Jörg Gsponer

Computational predictors of protein-binding sites within intrinsically disordered regions (IDRs) show highly inconsistent performance across high-quality benchmark datasets. To understand the origins of these discrepancies, we systematically compared predictors across three independent test sets: two CAID datasets updated with the latest DisProt annotations and a composite dataset (DBs) assembled from DIBS, FuzDB, IDEAL, and MFIB. Predictors trained predominantly on DisProt data achieved substantially higher AUCs on the CAID sets but performed poorly on the DBs. In contrast, predictors trained on older, low-quality PDB-based datasets showed balanced performance across all sets, with a slight preference for DBs. Predictors with mixed training exposure displayed intermediate behavior. Through controlled experiments using identical CNN architectures and feature analysis, we demonstrate that the dominant factor driving these performance differences is the intrinsic disorder propensity of the binding sites themselves. Binding residues in DisProt-based datasets exhibit markedly higher average disorder propensity scores than those in PDB-derived datasets. This previously unrecognized selection bias (literature studies preferentially characterizing more disordered binding sites, while PDB-derived annotations capture less disordered ones) effectively splits IDR-protein binding sites into two distinct categories. Predictors optimized on one category therefore generalize poorly to the other. Binding-site length and sequence conservation play only minor or negligible roles in explaining the observed inconsistencies. These findings highlight a critical limitation in current benchmarking practices and training strategies for IDR-binding site prediction, underscoring the need for more balanced and disorder-aware reference datasets. Finally, the diagnostic techniques introduced here could prove valuable beyond the specific application examined in this study.

Context: Insulin action is critical for energy homeostasis and its dysfunction in muscle cells is associated with type 2 diabetes. Insulin receptor (INSR) internalization and cell-surface dynamics at rest and during insulin exposure are incompletely understood in muscle cells. Objective: We aimed to characterized the INSR dynamics and interactions in muscle. Methods: We applied inter-domain tagged INSR, microscopy, immunoprecipitation, mass spectrometry, and AlphaFold multimer to comprehensively profile INSR internalization and interactions with or without insulin stimulation. Results: Using surface labeling and live-cell imaging, we observed robust basal internalization of INSR in C2C12 myoblasts, without an effect of added insulin. Mass spectrometry using INSR knockout cells as controls identified high-confidence binding partners, including proteins associated with internalization. We confirmed known interactors, including insulin-like growth factor 1 receptor, and also identified underappreciated INSR-binding factors, such as annexin A2. AlphaFold multimer analysis predicted potential INSR-binding sites of these proteins. Protein–protein interaction network mapping suggested links between INSR and caveolin-mediated endocytosis. INSR interacted with both caveolin and clathrin heavy chain (CLTC) in mouse skeletal muscle and C2C12 myoblasts. Whole-cell 2-dimensional super-resolution imaging revealed that high levels of insulin (20 nM) increased INSR colocalization with caveolin-1 (CAV1) but decreased its colocalization with CLTC. Single-particle tracking confirmed the colocalization of cell-surface INSR with both overexpressed CAV1-mRFP (monomeric red fluorescent protein) and CLTC-mRFP. INSR tracks that colocalized with CAV1 exhibited longer radii and lifetimes, regardless of insulin exposure, compared with noncolocalized tracks, whereas insulin further increased the lifetime of INSR/CLTC-colocalized tracks. Conclusion: Overall, these data suggest that muscle cells utilize both CAV1- and CLTC-dependent pathways for INSR mobilization and internalization.

Many proteins function by toggling between distinct conformations, yet most structure predictors have been trained on data that do not capture this conformational diversity. Here, we benchmarked AlphaFold2, AlphaFold3, and recent variants on autoinhibited proteins, a class of allosterically regulated, often multi-domain proteins that exist in equilibrium between active and autoinhibited states. Our analyses show that AlphaFold2 fails to reproduce the experimental structures of many autoinhibited proteins, which is reflected in reduced confidence scores. This contrasts sharply with its high-accuracy, high-confidence predictions of non-autoinhibited multi-domain proteins. When tested for its ability to capture conformational diversity, we found that AlphaFold2 performs better when combined with uniform subsampling of sequence alignments rather than local subsampling. BioEmu and AlphaFold3 improve upon these results, yet still struggle to accurately reproduce details of experimental structures. Together, our study underscores the persistent challenges of predicting protein structures shaped by complex energy landscapes.

Synthetic biology enabled the systematic engineering of bacteria for diverse applications, but their deployment in open environments raises concerns about their persistence and unintended ecological impacts. To address these challenges, genetic “expiry-date” circuits were designed to impose a tunable lifespan on bacteria. These circuits, structured as a feedforward activation network, regulate the timing of cell death by controlling the expression of Lysis E, enabling a programmed lifespan ranging from hours to days. The lifespan can be tailored by modifying the number of activation steps in the cascade. The circuits were optimized by reducing gene expression leakiness of Lysis E using a synthetic small regulatory RNA and combining it with an asd-based auxotrophic system. The bacteria harboring the “expiry-date” circuits resulted in a GMO escape rate below U.S. NIH release standards (<10–10). To validate the practical applicability of this system, a synthetic phenol-scavenging Escherichia coli was constructed, which possessed enhanced phenol tolerance and phenol-detoxification capability, and harbored the “expiry-date” circuits. The engineered bacteria detoxified 0.1 g/kg of phenol in soil within 4 days and self-destructed by day 5. These results support the circuit’s potential as a biocontainment strategy for the safe and controlled deployment of synthetic bacteria in real-world applications.

Identifying sites within intrinsically disordered regions (IDR) that bind to other proteins remains a significant challenge. Molecular Recognition Features (MoRFs) are a subset of segments in IDR that bind to other proteins, undergoing a disorder-to-order transition upon binding. This paper introduces MoRFchibi 2.0, a specialized prediction tool designed to identify the locations of MoRFs within protein sequences. Our results show that MoRFchibi 2.0 outperforms all existing MoRF and general predictors of protein-binding sites within IDRs, including top-performing models from CAID rounds 1, 2, and 3. Remarkably, MoRFchibi 2.0 surpasses predictors that utilize AlphaFold data and state-of-the-art protein language models, achieving superior ROC and Precision-Recall curves and higher success rates. MoRFchibi 2.0 generates output scores using an ensemble of convolutional neural network logistic regression models, followed by a reverse Bayes Rule to adjust for priors in the training data. These scores reflect MoRF probabilities normalized for the priors in the training data, making them individually interpretable and compatible with other tools utilizing the same scoring framework. Availability https://mc2.msl.ubc.ca/index.xhtml

Summary: Biomolecular condensates organize cellular environments and regulate key processes such as transcription. We previously showed that full-length androgen receptor (AR-FL), a major oncogenic driver in prostate cancer (PCa), forms nuclear condensates upon androgen stimulation in androgen-sensitive PCa cells. Disrupting these condensates impairs AR-FL transcriptional activity, highlighting their functional importance. However, resistance to androgen deprivation therapy often leads to castration-resistant prostate cancer (CRPC), driven by constitutively active splice variants like AR variant 7 (AR-V7). The mechanisms underlying AR-V7’s role in CRPC remain unclear. In this study, we characterized the condensate-forming ability of AR-V7 and compared its phase behavior with AR-FL across a spectrum of PCa models and in vitro conditions. Our findings indicate that cellular context can influence AR-V7’s condensate-forming capacity. Unlike AR-FL, AR-V7 spontaneously forms condensates in the absence of androgen stimulation and functions independently of AR-FL in CRPC models. However, AR-V7 requires a higher concentration to form condensates, both in cellular contexts and in vitro . We further reveal that AR-V7 drives transcription via both condensate-dependent and condensate-independent mechanisms. Using an AR-V7 mutant incapable of forming condensates, while retaining nuclear localization and DNA-binding ability, we reveal that the condensate-dependent regime activates part of the oncogenic KRAS pathway in CRPC models. Genes under this condensate-dependent regime were found to harbor significantly higher numbers of AR-binding sites and exhibited boosted expression in response to AR-V7. These findings uncover a previously unrecognized role of AR-V7 condensate formation in driving oncogenic transcriptional programs and shed light on its unique contribution to CRPC progression. Highlights: AR-V7 condensates form independently of both androgens and AR-FL in CRPC models. AR-V7 mediates condensate-dependent and independent transcription Condensate-dependent transcription enables boosted expression of oncogenic KRAS genes Condensate-dependent genes exhibit an exponential increase in expression, with a higher number of AR binding sites potentially playing a key role in their reliance on condensate formation. Graphical

Intrinsically disordered protein regions, IDRs, are observed in many eukaryotic proteins. They play critical roles in essentially all cellular processes because segments of these regions, known as linear interacting peptides (LIPs), are heavily involved in regulatory protein interactions across proteomes. This chapter presents an integrated summary of the results from the last two Critical Assessments of protein Intrinsic Disorder predictions, known as CAID events, on the computational prediction of LIP segments. Because the CAID community questioned the quality of the test dataset used by the first CAID event, we reannotated this dataset using more accurate annotations from the latest DisProt database release. Then, we compared the results of the first CAID with the updated data and the results of the second CAID event. Our comparison highlights the importance of data annotation on the evaluation outcome and provides recommendations for users of LIP predictors.

Autoinhibition, a crucial allosteric self-regulation mechanism in cell signaling, ensures signal propagation exclusively in the presence of specific molecular inputs. The heightened focus on autoinhibited proteins stems from their implication in human diseases, positioning them as potential causal factors or therapeutic targets. However, the absence of a comprehensive knowledgebase impedes a thorough understanding of their roles and applications in drug discovery. Addressing this gap, we introduce Autoinhibited Protein Database (AiPD), a curated database standardizing information on autoinhibited proteins. AiPD encompasses details on autoinhibitory domains (AIDs), their targets, regulatory mechanisms, experimental validation methods, and implications in diseases, including associated mutations and post-translational modifications. AiPD comprises 698 AIDs from 532 experimentally characterized autoinhibited proteins and 2695 AIDs from their 2096 homologs, which were retrieved from 864 published articles. AiPD also includes 42 520 AIDs of computationally predicted autoinhibited proteins. In addition, AiPD facilitates users in investigating potential AIDs within a query sequence through comparisons with documented autoinhibited proteins. As the inaugural autoinhibited protein repository, AiPD significantly aids researchers studying autoinhibition mechanisms and their alterations in human diseases. It is equally valuable for developing computational models, analyzing allosteric protein regulation, predicting new drug targets, and understanding intervention mechanisms AiPD serves as a valuable resource for diverse researchers, contributing to the understanding and manipulation of autoinhibition in cellular processes. Database URL: http://ssbio.cau.ac.kr/databases/AiPD.