Crystallization and initial X-ray crystallographic analysis of a de novo-designed protein with left-handed βαβ units

A newly designed protein featuring a rare left-handed βαβ motif has successfully been crystallized and characterized by preliminary X-ray diffraction. The computational design was conducted using a combination of Rosetta BluePrintBDR, ProteinMPNN and AlphaFold2, generating eight candidates based on predicted stability and folding accuracy. The final construct was expressed, purified and crystallized in space group P2₁. Complete X-ray diffraction data were collected on the BL2S1 beamline at the Aichi Synchrotron and processed to 1.95 Å resolution. Despite multiple attempts, molecular replacement using the AlphaFold2 model did not yield a conclusive solution, suggesting that alternative phasing methods or refined modeling approaches will be needed. This work highlights both the promise and the challenges of pushing protein biodesign into underexplored structural motifs and provides a foundation for future structural and functional investigations.

1. Introduction

The design and synthesis of de novo proteins have emerged as a powerful approach to elucidating protein structure and function (Huang et al., 2016). Such de novo-designed proteins enable the exploration of novel folds and functionalities that are not found in natural protein repertoires (Koga et al., 2012). This study reports the crystallization and initial X-ray crystallographic analysis of a newly designed protein featuring a rare left-handed βαβ motif.

The left-handed βαβ motif is only rarely observed in the Protein Data Bank (PDB), making it an intriguing target for de novo design (Murata et al., 2021). In principle, βαβ sub­structures can adopt either a right-handed or left-handed twist; however, surveys of the PDB reveal that the left-handed variant is almost entirely prohibited (Richardson, 1976). Because βαβ motifs occur so frequently in protein structures, appearing in more than half of all known domains, this absence severely restricts the diversity of naturally observed α/β folds (Finkelstein & Ptitsyn, 1987). Incorporating a left-handed βαβ motif could therefore unlock new structural possibilities that cannot be accessed by the right-handed motif alone, making it a key frontier for expanding the repertoire of functional protein designs.

Recent advances in computational protein design, including AlphaFold2, have significantly enhanced our ability to predict and model novel protein structures with high accuracy (Jumper et al., 2021). These breakthroughs were further recognized in 2024 with the award of the Nobel Prize in Chemistry to David Baker, Demis Hassabis and John Jumper for their pioneering work in computational protein design and protein structure prediction, a testament to the transformative potential of engineering novel protein architectures.

In this work, we employed AlphaFold2 alongside other computational tools to design a protein with a rare left-handed βαβ fold. The protein was purified, crystallized and subjected to X-ray diffraction experiments to characterize its structure and verify this unusual motif. While molecular replacement did not yield a solution, the crystallographic data presented here offer promising insights for future structural determination.

Our findings highlight the difficulties of solving protein motifs that are not routinely encountered in nature, whether mirror-image proteins or unusual structural motifs such as the left-handed βαβ fold. Further refinement of structure-determination methods and design strategies will be critical to overcome these challenges. By exploring unique secondary structures, we aim to expand the protein-design toolkit and uncover new possibilities in protein engineering.

2. Materials and methods

2.1. Protein-sequence design and model generation

The de novo protein featuring a left-handed βαβ motif was designed using a multifaceted computational pipeline. Firstly, Rosetta BluePrintBDR was employed with explicit constraints to favor backbone dihedral angles consistent with a left-handed βαβ geometry, ensuring that potential scaffolds would adopt this unusually twisted conformation (Fleishman et al., 2011). These backbones were then refined using ProteinMPNN, which optimized the side-chain composition to favor stable folding and solubility (Dauparas et al., 2022). Finally, AlphaFold2 assessed the likelihood of each candidate adopting the intended fold, with pLDDT (predicted local distance difference test) and pTM (predicted template modeling) scores guiding the selection of the most promising designs (Jumper et al., 2021). A total of eight sequences emerged from this workflow, each showing high predicted accuracy. Throughout iterative rounds of refinement, we closely monitored charge distribution, potential steric clashes and secondary-structure geometry, all to preserve a robust left-handed βαβ substructure. The top-performing construct, which scored highest on both the pLDDT and pTM metrics, was chosen for experimental validation (Takei & Ishida, 2022).

2.2. Molecular cloning, protein expression and purification

The gene encoding the designed protein (9.13 kDa) was synthesized (GenScript) and cloned into the pET-24a(+) vector (Novagen), adding a C-terminal His-tag. The recombinant vector was transformed into Escherichia coli T7 Express cells (New England Biolabs) grown in Luria–Bertani (LB) medium at 37°C. Protein expression was induced at an optical density (600 nm) of 0.6–0.8 by the addition of 0.5 mM isopropyl β-D-1-thiogalactopyranoside (Sigma–Aldrich) and was maintained for 4–6 h.

After harvesting cells by centrifugation at 8000g for 20 min at 4°C, the pellet was resuspended in phosphate-buffered saline (PBS; Thermo Fisher Scientific). Following sonication, the soluble fraction was clarified by centrifugation at 20 000g for 20 min (repeated twice) at 4°C. Ni–NTA affinity chromatography (Qiagen) was used to isolate the target protein, which was subsequently concentrated using a Vivaspin 3K ultrafiltration device (Cytiva). The His-tag remained uncleaved, as preliminary tests showed no detrimental effect on stability. The protein, at a concentration of ∼120 mg ml⁻¹, was stored at 4°C short-term and frozen at −80°C long-term without loss of solubility.

2.3. Crystallization

Initial crystallization trials used the sitting-drop vapor-diffusion method in 96-well plates (Molecular Dimensions). A 1:1 mixture of the protein solution (∼120 mg ml⁻¹) and the reservoir solution was screened at 20°C using Crystal Screen from Hampton Research. Small crystals appeared after a few days under several PEG-containing conditions. Subsequent manual optimization yielded a reproducible condition consisting of 100 mM NaCl, 20%(w/v) PEG 5000 pH 7.0. Within a week, the crystals reached dimensions of up to 300 µm in their longest axis (Fig. 1).

Figure 1 Optical micrograph of the de novo-designed protein crystals grown by sitting-drop vapor diffusion at 20°C. The crystals reached dimensions of up to 300 µm within a week. The octahedral shapes visible here reflect the reproducible crystallization conditions, which provided samples suitable for X-ray diffraction experiments.

Crystals were harvested using LithoLoops (Molecular Dimensions) and briefly soaked in 20%(v/v) glycerol as a cryoprotectant before flash-cooling in liquid nitrogen. This protocol minimized ice formation and crystal cracking, resulting in improved diffraction at the synchrotron.

2.4. Synchrotron data collection and analysis

X-ray data were collected on the BL2S1 beamline at the Aichi Synchrotron using a PILATUS 1M detector (Watanabe et al., 2017). An initial assessment showed diffraction to about 1.8 Å resolution. However, full data acquisition revealed significant radiation damage, reducing the effective resolution to 1.95 Å (Fig. 2). Data were collected at a wavelength of 0.7233 Å, employing a 0.1° oscillation with a 1.0 s exposure per image to cover 360° of total rotation. The total absorbed dose was estimated to be approximately 0.34 MGy, about one-tenth of the 3.0 MGy Garman dose limit, suggesting that the observed crystal decay may reflect an unusual sensitivity rather than purely dose-related damage (Zeldin et al., 2013).

Figure 2 A representative diffraction image of the de novo-designed protein crystal collected on the BL2S1 beamline at the Aichi Synchrotron. Clear diffraction spots are visible at lower resolutions, but they diminish in intensity toward the higher resolution range. The relatively high mosaicity of these crystals is evident in the broader, less distinct diffraction spots. This image captures one still from the overall fine-sliced data collection, highlighting both the diffraction quality and the challenges posed by mosaicity.

The data were indexed, integrated and scaled using the xia2 pipeline and related software from the CCP4 suite (Agirre et al., 2023). The crystals belonged to space group P2₁, with unit-cell parameters a = 34.28, b = 53.41, c = 37.63 Å, β = 113.10°. Analysis suggested the presence of one molecule in the asymmetric unit, consistent with the molecular weight of the protein. Attempts to solve the structure by molecular replacement using the AlphaFold2 model in multiple pipelines provided by CCP4 Cloud (Krissinel et al., 2022), including MOLREP (Vagin & Teplyakov, 2010) and Phaser (McCoy et al., 2007), were inconclusive. For the generation of the model coordinates, the B factors were recalculated by MOLREP, assuming the AlphaFold model. Additional optimizations and further model refinements may be required as no solution was obtained, or alternative phasing methods should be considered. Data-quality and refinement statistics are given in Table 1.

Table 1
Data-quality and refinement statistics

Values in parentheses are for the outer shell.

Beamline	BL2S1, Aichi Synchrotron
Detector	PILATUS 1M
Wavelength (Å)	0.72333
Temperature (K)	120.0
Crystal-to-detector distance (mm)	200.0
Oscillation per image (°)	0.1
Oscillation range (°)	360.0
Mosaicity (°)	0.61
Space group	P21
a, b, c (Å)	34.28, 53.41, 37.63
α, β, γ (°)	90, 113.10, 90
Resolution range (Å)	29.05–1.95 (1.99–1.95)
Total No. of reflections	59960 (3297)
No. of unique reflections	8930 (476)
Rmerge	0.1576 (0.7670)
CC1/2	0.9950 (0.8770)
Completeness (%)	97.20 (97.90)
〈I/σ(I)〉	6.70 (1.20)
Multiplicity	6.70 (6.90)
Wilson B factor (Å^2)	18.79

3. Results

3.1. Model generation

Of the eight de novo sequences generated by our computational pipeline, the one demonstrating the highest AlphaFold2 confidence scores (pLDDT and pTM) was selected for experimental testing (Fig. 3). In silico predictions indicated that this protein would adopt a well packed core consistent with a left-handed βαβ conformation, featuring extensive side-chain networks predicted to stabilize the backbone twist. Notably, no major steric clashes or destabilizing motifs were identified, suggesting that the designed sequence was capable of folding and remaining soluble under typical laboratory conditions.

Figure 3 Schematic representation of the de novo design workflow and the final AlphaFold2-predicted structure. (a) A flowchart outlines the sequential use of Rosetta BluePrintBDR, ProteinMPNN and AlphaFold2 to generate candidate sequences featuring a left-handed βαβ motif. (b) The top-scoring AlphaFold2 model is shown in two parts. In the upper ribbon diagram, the protein is color-coded by pLDDT (orange → yellow → light blue → dark blue) to indicate increasing local confidence; the average pLDDT value is 97.0 and the predicted TM-score (pTM) is 0.84, reflecting high overall confidence. Lighter, warmer hues mark lower-confidence regions that may deviate from the in silico model in the crystal structure. In the lower ribbon diagram, the left-handed βαβ motif is highlighted in orange, emphasizing its location and potential structural significance.

3.2. Purification, crystallization and diffraction

Upon expression in E. coli, the protein appeared as a single major band on SDS–PAGE, and its high solubility permitted its concentration to ∼120 mg ml⁻¹ without precipitation. In contrast to many de novo designs that can suffer from aggregation or instability, the present construct remained stable throughout purification and storage. Crystals reproducibly formed within a week under optimized conditions (100 mM NaCl, 20% PEG 5000 pH 7.0) and reached typical dimensions of ∼300 µm.

Data collection at the Aichi Synchrotron revealed diffraction spots extending to approximately 1.8 Å resolution. However, noticeable radiation-induced decay caused the final data set to be truncated to 1.95 Å resolution to give a reliable signal-to-noise ratio. This finding underscores the X-ray sensitivity of the crystals and suggests that multi-crystal or shorter-exposure strategies might be essential for future structure-determination efforts.

3.3. Data analysis and molecular replacement

Detailed data processing in DIALS confirmed the space group to be P2₁ with one molecule per asymmetric unit. Although the high-resolution shell revealed some reflections approaching 1.8 Å resolution, consistent beam damage required the final data set to be limited to 1.95 Å resolution to give a reliable signal-to-noise ratio. During data analysis, attempts were also made to integrate less than 360° of rotation, but no tangible improvements in data quality were observed (data not shown). Molecular-replacement attempts using the AlphaFold2-predicted model in multiple CCP4 Cloud pipelines (including MOLREP and Phaser) did not yield a definitive solution, pointing to possible deviations between the in silico model and the actual fold or limitations arising from radiation damage. Despite recalculating B factors for the model and performing additional optimizations of the AlphaFold2-generated model, no solution was obtained, suggesting the need for alternative phasing methods or further model refinements. Notably, no suitable homologous model could be identified in the PDB for this left-handed βαβ motif, leaving the AlphaFold2-based construct as our only feasible search template; attempts to use naturally occurring right-handed βαβ structures as proxies were similarly unsuccessful.

Interestingly, a recent large-scale study (Keegan et al., 2024) reported that only 3% of approximately 400 structures solved by SAD failed to yield a solution by molecular replacement when using an AlphaFold-based model. Our inability to obtain a clear solution therefore underscores the unusual difficulties posed by this system. Given the rarity of the left-handed βαβ motif in current structure databases, conventional MR search procedures may struggle to accommodate such unique structural features. Alternative phasing strategies, such as selenomethionine labeling or heavy-atom soaking, will be investigated in future work to address these challenges.

4. Conclusion

In summary, we have successfully designed, purified and crystallized a de novo protein predicted to feature a rare left-handed βαβ motif. Although X-ray data were collected to an effective resolution of 1.95 Å in space group P2₁, molecular replacement using the AlphaFold2 model was not conclusive. Future efforts will focus on mitigating radiation damage during data collection and pursuing alternative phasing methods (for example SeMet labeling). Resolving this unusual fold will be a key advance in protein bio-design, broadening our understanding of engineering novel structural motifs and opening doors to innovative applications in protein science.

Footnotes

‡These authors contributed equally to this work.