Florigen, encoded by FT genes, is synthesized in leaves and transported to the shoot apical meristem (SAM) to induce flower development1-3. At the SAM, 14-3-3 proteins are proposed to act as receptors for FT protein and to mediate the indirect interaction between FT and the basic leucine zipper (bZIP) transcription factor FD to form the florigen activation complex (FAC) that activates transcription of flowering genes4-6. Here we demonstrate a different mechanism of FAC assembly, diverse functions for the 14-3-3 proteins within the complex, and an unexpected spatiotemporal distribution of the FAC. We show that FT is not recruited by 14-3-3 alone, but that it interacts with the DNA-FD-14-3-3 complex through two interfaces, one of which binds DNA via the unstructured C terminus of FT. We also find that interaction of 14-3-3 proteins with the C terminus of phosphorylated FD reduces liquid phase condensation of the intrinsically disordered FD protein, allowing it to bind DNA, and that the 14-3-3 proteins strengthen DNA binding of FD by promoting dimerization, which ultimately results in the recruitment of FT. Unexpectedly, we also find that after FT movement to the shoot apex, FT and FD are co-transcribed in young floral primordia, forming a boundary with the suppressed bract and allowing formation of the FAC during the first stages of floral development. Our studies propose a new mechanism by which the florigen FT transcriptional complex is formed, and indicate distinct functions for the complex during SAM and floral primordium development.

The synthesis of enantiopure materials is vital for pharmaceutical and agrochemical industries owing to the inherently chiral nature of biological systems and the fact that two enantiomers can have markedly different biochemical properties1. In particular, enantioselective preparation of atropisomers is of great interest owing to their privileged status as chiral ligands and pharmacophores2-4. Although chromatographic- or crystallization-based methods are commonly used to separate atropisomers, we urgently need more efficient and economical approaches to access enantioenriched atropisomers5,6. The use of stereoconvergent methods to access molecules with point chirality is well established but we have not tapped the potential of stereoconvergent catalytic methods to arrive at enantioenriched atropisomers. Here we report deracemization activity of a P450 enzyme and explore its ability to deliver a stereoconvergent route towards enantioenriched atropisomers. Using a curated set of P450 variants, we found that a wide variety of symmetric and non-symmetrically substituted 2,2'-binaphthol (BINOL) building blocks can be deracemized to high enantiomeric purity. This deracemization activity is mechanistically distinct from the activity of previously reported P450 enzymes, which operate through enantioselective bond formation to afford enantioenriched atropisomers. By contrast, the deracemization process reported here is proposed to proceed through bond rotation. As engineered variants have complementary selectivity profiles and substrate scope, this biocatalytic platform should be readily tunable for any desired substitution pattern. We anticipate that these results will inspire new stereoconvergent approaches to synthesizing conformationally stable atropisomers.

Discoveries of fundamental limits for the rates of physical processes, from the speed of light to the Lieb-Robinson bound for information propagation1,2, often lead to breakthroughs in the understanding of the underlying physics. Here we observe such a limit for a paradigmatic many-body phenomenon, the spreading of coherence during the formation of a weakly interacting Bose-Einstein condensate3-18. We study condensate formation in an isolated homogeneous atomic gas19,20 that is initially far from equilibrium, in an incoherent low-energy state, and condenses as it relaxes towards equilibrium. Tuning the interatomic interactions that drive condensation, we show that the spreading of coherence through the system is initially slower for weaker interactions and faster for stronger ones, but always eventually reaches the same limit, at which the square of the coherence length grows at a universal rate given by the ratio of Planck's constant and the particle mass, or, equivalently, by the quantum of velocity circulation associated with a quantum vortex. These observations are robust to changes in the initial state, the gas density, and the system size. Our results provide benchmarks for theories of universality far from equilibrium21-34, are relevant for quantum technologies that rely on large-scale coherence, and invite similar measurements in other systems.

Rainfall and flooding frequently disrupt the lives of urban residents worldwide, posing substantial public health risks1,2. Rapid urbanization is exposing larger and more vulnerable populations to flooding3, while climate change intensifies rainfall patterns4,5 and rising sea levels impair drainage systems6-11. Despite the growing recognition and urgency of these hazards, the health impacts of rainfall remain poorly understood, and those of sea-level rise are entirely unquantified. Here we estimate the mortality consequences of rainfall in one of the world's largest cities-Mumbai, India. We integrate high-resolution data on rainfall, tides and mortality to analyse how unmanaged rainfall, and its interaction with tidal dynamics, contribute to urban health risks. We find that rainfall causes more than 8% of Mumbai's deaths during the monsoon season, and that more than 80% of this burden is borne by slum residents. Children face the biggest increase in mortality risk from rainfall, and women face a greater risk than men. We also demonstrate that mortality risk from rainfall increases sharply during high tides and use this relationship to evaluate how rising sea levels could amplify rainfall-induced mortality in the absence of adaptation. Our findings reveal that the mortality impacts of rainfall are an order of magnitude larger than is documented by official statistics12,13, highlighting the urgent need for investment in improved drainage, sanitation and waste-management infrastructure.

Gene flow between biological species is a common and often adaptive evolutionary phenomenon throughout the tree of life1-4. Given the pervasive nature of genetic exchange, a daunting challenge is how best to infer the correct relationships between species from the complex collection of histories arrayed across genomes5. The local rate of meiotic recombination influences the distribution of signatures of, and barriers to, gene flow during the early stages of speciation6. Still, a broader understanding of this relationship and its application to accurately discerning phylogeny is lacking due to a scarcity of recombination maps. Here we applied deep learning methods to genome alignments from 22 divergent placental mammal species to infer the evolution of the recombination landscape. We identified a large and evolutionarily conserved X-linked recombination desert constituting 30% of the chromosome. Recombination-aware phylogenomic analyses from 94 species revealed that the X-linked recombination desert is an ancient and recurrent barrier to gene flow and retains the species history when introgression dominates genome-wide ancestry. The functional basis for this supergene is manifold, enriched with genes that influence sex chromosome silencing and reproduction traits. Because the locus underpins reproductive isolation across ordinal lineages, it may represent a reliable marker for resolving challenging relationships across the mammalian phylogeny.

Chirality is central to life, and controlling the formation of one of a pair of mirror-image molecules (enantiomers) is a central tenet of synthetic chemistry. Although controlling stereogenic carbon1-3, silicon4,5, phosphorus6,7 and sulfur8,9 centres is commonplace, nitrogen centres in amines are not typically stable. Limited achievements in the enantioselective construction of nitrogen chirality have primarily been established in quaternary ammonium salts10-12 and bridged bicyclic amines13-17, which have a restricted pyramidal configuration. The asymmetric synthesis of non-bridged pyramidal nitrogen-chirogenic compounds suffers from a super-stoichiometric chiral source and exhibits poor stereoselectivity18-24. Here we present a catalytic enantioselective strategy for construction of acyclic nitrogen stereocentres via a chiral Brønsted acid-catalysed chlorination reaction. We designed a stereospecific intramolecular reaction to overcome the structural and configurational instabilities of nitrogen-chlorinated hydroxylamines. The resulting 2-alkoxy-1,2-oxazolidines showed good enantiopurities, and density functional theory calculations confirmed successful enantiocontrol of nitrogen chirality during the chlorination process. Furthermore, this strategy has been applied successfully to synthesize the enantioselective N-chloroaziridines with a configurationally stable nitrogen stereogenic centre. Control experiments provide evidence for an SN2 pathway for the intramolecular nucleophilic substitution event.

Necroptosis is a form of lytic cell death that is overactivated during infections and in inflammatory pathologies1. NINJ1 was recently found to be a mediator of plasma membrane rupture (PMR) during pyroptosis, toxin-induced necrosis, apoptosis, and ferroptosis2,3, but the mediator of PMR during necroptotic cell death remained unknown. Here, using a CRISPR-Cas9-based genome-wide knockout approach, we identify SIGLEC12 as a key mediator of necroptosis downstream of MLKL at the PMR step. Cells with knockdown or knockout of SIGLEC12 are defective in necroptosis-induced PMR and demonstrate ballooning morphology. During necroptosis, SIGLEC12 undergoes dephosphorylation, interacts with MLKL, forms cytosolic puncta and assembles into fibrils. Notably, SIGLEC12 is cleaved by TMPRSS4 during necroptosis to produce a 20-kDa fragment highly homologous to NINJ1, and this cleavage event is required and sufficient to induce PMR during necroptosis. A SIGLEC12 variant associated with cancer (Ser458Phe) and a variant found in the general human population (Arg528Trp) attenuate SIGLEC12 cleavage by TMPRSS4. Knockout of Siglec12 in mouse cells does not affect PMR, suggesting a species-specific role. Our identification of SIGLEC12 as a mediator of PMR expands our understanding of how programmed necrosis is executed and offers new approaches for targeting this proinflammatory form of cell death in human diseases.

Marburg virus (MARV) is a filovirus that causes a severe and often lethal hemorrhagic fever1,2. Despite the increasing frequency of MARV outbreaks, no vaccines or therapeutics are licensed for use in humans. Here, we designed mutations that improve the expression, thermostability, and immunogenicity of the prefusion MARV glycoprotein (GP) ectodomain trimer, which is the sole target of neutralizing antibodies and vaccines in development3-8. We discovered a fully human, pan-marburgvirus monoclonal antibody, MARV16, that broadly neutralizes all MARV isolates as well as Ravn virus and Dehong virus with 40 to 100-fold increased potency relative to previously described antibodies9. Moreover, MARV16 provides therapeutic protection in guinea pigs challenged with MARV. We determined a cryo-electron microscopy structure of MARV16-bound MARV GP showing that MARV16 recognizes a prefusion-specific epitope spanning GP1 and GP2, blocking receptor binding and preventing conformational changes required for viral entry. We further reveal the architecture of the MARV GP glycan cap, which shields the receptor binding site (RBS), underscoring architectural similarities with distantly related filovirus GPs. MARV16 and previously identified RBS-directed antibodies9-11 can bind MARV GP simultaneously. These antibody cocktails require multiple mutations to escape neutralization by both antibodies, paving the way for MARV therapeutics resilient to viral evolution. MARV GP stabilization along with the discovery of MARV16 advance prevention and treatment options for MARV.

A long-standing goal of artificial intelligence is to build systems capable of complex reasoning in vast domains, a task epitomized by mathematics with its boundless concepts and demand for rigorous proof. Recent AI systems, often reliant on human data, typically lack the formal verification necessary to guarantee correctness. By contrast, formal languages such as Lean1 offer an interactive environment that grounds reasoning, and reinforcement learning (RL) provides a mechanism for learning in such environments. We present AlphaProof, an AlphaZero-inspired2 agent that learns to find formal proofs through RL by training on millions of auto-formalized problems. For the most difficult problems, it uses Test-Time RL, a method of generating and learning from millions of related problem variants at inference time to enable deep, problem-specific adaptation. AlphaProof substantially improves state-of-the-art results on historical mathematics competition problems. At the 2024 IMO competition, our AI system, with AlphaProof as its core reasoning engine, solved three out of the five non-geometry problems, including the competition's most difficult problem. Combined with AlphaGeometry 23, this performance, achieved with multi-day computation, resulted in reaching a score equivalent to that of a silver medallist, marking the first time an AI system achieved any medal-level performance. Our work demonstrates that learning at scale from grounded experience produces agents with complex mathematical reasoning strategies, paving the way for a reliable AI tool in complex mathematical problem-solving.

Despite the adoption of the Paris Agreement ten years ago, fossil CO2 emissions continue to rise, pushing atmospheric CO2 levels to 423 ppm in 2024 and driving human-induced warming to 1.36°C, within years of breaching the 1.5°C limit 1,2. Accurate reporting of anthropogenic and natural CO2 sources and sinks is a prerequisite to tracking the effectiveness of climate policy and detecting carbon sink responses to climate change. Yet notable mismatches between reported emissions and sinks have so far prevented confident interpretation of their trends and drivers 1. Here, we present and integrate recent advances in observations and process understanding to address some long-standing issues in the global carbon budget estimates. We show that the magnitude of the natural land sink is substantially smaller than previously estimated, while net emissions from anthropogenic land-use change are revised upwards 1. The ocean sink is 15% larger than the land sink, consistent with new evidence from oceanic and atmospheric observations 3,4. Climate change reduces the efficiency of the sinks, particularly on land, contributing 8.3 ± 1.4 ppm to the atmospheric CO2 increase since 1960. The combined effects of climate change and deforestation turn Southeast Asian and large parts of South American tropical forests from CO2 sinks to sources. This underscores the need to halt deforestation and limit warming to prevent further loss of carbon stored on land. Improved confidence in assessments of CO2 sources and sinks is fundamental for effective climate policy.

It has long been proposed1, and then observed2,3, that faster rates of soil production occur beneath thinner soils. It remains uncertain, however, whether soil thickness is the driving variable regulating production from the top-down2-6, or whether soil thickness is simply responding to changes in bedrock weathering controlled from the bottom-up7,8. Answering this question is difficult because the feedbacks between soil production and soil erosion, the processes that jointly govern soil thickness2,9,10, respond to perturbations on timescales of thousands to millions of years11,12, timescales that are too long for scientists to observe directly. Here we leverage a space-for-time substitution at a transient mountain range along the San Andreas Fault13-15, where the remarkable tectonic setting allows for independent quantification of uplift, soil production and erosion. We show that, following a pulse of tectonic uplift, the conversion of rock to soil accelerates before the overlying soils thin, but at the same time that topographic stresses increase7 and the rock weakens16. This observation challenges the long-standing assumption that soil production rates are controlled predominantly by soil thickness1,2, and instead lends evidence for a bottom-up, rock strength control on soil production.

The challenges associated with the transition of life from water to land are profound1, yet they have been met in many distinct animal lineages2-5. These constitute a series of independent evolutionary experiments from which we can decipher the role of contingency versus convergence in the adaptation of animal genomes. Here we compare 154 genomes from 21 animal phyla and their outgroups to reconstruct the protein-coding content of the ancestral genomes linked to 11 animal terrestrialization events, and to produce a timescale of terrestrialization. We uncover distinct patterns of gene gain and loss underlying each transition to land, but similar biological functions emerged recurrently pointing to specific adaptations as key to life on land. We show that semi-terrestrial species evolved convergent functional patterns, in contrast with fully terrestrial lineages that followed different paths to land. Our timeline supports three temporal windows of land colonization by animals during the last 487 million years, each associated with specific ecological contexts. Although each lineage exhibits distinct adaptations, there is strong evidence of convergent genome evolution across the animal kingdom suggesting that, in large part, adaptation to life on land is predictable, linking genes to ecosystems.

Acetyl-coenzyme A (AcCoA) sits at the nexus of nutrient metabolism and shuttles between the canonical and non-canonical tricarboxylic acid cycle1,2, which is dynamically regulated by nutritional status, such as fasting3. Here we find that mitophagy is triggered after a reduction in cytosolic AcCoA levels through short-term fasting and through inhibition of ATP-citrate lyase (encoded by ACLY), mitochondrial citrate/malate antiporter (encoded by SLC25A1) or acyl-CoA synthetase short chain family member 2 (encoded by ACSS2), and the mitophagy can be counteracted by acetate supplementation. Notably, NOD-like receptor (NLR) family member X1 (NLRX1) mediates this effect. Disrupting NLRX1 abolishes cytosolic AcCoA reduction-induced mitophagy both in vitro and in vivo. Mechanically, the mitochondria outer-membrane-localized NLRX1 directly binds to cytosolic AcCoA within a conserved pocket on its leucine-rich repeat (LRR) domain. Moreover, AcCoA binds to the LRR domain and enhances its interaction with the nucleotide-binding and oligomerization (NACHT) domain, which helps to maintain NLRX1 in an autoinhibited state and prevents the association between NLRX1 and light chain 3 (LC3). Furthermore, we find that the AcCoA-NLRX1 axis underlies the KRAS-inhibitor-induced mitophagy response and promotes drug resistance, providing a metabolic mechanism of KRAS inhibitor resistance. Thus, cytosolic AcCoA is a signalling metabolite that connects metabolism to mitophagy through its receptor NLRX1.

Since late 2021, a panzootic of highly pathogenic H5N1 has devastated wild birds, agriculture and mammals. Here an analysis of 1,818 haemagglutinin sequences from wild birds, domestic birds and mammals reveals that the North American panzootic was driven by around nine introductions into the Atlantic and Pacific flyways, followed by rapid dissemination through wild, migratory birds. Transmission was primarily driven by Anseriformes, while non-canonical species acted as dead-end hosts. In contrast to the epizootic of 2015 (refs. 1,2), outbreaks in domestic birds were driven by around 46-113 independent introductions from wild birds that persisted for up to 6 months. Backyard birds were infected around 9 days earlier on average than commercial poultry, suggesting potential as early-warning signals for transmission upticks. We pinpoint wild birds as critical drivers of the epizootic, implying that enhanced surveillance in wild birds and strategies that reduce transmission at the wild-agriculture interface will be key for future tracking and outbreak prevention.

Crohn's disease often presents with fistulae, abnormal tunnels that connect the intestine to the skin or other organs. Despite their profound effect on morbidity, the molecular basis of fistula formation remains unclear, largely owing to the challenge of capturing intact fistula tracts and their inherent heterogeneity1-3. Here we construct a subcellular-resolution spatial atlas of 68 intestinal fistulae spanning diverse anatomical locations. We describe fistula-associated epithelial, immune and stromal cell states, revealing abnormal zonation of growth factors and morphogens linked to establishment of tunnelling anatomy. We identify fistula-associated stromal (FAS) fibroblasts, which are assembled in concentric layers: a proliferative, lumen-adjacent zone beneath neutrophil and macrophage-rich granulation tissue, an active lesion core of FAS cells and a quiescent, pro-fibrotic outer zone. We examine the architecture of the extracellular matrix in the fistula tract and demonstrate that FAS populations associate with distinct collagen structures, exhibiting properties ranging from proliferation, migration and extracellular matrix remodelling to dense collagen deposition and fibrosis. We define niches supporting epithelialization of fistula tunnels and a FAS-like population that is detected at the base of ulcers in non-penetrating Crohn's disease. Our study demonstrates that common molecular pathways and cellular niches underpin fistulae across intestinal locations, revealing the cellular protagonists of fistula establishment and persistence. This resource will inform the development of model systems and interventions to mitigate aberrant fibroblast activity while preserving their regenerative properties in Crohn's disease.

Coronal mass ejections (CMEs) are massive expulsions of magnetized plasma from a star and are the largest contributors to space weather in the Solar System1,2. CMEs play an important role in planetary atmospheric erosion, especially for planets that are close to their host star3-5. However, this conclusion remains controversial as there has not been an unambiguous detection of a CME from a star outside our Sun. Previous stellar CME studies have only inferred the presence of a CME through the detection of other types of stellar eruptive event6-9. A signature of a fast CME is a type II radio burst10,11, which is emitted from the shock wave produced as the CME travels through the stellar corona into interplanetary space. Here we report an analogue to a type II burst from the early M dwarf StKM 1-1262. The burst exhibits identical frequency, time and polarization properties to fundamental plasma emission from a solar type II burst. We demonstrate that the rate of these events with similar radio luminosity from M dwarfs is 0.84-0.69+1.94×10-3 per day per star. Our detection implies that we are no longer restricted to extrapolating the solar CME kinematics and rates to other stars, allowing us to establish observational limits on the impact of CMEs on exoplanets.

Attention deficit hyperactivity disorder (ADHD) is a childhood-onset neurodevelopmental disorder with a large genetic component1. It affects around 5% of children and 2.5% of adults2, and is associated with several severe outcomes3-11. Common genetic variants associated with the disorder have been identified12,13, but the role of rare variants in ADHD is mostly unknown. Here, by analysing rare coding variants in exome-sequencing data from 8,895 individuals with ADHD and 53,780 control individuals, we identify three genes (MAP1A, ANO8 and ANK2; P < 3.07 × 10-6; odds ratios 5.55-15.13) that are implicated in ADHD. The protein-protein interaction networks of these three genes were enriched for rare-variant risk genes of other neurodevelopmental disorders, and for genes involved in cytoskeleton organization, synapse function and RNA processing. Top associated rare-variant risk genes showed increased expression across pre- and postnatal brain developmental stages and in several neuronal cell types, including GABAergic (γ-aminobutyric-acid-producing) and dopaminergic neurons. Deleterious variants were associated with lower socioeconomic status and lower levels of education in individuals with ADHD, and a decrease of 2.25 intelligence quotient (IQ) points per rare deleterious variant in a sample of adults with ADHD (n = 962). Individuals with ADHD and intellectual disability showed an increased load of rare variants overall, whereas other psychiatric comorbidities had an increased load only for specific gene sets associated with those comorbidities. This suggests that psychiatric comorbidity in ADHD is driven mainly by rare variants in specific genes, rather than by a general increased load across constrained genes.

Marine picocyanobacteria are abundant photosynthetic organisms of global importance. They coexist in the ocean with cyanophages-viruses that infect cyanobacteria. Cyanophages carry many auxiliary metabolic genes acquired from their hosts that are thought to redirect host metabolism for the phage's benefit1-5. One such gene is nblA, which is present in multiple cyanophage families2,6-8. Under nutrient deprivation cyanobacterial NblA is responsible for inducing proteolytic degradation of the phycobilisome9-11, the large cyanobacterial photosynthetic light-harvesting complex. This increases the pool of amino acids available for essential tasks11, serving as a survival mechanism12. Ectopic expression of different cyanophage nblA genes results in host pigment protein degradation6,8,13. However, the benefit of the virus-encoded NblA for cyanophages and the broader impact on the host are unclear. Here, using a recently developed genetic manipulation system for marine cyanophages14, we reveal that viral NblA significantly accelerates the cyanophage infection cycle, directs degradation of the host phycobilisome and other proteins, and reduces host photosynthetic light-harvesting efficiency. Metagenomic analysis revealed that cyanophages carrying nblA are widespread in the oceans and comprise 35% and 65% of oceanic T7-like cyanophages in surface and deep photic zones, respectively. Our results show a large benefit of NblA to the cyanophage, while it exerts a negative effect on the host photosynthetic apparatus and host photosynthesis. These findings suggest that cyanophage NblA has an adverse global impact on light harvesting by oceanic picocyanobacteria.

The number of spatial omics technologies being developed is increasing1. However, a missing tool is one that can locate proteins in tissues in an untargeted manner at high spatial resolution and coverage. Here we present in situ imaging proteomics via expansion (iPEX), which integrates isotropic tissue magnification2 with matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging. iPEX provides scalable spatial resolution down to the micrometre scale and substantially increases the sensitivity of protein identification by 10-100-fold. Using the retina as a model, iPEX enabled the construction of spatial proteomic maps with high precision, the visualization of single-cell layers and extrasomatic structures and the identification of colocalized proteins. iPEX was readily applied to diverse tissues, including brain, intestine, liver and organoids, detecting 600-1,500 proteins at 1-5-µm effective pixel size. The application of iPEX to depict spatial proteomic maps in brains of mice with 5xFAD Alzheimer's disease revealed an early-onset mitochondrial aberrancy. Notably, in young mice, the peroxisomal acetyl-CoA acyltransferase ACAA1A-of which the N392S mutant is a monogenic risk factor in Alzheimer's disease3-was downregulated. ACAA1 depletion blocked the biosynthesis of long-chain polyunsaturated fatty acids, including docosahexaenoic acid, in multiple cellular contexts. These lipidome alterations were restored in cells overexpressing wild-type ACAA1 but not ACAA1(N392S), which suggests that the dysregulation of long-chain polyunsaturated fatty acids has an early role in neurodegeneration. Together, these results demonstrate that iPEX facilitates untargeted spatial proteomics at micrometre resolution for diverse applications.

Two-dimensional moiré materials are formed by artificially stacking atomically thin monolayers. Correlated and topological quantum phases can be engineered by precise choice of stacking geometry1-3. These designer electronic properties depend crucially on interlayer coupling and atomic registry4,5. An open question is how the atomic registry responds on ultrafast timescales to optical excitation and whether the moiré geometry can be dynamically reconfigured to tune emergent phenomena in real time. Here we show that femtosecond photoexcitation drives a coherent twist-untwist motion of the moiré superlattice in 2° and 57° twisted WSe2/MoSe2 heterobilayers, resolved directly by ultrafast electron diffraction. On above-band-gap photoexcitation, the moiré superlattice diffraction features are enhanced within 1 ps and subsequently suppressed several picoseconds after, deviating markedly from typical photoinduced lattice heating. Kinetic diffraction analysis, supported by simulations of the sample dynamics, indicates a peak-to-trough local twist angle modulation of 0.6°, correlated with a sub-THz frequency moiré phonon. This motion is driven by ultrafast charge transfer that transiently increases interlayer attraction. Our results could lead to ultrafast control of moiré periodic lattice distortions and, by extension, the local moiré potential that shapes excitons, polarons and correlation-driven behaviours.