DNA breathing integration with deep learning foundational model advances genome-wide binding prediction of human transcription factors.

It was previously shown that DNA breathing, thermodynamic stability, as well as transcriptional activity and transcription factor (TF) bindings are functionally correlated. To ascertain the precise relationship between TF binding and DNA breathing, we developed the multi-modal deep learning model EPBDxDNABERT-2, which is based on the Extended Peyrard-Bishop-Dauxois (EPBD) nonlinear DNA dynamics model. To train our EPBDxDNABERT-2, we used chromatin immunoprecipitation sequencing (ChIP-Seq) data comprising 690 ChIP-seq experimental results encompassing 161 distinct TFs and 91 human cell types. EPBDxDNABERT-2 significantly improves the prediction of over 660 TF-DNA, with an increase in the area under the receiver operating characteristic (AUROC) metric of up to 9.6% when compared to the baseline model that does not leverage DNA biophysical properties. We expanded our analysis to in vitro high-throughput Systematic Evolution of Ligands by Exponential enrichment (HT-SELEX) dataset of 215 TFs from 27 families, comparing EPBD with established frameworks. The integration of the DNA breathing features with DNABERT-2 foundational model, greatly enhanced TF-binding predictions. Notably, EPBDxDNABERT-2, trained on a large-scale multi-species genomes, with a cross-attention mechanism, improved predictive power shedding light on the mechanisms underlying disease-related non-coding variants discovered in genome-wide association studies.

CI:Mor interactions in the lysogeny switches of Lactococcus lactis TP901-1 and Staphylococcus aureus φ13 bacteriophages.

Aim: To structurally characterize in detail the interactions between the phage repressor (CI) and the antirepressor (Mor) in the lysis-lysogeny switches of two Gram-positive bacteriophages, the lactococcal TP901-1 and staphylococcal φ13. Methods: We use crystallographic structure determination, computational structural modeling, and analysis, as well as biochemical methods, to elucidate similarities and differences in the CI:Mor interactions for the two genetic switches. Results: By comparing a newly determined and other available crystal structures for the N-terminal domain of CI (CI-NTD), we show that the CI interface involved in Mor binding undergoes structural changes upon binding in TP901-1. Most importantly, we show experimentally for the first time the direct interaction between CI and Mor for φ13, and model computationally the interaction interface. The computational modeling supports similar side chain rearrangements in TP901-1 and φ13. Conclusion: This study ascertains experimentally that, like in the TP901-1 lysogeny switch, staphylococcal φ13 CI and Mor interact with each other. The structural basis of the interaction of φ13 CI and Mor was computationally modeled and is similar to the interaction demonstrated experimentally between TP901-1 CI-NTD and Mor, likely involving similar rearrangement of residue side chains during the formation of the complex. The study identifies one CI residue, Glu69, which unusually interacts primarily through its aliphatic chain with an aromatic residue on Mor after changing its conformation compared to the un-complexed structure. This and other residues at the interface are suggested for investigation in future studies.

Advancing Transcription Factor Binding Site Prediction Using DNA Breathing Dynamics and Sequence Transformers via Cross Attention.

Understanding the impact of genomic variants on transcription factor binding and gene regulation remains a key area of research, with implications for unraveling the complex mechanisms underlying various functional effects. Our study delves into the role of DNA's biophysical properties, including thermodynamic stability, shape, and flexibility in transcription factor (TF) binding. We developed a multi-modal deep learning model integrating these properties with DNA sequence data. Trained on ChIP-Seq (chromatin immunoprecipitation sequencing) data in vivo involving 690 TF-DNA binding events in human genome, our model significantly improves prediction performance in over 660 binding events, with up to 9.6% increase in AUROC metric compared to the baseline model when using no DNA biophysical properties explicitly. Further, we expanded our analysis to in vitro high-throughput Systematic Evolution of Ligands by Exponential enrichment (SELEX) and Protein Binding Microarray (PBM) datasets, comparing our model with established frameworks. The inclusion of DNA breathing features consistently improved TF binding predictions across different cell lines in these datasets. Notably, for complex ChIP-Seq datasets, integrating DNABERT2 with a cross-attention mechanism provided greater predictive capabilities and insights into the mechanisms of disease-related non-coding variants found in genome-wide association studies. This work highlights the importance of DNA biophysical characteristics in TF binding and the effectiveness of multi-modal deep learning models in gene regulation studies.

Examining DNA breathing with pyDNA-EPBD.

MOTIVATION: The two strands of the DNA double helix locally and spontaneously separate and recombine in living cells due to the inherent thermal DNA motion. This dynamics results in transient openings in the double helix and is referred to as "DNA breathing" or "DNA bubbles." The propensity to form local transient openings is important in a wide range of biological processes, such as transcription, replication, and transcription factors binding. However, the modeling and computer simulation of these phenomena, have remained a challenge due to the complex interplay of numerous factors, such as, temperature, salt content, DNA sequence, hydrogen bonding, base stacking, and others. RESULTS: We present pyDNA-EPBD, a parallel software implementation of the Extended Peyrard-Bishop-Dauxois (EPBD) nonlinear DNA model that allows us to describe some features of DNA dynamics in detail. The pyDNA-EPBD generates genomic scale profiles of average base-pair openings, base flipping probability, DNA bubble probability, and calculations of the characteristically dynamic length indicating the number of base pairs statistically significantly affected by a single point mutation using the Markov Chain Monte Carlo algorithm. AVAILABILITY AND IMPLEMENTATION: pyDNA-EPBD is supported across most operating systems and is freely available at https://github.com/lanl/pyDNA_EPBD. Extensive documentation can be found at https://lanl.github.io/pyDNA_EPBD/.

Examining DNA Breathing with pyDNA-EPBD.

Motivation: The two strands of the DNA double helix locally and spontaneously separate and recombine in living cells due to the inherent thermal DNA motion.This dynamics results in transient openings in the double helix and is referred to as "DNA breathing" or "DNA bubbles." The propensity to form local transient openings is important in a wide range of biological processes, such as transcription, replication, and transcription factors binding. However, the modeling and computer simulation of these phenomena, have remained a challenge due to the complex interplay of numerous factors, such as, temperature, salt content, DNA sequence, hydrogen bonding, base stacking, and others. Results: We present pyDNA-EPBD, a parallel software implementation of the Extended Peyrard-Bishop- Dauxois (EPBD) nonlinear DNA model that allows us to describe some features of DNA dynamics in detail. The pyDNA-EPBD generates genomic scale profiles of average base-pair openings, base flipping probability, DNA bubble probability, and calculations of the characteristically dynamic length indicating the number of base pairs statistically significantly affected by a single point mutation using the Markov Chain Monte Carlo (MCMC) algorithm.

Flexible linker modulates the binding affinity of the TP901-1 CI phage repressor to DNA.

Temperate bacteriophages can switch between two life cycles following infection of a host bacterium: the lytic or lysogenic life cycle. The choice between these is controlled by a bistable genetic switch. We investigated the genetic switch of the lactococcal temperate bacteriophage, TP901-1, which is controlled by two regulatory proteins, the Clear 1 (CI) repressor and modulator of repression (MOR) antirepressor. CI consists of a DNA-binding N-terminal domain and a C-terminal domain responsible for oligomerization, connected by a flexible interdomain linker. Full-length CI is hexameric, whereas the truncated version CI with 58 C-terminal residues truncated (CIΔ58), missing the second C-terminal subdomain, is dimeric, but binds with the same affinity as full-length CI to the OL operator site, responsible for lytic genes transcription repression. Three variants of CIΔ58 with shorter, longer, and PP substituted linkers were produced and confirmed by circular dichroism spectroscopy and nanodifferential scanning fluorimetry to be well folded. With small-angle X-ray scattering, we delineated the conformational space sampled by the variants and wild-type in solution and found that shortening and lengthening the linker decrease and increase this, respectively, as also substantiated by molecular dynamics and as intended. Isoelectric focusing electrophoresis confirmed that all variants are able to bind to the MOR antirepressor. However, using electrophoretic mobility shift assays, we showed that shortening and lengthening the linker lead to a 94 and 17 times decrease in affinity to OL , respectively. Thus, an appropriate linker length appears to be crucial for appropriate DNA-binding and subsequent TP901-1 genetic switch function.

Engineering the substrate binding site of the hyperthermostable archaeal endo-ß-1,4-galactanase from Ignisphaera aggregans.

BACKGROUND: Endo-ß-1,4-galactanases are glycoside hydrolases (GH) from the GH53 family belonging to the largest clan of GHs, clan GH-A. GHs are ubiquitous and involved in a myriad of biological functions as well as being widely used industrially. Endo-ß-1,4-galactanases, in particular hydrolyse galactan and arabinogalactan in pectin, a major component of the primary plant cell wall, with important functions in plant defence and application in the food and other industries. Here, we explore the family's biological diversity by characterizing the first archaeal and hyperthermophilic GH53 galactanase, and utilize it as a scaffold for engineering enzymes with different product lengths. RESULTS: A galactanase gene was identified in the genome of the anaerobic hyperthermophilic archaeon Ignisphaera aggregans, and the isolated catalytic domain expressed and characterized (IaGal). IaGal presents the typical (ßα)8 barrel structure of clan GH-A enzymes, with catalytic carboxylates at the end of the 4th and 7th barrel strands. Its activity optimum of at least 95 °C and melting point over 100 °C indicate extreme thermostability, a very advantageous property for industrial applications. If enzyme depletion is reduced, so is the need for re-addition, and thus costs. The main stabilizing features of IaGal compared to other structurally characterized members are π-π and cation-π interactions. The length of the substrate binding site-and thus produced oligosaccharide products-is intermediate compared to previously characterized galactanases. Variants inspired by the structural diversity in the GH53 family were rationally designed to shorten or extend the substrate binding groove, in order to modulate product length. Subsite-deleted variants produced shorter products than IaGal, as do the fungal galactanases inspiring the design. IaGal variants engineered with a longer binding site produced a less expected degradation pattern, though still different from that of wild-type IaGal. All variants remained extremely stable. CONCLUSIONS: We have characterized in detail the most thermophilic endo-ß-1,4-galactanase known to date and successfully engineered it to modify the degradation profile, while maintaining much of its desirable thermostability. This is an important achievement as oligosaccharide products length is an important property for industrial and natural GHs alike.

Revealing the mechanism of repressor inactivation during switching of a temperate bacteriophage.

Temperate bacteriophages can enter one of two life cycles following infection of a sensitive host: the lysogenic or the lytic life cycle. The choice between the two alternative life cycles is dependent upon a tight regulation of promoters and their cognate regulatory proteins within the phage genome. We investigated the genetic switch of TP901-1, a bacteriophage of Lactococcus lactis, controlled by the CI repressor and the modulator of repression (MOR) antirepressor and their interactions with DNA. We determined the solution structure of MOR, and we solved the crystal structure of MOR in complex with the N-terminal domain of CI, revealing the structural basis of MOR inhibition of CI binding to the DNA operator sites. 15N NMR Carr-Purcell-Meiboom-Gill (CPMG) relaxation dispersion and rotating frame R1ρ measurements demonstrate that MOR displays molecular recognition dynamics on two different time scales involving a repacking of aromatic residues at the interface with CI. Mutations in the CI:MOR binding interface impair complex formation in vitro, and when introduced in vivo, the bacteriophage switch is unable to choose the lytic life cycle showing that the CI:MOR complex is essential for proper functioning of the genetic switch. On the basis of sequence alignments, we show that the structural features of the MOR:CI complex are likely conserved among a larger family of bacteriophages from human pathogens implicated in transfer of antibiotic resistance.

Formation and Structure of Fluorescent Silver Nanoclusters at Interfacial Binding Sites Facilitating Oligomerization of DNA Hairpins.

Fluorescent, DNA-stabilized silver nanoclusters (DNA-AgNCs) are applied in a range of applications within nanoscience and nanotechnology. However, their diverse optical properties, mechanism of formation, and aspects of their composition remain unexplored, making the rational design of nanocluster probes challenging. Herein, a synthetic procedure is described for obtaining a high yield of emissive DNA-AgNCs with a C-loop hairpin DNA sequence, with subsequent purification by size-exclusion chromatography (SEC). Through a combination of optical spectroscopy, gel electrophoresis, inductively coupled plasma mass spectrometry (ICP-MS), and small-angle X-ray scattering (SAXS) in conjunction with the systematic study of various DNA sequences, the low-resolution structure and mechanism of the formation of AgNCs were investigated. Data indicate that fluorescent DNA-AgNCs self-assemble by a head-to-head binding of two DNA hairpins, bridged by a silver nanocluster, resulting in the modelling of a dimeric structure harboring an Ag12 cluster.

Repression of the lysogenic PR promoter in bacteriophage TP901-1 through binding of a CI-MOR complex to a composite OM-OR operator.

A functional genetic switch from the lactococcal bacteriophage TP901-1, deciding which of two divergently transcribing promoters becomes most active and allows this bi-stable decision to be inherited in future generations requires a DNA region of less than 1 kb. The fragment encodes two repressors, CI and MOR, transcribed from the PR and PL promoters respectively. CI can repress the transcription of the mor gene at three operator sites (OR, OL, and OD), leading to the immune state. Repression of the cI gene, leading to the lytic (anti-immune) state, requires interaction between CI and MOR by an unknown mechanism, but involving a CI:MOR complex. A consensus for putative MOR binding sites (OM sites), and a common topology of three OM sites adjacent to the OR motif was here identified in diverse phage switches that encode CI and MOR homologs, in a search for DNA sequences similar to the TP901-1 switch. The OR site and all putative OM sites are important for establishment of the anti-immune repression of PR, and a putative DNA binding motif in MOR is needed for establishment of the anti-immune state. Direct evidence for binding between CI and MOR is here shown by pull-down experiments, chemical crosslinking, and size exclusion chromatography. The results are consistent with two possible models for establishment of the anti-immune repression of cI expression at the PR promoter.

Neurotrophic Effects of Vascular Endothelial Growth Factor B and Novel Mimetic Peptides on Neurons from the Central Nervous System.

Vascular endothelial growth factor B (VEGFB) is a pleiotropic trophic factor, which in contrast to the closely related VEGFA is known to have a limited effect on angiogenesis. VEGFB improves survival in various tissues including the nervous system, where the effect was observed mainly for peripheral neurons. The neurotrophic effect of VEGFB on central nervous system neurons has been less investigated. Here we demonstrated that VEGFB promotes neurite outgrowth from primary cerebellar granule, hippocampal, and retinal neurons in vitro. VEGFB protected hippocampal and retinal neurons from both oxidative stress and glutamate-induced neuronal death. The VEGF receptor 1 (VEGFR1) is required for VEGFB-induced neurotrophic and neuroprotective effects. Using a structure-based approach, we designed short peptides, termed Vefin1-7, mimicking the binding interface of VEGFB to VEGFR1. Vefins were analyzed for their secondary structure and binding to VEGF receptors and compared with previously described peptides derived from VEGFA, another ligand of VEGFR1. We show that Vefins have neurotrophic and neuroprotective effects on primary hippocampal, cerebellar granule, and retinal neurons in vitro with potencies comparable to VEGFB. Similar to VEGFB, Vefins were not mitogenic for MCF-7 cancer cells. Furthermore, one of the peptides, Vefin7, even dose-dependently inhibited the proliferation of MCF-7 cells in vitro. Unraveling the neurotrophic and neuroprotective potentials of VEGFB, the only nonangiogenic factor of the VEGF family, is promising for the development of neuroprotective peptide-based therapies.

Structure and Dynamics of a Promiscuous Xanthan Lyase from Paenibacillus nanensis and the Design of Variants with Increased Stability and Activity.

We have characterized the structure and dynamics of the carbohydrate-modifying enzyme Paenibacillus nanensis xanthan lyase (PXL) involved in the degradation of xanthan by X-ray crystallography, small-angle X-ray scattering, and hydrogen/deuterium exchange mass spectrometry. Unlike other xanthan lyases, PXL is specific for both unmodified mannose and pyruvylated mannose, which we find is correlated with structural differences in the substrate binding groove. The structure of the full-length enzyme reveals two additional C-terminal modules, one of which belongs to a new non-catalytic carbohydrate binding module family. Ca2+ are critical for the activity and conformation of PXL, and we show that their removal by chelating agents results in localized destabilization/unfolding of particularly the C-terminal modules. We use the structure and the revealed impact of Ca2+ coordination on conformational dynamics to guide the engineering of PXL variants with increased activity and stability in a chelating environment, thus expanding the possibilities for industrial applications of PXL.

NCAM2 Fibronectin type-III domains form a rigid structure that binds and activates the Fibroblast Growth Factor Receptor.

NCAM1 and NCAM2 have ectodomains consisting of 5 Ig domains followed by 2 membrane-proximal FnIII domains. In this study we investigate and compare the structures and functions of these FnIII domains. The NCAM1 and -2 FnIII2 domains both contain a Walker A motif. In NCAM1 binding of ATP to this motif interferes with NCAM1 binding to FGFR. We obtained a structural model of the NCAM2 FnIII2 domain by NMR spectroscopy, and by titration with an ATP analogue we show that the NCAM2 Walker A motif does not bind ATP. Small angle X-ray scattering (SAXS) data revealed that the NCAM2 FnIII1-2 double domain exhibits a very low degree of flexibility. Moreover, recombinant NCAM2 FnIII domains bind FGFR in vitro, and the FnIII1-2 double domain induces neurite outgrowth in a concentration-dependent manner through activation of FGFR. Several synthetic NCAM1-derived peptides induce neurite outgrowth via FGFR. Only 2 of 5 peptides derived from similar regions in NCAM2 induce neurite outgrowth, but the most potent of these peptides stimulates neurite outgrowth through FGFR-dependent activation of the Ras-MAPK pathway. These results reveal that the NCAM2 FnIII domains form a rigid structure that binds and activates FGFR in a manner related to, but different from NCAM1.

Structural basis of the bacteriophage TP901-1 CI repressor dimerization and interaction with DNA.

Temperate bacteriophages are known for their bistability, which in TP901-1 is controlled by two proteins, CI and MOR. Clear 1 repressor (CI) is hexameric and binds three palindromic operator sites via an N-terminal helix-turn-helix domain (NTD). A dimeric form, such as the truncated CI∆58 investigated here, is necessary for high-affinity binding to DNA. The crystal structure of the dimerization region (CTD1 ) is determined here, showing that it forms a pair of helical hooks. This newly determined structure is used together with the known crystal structure of the CI-NTD and small angle X-ray scattering data, to determine the solution structure of CI∆58 in complex with a palindromic operator site, showing that the two NTDs bind on opposing sides of the DNA helix.

Evaluating the role of coherent delocalized phonon-like modes in DNA cyclization.

The innate flexibility of a DNA sequence is quantified by the Jacobson-Stockmayer's J-factor, which measures the propensity for DNA loop formation. Recent studies of ultra-short DNA sequences revealed a discrepancy of up to six orders of magnitude between experimentally measured and theoretically predicted J-factors. These large differences suggest that, in addition to the elastic moduli of the double helix, other factors contribute to loop formation. Here, we develop a new theoretical model that explores how coherent delocalized phonon-like modes in DNA provide single-stranded "flexible hinges" to assist in loop formation. We combine the Czapla-Swigon-Olson structural model of DNA with our extended Peyrard-Bishop-Dauxois model and, without changing any of the parameters of the two models, apply this new computational framework to 86 experimentally characterized DNA sequences. Our results demonstrate that the new computational framework can predict J-factors within an order of magnitude of experimental measurements for most ultra-short DNA sequences, while continuing to accurately describe the J-factors of longer sequences. Further, we demonstrate that our computational framework can be used to describe the cyclization of DNA sequences that contain a base pair mismatch. Overall, our results support the conclusion that coherent delocalized phonon-like modes play an important role in DNA cyclization.

Expression, refolding and spectroscopic characterization of fibronectin type III (FnIII)-homology domains derived from human fibronectin leucine rich transmembrane protein (FLRT)-1, -2, and -3.

The fibronectin leucine rich transmembrane (FLRT) protein family consists in humans of 3 proteins, FLRT1, -2, and -3. The FLRT proteins contain two extracellular domains separated by an unstructured linker. The most membrane distal part is a leucine rich repeat (LRR) domain responsible for both cis- and trans-interactions, whereas the membrane proximal part is a fibronectin type III (FnIII) domain responsible for a cis-interaction with members of the fibroblast growth factor receptor 1 (FGFR1) family, which results in FGFR tyrosine kinase activation. Whereas the structures of FLRT LRR domains from various species have been determined, the expression and purification of recombinant FLRT FnIII domains, important steps for further structural and functional characterizations of the proteins, have not yet been described. Here we present a protocol for expressing recombinant FLRT-FnIII domains in inclusion bodies in Escherichia coli. His-tags permitted affinity purification of the domains, which subsequently were refolded on a Ni-NTA agarose column by reducing the concentration of urea. The refolding was confirmed by circular dichroism (CD) and 1H-NMR. By thermal unfolding experiments we show that a strand-strand cystine bridge has significant effect on the stability of the FLRT FnIII fold. We further show by Surface Plasmon Resonance that all three FnIII domains bind to FGFR1, and roughly estimate a Kd for each domain, all Kd s being in the µM range.

Structural and dynamics studies of a truncated variant of CI repressor from bacteriophage TP901-1.

The CI repressor from the temperate bacteriophage TP901-1 consists of two folded domains, an N-terminal helix-turn-helix DNA-binding domain (NTD) and a C-terminal oligomerization domain (CTD), which we here suggest to be further divided into CTD1 and CTD2. Full-length CI is a hexameric protein, whereas a truncated version, CI∆58, forms dimers. We identify the dimerization region of CI∆58 as CTD1 and determine its secondary structure to be helical both within the context of CI∆58 and in isolation. To our knowledge this is the first time that a helical dimerization domain has been found in a phage repressor. We also precisely determine the length of the flexible linker connecting the NTD to the CTD. Using electrophoretic mobility shift assays and native mass spectrometry, we show that CI∆58 interacts with the OL operator site as one dimer bound to both half-sites, and with much higher affinity than the isolated NTD domain thus demonstrating cooperativity between the two DNA binding domains. Finally, using small angle X-ray scattering data and state-of-the-art ensemble selection techniques, we delineate the conformational space sampled by CI∆58 in solution, and we discuss the possible role that the dynamics play in CI-repressor function.

The role of structural parameters in DNA cyclization.

BACKGROUND: The intrinsic bendability of DNA plays an important role with relevance for myriad of essential cellular mechanisms. The flexibility of a DNA fragment can be experimentally and computationally examined by its propensity for cyclization, quantified by the Jacobson-Stockmayer J factor. In this study, we use a well-established coarse-grained three-dimensional model of DNA and seven distinct sets of experimentally and computationally derived conformational parameters of the double helix to evaluate the role of structural parameters in calculating DNA cyclization. RESULTS: We calculate the cyclization rates of 86 DNA sequences with previously measured J factors and lengths between 57 and 325 bp as well as of 20,000 randomly generated DNA sequences with lengths between 350 and 4000 bp. Our comparison with experimental data is complemented with analysis of simulated data. CONCLUSIONS: Our data demonstrate that all sets of parameters yield very similar results for longer DNA fragments, regardless of the nucleotide sequence, which are in agreement with experimental measurements. However, for DNA fragments shorter than 100 bp, all sets of parameters performed poorly yielding results with several orders of magnitude difference from the experimental measurements. Our data show that DNA cyclization rates calculated using conformational parameters based on nucleosome packaging data are most similar to the experimental measurements. Overall, our study provides a comprehensive large-scale assessment of the role of structural parameters in calculating DNA cyclization rates.

Allostery through protein-induced DNA bubbles.

Allostery through DNA is increasingly recognized as an important modulator of DNA functions. Here, we show that the coalescence of protein-induced DNA bubbles can mediate allosteric interactions that drive protein aggregation. We propose that such allostery may regulate DNA's flexibility and the assembly of the transcription machinery. Mitochondrial transcription factor A (TFAM), a dual-function protein involved in mitochondrial DNA (mtDNA) packaging and transcription initiation, is an ideal candidate to test such a hypothesis owing to its ability to locally unwind the double helix. Numerical simulations demonstrate that the coalescence of TFAM-induced bubbles can explain experimentally observed TFAM oligomerization. The resulting melted DNA segment, approximately 10 base pairs long, around the joints of the oligomers act as flexible hinges, which explains the efficiency of TFAM in compacting DNA. Since mitochondrial polymerase (mitoRNAP) is involved in melting the transcription bubble, TFAM may use the same allosteric interaction to both recruit mitoRNAP and initiate transcription.