++	+	unfolded
At2g20490	+	−	−

Figure 2.

Two-dimensional NMR analysis of control workgroup proteins. Samples from the control workgroup were subject to ¹H-¹⁵N HSQC NMR. Spectra were acquired in ∼80 min using 16 transients per FID at 25°C on a Bruker Avance 600 MHz NMR equipped with a 5 mm TCI Cryoprobe. Purity of samples generated by the Maxwell-16 is illustrated by SDS-PAGE (inset). Each gel contains samples of depleted lysate (DL), each wash step (W1–W4) and pure protein in the elution cuvette (EC) prior to exchange into the selected NMR buffer.

Rapid purification and screening using the Maxwell-16 permits empirical optimization of protein expression constructs. This is especially valuable when screening fragments of large proteins for which domain boundaries are defined imprecisely. To evaluate the utility of the Maxwell-16 in this application, we designed a domain workgroup consisting of a series of constructs that extracted individual or multiple PDZ domains from human Par-3 and the Dbl homology domain or Src homology 3 (SH3) domain from rat β-PIX. Once the expression constructs were in hand, ¹⁵N-labeled proteins were expressed and purified using the optimized Maxwell-16 protocol as summarized in Table II. Each construct was analyzed by 2D HSQC NMR where the β-PIX SH3 domain and the hPar-3 PDZ2 domain were judged to be folded, five constructs were unfolded, and 11 constructs contained no detectable signal (Fig. 3, Table II). SDS-PAGE analysis of the 18 constructs showed that three failed at the protein expression level, five additional proteins failed at the purification level and 10 yielded soluble protein suitable for NMR analysis. Of those 10 NMR samples, three yielded no detectable signal, consistent with aggregation or limited solubility.

Table II.

Maxwell-16 Screening of 18 Domain Constructs from hPar3 and β-PIX

Construct	Expression temperature	Expression	Purification	HSQC
hPar3 (1–135)	37°C	+++	++	N.S
hPar3 (136–270)	37°C	++	++	N.S
hPar3 (136–367)	37°C	+++	−	N.S
hPar3 (271–367)	37°C	++	−	N.S
hPar3 (271–549)	37°C	−	−	N.S
hPar3 (271–689)	37°C	+	−	N.S
hPar3 (451–549)	37°C	+++	+++	+
hPar3 (451–689)	37°C	+	+	−
hPar3 (584–689)	37°C	+++	+++	N.S
hPar3 (690–840)	37°C	multiple bands	multiple bands	−
hPar3 (739–958)	37°C	−	−	−
hPar3 (942–1025)	37°C	++	+	−
hPar3 (1026–1230)	37°C	+	−	N.S
hPar3 (1232–1353)	37°C	+	multiple bands	−
β-PIX DH (82–289) 8HT	37°C	+++	+	N.S
β-PIX DH (82–289) 8HT	15°C	++	++	N.S
β-PIX SH3 (10–63) 8HT	37°C	+++	+	N.S
β-PIX SH3 (10–63) 8HT	15°C	++	++	+

All eighteen constructs were subjected to ¹H-¹⁵N HSQC NMR. Spectra were graded as + for a folded protein, − for an unfolded or partially folded protein, or N.S. for a spectrum that did not show any signal. Gray shading indicates proteins that failed in expression or purification as determined by SDS-PAGE analysis.

Figure 3.

Two-dimensional NMR analysis of hPar-3 and β-PIX domain constructs. All 18 of the hPar-3 and β-PIX constructs were analyzed by ¹H-¹⁵N HSQC NMR. Spectra were acquired in ∼80 min using 16 transients per FID at 25°C on a Bruker Avance 600 MHz NMR equipped with a 5 mm TCI Cryoprobe. HSQC spectra for constructs that are folded, partially folded or unfolded are labeled in black and those that contain no signal are labeled in gray. The two proteins that are suitable for NMR structure determination are indicated with an asterisk.

Our results indicate that ¹⁵N-labeled protein automatically isolated from 60 mL of bacterial cell culture using a single channel of the Maxwell-16 is sufficient for evaluation by 2D NMR in a structural genomics environment. For targets proceeding to 3D structure determination, a 0.5 mL sample of ¹³C/¹⁵N-labeled protein at a concentration ≥ 0.5 mM is typically obtained by a large-scale (1–2 L) cell culture and manual purification process. However, because HSQC screening in a 3-mm NMR tube requires a sample volume of only 0.2 mL with little or no signal loss relative to a 5-mm tube, we speculated that ¹³C/¹⁵N protein production on the Maxwell-16 might completely eliminate the need for a large-scale production pipeline. Based on the NMR screening results for hPar-3 domains, we expressed a 1 L ¹⁵N/¹³C culture of the second PDZ domain (PDZ2) and purified it on the Maxwell-16. Yield and purity of protein from equivalent 60 mL aliquots was uniform across all 16 channels (Supporting Information Fig. 3), each of which produced ∼0.4 mg of protein. The overall yield of PDZ2 purified using the Maxwell-16 was similar to conventional batch IMAC purification (∼8 mg/L). A 0.2 mL NMR sample containing 1 mM PDZ2 pooled from five Maxwell-16 channels was subjected to our 3D NMR data collection and structure determination protocol²⁰ using a standard 5-mm cryogenic probe at 500 MHz [Fig. 4(a)]. While NMR data acquired on miniaturized samples using a microcoil probe suffers from reduced signal-to-noise ratios,²¹ spectra acquired on the 1 mM sample from robotic purification (0.2 mL in a 3 mm tube) were equal or superior to equivalent spectra acquired on the same instrument using 0.5 mL samples at the same protein concentration in a 5 mm tube [Fig. 4(b)]. We used automated methods to assign all ¹H, ¹⁵N, and ¹³C shifts and solve the structure [Fig. 4(c)] to high resolution (Table III).

Figure 4.

Validation of the optimized Maxwell-16 protocol. (a) An assigned ¹H-¹⁵N-HSQC of hPar3-PDZ2 shows 113 of 114 expected resonances. Data collection was done at 25°C in 3 mm sample cells with 4 transients per FID using a 1 mM sample of hPar3 PDZ2 in 20 mM sodium phosphate pH 6.5 containing 50 mM sodium chloride, 0.02% sodium azide, and 10% (v/v) ²H₂O. (b) Comparison of four ¹⁵N-edited 3D NOESY-HSQC strips from spectra collected on hPar3 PDZ2 (1 mM) prepared by a conventional large-scale production protocol (left) or the Maxwell-16 protocol (right). Data for the standard sample was acquired in a 5-mm Shigemi tube at 600 MHz, whereas data collection for the Maxwell-16 sample was acquired in a 3-mm NMR tube at 500 MHz with equal data collection times. Spectra of equivalent or higher quality were obtained from the Maxwell-16 sample despite a 2.5-fold reduction in the active NMR sample by mass and volume. Similar results were observed for the 3D triple-resonance experiments. (c) Ribbon diagram of the hPar3 PDZ2 NMR structure solved protein generated by five channels of the Maxwell-16. Structural statistics for the NMR ensemble are presented in Table III.

Table III.

Statistics for the 20 hPar3-PDZ2 Conformers

Experimental constraints
Distance constraints
Long	520
Medium [1 < (i − j) ≤ 5]	178
Sequential [(i − j) = 1]	296
Intraresidue [i = j]	351
Total	1318
Dihedral angle constraints (φ and ψ)	99
Average atomic R.M.S.D. to the mean structure (Å)
Residues 10-29, 40-98
Backbone (Cα, C′, N)	0.47 ± 0.07
Heavy atoms	0.96 ± 0.06
Deviations from idealized covalent geometry
Bond lengths RMSD (Å)	0.018
Torsion angle violations RMSD (°)	1.4
WHATCHECK quality indicators
Z-score	−1.36 ± 0.13
RMS Z-score
Bond lengths	0.86 ± 0.03
Bond angles	0.70 ± 0.03
Bumps	0 ± 0
Lennard-Jones energya (kJ mol−1)	−1850 ± 69
Constraint violations
NOE distance Number > 0.5 Å	0 ± 0
NOE distance RMSD (Å)	0.027 ± 0.001
Torsion angle violations Number > 5°	0 ± 0
Torsion angle violations RMSD (°)	0.508 ± 0.092
Ramachandran statistics (% of all residues)
Most favored	80.9 ± 2.9
Additionally allowed	15.2 ± 2.3
Generously allowed	1.4 ± 1.3
Disallowed	2.5 ± 0.9

^aNonbonded energy was calculated in XPLOR-NIH.

Small-scale robotic purification is a robust alternative to conventional protein production schemes that permits rapid expression screening of many different protein targets or large panels of related domain constructs.⁸ While it may not be necessary or advisable to acquire NMR data on every target (e.g., completely insoluble proteins or those that fail to express), the cost of ¹⁵N enrichment in a 50 mL culture is less than one dollar per target and the HSQC spectrum is the most critical step in validation for 3D structure determination. By linking expression testing with NMR screening in a single process, the most promising candidates can be identified more rapidly and at lower cost. Importantly, we showed that a small-scale robotic purification strategy could replace a large-scale structural genomics pipeline by solving the NMR structure of a PDZ domain prepared using the Maxwell-16. While an additional polishing step would be advisable in some instances, the optimized yield from a single Maxwell-16 cartridge (∼0.5 mg) is sufficient for nanoliter-scale crystallization screening and this approach could also be adapted to boost structure production by X-ray crystallography.

Materials and Methods

Expression of recombinant proteins

Expression vectors for CoB,¹⁷ At3g17210,²² At5g22580,¹⁵ At1g16640,¹⁹ and ZNF24¹⁸ were constructed as previously described. Expression vectors for At3g29075 and At3g05570 were cloned into pET15b by the CESG as described for At3g17210 and At5g22580.^15,22 At2g20490 was cloned into a modified pQE30 vector as described for CoB.¹⁷

The identification of domain boundaries in β-PIX and hPar-3 was facilitated by sequence alignment of homologs and previously published works. Selected domains were PCR amplified from full-length clones using a 5′ primer containing a BamHI restriction site and a 3′ primer containing a HindIII. PCR products were cut with BamHI and HindIII, gel purified and ligated into a pQE30 vector (Qiagen) modified to contain a (His)₈ metal affinity tag and a tobacco etch virus (TEV) protease site for removal of the affinity tag.¹⁹ All expression vectors were verified by DNA sequencing. A full-length rat β-PIX clone was kindly provided by Dr. Andrey Sorokin (Medical College of Wisconsin, Milwaukee, WI), and the full-length human Par-3 was kindly provided by Dr. Ian Macara (University of Virginia, Charlottesville, VA).

Expression

The pQE based and pET based expression constructs were transformed into E. coli strain SG13009[pRPEP4] (Qiagen) and BL21(DE3), respectively. Cell were grown in shake flasks using LB media containing the appropriate antibiotics until the cell density reached OD₆₀₀= 0.7. Protein expression was induced by the addition of isopropyl-β-d-thiogalactopyranoside to a final concentration of 1 mM. Final OD₆₀₀ values ranged from 0.8–1.8. Following induction, incubation of the cells was continued for 5 h at 37°C or for 20 h at 15°C. Cells were harvested by centrifugation at 5000g for 10 min. Cell pellets were stored at −80°C. For uniform ¹⁵N labeling, cells were grown in M9 media containing ¹⁵N-ammonium chloride as the sole nitrogen source. Uniform ¹⁵N/¹³C labeling was accomplished by providing ¹³C-glucose as the sole carbon source concurrent with ¹⁵N-ammonium chloride.

Optimization of the Maxwell-16 protocol

The Maxwell-16 utilizes a seven-well cartridge and a separate elution cuvette to complete the automated purification of His-tagged proteins using MagneHis resin (Promega). Initial purification studies utilized a cell pellet from 50–60 mL of cell culture resuspended in 950 μL of resuspension buffer (50 mM sodium phosphate pH 7.4, 300 mM sodium chloride, 10 mM imidazole, 0.1% (v/v) 2-mercaptoethanol, 1 mM phenylmethylsufonyl fluoride, and 50 μg of DNase) and 110 μL of FastBreak cell lysis solution (Promega). The cell suspension was placed in the first well of the Maxwell-16 cartridge. Well 2–6 contained 1 mL of wash buffer (50 mM sodium phosphate pH 7.4, 300 mM sodium chloride, and 10 mM imidazole), with 30 μL of MagneHis resin added to well 2. The elution cuvette contained 1 mL of elution buffer (50 mM sodium phosphate pH 7.4, 300 mM sodium chloride, 500 mM imidazole, and 0.02% sodium azide) and well 7 contained the plunger. The automated protein purification protocol was selected from the Maxwell-16 menu and the results were evaluated by SDS-PAGE. To improve the yield of purified protein, the volume of MagneHis resin and the volume and composition of the wash and elution buffers were evaluated using ZNF24, At1g17210, and CoB as test cases. These optimizations resulted in two modifications to our initial protocol. First, the highest yields of purified protein were obtained using 150 μL of MagneHis resin. Second, the pH of the elution buffer was lowered from 7.4 to 6.5. All the subsequent purifications of isotopically labeled proteins for NMR were done using 50–60 mL of cell culture and the modified Maxwell-16 protocol. Yields of pure protein from a single cartridge determined by BCA assay or absorbance at 280 nm ranged from 0.2–0.5 mg, corresponding to 4–10 mg/L. Upon completion of the Maxwell-16 protocol, purified proteins were buffer exchanged into 20 mM sodium phosphate pH 6.5 containing 50 mM sodium chloride, 0.02% sodium azide, and 10% (v/v) ²H₂O and concentrated to a final volume of 200 μL.

NMR spectroscopy

1D ¹H and 2D ¹H-¹⁵N HSQC NMR spectra were acquired at 25°C in 3 mm sample cells on a Bruker 600 MHz spectrometer equipped with a 5 mm triple-resonance CryoProbe™. All HSQCs were collected with 16 transients per FID and processed with NMRPipe software.²³

Structure determination

All 3D NMR data was acquired on a Bruker 500 MHz spectrometer equipped with a 5 mm triple-resonance CryoProbe™. The uniformly ¹⁵N/¹³C-hPar3 PDZ2 sample was prepared at a concentration of 1 mM in 20 mM sodium phosphate pH 6.5 containing 50 mM sodium chloride, 0.02% sodium azide and 10% (v/v) ²H₂O and placed in a 3 mm sample cell. Backbone resonance assignments were verified from our previous structure determination using 3D HNCO, HNCA, and HNCACB spectra. Sidechain assignments were verified using HCCH-TOCSY data and a ¹³C-edited NOESY-HSQC spectrum optimized for aromatic groups. Distance constraints were obtained from 3D ¹⁵N-edited NOESY-HSQC and ¹³C-edited NOESY-HSQC spectra (τ_mix = 80 ms). Backbone ψ and φ dihedral angle constraints were generated from secondary shifts of the ¹H^α, ¹³C^α, ¹³C^β, ¹³C′, and ¹⁵N nuclei shifts by the program TALOS.²⁴ Initial structures were generated using the NOEASSIGN module of the torsion angle dynamics program CYANA,^25,26 followed by iterative manual refinement to eliminate consistently violated restraints. Of the final 100 structures calculated, the 20 conformers with the lowest target function values were selected and subjected to a molecular-dynamics protocol in explicit solvent using XPLOR-NIH.²⁷ All time-domain NMR data and chemical shift assignments have been deposited in BioMagResBank (accession code: 16520) and the Protein Data Bank (accession code: 2kom).