De novo design of protein binders targeting the human sweet taste receptor as potential sweet proteins

Abstract: Excessive consumption of dietary sugars is a major contributor to metabolic disorders, driving global interest in finding alternative sweeteners with reduced caloric impact. Natural sweet proteins, such as brazzein, offer exceptional sweetness intensity with little caloric contribution. However, their widespread use is limited by restricted natural diversity, low stability, and high production costs. Recent advances in structural biology and de novo protein design provide new opportunities to overcome these limitations through rational engineering. In this study, we report an integrated computational pipeline for the de novo design of protein binders targeting the human sweet taste receptor subunit TAS1R2, a key component of the heterodimeric class C G protein-coupled receptor mediating sweetness perception. The workflow combines diffusion-based backbone generation (RFdiffusion), neural network-guided sequence design (ProteinMPNN), structure-based filtering using Boltz-1, and binding energy evaluation via MM/GBSA calculations. Using the recently resolved cryo-EM structure of the TAS1R2 receptor, protein binders were designed to target both the Venus Flytrap Domain and the cysteine-rich domain of TAS1R2. A few designed binders exhibited favorable structural confidence and predicted binding energetics. In particular, Binder2 exhibited brazzein-like structural plausibility through specific short-range CRD contacts, while Binder1 displayed the strongest predicted binding affinity. Structural analyses of the binder-receptor complex revealed distinct binding modes and secondary structure profiles among the designs. This study demonstrates the feasibility of de novo designing protein binders that emulate key functional properties of natural sweet proteins, establishing a computational framework for the rational development of next-generation protein-based sweeteners.

• The paper introduces a novel computational pipeline that integrates diffusion models and neural network-guided sequence design to engineer de novo protein binders for TAS1R2.
• The study reports that designed candidates, notably Binder1 and Binder2, achieve competitive binding energies and structural metrics compared to natural sweet proteins like brazzein.
• The paper highlights the potential for de novo design strategies to yield tunable, stable, and scalable sweet protein candidates addressing metabolic and sensory challenges.

De Novo Design of Protein Binders Targeting TAS1R2 as Potential Sweet Proteins

Background and Motivations

The growing prevalence of metabolic disorders linked to excessive sugar consumption has intensified the search for alternative sweeteners that can deliver high perceived sweetness with minimal caloric impact. While traditional synthetic sweeteners effectively reduce caloric content, their adoption is hampered by adverse sensory properties and potential safety concerns. Natural sweet proteins such as brazzein present an attractive alternative, combining high sweetness intensity with favorable nutritional profiles. However, their industrial adoption is constrained by limited stability, low diversity, and challenging production economics. Recent advances in high-resolution receptor structure determination and generative protein design methodologies have opened new avenues for the rational engineering of protein-based sweeteners with tailor-made properties.

Computational Pipeline for De Novo Protein Binder Design

The study describes an integrated computational strategy to design de novo protein binders targeting the human sweet taste receptor subunit TAS1R2—a class C GPCR critical for mediating human sweetness perception. The pipeline comprises the following sequential steps:

1. Structural Template Preparation: Utilization of the recent cryo-EM structure of the TAS1R2-TAS1R3 heterodimer and AlphaFold2-predicted models provided high-fidelity templates for downstream design.
2. Backbone Generation (RFdiffusion): Employing RFdiffusion, a diffusion-based generative model, the authors generated a large library of 128-residue backbone scaffolds conditioned to interact with either the Venus Flytrap Domain (VFTD) or Cysteine-Rich Domain (CRD) of TAS1R2.
3. Sequence Design (ProteinMPNN): For each backbone, ProteinMPNN optimized amino acid sequences to fit the geometric and physiochemical constraints of the scaffold, followed by iterative structural relaxation (PyRosetta FastRelax).
4. Structural Evaluation (Boltz-1): Boltz-1, an AlphaFold3-based predictor, assessed and ranked complex models using confidence metrics (ipTM, pLDDT, confidence_score) to filter out suboptimal designs.
5. Binding Energy Prediction (Uni-GBSA): The top candidate complexes were subjected to MM/GBSA binding free energy calculations, offering quantitative affinity predictions to prioritize final designs.

Structural and Functional Outcomes

Boltz-1 evaluation revealed that several de novo designed binders achieved structural confidence and interface metrics comparable to, or approaching, those of the natural sweet protein brazzein. In particular:

• Binder2: Demonstrated ipTM and confidence scores on par with brazzein, suggesting a plausible and stable binding orientation. Structurally, Binder2 formed specific CRD contacts (ARG37/GLY509, THR55/ALA228), underpinning its receptor engagement profile. The architecture of Binder2 was strongly helix-dominated (92% helix), indicative of high-order structural design.
• Binder1: Exhibited the most favorable predicted binding free energy (-115.6 kcal/mol), surpassing brazzein (-70.7 kcal/mol) and the other designed candidates. However, structural plausibility metrics lagged those of Binder2 and brazzein, highlighting the trade-off between binding energetics and modeling confidence.

Secondary structure composition across designs was dominated by α-helices, reflecting biases in generative protocols toward helical scaffolds; only a minority of designs featured significant β-strand content.

Comparative Analysis and Implications

The results position Binder2 as the highest-confidence de novo TAS1R2 binder emulating brazzein’s structural and interface features, while Binder1 claims an exceptionally strong predicted binding affinity, albeit with reduced structural validation. This juxtaposition highlights an inherent tension in computational protein design: maximizing interface energetics can come at the expense of foldability and model confidence, necessitating multidimensional optimization in future frameworks.

By targeting both the VFTD and CRD, the design strategy mirrors known binding modes of natural sweet proteins but extends the diversity of possible binder interaction profiles. The framework establishes a versatile blueprint for the rational engineering of high-potency, low-calorie sweet proteins that can be tuned for stability, solubility, and production in heterologous hosts.

Theoretical and Practical Implications

From a theoretical standpoint, this work exemplifies the maturation of generative design protocols (diffusion models, neural network-guided sequence search) as practical tools for engineering functional proteins that interact with membrane protein complexes of significant physiological and industrial relevance. The method integrates orthogonal advances in structural biology, generative modeling, and binding free energy estimation, providing a foundation for iterative improvement.

Practically, successful realization of de novo sweet protein binders could overcome supply chain, production, and intellectual property limitations faced by current natural sweeteners. Moreover, the systematic approach described is extensible to other chemoreceptors and taste modalities, facilitating broader application in food technology and beyond.

Future Directions

Key areas for further research include:

• Experimental Validation: Synthesis and in vitro functional assays of Binder1 and Binder2 to confirm predicted binding and sweetness perception.
• Stability and Scalability: Assessment of thermal, pH, and proteolytic stability, as well as large-scale production feasibility in suitable expression systems.
• Expanding Target Space: Application of the design paradigm to additional domains of TAS1R2 or multivalent targeting of the TAS1R2-TAS1R3 heterodimer for increased potency and selectivity.
• Integration with Sensory Neuroscience: Deeper exploration of the downstream signaling and central processing effects initiated by novel protein binders to fine-tune sweetness profiles and minimize off-target effects.

Conclusion

The study systematically demonstrates the feasibility of de novo computational design of protein binders targeting the human sweet taste receptor, leveraging a multi-tiered generative and screening pipeline. By bridging advances in protein structure prediction, sequence design, and energetic evaluation, the framework delivers lead candidates with favorable predicted affinity and structural credence relative to the natural sweet protein brazzein. These findings lay the groundwork for developing next-generation, protein-based sweeteners tailored for desirable physicochemical properties, scalable production, and robust sensory profiles, with broader implications for protein engineering and synthetic biology (2601.14574).