Christian Poitras

Activation-induced deaminase (AID) drives antibody diversification through class-switch recombination and somatic hypermutation of the immunoglobulin (Ig) genes, but its off-target mutagenic activity contributes to B cell lymphoma^1,2,3,4,5. How AID selectively mutates Ig loci and a restricted set of other genes remains unknown. Transcription is required for AID activity, but most transcribed genes are not mutated^1,5,6,7, and AID occupies more loci than it mutates^7,8,9. Here we identify the super elongation complex histone readers MLLT1 and MLLT3 as determinants of selective AID activity. Combined loss of MLLT1 and MLLT3 abolishes all AID-dependent mutagenic processes, recapitulating AID deficiency. AID-mutated genomic regions in mouse and human B cells are precisely marked by high MLLT1 and MLLT3 occupancy (MLLT1/MLLT3^high). Transcriptional changes after deleting both readers are modest and cannot explain the AID activity loss. Mechanistically, MLLT1 and MLLT3 are dispensable for global AID chromatin tethering but locally enrich AID downstream from promoter regions. Both readers bind to AID and act redundantly, but MLLT1 has a dominant role in mice through its intrinsically disordered region, which promotes condensates that selectively concentrate AID. Fusing AID to MLLT1 or MLLT3 is sufficient to restore class-switch recombination and mutagenesis in Mllt1^−/−Mllt3^−/− cells. These findings reveal that local MLLT1/MLLT3^high-dependent enrichment licenses AID at a restricted subset of genomic regions by spatially confining and concentrating its activity, probably through condensate formation.

TFIIH is an essential transcription initiation factor for RNA polymerase II (RNApII). This multi-subunit complex comprises two modules that are physically linked in Saccharomyces cerevisiae by the subunit Tfb3 (MAT1 in metazoans). The Core Module, with two DNA-dependent ATPases and several additional subunits, promotes DNA unwinding. The Kinase Module phosphorylates the C-terminal domain (CTD) of RNApII subunit Rpb1, initiating a cycle of CTD modifications that coordinate the exchange of initiation and elongation factors. Why these two disparate activities are bundled into one factor is not obvious, but the connection may provide temporal coordination during early initiation. When Tfb3 is split into two parts to uncouple the TFIIH modules, the resulting cells are viable but grow very slowly. Chromatin immunoprecipitation of the split TFIIH shows that the Core Module, but not the Kinase, is properly recruited to promoters. Instead of the normal promoter-proximal peak, high CTD Serine 5 phosphorylation is seen throughout transcribed regions. Therefore, coupling the TFIIH modules is necessary to localize and limit CTD kinase activity to early stages of transcription. These results are consistent with the idea that the two TFIIH modules began as independent functional entities that later became connected by Tfb3 during early eukaryotic evolution.

The developing limb is a classical model system for studying genetic aspects of pattern formation and mechanisms underlying tissue differentiation. The transition from early to differentiated cell fates during limb bud development requires the precise spatial and temporal regulation of gene activation and silencing programs. However, how transcriptional silencing of early genes is achieved remains poorly understood. Here, we investigate the role of the Polycomb Repressive Complexes (PRC1 and PRC2) in this transcriptional transition in the developing mouse forelimb. We show that the progression from early (E10) to late (E12.5 distal) forelimb stages is associated with shifts in promoter-proximal SUZ12 occupancy at early developmental genes. Conditional inactivation of Eed results in derepression and ectopic distal expression of early genes, which occurs despite the persistence of RING1B and H2AK119Ub. We further identified that SUZ12/RING1B co-occupied targets were also bound by RYBP-containing variant PRC1, indicating parallel recruitment of canonical and variant PRC1. Functional analyses of the combined inactivation of EED and RING1B reveal severe limb malformations and significant gene derepression, exceeding the effects of single EED or RING1B inactivation. Together, our findings demonstrate that the functional convergence of canonical and variant PRC1 at key developmental loci is required to maintain robust transcriptional silencing and regulate transitions in gene expression programs required for proper limb morphogenesis.

TFIIH is an essential transcription initiation factor for RNA polymerase II (RNApII). This multi-subunit complex comprises two modules that are physically linked in S. cerevisiae by the subunit Tfb3 (MAT1 in metazoans). The Core Module, with two DNA-dependent ATPases and several additional subunits, promotes DNA unwinding. The Kinase Module phosphorylates the C-terminal domain (CTD) of RNApII subunit Rpb1, initiating a cycle of CTD modifications that coordinate exchange of initiation and elongation factors. Why these two disparate activities are bundled into one factor is not obvious, but the connection may provide temporal coordination during early initiation. When Tfb3 is split into two parts to uncouple the TFIIH modules, the resulting cells are viable but grow very slowly. Chromatin immunoprecipitation of the split TFIIH shows that the Core Module, but not the Kinase, is properly recruited to promoters. Instead of the normal promoter-proximal peak, high CTD Serine 5 phosphorylation is seen throughout transcribed regions. Therefore, coupling the TFIIH modules is necessary to localize and limit CTD kinase activity to early stages of transcription. These results are consistent with the idea that the two TFIIH modules began as independent functional entities that later became connected by Tfb3 during early eukaryotic evolution.

Understanding how viruses engage host cell surfaces is fundamental to infection biology and therapeutic development. While vaccines remain central to prevention, recent global crises have emphasized the need for complementary antiviral strategies that can be mobilized rapidly against both known and emerging pathogens. In this context, artificial intelligence (AI) systems for biomolecular structure prediction, culminating in AlphaFold 3, are reshaping what is experimentally and conceptually achievable. Here, we present “Interactys-AI”, a framework designed to exploit AI-based structural modeling to systematically map virus–host protein–protein interactions (PPIs) and connect them to actionable drug repurposing opportunities. Beyond a technical workflow, Interactys-AI reflects a broader transformation toward predictive and anticipatory antiviral discovery. We describe the conceptual foundations of the platform, its implementation, and its application to influenza A H5N1 hemagglutinin. We further discuss how structural AI may redefine preparedness strategies, highlight current limitations, and outline future directions toward real-time therapeutic hypothesis generation.

Activation-induced deaminase (AID) mutates immunoglobulin genes to initiate antibody diversification by class switch recombination and somatic hypermutation but can also target non-immunoglobulin loci with oncogenic consequences. The mechanisms determining gene susceptibility to AID remain unclear. Here, we show that the H3K79 histone methyltransferase DOT1L is proximal to nuclear AID and promotes both class switch recombination and off-target AID activity, including Igh-cMyc translocations in mouse B cells. AID-mutated genes display high DOT1L activity. In the absence of DOT1L, nascent transcription largely increases despite reduced RNA Polymerase II (RNAPII) occupancy. Integrative genomic analyses reveal that DOT1L locally restricts transcription elongation velocity proportionally to H3K79me2/3 levels and extends RNAPII pausing. These transcriptional conditions enhance AID occupancy and thereby its activity. Our findings provide a harmonizing explanation for the bidirectional gene expression changes observed in DOT1L-deficient cells, and link attenuated transcriptional elongation velocity and prolonged RNAPII pausing to productive AID engagement at target loci.

H5N1 is a highly pathogenic avian influenza virus of major global concern. Since 2023, it has circulated widely among wild and farmed birds, with increasing spillover into mammals, including minks, seals, and cattle, and sporadic infections in humans in Chile, the UK, and the USA. The risk of a future pandemic is considered high because ongoing viral evolution could enable efficient human-to-human transmission. The hemagglutinin (HA) glycoprotein is the principal determinant of host range, mediating viral attachment and entry through interactions with sialylated glycans and potentially additional host surface proteins. Here, we developed an artificial intelligence (AI)-based pipeline integrating structural modeling, protein–protein interaction prediction, and biological filtering to identify human cell surface proteins with high likelihood of interacting with H5N1 HA. These interactions may contribute to viral entry and tropism and therefore represent promising candidates for experimental validation and therapeutic targeting. Our findings highlight the utility of AI-driven pipelines in accelerating the discovery of host factors relevant to pandemic influenza viruses.

Cell-based interception and precision medicine is a novel approach aimed at improving healthcare through the early detection and treatment of diseased cells. Here, we describe our recent progress towards developing cell-based interception and precision medicine to detect, understand, and advance the development of novel therapeutic approaches through a single-cell omics and drug screening platform, as part of a multi-laboratory collaborative effort, for a group of neurodegenerative disorders named leukodystrophies. Our strategy aims at the identification of diseased cells as early as possible to intercept progression of the disease prior to severe clinical impairment and irreversible tissue damage.

Activation induced deaminase (AID) initiates class switch recombination (CSR) and somatic hypermutation of the antibody genes but also mutates “off-target” genes, with oncogenic consequences. Understanding the mechanisms that render such loci susceptible to AID is therefore necessary. We identify the H3K79 histone methyltransferase DOT1L as a nuclear AID-proximal factor required for CSR and AID off-target activity, including IgH-cMyc translocations. Mechanistically, DOT1L limits promoter-proximal transcriptional pause release. DOT1L inactivation reduced promoter-proximal paused RNA polymerase II (RNAPII) occupancy, while concomitantly increasing RNAPII Serine 2 phosphorylation and reducing cohesin at gene bodies. These changes were proportional to the amount of H3K79me3 lost and perturbed transcription elongation, as evidenced by downregulation of highly expressed long genes but upregulation of shorter primed genes. Thus, DOT1L restricting promoter-proximal pause release can explain the hitherto puzzling predominance of gene upregulation in DOT1L-deficient mammalian cells and establishes regulated pause release as a critical step for AID activity.