Ezgi Karaca

CASK-related disorders comprise two main phenotypes: severe microcephaly with pontocerebellar hypoplasia (MICPCH), typically caused by truncating variants, and milder X-linked intellectual disability (XLID), usually associated with missense variants. We aimed to characterize this molecular spectrum within an underrepresented national cohort. We evaluated 18 affected individuals from 16 unrelated families with CASK variants identified through exome or genome sequencing. Clinical, neurodevelopmental and neuroimaging findings were systematically reviewed. Selected missense variants were evaluated by in silico prediction tools and structural modeling. We identified 16 distinct CASK variants, 11 of which are novel, including frameshift, stop-gain, splice-site, and missense variants, as well as one large deletion. Consistent with known genotype-phenotype correlations, null alleles resulted in the severe MICPCH phenotype, while presumed hypomorphic alleles were associated with XLID. However, unlike previous reports, our MICPCH patients exhibited a higher frequency of dysmorphic facial features (86%), skeletal anomalies (57%) and sensorineural hearing loss (43%). Furthermore, contrary to the literature suggesting a milder epilepsy in females due to X-inactivation, 63% of our female patients with MICPCH presented with recurrent seizures. This cohort broadens both the molecular and phenotypic spectrum of CASK-related disorders. Findings reinforce that truncating variants predominantly cause MICPCH, while missense variants can lead to various outcomes depending on domain-level structural impact. We demonstrate that the CASK phenotypic spectrum is broader than previously recognized, frequently involving distinctive facial dysmorphism, skeletal defects and hearing impairment, suggesting that CASK plays a critical role not only in synaptic function but also in wider developmental processes.

AlphaFold models provide static structural predictions, limiting their use in interpreting flexible regions in low‐resolution cryo‐EM maps. Here, we assess whether AlphaFold‐generated distograms can instead reveal conformational flexibility, focusing on binding‐induced hinge motions. For this, we examined the key metabolic AK2/AIFM1 complex, where molecular dynamics and cryo‐EM confirm a hinge motion in AK2 upon binding. Notably, this motion is captured in the AlphaFold2/3 distograms of apo AK2, even though it is absent in the predicted structures. By extending our analysis to other systems, we demonstrate that distograms may offer a valuable, model‐independent method for interpreting ambiguous hinge regions in cryo‐EM maps. Impact statement: We reveal that AlphaFold distograms can successfully predict binding‐induced hinge motions. This establishes distograms as a valuable, structure‐free metric for identifying alternative conformational states, aiding the interpretation of ambiguous densities in cryo‐EM maps.

Static structural models often fail to capture the dynamic mechanisms of protein interactions. To address this, we introduce DynaPIN, an open-source pipeline for extracting dynamic interface fingerprints from molecular simulations. DynaPIN unifies quality control metrics, interface prediction accuracy assessment, and atomistic interaction analysis into a single automated workflow, accessible at https://github.com/CSB-KaracaLab/DynaPIN . A key feature is our interface-specific analysis centered on a Dynamic Interface definition, which classifies residues based on the persistence of their interaction status over the simulation. We applied DynaPIN to representative rigid, medium, and difficult targets from the DynaBench dataset, an MD simulation resource for Docking Benchmark 5.5. Our results show that interface flexibility diverges from static accuracy classifications established in Docking Benchmark 5.5, as explored before. All in all, by providing standardized, frame-resolved outputs, DynaPIN’s aim is to facilitate mechanistic studies and generate standardized unbiased data for future dynamics-aware artificial intelligence models.

Oestrogen receptor alpha (ERα)-positive (ER+) breast cancers are driven by the binding of 17β-oestradiol (E2) to ERα, which transcriptionally regulates target genes. Although microarrays and conventional RNA sequencing have identified E2 target genes, pre-designed probes and short read lengths are limited in their ability to accurately capture complex transcript structures. Long-read sequencing offers a solution by spanning entire transcripts, providing a more complete view of the transcriptome. Here, we employed nanopore long-read direct RNA sequencing (DRS) complemented with 3'-end sequencing, in vitro experiments, and deep learning-based protein modelling to explore the landscape of the E2-responsive transcriptome and protein level implications. Our analysis revealed a range of E2-responsive non-coding and coding isoforms, including intronically polyadenylated (IPA) mRNAs. One of these IPA isoforms was detected for TLE1, which assists ERα-chromatin interactions for a subset of E2 targets. The IPA isoform produces a C-terminus truncated protein that lacks the WDR interaction domain but retains dimerization/tetramerization capacity through its intact N-terminal Q-domain. Structural modelling and protein-based assays confirmed the dimerization potential and nuclear localization of the truncated protein. Functional assays showed that the overexpression of truncated TLE1 reduced the E2-induced upregulation of TFF1 and GREB1, E2-responsive genes, thereby disrupting transcriptional regulation. Importantly, a lower IPA isoform ratio is associated with worse survival in ER+ breast cancers, highlighting clinical relevance. Our study revealed new layers of complexity in the E2-regulated transcriptome, providing insights into truncated proteins. These findings contribute to a deeper understanding of gene regulation and may help the development of new therapeutic strategies.

DNMT3A and DNMT3B are closely related DNA methyltransferases that catalyze de novo CpG methylation and have distinct preferences for flanking sequences. Despite sharing 91% sequence similarity within their catalytic domains, these paralogs show non-overlapping genomic targeting and divergent biological roles. To uncover the mechanistic basis of this specificity, we performed 16µs of all-atom molecular dynamics simulations on DNMT3A and DNMT3B complexes bound to CpG substrates with varied +2 flanking bases (i.e., CGX). To resolve their base- and shape-readout mechanisms at atomistic detail, we introduced a Comparative Dynamics Analysis (CDA) framework. Our CDA approach revealed that DNMT3A relies on a rigid, sequence-specific hydrogen bonding network and shape-constrained electrostatic anchoring, whereas DNMT3B employs a more flexible and distributed interface, allowing broader substrate tolerance. This represents the first systematic analysis of shape readout in DNMT3 enzymes and demonstrates how flanking sequence specificity is dynamically encoded by two nearly identical proteins. Our findings not only clarify how closely related DNA-modifying enzymes diverge in recognition strategies, but also lay the foundation for future efforts to engineer paralog-specific protein-DNA interactions.

Estrogen receptor alpha (ERα)-positive (ER+) breast cancers are driven by 17β-estradiol (E2) binding to ERα, which transcriptionally regulates downstream target genes. Although microarrays and conventional RNA sequencing have identified E2 target genes, pre-designed probes and short read lengths are limited in accurately capturing complex transcript structures. Long-read RNA sequencing offers a solution by spanning entire transcripts, providing a more complete view of the transcriptome. Here, we employed nanopore long-read direct RNA sequencing (DRS) complemented with 3’-end sequencing, in vitro experiments, and deep learning-based protein modeling to explore the intricate landscape of E2-responsive transcriptome and protein level implications. Our analysis revealed a range of E2-responsive non-coding and coding isoforms, including intronically polyadenylated (IPA) mRNAs. One of these IPA isoforms was detected for TLE1 (Transducin-like enhancer protein 1), which positively assists ERα-chromatin interactions for a subset of E2 target genes. The IPA isoform produces a C-terminus truncated protein, lacking the WDR interaction domain, but retains dimerization/tetramerization capacity through its intact N-terminus Q-domain. Structural modeling and protein-based assays confirmed the truncated protein’s dimerization potential and nuclear localization. Functional assays showed that overexpression of truncated TLE1 reduces the E2-induced upregulation of GREB1 , an E2-responsive gene, thereby disrupting transcriptional regulation. Importantly, a lower IPA isoform ratio is associated with worse survival in ER+ patients, highlighting clinical relevance. Our study uncovers new layers of complexity in the E2-regulated transcriptome, providing insights into truncated proteins. These findings contribute to a deeper understanding of gene regulation and may help the development of new therapeutic strategies.

Morphogen gradients direct tissue patterning by inducing dose dependent transcriptional responses, yet ligand producing cells often respond differently from their neighbors. Using the zebrafish lateral line organogenesis model, we uncover a cell autonomous role for the paracrine ligand Fgf3. Transcriptomic profiling and quantitative single molecule imaging identify target genes, including the chemokine scavenger cxcr7b, whose expression decreases both when FGF receptor signaling is inhibited and when Fgf3 is overexpressed. High resolution live imaging reveals nuclear accumulation of Fgf3 in producing cells, whereas neighbors receive only extracellular ligand, a feature also observed in other embryonic tissues. Mosaic gain of function and nanobody mediated degradation demonstrate that the nuclear pool of Fgf3 autonomously represses specific targets without impairing canonical receptor signaling. Structure guided comparative assays indicate nuclear targeting as a latent property of several paracrine FGFs. Dual secreted nuclear functionality of FGF ligands may represent an intrinsic symmetry breaking mechanism during organogenesis.

AlphaFold models provide static structural predictions, limiting their use in interpreting flexible regions in low-resolution cryo-EM maps. Here, we assess whether AlphaFold-generated distograms can instead reveal conformational flexibility, focusing on binding-induced hinge motions. For this, we examined the key metabolic AK2/AIFM1 complex, where molecular dynamics and cryo-EM confirm a hinge motion in AK2 upon binding. Notably, this motion is captured in the AlphaFold2/3 distograms of apo AK2, even though it is absent in the predicted structures. By extending our analysis to other systems, we demonstrate that distograms may offer a valuable, model-independent method for interpreting ambiguous regions in cryo-EM maps.