Design
At the outset of our project, we faced a major challenge: the absence of experimental structural data for the C-terminus of progerin. This region is intrinsically disordered, making it inaccessible to crystallography and only marginally informative through cryo-EM. To overcome this limitation, we employed AlphaFold3 to predict the structure of progerin, using these models as an initial indication of the possible conformations of the C-terminal region and as a starting point for our computational strategy.
Build
To address this challenge, we established a computational pipeline that integrates different levels of modeling and analysis. Our approach combines AI-based structure prediction and design with molecular docking and affinity estimation to enhance the design process. Specifically, AlphaFold3, RFdiffusion, and ProteinMPNN were used in succession for structural modeling, binder backbone generation, and sequence design, while HADDOCK and ClusPro were applied for docking simulations, followed by PRODIGY for the estimation of binding affinities. The pipeline was conceived to move from structural modeling and refinement of the target, through the design of potential binders, and finally to the evaluation of their interactions, providing an integrated framework for iterative improvement.
Test
We applied this pipeline by generating full-length structural predictions of progerin using AlphaFold3 and analyzing the outputs with Predicted Alignment Error (PAE) plots. Most of the protein was modeled with high confidence, with well-defined regions showing strong cross-correlation. In contrast, the C-terminal region displayed very low confidence, consistent with its intrinsically disordered nature. This confirmed that AlphaFold was unable to assign a stable fold to this segment, leaving a crucial gap in the model and highlighting the need for a refined approach.
Learn
At this point, we sought expert advice to interpret our results and improve our strategy. Professor Roberto Steiner guided us through the analysis of the PAE plots, consulting with a researcher from DeepMind. Based on their input, we shifted our focus to the isolated C-terminal region, generating AlphaFold3 predictions of the fragment at different lengths. This modification improved the confidence values, which, although still moderate, produced more coherent conformations. However, because AlphaFold is context-dependent, these structures should not be viewed as definitive but rather as snapshots of possible folding states. Interpreting them as a library of candidate conformations enabled us to refine our computational pipeline and expand the set of potential interactors to be designed against progerin.
Design
Building upon the limitations discovered in the first cycle, we refined our design approach.
In this cycle, we decided to focus on a series of C-terminal fragments. By narrowing the scope of the target region, we aimed to improve the reliability of the predicted structures.
Build
We updated our pipeline to focus on the C-terminal fragments of progerin, as well as exploring AlphaFold3’s capability to predict shorter fragments. For the structural modeling, we still employed AlphaFold3 predictions at varying fragment lengths. This fragment-focused design approach enabled us to improve the precision of predicted structures.
Test
Structural outputs from AlphaFold3 showed improved PAE plots with greater coherence in the folded conformations. While the confidence in the C-terminal region was still moderate, we observed clearer patterns and more stable structures compared to the first cycle.
Learn
Narrowing our focus to the C-terminal fragment allowed us to obtain more reliable structural predictions and enabled us to design better-targeted interactors. Additionally, the iterative process of generating multiple AlphaFold3 predictions for fragments of different lengths provided us with a library of candidate conformations. This not only improved our understanding of the C-terminal flexibility but also highlighted the need for contextual interpretation of the predicted models, treating them as snapshots rather than definitive structures.
Design
In the first stage of our structural prediction workflow for Progerin, we considered two different tools: AlphaFold and SWISS-MODEL. Our goal was to compare the structural outputs obtained from both approaches. These platforms rely on fundamentally different strategies: while SWISS-MODEL generates structures based on available templates in protein databases (homology modeling), AlphaFold predicts protein conformations de novo using deep learning and evolutionary information.
Build
We performed structural predictions for both Progerin and Lamin A using SWISS-MODEL and AlphaFold.
Test
Model validation was performed through MolProbity and by examining internal quality parameters (e.g., pLDDT and pTM for AlphaFold, GMQE and QMEAN for SWISS-MODEL).
SWISS-MODEL predictions exhibited low confidence scores and poor model coverage, likely due to the absence of suitable structural templates for Progerin in existing databases.
Conversely, AlphaFold produced high-confidence structures, capturing the overall domain organization and maintaining consistent secondary structure elements, making them more reliable for further applications such as docking and interactor design.
Learn
Given the higher reliability of AlphaFold predictions, we selected AlphaFold as our primary tool for structural prediction.
Design
Having selected AlphaFold as our primary tool for structural prediction due to its higher reliability, we adjusted the pipeline accordingly.
Build
We performed structural prediction of full-length progerin and its C-terminal fragments with AlphaFold.
Test
Model validation was carried out using MolProbity to check for structural quality, including the verification of stereochemistry, Ramachandran plot analysis, and potential clashes.
Learn
Our analysis confirmed that the AlphaFold-predicted structures were better in terms of structural quality, and that we could proceed with further experimental work.
Design
One of our first challenges was to capture the three-dimensional structure of progerin, the mutant protein responsible for Hutchinson–Gilford Progeria Syndrome. A key difference from lamin A is that progerin permanently retains its farnesyl group, a small but crucial post-translational modification that anchors the protein to the nuclear membrane.
We realized that any attempt to design a specific interactor would be incomplete without accounting for this feature. For this reason, our design strategy started by incorporating the farnesyl group into the predicted structures of progerin, making the model as close as possible to its biological form.
Build
After generating the initial structural model, we moved into “molecular carpentry.” Using ChimeraX, we carefully attached the farnesyl group—sourced from PubChem (entry 444108)—to the protein. This manual step was more than cosmetic: it was the moment where our predicted model began to resemble the biologically relevant form of progerin.
Test
We then processed our model using the bioinformatic tools integrated into our pipeline for protein relaxation. To our surprise, the system stalled: post-translational modifications (PTMs) were a blind spot for these otherwise powerful algorithms. The output was unusable, highlighting an important limitation of current computational pipelines. In other words: we had built something too realistic for the software to handle.
Learn
This obstacle became an opportunity. Instead of abandoning the PTM, we stepped back and asked ourselves: what is the real effect of the farnesyl group on progerin?
The answer: it acts as a molecular anchor, tethering the farnesyl–cysteine to the inner nuclear membrane and reducing the local flexibility of the protein. With this insight, we adapted our approach. By simulating the association between the farnesyl group and the nuclear membrane through semi-flexible docking with restraints, we struck a balance: the C-terminus remained fully flexible, except for the rigidly attached farnesyl–cysteine. Moreover, we excluded the farnesyl–cysteine from the RFdiffusion hotspot definition used for designing the interactors. This prevented the algorithm from incorrectly interpreting the chemical environment around the absent farnesyl group, thereby avoiding computational artifacts in the design of the interactors.
This adjustment transformed a computational limitation into a refined, biologically grounded model, paving the way for more reliable interactor design.
Design
Building on the insights from the first cycle, we redefined our structural approach. Instead of explicitly modeling the farnesyl group, we decided to represent its effect on local rigidity through spatial restraints and reduced conformational freedom at the C-terminal cysteine. Our goal was to create a model that realistically mimics the farnesyl-anchored configuration of progerin, while remaining compatible with computational tools such as AlphaFold3, RFdiffusion, and HADDOCK. We also revised the interactor design pipeline to ensure that generated peptides targeted the accessible portion of the C-terminal region, rather than the membrane-facing farnesylated site.
Build
We implemented this design by introducing positional restraints in the molecular docking stage to mimic membrane tethering. The C-terminal cysteine was treated as a pseudo-anchored residue, while the remaining amino acids were free to relax naturally. At the same time, we updated the RFdiffusion and ProteinMPNN pipelines by explicitly excluding the terminal residues involved in membrane association from the diffusion hotspot definition. This modification guided the AI models to focus binding toward the biologically accessible surfaces of progerin.
Test
We re-ran docking under these revised conditions. The resulting structures displayed consistent folding of the C-terminal tail, while maintaining compatibility with all downstream validation steps.
Learn
Through this second cycle, we learned how to translate biological realism into computationally tractable constraints. By abstracting the physical effect of the farnesyl anchor instead of directly modeling it, we overcame a key bottleneck in AI-driven protein design workflows. This approach allowed us to preserve the biological relevance of progerin’s membrane association while unlocking full compatibility with modern prediction and docking platforms.
Design
In defining our pipeline, we anticipated the need to relax the structure after prediction, but we were uncertain about how to outline this necessary step. During the design phase, we decided to employ the two methods most accessible to us: Amber from the NeuroSnap package and Rosetta Relax from ROSIE. However, we were unsure how to use them and in which order.
Build
We performed structural relaxations on the entire progerin structure and some fragments using Amber and RosettaRelax.
Test
We validated the predicted structures using MolProbity, a widely adopted tool for structural quality assessment. We summarized some of our analyses and compared different relaxation methods (Comparison of different relaxation methods). To evaluate the effect of relaxation, we first applied each tool independently and then tested combinations of both methods in sequence. This approach allowed us to identify not only the improvements introduced by each tool but also the optimal order of application for achieving the most reliable structural refinement.
Learn
At the end of our analyses, we identified the most effective relaxation strategy to achieve both optimal energy minimization and a stereochemically sound structure. The best approach consisted of three consecutive steps: Amber → RosettaRelax → Amber. We compared the resulting models by evaluating their MolProbity scores and energy values.
Design
In the previous cycle, we established a relaxation pipeline to obtain an optimized structure for our goal. We decided to continue using the relaxation sequence Amber → RosettaRelax → Amber.
Build
We applied the relaxation procedure to all the structures of Progerin and its fragments.
Test
We performed a new cycle of validation for the relaxed structures, again using MolProbity to check the quality of the predicted structures. During this phase, we focused on analyzing the improvements introduced by each relaxation step and the combination of methods to ensure there were no artifacts or distortions. In this phase, we also analyzed the energy progress in relation to structural stability.
Learn
We found that the relaxation sequence chosen in the previous cycle continues to work effectively. Our analysis confirmed that our Amber → RosettaRelax → Amber approach produces the best structures, which have been selected for subsequent docking studies and interactor design. Despite the good performance, we are considering the implementation of additional relaxation cycles, such as multiple iterations with Amber, to explore potential further improvements.
Design
In this stage, we aimed to assess how the fusion of our interactors with the RING domain affected their structure and binding performance.
To ensure consistency with previous analyses, we applied the same computational pipeline and docking methodology established in earlier cycles, this time integrating the RING domain into the constructs.
This step was essential to understand whether the addition of RING influenced the folding stability and binding specificity of the interactors toward progerin and lamin A.
Build
All selected interactors were re-processed through our structural workflow, now including structure prediction with the RING domain, relaxation, validation, and docking simulations against both target proteins. Each model was evaluated through standardized quality parameters to ensure cross-dataset comparability and to detect any RING-related conformational effects.
Test
Docking outputs were analyzed using binding affinity (Kd) as the primary quantitative descriptor.
For each fusion construct, we calculated the Kdratio = Kd(progerin) / Kd(lamin a) to determine binding preference. A Kd ratio < 1 indicates a stronger affinity for progerin, while a Kd ratio > 1 indicates a stronger affinity for lamin A.
The full dataset and comparative analysis are reported in the Model section of our wiki.
Learn
From this analysis, we observed that some promising interactors, lost their proper folding when fused to the RING domain, leading to a modification in binding affinity for progerin, in some cases leading to a loss of specificity. This suggests that steric interference or linker rigidity may modify structural integrity. We therefore learned that further optimization — for example, evaluating different linker lengths or flexible sequences — will be necessary to preserve the correct folding while maintaining RING functionality. In the next cycle, we plan to build on the insights gained from this "learn" phase by further optimizing the linker length and assessing the impact of the RING domain. Specifically, we aim to experiment with varying linker lengths and flexible sequences to ensure proper folding while maintaining the functionality of the RING domain.
Design
To validate the Syn-Xpress™ SILVER cell-free protein synthesis (CFPS) system, we aimed to establish conditions for efficient expression of our constructs. The first objective was to test protein synthesis at 30 °C, a temperature that more closely resembles human physiological conditions. Although we were aware that the enzymatic machinery of the kit is derived from E. coli, we wanted to assess whether the system would still function efficiently under these conditions.
Build
Reagents (Syn-Xpress™ SILVER mix and plasmid DNA) were thawed on ice and handled carefully, avoiding vortex agitation. Reactions were prepared in 50 µL final volume, containing 40 µL of Syn-Xpress™ SILVER mix (final 1×), 15 ng/µL of plasmid DNA, and nuclease-free water. Tubes were sealed with sufficient headspace for oxygenation and incubated with shaking at 150 rpm.
Test
The reaction was incubated for 6 h at 30 °C. mEGFP expression was measured using a Thermo Scientific™ Varioskan™ LUX reader. Fluorescence values confirmed protein synthesis, with the positive control yielding 260 RFU.
Learn
Although protein was expressed at 30 °C, the fluorescence signal was lower than expected. This suggested that optimal expression of our constructs might occur at a different temperature, prompting a second cycle of optimization.
Design
Based on the reduced expression observed at 30 °C, we decided to lower the incubation temperature to 20 °C. This was expected to favor proper folding and stability of proteins synthesized in vitro, especially small peptides like SpyTag and SpyCatcher.
Build
The same CFPS workflow was followed: 50 µL reactions with 40 µL Syn-Xpress™ SILVER mix (final 1×), 15 ng/µL plasmid DNA, and nuclease-free water. Reactions were incubated with shaking at 150 rpm.
Test
The reaction was incubated for 6 h at 20 °C. mEGFP expression was quantified with the same fluorescence assay. The positive control yielded 859 RFU, a much stronger signal compared to 30 °C.
Learn
Lowering the incubation temperature significantly improved protein yield. For this reason, subsequent CFPS experiments with SpyTag and SpyCatcher were conducted at 20 °C, providing both higher expression and better reproducibility of results.
Design
One of the central objectives of our project was to establish a yeast-based model for Hutchinson–Gilford Progeria Syndrome (HGPS). Since our work focuses on studying protein–protein interactions, we identified the Yeast Two-Hybrid (Y2H) system as a suitable platform to search for and test potential interactors of progerin. By implementing this assay, we aimed to investigate both progerin’s interaction partners and its potential for self-association, a phenomenon already observed in vivo.
Build
Our initial design strategy was to express progerin in both Y2H plasmids, pGAD (bait) and pGBK (prey), fused with either the activation domain (AD) or the DNA binding domain (BD), respectively. This setup would allow us not only to test possible heterologous interactions, but also to directly test whether progerin could self-interact in the yeast model.
Test
We transformed Saccharomyces cerevisiae Y190 with the two Y2H plasmids pGBK and pGAD, carrying the full-length sequence of progerin (this initial version was not RFC10 optimised), in frame with BD and AD, respectively. Although our plan was to engineer the yeasts to express both plasmids simultaneously, this was not achieved due to two main issues:
- We did not observe consistent growth of the transformed yeast cultures;
- In the colonies that were able to grow on selective media, progerin expression was suppressed, as confirmed by Western blot analysis
Learn
The cytotoxicity observed with full-length progerin prompted us to revise our strategy and to restrict expression to its C-terminal portion (residues 430–614 and 545–614). This region is particularly relevant, as it contains the sequence differences that distinguish progerin from its wild type counterpart, lamin A. Specifically, we selected:
- Residues 430–614 BBa_25HM35ST, a longer sequence including the immunoglobulin-like domain;
- Residues 545–614 BBa_25IF4C4Z, a shorter sequence corresponding to the unique C-terminal tail
This allowed us to successfully express at least the C-term of progerin in the pGBK vector and proceed with further experiments.
- [1] Zhang, N., Hu, Q., Sui, T., Fu, L., Zhang, X., Wang, Y., Zhu, X., Huang, B., Lu, J., Li, Z., & Zhang, Y. (2023). Unique progerin C-terminal peptide ameliorates Hutchinson-Gilford progeria syndrome phenotype by rescuing BUBR1. Nature aging, 3(2), 185–201. https://doi.org/10.1038/s43587-023-00361-w
Design
Our goal was to insert our DNA fragments into the NanoBiT (for more information, check our experiments page) vectors to generate functional fusion constructs. We designed the ligation strategy using EcoRI and XhoI restriction sites to open the Multiple Cloning Site (MCS) and enable the directional insertion of our target sequences.
Build
We transformed E. coli DH5α with the ligation mixture and selected positive colonies for further analysis. Plasmids were extracted through mini-prep and purified according to standard protocols.
Test
Electrophoresis analysis of the purified plasmids revealed that most samples contained the recircularized vector rather than the desired ligation product. This indicated that the ligation had not occurred efficiently, likely due to the persistence of linearized backbone DNA in the mixture.
Learn
From this outcome, we understood the necessity of an additional purification step to remove unligated vector backbones. To improve ligation efficiency, we planned a gel purification to separate the opened plasmid backbone from the MCS fragment before repeating the cloning procedure.
Design
To overcome the recircularization problem, we aimed to increase ligation specificity by removing any residual backbone contaminants. A 1% agarose gel purification step was added to isolate the linearized NanoBiT vector digested with EcoRI and XhoI.
Build
Following gel extraction, purified backbones and inserts were quantified and used for a new ligation reaction under optimized molar ratios. The ligation products were then transformed into E. coli DH5α competent cells.
Test
Colonies were screened and analyzed by mini-prep and gel electrophoresis. The resulting plasmids showed bands corresponding to the expected size of the recombinant constructs, confirming successful ligation.
Learn
Introducing the gel purification step successfully eliminated the background of recircularized vector and improved cloning efficiency. This optimized workflow was subsequently adopted for all following NanoBiT construct assemblies.
Design
We planned to design and conduct our experiments using fibroblast cell lines derived from HGPS patients, kindly provided by Prof. Lattanzi. These cells represent a valuable model that naturally expresses progerin. Using patient-derived fibroblasts allows us to directly evaluate the physiological relevance of our degradation system and to assess its ability to specifically target and reduce progerin levels in a disease-relevant context.
Build
We cultured the cells in T25 flasks to monitor their growth dynamics, morphology, and overall health over time. This setup allowed us to carefully observe their adherence, proliferation, and characteristic morphological features under standard culture conditions.
Test
We allowed the cells to grow until they reached full confluence, ensuring that a uniform and continuous monolayer was formed across the entire flask surface. This step was essential to guarantee consistent cell density and physiological conditions for all subsequent experiments.
Learn
The cells did not proliferate as expected and failed to reach confluence in the T25 flasks, likely due to their reduced replicative capacity and altered metabolic state typical of HGPS-derived fibroblasts. As a result, we decided to adapt our experimental strategy by switching to healthy control fibroblasts provided by Prof. Lattanzi. These cells allow us to artificially express progerin and reproduce the pathological phenotype in a controlled manner.
Design
We planned to design our experiments using healthy fibroblast control cell lines provided by Prof. Lattanzi, which served as essential negative controls throughout the study. These cells, derived from unaffected donors, display normal nuclear architecture and lamin A expression, making them an ideal baseline for comparison with HGPS-derived or progerin-expressing fibroblasts. Furthermore, we intended to transfect these control fibroblasts with a progerin-expressing construct to induce the pathological phenotype, thereby enabling a controlled and reproducible model for our experiments.
Build
We cultured the cells in T25 flasks to monitor their growth dynamics, morphology, and overall health over time. This setup allowed us to carefully observe their adherence, proliferation, and characteristic morphological features under standard culture conditions.
Test
We allowed the cells to grow until they reached full confluence, ensuring that a uniform and continuous monolayer was formed across the entire flask surface. This step was essential to guarantee consistent cell density and physiological conditions for all subsequent experiments.
Learn
The control fibroblasts did not proliferate properly and were unable to reach confluence in the culture flasks, likely due to their nature as a primary cell line.. To ensure reproducibility and maintain optimal culture conditions, we therefore decided to switch to a commercially available human fibroblast cell line (MRC-5). These cells, derived from normal lung tissue, offer a robust and well-characterized model for transfection experiments. By introducing a progerin-expressing construct into MRC-5 cells, we aimed to recreate the progeria-like phenotype in a controlled and viable system, suitable for evaluating the efficiency of our designed degradation approach.
Design
We planned to design our experiments using the MRC-5 cell line, which we selected because it consists of human fibroblasts, making it suitable for modeling cellular processes related to progerin expression. By transfecting MRC-5 cells with progerin-expressing constructs, we aimed to reproduce the molecular and morphological features of HGPS in a controlled experimental context, providing a reliable model to evaluate the efficiency and specificity of our degradation system.
Build
We cultured the cells in T25 flasks to monitor their growth dynamics, morphology, and overall health over time. This setup allowed us to carefully observe their adherence, proliferation, and characteristic morphological features under standard culture conditions.
Test
We allowed the cells to grow until they reached full confluence, ensuring that a uniform and continuous monolayer was formed across the entire flask surface. This step was essential to guarantee consistent cell density and physiological conditions for all subsequent experiments.
Learn
The MRC5 cells were able to reach confluence in the flasks; but they exhibited very low transfection efficiency, making them unsuitable for reproducible expression studies. To overcome these limitations, we decided to use a commercially available human epithelial cell line (HEK 293T). These cells are widely used due to their high proliferative rate and exceptional transfection efficiency. By transfecting 293T cells with a progerin-expressing construct, we aimed to establish a reliable and scalable cellular model that would allow us to efficiently test our system’s ability to induce the degradation of progerin while maintaining optimal experimental reproducibility.