\n \n Mycoplasma pneumoniae\n \n , responsible for approximately 30% of community-acquired human pneumonia, needs to extract lipids from the host environment for survival and proliferation. Here, we report a comprehensive structural and functional analysis of the previously uncharacterized protein P116 (MPN_213). Single-particle cryo-electron microscopy of P116 reveals a homodimer presenting a previously unseen fold, forming a huge hydrophobic cavity, which is fully accessible to solvent. Lipidomics analysis shows that P116 specifically acquires essential lipids such as phosphatidylcholine, sphingomyelin and cholesterol. Structures of different conformational states reveal the mechanism by which lipids are transported. This finding immediately suggests a way to control Mycoplasma infection by interfering with lipid uptake.\n
\n\n \n Mycoplasma pneumoniae\n \n is a facultative intracellular human pathogen causing community-acquired pneumonia that can manifest severe systemic effects (\n \n 1\n \n ). Unlike other respiratory pathogens,\n \n M. pneumoniae\n \n has no approved vaccine (\n \n 2\n \n ).\n \n Mycoplasmas\n \n lack a cell wall and have the smallest known genomes (\n \n 3\n \n ).\n \n M. pneumoniae\n \n , with a 816 kb genome, is a model organism for a minimal cell (\n \n 4\n \n ). Many of the metabolic pathways required to synthesize essential products are absent, which makes an uptake by specialized mechanisms necessary. In fact,\n \n M. pneumoniae\n \n cannot synthesize several of the lipids that are important components of the cell membrane, such as sphingomyelin, phosphatidylcholine and cholesterol (\n \n 5\n \n ). Instead, it must take up lipids from the host environment and adapts its membrane composition depending on the medium in vitro (\n \n 6\n \n \u2013\n \n 8\n \n ). Cholesterol in particular, which is present in only a few prokaryotes, is essential for\n \n M. pneumoniae\n \n cells and several other\n \n Mycoplasma spp.\n \n (\n \n 6\n \n ). It is the most abundant lipid in the membranes, accounting for 35\u201350% of the total lipid fraction (\n \n 6\n \n ). To date, it is unclear how\n \n Mycoplasma spp.\n \n and other prokaryotic species achieve lipid uptake from the environment.\n
\n\n In this work, we report the structural and functional characterization of P116, a strongly immunogenic and essential protein for the viability of\n \n M. pneumoniae\n \n cells. P116 was previously uncharacterized, although it has been reported to potentially contribute to adhesion to host cells (\n \n 9\n \n ). Despite the essential role of P116 the\n \n M. pneumoniae\n \n genome contains only a single copy of\n \n p116\n \n (\n \n mpn213\n \n ). This is in contrast to the most immunogenic protein P1, which is not essential but contains multiple copies on the genome (\n \n 10\n \n ). To elucidate the role of P116, we first determined the structure of the ectodomain by single-particle cryo-electron microscopy (cryoEM). The structure has a novel fold (with no matches in the Protein Data Bank) featuring a uniquely large hydrophobic cavity that is fully accessible to solvent. Mass spectrometry and other analytical techniques identify ligands found in the cavity as several different lipids (incl. cholesterol), some of which are essential. Based on these findings, we describe the mechanism by which\n \n Mycoplasmas\n \n can extract lipids from the environment and possibly also deposit them in their own membrane, thus explaining the essential role of P116 for the survival of\n \n M. pneumoniae\n \n cells.\n
\n\n \n P116 is abundant on the cell surface\n \n
\n\n A construct predicted to span the whole ectodomain of P116 from\n \n M. pneumoniae\n \n (residues 30\u2013957)\n \n \n was overexpressed in\n \n Escherichia coli\n \n and purified by His-tag affinity and gel filtration chromatography (Materials and Methods and\n \n Supplementary Figure 1\n \n ). Immunolabeling with both polyclonal and monoclonal antibodies against this construct showed an intense and uniform distribution of labeling across the whole surface of\n \n M. pneumoniae\n \n cells (\n \n Figure 1a\n \n ), with adhesion and motility unaffected by the antibodies (\n \n Supplementary Table I and Supplementary Movie 1-3\n \n ). This distribution contrasts with that of P1, an adhesion protein that concentrates at the tip of the cell and has strong effects on adhesion and motility (\n \n 11, 12\n \n ).\n
\n\n \n P116 has a novel fold with a lipid-accessible cavity\n \n
\n\n The structure of P116 (30\u2013957) was determined by single-particle cryoEM at 3.3 A\u030a resolution (according to the gold standard criterion of FSC 0.143;\n \n Supplementary Table II\n \n ,\n \n Supplementary Figure 2\n \n ). It is an elongated homodimer of ~240 \u00c5 along its longest axis, which adopts an arched shape with an arc-radius of 200 \u00c5 (\n \n Figure 1b\n \n ,\n \n Supplementary Movies 4, 5\n \n ). Each monomer consists of two distinct subunits: A N-terminal domain (residues 60\u2013245), situated distal to the dimer axis, and a core domain (residues 246\u2013867). Proximal to the dimer axis is the dimerization interface (\n \n Figure 1b\n \n ,\n \n Supplementary Figure 3\n \n ), which is very well resolved. In addition, the N-terminal domain has significant hinge mobility with respect to the core domain, which reduced the local resolution of the cryoEM map (\n \n Supplementary Figure 2\n \n ), making model building difficult for the most distal parts of the construct (see Materials and Methods and\n \n Supplementary Figure 4\n \n ). The homodimer displays significant flexibility with many vibrational modes, as classification illustrates (\n \n Supplementary Figure 5\n \n ). Finally, some residues at the N- and C-termini of the construct (30\u201359 and 868\u2013957, respectively) were not visible in the cryoEM maps. The flexibility of the homodimer involves a change in the curvature of approximately 100 \u00c5, wringing along the axis perpendicular to the dimer axis by ~80 degrees, and bending up to 20 degrees (\n \n Supplementary Figure 5, Supplementary Movie 6\n \n ).\n
\n\n The core domain resembles a half-opened left hand, with four contiguous antiparallel \u03b1-helices corresponding to the four fingers and the N-terminal domain the thumb (\n \n Figure 2a\n \n ). The helices corresponding to the wrist form the dimer interface, and a conserved tryptophan residue (Trp681) interacts tightly with the neighboring monomer. In the variant Trp681Ala, the rate of dimers to monomers is 1:4, compared to only dimers without the mutation (\n \n Supplementary Figure 3b\n \n ). The palm of the hand includes a long and well defined central \u03b1-helix, the bridge helix (residues 268\u2013304), and a rigid \u03b2-sheet of five antiparallel strands that extends to the N-terminal domain (\n \n Figure 2b\n \n ). The hand appears in a half-opened state with a large elongated cleft across the whole core domain (\n \n Figure 2c\n \n ). The inner part of the hand (i.e. the fingers and palm) forms a large cavity that measures 62 \u00c5 proximal to distal and 38 \u00c5 anterior to posterior with a volume of ~18,000 \u00c5\n \n 3\n \n . The cavity is completely hydrophobic although fully accessible to the solvent (\n \n Figure 2c\n \n ,\n \n Supplementary Movie 7)\n \n .\u00a0In addition, the core has two access points, one at the dorsal side and one at the distal side (\n \n Figure 3a\n \n ). Using the DALI server, we found only very weak structural relationships between P116 and all other experimentally determined protein structures available in the Protein Data Bank, which shows that P116 has a new, unique fold.\n
\n\n The N-terminal domain is compact and organized around a cluster of aromatic residues, at the center of which is the only tryptophan residue of the domain (Trp121). The N-terminal and core domains of P116 superimpose for 126 equivalent residues (68% of the N-terminal domain), suggesting that P116 might have been generated by duplication of an ancestor domain\n \n .\n \n The common secondary structural elements in the N-terminal and core domains consist of a \u03b2-sheet and the two helices preceding the sheet (\n \n Figure 2b\n \n ). The core domain is much larger than the N-terminal domain mainly due to two insertions containing twelve and four helices, respectively.\n
\n\n For the inner part of the P116 core domain,\n \n \n the cryoEM maps show prominent elongated densities (with a length of 10\u201319 \u00c5 and a width of 4 \u00c5) that fill most of the hydrophobic areas (\n \n Figure 3a\n \n ,\n \n Supplementary Movies 8, 9\n \n ). These elongated densities, which are unaccounted for, cannot be explained by the protein residues missing in the model. Instead, the mass excess of ~13 kDa, consistently measured by multiple angle light scattering (MALS) and mass spectrometry for P116 in different preparations, could be explained by the presence of ligand molecules bound to P116 (\n \n Figure 4a\n \n ). Initial mass spectrometry analysis of the same samples from which the structure of P116 was determined (see Materials and Methods) showed the presence of several lipid species, including phosphatidylcholine and sphingomyelin, which are essential for\n \n M. pneumoniae\n \n \n 3\n \n , and of wax esters (\n \n Figure 4b and Supplementary Figure 6\n \n ).\n
\n\n P116 orthologues were found in at least eight other\n \n Mycoplasma\n \n \n spp.\n \n including\n \n M. genitalium\n \n and\n \n M. gallisepticum\n \n . The amino acids lining the hydrophobic cavity are largely conserved (either identical or with similar characteristics) (\n \n Supplementary Figure 7a\n \n ). Modeling the orthologues of P116 with AlphaFold (\n \n 14\n \n ) results in all the models having a similar tertiary structure, in which a large core domain is flanked by a smaller N-terminal domain, but the relative position of the domains does not closely match the experimental structure (\n \n Supplementary Figure 7b\n \n ).\n
\n\n \n The conformation of empty P116 cannot accommodate lipid binding\n \n
\n\n To obtain \u2018empty\u2019 P116 that was free of any bound ligands, we treated the P116 samples with the detergent Triton-X 100 (see below and Materials and Methods). Mass spectrometry confirmed a massive reduction of lipids in the sample (\n \n Figure 4b)\n \n . The structure of the empty P116 sample was solved by cryoEM at 4 \u00c5 resolution (\n \n Supplementary Figure 8\n \n ). Its overall topology is almost identical to that of the original P116 sample, with the difference that the cavity is closed as a result of fingers 1, 2 and 3 being closer to the palm by 8, 13 and 12 \u00c5, respectively, and finger 4 moving 11 \u00c5 sideways to retain the distal core access to the palm (\n \n Figure 3b\n \n ,\n \n Supplementary Movies 10\n \n ,\n \n 11\n \n , and\n \n Supplementary Figure 9\n \n ). These changes reduce the volume within the core domain from ~18,000 \u00c5\n \n 3\n \n to ~6,300 \u00c5\n \n 3\n \n . The unoccupied volume between the fingers and palm reduces to two pockets that are large enough for lipids to pass through but appear unoccupied in the cryoEM density. A comparison of the filled and empty P116 structures shows that the original densities that were unaccounted for create massive steric clashes in the closed configuration of the fingers, demonstrating that the cavity can no longer accommodate lipids (\n \n Supplementary Movie 12\n \n ). In the empty P116, the dimerization interface is shifted towards the dorsal side of the molecule by 10 \u00c5, resulting in a contraction that changes the arc radius of the dimer from 500 to 600 \u00c5 and shifts the N-terminal domain towards the dimerization interface.\n
\n\n \n Refilled P116 is structurally identical to the purified sample\n \n
\n\n We next refilled the empty P116 samples by incubating them either with fetal bovine serum (FBS) or with high-density lipoproteins (HDL) and then re-purified them by affinity chromatography. Media containing FBS is a common growing broth for\n \n M. pneumoniae\n \n cultures, although lipoproteins, in particular HDL, are efficient substitutes for serum in mycoplasma culture media, likely because lipoproteins can provide the key lipids, in particular cholesterol, which is essential for mycoplasma cells (\n \n 15\n \n ). We solved the structure of the refilled P116 samples at 3.5 \u00c5 resolution using cryoEM. The structure of the refilled P116 is practically identical at 3.5 \u00c5 resolution to the structure of the original P116 sample, including densities at the palm of the hand that can be assigned to ligands. Mass spectrometry of the refilled samples shows the clear presence of lipids (\n \n Figure 4b\n \n ). Classes of subunits of the dimer show a wringing of ~80 degrees (\n \n Figure 3d\n \n ,\n \n Supplementary\n \n \n Figure 5\n \n and\n \n Supplementary Movie 6\n \n ).\n
\n\n \n P116 is conformationally flexible\n \n
\n\n In the original P116, empty P116 and refilled P116 samples, the structure appears predominantly as a homodimer. In all cases, the homodimer exhibits significant flexibility. Most prominently, the empty structure has a different arc radius than those of the original and refilled structures. In the original and refilled structures, a wringing motion is visible: each monomer is twisted in the opposite direction along the axis perpendicular to the dimer axis (\n \n Figure 3d\n \n ,\n \n Supplementary Movie 6\n \n and\n \n Supplementary Figure 5\n \n ). In all P116 structures, the N-terminal domain is the most flexible. Within the core domain, temperature factors are higher at the fingertips, indicating the movement of the antiparallel \u03b1-helices. When the fingers approach the palm, this results in a closing of the hand and a clash with the densities therein (\n \n Supplementary Movie 12\n \n ).\n
\n\n \n P116 ligands include essential lipids\n \n
\n\n We next set out to characterize the possible ligands within P116. We first measured the rate of radioactivity transfer to P116 after incubation with HDL particles containing either tritium-labeled cholesterol ([\n \n 3\n \n H]cholesterol) or tritium-labeled cholesteryl oleate as a representative of cholesterol esters (\n \n Table I\n \n ). A significant fraction of the HDL-[\n \n 3\n \n H]-radiotracer was detected in the post-incubated and purified P116 fractions, indicating a net transfer of both cholesterol and cholesterol ester between HDL and P116. The total absence of the most abundant HDL protein (APOA1), cross-checked by immune detection, verified that no HDL remnants had contaminated the purified P116 fractions. The highest rate of radiotracer transfer was achieved when [\n \n 3\n \n H]cholesterol-containing HDLs were mixed with empty P116. Transfer of [\n \n 3\n \n H]cholesterol was also present, although reduced, when the original P116 was incubated with labeled HDL. Transfer of [\n \n 3\n \n H]cholesterol esters to P116 would require a direct interaction between HDL and P116, as these esters are buried in the core of the HDL particles (\n \n Table I\n \n ). Passive cholesterol transport has been reported from cellular membranes to HDL or from LDL to HDL (\n \n 16\n \n ), but the concept that bacteria can exploit such a mechanism is completely new. The net flux of cholesterol is bidirectional and is governed by the cholesterol gradient between acceptor and donor molecules.\n
\n\n We then conducted a detailed liquid chromatography-coupled mass spectrometry (LC-MS) analysis. We identified more than 500 lipid species in the samples and found striking differences between the original, empty and refilled P116 samples (\n \n Figure 4c, 4d\n \n ). Characterization of the lipids in the original and refilled P116 samples showed the presence of phosphatidylcholine and sphingomyelin lipids, among others, which are essential for\n \n M. pneumoniae\n \n . While these analyses found wax esters in the original P116, far fewer were found in the refilled P116. Wax esters are not known to be required by\n \n M. pneumoniae,\n \n although some pathogenic bacteria use wax esters as a carbon source (\n \n 17, 18\n \n ). However, wax esters are part of the cultivation medium of the\n \n E. coli\n \n strain in which P116 was produced. These findings are in agreement with the fact that\n \n M. pneumoniae\n \n takes up and incorporates many lipid species and adapts its membrane composition to the available lipid spectrum. In the P116 samples refilled from FBS, a clear accumulation of the essential lipids phosphatidylcholine and sphingomyelin, as well as cholesterol molecules, can be seen (\n \n Figure 4c, 4d\n \n and\n \n Supplementary Table III\n \n ). These findings are in strong agreement with the functional data from the tritium-labeled cholesterol assay. Taken together, our lipidomics analyses revealed that P116 can bind to different lipid species. While in purified P116, PCs and SMs comprised only a small part of the present lipids, the most abundant species being PGs, PEs and wax esters, the refilled P116 preferentially bound to PCs, SMs, and cholesterols. Notably, the composition of the lipid species in the refilled P116 was strikingly different than the serum lipid distribution. For example, highly abundant uncharged TGs did not bind to P116. Thus, P116 although displaying a large bandwidth of lipid uptake, it does show a preference for selective lipid species (\n \n Figure 4c, 4d\n \n and\n \n Supplementary Table III\n \n ).\n
\n\n \n P116 binds HDL between its N-terminal and core domains\n \n
\n\n Next, we performed cryoEM on a sample containing empty P116 and HDL. Of ~46,000 particles that were identified as HDL, ~25,000 were attached to P116. The resulting density at a resolution of 9 \u00c5 shows P116 interacting directly with HDL at the region between the N-terminal domain and the core (\n \n Figure 4f\n \n ). In the reconstruction the density of P116 resembles the filled conformation, and the structure can be well fitted to the density map. Cryo-electron tomograms of whole\n \n M. pneumoniae\n \n cells indicate a similar arrangement of P116 with respect to the\n \n Mycoplasma\n \n membrane, although an unambiguous identification of the involved complexes is challenging due to the modest resolution (\n \n Supplementary Figure 10\n \n ).\n
\n\n P116 is essential for the viability of the human pathogen\n \n M. pneumoniae\n \n (\n \n 4\n \n ) and is the target of a strong antigenic response in infected patients (\n \n 19\n \n ). The P116 structure has a previously unseen fold with a uniquely large hydrophobic cavity filled with ligands. Mass spectrometry and radioactivity transfer experiments confirm a lipid extraction from serum (FBS) and HDL. Further, the ligands are identified as essential lipids for the survival of the cells. In fact, we found a high specificity towards Cholesterol, PCs and SMs, which represent the most abundant membrane lipids in\n \n M. pneumoniae\n \n (\n \n 8\n \n ). Crosslinking mass spectrometry studies indicate one weak aminoacid-pair interaction between P116 and MPN161 (a protein of unknown function) (\n \n 20\n \n ). Thus, while the involvement of other proteins in incorporating the extracted lipids into the\n \n Mycoplasma\n \n membrane cannot be excluded, it appears likely, given the observed conformational cycle upon lipid uptake, that P116 is also responsible for incorporation. Altogether, the P116 structure, along with our insights into different P116 conformations and the P116 complex formation with HDL, reveals a mechanism by which\n \n Mycoplasmas\n \n extract lipids from their environment and most likely incorporate them into their own membrane.\n
\n\n The transition from a full to an empty P116 molecule involves a\u2009~\u200970% volume reduction of the hydrophobic cavity in concert with a wringing motion of the core domains. During this wringing motion, in which the monomers are each twisted in the opposite direction around their long axis, the hydrophobic cavities face almost opposite directions. Since, the N-terminal domain is in close proximity to the C-terminus with which the protein is anchored in the\n \n Mycoplasma\n \n membrane in vivo, the core domain is the one experiencing the high flexibility seen in our data sets. This enables an alternating motion of the core domain, in which each time one monomer of the core domain faces the\n \n Mycoplasma\n \n membrane (i.e. the one transferring lipids to the membrane) and the other monomer faces the environment (i.e. the one extracting lipids from the environment). This wringing motion can be repeated in a continuous manner. In this way, P116 could undergo a rolling movement on the\n \n Mycoplasma\n \n membrane, thus facilitating the transport of cholesterol and other essential lipids in an apparently simple and newly discovered way for lipid transporters.\n
\n\n \n Mycoplasmas\n \n have a minimal genome and are capable of incorporating many different lipids into their membrane (\n \n 6\n \n ,\n \n 7\n \n ). The lipid-binding versatility shown by P116 enables a single molecular system to cope with the transport of diverse lipids required by\n \n Mycoplasmas\n \n . Although only\n \n Mycoplasmas\n \n share genes with similar sequences to P116, other microorganisms that require uptake of lipids from the environment, including clinically relevant bacterial species such as\n \n Borrelia burgdorferi\n \n may have similar -not yet discovered- systems to regulate their cholesterol homeostasis. Whether P116 shares functional similarities with other transfer proteins such as human cholesteryl ester transfer and phospholipid transfer proteins (\n \n 21\n \n ,\n \n 22\n \n ) requires further investigation. However, the diversity and amount of lipids that P116 can bind appear to be unmatched by any other known prokaryotic or eukaryotic lipid carrier. Interestingly, despite the broad lipid range P116 still shows a high specificity, largely enriching certain lipids (SM, PC and cholesterol) while excluding others (TGs).This new understanding of bacterial lipid uptake opens possibilities for treatment of mycoplasma infections and may, for the first time (\n \n 2\n \n ), allow the creation of a vaccine against\n \n Mycoplasma pneumoniae.\n \n
\n| \n \n \n \n Table I: Relative transfer of (esterified) cholesterol from HDL to P116\n \n \n \n | \n |||
| \n \n \n \n \n \n \n \n \n | \n \n \n \n % of [\n \n 3\n \n H]cholesterol transferred/mL\n \n \n | \n \n \n \n nmol\n \n \n\n \n cholesterol transferred/mL/h\n \n \n | \n \n \n \n nmol cholesterol transferred/mg P116\n \n \n\n \n *\n \n \n | \n
| \n \n \n \n \n HDL to empty P116\n \n \n \n \n | \n \n \n \n \n \n | \n \n \n \n \n \n | \n \n \n \n \n \n | \n
| \n \n \n \n \n Free cholesterol\n \n \n \n \n | \n \n \n \n 13.12\n \n \n | \n \n \n \n 13.52\n \n \n | \n \n \n \n 59.49 (6.3)\n \n \n | \n
| \n \n \n \n \n Esterified cholesterol\n \n \n \n \n | \n \n \n \n 6.98\n \n \n | \n \n \n \n 7.22\n \n \n | \n \n \n \n 31.75 (3.3)\n \n \n | \n
| \n \n \n \n \n HDL to original P116\n \n \n \n \n | \n \n \n \n \n \n | \n \n \n \n \n \n | \n \n \n \n \n \n | \n
| \n \n \n \n \n Free cholesterol\n \n \n \n \n | \n \n \n \n 7.89\n \n \n | \n \n \n \n 7.42\n \n \n | \n \n \n \n 32.63 (3.4)\n \n \n | \n
| \n \n \n \n \n Esterified cholesterol\n \n \n \n \n | \n \n \n \n 6.32\n \n \n | \n \n \n \n 6.01\n \n \n | \n \n \n \n 26.44 (2.8)\n \n \n | \n
\n \n \n \n * Numbers in parentheses are the estimated number of cholesterol molecules transferred per P116 subunit (assuming a Mw of ~105 KDa for the construct).\n \n
\n\n \n Bacterial strains, tissue cultures and growth conditions\n \n
\n\n \n M. pneumoniae\n \n M129 strain was grown in cell culture flasks containing SP4 medium and incubated at 37\u00b0C and 5% CO\n \n 2\n \n . Surface-attached mycoplasmas were harvested using a cell scraper and resuspended in SP4 medium. To grow mycoplasma cells on IBIDI 8-well chamber slides, each well was seeded with about 10\n \n 5\n \n CFUs and incubated for 12\u201324 h in 200 \u03bcL SP4 supplemented with 3% gelatin.\n
\n\n NSI myeloma cells(\n \n 23\n \n ) were grown in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) and 50 \u03bcg mL\n \n -1\n \n gentamycin (complete RPMI). Hybridomas were selected in complete RPMI supplemented with HAT media and BM-Condimed (Sigma Aldrich, St. Louis, USA).\n
\n\n \n Cloning, expression, and purification of P116 constructs\n \n
\n\n Regions corresponding to the MPN213 gene from\n \n M. pneumoniae\n \n were amplified from synthetic clones using different primers for each construct: P116F\n \n 30\n \n and P116R\n \n 957\n \n for P116(30\u2013957); P116F\n \n 13\n \n and P116R\n \n 957\n \n for P116(13\u2013957); P116F\n \n 212\n \n and P116R\n \n 862\n \n for P116(212\u2013862); and P116W\n \n 681\n \n to introduce mutation W681A PCR fragments were cloned into the expression vector pOPINE (gift from Ray Owens; plasmid #26043, Addgene, Watertown, USA) to generate constructs, with a C-terminal His-tag. Recombinant proteins were obtained after expression at 22\u00b0C in B834 (DE3) cells (Merck , Darmstadt, Germany), upon induction with 0.6\u2009mM IPTG at 0.8 OD600. Cells were harvested and lysed by French press in binding buffer (20 mM TRIS-HCl pH: 7.4, 40\u2009mM imidazole and 150 mM NaCl) and centrifuged at 49,000\u2009\u00d7\n \n g\n \n at 4\u00b0C. Supernatant was loaded onto a HisTrap 5\u2009ml column (GE Healthcare , Chicago, USA) that was pre-equilibrated in binding buffer and elution buffer (20 mM TRIS-HCl pH: 7.4, 400\u2009mM imidazole and 150 mM NaCl). Soluble aliquots were pooled and loaded onto a Superdex 200 GL 10/300 column (GE Healthcare, Chicago, USA) in a protein buffer (20 mM TRIS-HCl\u2009pH 7.4 and 150\u2009mM NaCl).\n
\n\n To obtain empty P116, 2.6% Triton X-100 was added to the protein sample and incubated for 1.5 h at room temperature. Subsequent purification followed the same methodology described above, but also included a wash step with the binding buffer plus 1.3% Triton X-100, followed by extensive washing with at least 20 column volumes of wash buffer (20 mM TRIS-HCl pH: 7.4, 20\u2009mM imidazole) before eluting the samples from the column. P116 was concentrated with Vivaspin 500 centrifugal concentrators (10,000 MWCO PES, Sartorius, G\u00f6ttingen, Germany) to a final concentration of >0.5 mg/mL.\n
\n\n To refill P116 with lipids, the empty protein was incubated with approximately 1 ml FBS per mg P116 for 2 h at 30\u00b0C while still bound on the column. After extensive washing with at least 40 column volumes of wash buffer, elution and concentration were performed as described above.\n
\n\n \n HDL isolation and determination of cholesterol transfer rate\n \n
\n\n Human HDL (density 1.063\u20131.210 g/mL) was isolated from plasma of healthy donors via sequential gradient density ultracentrifugation, using potassium bromide for density adjustment, at 100,000\n \n g\n \n for 24 h with an analytical fixed-angle rotor (50.3, Beckman Coulter, Fullerton, CA, USA). The amount of cholesterol and apolipoprotein A1 were determined enzymatically and by an immunoturbidimetric assay, respectively, using commercial kits adapted for a COBAS 6000 autoanalyzer (Roche Diagnostics, Rotkreuz, Switzerland). Radiolabeled HDLs were prepared as previously described (\n \n 24\n \n ). Briefly, 10 \u03bcCi of either [1,2-\n \n 3\n \n H(N)] free cholesterol or [1,2-\n \n 3\n \n H(N)]cholesteryl oleate (Perkin Elmer, Boston, MA) were mixed with absolute ethanol, and the solvent was dried under a stream of N\n \n 2\n \n . HDL (0.5 mL, 2.25 g/L of ApoA1) was added to the tubes containing the radiotracers, as appropriate, and then incubated for 16 h in a 37\u00b0C bath. The labeled HDLs (both\n \n 3\n \n H-cholesterol-containing and\n \n 3\n \n H-cholesteryl oleate-containing HDLs) were re-isolated by gradient density ultracentrifugation at 1.063\u20131.210 g/mL and dialyzed against PBS via gel filtration chromatography. Specific activities of\n \n 3\n \n H-cholesterol-containing and\n \n 3\n \n H-cholesteryl oleate-containing HDLs were 1221 and 185 counts per minute (cpm)/nmol, respectively. The cholesterol transfer to P116 (1 g/L) was measured after adding either [\n \n 3\n \n H] free cholesterol-containing or [\n \n 3\n \n H]cholesteryl oleate-containing HDL (0.5 g/L of APOA1) and incubating for 2 h at 37\u00b0C. HDL and P116 were separated by a\u00a0HisTrap HP affinity column. The radioactivity associated with each P116 and HDL fraction was measured via liquid scintillation counting. The percentage of [\n \n 3\n \n H]cholesterol transferred per mL was determined for each condition. The specific activities for each radiotracer were used to calculate the amount of free cholesterol and cholesteryl ester transferred from HDL to P116.\n
\n\n \n Size exclusion chromatography and multi-angle light scattering (SEC-MALS)\n \n
\n\n Molecular weights were measured from P116 samples using a Superose 6 10/300 GL (GE Healthcare, Chicago, USA) column in a Prominence liquid chromatography system (Shimadzu, Kyoto, Japan) connected to a DAWN HELEOS II multi-angle light scattering (MALS) detector and an Optilab T-REX refractive index (dRI) detector (Wyatt Technology, Santa Barbara, USA). ASTRA 7 software (Wyatt Technology) was used for data processing and analysis. An increment of the specific refractive index in relation to concentration changes (dn/dc) of 0.185 mL/g (typical of proteins) was assumed for calculations.\n
\n\n \n Matrix-assisted laser desorption/ionization-mass spectrometry (MALDI-TOF)\n \n
\n\n All samples were mixed in a 1:1 ratio with either DHB or sDHB (Bruker Daltonics, Germany) matrix solution (50 mg\u00b7ml\n \n -1\n \n in 50% Acetonitrile (ACN), 50% water and 0.1% TFA). Subsequently 1 \u03bcl aliquots of the mixture were deposited on a BigAnchor MALDI target (Bruker Daltonics, Germany) and allowed to dry and crystallize at ambient conditions. Unless stated otherwise, all reagents and solvents were obtained from Sigma Aldrich, Germany.\n
\n\n MS spectra were acquired on a rapifleX MALDI-TOF/TOF (Bruker Daltonics, Germany) in the mass range from 20.000-120.000 m/z in linear positive mode and in the mass range from 100-1600 m/z in reflector positive mode. The Compass 2.0 (Bruker, Germany) software suite was used for spectra acquisition and processing.\n
\n\n \n Lipidomics analysis (LC-TIMS-MS/MS)\n \n
\n\n Samples were extracted using a modified MTBE/Methanol extraction protocol, and submitted to LC-nanoESI-IMS-MS/MS analysis using a Bruker NanoElute UHPLC coupled to a Bruker TimsTOF Pro 2 mass spectrometer operated in DDA-PASEF mode. In brief, 40 min gradients on PepSep C18 columns (1.9A, 75\u00b5m ID, 15cm length) were recorded in positive and negative ion mode. Data were analysed using the MS-DIAL pipeline (version 4.9).\n
\n\n \n Single-particle cryoEM\n \n
\n\n For single-particle cryoEM, a 3.5 \u00b5l drop of purified P116 (100\u2013400 \u00b5g/mL in 20 mM Tris, pH 7.4 buffer or 600 \u00b5g/mL in 20 mM Tris, 2 mM CHAPSO, pH 7.4 buffer) or P116 mixed with HDL (250 \u00b5g/mL P116 and 1116 \u00b5g/mL HDL in 20 mM Tris, pH 7.4 buffer) was applied to a (45 s) glow-discharged R1.2/1.3 C-flat grid (Electron Microscopy Science, Hatfield, USA), and plunge-frozen in liquid ethane (Vitrobot Mark IV, Thermo Scientific, Waltham, USA) at 100% relative humidity, 4 \u00b0C, nominal blot force \u20133, wait time 45 s, with a blotting time of 12\u00a0s. Before freezing, Whatman 595 filter papers were incubated for 1 h in the Vitrobot chamber at 100% relative humidity and 4\u00b0C.\n
\n\n Dose-fractionated Movies of P116, P116 refilled and P116 mixed with HDL were collected with SerialEM v3.8 (\n \n 25\n \n ) at a nominal magnification of 130,000x (1.05 \u00c5 per pixel) in nanoprobe EFTEM mode at 300 kV with a Titan Krios (Thermo Scientific, Waltham, USA) electron microscope equipped with a GIF Quantum S.E. post-column energy filter in zero loss peak mode and a K2 Summit detector (Gatan Inc., Pleasanton, USA). For P116, P116 refilled and P116 with HDL a total of 4376, 4019 and 3114 micrographs with 34, 29 and 30 frames per micrograph and a frame time of 0.2 s were collected. The camera was operated in dose-fractionation counting mode with a dose rate of ~8 electrons per \u00c5\n \n 2\n \n s\n \n -1\n \n , resulting in a total dose of 50 electrons per \u00c5\n \n 2\n \n s\n \n -1\n \n . Defocus values ranged from \u20131 to \u20133.5 \u00b5m.\n
\n\n For P116 empty, dose-fractionated Movies were collected using EPU 2.12 (Thermo Scientific, Waltham, USA) at a nominal magnification of 105,000x (0.831 \u00c5 per pixel) in nanoprobe EFTEM mode at 300 kV with a Titan Krios G2 electron microscope (Thermo Scientific, Waltham, USA), equipped with a BioQuantum-K3 imaging filter (Gatan Inc., Pleasanton, USA), operated in zero loss peak mode with 20 eV energy slit width. In total 15,299 micrographs with 50 frames per micrograph and frame time of 0.052 s were collected. The K3 camera was operated in counting mode with a dose rate of ~ 16 electrons per A\n \n 2\n \n s\n \n -1\n \n , resulting in a total dose of 50 electrons per \u00c5\n \n 2\n \n s\n \n -1\n \n . Defocus values ranged from -0.8 to -3.5 \u00b5m.\n
\n\n CryoSPARC v3.2 (\n \n 26\n \n ) was used to process the cryoEM data, unless stated otherwise. Beam-induced motion correction and CTF estimation were performed using CryoSPARC\u2019s own implementation. Particles were initially clicked with the Blob picker using a particle diameter of 200\u2013300 \u00c5. Particles were then subjected to unsupervised 2D classification. For the final processing, the generated 2D averages were taken as templates for the automated particle picking, for the processing of P116 with HDL no template picking was performed. In total, 3,463,490, 4,532,601 particles, 2,930,863 particles and 262,981 particles were picked and extracted with a binned box size of 256 pixels for P116, P116 empty, P116 refilled and P116 with HDL respectively. False-positive picks were removed by two rounds of unsupervised 2D classification. The remaining 1,324,330 particles (P116), 1,140,275 particles (P116 empty), 1,311,526 particles (P116 refilled) and 46,277 particles (P116 with HDL) were used to generate an ab initio reconstruction with three classes followed by a subsequent heterogeneous refinement with three classes. For the final processing, 1,315,362 particles (P116), 633,332 particles (P116 empty), 1,311,526 particles (P116 refilled) and 46,277 particles (P116 with HDL) were used. For the remaining particles, the beam-induced specimen movement was corrected locally.\n
\n\n The CTF was refined per group on the fly within the non-uniform refinement. The obtained global resolution of the homodimer was 3.3 \u00c5 (P116), 4 \u00c5 (P116 empty), and 3.5 \u00c5 (P116 refilled) (\n \n Supplementary Figure 2 & 8 and Supplementary Table II\n \n ). To analyze the sample in regard to its flexibility the particles were subjected to the 3D variability analysis of cryoSPARC which was used to display the continuous movements of the protein.\n
\n\n \n Cryo-electron tomography of\n \n M. pneumoniae\n \n \n
\n\n \n M. pneumoniae\n \n M129 cells of an adherently growing culture were scraped in a final volume of 1 ml of SP4 medium and washed three times in PBS. This solution was mixed with fiducial markers (Protein A conjugated to 5 nm colloidal gold: Cell biology department, University Medical Center Utrecht, The Netherlands). From this stock a 3.5 \u00b5l drop was applied to a (45 s) glow-discharged R1.2/1.3 C-flat grid (Electron Microscopy Science, Hatfield, USA), and plunge-frozen in liquid ethane (Vitrobot Mark IV, Thermo Scientific, Waltham, USA) at 100% relative humidity, 4 \u00b0C, nominal blot force \u20131, with a blotting time of 10 s.\n
\n\n Tilt-series were recorded using SerialEM v3.8 (\n \n 25\n \n ) at a nominal magnification of 105,000x (1.3 \u00c5 per pixel) in nanoprobe EFTEM mode at 300 kV with a Titan Krios (Thermo Scientific, Waltham, USA) electron microscope equipped with a GIF Quantum S.E. post-column energy filter in zero loss peak mode and a K2 Summit detector (Gatan Inc., Pleasanton, USA). The total dose per tomogram was 120 e\n \n -\n \n / \u00c5\n \n 2\n \n , the tilt series covered an angular range from -60\u00b0 to 60\u00b0 with an angular increment of 3\u00b0 and a defocus set at -3 \u00b5m. Tomograms were reconstructed by super-sampling SART (\n \n 27\n \n ) with a 3D CTF correction (\n \n 28\n \n ).\n
\n\n \n P116 model building and refinement\n \n
\n\n The initial tracing of the core domain was performed manually with Coot (\n \n 29\n \n ). It contained numerous gaps and ambiguities that were slowly polished by alternating cycles of refinement using the \u201cReal Space\u201d protocol in the program Phenix (\n \n 30, 31\n \n ) and manual reinterpretation and rebuilding with Coot. The tracing and assignment of specific residues in the N-terminal domain were very difficult due to the low local resolution of the map for this domain, and only a partial interpretation was achieved. Using Robetta and AlphaFold (\n \n 14\n \n ) we obtained different predictions of the N-terminal domain structure using different parts of the sequence. The highest ranked predictions, selected using the partial experimental structure already available, were obtained with AlphaFold for residues 81\u2013245, which allowed us to complete the building of the N-terminal domain according to the cryoEM map. The RMS deviation between the AlphaFold prediction and the experimental model was 2.6 \u00c5 for 104 (63%) structurally equivalent residues. Some residues at the N-end of the N-terminal domain were difficult to identify and were represented as alanines in the final model. The whole P116 model was then refined using Phenix, and the final refined structure was deposited in the EMDB with code XXXX\n \n (Supplementary Table II)\n \n .\n
\n\n \n Polyclonal and monoclonal antibody generation\n \n
\n\n Two BALB/C mice were serially immunized with four intraperitoneal injections, each one containing 150 \u03bcg of recombinant P116 ectodomain (residues 30\u2013957) in 200 \u03bcL of PBS with no adjuvants. The last injection was delivered four days before splenectomy. Isolated B lymphocytes from the immunized mice were fused to NSI myeloma cells (\n \n 23\n \n ) to obtain stable hybridoma cell lines producing monoclonal antibodies, as previously described (\n \n 32\n \n ) .Supernatants from hybridoma cell lines derived from single fused cells were first investigated by indirect ELISA screening against the recombinant P116 ectodomain. Positive clones were also tested by Western blot against protein profiles from\n \n M. pneumoniae\n \n cell lysates and by immunofluorescence using whole, non-permeabilized M. pneumoniae cells (see below). Only those clones with supernatants revealing a single 116 kDa band in protein profiles and also exhibiting a consistent fluorescent staining of\n \n M. pneumoniae\n \n cells were selected and used in this work. Polyclonal sera were obtained by cardiac puncture of properly euthanized mice just before splenectomy and titred using serial dilutions of the antigen. The titer of each polyclonal serum was determined as the IC\n \n 50\n \n value from four parameter logistic plots and found to be approximately 1/4000 for both sera. Polyclonal anti-P1 antibodies were obtained by immunizing two BALB/C mice with recombinant P1 proteins (\n \n 33\n \n ), respectively, as described above. The titers obtained for polyclonal anti-P1 antibodies were approximately 1/2500 and 1/3000, respectively.\n
\n\n \n Immunofluorescence microscopy\n \n
\n\n The immunofluorescence staining of mycoplasma cells on chamber slides was similar to previously described (\n \n 34\n \n ), with several modifications. Cells were washed with PBS containing 0.02% Tween 20 (PBS-T) prewarmed at 37\u00b0C, and each well was fixed with 200 \u03bcL of 3% paraformaldehyde (wt/vol) and 0.1% glutaraldehyde. Cells were washed three times with PBS-T, and slides were immediately treated with 3% BSA in PBS-T (blocking solution) for 30 min. The blocking solution was removed, and each well was incubated for 1 h with 100 \u03bcL of the primary antibodies diluted in blocking solution. For P116 polyclonal sera, we used a 1/2000 dilution; a 1/10 dilution was used for monoclonal antibodies from hybridoma supernatants. Wells were washed three times with PBS-T and incubated for 1 h with a 1/2000 dilution of a goat anti-mouse Alexa 555 secondary antibody (Invitrogen, Waltham, USA) in blocking solution. Wells were then washed three times with PBS-T and incubated for 20 min with 100 \u03bcL of a solution of Hoechst 33342 10 \u03bcg/\u03bcL in PBS-T. Wells were finally washed once with PBS-T and replenished with 100 \u03bcL of PBS before microscopic examination. Cells were observed by phase contrast and epifluorescence in an Eclipse TE 2000-E inverted microscope (Nikon, Tokyo, Japan). Phase contrast images, 4',6-diamidino-2-phenylindole (DAPI, excitation 387/11 nm, emission 447/60 nm) and Texas Red (excitation 560/20 nm, emission 593/40 nm) epifluorescence images were captured with an Orca Fusion camera (Hamamatsu, Hamamatsu, Japan) controlled by NIS-Elements BR software (Nikon, Tokyo, Japan).\n
\n\n \n Time-lapse microcinematography\n \n
\n\n The effect of anti-P116 antibodies and anti-P1 polyclonal serum on mycoplasma cell adhesion was investigated by time-lapse cinematography of\n \n M. pneumoniae\n \n cells growing on IBIDI 8-well chamber slides. Before observation, medium was replaced with PBS containing 10% FBS and 3% gelatin prewarmed at 37\u00b0C. A similar medium has been used to test the effect of P1 antibodies on mycoplasma adhesion and gliding motility(\n \n 35\n \n ). After incubation for 10 min at 37\u00b0C and 5% CO\n \n 2\n \n , the slide was placed in a Nikon Eclipse TE 2000-E inverted microscope equipped with a Microscope Cage Incubation System (Okolab, Pozzuoli, Italy) at 37\u00b0C. Images were captured at 0.5 s intervals for a total observation time of 10 min. After the first 60 s of observation, the different antibodies were dispensed directly into the wells. The frequencies of motile cells and detached cells before the addition of antibodies were calculated from the images collected between 0 and 60 s of observation. The frequencies of motile cells and detached cells after the addition of antibodies were calculated from the images collected in the last minute of observation.\n
\n\n supplemental material\n
\n \n\n Supplementary Table III: Identified lipid compounds from of P116, P116-empty, P116-refilled and serum in positive and negative mode as used for heatmap generation.\n
\n \n\n movie 1\n
\n \n\n movie 2\n
\n \n\n movie 3\n
\n \n\n movie 4\n
\n \n\n movie 5\n
\n \n\n movie 6\n
\n \n\n movie 7\n
\n \n\n movie 8\n
\n \n\n movie 9\n
\n \n\n movie 10\n
\n \n\n movie 11\n
\n \n\n movie 12\n
\n \n\n Activating signal co-integrator complex (ASCC) supports diverse genome maintenance and gene expression processes. Its ASCC3 subunit is an unconventional nucleic acid helicase, harboring tandem Ski2-like NTPase/helicase cassettes crucial for ASCC functions. Presently, the molecular mechanisms underlying ASCC3 helicase activity and regulation remain unresolved. Here, we present cryogenic electron microscopy, DNA-protein cross-linking/mass spectrometry as well as\n \n in vitro\n \n and cellular functional analyses of the ASCC3-ASC1/TRIP4 sub-module of ASCC. Unlike the related spliceosomal SNRNP200 RNA helicase, ASCC3 can thread substrates through both helicase cassettes. ASC1 docks on ASCC3\n \n via\n \n a zinc finger domain and stimulates the helicase by positioning a C-terminal ASC1-homology domain next to the C-terminal helicase cassette of ASCC3, likely assisting the DNA exit. ASC1 binds ASCC3 mutually exclusively with the DNA/RNA dealkylase, ALKBH3, directing ASCC for specific processes. Our findings define ASCC3-ASC1/TRIP4 as a tunable motor module of ASCC that encompasses two cooperating ATPase/helicase units functionally expanded by ASC1/TRIP4.\n
\n\n Human activating signal co-integrator complex (ASCC) has been implicated in a surprisingly diverse range of genome maintenance and gene expression processes, including transcriptional regulation, DNA repair and ribosome quality control. ASCC was originally described to comprise four subunits,\n \n i.e.\n \n , activating signal co-integrator 1/thyroid receptor-interacting protein 4 (ASC1/TRIP4; \u201cASC1\u201d in the following), ASCC1, ASCC2 and ASCC3.\n \n \n 1\n \n \n However, different sets of ASCC subunits have been implicated in different ASCC-dependent processes, suggesting that ASCC\u2019s subunit composition or requirements may differ for different cellular functions. By associating with basal transcription factors\n \n \n 2\n \n \n , nuclear receptors\n \n \n 2\n \n ,\n \n 3\n \n \n and/or various co-activators\n \n \n 1\n \n ,\n \n 2\n \n ,\n \n 4\n \n ,\n \n 5\n \n \n , ASCC is thought to establish distinct transcription co-activator complexes in response to different cellular conditions\n \n \n 2\n \n ,\n \n 4\n \n \n . Moreover, the ASCC3 subunit has been identified as a modulator of antiviral type I interferon-stimulated genes during infections by positive-strand RNA viruses.\n \n \n 6\n \n \n ASCC2 and, in particular, ASCC3 have also been implicated in the suppression of long mRNA isoforms, due to a decrease in transcription elongation rates and instigation of alternative last exon splicing, upon UV irradiation or exposure to agents that give rise to bulky DNA lesions; a short\n \n ascc3\n \n transcript, itself originating from alternative last exon splicing, in turn acts as a long non-coding RNA during transcriptional recovery.\n \n \n 7\n \n ,\n \n 8\n \n \n
\n\n ASCC3, supported by ASC1, ASCC1 and ASCC2, is also involved in ribosome and translation quality control pathways.\n \n \n 9\n \n \u2013\n \n 12\n \n \n In contrast, only ASCC1, ASCC2 and ASCC3, have additionally been found to be important for DNA alkylation damage repair\n \n \n 13\n \n \u2013\n \n 15\n \n \n , for which the factors associate with the single-stranded (ss) DNA/RNA-specific \u03b1-ketoglutarate/iron-dependent dioxygenase, ALKBH3;\n \n \n 16\n \n ,\n \n 17\n \n \n ASC1 has not yet been implicated in DNA alkylation damage repair. Finally, ASCC, possibly in different constellations, may also help mediate RNA modification/repair processes. For example, a proteomics analysis suggested that ASC1, ASCC1, ASCC2 and ASCC3 interact with ZCCHC4, a methyl-transferase that introduces a m\n \n 6\n \n A modification at position A4220 of 28S rRNA.\n \n \n 18\n \n \n Furthermore, ASCC3 is required for efficient, ALKBH3-dependent removal of m\n \n 1\n \n A and m\n \n 3\n \n C modifications from mRNAs, and for alkylation-induced P-body formation.\n \n \n 19\n \n \n
\n\n ASCC3 contributes to all of the above ASCC-related functions. It is a large nucleic acid-dependent NTPase that can act as a 3\u2019-to-5\u2019 translocase/helicase.\n \n 13,20\n \n NTPase-fueled remodeling of nucleic acids or nucleic acid-protein complexes by ASCC3, therefore, likely constitute central activities for all of ASCC\u2019s diverse cellular roles. For example, during DNA alkylation damage repair, ASCC3 generates single-stranded DNA for dealkylation by ALKBH3.\n \n \n 13\n \n ,\n \n 14\n \n \n Furthermore, mutations in conserved NTPase/helicase motifs of ASCC3 interfere with ASCC-mediated splitting of stalled ribosomes during ribosome or translation quality control\n \n \n 9\n \n \u2013\n \n 12\n \n \n , and ASCC3 has been suggested to disassemble ribosomes collided on alkylated mRNAs for dealkylation by ALKBH3\n \n 19\n \n .\n
\n\n ASCC3 is an unconventional nucleic acid-dependent NTPase that is closely related to the spliceosomal RNA helicase, U5 small nuclear ribonucleoprotein 200 kDa (SNRNP200/BRR2). ASCC3 and SNRNP200 contain a tandem array of Ski2-like helicase cassettes (N-terminal cassette, NC; C-terminal cassette, CC), preceded by ~\u2009400-residue N-terminal regions that can auto-inhibit the helicase activities.\n \n \n 20\n \n \u2013\n \n 22\n \n \n In SNRNP200, only the NC is an active NTPase and helicase, while the CC acts as an intra-molecular helicase co-factor.\n \n \n 21\n \n ,\n \n 23\n \n \n In contrast, both helicase cassettes in ASCC3 may be enzymatically active.\n \n \n 12\n \n ,\n \n 13\n \n ,\n \n 20\n \n \n However, presently the molecular mechanisms underlying ASCC3 nucleic acid translocase/helicase activities and its regulation are poorly understood.\n
\n\n Here, we find a hitherto unobserved mechanism of nucleic acid translocation/unwinding in ASCC3 and reveal that it is regulated by ASC1. Using cryogenic electron microscopy (cryoEM)/single-particle analysis (SPA) and DNA-protein cross-linking/mass spectrometry (CLMS)-based structural analyses as well as systematic protein interaction, DNA binding and unwinding assays, we show that ASCC3 can thread DNA through both of its helicase cassettes. ASC1 docks to the ASCC3 CC\n \n via\n \n a zinc finger (ZnF) domain, positioning its ASC1-homology (ASCH) domain such that it can engage DNA exiting from ASCC3. We also present evidence that ASC1 and ALKBH3 engage ASCC3 in a mutually exclusive manner and that ASC1 does not affect ASCC-dependent DNA alkylation damage repair, suggesting that ASC1 and ALKBH3 are facultative, process-specific ASCC subunits or auxiliary proteins.\n
\n\n ASCC1 and ASCC2 directly interact with ASCC3,\n \n \n 11\n \n ,\n \n 15\n \n ,\n \n 20\n \n \n suggesting that ASCC3 forms the main scaffold for the ASCC and possibly an interaction platform for ASCC-auxiliary proteins. We therefore tested whether ASC1 also directly binds ASCC3\n \n in vitro\n \n . While ASC1 did not stably interact with the ASCC3 N-terminal region (ASCC3\n \n NTR\n \n , residues 1-400), it co-eluted with the helicase region of ASCC3 (ASCC3\n \n HR\n \n , residues 401\u20132202) in analytical size-exclusion chromatography (SEC; Fig.\n \n 1\n \n a,b)\n
\n\n Next, we reconstituted an ASCC3\n \n HR\n \n -ASC1 complex and determined its atomic structure\n \n via\n \n cryoEM/SPA at a nominal resolution of 3.4 \u00c5 (Fig.\n \n 2\n \n a; Supplementary Fig.\u00a01; Supplementary Fig.\u00a02; Supplementary Table\u00a01). In the cryoEM reconstruction, we could trace residues 401\u20132183 of ASCC3\n \n HR\n \n as well as residues 168\u2013219 and 375\u2013580 of ASC1 (Fig.\n \n 2\n \n b,c), capitalizing on AlphaFold-predicted models\n \n \n 24\n \n \n . ASCC3\n \n HR\n \n adopts a structure very similar to the helicase region of SNRNP200 (root mean square deviation [rmsd] of 3.1 \u00c5 for 1,504 pairs of C\u03b1 atoms compared to isolated SNRNP200\n \n HR\n \n ; PDB ID 4F91; Supplementary Fig.\u00a03)\n \n \n 21\n \n \n . Like SNRNP200, both ASCC3 helicase cassettes contain consecutive dual RecA-like (RecA1, RecA2), winged-helix (WH), helical bundle (HB), helix-loop-helix (HLH) and immunoglobulin-like (IG) domains and associate to form a compact helicase region (Fig.\n \n 2\n \n c). An extended, irregularly structured linker (residue 1296\u20131306) connects the IG domain of the NC to the RecA1 domain of the CC, running closely along the body of the ASCC3 CC (Fig.\n \n 2\n \n a-c).\n
\n\n ASC1 exclusively associates with the CC of ASCC3\n \n HR\n \n (Fig.\n \n 2\n \n a,b). Residues 168\u2013219 of ASC1 fold into a dual-ZnF domain, with residues C171/C173/H178/C192 and C184/C187/C200/C203 each coordinating a zinc ion (Fig.\n \n 2\n \n b). The ZnF domain of ASC1 rests on top of the RecA1 domain of the ASCC3 CC, neighboring the extended linker to the NC (Fig.\n \n 2\n \n b) and spanning\u2009~\u2009757 \u00c5\n \n \n 2\n \n \n of interface area, with hydrophobic interactions in the center and hydrophilic interactions at the periphery (Fig.\n \n 2\n \n d, left). ASC1 residues 375\u2013424 lack a globular fold and regular secondary structure elements, except for a short helical region in residues 398\u2013405. They form a lasso-like structure around a protruding edge of the C-terminal ASCC3 WH domain (Fig.\n \n 2\n \n b; Fig.\n \n 2\n \n d, middle), with residues 411\u2013424 inserted deeply into a groove between the RecA1, WH, HB and IG domains of the ASCC3\n \n HR\n \n CC, spanning\u2009~\u20091,914 \u00c5\n \n \n 2\n \n \n of interface area with ASCC3\n \n HR\n \n . ASC1 residues 411\u2013424 form a support for the C-terminal ASCH domain of ASC1 (residues 425\u2013578) that further interconnects the C-terminal ASCC3 RecA1, WH and IG domains (Fig.\n \n 2\n \n b; Fig.\n \n 2\n \n d, right), spanning an additional\u2009~\u20091,321 \u00c5\n \n \n 2\n \n \n of interface area with ASCC3\n \n HR\n \n .\n
\n\n \n The ZnF domain is required for stable docking of ASC1 on ASCC3\n \n \n \n HR\n \n \n \n in vitro\n \n
\n\n Based on the structure, we designed various ASC1 fragments to probe the importance of different regions for stable complex formation with ASCC3\n \n HR\n \n . Consistent with the cryoEM structure, the N-terminal 80 residues of ASC1 did not sustain a stable interaction with ASCC3\n \n HR\n \n (Fig.\n \n 1\n \n c), while ASC1 residues 152\u2013581, encompassing the ZnF domain, the lasso-like peptide and the ASCH domain, co-migrated with ASCC3\n \n HR\n \n in analytical SEC (Fig.\n \n 1\n \n d). An N-terminal ASC1 region including the ZnF domain (residues 1-230) or the ZnF domain alone (residues 152\u2013230) also stably bound ASCC3\n \n HR\n \n (Fig.\n \n 1\n \n e). In contrast, C-terminal ASC1 residues 281\u2013403, 403\u2013581 or 281\u2013581, containing the lasso-like peptide, the ASCH domain or both, did not support stable complex formation with ASCC3\n \n HR\n \n (Fig.\n \n 1\n \n f), although these regions span a considerably larger interface with ASCC3\n \n HR\n \n than the ZnF domain (see above). Thus, only the ZnF domain of ASC1 is required for stable complex formation\n \n in vitro\n \n , and only upon anchoring\n \n via\n \n the ZnF domain, the C-terminal ASCH domain and the preceding peptide region of ASC1 are stably docked on the ASCC3 CC.\n
\n\n \n Cellular interaction tests corroborate\n \n \n in vitro\n \n \n interaction patterns\n \n
\n\n To test the importance of ASC1 regions for the interaction with ASCC3 and other ASCC subunits in cells, we generated stably transfected Flp-In\u2122 T-REx\u2122 293 cell lines for the inducible expression of N- or C-terminally Flag-tagged versions of full-length ASC1 or truncation variants lacking either N-terminal regions including the ZnF domain (ASC1\n \n \u03941\u2212\u2009276\n \n ) or lacking the C-terminal ASCH domain (ASC1\n \n \u0394403\u2212\u2009581\n \n ). Immuno-fluorescence microscopy showed that all constructs were located to both the cytosol and the nucleus (Fig.\n \n 3\n \n a). We then immuno-precipitated the Flag-tagged ASC1 variants with \u03b1-Flag antibodies and probed the eluates for the presence of other ASCC subunits by western blot. Irrespective of the position of the tag, ASC1 and ASC1\n \n \u0394403\u2212\u2009581\n \n (lacking the ASCH domain) co-precipitated ASCC1, ASCC2 and ASCC3 (Fig.\n \n 3\n \n b). In contrast, no interaction with these ASCC subunits was detected by co-precipitation with ASC1\n \n \u03941\u2212\u2009276\n \n (lacking the ZnF domain; Fig.\n \n 3\n \n b).\n
\n\n To further test the relevance of ASCC3\n \n HR\n \n -ASC1 contacts observed in our cryoEM structure for the interaction of ASCC3 and ASC1 in cells, we transfected 293T cells for the expression of N-terminally HA-tagged versions of ASC1. In these ASC1 variants, either the ZnF domain was precisely deleted (\u0394ZnF; deletion of residues 168\u2013219), three residues that engage in hydrophobic interactions with ASCC3\n \n HR\n \n were exchanged for alanines (LLI-AAA, ASC1\n \n L174A\u2009\u2212\u2009L180A\u2212I190A\n \n ; Fig.\n \n 2\n \n d, left) or two cysteines coordinating the first (C171) and second (C184) Zn\n \n 2+\n \n ion were exchanged for alanines (CC-AA, ASC1\n \n C171A\u2009\u2212\u2009C184A\n \n ). While wild type (wt) ASC1 efficiently co-immuno-precipitated endogenous ASCC3, the \u0394ZnF and CC-AA variants entirely lost the ability to immuno-precipitate ASCC3, and the ASCC3 interaction of the LLI-AAA variant was strongly reduced (Fig.\n \n 3\n \n c).\n
\n\n Together, the results of these cellular interaction studies are fully in line with the\n \n in vitro\n \n ASCC3\n \n HR\n \n -ASC1 interaction profiles. They confirm that the ZnF domain of ASC1 is the main ASCC3-interacting domain of ASC1,\n \n via\n \n which ASC1 also seems to be incorporated into the ASCC, and suggest that ASC1, ASCC1 and ASCC2 can concomitantly interact with ASCC3. The observations also confirm that our cryoEM structure closely represents the mode of interaction of ASCC3 and ASC1 in cells.\n
\n\n To test the effect of ASC1 on the helicase activity of ASCC3\n \n HR\n \n , we conducted fluorescence-based unwinding assays in a stopped-flow device (Fig.\n \n 4\n \n a). As this assay tested multiple rounds of unwinding, the observed time traces were fit to a double exponential equation, and amplitude-weighted unwinding rate constants (\n \n k\n \n \n \n uaw\n \n \n ) were calculated for the comparison of unwinding efficiencies.\n \n \n 25\n \n \u2013\n \n 27\n \n \n ASCC3\n \n HR\n \n alone efficiently unwound the substrate DNA (\n \n k\n \n \n \n uaw\n \n \n = 0.024 s\n \n \u2212\u20091\n \n ), but unwinding was further stimulated 2.3-fold by ASC1 (\n \n k\n \n \n \n uaw\n \n \n = 0.054 s\n \n \u2212\u20091\n \n ; Fig.\n \n 4\n \n b; Supplementary Table\u00a02). In contrast, both ASC1\n \n 1\u2013230\n \n (encompassing the ZnF domain), which stably bound ASCC3\n \n HR\n \n in analytical SEC, as well as ASC1\n \n 403\u2013581\n \n (encompassing the ASCH domain and preceding peptide), which did not co-migrate with ASCC3\n \n HR\n \n in analytical SEC, only marginally affected the ASCC3\n \n HR\n \n helicase activity (\n \n k\n \n \n \n uaw\n \n \n = 0.030 s\n \n \u2212\u20091\n \n and 0.035 s\n \n \u2212\u20091\n \n , respectively; Fig.\n \n 4\n \n b; Supplementary Table. 2). Thus, while the ASC1 ZnF domain alone can stably bind to ASCC3\n \n HR\n \n , it does not efficiently activate ASCC3\n \n HR\n \n helicase activity, for which the lasso-like peptide and ASCH domain are also required.\n
\n\n Next, we asked which helicase cassette of ASCC3\n \n HR\n \n is preferentially regulated by ASC1. To this end, we employed ASCC3\n \n HR\n \n variants, in which a crucial motif II aspartate of the NC (D611) or CC (D1453) was exchanged for an alanine (Fig.\n \n 4\n \n c), abrogating NTPase/helicase activity in the respective cassette.\n \n \n 20\n \n \n ASCC3\n \n HR,D1453A\n \n , bearing an inactive CC, unwound DNA at a reduced rate (\n \n k\n \n \n \n uaw\n \n \n = 0.011 s\n \n \u2212\u20091\n \n ), while the unwinding activity of ASCC3\n \n HR,D611A\n \n , containing an inactive NC, was strongly reduced (\n \n k\n \n \n \n uaw\n \n \n n.d.; Fig.\n \n 4\n \n d), suggesting that both cassettes are required for full ASCC3 helicase activity. Only the construct bearing an inactive CC was stimulated by ASC1 to quantifiable levels (ASCC3\n \n HR,D1453A\n \n -ASC1\n \n k\n \n \n \n unw\n \n \n = 0.024 s\n \n \u2212\u20091\n \n ; Fig.\n \n 4\n \n d; Supplementary Table\u00a02).\n
\n\n NTPase activity associated with both ASCC3\n \n HR\n \n cassettes was further corroborated by DNA-stimulated ATPase assays. ASCC3\n \n HR,D1453A\n \n (inactive CC) and ASCC3\n \n HR,D611A\n \n (inactive NC) exhibited\u2009~\u200928 and ~\u200973% of the DNA-stimulated ATPase activity of wt ASCC3\n \n HR\n \n , while the DNA-stimulated ATPase activity of the ASCC3\n \n HR,DD611/1453AA\n \n variant, with motif II changes in both cassettes, was negligible (Fig.\n \n 4\n \n e; Supplementary Fig.\u00a04). As expected if the implemented residue exchanges selectively abrogated ATPase activity in the respective cassette, the ATPase activity of ASCC3\n \n HR,D1453A\n \n (inactive CC) closely matched the ATPase activity of the isolated wt NC (Fig.\n \n 4\n \n e; Supplementary Fig.\u00a04) As we failed to produce the ASCC3 CC in isolation, a similar comparison could not be drawn between ASCC3\n \n HR,D611A\n \n (inactive NC) and isolated wt CC. Irrespectively, in contrast to the helicase activity, the stimulated ATPase activity of ASCC3\n \n HR\n \n was not further enhanced by ASC1 (Fig.\n \n 4\n \n e). Thus, ASC1 activates ASCC3\n \n HR\n \n helicase activity without affecting its ATPase activity.\n
\n\n We failed to obtain cryoEM structures of ASCC3\n \n HR\n \n or ASCC3\n \n HR\n \n -ASC1 in complex with ssDNA or with dsDNA bearing a 3\u2019-ss overhang. Modeling of putative ssDNA binding to the NC and CC of ASCC3\n \n HR\n \n by superimposing a structure of the Hel308 DNA helicase in complex with DNA (PDB ID 2P6R)\n \n \n 28\n \n \n on both ASCC3\n \n HR\n \n cassettes indicated that ssDNA could be threaded consecutively through both helicase cassettes and might exit the CC close to the ASC1 ASCH domain (Fig.\n \n 5\n \n a). Positive electrostatic surface potential is in agreement with the modeled path of ssDNA, in particular for the ASCC3\n \n HR\n \n NC (Fig.\n \n 5\n \n a). The model suggested that a minimum of 24 nucleotides (nts) of ssDNA are required to traverse the two cassettes and ASC1. In contrast, lateral entry of ssDNA to the CC, circumventing the NC, is blocked in the conformation of ASCC3\n \n HR\n \n observed in our cryoEM structure. A requirement for DNA to enter the CC\n \n via\n \n the preceding NC would be consistent with the larger effect on helicase activity we observed upon inactivating the NC alone as compared to a ASCC3\n \n HR\n \n variant containing only an inactive CC (see Fig.\n \n 4\n \n ).\n
\n\n To test if, during unwinding, ASCC3\n \n HR\n \n and ASCC3\n \n HR\n \n -ASC1 might thread single-stranded DNA through both helicase cassettes, and in the latter case along the ASC1 ASCH domain, we conducted ultra-violet (UV) irradiation-induced cross-linking of ASCC3\n \n HR\n \n and ASCC3\n \n HR\n \n -ASC1 to variable-length, single-stranded oligo-T DNAs (T\n \n 12\n \n , T\n \n 24\n \n , T\n \n 36\n \n , T\n \n 48\n \n ; Fig.\n \n 5\n \n b). Both ASCC3\n \n HR\n \n and ASCC3\n \n HR\n \n -ASC1 did not efficiently cross-link to T\n \n 12\n \n ssDNA and showed stepwise increased cross-linking to T\n \n 24\n \n , T\n \n 36\n \n and T\n \n 48\n \n DNAs (cross-link efficiencies of ~\u200930%, 80% and 90%, respectively; Fig.\n \n 5\n \n b,c). ASC1 alone did not efficiently cross-link to any of the DNA samples. These observations are consistent with the notion that a ssDNA region sufficiently long to traverse both cassettes is required for DNA to be efficiently engaged by ASCC3\n \n HR\n \n or ASCC3\n \n HR\n \n -ASC1.\n
\n\n Next, we subjected ASCC3\n \n HR\n \n or ASCC3\n \n HR\n \n -ASC1 after UV-induced cross-linking to T\n \n 48\n \n ssDNA to DNase/protease digestion followed by mass spectrometric analysis of cross-linked peptide-DNA conjugates. We observed one cross-linked peptide each in ASC1 (region connecting ZnF and lasso), the RecA1 domain of the ASCC3\n \n HR\n \n NC (corresponding to helicase motif Ia), the N-terminal WH domain and the C-terminal WH domain, as well as two cross-linked peptides in the CC IG domain (Table\n \n 1\n \n ). With exception of the ASC1 peptide, we could identify one or two specific cross-linked residues in these peptides (Table\n \n 1\n \n ; RecA1\n \n NC\n \n , M546; WH\n \n NC\n \n , Y988; WH\n \n CC\n \n , Y1821 and Y1822; IG\n \n CC\n \n , C2101 and Y2135). The cross-linked residues and the modeled cross-linked ASC1 peptide are positioned closely along the path of the modeled DNA (Fig.\n \n 5\n \n d). Together, these observations are consistent with the idea that during unwinding, ssDNA is threaded through both helicase cassettes and along ASC1 in the vicinity of the ASCH domain. It is, however, also possible that ASCC3\n \n HR\n \n may undergo conformational changes upon binding to ssDNA of sufficient length, so that the substrate can engage the NC and CC independently.\n
\n\n \n \n Table 1: DNA-protein cross-links identified by MS.\n \n \n
\n| \n \n \n \n Cross-linked\n \n \n \n\n \n \n pepide\n \n 1\n \n \n \n \n | \n \n \n \n \n Cross-linked residue\n \n \n \n | \n \n \n \n \n Trial\n \n \n \n | \n \n \n \n \n Domain or region\n \n \n \n | \n \n \n \n \n Motif\n \n 2\n \n \n \n \n | \n
| \n \n \n \n ASC1\n \n \n \n | \n ||||
| \n \n \n 257-\n \n \n SGLEK\n \n \n -261\n \n \n | \n \n \n \n n.i.\n \n 3\n \n \n \n | \n \n \n \n 1\n \n \n | \n \n \n \n ZnF-lasso linker\n \n \n | \n \n \n \n -\n \n \n | \n
| \n \n \n \n ASCC3\n \n HR\n \n \n \n \n | \n ||||
| \n \n \n 541-ALAAE\n \n \n \n M\n \n \n \n TDYFSR-552\n \n \n | \n \n \n \n M546\n \n \n | \n \n \n \n 2\n \n \n | \n \n \n \n RecA1\n \n NC\n \n \n \n | \n \n \n \n Ia\n \n \n | \n
| \n \n \n 984-\n \n \n TASH\n \n \n \n Y\n \n \n \n \n Y\n \n IK\n \n \n -991\n \n \n | \n \n \n \n Y988\n \n \n | \n \n \n \n 1,2\n \n \n | \n \n \n \n WH\n \n NC\n \n \n \n | \n \n \n \n -\n \n \n | \n
| \n \n \n 1818-IAS\n \n \n \n Y\n \n \n \n YYLK-1825\n \n \n | \n \n \n \n Y1821\n \n \n | \n \n \n \n 2\n \n \n | \n \n \n \n WH\n \n CC\n \n \n \n | \n \n \n \n -\n \n \n | \n
| \n \n \n 1818-IASY\n \n \n \n Y\n \n \n \n YLK-1825\n \n \n | \n \n \n \n Y1822\n \n \n | \n \n \n \n 2\n \n \n | \n \n \n \n WH\n \n CC\n \n \n \n | \n \n \n \n -\n \n \n | \n
| \n \n \n 2096-GKPES\n \n \n \n C\n \n \n \n AVTPR-2106\n \n \n | \n \n \n \n C2101\n \n \n | \n \n \n \n 2\n \n \n | \n \n \n \n IG\n \n CC\n \n \n \n | \n \n \n \n -\n \n \n | \n
| \n \n \n 2133-VG\n \n \n \n Y\n \n \n \n IR-2137\n \n \n | \n \n \n \n Y2135\n \n \n | \n \n \n \n 1,2\n \n \n | \n \n \n \n IG\n \n CC\n \n \n \n | \n \n \n \n -\n \n \n | \n
\n \n \n 1\n \n \n \n Cross-linked residue(s) colored as in (\n \n c\n \n ) and underlined\n \n
\n\n \n \n 2\n \n \n \n NC helicase motif Ia, residues 536-546\n \n
\n\n \n \n 3\n \n \n \n n.i., not identified\n \n
\n\n
\n
\n Present data suggest that ASCC core subunits may associate with different auxiliary proteins to participate in distinct genome maintenance and gene expression processes. More specifically, ASC1 has so far been found associated with ASCC-dependent transcription regulation\n \n \n 1\n \n ,\n \n 2\n \n ,\n \n 4\n \n ,\n \n 5\n \n \n and ribosome quality control\n \n \n 9\n \n ,\n \n 11\n \n ,\n \n 12\n \n \n , while ALKBH3 is associated with ASCC3 during DNA dealkylation repair\n \n \n 13\n \n ,\n \n 14\n \n \n . We therefore wondered whether ASC1 and ALKBH3 might bind ASCC3 in a mutually exclusive manner. To test this notion, we conducted competitive SEC-based interaction studies. ASC1 and ALKBH3 did not co-migrate during SEC (Fig.\n \n 6\n \n a). A portion of ALKBH3 stably associated with ASCC3\n \n HR\n \n in SEC, but failed to be incorporated into a pre-formed ASCC3\n \n HR\n \n -ASC1 complex (Fig.\n \n 6\n \n a). These findings suggest that ASC1 and ALKBH3 engage ASCC3\n \n HR\n \n in a mutually exclusive manner, possibly by taking advantage of overlapping binding sites, and that ASC1 might associate more strongly with ASCC3\n \n HR\n \n than ALKBH3.\n
\n\n To further test the idea that either ASC1 or ALKBH3 associates with ASCC core subunits depending on the particular ASCC-dependent cellular process, we explored the effect of ASC1 on DNA dealkylation damage repair, where ALKBH3 is known to be involved. To this end, we knocked out ASC1\n \n via\n \n CRISPR/Cas9-based genome engineering in human PC-3 cells (Fig.\n \n 6\n \n b) and tested the response of the edited and parental cells to methyl methanesulfonate (MMS) treatment. ASC1 knockout (KO) did not impact cell survival in the presence of even high concentrations of MMS (Fig.\n \n 6\n \n c), suggesting that ASC1 may not be involved in ASCC3/ALKBH3-mediated DNA dealkylation\n \n \n 13\n \n \n . Together, these observations suggest that ASC1 represents a process-specific ASCC subunit that regulates ASCC3 helicase activity during ASCC-dependent transcriptional events and ribosome rescue, but may be replaced by ALKBH3 during ASCC-dependent DNA dealkylation damage repair.\n
\n\n ASCC is a multi-functional complex. While apparently different sets of ASCC core and auxiliary factors participate in different ASCC-dependent processes, the large nucleic acid helicase, ASCC3, seems to provide crucial molecular motor activity for all of ASCC\u2019s multiple functions. ASCC3 has striking homology to the spliceosomal RNA helicase, SNRNP200, and the two proteins represent the only known human members of a unique sub-family of Ski2-like helicases that possess tandem helicase cassettes. In SNRNP200, only the NC is an active ATPase/RNA helicase, while the CC acts as an intra-molecular modulator of the NC helicase.\n \n \n 21\n \n ,\n \n 29\n \n \n
\n\n Here, we show by cryoEM-based structural analysis that ASCC3 indeed contains a dual-cassette helicase region that closely resembles the analogous region of SNRNP200, at least in the absence of factors other than ASC1. In line with previous observations\n \n \n 9\n \n \u2013\n \n 11\n \n ,\n \n 13\n \n ,\n \n 20\n \n \n , our systematic DNA unwinding and ATPase assays strongly suggest that, in contrast to SNRNP200, both ASCC3 cassettes are active ATPases and helicases. Our DNA-protein CLMS analyses are consistent with a model in which ASCC3 translocates relative to ssDNA during DNA unwinding, threading one DNA strand consecutively through both helicase units. In principle, our data would also be consistent with the two ASCC3 helicase cassettes unwinding DNA independently of each other. However, in the ASCC3\n \n HR\n \n conformation observed here, direct accommodation of ssDNA at the CC is blocked by the NC. Thus, for the latter scenario, ASCC3 would have to undergo a large conformational rearrangement that leads to a separation of its helicase cassettes if ssDNA were to be captured by the CC without being first threaded through the NC. As ASCC3 interacts with different substrate complexes and auxiliary proteins in different functional contexts, which could provoke conformational changes in ASCC3, it is conceivable that in certain scenarios the helicase activity of either individual cassette is employed, while in others the two helicase cassettes operate in tandem. Furthermore, in a given functional scenario the two cassettes may even translocate the same or different nucleic acid molecules (see also below).\n
\n\n The CC of SNRNP200 serves as an interaction platform for numerous proteins, several of which inhibit its NC helicase activity from a distance\n \n \n 30\n \n \u2013\n \n 32\n \n \n . In contrast, the C-terminal Jab1 domain of the large spliceosomal PRPF8 scaffold that can activate the SNRNP200 helicase directly binds the active NC.\n \n \n 33\n \n ,\n \n 34\n \n \n Here, we find that similar to the situation in SNRNP200, the ASCC3 CC serves as a binding platform for the ASC1 protein. ASC1 predominantly latches onto ASCC3\n \n via\n \n its ZnF domain, allowing the positioning of an ASCH domain close to the presumed DNA exit of the ASCC3 CC with the help of the intervening lasso peptide. However, unlike many proteins that bind the SNRNP200 CC, we show that ASC1 stimulates ASCC3 helicase activity. The ZnF docking domain is insufficient for helicase stimulation, which also requires C-terminal ASC1 regions including the ASCH domain. While we cannot yet pinpoint the precise molecular mechanism, by which ASC1 stimulates ASCC3, our DNA-protein CLMS data support the notion that the ASCH domain or neighboring regions may facilitate DNA exit from the ASCC3 CC. Indeed, the ASCH domain belongs to a large family of domains that bind nucleic acids.\n \n \n 35\n \n ,\n \n 36\n \n \n
\n\n Cooperation between both helicase cassettes and activation of ASCC3 helicase activity by ASC1 may be required to unfold sufficiently strong or appropriately coordinated motor activity during transcription regulatory processes and ribosome quality control, where both ASCC3 and ASC1 are involved. While the targets of ASCC3\u2019s motor activity during transcriptional regulation are presently unknown, during ribosome quality control, ASCC3\u2019s ATP-dependent motor activity is essential for the disassembly of the lead ribosome in collided di-somes or polysomes into ribosomal subunits.\n \n \n 9\n \n ,\n \n 11\n \n ,\n \n 12\n \n \n As no DNA is involved in this process, ASCC3 most likely operates by engaging and translocating mRNA or rRNA regions. Indeed, we know that ASCC3 can also unwind RNA duplexes\n \n in vitro\n \n , suggesting that it is also an RNA translocase, but its RNA helicase/translocase activity is much less efficient than its DNA helicase/translocase activity (unpublished). Our analyses show that inactivation of either ASCC3 cassette leads to partial loss of ASCC3 helicase activity. Thus, splitting of ribosomes by translocating on a sub-optimal mRNA or rRNA substrate may require (a) ASCC3 resorting to a translocation mode that involves both active cassettes on the same or on different RNA molecules, (b) additional stimulation by ASC1 and/or (c) stimulation by another accessory factor that promotes ASCC3 RNA translocase/helicase function.\n
\n\n Recent cryoEM structures of yeast ribosome quality control trigger complex (RQT)-ribosome complexes revealed that prior to ribosome splitting the yeast ASCC3 ortholog, Slh1p, can adopt a more open conformation with fewer direct interactions between the two helicase cassettes as observed in our human ASCC3-ASC1 complex structure.\n \n \n 37\n \n \n While in the imaged conformations both Slh1p helicase cassettes are potentially accessible to an RNA substrate, no corresponding substrate density was observed at either Slh1p cassette.\n \n \n 37\n \n \n In the observed conformations, mRNA could apparently be accommodated directly at the Slh1p CC, but an Slh1p variant harboring an ATPase-deficient NC (Slh1p\n \n K361R\n \n ) was required to capture RQT-ribosome complexes at a stage preceding ribosome splitting\n \n \n 37\n \n \n , indicating that the NC ATPase/helicase activity is also required for the splitting reaction. Thus, whether both cassettes or only one of them translocate mRNA or whether one cassette engages mRNA while the other operates on an rRNA region during ribosome splitting remains to be elucidated.\n
\n\n Findings reported here also underscore the notion that ASCC exhibits compositional dynamics that allow it to participate in different processes. We find that ASC1, which collaborates with ASCC3 during transcriptional and ribosome quality control, binds to ASCC3 in a manner that is mutually exclusive to ALKBH3, which capitalizes on the ASCC3 helicase activity during DNA alkylation damage repair. Consistent with the idea of these two factors associating with ASCC3 in different functional scenarios, we also show that ASC1 does not impact cell sensitivity to an alkylating agent, unlike ALKBH3 or other subunits of the ASCC complex\n \n \n 13\n \n ,\n \n 15\n \n \n . As ASC1 seems to associate more stably with ASCC3\n \n HR\n \n than ALKBH3, it remains to be seen if additional factors may aid ALKBH3 in displacing ASC1 for DNA dealkylation damage repair. Additional interactors may favor a conformation of ASCC3 that exhibits altered ALKBH3 affinity. It is also possible, that the protein interactions of ASCC3 may be dynamically regulated by specific post-translational modifications or by the recruitment of subsets of factors to specific sub-cellular compartments. Both of the latter principles have been shown to play a role during ASCC-related cellular processes.\n \n \n 9\n \n ,\n \n 10\n \n ,\n \n 14\n \n ,\n \n 15\n \n ,\n \n 19\n \n ,\n \n 38\n \n ,\n \n 39\n \n \n
\n\n DNA fragments encoding ASCC3\n \n HR\n \n (wt, D611A, D1453A or D611A-D1453A) or ASCC3\n \n NC\n \n were cloned into a pFL vector for expression as N-terminally His\n \n 10\n \n -tagged, TEV-cleavable proteins\n \n via\n \n recombinant baculoviruses in insect cells as described previously.\n \n \n 20\n \n \n A DNA fragment encoding full-length (FL) ASC1 was PCR-amplified from a synthetic gene (IDT; Supplementary Table\u00a03) and inserted into the pETM-11 or pIDS vectors (EMBL, Heidelberg) for expression as an N-terminally His\n \n 6\n \n -tagged, TEV-cleavable protein. See Supplementary Table\u00a04 for PCR primers used. The pIDS-\n \n asc1\n \n \n FL\n \n construct was Cre-recombined with pFL-\n \n ascc3\n \n \n HR\n \n for co-expression\n \n via\n \n a recombinant baculovirus in insect cells. DNA fragments encoding ASC1\n \n 1\u201380\n \n , ASC1\n \n 1\u2013230\n \n , ASC1\n \n 152\u2013230\n \n , ASC1\n \n 152\u2013581\n \n , ASC1\n \n 281\u2013403\n \n , ASC-1\n \n 281\u2013581\n \n or ASC-1\n \n 403\u2013581\n \n were amplified\n \n via\n \n PCR from the pETM-11-\n \n asc1\n \n \n FL\n \n , and re-cloned into the pETM-11 vector. A DNA fragment encoding full-length ALKBH3 was PCR-amplified from a cDNA library of human HeLa cells and inserted into the pETM-11 vector for expression as an N-terminally His\n \n 6\n \n -tagged, TEV-cleavable protein. All constructs were verified by Sanger sequencing.\n
\n\n For the preparation of an ASC1 sgRNA vector, we followed a previously established method\n \n \n 40\n \n \n , cloning the target sequence into the pLenti-CRISPRV2 vector\n \n \n 41\n \n \n . Primers used for generating the DNA fragment containing the target sequence is shown in Supplementary Table\u00a04.\n
\n\n For expression of HA-tagged ASC1 variants, DNA fragments encoding wt or \u0394ZnF ASC1 were cloned into pENTR-3C using a synthetic gene (IDT; Supplementary Table\u00a03). Vectors encoding HA-tagged variants ASC1\n \n L174A\u2009\u2212\u2009L180A\u2212I190A\n \n or ASC1\n \n C171A\u2009\u2212\u2009C184A\n \n were created using the In-Fusion Snap Assembly mutagenesis kit (Takara Bio #683945). Each construct was then cloned into pHAGE-HA-Blast vector\n \n \n 14\n \n \n \n via\n \n Gateway recombination. All constructs were verified by Sanger sequencing.\n
\n\n Stably transfected Flp-In\u2122 T-REx\u2122 293 cell lines for the tetracycline-inducible expression of ASC1 variants with N-terminal 2xFlag-His\n \n 6\n \n or C-terminal His\n \n 6\n \n -2xFlag tags were generated according to the manufacturer\u2019s guidelines.\n \n \n 20\n \n \n Transfection of the parental cell line was done using X-tremeGENE HP DNA Transfection Reagent (Sigma Aldrich). After hygromycin-based selection of cells that had genomically integrated the expression cassette, tetracycline-induced expression of the tagged proteins was confirmed by western blotting using a monoclonal \u03b1-Flag M2 antibody (Sigma Aldrich #F3165; 1:7500). For expression of HA-tagged ASC1 variants, the pHAGE-HA-ASC1 vectors encodin HA-tagged ASC1\n \n wt\n \n , ASC1\n \n \u0394ZnF\n \n , ASC1\n \n L174A\u2009\u2212\u2009L180A\u2212I190A\n \n or ASC1\n \n C171A\u2009\u2212\u2009C184A\n \n , were transfected into 293T cells using Transit293 transfection reagent (Mirus Bio).\n
\n\n The ASC1 sgRNA expression vector was transfected into the Lenti-X 293T cell line (Takara Bio) together with psPAX2 and pCMV-VSVG (Addgene) for lentivirus production. The virus-containing culture medium was collected 72 h post-transfection. Human PC-3 cells were infected with the viral medium and individual clones were selected in 96-well plates. The single KO colonies were analyzed by western blot using an \u03b1-ASC1 antibody (sc-376916, Santa Cruz).\n
\n\n ASCC3\n \n HR\n \n variants and ASCC3\n \n NC\n \n were produced in High Five cells as described previously.\n \n \n 20\n \n \n Cell pellets were re-suspended in 20 mM HEPES-NaOH, pH 7.5, 500 mM NaCl, 10 mM imidazole, 1 mM DTT, 8.6% (v/v) glycerol (lysis buffer 1), supplemented with cOmplete\u2122 protease inhibitors (Roche) and lysed by sonication using a Sonopuls Ultrasonic Homogenizer HD (Bandelin). The lysate was cleared by centrifugation and filtration. The protein of interest (POI) was captured on Ni\n \n 2+\n \n -NTA resin in a gravity flow column, washed with lysis buffer 1 and eluted with lysis buffer 1 containing 400 mM imidazole. Fractions enriched for the POI were supplemented with 1/10 (w/w) TEV protease and dialyzed against 20 mM HEPES-NaOH, pH 7.5, 500 mM NaCl, 1 mM DTT, 8.6% (v/v) glycerol (dialysis buffer) overnight. The sample was then diluted to 100 mM NaCl and loaded onto a HiTrap Heparin HP column (Cytiva), pre-equilibrated with lysis buffer 1 containing 100 mM NaCl. After washing with lysis buffer 1 containing 100 mM NaCl, the POI was eluted with a linear gradient to lysis buffer 1 containing 1.5 M NaCl. The fractions containing the POI were pooled and concentrated with a centrifugal concentrator (100 kDa molecular mass cut-off). The concentrated sample was further purified by SEC on a Superdex 200 10/300 GL column (Cytiva) in 20 mM HEPES-NaOH, pH 7.5, 250 mM NaCl, 5% (v/v) glycerol, 1 mM DTT (SEC buffer). Fractions containing the POI were combined, concentrated, aliquoted, flash-frozen in liquid nitrogen and stored at -80\u00b0C.\n
\n\n For preparation of the ASCC3\n \n HR\n \n -ASC1\n \n FL\n \n complex, ASC1\n \n FL\n \n was co-produced with ASCC3\n \n HR\n \n in High Five cells. Cell pellets were re-suspended in lysis buffer 1 supplemented with cOmplete\u2122 protease inhibitors, 1 mM DTT and 20 mM imidazole. The samples were lysed by sonication, then the suspension was centrifuged at 56,000 x g for 1 h, the soluble extract was further filtered through 0.8 \u00b5M pore size membrane filters (Millipore). The filtered fractions were collected and incubated with Ni\n \n 2+\n \n -NTA resin pre-equilibrate with lysis buffer 1 for 2 h with gentle rotation at 4\u00b0C. POI-bound resin was loaded on a gravity flow column, washed with lysis buffer 1 and the POI was eluted with lysis buffer 1 containing 400 mM imidazole. To remove the His\n \n 6/10\n \n -tags, 1/10 (w/w) of TEV protease was added and the sample and dialyzed against dialysis buffer overnight. Subsequently, the sample was diluted to 50 mM NaCl and loaded on a 5 ml StrepTrap HP column (Cytiva) pre-equilibrated with lysis buffer 1 containing 50 mM NaCl. After washing with lysis buffer 1 containing 50 mM NaCl, the POI was eluted in a linear gradient to lysis buffer 1 containing 1.5 M NaCl. Fractions containing the POI were combined, diluted to 50 mM NaCl, loaded on a 5 ml HiTrap Heparin HP column, washed and eluted in a linear gradient with lysis buffer 1 containing 1.5 M NaCl. Fractions containing the POI were pooled, concentrated and further purified by SEC on a Superdex 200 10/600 GL column (Cytiva) in 20 mM HEPES-NaOH, pH 7.5, 300 mM NaCl, 1 mM DTT. Fractions containing the POI were combined, concentrated, aliquoted, flash-frozen in liquid nitrogen and stored at -80\u00b0C.\n
\n\n For production of isolated ASC1 variants, the corresponding pETM-11 vectors were transformed into\n \n Escherichia coli\n \n BL21 (DE3) cells by electroporation for protein production\n \n via\n \n auto-induction at 18\u00b0C.\n \n \n 42\n \n \n Cells were harvested when cultures reached an optical density (600 nm) of 10. Cell pellets were re-suspended in lysis buffer 1 and supplemented with cOmplete\u2122 protease inhibitors. After sonication, the lysate was cleared by centrifugation. The POI was captured on Ni\n \n 2+\n \n -NTA resin in a gravity flow column, washed with lysis buffer 1 and eluted with lysis buffer 1 containing 400 mM imidazole. Fractions enriched for the POI were supplemented with 1/10 (w/w) TEV protease and dialyzed against dialysis buffer overnight. Dialyzed samples were passed through a Ni\n \n 2+\n \n -NTA gravity flow column to remove the cleaved His\n \n 6\n \n -tag and TEV. For ASC1\n \n FL\n \n , ASC1\n \n 152\u2013581\n \n , ASC1\n \n 281\u2013581\n \n and ASC1\n \n 403\u2013581\n \n fragments, the samples were diluted to 100 mM NaCl, loaded on a HiTrap Heparin HP column, washed and eluted in a linear gradient to lysis buffer 1 containing 1.5 M NaCl. Fractions containing the POI were combined, concentrated and further purified on a Superdex 200 16/600 GL column in SEC buffer.\n
\n\n For purification of the ASC1\n \n 1\u201380\n \n , ASC1\n \n 1\u2013230\n \n , ASC1\n \n 152\u2013230\n \n and ASC1\n \n 281\u2013403\n \n fragments, the Heparin column step was omitted and the final gel filtration was conducted in 20 mM HEPES-NaOH, pH 7.5, 150 mM NaCl, 1 mM DTT on a HiLoad 16/60 Superdex 75 pg column (Cytiva).\n
\n\n For production of ALKBH3, the corresponding pETM-11 vector was transformed into\n \n E. coli\n \n C2566 cells by electroporation for protein production\n \n via\n \n IPTG induction at 37\u00b0C. Cell pellets were re-suspended in 20 mM TRIS-HCl, pH 7.5, 500 mM NaCl, 10 mM imidazole, 1 mM DTT, 0.1 mM PMSF (lysis buffer 2), and lysed by sonication. The lysate was cleared by centrifugation. The supernatant was loaded onto a Ni\n \n 2+\n \n -NTA column, washed with lysis buffer 2 and the POI was eluted with a linear gradient to lysis buffer 2 containing 400 mM imidazole. Fractions enriched for the POI were combined, supplemented with 1/20 (w/w) TEV protease and dialyzed against dialysis buffer overnight. The sample was then diluted to 100 mM NaCl and loaded onto a HiTrap Heparin HP 5 ml column (Cytiva), pre-equilibrated with dialysis buffer containing 100 mM NaCl. After washing with dialysis buffer containing 100 mM NaCl, the POI was eluted with a linear gradient to dialysis buffer containing 1.5 M NaCl. The fractions containing the POI were pooled and concentrated with a centrifugal concentrator (10 kDa molecular mass cut-off). The concentrated sample was further purified by SEC on a Superdex 75 10/60 GL column (Cytiva) in 20 mM TRIS-HCl, pH 7.5, 250 mM NaCl, 1 mM DTT. Fractions containing the POI were combined, concentrated, aliquoted, flash-frozen in liquid nitrogen and stored at -80\u00b0C.\n
\n\n Analytical SEC-based interaction tests were conducted in 20 mM HEPES-NaOH, pH 7.5, 250 mM NaCl, 5% (v/v) glycerol, 1 mM DTT. 100 pmol of ASCC3\n \n HR\n \n were mixed with other proteins in a two to ten-fold molar excess in a final reaction volume of 80 \u00b5l. After incubation of the mixtures on ice for 30 min, the samples were loaded on a Superdex 200 3.2/300 analytical size exclusion column (Cytiva). 50 \u00b5l fractions were collected and subjected to SDS-PAGE analysis. Protein bands were visualized by Coomassie staining except for gels containing ASC1\n \n 1\u201380\n \n or ASC1\n \n 152\u2013230\n \n , which were imaged by silver staining.\n
\n\n For testing competitive binding of ASC1 and ALKBH3 to ASCC3\n \n HR\n \n , 120 pmol of ASCC3\n \n HR\n \n (or of pre-formed ASCC3\n \n HR\n \n -ASC1 complex) were mixed with 360 pmol each of ASC1 and ALKBH3 (or of ALKBH3) in a volume of 100 \u00b5l. After 30 min of incubation on ice, the samples were loaded on a Superdex 200 3.2/300 analytical size exclusion column. 50 \u00b5l fractions were collected and subjected to SDS-PAGE analysis. The proteins were visualized by Coomassie staining.\n
\n\n DNA duplex unwinding activity was assessed in fluorescence-based stopped-flow experiments on a SX-20MV spectrometer (Applied Photophysics).\n \n \n 26\n \n ,\n \n 27\n \n \n The DNA substrate contained a 12-base pair duplex region and a 31-nucleotide 3\u2019-ss overhangs, with an Alexa 488 fluorophore on the short strand and an Atto 540 Q quencher on the complementary strand ([(Atto 540 Q]5\u2019-\n \n GGCCGCGAGCCG\n \n GAAATTTAATTATAAACCAGACCGTCTCCTC-3\u2019; 5\u2019-\n \n CGGCTCGCGGCC\n \n -3\u2019[Alexa 488]; duplex region in bold). Reactions were carried out in 40 mM HEPES-NaOH, pH 7.5, 80 mM NaCl, 0.5 mM MgCl\n \n 2\n \n at 30\u00b0C. 250 nM protein or protein complex were pre-incubated with 50 nM DNA duplex for 5 min. 60 \u00b5l of the protein-DNA mixture were rapidly mixed with 60 \u00b5l of 4 mM ATP/MgCl\n \n 2\n \n , and the excited Alexa 488 fluorescence signal was recorded for 20 min using a 495 nm cutoff filter (KV 495, Schott). For each experiment, at least two individual traces were averaged, baseline-corrected by the fluorescence immediately after addition of ATP and normalized to the baseline-corrected maximum fluorescence. Data for ASCC3\n \n HR,D611A\n \n -based unwinding had been reported previously\n \n \n 20\n \n \n and are reproduced here to facilitate direct comparison. Data were plotted using GraphPad Prism 6.0 and fitted to a double exponential equation (fraction unwound\u2009=\n \n A\n \n \n fast\n \n *(1 \u2013 exp(\u2013\n \n k\n \n \n fast\n \n *\n \n t\n \n ))\u2009+\n \n A\n \n \n slow\n \n * (1 \u2013 exp(\u2013\n \n k\n \n \n slow\n \n *\n \n t\n \n ));\n \n A\n \n , total unwinding amplitude;\n \n k\n \n , unwinding rate constants [s\n \n \u2212\u20091\n \n ];\n \n t\n \n , time [s]).\n \n \n 25\n \n \n Amplitude-weighted unwinding rate constants were calculated as\n \n k\n \n \n uaw\n \n = (\n \n A\n \n \n fast\n \n *\n \n k\n \n \n \n fast\n \n \n \n 2\n \n +\n \n A\n \n \n slow\n \n *\n \n k\n \n \n \n slow\n \n \n \n 2\n \n ) / (\n \n A\n \n \n fast\n \n *\n \n k\n \n \n \n fast\n \n \n +\n \n A\n \n \n slow\n \n *\n \n k\n \n \n \n slow\n \n \n ).\n \n 25\n \n
\n\n Thin layer chromatography (TLC)-based ATPase assays were performed using [\u03b1-\n \n 32\n \n P]ATP (Hartmann Analytic).\n \n \n 26\n \n ,\n \n 27\n \n \n To quantify DNA-stimulated ATPase activity, 0.5 \u00b5M protein or protein complex were combined with 1 mM of a 43-nt ssDNA (5\u2019-GGCCGCGAGCCGGAAATTTAATTATAAACCAGACCGTCTCCTC-3\u2019). 0.5 \u00b5M protein or protein complex or equivalent protein-DNA mixtures were incubated with 1 mM [\u03b1-\n \n 32\n \n P]ATP in 50 mM HEPES-NaOH, pH 7.5, 80 mM NaCl, 5 mM MgCl\n \n 2\n \n , 2 mM DTT at 30\u00b0C for up to 60 min. 5 \u00b5l of sample were withdrawn at selected time points and reactions were quenched with 5 \u00b5l of 100 mM EDTA. 0.8 \u00b5l of the samples were spotted on a PEI-cellulose TLC plate and chromatographed with 1 M acetic acid, 0.5 M LiCl, 20% (v/v) ethanol. The corresponding ADP and ATP spots were visualized using a Storm 860 phosphorimager (GMI, USA) and quantified using ImageQuant software (version 5.2; Cytiva). Data were plotted and analyzed using Prism software (Graphpad, version 5), the ATPase activity was calculated as the number of hydrolyzed ATP molecules per protein molecule per minute, by fitting quantified data to the equation\n \n V\n \n = (\n \n A\n \n \n fast\n \n *\n \n V\n \n \n fast\n \n \n 2\n \n +\n \n A\n \n \n slow\n \n *\n \n V\n \n \n slow\n \n \n 2\n \n ) / (\n \n A\n \n \n fast\n \n *\n \n V\n \n \n fast\n \n +\n \n A\n \n \n slow\n \n *\n \n V\n \n \n slow\n \n );\n \n A\n \n \n fast\n \n and\n \n A\n \n \n slow\n \n , amplitudes of ATP hydrolyzed in the rapid and slow phase, respectively;\n \n V\n \n \n fast\n \n and\n \n V\n \n \n slow\n \n , rates of the rapid and slow hydrolysis phases [min\n \n \u2212\u20091\n \n ];\n \n V\n \n , ATP hydrolyzed as a function of time [min\n \n \u2212\u20091\n \n ].\n
\n\n The sub-cellular localizations of the Flag/His-tagged versions of ASC1 were determined by immuno-fluorescence.\n \n \n 43\n \n \n 293 cell lines expressing Flag-tagged ASC1 variants were grown on coverslips and fixed using 4% (v/v) paraformaldehyde for 20 min before permeabilization using 0.1% (v/v) Triton-X-100 in PBS for 20 min. Cells were blocked using PBS supplemented with 10% (v/v) fetal bovine serum (FBS) and 0.1% (v/v) Triton-X-100 for 1 h, then treated for 2 h with an FITC-conjugated \u03b1-Flag M2 antibody (Sigma Aldrich F4049; 1:200) diluted in PBS containing 10% FBS and 0.1% Triton-X-100. Cells were washed, and coverslips were mounted using mounting media containing DAPI. Cells were imaged using a Nikon Ti2 2-E inverted microscope.\n
\n\n 293 cells expressing N- or C-terminally Flag/His-tagged versions of full-length or truncated ASC1 or the Flag tag were lysed by sonication in IP buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.5 mM EDTA, 0.1% (v/v) Triton-X-100, 10% (v/v) glycerol and cOmplete\u2122 protease inhibitors. Lysates were cleared of debris by centrifugation at 20,000 x g for 10 min, then the cleared lysates were incubated with \u03b1-Flag M2 magnetic beads (Sigma-Aldrich #M8823) for 2 h. The matrix was washed five times with IP buffer and complexes were eluted using 3xFlag peptide (Sigma Aldrich #SAE0194). Proteins were precipitated using 20% (w/v) trichloroacetic acid (TCA) and separated by SDS-PAGE. Western blotting was performed using antibodies against the Flag tag (Sigma-Aldrich F3165; 1:7500), ASCC1 (Proteintech #12301-1-AP; 1:500), ASCC2 (Proteintech #11529-1-AP; 1:1000) and ASCC3 (Proteintech #17627-1-AP; 1:1000).\n
\n\n For immuno-precipitation of HA-tagged ASC1 variants (ASC1\n \n wt\n \n , ASC1\n \n \u0394ZnF\n \n , ASC1\n \n L174A\u2009\u2212\u2009L180A\u2212I190A\n \n or ASC1\n \n C171A\u2009\u2212\u2009C184A\n \n ), the transfected 293T cells were resuspended in ice cold, high salt co-IP buffer (50 mM Tris-HCl, pH 7.9, 300 mM KCl, 10% [v/v] glycerol, 1% [w/v] Triton X-100, 1 mM DTT) supplemented with protease inhibitors. The cells were then lysed by sonication and allowed to rotate at -4\u00b0C to complete lysis. Lysates were cleared by centrifugation and diluted to 150 mM KCl using co-IP buffer without KCl. Anti-HA beads (Santa Cruz Biotechnology, sc-7392 AC) were then added to the samples, and after incubation at 4\u00b0C for 3.5 h, the beads were centrifuged and washed multiple times with 150 mM KCl co-IP buffer. Bound proteins were eluted with SDS-PAGE loading buffer and boiled before analysis\n \n via\n \n SDS-PAGE/western blot using antibodies against the HA-tag (Abcam EPR22819-101, 1:4000) and ASCC3 as described previously\n \n \n 13\n \n \n .\n
\n\n The wt and ASC1 KO PC-3 cells were plated on a 96-well plate with 3,500 cells per well. Cells were exposed to media containing variable concentrations of MMS for 24 h at 37\u00b0C. Then, cells were recovered with fresh culture medium for an additional 48 h at 37\u00b0C. Cell viability was measured by using the MTS assay (Promega).\n
\n\n The ASCC3\n \n HR\n \n -ASC1 complex was prepared freshly in buffer 20 mM HEPES-NaOH, pH 7.5, 300 mM NaCl, 1 mM DTT, and concentrated to 4.15 mg/ml using a 50k ultra centrifugal filter (Merck). The sample was supplemented with 0.01% (w/v) n-dodecyl \u03b2-maltoside promptly before vitrification. 3.8 \u00b5l of the sample were applied to glow-discharged holey carbon R1.2/1.3 copper grids (Quantifoil Microtools, Germany) and plunge-frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher) equilibrated at 10\u00b0C and 100% humidity.\n
\n\n Data acquisition was conducted on a FEI Titan Krios G3i TEM operated at 300 kV equipped with a Falcon 3EC detector. Movies were taken for 40.57 s accumulating a total electron flux of ~\u200940 el/\u00c5\n \n 2\n \n in counting mode at a calibrated pixel size of 0.832 \u00c5/px distributed over 33 fractions.\n
\n\n All image analysis steps were done with cryoSPARC (version 3.2.2)\n \n \n 44\n \n \n . Movie alignment was done with patch motion correction generating Fourier-cropped micrographs (pixel size 1.664 \u00c5/px), CTF estimation was conducted by Patch CTF. Class averages of manually selected particle images were used to generate an initial template for reference-based particle picking from 6,022 micrographs. 2,818,857 particle images were extracted with a box size of 160 px and Fourier-cropped to 80 px for initial analysis. Reference-free 2D classification was used to select 1,590,881 particle images for further analysis.\n \n Ab initio\n \n reconstruction using a small subset of particles was conducted to generate an initial 3D reference for consecutive iterations of 3D heterogeneous refinement. 597,971 particle images were re-extracted with a box of 160 px and subjected to non-uniform refinement followed by CTF refinement. Another heterogeneous refinement round was applied to select 473,863 particle images for re-extraction at full spatial resolution after local motion correction (box size 320 px, 0.832 \u00c5/px). A final heterogeneous refinement run was conducted to select 244,064 particle images for non-uniform refinement and generate the final reconstruction at a global resolution of 3.4 \u00c5, locally extending down to 2.5 \u00c5.\n
\n\n AlphaFold-predicted models\n \n \n 24\n \n \n of ASCC3\n \n HR\n \n and of regions of ASC1 were manually placed in the cryoEM reconstruction and adjusted by rigid body fitting and segmental real-space refinement using Coot (version 0.8.9.1)\n \n \n 45\n \n \n . The model was refined by iterative rounds of real space refinement in PHENIX (version 1.17.1)\n \n \n 46\n \n \n and manual adjustment in Coot. Manual adjustments also took advantage of locally refined, focused cryoEM reconstructions. The structural model was evaluated with Molprobity (version 4.5.1)\n \n \n 47\n \n \n . Interface areas were analyzed\n \n via\n \n the PISA server (version 1.52)\n \n \n 48\n \n \n . Structure figures were prepared using ChimeraX (version 1.4)\n \n \n 49\n \n \n and PyMOL (version 1.8; Schr\u00f6dinger, LLC).\n
\n\n UV cross-linking was employed to generated zero length cross-links between protein and bound ssDNA oligos (T\n \n 12\n \n , T\n \n 24\n \n , T\n \n 36\n \n , T\n \n 48\n \n ). DNA oligos were 5\u2019-end labeled using [\u03b3-\n \n 32\n \n P]ATP and T4 polynucleotide kinase using a standard protocol. 10 \u00b5l reaction mixtures containing 100 nM (\u201c1\u201d in Fig.\n \n 5\n \n b) or 200 nM (\u201c2\u201d in Fig.\n \n 5\n \n b) protein or protein complex and 4.3 nM radio-labeled DNA probe were incubated in a 72-well microbatch plate (Greiner) in 50 mM HEPES-NaOH, pH 7.5, 80 mM NaCl, 5 mM MgCl\n \n 2\n \n , 2 mM DTT on ice for 5 min, then the samples were exposed to 254 nm UV irradiation for 10 min (Ultra-violet cross-linker, Amersham Life Science). Cross-linked samples were separated by SDS-PAGE and visualized by autoradiography using a Storm 860 phosphorimager.\n
\n\n For identifying cross-linked peptides and residues, 6.7 nM unlabeled T\n \n 48\n \n ssDNA were cross-linked to 200 nM ASCC3\n \n HR\n \n or ASCC3\n \n HR\n \n -ASC1 in 48 x 10 \u00b5l reactions as above and ethanol precipitated. Subsequent analyses were conducted in duplicates. The pellets were dissolved in 50 \u00b5l 4 M urea and diluted to 1 M Urea with 50 mM Tris-HCl, pH 7.5. To digest the DNA, 1 \u00b5l Universal nuclease (Pierce) and 1 \u00b5l Nuclease P1 (New England Biolabs) were added to the samples, followed by incubation at 37\u00b0C for 3 h. Protein digestion was performed with 1 \u00b5g of trypsin (Promega) overnight at 37\u00b0C. The samples were acidified with formic acid (FA; final concentration 0.1% [v/v]), and acetonitrile (ACN) was added to 5% (v/v) final concentration. Non cross-linked nucleotides were depleted by C18 reversed-phase chromatography with Harvard Apparatus MicroSpin columns. Sample was eluted by stepwise application of 50% (v/v) and 80% (v/v) ACN. Cross-linked peptides were enriched over linear peptides by TiO2 self-packed tip columns with 5% (v/v) glycerol as a competitor as described previously\n \n \n 50\n \n \n . The samples were dried under vacuum and resuspended in 10 to 15 \u00b5l of 2% (v/v) ACN, 0.05% (v/v) trifluoroacetic acid. 7 or 8 \u00b5l (first or second analysis) were used for LC-MS analysis.\n
\n\n Chromatographic separation was achieved with Dionex Ultimate 3000 UHPLC (Thermo Fischer Scientific) coupled with a C18 column packed in-house (ReproSil-Pur 120 C18-AQ, 1.9/3 \u00b5m pore size, 75 \u00b5m inner diameter, 30 cm length, Dr. Maisch GmbH). The flow rate was set to 300 nl/min, and a 44 min linear gradient was formed with mobile phase A (0.1% [v/v] FA) and B (80% [v/v] ACN, 0.08% [v/v] FA) from 8% or 10% (first or second analysis) to 45% mobile phase B. Data acquisition of eluting peptides was performed with Orbitrap Exploris 480 (Thermo Fischer Scientific). The resolution for survey scans was set to 120,000, the maximum injection time to 60 ms, the automatic gain control target to 100% or 250% (first or second analysis) and the dynamic exclusion to 9 s. Analytes selected for fragmentation were isolated with a 1.6 m/z window and fragmented with a normalized collision energy of 28. MS/MS spectra were acquired with a resolution of 30,000, a maximum injection time of 120 ms and an automatic gain control target of 100%.\n
\n\n Cross-link data analysis of the resulting raw files was performed with the OpenNuXL node of OpenMS (version 3.0.0)\n \n \n 51\n \n \n . Default general settings were used and the preset DNA-UV Extended was selected. The sequences of the proteins in the sample were provided as a database. The maximum length of DNA adducts was set to 3 and poly-T was used as sequence. The resulting .idxml files were used for annotation, and spectra were manually validated.\n
\n\n The cryoEM reconstruction of the ASCC3\n \n HR\n \n -ASC1 complex has been deposited in the Electron Microscopy Data Bank (\n \n \n https://www.ebi.ac.uk/pdbe/emdb\n \n \n \n \n ) under accession code EMD-15521 (\n \n \n https://www.ebi.ac.uk/pdbe/entry/emdb/EMD-15521\n \n \n \n \n ). Structure coordinates have been deposited in the RCSB Protein Data Bank (\n \n \n https://www.rcsb.org\n \n \n \n \n ) with accession code 8ALZ (\n \n \n https://www.rcsb.org/structure/8ALZ)\n \n \n \n \n .\n \n 52\n \n The DNA-protein CLMS data have been deposited in the ProteomeXchange Consortium (\n \n \n http://www.proteomexchange.org\n \n \n \n \n )\n \n via\n \n the PRIDE\n \n \n 53\n \n \n partner repository (\n \n \n https://www.ebi.ac.uk/pride/\n \n \n \n \n ) under dataset identifier PXD036106 (\n \n \n https://www.ebi.ac.uk/pride/archive/projects/PXD036106\n \n \n \n \n ). All other data are contained in the manuscript or the Supplementary Information. Source data are provided with this paper.\n
\n\n Supplementary Tables and Figures\n
\n \n\n Supplementary Dataset 1\n
\n \n\n Supplementary Dataset 2\n
\n \n\n Supplementary Dataset 3\n
\n \n\n Supplementary Dataset 4\n
\n \n\n Supplementary Dataset 5\n
\n \n\n Supplementary Dataset 6\n
\n \n\n Supplementary Dataset 7\n
\n \n\n Supplementary Dataset 8\n
\n \n\n Space charge layers (SCLs) formed at grain boundaries (GBs) are considered to critically influence the properties of polycrystalline materials such as ion conductivities. Despite the extensive researches on this issue, the presence of GB SCLs and their relationship with GB orientations, atomic-scale structures and impurity/solute segregation behaviors remain controversial, primarily due to the difficulties in directly observing charge distribution at GBs. In this study, we directly observe electric field distribution across the well-defined yttria-stabilized zirconia (YSZ) GBs by tilt-scan averaged differential phase contrast scanning transmission electron microscopy. Our observation clearly reveals the existence of SCLs across the YSZ GBs with nanometer precision, which are significantly varied depending on the GB orientations and the resultant core atomic structures. Moreover, the magnitude of SCLs show a strong correlation with yttrium segregation amounts. This study provides critical insights into the complex interplay between SCLs, orientations, atomic structures and segregation of GBs in ionic crystals.\n
\n\n The study of space charge layers (SCLs) in oxide grain boundaries (GBs) has attracted significant attention due to their impact on various material properties, such as crystal growth, ion conductivity, and electron conductivity\n \n \n 1\n \n ,\n \n 2\n \n ,\n \n 3\n \n \n . Many solid-state O-ion and Li-ion conductors are material systems where GB SCLs play a crucial role in determining transport properties\n \n \n 3\n \n ,\n \n 4\n \n ,\n \n 5\n \n ,\n \n 6\n \n \n . Yttria-stabilized cubic zirconia (YSZ), for instance, is widely used as an electrolyte for solid oxide fuel cells due to its high oxygen ionic conductivity and thermal stability\n \n \n 7\n \n ,\n \n 8\n \n \n . However, numerous studies have indicated that the ionic conductivity across the GBs is significantly lower than that within the grains, and this has been attributed to the formation of SCLs\n \n \n 9\n \n ,\n \n 10\n \n \n . According to the SC theory\n \n \n 4\n \n ,\n \n 10\n \n ,\n \n 11\n \n \n , when the GB core of YSZ is assumed to be positively charged, the positively charged mobile carriers, in this case oxygen vacancies, can be expelled out from the GB core region. Such depletion of oxygen vacancies and the resultant potential barriers can retard oxygen ion conductivity across the YSZ GBs\n \n \n 4\n \n \n . A thorough analysis of SCLs is thus essential for advancing our understanding of the properties and developing ion conductors with better performances.\n
\n\n So far, some experimental and computational methods have been employed to investigate SCLs\n \n \n 11\n \n ,\n \n 12\n \n \n . Conventionally, SCLs have primarily been indirectly investigated through macroscopic measurements of polycrystalline materials using electrochemical impedance spectroscopy\n \n \n 13\n \n \n . However, such macroscopic measurements are unable to identify the individual contribution of GBs in polycrystalline materials. Since each GB possesses distinct characteristics depending on its orientation and resultant atomic structures\n \n \n 14\n \n ,\n \n 15\n \n ,\n \n 16\n \n ,\n \n 17\n \n \n , the SCLs are expected to significantly vary in each GB. Thus, there is a pressing need for the direct observation of the SCLs of individual GBs in conjunction with their orientations, atomic structures and chemistry, in order to design and develop materials with better properties. To address this need, direct observation of SCLs using transmission electron microscopy (TEM) or scanning TEM (STEM)\n \n \n 18\n \n ,\n \n 19\n \n \n has been attempted continuously. However, in the previous studies using S/TEM, the effect of diffraction contrast, which arises from the changes in local structures and distortions around GBs, has made it extremely difficult to extract true and quantitative SCL signals. As a result, comprehensive understanding of SCLs at GBs still remains elusive.\n
\n\n Previous studies have demonstrated that differential phase contrast (DPC) STEM can directly visualize local electromagnetic field distribution inside materials from nanometer to sub-atomic scale\n \n \n 20\n \n ,\n \n 21\n \n ,\n \n 22\n \n ,\n \n 23\n \n ,\n \n 24\n \n \n . In recent years, tilt-scan averaged DPC STEM (tDPC STEM) has been developed for minimizing diffraction contrast effects and extracting true electromagnetic field signals at crystalline interfaces\n \n \n 23\n \n ,\n \n 25\n \n ,\n \n 26\n \n ,\n \n 27\n \n \n . Figure\n \n 1\n \n shows a schematic illustration of tDPC STEM technique. In tDPC, the incident-beam-tilt conditions are systematically changed while the incident electron probe is stationary at the same sample position. Subsequently, the bright field (BF) disks under multiple beam-tilt conditions are averaged on the detector plane. To a good approximation, the Coulomb deflection of the BF disk by electric fields is insensitive to minor changes in the beam-tilt condition. By contrast, the diffraction contrast is highly sensitive to even slight changes in the beam-tilt condition. Therefore, the electric field component in the DPC signals is mostly unchanged and reinforced by tDPC, whereas the diffraction contrast component changed by beam tilting is effectively averaged and suppressed. Thus, tDPC STEM can extract true electric field signals by suppressing diffraction contrast. By employing tDPC STEM, we recently succeeded in extracting the true electric field signals related to the charge inhomogeneity across semiconductor heterointerfaces quantitatively\n \n \n 23\n \n \n . Moreover, other STEM methods such as high-angle annular dark field (HAADF)\n \n \n 28\n \n \n and energy dispersive X-ray spectroscopy (EDS)\n \n \n 29\n \n \n , can also be used in DPC STEM, which provides us a comprehensive understanding towards the correlation between local atomic structures, chemistry and electromagnetic fields.\n
\n\n In the present study, we employed the tDPC STEM to directly observe the SCLs formed at four different model YSZ GBs. The amount of the SC and the core charge was quantified by fitting the experimental electric field profiles with a linear model\n \n \n 3\n \n \n . We show that it is now possible to directly and quantitatively characterize SCLs at individual GBs. Furthermore, the atomic structures and compositions of these model GBs were thoroughly investigated by HAADF STEM and STEM-EDS, which allow us to establish the one-by-one correlations between atomic structures, segregation behaviors and core charges associated with SCLs.\n
\n\n The atomic structures and segregation behaviors of the four GBs were first studied. Figure\n \n 2\n \n shows HAADF STEM images of the four coincident-site-lattice YSZ GBs of\n \n a\n \n ,\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n ,\n \n b\n \n ,\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n ,\n \n c\n \n ,\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n and\n \n d\n \n ,\n \n \n \\(\\text{\u22113}\\left[\\text{110}\\right]\\text{/(111)}\\)\n \n \n , respectively. The bicrystal fabrication procedure is described in the Methods. The GB core atomic structures are clearly resolved, which are consistent with the previous studies\n \n \n 30\n \n ,\n \n 31\n \n ,\n \n 32\n \n ,\n \n 33\n \n ,\n \n 34\n \n \n . Atomic-resolution STEM-EDS elemental maps across the GBs are shown in Supplementary Fig.\n \n S1\n \n . Quantitative line profiles of the EDS maps are also shown in Supplementary Fig. S2. Segregation of Y to the GB cores was observed for all the GBs, which was strongly dependent on the GB orientations. It is noted that, although no obvious preferential segregation sites exist in the atomic-scale maps, strong Y segregation was detected in the \u22115[001]/(310) and \u22119[110]/(221) GBs as a total. On the other hand, the\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n shows very small Y segregation. In the\n \n \n \\(\\text{\u22113}\\left[\\text{110}\\right]\\text{/(111)}\\)\n \n \n , Y segregation was clearly detected at specific GB atomic sites. This preferential segregation should be triggered by the strain relaxation of Y\n \n 3+\n \n with a larger ionic size than Zr\n \n 4+ 33\n \n (more details will be discussed later). It is also noted that Al and Si impurity segregations were observed for all the GBs except for the\n \n \n \\(\\text{\u22113}\\left[\\text{110}\\right]\\text{/(111)}\\)\n \n \n . These impurity atoms may be introduced during the bicrystal fabrication processes\n \n \n 35\n \n ,\n \n 36\n \n \n .\n
\n\n Next, the electric field distribution across the GBs were analyzed by tDPC STEM. Figure\n \n 3\n \n shows the horizontal electric field component images and the corresponding line profiles (averaging across the field of view) of the\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n ,\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n GBs, respectively. From the line profiles, sharp and narrow positive and negative electric field peaks (indicated by red arrows) corresponding to the electric fields converging towards the GB cores, were clearly observed for both GB cores. In addition, in the \u22115[001]/(310) GB, a gentle and wide leftward electric field peak on the left side of the GB, and a gentle and wide rightward electric field peak on the right side of the GB were observed, as indicated by blue arrows (Fig.\n \n 3\n \n a and\n \n b\n \n ). These signals indicate the presence of a divergent electric field from the GB cores, with approximately 15 nm width on each side of the GB. Conversely, almost no divergent electric fields are found in the\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n GB (Fig.\n \n 3\n \n c and\n \n d\n \n ).\n
\n\n To interpret these electric field profiles, possible electrostatic potential and associated electric field profiles across GBs with SCLs are schematically illustrated in Fig.\n \n 4\n \n . From electrostatic point of view, if a GB core is positively charged and negative SCLs exist in adjacent to it, the GB will exhibit positive potential, and result in a diverging electric field from the GB core (Fig.\n \n 4\n \n a). On the other hand, GB electrostatic potential and resultant electric field profile can also be affected by the structural feature of the GB core. Since the local atomic density of the GB core can be intrinsically lower than that of bulk and the thickness of GB is often thinner than the bulk regions due to the selective etching by ion milling, there can be a dip in the mean inner potential along the GB core even though the GB is not charged. As a result, the GB exhibits a sharp converging electric field towards its core (Fig.\n \n 4\n \n b). Consider a case of positively charged GB core and negatively charged SCL in real case, where the above two effects coexist in the GB, the overlap of the two potentials shown in Fig.\n \n 4\n \n a and\n \n b\n \n is expected, and an overlapped electric field profiles should be observed as a total electric field profile by tDPC STEM (Fig.\n \n 4\n \n c). Our experimental observations are in consistence with such schematic profile shown in Fig.\n \n 4\n \n c, indicating that the core of GBs in YSZ is positively charged. Furthermore, it can be concluded that a large amount of SCs exists in\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n GB, while the amount of SCs is very small in\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n GB.\n
\n\n Figure\n \n 5\n \n shows the line profiles of the horizontal electric field component images for all the GBs studied here. Note that the plot range of Fig.\n \n 5\n \n is modified from that in Fig.\n \n 3\n \n in order to emphasize the diverging electric fields of each GB. The original profiles are shown in Supplementary Fig. S3. Strong converging electric field peaks are found at the cores for all the GBs, which can be attributed to the structural-induced abrupt mean inner potential decreases at the cores as discussed above. On the other hand, the diverging electric fields surrounding the GB cores were significantly different among the four GBs. The magnitude of the diverging electric fields follows in the order of\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n >\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n >\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n >\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n . The differences in the diverging electric fields can be attributed to the amount of core charges, which equal to the accumulated SCs.\n
\n\n Next, we attempted to quantify the SCs and core charges from the distribution of the diverging electric fields. Here, we fitted the electric field line profiles obtained via the tDPC STEM experiment shown in Fig.\n \n 5\n \n using a linear model\n \n \n 3\n \n ,\n \n 37\n \n \n . The linear model assumes that the intrinsic core charges,\n \n \n \\({\\sigma }_{\\text{c}\\text{o}\\text{r}\\text{e}}\\)\n \n \n , exist as a sheet charge at the GB core and that the volume density of associated SCs,\n \n \n \\({\\rho }_{\\text{S}\\text{C}}\\)\n \n \n , around the GB remains constant within the SCLs. Let\n \n \n \\(x\\)\n \n \n direction be perpendicular to the GB, and the GB core be placed at\n \n \n \\(x=0\\)\n \n \n . Then, the electric field around the GB,\n \n \n \\({E}_{\\text{G}\\text{B}}\\)\n \n \n , can be described as:\n
\n\n \n \n \\(\\begin{array}{c}{E}_{\\text{G}\\text{B}}=\\left\\{\\begin{array}{c}0 x<-{l}_{\\text{S}\\text{C}}\\\\ -\\frac{{\\rho }_{\\text{S}\\text{C}}}{{l}_{\\text{S}\\text{C}}}x-{\\rho }_{\\text{S}\\text{C}} -{l}_{\\text{S}\\text{C}} \\le x<0,\\\\ -\\frac{{\\rho }_{\\text{S}\\text{C}}}{{l}_{\\text{S}\\text{C}}}x+{\\rho }_{\\text{S}\\text{C}} 0<x\\le {l}_{\\text{S}\\text{C}},\\\\ 0 {l}_{\\text{S}\\text{C}}<x\\end{array} \\right.\\#\\left(1\\right)\\end{array}\\)\n \n \n
\n\n where\n \n \n \\({l}_{\\text{S}\\text{C}}\\)\n \n \n is the width of SCLs. In order to satisfy the charge neutrality condition, the sheet positive charges at the GB core and the total amount of negative SCs are set to be equal. Then,\n \n \n \\({\\sigma }_{\\text{c}\\text{o}\\text{r}\\text{e}}\\)\n \n \n and\n \n \n \\({\\rho }_{\\text{S}\\text{C}}\\)\n \n \n can be described as,\n
\n\n We first obtained\n \n \n \\({\\rho }_{\\text{S}\\text{C}}\\)\n \n \n by fitting the divergent electric field profiles without the core region using Eq.\u00a0(1). Although structurally induced abrupt mean inner potential decrease at the cores is superimposed in the electric field profiles (Fig.\n \n 5\n \n ), which is difficult to quantify, the sheet core charges\n \n \n \\({\\sigma }_{\\text{c}\\text{o}\\text{r}\\text{e}}\\)\n \n \n can still be quantified using Eq.\u00a0(2). Table\n \n 1\n \n shows the results of charge quantification for each GB, along with the amount of Y segregation estimated by STEM-EDS. The total amount of sheet core charges was different, resulting in the order of\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n >\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n >\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n >\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n . These findings clearly show that the charge distribution around the GB significantly differs depending on the GB orientation. Moreover, even in the GBs with the same sigma values (\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n and\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n ) and with the same rotation axis (\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n and\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n , and\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n and\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n ), the amount of SCs and core charges significantly differ.\n
\n| \n \n Grain boundary\n \n | \n \n \n Core charge\n \n\n [electron cm\n \n -\n \n 2\n \n \n ]\n \n | \n \n \n Space charge\n \n\n [electron cm\n \n -\n \n 3\n \n \n ]\n \n | \n \n \n Y segregation\n \n\n [%]\n \n | \n
|---|---|---|---|
| \n \n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n \n | \n \n \n 5.5\u2009\u00b1\u20090.1\u00d710\n \n 13\n \n \n | \n \n \n 1.47\u2009\u00b1\u20090.05\u00d710\n \n 19\n \n \n | \n \n \n 24\u2009\u00b1\u20094\n \n | \n
| \n \n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n \n | \n \n \n 6.0\u2009\u00b1\u20090.2\u00d710\n \n 13\n \n \n | \n \n \n 1.2\u2009\u00b1\u20090.1\u00d710\n \n 19\n \n \n | \n \n \n 25\u2009\u00b1\u20094\n \n | \n
| \n \n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n \n | \n \n \n 1.4\u2009\u00b1\u20090.1\u00d710\n \n 13\n \n \n | \n \n \n 0.28\u2009\u00b1\u20090.05\u00d710\n \n 19\n \n \n | \n \n \n 10\u2009\u00b1\u20094\n \n | \n
| \n \n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n \n | \n \n \n 0.1\u2009\u00b1\u20090.2\u00d710\n \n 13\n \n \n | \n \n \n 0.0\u2009\u00b1\u20090.2\u00d710\n \n 19\n \n \n | \n \n \n 35\u2009\u00b1\u20093\n \n | \n
\n Next, we explore the relationship between the charge distribution and Y segregation amounts at the GBs. In the four GBs, different amounts of Y segregation have been observed (Supplementary Fig.\n \n S1\n \n and the previous studies\n \n \n 32\n \n ,\n \n 33\n \n ,\n \n 34\n \n \n ), and the segregation amount is summarized in Table\n \n 1\n \n . For the\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n GB, Y segregation at very specific GB atomic sites has been observed\n \n \n 33\n \n \n . The segregation structure of the\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n GB can be perfectly reproduced via Monte Carlo simulations combined with static lattice calculations without taking into account any GB core charges. It has been pointed out that the undercoordinated GB atomic sites are considered to be the origin of Y segregation, in order to minimize the local strain\n \n \n 33\n \n \n . Such scenario also agrees with our present result, that the SCs and core charges at the\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n GB were essentially negligible within the measurement error of the present tDPC STEM. These results suggest that Y segregation to the\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n GB is induced solely by the elastic energy minimization mechanism, and has little correlation to GB core charges. On the other hand, it is found that the amount of Y segregation shows a positive correlation with the amount of core charges in the\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n ,\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n and\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n GBs. The\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n and\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n GBs show pronounced Y segregation, despite the absence of specific GB core segregation sites like the\n \n \n \\(\\text{\u22113}\\left[\\text{11}\\text{0}\\right]\\text{/(111)}\\)\n \n \n GB. At the\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n GB some preferential segregation sites could be observed, but Y segregation in total is notably smaller than the\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n and\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n GBs. These results suggest that the electrostatic interaction between positively charged GB core and Y might be the major origin of Y segregation in general GBs. It is thus revealed that Y segregation behaviors can be determined by the balance of both electrostatic and elastic interactions, which are strongly dependent on the GB orientations and resultant core atomic structures.\n
\n\n Finally, we briefly discuss the origin of the positive core charges. The origin of the positive core charges in YSZ GBs has conventionally been explained by the difference in the chemical potential of oxygen vacancies at the GB core and within the grains\n \n \n 11\n \n \n . In this scenario, oxygen vacancies should abruptly accumulate at the GB core. The oxygen distribution has been studied in detail in our previous report\n \n \n 34\n \n \n . STEM EDS in that study did not detect such abrupt oxygen vacancy accumulation in those GBs\n \n \n 34\n \n \n . Other mechanisms of positive core charges have been also proposed, including the existence of unintentional impurities such as Si or Al\n \n \n 18\n \n ,\n \n 38\n \n \n , or non-stoichiometric anion-cation broken bonds at the GB cores\n \n \n 39\n \n \n . Our STEM-EDS mapping clearly show that Si and Al impurities indeed segregated to the GBs (Supplementary Fig. S2). If these impurity cations substitute Zr sites, they should not cause positive formal charges at the GB cores. Conversely, if Al substitutes Zr, it should cause negative formal charges. On the other hand, if these impurities occupy interstitial sites, they should cause positive charges. The Si and Al EDS maps suggest that some of the segregated impurity atoms may occupy the interstitial sites at the GB cores. Thus, the impurity segregation at the core can be one of the origins for the positive core charges. On the other hand, the two GBs which show the large SCLs,\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n and\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n GBs, exhibit incoherent GB core atomic structures (Fig.\n \n 2\n \n ). These results may still support the broken bond mechanism as the origin of the positive core charges. As the bonding environment at GB is generally much different from those within the bulk, our conclusion could be applicable to those general GBs inside practical polycrystalline materials. To address the true origin of the positive core charges quantitatively, first-principles calculations with very long-range cells that account for impurity segregation and charge inhomogeneities should be necessary, which is technically difficult to realize at the moment and beyond our scope of this study. However, the ability to directly observe SCLs at individual, well-defined GBs in conjunction with their atomic-scale structures and chemistry finally open the possibility for fundamental understanding of the correlation between SCLs, atomic structures and segregation behaviors of GBs in many oxide materials.\n
\n\n We directly observed the electric field distribution across the four different YSZ GBs using tDPC STEM. We found that the SCLs are formed at the YSZ GBs, but the magnitude is strongly dependent on their GB orientations. Moreover, the amount of SCs and the amount of Y segregation show a positive correlation for most of the GBs, where the segregation is mainly dominated by the electrostatic effects. The present approach of directly observing SCLs should pave the way for fundamental understanding of the complex interplays between GB orientations, GB core atomic structures, impurity/solute segregation and SCLs in many oxide materials and devices.\n
\n\n We fabricated four well-defined YSZ bicrystals as model samples by a diffusion bonding method. Two YSZ (10 mol% Y\n \n 2\n \n O\n \n 3\n \n ) single crystals were precisely cut and joined at 1600\u00b0C for 15h in air\n \n \n 32\n \n \n , forming the bicrystals with\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(310)}\\)\n \n \n ,\n \n \n \\(\\text{\u22115}\\left[\\text{001}\\right]\\text{/(210)}\\)\n \n \n ,\n \n \n \\(\\text{\u22119}\\left[\\text{110}\\right]\\text{/(221)}\\)\n \n \n , and\n \n \n \\(\\text{\u22113}\\left[\\text{110}\\right]\\text{/(111)}\\)\n \n \n GBs. TEM specimens were prepared via mechanical polishing and Ar ion-beam milling. The sample thickness was estimated to be approximately 30 to 80 nm for each GB using STEM electron energy loss spectroscopy\n \n \n 40\n \n \n .\n
\n\n The atomic structure and Y segregation of the GBs were examined by HAADF-STEM and STEM-EDS using an aberration-corrected STEM with dual-EDS detectors (JEM-ARM200CF, JEOL). The accelerating voltage and the convergence semi-angle were set to 200 keV and 24 mrad, respectively. NSS3 spectral analysis software (Thermo Fisher Scientific Inc.) was used to perform the EDS analysis. The details of the experimental conditions were as per the previous studies\n \n \n 33\n \n ,\n \n 34\n \n \n .\n
\n\n Electric field distribution was mapped via tDPC STEM using the magnetic-field-free atomic resolution STEM with 40-segmented detector\n \n \n 22\n \n \n and tilt-scan system\n \n \n 26\n \n \n (JEM-ARM200CF equipped with a magnetic-field-free objective lens, JEOL)\n \n \n 41\n \n \n . The accelerating voltage and the convergence semi-angle were set to be 200 keV and 2 mrad, respectively. The probe current and the expected probe size were approximately 12 pA and 0.7 nm, respectively. 61-beam-tilt conditions, of which the maximum-tilt angle was set to be 8 mrad, were generated by the tilt coils above the probe corrector, and the tilted beam converged into one BF disk on the detector by the other tilt coils below the objective lens\n \n \n 23\n \n \n . The center of mass (CoM) of the BF disk was measured for imaging quantitative electric field maps inside the specimens\n \n \n 42\n \n \n . The CoM value was measured by weighting the electron intensity of each detector segment by the geometric center of mass of the detector segment at each raster position. The tDPC images were denoised by excluding signals incompatible with the Poisson equation via discrete cosine transformation\n \n \n 43\n \n \n . The residual diffraction contrast was evaluated as per the previous study\n \n \n 23\n \n ,\n \n 27\n \n \n using electron-diffraction simulation\n \n \n 44\n \n \n .\n
\n\n SCs and core charges were quantified by fitting the experimental electric field profiles to the Eq.\u00a0(1), convolved with the probe size, using the Markov chain Monte Carlo method with the Metropolis-Hasting algorithm\n \n \n 45\n \n \n .\n
\n\n Image calculation, analysis, fitting, and display of the results were performed using common Python3 packages such as Numpy and Scikit-image.\n
\n\n Supplementary figure\n
\n \n\n The Arctic is rapidly losing its sea ice cover while the region warms faster than anywhere else on Earth. As larger areas become ice-free for longer, winds strengthen and interact more with open waters. Higher waves can increase coastal erosion and flooding, threatening communities and releasing permafrost carbon. However, the future trajectory of these changes remains poorly understood as instrumental observations and geological archives remain rare and short. Here, we address this critical knowledge by presenting the first continuous Holocene-length reconstruction of Arctic wind and wave strength using coastal lake sediments from Svalbard. Exposed to both polar Easterlies and Westerly storm tracks, sheltered by a bedrock barrier, and subjected to little post-glacial uplift, our study site provides a uniquely stable baseline to assess long-term changes in the region's dominant wind systems. To do so with high precision, we rely on multiple independent lines of proxy evidence for wind- and wave-blown sediment input. Our reconstructions reveal quasi-cyclic wind maxima during regional cold periods, and therefore challenge the prevalent view that a warmer less icy future Arctic will be stormier.\n
\n\n The Arctic responds faster to on-going climate change than any other region on Earth\n \n \n 1\n \n \u2013\n \n 3\n \n \n . Over the past forty years, amplified warming has progressed nearly four times faster than the global average\n \n \n 2\n \n \n . This dramatic transformation is most visibly manifested by a rapid decline of the Arctic\u2019s sea ice cover\n \n \n 4\n \n ,\n \n 5\n \n \n . As a result, larger areas remain ice-free longer, and the fetch \u2013 the distance over which wind can interact and transfer energy to surface waters \u2013 expands\n \n \n 6\n \n \u2013\n \n 9\n \n \n . Associated increases in wave height and frequency are further exacerbated by the vulnerability of thinning remnant ice to wind-driven fracturing\n \n \n 10\n \n \n .\n
\n\n While observations are rare and mostly local, these changes have increased wave height by up to 30 cm per decade in some Arctic areas\n \n \n 11\n \n \n . Worryingly, the impact of wave energy, heightened by permafrost degradation and sea-level rise, is the main driver of coastal erosion along vast tracts of Arctic shoreline\n \n \n 12\n \n \u2013\n \n 18\n \n \n . Already, retreat rates have increased by more than 50% along some permafrost-rich coastal sections in the past two decades\n \n \n 13\n \n \n . Besides posing a threat to coastal environments and communities\n \n \n 16\n \n \n , erosion is also associated with the release of more carbon than all the region's rivers combined\n \n \n 19\n \n \n , which could result in significant greenhouse gas emissions\n \n \n 20\n \n \n . Notwithstanding uncertainties, pan-Arctic coastal erosion rates may double by the end of this century under the business-as-usual greenhouse gas emissions scenario\n \n \n 17\n \n \n .\n
\n\n Despite the afore-mentioned environmental and socio-economic impacts, the magnitude of future changes in Arctic storminess under warmer and less icy conditions remains poorly constrained. This is well-illustrated by the divergence between climate models \u2013 while some suggest a weakening and southward migration of the mid-latitude Westerlies\n \n \n 21\n \n \n , others present evidence for a poleward shift of these storm tracks\n \n \n 22\n \n \n . Moreover, the contribution of mechanical wave erosion to coastal erosion is poorly parameterized, and estimates differ up to 20%\n \n 17\n \n .\n
\n\n As climate models are calibrated with observations, projections become more uncertain when variability exceeds the range of the brief instrumental record\n \n \n 23\n \n \n . Paleoenvironmental data from geological archives are well-suited to fill this critical knowledge gap, by providing us with longer-term baseline data on the links between changes in sea ice, storminess, and coastal erosion under different climate conditions. Arctic coastal deposits are often well-preserved as isostatic uplift rates have typically outpaced global sea-level rise since deglaciation\n \n \n 24\n \n ,\n \n 25\n \n \n , potentially preserving coastal sediment sequences that cover the full Holocene. By effectively trapping eolian particles and sea-spray aerosols\n \n \n 26\n \n ,\n \n 27\n \n \n , sediments from coastal lakes are prime archives to record past changes in wind strength and wave height \u2013 henceforth reffered to as\n \n storminess\n \n . Critically, a new generation of high-fidelity sediment core scanning techniques and geochronological advances allow us to reconstruct past changes on human-relevant (decades to centuries) timescales\n \n \n 27\n \n \n . However, while several lake sediment-based North Atlantic wind reconstructions have been published in recent years\n \n \n 29\n \n ,\n \n 33\n \n ,\n \n 34\n \n \n , the potential of coastal Arctic lakes to record changes in paleostorminess remains under-utilized\n \n \n 28\n \n \n .\n
\n\n Here, we present the first continuous Holocene-length lake sediment-based paleostorminess reconstruction from Svalbard \u2013 an Arctic climate change hotspot\n \n \n 31\n \n \n . This Archipelago is uniquely sensitive to changes in key drivers of storminess as both warming and sea ice melt rates exceed the regional average\n \n \n 32\n \n ,\n \n 35\n \n \n . We analyze a\u2009~\u20099,500 year long sediment sequence from the Archipelago`s southern tip \u2013 S\u00f8rkapp\u00f8ya island. Protected by a bedrock barrier and exposed to little post-glacial emergence\n \n \n 36\n \n \n , our study site \u2013 coastal Lake Steinbruvatnet \u2013 provides a stable baseline to assess Holocene changes. To rigorously reconstruct storminess on multicentennial to millennial timescales, we pioneer a multi-proxy approach that combines independent geochemical (X-Ray Fluorescence; XRF), visual (Computed Tomography; CT), and granulometric (End-Member Modelling Analysis; EMMA) lines of evidence for wind- and wave-transported particles in a geostatistical (Principal Component Analysis; PCA) framework. Our findings suggest that Holocene wave and wind maxima occurred during cold periods, and thus challenge the widely held notion that a warmer and less icy future Arctic will be stormier.\n
\n\n S\u00f8rkapp\u00f8ya is a 7 km-long island off the southern tip of Spitsbergen \u2013 the biggest island of the Svalbard Archipelago (Fig.\n \n 1\n \n a-b). In contrast with other parts of the Archipelago\n \n \n 36\n \n \n , adjacent S\u00f8rkapp Land has experienced only modest sea-level changes after deglaciation\u2009~\u200911,000\u20139,000 cal. yrs B.P. as shoreline uplift has not exceeded 10 m over the last 6,500 years\n \n \n 37\n \n ,\n \n 38\n \n \n , enhancing the preservation-potential of Holocene-length archives of coastal change (Fig.\n \n 1\n \n b)\n \n \n 36\n \n \n . The bedrock is composed of Palaeozoic and Mesozoic sedimentary and low-grade metamorphic rocks\n \n \n 39\n \n ,\n \n 40\n \n \n . In addition, large areas are covered by unconsolidated Quaternary deposits\n \n \n 39\n \n \n . The coastal geomorphology of the island is dominated by three major components:\n \n I)\n \n rocky ridges and spurs in the West (Fig.\n \n 1\n \n d and Supplementary Fig.\n \n S1\n \n b), which often constitute the structural anchor for\n \n II)\n \n uplifted beach ridges to the East (Fig.\n \n 1\n \n d), and\n \n III\n \n ) numerous coastal lakes separated from the sea by hooked spits and barriers (Fig.\n \n 1\n \n d).\n
\n\n Coastal lakes effectively capture the products of wave- and wind-blown input like sea-spray aerosols and minerogenic grains\n \n \n 26\n \n ,\n \n 27\n \n ,\n \n 33\n \n \n . To harness this potential, we target Lake Steinbruvatnet for this study (Fig.\n \n 1\n \n c), which is distinct from other lakes on S\u00f8rkapp\u00f8ya because of its unique setting. Notably, the basin is facing both polar Easterlies and Westerly storm tracks (Fig.\n \n 1\n \n b), so that wind- and wave-blown input might derive from both systems. Indeed, the presence of sandy shadow dunes to the West of Steinbruvatnet and silt sheets to the East of the lake indicate efficient inland eolian transport of sediment (Supplementary Fig.\n \n S1\n \n b-c). The observed East-West grain size difference can be traced back to the source of mobilized material: the West coast is characterized by a high (2\u20134 m) gravel-dominated storm ridge perched on a rocky shore platform (Fig.\n \n 1\n \n d; Supplementary Fig.\n \n S1\n \n b), while the East coast is characterized by a flatter\u2009~\u200950 m wide beach where many silty and sandy deposits can be found (Fig.\n \n 1\n \n d and Supplementary Fig.\n \n S1\n \n c). Moreover, the lake is protected from erosion and disturbance by storm surges as it is situated 5 m above sea-level (a.s.l.), and sheltered by an 8 m a.s.l. rocky ridge to the West as well as a 1 km wide beach ridge plain to the East (Fig.\n \n 1\n \n d). Finally, Lake Steinbruvatnet lacks an out- or inlet, limiting the potential for non-eolian catchment-derived minerogenic input, and is unaffected by the water level fluctuations seen in other local lakes.\n
\n\n Climatologically, available wind observations measured from 2013 onwards on S\u00f8rkapp\u00f8ya reveal that the easterlies dominate during wintertime (DJF), while wind directions are equally distributed in summer (JJA)\n \n \n 41\n \n \n . The westerlies are, however, generally weaker as wind speeds rarely (0.5% of the time) reach gale force, whereas the easterlies do so during on 10% of winter days\n \n \n 41\n \n \n . Also, timeseries analysis of Sentinel-2 satellite imagery reveals that the lake is ice-covered for ~\u20099 months per year\n \n \n 42\n \n \n . At present, our study area is situated close to the rapidly retreating seasonal sea ice maximum (Fig.\n \n 1\n \n b)\n \n \n 43\n \n \n , while biomarker (IP\n \n 25\n \n ) evidence suggests that seasonal sea ice only became widespread during the last millennium of the Holocene\n \n \n 44\n \n \n . On-going changes are closely linked to the 1\u00b0C per decade warming trend observed in the region\n \n \n 35\n \n \n . Today, local mean air temperatures remain below zero at -3.7\u02daC and the annual amount of precipitation averages 478 mm per year at nearby Hornsund station\n \n \n 45\n \n \n .\n
\n\n
\n\n All 9 radiocarbon dates taken from the core 601-21-6 GC (see Coring paragraph of our\n \n methods\n \n section and Table\n \n 2\n \n ) were incorporated in a linearly interpolated age model with the help of version 2.5 of the Clam R package\n \n \n 56\n \n \n . Ages were calibrated with IntCal20 curve and reported with a 2 sigma (2\u03c3) uncertainty range (cal. yrs B.P.; see Fig.\n \n 2\n \n and Table\n \n 2\n \n )\n \n \n 57\n \n \n . Stratigraphically inverted old ages were identified as outliers: we note that the age of all but one (Poz-150330: flagged by the lab because of a sample size that fell short of requirements for precise dating) of these cluster\u2009~\u200910,500 cal. yrs B.P. As only terrestrial plant macrofossils were dated (see Chronology paragraph in our\n \n methods\n \n section), we argue that these anomalous ages derive from reworked land deposits. In support of this evidence,\n \n \n 38\n \n \n suggest that local sea-level was up to 5 meters lower than today around this time, based on wave-scoured peat remains from adjacent S\u00f8rkapp Land that also date to ~\u200910,000 cal. yrs B.P. The same authors show that a transgression culminated in our study area close to the elevation of Lake Steinbruvatnet ca. 8,000 cal. yrs B.P. Based on these findings, we argue that the distinct decline in sediment accumulation rates (SARs) after ca. 7,900 cal. yrs B.P. (Fig.\n \n 2\n \n ) may be linked to progressive emergence from the sea \u0336 as the marine processes that often supply sediments in coastal lakes like Steinbruvatnet waned\n \n 2\n \n 9\n \n \n , accumulation slowed. Further supporting this evidence, we also note a decline in the Ca/Ti ratio, an often-used indicator of marine influence in similar settings\n \n 5\n \n 8\n \n \n . Based on its insignificant (anti)correlation with independent minerogenic indicators like CT density (Table\n \n 1\n \n ), we preclude that Ca derives from the traces of silicified limestone that underly parts of the catchment (also see our setting paragraph)\n \n 3\n \n 9\n \n \n . Regardless, both the afore-mentioned outliers as well as our ca. 9,600 cal. yrs B.P. basal age (Table\n \n 2\n \n ) suggest that the Steinbruvatnet catchment was isolated from the ocean multiple millennia earlier than previously reported\n \n 3\n \n 6\n \n \n . Finally, we note that sedimentation rates between our uppermost radiocarbon ages are indistinguishable from the values inferred for the core top (Fig.\n \n 2\n \n ), which is also constrained by the year of sediment collection \u2013 2021 C.E. This strengthens our confidence in the presented model and suggests that sediments from the top 7 cm of our core are reworked but complete.\n
\n\n As outlined the previous section, visual assessment reveals that sediment from the uppermost 7 cm of the investigated Steinbruvatnet record (core 601-21-6 GC: see methods) has been homogenized (Suppl. Fig. S2). We argue that this lack of structure stems from reworking, likely due to post-coring disturbance (mixing the sediment-water interface). We therefore exclude the uppermost 7 cm from further analysis. Following visual assessment of the record, we identify two main facies (Fig.\n \n 3\n \n a): dark brown background sediments (facies 1), and lighter-colored clastic horizons that vary in thickness from ~\u20092 mm to ~\u20092 cm (facies 2). The latter layers are also automatically captured by WlCount (see Statistics paragraph in our methods)\n \n \n 59\n \n \n . Our multi-proxy analysis furthermore reveals that facies 1 is organic, as reflected by higher Loss on Ignition (LOI) values and XRF incoherent/coherent (inc./coh.) counts (Fig.\n \n 2\n \n a, c), an often-used productivity proxy\n \n \n 60\n \n ,\n \n 61\n \n \n . In contrast, facies 2 layers are dense \u2013 as reflected by elevated Dry Bulk Density (DBD) and CT grayscale values\n \n \n 62\n \n \n , and minerogenic \u2013 as reflected by higher Total Scatter Normalized (TSN; see Stratigraphy paragraph in our\n \n methods\n \n section) XRF Titanium (Ti) counts\n \n \n 27\n \n ,\n \n 63\n \n \n , a conservative element that is broadly applied as an indicator of clastic terrigenous input\n \n \n 60\n \n \n . Ti is highly correlated with DBD and CT grayscale values (\u03c1\u2009=\u20090.81 and 0.61, respectively, n\u2009=\u200996,\n \n p\n \n =\u20090.000; Table\n \n 1\n \n ). Based on the characteristics of both facies and the distinct difference between them (Supplementary Fig. S3), we argue that facies 2 layers were deposited in a higher energy environment. Considering the coastal setting of the investigated Lake Steinbruvatnet and the absence of in- and outlets to mobilize catchment material (see our setting paragraph), we favor an eolian origin. This interpretation is supported by the strong co-variance between Ti and Bromine (Br) \u2013 a sea-spray indicator\n \n \n 64\n \n ,\n \n 65\n \n \n , here normalized to scattering (inc./coh.) rates to account for its association with productivity (Table\n \n 1\n \n )\n \n \n 60\n \n ,\n \n 61\n \n \n . Moreover, although Steinbruvatnet sits just 5 m above modern sea-level, we argue that the lake has not been directly impacted by storm surges since the onset of lacustrine sedimentation\u2009~\u20099,500 cal. yrs B.P. (see our setting paragraph), due to the absence of diagnostic features like erosive contacts between both facies, marine fossils, rip-up clasts, or event deposits in the catchment area\n \n \n 66\n \n ,\n \n 67\n \n \n . Finally, our record is devoid of the gravels that dominate the western beach (see setting, and Supplementary Fig.\n \n S1\n \n b) \u2013 the likeliest source of storm surges due to its proximity to Lake Steinbruvatnet and exposure to the North Atlantic (Greenland Sea) swell (Fig.\n \n 1\n \n d).\n
\n\n Complementing the above evidence that identifies variations in eolian input (facies 2), we explore the use of particle size distributions as a wind strength indicators after\n \n \n 68\n \n \n . Compared to other beach-proximal storm-influenced settings\n \n \n 69\n \n \n , the silt-dominated mean grain size (MGS) distribution in Steinbruvatnet lake is comparatively fine and also stable (Fig.\n \n 3\n \n c). Our End-Member Modelling Analysis (EMMA; see Stratigraphy paragraph in our\n \n methods\n \n section and Supplementary Fig. S4) output provides a possible explanation\n \n \n 70\n \n \n , by demonstrating that particle size distributions are diluted with the silts of End-Member (EM) 1 that often dominate reworked glacigenic soils found in unvegetated ice-proximal polar settings like our study area\n \n \n 71\n \n \n . Indeed, the granulometry of a catchment sample taken from the eolian silt sheets that are found towards the East coast (CS 1; see Fig.\n \n 1\n \n c and setting) confirm that particles transported by the polar Easterlies are dominated by this size fraction (Supplementary Fig. S4). In contrast, sand-dominated End-Member (EM) 3, which exerts a strong influence on mean grain size (MGS) distributions and co-varies with coarser grain size indicator Zr/K (Table\n \n 1\n \n )\n \n \n 60\n \n ,\n \n 72\n \n \n , has a near-identical particle size distribution as catchment sample CS 2 (Supplementary Fig. S4). This material was extracted from one of the active dunes to the West of Lake Steinbruvatnet (see Supplementary Fig.\n \n S1\n \n and setting) and thus likely transported by the Westerly storm tracks. Following from the above grain size evidence, we argue that the fine silt-dominated input of EM 1 reflects input mobilized by the polar Easterlies, while coarser sand-dominated EM 3 is transported to Lake Steinbruvatnet by the Westerly storm tracks. Finally, the presented CT imagery of Fig.\n \n 3\n \n and Supplementary Fig. S3 reveal the sporadic presence of rounded pebbles and cobbles that were likely ice rafted from the lake beach and deposited as drop stones. Between 10.4\u201312.8 cm these did not allow us to acquire sediment CT grayscale values (Fig.\n \n 3\n \n ).\n
\n| \n | \n\n \n CT\n \n | \n \n \n DBD\n \n | \n \n \n inc./coh.\n \n | \n \n \n LOI\n \n | \n \n \n Zr/K\n \n | \n \n \n MGS\n \n | \n \n \n EM 1\n \n | \n \n \n EM 3\n \n | \n \n \n Ti\n \n | \n \n \n Ca/Ti\n \n | \n
|---|---|---|---|---|---|---|---|---|---|---|
| \n \n \n DBD\n \n \n | \n \n \n 0.60\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n \n inc./coh.\n \n \n | \n \n \n -0.64\n \n | \n \n \n -0.80\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n \n LOI\n \n \n | \n \n \n -0.56\n \n | \n \n \n -0.93\n \n | \n \n \n 0.71\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n \n Zr/K\n \n \n | \n \n \n -0.41\n \n | \n \n \n -0.73\n \n | \n \n \n 0.62\n \n | \n \n \n 0.67\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n \n MGS\n \n \n | \n \n \n \n -0.13\n \n \n | \n \n \n \n -0.07\n \n \n | \n \n \n 0.28\n \n | \n \n \n \n -0.06\n \n \n | \n \n \n 0.40\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n \n EM 1\n \n \n | \n \n \n 0.25\n \n | \n \n \n \n 0.17\n \n \n | \n \n \n -0.32\n \n | \n \n \n \n -0.09\n \n \n | \n \n \n -0.46\n \n | \n \n \n -0.90\n \n | \n \n | \n\n | \n\n | \n\n | \n
| \n \n \n EM 3\n \n \n | \n \n \n \n -0.03\n \n \n | \n \n \n \n -0.03\n \n \n | \n \n \n 0.21\n \n | \n \n \n \n -0.14\n \n \n | \n \n \n 0.29\n \n | \n \n \n 0.86\n \n | \n \n \n -0.66\n \n | \n \n | \n\n | \n\n | \n
| \n \n \n Ti\n \n \n | \n \n \n 0.61\n \n | \n \n \n 0.81\n \n | \n \n \n -0.88\n \n | \n \n \n -0.71\n \n | \n \n \n -0.75\n \n | \n \n \n -0.34\n \n | \n \n \n 0.38\n \n | \n \n \n -0.27\n \n | \n \n | \n\n | \n
| \n \n \n Ca/Ti\n \n \n | \n \n \n -0.14\n \n | \n \n \n \n -0.03\n \n \n | \n \n \n 0.22\n \n | \n \n \n \n 0.05\n \n \n | \n \n \n 0.06\n \n | \n \n \n \n 0.06\n \n \n | \n \n \n \n 0.00\n \n \n | \n \n \n \n -0.05\n \n \n | \n \n \n -0.29\n \n | \n \n | \n
| \n \n \n Br/(inc./coh.)\n \n \n | \n \n \n 0.64\n \n | \n \n \n 0.80\n \n | \n \n \n -0.99\n \n | \n \n \n -0.71\n \n | \n \n \n -0.63\n \n | \n \n \n -0.26\n \n | \n \n \n 0.31\n \n | \n \n \n \n -0.19\n \n \n | \n \n \n 0.88\n \n | \n \n \n -0.20\n \n | \n
\n To distil information from multiple of the afore-mentioned proxies that capture changes in wind regime in investigated Lake Steinbruvatnet, we relied on Principal Component Analysis (PCA; see Statistics paragraph in methods). To do so on human-relevant (decades to centuries) timescales, we decided to only include \u00b5m-scale resolution scanning data as the 0.3 cm sampling diameter of measured physical parameters exceeds the width of many facies 2 layers (see Stratigraphy paragraph in methods). We argue that this can be legitimized by the strong correlation between physical and scanning measures of organic content (LOI vs. inc./coh.), density (DBD vs. CT), and grain size (MGS vs. Zr/K) \u2013 see Table\n \n 1\n \n . As outlined in our Stratigraphy paragraph in methods, except for Br and Calcium (Ca), we excluded XRF elements with a Signal-to-Noise ratio lower than 2\n \n 63,74\n \n . Based on the observed co-variance of our first principal component (PC 1) with minerogenic indicators Rubidium (Rb), Strontium (Sr) as well as Ti (Fig.\n \n 4\n \n ), and the association of the latter element with the fine-grained (EM 1-dominated) and therefore (CT) dense input (see Fig.\n \n 3\n \n and Table\n \n 1\n \n ), which is only found along the eastern shores of S\u00f8rkapp\u00f8ya (see setting and Supplementary Fig. S4), we associate PC 1 with eolian input from the polar Easterlies. The strong correlation between PC 1 scores and sea-spray indicator Br/(inc./coh.) counts (\u03c1\u2009=\u20090.93, n\u2009=\u20094,729,\n \n p\n \n =\u20090.00) provides additional information (also see Supplementary Fig. S5). As these aerosols derive from open ocean waters, and the adjacent northern Barents Sea has been seasonally ice-covered throughout the Holocene\n \n \n 50\n \n \n , the relation between PC 1 and Br/(inc./coh.) hints at a summer season signal. In contrast, both coarse grain size indicator Zr/K as well as CT grayscale values\n \n \n 60\n \n ,\n \n 72\n \n \n , which are also often impacted by changes in grain size\n \n \n 75\n \n \n , have stronger PC 2 loadings (see Fig.\n \n 4\n \n ). As mentioned, and shown (see Supplementary Fig. S4), minerogenic input of this size fraction (EM 3) is only available on the more proximal West coast (see setting and Fig.\n \n 1\n \n d). Therefore, we argue that PC 2 tracks changes in the Westerly storm tracks. The coarseness of minerogenic material sourced from the West may also help explain why there is no correlation (\u03c1\u2009=\u20090.07, n\u2009=\u20094,729,\n \n p\n \n =\u20090.00) between Br/(inc./coh.) and PC 2: the winds speeds required to mobilize these grains are too great to allow small sea-spray drops to fall out of suspension. Regardless, based on observational evidence that westerly winds only prevail between June and August on S\u00f8rkapp\n \n \n 41\n \n \n , we contend that PC 2 also captures a summer-dominated signal. Finally, we note that the aforesaid 10.4\u201312.8 cm clast-related CT data gap is also reflected in our PCA data.\n
\n\n Following from the above, we argue that the presented multi-proxy evidence from Lake Steinbruvatnet captures changes in Easterly (PC 1) and Westerly (PC 2) wind strength between ca. 9,500\u2013800 cal. yrs B.P. Due to the centennial-scale uncertainties of our age model and a lack of undisturbed surface sediments to build modern analogs using historical storms (Fig.\n \n 2\n \n and core chronology)\n \n \n 26\n \n \n , it is not possible to ascertain whether our PC maxima reflect windy events or phases of stronger winds. However, we favor the latter scenario after\n \n \n 27\n \n \n , as our proxies do not behave in a binary fashion and inferred eolian input dominates sedimentation throughout the record (Fig.\n \n 3\n \n ). Also, while our PCs are associated with each of the wind systems that impact our study area today\n \n \n 41\n \n \n , we cannot assess absolute changes in their respective strength, due the afore-mentioned lack of observation-based validation, as well as differences in eolian particle size and transport distance (Fig.\n \n 1\n \n , Supplementary Figs. S1 and S4). Therefore, we will discuss our reconstructions as relative variations in Easterly and Westerly wind strength through time. To further validate our interpretations and assess their broader representativeness, we compare our records to other wind reconstructions. Such efforts are, however, often hampered by\n \n 1\n \n ) data scarcity \u2013 as Holocene-length extra-tropical storm reconstructions remain scarce,\n \n 2\n \n ) baseline shifts \u2013 fluxes of shore-derived eolian input are affected by sea-level changes, and\n \n 3\n \n ) age uncertainty \u2013 windy phases may mis-align between sites because of chronological errors\n \n \n 26\n \n ,\n \n 76\n \n \n . All these factors affect the Arctic region disproportionally due to its remoteness, complex isostatic uplift history, and a general scarcity of radiocarbon (\n \n \n 14\n \n \n C) dateable material. To help overcome these challenges, we primarily focus our comparison on two regionally relevant records that resolve change on multi-centennial timescales like our PC-based wind indicators (Fig.\n \n 5\n \n ), and cover most of the Holocene like our data \u2013 published reconstructions rarely extend beyond the Mid-Holocene\n \n \n 76\n \n \n .\n \n Firstly\n \n , the only existing continuous Holocene reconstruction of the Easterly winds in the North Atlantic by\n \n \n 46\n \n \n , which covers the last 8,700 cal. yrs B.P. and remains unaffected by post-glacial uplift as measured eolian input derives from local soils. And\n \n secondly\n \n , the stacked chronology of past Westerly wind activity in coastal northwest Europe by\n \n \n 47\n \n \n , which identifies Holocene Storm Periods (HSPs) in nine coastal records since 6,500 cal. yrs B.P., when regional sea-level change stabilized following melt of the Laurentide Ice Sheet\n \n \n 77\n \n \n .\n
\n\n As outlined in the introduction, this study fundamentally seeks to deepen our understanding of the links between Arctic climate and storminess. To further highlight wind strength maxima, we developed a so-called Storm Magnitude Index (SMI) by calculating the area under our PC curves after\n \n \n 63\n \n \n . For this purpose, we\n \n 1\n \n ) detrended PC values to negate the impact of sea-level change on coastal distance and therefore eolian material fluxes to warrant assessment against a stable baseline (see Statistics paragraph in methods) after\n \n \n 26\n \n \n , before\n \n 2\n \n ) identifying extremes as peaks that exceed the mean (\u00b5)\u2009+\u2009one standard deviation (\u03c3) bound of our PC values, and finally\n \n 3\n \n ) calculating the definite integral for each of these stormy intervals using the trapezoidal rule.\n
\n\n Considering the afore-mentioned challenges that complicate comparison between sites, our PC-derived wind reconstructions from Lake Steinbruvatnet bear a striking resemblance to the selected regionally relevant reconstructions. As can be seen in Fig.\n \n 5\n \n a-b, PC 1 reproduces each Easterly wind event captured by\n \n \n 46\n \n \n , despite differences in signal amplitude that we tentatively attribute to\n \n 1\n \n ) location \u2013 both sites sit\u2009~\u20092,000 km apart and therefore differently with regard to the average storm track position (Fig.\n \n 1\n \n a), and\n \n 2\n \n ) proxy \u2013 the silt-sized particles associated with the Easterly winds in Steinbruvatnet lake require less wind energy for transport than the sand-sized soil grains reported by\n \n \n 46\n \n \n . The resemblance between both reconstructions is also reflected by a moderately positive \u03c1 of 0.35 (\n \n p\n \n =\u20090.000), which can be considered a strong correlation for noisy paleostorm records\n \n \n 29\n \n \n . Similarly, as seen in Fig.\n \n 5\n \n c-d, our PC 2 SMI maxima broadly coincide with the Westerly wind HSPs reported by\n \n \n 47\n \n \n , although overlap with HSP III around 3,300-2,400 cal. yrs B.P. is marginal. However, when looking at our detrended PC 2 scores, we see that stronger Westerly winds prevail throughout this period. This difference in signal amplitude can be explained by the sensitivity of the investigated systems \u2013 the foreshore archives analyzed by\n \n \n 47\n \n \n are more exposed to wind impacts than sheltered backshore sites like Lake Steinbruvatnet\n \n \n 26\n \n \n . In summary, despite site-specific differences, our Easterly and Westerly wind maxima are regionally consistent, which strengthens our confidence in their interpretation. Also, as pointed out by\n \n \n 47\n \n \n , Holocene extremes of both systems coincide (Fig.\n \n 5\n \n ).\n
\n\n Our PC-based reconstructions reveal five regionally consistent multi-centennial Holocene maxima in between 1,000\u20132,000 cal. yrs B.P., as well as around 3,500, 4,300, 5,500, 7,400 and 8,200 cal. yrs B.P. (Fig.\n \n 5\n \n ). As shown by\n \n 46\n \n and\n \n 47\n \n , these stormy phases coincide with North Atlantic cooling periods. By extending this association between cold and windy conditions into a study area that is seasonally sea ice-covered, our findings challenge the view that a warmer and less icy Arctic will become stormier \u2013 the premise of this study (see introduction). This notion is supported by local evidence, which reveals that\n \n 1\n \n ) Easterly winds were most intense during the Late Holocene, when sea surface temperatures were relatively low and severe sea ice conditions persisted in up-wind Barents Sea\n \n \n 50\n \n ,\n \n 51\n \n ,\n \n 78\n \n \n , while\n \n 2\n \n ) Westerly wind strength does not exhibit a clear relation with either temperature or sea ice conditions as SMI maxima occur throughout the Holocene (Figs.\n \n 1\n \n and\n \n 5\n \n )\n \n \n 48\n \n \n . Based on the outlined evidence, we question that a warmer less icy future Arctic will be wavier and windier, as is often suggested\n \n \n 17\n \n \n . If true for areas other than our study site, this has implications for the perceived sensitivity of regional shorelines to coastal erosion and ensuing societal impacts like infrastructure damage and carbon release\n \n \n 13\n \n ,\n \n 19\n \n ,\n \n 79\n \n \n . At the same time, we would like to stress the compounding impacts of two processes that will affect Arctic coastal dynamics independent from changes in wind and wave energy: permafrost degradation and sea-level rise, which are mostly thermally driven\n \n \n 17\n \n \n .\n
\n\n As outlined above, SMI maxima coincide with Easterly and Westerly wind intensity extremes that occurred during regional cooling intervals as shown by\n \n 46\n \n and\n \n 47\n \n . Both of these studies associate these recurring cold and windy phases with quasi-periodic\u2009~\u20091,500-year North Atlantic ice rafting events\n \n \n 80\n \n \n . Wavelet transformation (see Statistics paragraph in our methods) reveals that this cycle also dominates the spectral signature of both our PC-based wind reconstructions (Supplementary Fig. S6). While pervasive and debated for decades\n \n \n 80\n \n ,\n \n 81\n \n \n , differences in phasing and frequency have hampered confident attribution of these oscillations\n \n \n 82\n \n \n . The Steinbruvatnet record adds to our understanding of this enigmatic periodicity in two notable ways.\n \n Firstly\n \n , by investigating the spectral signature of the Easterly winds for the first time, we find that our PC 1-based reconstruction and the Icelandic record by\n \n \n 46\n \n \n share a common and consistent phase relation (r\u2009=\u20090.95,\n \n p\n \n =\u20090.000) that differs from the afore-mentioned\u2009~\u20091,500-year North Atlantic ice rafting events\n \n \n 53\n \n \n . As demonstrated by the extracted\u2009~\u20091,500-year frequency components shown in Supplementary Fig. S7, a max. ~500-year offset suggests a different forcing mechanism. A similar observation was made with regard to the Holocene behavior of the Arctic Oscillation \u2013 the pressure anomaly between the North Pole and 20\u00b0N that increases polar Easterly wind strength in Iceland and Svalbard when in a positive phase\n \n \n 83\n \n \u2013\n \n 85\n \n \n . When comparing the ~\u20091,500-year component of this timeseries to those captured by our PC 1-based Easterly wind reconstruction and that of\n \n \n 83\n \n \n (Supplementary Fig. S7), we observe a strong (r\u2009=\u20090.65,\n \n p\n \n =\u20090.000) correlation. Based on our evidence, we argue that Holocene Easterly wind strength was controlled by the Arctic Oscillation, and not by the same internal mechanism as proposed for the Westerlies by\n \n \n 47\n \n \n , despite a common periodicity.\n \n Secondly\n \n , our PC 2-based reconstruction extends the temporal evolution of the spectral signal of the Westerly winds beyond the 6,000-year perspective provided by\n \n \n 47\n \n \n . And while our data support the conclusion of this study that Westerly wind and ice rafting are phase-locked throughout this period (r\u2009=\u20090.76,\n \n p\n \n =\u20090.000), they also show that this relation breaks down further back in time (Supplementary Fig. S7d-e), as shorter cycles become more dominant (Supplementary Fig. S6). This so-called Mid-Holocene transition has been detected in numerous proxy reconstructions from around the world\n \n \n 86\n \n \n , and is often associated with the concurrent stabilization of large-scale climate boundary conditions like sea-level and ice sheet extent\n \n \n 82\n \n \n . But unlike these studies, we find no conclusive evidence that the ~\u20091,000-year cycle characteristic of solar forcing dominates during the Early Holocene (Supplementary Fig. S6). However, regarding higher-frequency variability, our Westerly wind reconstruction does indicate an increase in the amplitude of decadal-scale variability during this period (Fig.\n \n 5\n \n ). This observation contrasts with the results of\n \n \n 87\n \n \n , who infer dampened Westerlies variability during the Early Holocene based on changes in varve thickness, although we should note that this reconstruction records winter conditions, whereas our data is biased towards summer (see the Holocene evolution of Steinbruvatnet). If true, our findings might also be relevant for the predictability of regional wind change, as the future will be shaped by melting ice sheets and warming like the Early Holocene.\n
\n\n We analyse 106 cm long sediment core 601-21-6 GC, which was collected from Lake Steinbruvatnet in August 2021 at a depth of 1.86 m using a Uwitec gravity corer. We targeted a flat section in the central part of the lake to avoid disturbances (Fig.\n \n 1\n \n c). Following fieldwork, the core was split lengthwise, then visually logged and photographed with a RGB line camera, prior to non-destructive core scanning and subsequent physical sampling.\n
\n\n Following logging, we first conducted several non-destructive scanning analyses. X-Ray Fluorescence (XRF) scanning was performed on an ITRAX scanner at the EARTHLAB facility of the University of Bergen (UiB) to map fluxes of eolian minerogenic elements and sea-spray aerosols\n \n \n 27\n \n \n . To measure minerogenic input with higher atomic numbers with greater precision, the scanner was fitted with a Molybdenum (Mo) tube set to 40 kV and 10 mA. Down-core measurements were generated for 34 elements at 200 \u00b5m intervals. We excluded elements with a Signal-to-Noise ratio (SNR; \u00b5/\u03c3) lower than 2 after\n \n \n 63\n \n ,\n \n 74\n \n \n , and those with a low sensitivity to the fitted Mo tube, except for marine indicators Calcium (Ca) and Bromine (Br)\n \n \n 58\n \n ,\n \n 64\n \n ,\n \n 65\n \n \n . All XRF data are presented as Total Scatter Normalized (TSN) ratios after\n \n \n 27\n \n ,\n \n 63\n \n \n , to account for variations in organic and water content. Computed Tomography (CT) scanning was applied to visualise sediment structures like storm layers in 3-D\n \n \n 63\n \n ,\n \n 88\n \n \n , and to determine ensuing variations in density captured by CT grayscale values\n \n \n 62\n \n ,\n \n 63\n \n \n . CT scanning was performed on a ProCon X-ray CT-ALPHA scanner, operated at 110 kV and 810 \u00b5A, with a 267 ms exposure time to generate ca. 100 \u00b5m resolution 24-bit scans. Scans were then processed with version 9 of the Thermo Fisher Avizo software to generate 2-D orthoslices and 3-D reconstructions.. Subsequently, we used CT orthoslices to verify initial visual logging and create a schematic lithostratigraphy after\n \n \n 33\n \n \n . To this end, we relied on the image trace operator in Adobe Illustrator CC 2015\n \n 89\n \n . Next, we performed destructive physical analyses to measure down-core variations in organic content, density, and grain size distribution. Based on CT imagery and visual assessment, we extracted 97 samples with a 0.3 cm wide 1 ml syringe from minerogenic (facies 2) layers (n\u2009=\u200979) as well as organic background (facies 1) sediment (n\u2009=\u200918) at irregular 0.2\u20133.5 cm intervals. All samples were dried 12 h at 105\u00b0C and then combusted for 4 h at 550\u00b0C to determine Dry Bulk Density (DBD; g/cm\n \n 3\n \n ) and Loss on Ignition (LOI; %; a measure of organic content)\n \n \n 90\n \n ,\n \n 91\n \n \n . Grain size, a commonly used indicator of wind strength\n \n \n 68\n \n \n , was measured on all 97 sediment samples from the core, as well as 2 catchment samples from active eolian deposits near the eastern (CS 1) and western (CS 2) beaches (see Fig.\n \n 1\n \n c and the Holocene evolution of Steinbruvatnet), using a Malvern Mastersizer 3000 with a Hydro SV dispersion unit. Each sample was measured 5 times to warrant reproducibility. Following the recommendations of\n \n \n 73\n \n \n , sample particle size distributions were processed using the GRADISTAT software and expressed as metric (\u00b5m) Folk and Ward measures. Finally, End-Member Modelling Analysis (EMMA) was applied to the core samples to unmix particle size distributions and their sediment sources\n \n \n 70\n \n \n . The analysis was run with the AnalySize 9.3 tool in MATLAB\n \n \n 92\n \n \n . We used the non-parametric HALS-NMF algorithm which is well-suited for improving the unmixing accuracy\n \n \n 93\n \n \n and thus identifying End-Members (EMs) and their abundances\n \n \n 92\n \n \n .\n
\n\n We relied on radiocarbon (\n \n \n 14\n \n \n C) dating to establish age control. To allay concerns about freshwater reservoir effects, we only picked terrestrial plant fragments (leaves and stems). The material was extracted by wet sieving through 250 and 125 \u00b5m meshes, before overnight drying at 50\u00b0C. In total, 9 samples were taken from 601-21-6 GC at semi-regular intervals and submitted for Accelerator Mass Spectrometer (AMS) dating in the Pozna\u0144 Radiocarbon Laboratory, Poland (Poz)\n \n \n 94\n \n \n , and the Tandem Laboratory at Uppsala University, Sweden (Ua; Table\n \n 2\n \n ). The latter was chosen because several samples were particularly small (2.7\u20137.1 mg).\n
\n| \n \n Lab code\n \n | \n \n \n Depth (cm)\n \n | \n \n \n Material\n \n | \n \n \n mg C\n \n | \n \n \n \n \n 14\n \n \n C age (yrs B.P.)\n \n | \n \n \n Error (yrs)\n \n | \n \n \n Cal. yrs B.P.\n \n | \n
|---|---|---|---|---|---|---|
| \n \n Ua-76380\n \n | \n \n \n 7.5\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 2.7\n \n | \n \n \n 9,312*\n \n | \n \n \n 45\n \n | \n \n \n 10,601\u2009\u2212\u200910,369*\n \n | \n
| \n \n Ua-76381\n \n | \n \n \n 15.5\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 7.1\n \n | \n \n \n 9,461*\n \n | \n \n \n 42\n \n | \n \n \n 10,790\u2009\u2212\u200910,574*\n \n | \n
| \n \n Poz-150330\n \n | \n \n \n 18.75\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 0.9\n \n | \n \n \n 6,290*\n \n | \n \n \n 35\n \n | \n \n \n 7,289-7,158*\n \n | \n
| \n \n Poz-149887\n \n | \n \n \n 29.25\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 11.8\n \n | \n \n \n 3,475\n \n | \n \n \n 35\n \n | \n \n \n 3,839-3,682\n \n | \n
| \n \n Poz-150331\n \n | \n \n \n 53.25\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 9.7\n \n | \n \n \n 6,180\n \n | \n \n \n 40\n \n | \n \n \n 7,166-6,953\n \n | \n
| \n \n Poz-149888\n \n | \n \n \n 65.75\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 13.7\n \n | \n \n \n 7,130\n \n | \n \n \n 50\n \n | \n \n \n 8,024\u2009\u2212\u20097,915\n \n | \n
| \n \n Ua-76382\n \n | \n \n \n 79.25\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 6.0\n \n | \n \n \n 7,714\n \n | \n \n \n 38\n \n | \n \n \n 8,552-8,415\n \n | \n
| \n \n Ua-76383\n \n | \n \n \n 85.25\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 4.3\n \n | \n \n \n 9,376*\n \n | \n \n \n 46\n \n | \n \n \n 10,718\u2009\u2212\u200910,495*\n \n | \n
| \n \n Poz-145555\n \n | \n \n \n 105.75\n \n | \n \n \n \n Terrestrial plants\n \n \n | \n \n \n 36.2\n \n | \n \n \n 8,650\n \n | \n \n \n 50\n \n | \n \n \n 9,742-9,532\n \n | \n
\n XRF and CT output was resampled on a common 0.5 cm with the lower-resolution physical analyses to allow multivariate statistical analysis. For this purpose, we employed a 0.3 cm (15 point) Gaussian smoothing operator to account for the width of the syringe used to extract samples (see Stratigraphy paragraph in methods), before resampling at 0.5 cm intervals using linear interpolation in version 4 of the PAST software\n \n \n 95\n \n \n . We used the same program for the calculation of Spearman\u2019s rank correlation coefficient (\u03c1). To explore shared gradients of change captured by our independently measured eolian indicators, we carried out a Principal Component Analysis (PCA) on selected proxy parameters, using version 5 of the CANOCO software\n \n \n 96\n \n \n . The input was centred and standardized before analysis, following software recommendations. Following\n \n \n 26\n \n \n , we detrended PCA output to account for the fact that (Early) Holocene sea-level changes influenced the distance between the lake and the coast\n \n \n 36\n \n \n , and thus fluxes of eolian material (see Core chronology). To this end, we relied on the remove trend transformation in version 4 of PAST\n \n \n 95\n \n \n . We also used this software to\n \n 1)\n \n smooth PC 1\u20132 and Br/(inc./coh.) values using a 150-year moving average,\n \n 2\n \n ) cross-correlate selected timeseries, and\n \n 3\n \n ) perform continuous wavelet transform (CWT) analysis to detect spectral signatures\n \n \n 95\n \n \n . To do so, we used a Morlet mother wavelet, following the recommendations of\n \n \n 82\n \n ,\n \n 83\n \n ,\n \n 95\n \n \n . Finally, we applied WlCount \u2013 a semi-automatic lamination detection and counting software by\n \n \n 59\n \n \n . By extracting visual information from the entire width of the CT image of investigated core 601-21-6 GC (see Fig.\n \n 3\n \n ), this tool complements other down-core scanning data, which were acquired along specific down-core lines.\n
\n\n Supplementary Dataset 1\n
\n \n\n Supplementary Information\n
\n \n\n The Tethyan orogenic belt underwent prolonged tectonic evolution and hosts numerous world-class porphyry copper deposits. Notably, most porphyry deposits are associated with Cenozoic continental collision, while fewer are formed during Mesozoic subduction. Here we integrate detrital zircon oxybarometry with geochemical data, stratigraphy, sea-level and temperature fluctuations, and major geological events. Our results reveal a stark redox transition\u2013from anoxic during Mesozoic subduction to oxidized during Cenozoic collision. We propose that subduction of organic-rich, reduced sediments in the Mesozoic suppressed the oxidation state of arc magmas, locking chalcophile elements in the lower crust and inhibiting the formation of subduction-related porphyry Cu deposits. In contrast, the subduction of more oxidized sediments during the Cenozoic elevated oxygen fugacity, releasing stored metals and promoting extensive formation of porphyry Cu deposits during continental collision. These findings underscore the critical role of sediment redox state and subduction history in governing porphyry mineralization along the Tethyan belt.\n
\n\n The Tethyan orogenic belt, extending for 12,000 km from the Pyrenees through the Alps and the Turkish\u2013Iranian plateau, across Pakistan and the Himalayan\u2013Tibetan Plateau, and into the Indochina Peninsula (Fig.\n \n 1\n \n a), is one of the world\u2019s most important porphyry Cu belts\n \n \n 1\n \n \u2013\n \n 4\n \n \n . This orogen formed through successive stages of continental breakup, ocean basin formation, and eventual collision between Gondwana-derived continents and Eurasia (Supplementary Fig.\u00a01 and Supplementary Material). The Tethys realm can be divided chronologically into the Proto-, Paleo-, and Neo-Tethys oceans, formed during the Early Paleozoic, Late Paleozoic, and Mesozoic, respectively\n \n \n 5\n \n \n . Of particular interest is the Neo-Tethys Ocean, which opened in the Early Permian and underwent protracted subduction during the Mesozoic, culminating in extensive Cenozoic continental collisions that led to the formation of porphyry Cu deposits in regions such as the Gangdese and Yulong belts (Tibet), the Kerman belt (Iran), and the Anatolides belt (Turkey)\n \n \n 1\n \n ,\n \n 2\n \n ,\n \n 5\n \n \u2013\n \n 8\n \n \n (Fig.\n \n 1\n \n a and Supplementary Material).\n
\n\n
\n\n Despite a long Mesozoic subduction history, significantly fewer and smaller porphyry Cu deposits formed during subduction-related magmatism than during Cenozoic collisional or post-collisional magmatic stages in the Tethyan belt (Fig.\n \n 1\n \n b and\n \n 2\n \n c). The reason for this disparity remains elusive. A key control on porphyry copper formation is the oxidation state of the magmatic system, i.e., magmatic oxygen fugacity, commonly expressed as \u2206FMQ (the log of oxygen fugacity relative to the fayalite\u2013magnetite\u2013quartz buffer). Oxidized arc magmas (\u2206FMQ\u2009\u2248\u2009+\u20091 to +\u20092) favor the mobilization of chalcophile elements, enabling their transport to the upper crust\n \n \n 9\n \n ,\n \n 10\n \n \n . Detrital zircon provides a valuable record of magmatic redox conditions over geologic timescales. Global geological events\u2013particularly oceanic anoxic events and seawater incursion events\u2013can influence the geochemical composition of subducted sediments\n \n \n 11\n \n ,\n \n 12\n \n \n , altering mantle wedge oxygen fugacity and affecting porphyry mineralization potential.\n
\n\n
\n\n Here, we apply a multidisciplinary approach to track the oxygen fugacity evolution of Tethyan magmas from the Early Jurassic to the Miocene. Using the novel zircon oxybarometer, we reconstruct \u2206FMQ through time and integrate these data with sedimentary rocks, global sea-level and temperature changes, whole-rock geochemical data, and the timing of major geological events. Our findings illuminate the mechanisms responsible for redox variations in the mantle wedge and their implications for subduction- vs. collision-related porphyry Cu deposits.\n
\n\n Detrital zircon oxybarometry reveals an overall increase in \u2206FMQ from the Early Jurassic through the Miocene in the Tethyan belt (Fig.\n \n 2\n \n d). Although data density during the Jurassic to Early Cretaceous is relatively low and thus more variable, Mesozoic \u2206FMQ values are dominantly below +\u20091. By contrast, Cenozoic \u2206FMQ values commonly exceed\u2009+\u20091, consistent with a transition to more oxidized magmatic conditions (Fig.\n \n 2\n \n d).\n
\n\n The scarcity of Mesozoic subduction-related porphyry deposits in the Tethyan belt contrasts with the abundance of Cenozoic collision-related porphyry deposits (Fig.\n \n 2\n \n c). The porphyry Cu deposits are formed in the upper crust, usually at a depth of 1 to 5 km\n \n \n 13\n \n \n , and globally occur mainly in the Phanerozoic, particularly in the Cenozoic\n \n \n 14\n \n \n . While some workers have suggested that extensive erosion may have removed older subduction-related porphyry Cu deposits\n \n \n 15\n \n ,\n \n 16\n \n \n , there is no credible evidence for widespread pre-Cenozoic porphyry Cu mineralization before the continental collision in the Tethyan belt.\n
\n\n The fertile intrusions for porphyry Cu formation worldwide are oxidized with \u2206FMQ\u2009+\u20091 to +\u20092\n \n 9,17,18\n \n . By contrast, the notably low \u2206FMQ (\u2009<\u2009+\u20091) during the Jurassic and Cretaceous suggests that arc magmas were relatively reduced, inhibiting the transport of chalcophile elements into upper crustal levels and thus limiting porphyry Cu formation (Fig.\n \n 2\n \n d). Under low \u0192O\n \n 2\n \n , sulfur tends to form sulfides that sequester metals (e.g. Cu, Au) in the lower crust via sulfide accumulation, resulting in rare mineralization in the upper crust\n \n \n 9\n \n ,\n \n 10\n \n ,\n \n 19\n \n \n . This process parallels the anoxic conditions inferred for the Paleo-Tethys Ocean basin during the Permian when the basin was in a restricted environment near the equator\n \n \n 20\n \n ,\n \n 21\n \n \n . The correlation between redox state and porphyry Cu deposit frequency (Fig.\n \n 2\n \n c\u2013d) supports redox-driven control over the spatiotemporal distribution of Tethyan porphyry Cu mineralization.\n
\n\n Oceanic subduction-related arc magmas are commonly oxidized due to slab-derived fluids\n \n \n 22\n \n \n , but the Neo-Tethys domain appears to have bucked this trend during the Mesozoic, exhibiting lower \u2206FMQ (Fig.\n \n 2\n \n d). To elucidate this discrepancy, we compiled data on sedimentary rocks, global sea-level and temperature changes, whole-rock geochemical data, and major global geological events (Fig.\n \n 2\n \n \u2013\n \n 4\n \n ).\n
\n\n
\n\n Multiple Jurassic and Cretaceous oceanic anoxic events are recognized in the Tethyan belt, including the early Toarcian (~\u2009183 Ma), Callovian (~\u2009166 Ma), early Aptian (~\u2009120 Ma), early Albian (~\u2009111 Ma), late Albin (~\u2009102 Ma and ~\u2009100Ma), Cenomanian\u2013Turonian (~\u200993 Ma), and late Coniacian to Santonian (~\u200986 Ma)\n \n \n 23\n \n \u2013\n \n 28\n \n \n (Fig.\n \n 2\n \n c). These events coincided with massive deposition of organic-rich black shales in marine and terrestrial settings\n \n \n 23\n \n \n . The compiled stratigraphic columns in the Tethyan domain show abundant organic matter-rich sediments in the Mesozoic (Fig.\n \n 3\n \n ). Subduction of such reduced sediments would release CH\n \n 4\n \n -rich fluids into the mantle wedge, decreasing the oxidation state of arc magmas\n \n \n 29\n \n ,\n \n 30\n \n \n . The temporal correlation of low \u2206FMQ values with oceanic anoxic events suggests that organic-rich subducted slabs played a pivotal role in producing relatively reduced magmas during the Mesozoic (Fig.\n \n 2\n \n c\u2013d).\n
\n\n
\n\n This is further supported by whole-rock elemental and isotopic data for mafic rocks in the Tethyan domain (Supplementary Table\u00a03). The mantle can be modified by the subducting slab and overlying sediments. Different sediment melt (pelagic or terrigenous) and fluid display distinct elemental and isotopic signatures. As a result, some specific geochemical signatures can be used to decipher the contribution of fluid and sediment melt to a depleted mantle wedge\n \n \n 31\n \n ,\n \n 32\n \n \n . Slab fluid has higher concentrations of fluid-mobile elements such as large ion lithophile elements (e.g., Ba, Rb, Sr, K) but lower concentrations of light rare earth elements, thorium, and high field strength elements (e.g., Nb, Ta, Zr, and Hf) compared to sediment melt\n \n \n 31\n \n ,\n \n 32\n \n \n . Furthermore, owing to the \u2018zircon effect\u2019, melted terrigenous sediments will have lower Lu/Hf ratios due to the abundant detrital zircons that enrich Hf compared to melted pelagic sediments that lack detrital zircons\n \n \n 33\n \n \n . The high Ba/La, Lu/Hf ratios (Fig.\n \n 4\n \n a\u2013b), Sr isotopic ratios and low Nd isotopic ratios (Fig.\n \n 4\n \n c) of Tethyan mafic rocks produced during Mesozoic oceanic subduction suggest more pelagic sediment melts (less than 4%) and/or slab fluid component was added into the mantle magmatic source reservoir. The enriched isotopic features are mainly attributed to mantle source metasomatism rather than crustal contamination, as there are no negative or positive covariation patterns between whole-rock\n \n 87\n \n Sr/\n \n 86\n \n Sr\n \n (i)\n \n ,\n \n 143\n \n Nd/\n \n 144\n \n Nd ratios and SiO\n \n 2\n \n contents (Supplementary Fig.\u00a02).\n
\n\n Frequent seawater incursion events during the Mesozoic, coupled with elevated global sea levels and warmer temperatures (Fig.\n \n 2\n \n a), facilitated extensive organic carbon burial and the formation of hydrocarbon source rocks\n \n \n 11\n \n ,\n \n 12\n \n ,\n \n 34\n \n \u2013\n \n 36\n \n \n . Rising sea levels played a critical role in controlling organic facies deposition\n \n \n 37\n \n \n . During the Jurassic and Cretaceous, higher global temperatures\n \n \n 38\n \n \n and sea-level rises\n \n \n 39\n \n \n (Fig.\n \n 2\n \n a) led to increased nutrient influx from continental erosion into the oceans. This influx enhanced marine primary productivity, creating optimal conditions for significant organic matter burial in continental margin basins. Consequently, these conditions drove the formation of extensive hydrocarbon source rocks, establishing the Jurassic and Cretaceous as the world\u2019s most prolific periods for oil and gas generation (Fig.\n \n 2\n \n b).\n
\n\n The degree of restriction in sedimentary basins\u2013classified as open or restricted\u2013serves as a proxy for oceanic anoxia\n \n \n 40\n \n \n . Restricted ocean basins, often enclosed by surrounding land, are more likely to develop anoxic conditions. In contrast, open marine basins are less prone to anoxia\n \n \n 40\n \n \n . Although this classification does not take into account detailed geochemical constraints or ocean dynamics (e.g., upwelling, surface currents, or salinity), certain trends can be inferred from these maps. During the Jurassic, the Neo-Tethys Ocean was predominantly in an open state, resulting in limited anoxia (Fig.\n \n 2\n \n e). However, as the oceanic subduction progressed, the Neo-Tethys basins gradually closed from the Jurassic to the Cretaceous, becoming increasingly restricted and anoxic (Fig.\n \n 2\n \n e). These periods of basin closure, accompanied late Mesozoic subduction, preceded Cenozoic continent-continent collision, align with widespread oceanic anoxic events and lower zircon \u2206FMQ values, and thus are consistent with an anoxic environment (Fig.\n \n 2\n \n c\u2013e).\n
\n\n In summary, the Neo-Tethys Ocean evolved into a warm, high sea-level, progressively closing restricted ocean basin during the Jurassic and Cretaceous. This environment fostered the deposition of abundant reduced organic matter, the subduction of which would undoubtedly decrease the oxygen fugacity in the mantle wedge (Fig.\n \n 2\n \n ).\n
\n\n Cenozoic magmas in the Tethyan belt are predominantly oxidized with \u2206FMQ\u2009>\u2009+\u20091 (Fig.\n \n 2\n \n d). Several mechanisms have been proposed to explain this elevated Cenozoic oxidation state:\n
\n\n (1) Subduction of oxidized continental sediments. After the Indian, Arabian, and African continents collided with Eurasia, the subduction of oxidized continental sediments (e.g., carbonates and evaporites, both are oxidants) likely elevated the oxidation state of the mantle wedge (Fig.\n \n 3\n \n ). This process occurs through the dehydration of these sediments, releasing oxidized fluids that contribute to mantle oxidation\n \n \n 2\n \n ,\n \n 41\n \n \u2013\n \n 45\n \n \n .\n
\n\n (2) Injection of oxidized ultrapotassic rocks into the lower crust. Ultrapotassic rocks play an increasingly recognized role in porphyry Cu mineralization in collision settings\n \n \n 46\n \n ,\n \n 47\n \n \n . As mantle-derived ultrapotassic magmas ascend from a paleo depth of ~\u200956 km to ~\u200915 km, their \u0394FMQ increases from +\u20090.8 to +\u20093.0\n \n 48\n \n . However, their capacity to significantly oxidize magma in the lower crust remains limited.\n
\n\n (3) Auto-oxidation during amphibole and/or garnet fractionation. Garnet fractionation, favored by crust thickening induced by continental collision, preferentially removes Fe\n \n 2+\n \n from magma, thus auto-oxidizing the residual melt\n \n \n 49\n \n \u2013\n \n 51\n \n \n . Amphibole, another common Fe-bearing mineral in hydrous melt, also contributes to magma oxidation\n \n \n 52\n \n \n . Its Fe\n \n 3+\n \n /\u01a9Fe ratio decreases during magma evolution, enriching Fe\n \n 3+\n \n in the residual melt\n \n \n 53\n \n \n . However, these auto-oxidation processes induced by garnet and amphibole fractionation under high pressure and hydrous conditions primarily occur during crustal differentiation rather than in the mantle source.\n
\n\n Overall, the tectonic switch from subduction of reduced pelagic to oxidized terrigenous sediments emerges as a dominant factor in increasing the oxidation state of magmas. Whole-rock elemental and isotopic data further support this inference. The high Th/Yb, Th/La, and Sr isotopic ratios, and low Lu/Hf and Nd isotopic ratios of mafic rocks emplaced during continental collision indicate increased involvement of terrigenous sediments in the mantle source (Fig.\n \n 4\n \n ). Monte Carlo isotopic mixing model between the depleted mantle and terrigenous sediments shows about 0.5\u20134% of terrigenous sediment melts were added into the mantle source (Fig.\n \n 4\n \n d). Moreover, the light Mg-Ca isotopic compositions of Tethyan post-collisional potassic-ultrapotassic rocks are interpreted as evidence for CO\n \n 2\n \n -related metasomatism in the mantle source replacing the earlier CH\n \n 4\n \n -related metasomatism\n \n \n 54\n \n \u2013\n \n 56\n \n \n .\n
\n\n The temporal distribution of Mesozoic subduction-related versus Cenozoic collision-related porphyry Cu deposits in the Tethys domain is highly uneven, with the majority of large deposits forming during the Cenozoic. Previous studies proposed a genetic link between these two types of porphyry Cu deposits. Magmas from earlier oceanic subduction emplaced metal-rich sulfide cumulates in the lower crust and these were suggested to have later remelted during continental collision to provide abundant metals for collision-related porphyry Cu deposits\n \n \n 57\n \n ,\n \n 58\n \n \n .\n
\n\n A critical aspect of this model requires the transition from reduced to oxidized conditions, a feature which is documented with the new evidence presented in this study. Thus, a reduced mantle wedge and associated arc magmas are present during oceanic subduction, which is followed by an increased oxidation state during continental collision. The initial reduced mantle wedge is largely attributed to the subduction of organic-rich sediments. Similar conditions have been observed in Japan\n \n \n 30\n \n ,\n \n 59\n \n \n and the Paleo-Tethys Ocean basins\n \n \n 20\n \n ,\n \n 21\n \n \n . Low oxygen fugacity traps chalcophile elements in the lower crust during magma fractionation, resulting in the emplacement of barren arc magmas in the upper crust (Fig.\n \n 5\n \n a).\n
\n\n
\n\n For these metal-rich sulfide cumulates to be remobilized, the oxidation state must increase. Subduction of oxidized continental sediments during collision releases oxidized fluids, converting residual sulfides in the lower crust to sulfates, thereby liberating metals for porphyry mineralization (Fig.\n \n 5\n \n b). This process underscores the interplay between subduction and collision in the formation of porphyry Cu deposits.\n
\n\n Our findings highlight the crucial role of the nature of subducted sediments in modulating the redox state and controlling the formation of porphyry Cu deposits. Given the importance of oxidation and hydration state for magmas to form porphyry Cu deposits, the key for Cu exploration is to identify areas or periods that may host oxidized and hydrous shallowly emplaced igneous complexes. Zircon is a common accessory mineral in felsic igneous provinces. Owing to its mechanical and chemical robustness, zircon can not only provide reliable geochronologic data but also contains metallogenic information, such as oxidation and hydration states\n \n \n 17\n \n ,\n \n 18\n \n \n , which is becoming a cost-competitive exploration tool for porphyry Cu deposits. Temporally, detrital zircons can be used to trace the evolution of redox and hydration state for a long period of geological history. This study provides an example of the application of detrital zircons in the Tethyan domain. Higher oxidation and hydration state revealed by zircon Ce-U-Ti (Fig.\n \n 2\n \n ) and the ratio of europium anomaly to ytterbium (Supplementary Fig.\u00a03) in Cenozoic intrusions implies higher Cu prospectivity. Spatially, detrital zircons can provide a much wider halo than other more conventional geochemical exploration media, particularly when applied in paleo-watersheds to identify potential fertile units upstream from the sampling site\n \n \n 18\n \n ,\n \n 60\n \n ,\n \n 61\n \n \n . This exploration tool is also applicable in other orogens or geological periods globally.\n
\n\n \n Materials.\n \n The zircon age and trace elements are compiled from the global detrital zircon database\n \n \n 62\n \n \n and our new data in Tibet. The complete database includes 13981 detrital zircons.\n
\n\n \n Zircon U\u2013Pb dating and trace element analyses.\n \n Zircon grains were separated from two rivers in Tibet and mounted in epoxy resin for age and trace element determinations. Zircon U\u2013Pb dating and trace element analyses were conducted simultaneously by laser ablation\u2013inductively coupled plasma\u2013mass spectrometry (LA\u2013ICP\u2013MS) in the Mineral and Fluid Inclusion Microanalysis Laboratory, Institute of Geology, Chinese Academy of Geological Sciences, Beijing, China. The NWR 193\n \n UC\n \n laser ablation system (Elemental Scientific Lasers, USA) was equipped with a Coherent Excistar 200 excimer laser and a two-volume ablation cell. The laser ablation system was coupled to an Agilent 7900 ICP\u2013MS instrument (Agilent, USA). Zircons were mounted in epoxy resin discs, polished to expose grain interiors, cleaned ultrasonically in ultrapure water, and then cleaned again before analysis using analytical-grade methanol. Pre-ablation was conducted before each spot analysis using five laser shots (~\u20090.3 \u00b5m in depth) to remove potential surface contamination. The analyses used a 30 \u00b5m laser beam diameter with a laser frequency of 8 Hz and fluence of 2 J/cm\n \n 2\n \n . The Iolite software package was used for data reduction\n \n \n 63\n \n \n . Zircons 91500 and GJ-1 were the primary and secondary reference materials, respectively. The exponential function was used to correct for down-hole fractionation\n \n \n 64\n \n \n . NIST 610 and\n \n 91\n \n Zr were used to calibrate the trace element concentrations as an external reference material and internal standard, respectively. Zircon age and trace element data are listed in Supplementary Table\u00a02.\n
\n\n \n Calculation of oxygen fugacity.\n \n Data points outside the Tethys domain were excluded. Zircon ages younger than 200 Ma are used for the following discussion to minimize the influence of the Paleo-Tethys Oceanic subduction because the closure of the Paleo-Tethys Ocean occurred in the Late Permian\u2013Late Triassic\n \n \n 5\n \n \n . Zircons with a Th/U ratio of <\u20090.1 were used to exclude metamorphic zircon\n \n \n 65\n \n \n . Moreover, zircons derived from S-type granites were screened out using zircon P and REE contents\n \n \n 66\n \n \n . After screening, 3010 zircon grains were used for reconstructing the Tethyan oxygen fugacity variation (Supplementary Table\u00a02). The novel magmatic oxybarometer using ratios of Ce, U, and Ti in zircon is independent of temperature and pressure, with a standard error of \u00b1\u20090.6 log unit \u0192O\n \n 2\n \n \n 67\n \n . The \u2206FMQ values can be calculated by the equation:\n
\n\n \u2206FMQ\u2009=\u20093.998 (\u00b1\u20090.124) \u00d7 log [Ce/\u221a (U\n \n i\n \n \u00d7 Ti)]\u2009+\u20092.284 (\u00b1\u20090.101) (1)\n
\n\n where U\n \n i\n \n denotes age-corrected initial U content\n \n \n 67\n \n \n . The \u2206FMQ values are plotted as binned averages (bin size\u2009=\u20095 Myr).\n
\n\n \n Monte Carlo isotopic simulation.\n \n The Sr-Nd concentrations and isotopic ratios of the sediment\u2013depleted mantle mixture can be calculated using the following equations:\n
\n\n Cx\u2009=\u2009Cs\u00d7f\u2009+\u2009Cm\u00d7(1-f) (2)\n
\n\n Rx\u2009=\u2009Rs\u00d7f\u00d7Cs/Cx\u2009+\u2009Rm\u00d7(1-f)\u00d7Cm/Cx (3)\n
\n\n where x is Sr or Nd, Cx, Cs, and Cm are element concentrations of x after mixing, sediment, and depleted mantle, respectively. Rx, Rs, and Rm are isotopic ratios of x after mixing, sediment, and depleted mantle, respectively, and f is the proportion of sediment-derived melt.\n
\n\n To obtain as many as possible mixture scenarios, random two pelagic or terrigenous sediments were selected and mixed to create a new pelagic or terrigenous sediment end-member. Then the new sediment end-member was mixed with the depleted mantle in random proportions. 30000 pelagic sediment\u2013mantle mixtures and 30000 terrigenous sediment\u2013mantle mixtures were generated using this method. The simulation results are shown in Fig.\n \n 4\n \n c\u2013d.\n
\n\n Supplementary Table 1\n
\n \n\n Supplementary Table 2\n
\n \n\n Supplementary Table 3\n
\n \n\n Supplementary Table 4\n
\n \n\n Supplementary Materials for Types of Subducted Material Controlling Tethyan Porphyry Copper Mineralization\n
\n \n\n Although cities have risen to prominence as climate actors, emissions data scarcity has been the primary challenge to evaluating their performance. Here we develop a scalable, replicable machine learning methodology for evaluating the mitigation performance for nearly 50,000 local and municipal actors in the European Union from 2001\u20132018. We find that participation in one of the largest voluntary transnational climate initiatives is associated with a 1.6 percent reduction in annual emissions. Overall, these cities representing 301\u00a0million inhabitants have reduced nearly 186\u00a0million tons of carbon dioxide emissions. Compared to only 35 percent of external cities that have reduced emissions, 84 percent of cities participating in transnational climate governance have reduced emissions over the same time period. Participating cities reporting emissions data on average have higher annualized per capita reduction compared to cities without reported emissions. These findings provide quantitative evidence urban climate governance initiatives\u2019 effect on global climate mitigation.\n
\n\n Cities have in recent years risen in prominence on the global sustainability policy agenda, as researchers and policy-makers have increasingly focused on urban jurisdictions as powerful policy actors in their own right. More than 10,000 of the world\u2019s cities are pledging various forms of climate mitigation, adaptation, and financing actions, and in many instances these municipalities participate in multiple voluntary transnational climate initiatives.\n \n \n 1\n \n \n As part of these initiatives\u2019 requirements, in accordance with national government directives\n \n \n 2\n \n \n , or on their own volition, cities articulate strategies and policies to tackle climate change mitigation and, less frequently, adaptation. Cities predominantly put forth mitigation strategies centered on greenhouse gas emission reduction targets, often achieved through policies focused on increasing the use of sustainable transport, enhancing the efficiency of lighting in public and municipal buildings, adopting energy efficiency standards, promoting climate awareness to encourage citizen action, and other areas\n \n \n 3\n \n ,\n \n 4\n \n \n .\n
\n\n There are thousands of current strategies and policies detailing urban mitigation efforts, yet, as Milojevic-Dupont & Creutzig (2021)\n \n \n 5\n \n \n point out, there is little understanding of these actions\u2019 effects. These knowledge gaps cause policymakers to be \u201cdisoriented on which measures are adequate and impactful\u201d in urban areas and uncertain which \u201ceveryday decisions\u201d regarding planning or infrastructure investments should be made to achieve mitigation targets. Little is known about the emission reductions from common urban climate policies and strategies, a missing block of vital information acknowledged in Chap.\u00a012 on Human Settlements in the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC)\n \n \n 5\n \n ,\n \n 6\n \n \n .\n
\n\n Scholars have argued that cities\u2019 involvement in transnational climate governance \u201ccan accelerate their actions to curb GHG emissions under certain conditions\u201d\n \n \n 7\n \n \n . The evidence in support of this claim is scarce, making it hard to predict precisely what conditions would have this effect. Transnational climate initiatives typically require reporting of climate action plans and regular monitoring in the form of emissions inventories to assess whether mitigation goals are met, yet in practice only a small fraction of subnational actors meet these requirements\n \n \n 8\n \n ,\n \n 9\n \n \n . Hsu et al. (2020)\n \n \n 10\n \n \n found that out of more than 9,000 cities that were signatories to the EU Covenant of Mayors for Climate and Energy (EUCoM) initiative, only approximately 15 percent had reported any emissions data, and even fewer (around 11 percent) had reported both a baseline emissions inventory and an additional year of inventory emissions data needed to track progress towards voluntary reduction targets. When emissions data are available, they are frequently incomparable due to the limited availability of datapoints, a general lack of transparency regarding underlying methodologies, and the lack of standardized accounting approaches. Ibrahim et al. (2012)\n \n \n 11\n \n \n evaluated seven distinct city-scale greenhouse gas emissions inventory protocols and methodologies and concluded that a common reporting standard or approach is needed for cities. Differences in the various standards\u2019 definitions \u2013 e.g. for emission scopes, particularly in Scope 3 supply chain emissions \u2013 must be addressed so that participants emissions\u2019 data can be appropriately compared.\n
\n\n Recent advances in machine learning (ML), the application of computational algorithms usually applied to large-scale datasets to simulate human learning, help us overcome these tricky emissions data challenges.\n \n \n 12\n \n \n In this study, we employ a ML-driven approach to estimating and evaluating the performance for nearly 50,000 local and municipal actors in the European Union from 2001\u20132018. Our method develops a process for identifying spatial boundaries and geospatial predictors for each local and municipal government participating in the EUCoM, one of the largest voluntary transnational climate governance initiatives, and then utilizing the self-reported carbon emissions inventory data from 6,114 participating EUCoM cities as training data in an extreme gradient boosting model. To our knowledge, our resulting dataset is the most comprehensive time series dataset used to evaluate city-level carbon emissions and mitigation performance. We apply these data to evaluate the performance of three groups of European cities: \u201creporting\u201d cities that have reported at least one year of emissions data; \u201cparticipating\u201d cities that have pledged voluntary climate action but have not reported any emissions data; and last, \u201cexternal\u201d cities representing local administrative units (LAUs) that are not participants.\n
\n\n Figure\n \n 1\n \n a shows the correlation between the city-level dependent (i.e., self-reported \u201cemissions\u201d) and independent variables (i.e., heating degree days, fossil-fuel CO2, GDP per capita, etc.). We found a strong positive correlation between reported emissions inventory data and stationary fossil-fuel CO\n \n 2\n \n emissions from the Open-Data Inventory for Anthropogenic Carbon dioxide (ODIAC)\n \n \n 13\n \n \n (r\n \n 2\n \n =\u2009.81), as well as between emissions and population (r\n \n 2\n \n =\u2009.89). Population and stationary fossil-fuel CO\n \n 2\n \n emissions were also highly correlated (r\n \n 2\n \n =\u2009.79), confirming prior studies that demonstrate through the use of nighttime lights intensity the relationships between these data and energy consumption, economic activity, and fossil-fuel emissions.\n \n \n 14\n \n \n Our analysis did not show strong relationships between self-reported emissions data and GDP per capita (r\n \n 2\n \n =\u20090.03) or with fine particulate air pollution (PM\n \n 2.5\n \n ; r\n \n 2\n \n =\u20090). We determined that stationary fossil-fuel CO\n \n 2\n \n emissions and population were the primary predictors of cities\u2019 self-reported emissions data with the highest contribution or importance to our emissions model (Fig.\n \n 1\n \n b). Figure\n \n 1\n \n b shows the gain value of the importance of each of the top six features we considered. The gain values are determined by the amount each attribute split improves the model\u2019s performance, weighted by the number of observations for the node. See Methods for more description about the grid search process and parameter tuning to determine the final model.\n
\n\n We predicted emissions for nearly 50,000 cities where we had underlying spatial data. Figure\n \n 2\n \n presents scatterplots of cities\u2019 self-reported emissions data compared to our model\u2019s predicted emissions data. The resulting r\n \n 2\n \n =\u20090.94 indicates our model is strongly predictive overall of cities\u2019 self-reported emissions inventories. We further validated our predicted emissions with other studies that report emissions data for European cities, including Moran et al. (2022)\n \n \n 15\n \n \n , who estimate 2018 direct (Scope 1) emissions for more than 100,000 European cities and Nangini et al. (2019)\n \n \n 16\n \n \n , who combine self-reported inventories with other data for 343 global cities. We found fair correlation (r\n \n 2\n \n =\u2009.50 with Moran et al., 2022; r\n \n 2\n \n =\u2009.62 with Nangini et al., 2019) between our predicted data and these other studies (Supplementary Fig.\u00a07). Since most of the cities that report emissions data are small (mean population\u2009=\u200939,234; median population\u2009=\u20095,465), we find that our model tends to perform slightly better for larger cities (r\n \n 2\n \n =\u2009.98), although this trend, and high correlation coefficient, appears to be largely driven by a few very large global cities like London and Berlin. For smaller cities, which comprise the majority of EUCoM cities and those reporting emissions inventories used to train our model, the model tends to underpredict self-reported emissions (r\n \n 2\n \n =\u2009.77). As explained further in the Methods and Supplementary Information, we configured multiple models (e.g., separate models for large vs. small cities), but none performed as well in terms of minimizing error (i.e., RMSE) and achieving a high correlation (i.e., r-squared) between self-reported and predicted emissions. Figure\n \n 2\n \n b also shows the self-reported emissions data vs. predicted emissions data by country, which allows closer examination of potential eccentricities in our model or the predicted data. For instance, there are several cities in France where our model overpredicts their emissions. Further inspection of one of these outliers, Lyon, a city of 445,000 people in France, reports an emissions inventory of around 22,000 tons, resulting in per capita emissions of less than 0.05 tons, far below the national average of 5.4 tons per person.\n \n \n 17\n \n \n
\n\n Our model provides annual emissions predictions for 2001 to 2018, the latest year for which we have fossil-fuel based CO\n \n 2\n \n emissions. Figure\n \n 3\n \n selects three illustrative time series plots for three cities of varying population size: Waimes in Belgium (population\u2009=\u20098,711), Tolosa in Spain (population\u2009=\u200917,349), and London in the United Kingdom (population\u2009=\u20098.6\u00a0million). In the case of Waimes, the city reported one baseline emissions inventory for the year 2006. Our model predicts slightly higher emissions (207 tons or 0.03 tons per capita) than its reported inventory. For Tolosa and London, both cities reported both a baseline and monitoring emissions inventory, and our predicted emissions show similar trends for both actors. Our model slightly underpredicts Tolosa\u2019s baseline emissions in 2007 (0.2 tons per capita) and inventory emissions in 2015 (0.01 tons per capita). For London, a similar trend emerges - our model slightly underpredicts the city\u2019s 2008 baseline emissions (0.05 tons per capita) and 2013 inventory emissions (0.01 tons per capita). On average, our model tends to slightly overpredict emissions (0.9\u2009\u00b1\u20092.23 tons per capita) compared to cities\u2019 self-reported emissions.\n
\n\n Utilizing the predicted emissions from our model, we analyzed trends in annual per capita emissions reduction over the time period from 2001 to 2018 for cities participating in the EUCoM that report emissions data (reporting cities), those that do not report (participating cities), and for all LAUs in Europe (external cities).\n
\n\n Overall, we find that EUCoM cities have reduced emissions from 2001 to 2018 compared to external cities in the European Union that are not signatories (-1.22\u2009\u00b1\u20092.00 vs. 5.21\u2009\u00b1\u200911.03 annual per capita emissions trend; Table\n \n 2\n \n ). While 84 percent of EUCoM cities have reduced emissions during this time period, only 35 percent of external cities achieved a negative trend in emissions reductions. We interpret these emission trend differences between EUCoM cities and external LAUs with caution, however, noting the differences most notably in population between EUCoM (34,270\u2009\u00b1\u2009199,844 inhabitants for reporting cities; 34,693\u2009\u00b1\u2009161.567 for participating cities) and external LAUs, which tend to be on average must smaller (8,348\u2009\u00b1\u200922,851 inhabitants) (Table\n \n 1\n \n ; Supplementary Fig.\u00a06). Descriptive statistics (Table\n \n 1\n \n ) and distributions (Supplementary Fig.\u00a06) describing the three groups of cities in our analysis illustrate that EUCoM cities tend to have more sizeable stationary fossil-fuel carbon dioxide emissions and be larger in population and population density than external cities, which could explain differences in their emissions trends.\n
\n\n Within the EUCoM cities, we find that nearly 8,000 participating cities with 301\u00a0million inhabitants have reduced emissions 185.82\u00a0million tons between 2001\u20132018. Based on our quasi-experimental interrupted time series analysis, which models whether a policy intervention or program may have resulted in a measurable change in an outcome variable after its implementation,\n \n \n 18\n \n \n we find that joining the EUCoM is associated with a -1.64 (se: 0.13) percent annual per capita reduction, when accounting for differences by country and holding GDP per capita, per capita emissions, and population density constant (Table\n \n 4\n \n ). Thirty-eight percent of participating EUCoM cities achieved a greater annualized per capita emissions reduction after they joined the EUCoM, on average 3.67\u2009\u00b1\u20095.66 percent more than the year prior to their adhesion year.\n
\n\n Whether EUCoM cities self-report emissions data may be a predictor of mitigation performance. Seventy-five percent of EUCoM cities have reported at least one year of emissions data. At the country level, we observed large variation in the percentage of EUCoM cities reporting inventory data \u2013 e.g. 96 percent of Slovenia\u2019s 28 cities have reported at least one year of inventory data; while only 10 percent of nearby Slovakia\u2019s 29 cities evaluated have reported (Table\n \n 3\n \n ). We observed a performance gap between reporting EUCoM cities and participating but not reporting cities (mean difference\u2009=\u20091.52; p\u2009<\u20090.01; Table\n \n 2\n \n ). Despite being comparable in terms of population, population density, and GDP per capita (Table\n \n 1\n \n ; Supplementary Fig.\u00a06), reporting cities on average reduced per capita emissions 1.6\u2009\u00b1\u20092.0 from 2001 to 2018, while participating cities exhibited no or minimal reductions (-0.08\u2009\u00b1\u20091.5).\n
\n\n The EUCoM required cities to adopt at minimum a 20 percent reduction target by 2020 and at least a 40 percent reduction target by 2030, and we incorporated this information in two ways. First, we identified participating EUCoM cities as adopting \u201cambitious\u201d (i.e., greater than 20 percent reduction by 2020 or beyond the EU\u2019s own 2020 target) or \u201cunambitious\u201d (i.e., adopting the minimum target). This classification allowed us to investigate whether participation in the EUCoM signals fundamental differences in participating cities compared to others (e.g., underlying structural differences that may predispose them to achieving certain outcomes). If our model is able to predict unambitious and ambitious reporting cities\u2019 equally well, this result would suggest that the model is valid for external cities that are equally \u201cunambitious\u201d (i.e., have not exceeded the EU\u2019s 20 percent reduction target). We did not find the designation of an \u201cambitious\u201d (i.e., greater than 20 percent reduction by 2020 or beyond the EU\u2019s own 2020 target) emissions target a contributor to our predictive model (Fig.\n \n 1\n \n b), nor did we find differences in our model\u2019s predictions of ambitious or unambitious cities\u2019 emissions (Supplementary Fig.\u00a05; see Methods:Limitations). Participating EUCoM cities that adopted \u201cambitious\u201d 2020 emissions reduction targets that exceed the EU\u2019s, however, achieved higher annual per capita emissions reductions of -1.53\u2009\u00b1\u20092.7 (n\u2009=\u20093,570), compared to those that have adopted only the minimum (-0.47\u2009\u00b1\u20092.4; n\u2009=\u20093,964) (Table\n \n 2\n \n ). Second, we used our predicted emissions data to determine whether participating cities were on track to achieving their targets, replicating the method we used in Hsu et al. (2020). Fifty-five percent of participating cities were on track to achieving their emissions reduction targets, with Scandinavian countries in the lead (87 percent in Denmark; 67 percent in Finland and Norway). Spain also boasts a large proportion of cities on track, with 74 percent. Twenty-nine percent of participating cities were not making sufficient progress towards their targets, while 16 percent have increasing emissions.\n
\n\n We observe differences in performance by country. Figures\n \n 4\n \n and\n \n 5\n \n compare the performance of participating EUCoM cities versus all other LAUs by country. In some countries, EUCoM cities, such as those in the Netherlands and Malta, on average have had higher annual per capita reduction trends than their non-EUCoM counterparts, although participating EUCoM cities in Netherlands tend to be larger than external cities (121,606\u2009\u00b1\u2009177,804 for participating vs. 35,594\u2009\u00b1\u200926,906 for external cities). In others, such as Denmark and the United Kingdom, EUCoM cities appear to be underperforming compared to their counterparts (Fig.\n \n 4\n \n ), as evidenced by comparing the distributions of annual per capita emissions reductions for both groups of cities. This result may reflect the fact that the national governments of Denmark and the United Kingdom require local climate action plans from municipalities (Reckien et al., 2018). Italy and Spain, where most of the EUCoM cities are located, appear to have relatively comparable performance for both groups, despite the significant percentage of emissions covered by EUCoM cities in both countries (Italy\u2009=\u200960 percent; Spain\u2009=\u200944 percent; Table\n \n 3\n \n ). Countries where cities perform similarly are closer to the diagonal line in Supplementary Fig.\u00a08, suggesting that the mean annual per capita emissions reduction trends are similar among EUCoM and external cities. Countries above the diagonal are those where EUCoM cities have achieved greater annual per capita emissions reductions than their non-EUCoM counterparts and include countries like Albania, Norway, Malta, Germany, Poland, among others. While acknowledging the limitations of our model in performing out of sample as well as the inherent differences and similarities between EUCoM cities and external cities, the findings point to the need for further data collection and research in this direction.\n
\n\n Despite a measurable increase in urban climate governance scholarship over the past decade, gaps in understanding outcomes for transnational climate initiatives have persisted, particularly for smaller cities and on a systematic basis.\n \n \n 19\n \n \n Part of this gap is due to data availability and comparability, which limit researchers\u2019 ability to trace causal impacts or linkages between the processes and institutions of transnational urban climate governance initiatives to outcomes.\n \n \n 19\n \n ,\n \n 20\n \n \n To address this shortcoming, this study has developed a machine learning (ML)-based framework to predict nearly 50,000 European cities\u2019 emissions on an annual basis from 2001 to 2018 to evaluate cities adhering to one of the largest transnational climate governance initiatives. By utilizing globally gridded, spatially explicit predictor variables that are measured consistently and regularly and available self-reported emissions inventories, our ML-based model is able to explain 94 percent of the variation (r\n \n 2\n \n =\u20090.94) between self-reported emissions inventory data from recording EUCoM cities and predicted emissions values, validated through comparisons with other studies that have produced city-level carbon emission estimates for a single year. We provide clear evidence that participating in the EUCoM is associated with a 1.6 percent reduction in annual per capita emissions. Compared to only 35 percent of external cities that have reduced emissions, 84 percent of cities participating in transnational climate governance have reduced emissions over the same time period. Participating cities that reported emissions inventory data on average have achieved higher annualized per capita reduction compared to participating cities without reported emissions data. Our method and resulting dataset allow for the largest-scale examination of municipal and local government climate emissions over time, shedding light on the impact of urban climate governance initiatives that was previously unattainable due to the lack of comparable, consistent data.\n
\n\n Our findings that participating EUCoM cities observe emissions reductions after they adhere to the initiative and compared to external counterparts provides, to our knowledge, the largest-scale evidence suggesting an association between participating in a transnational climate initiative and direct mitigation impacts, although we lack full understanding of the causal mechanisms driving these results. We observe cities a measurable decrease in annual per capita emissions changes around the year in which participating cities join the EUCoM, on average 1.6 percent when controlling GDP per capita, population density, and per capita emission levels constant. Since emissions reductions are generally easier to achieve at the outset when cities design climate action plans to tackle easier-to-achieve reductions through energy efficiency gains, conducting energy audits of buildings, and purchasing more fuel-efficient vehicles,\n \n \n 21\n \n ,\n \n 22\n \n \n their transformations tend to follow an \u201cS-shape,\u201d where initial gains then slow down as incremental gains in reductions become more difficult to achieve or have already been met.\n \n \n 23\n \n \n Fig.\n \n 5\n \n illustrates similar trends in annual per capita emissions, where magnitudes reduce as time passes from the adhesion year, suggesting deeper transformational changes needed for cities adopting longer-term, decarbonization goals.\n \n \n 24\n \n \n
\n\n Although the ITS design does not rule out the possibility that there could be some other unobservable or unmeasurable factor driving these results (see Methods), the finding that a majority (84 percent) of the EUCoM cities have reduced emissions in the observed time period echoes the results of our 2020 study of 1,066 EUCoM cities that have reported at least two emissions inventories. There, we found that 60 percent of cities were on track to achieve their 2020 emissions reduction targets, whereas this study found 55 percent to be on track. Our results provide support and clarity to previous studies evaluating the impact of transnational climate initiatives and cities\u2019 mitigation performance. Kona et al. (2016), for example, estimated that 6,201 EUCoM cities, representing 213\u00a0million inhabitants, could reduce emissions by 254\u00a0million tons CO\n \n 2\n \n e in 2020 based on their pledged commitments, which were on average 7 percent higher than the 20 percent reduction target for the EU. The authors analyzed 315 reporting cities and found that they had reduced emissions by 23 percent on average. Since our analysis demonstrates reporting cities are driving most of the reductions compared to participating cities, the anticipated 254\u00a0million estimated tons in reductions in 2020 would largely hinge on reporting cities delivering these reductions. Yet at the time they made this report less than 5 percent of EUCoM cities had reported a baseline and monitoring emissions report. Our study, therefore, contributes the first wide-scale evidence of the scale and scope of cities\u2019 mitigation contributions and the associated effect of participating in urban climate governance initiatives like the EUCoM.\n
\n\n While our study does not speak to causal mechanisms of the predicted emissions, nor whether there are endogenous conditions that may explain why EUCoM cities have experienced on average greater annual per capita reductions than their external non-EUCoM counterparts, it does suggest some insights relevant for urban climate governance and transnational climate initiatives. First, since emissions inventories and monitoring protocols are considered hallmarks of effective local governments\u2019 climate mitigation plans,\n \n \n 8\n \n \n the ability to monitor and report emissions are likely indicators of capacity and achievement. We measured significant differences between annualized per capita emissions reductions between reporting cities and participating cities that fail to report any emissions data. Second, while assessing emissions trends, as an outcome variable does not provide a \u201cmeasure of effort\u201d\n \n \n 25\n \n \n nor describe the myriad inputs and factors that have led to a particular outcome, monitoring and reporting emissions inventories indicates a \u201cmeans of implementation\u201d\n \n \n 26\n \n \n for evaluating an entity\u2019s progress towards a climate policy outcome like climate mitigation. Data describing mitigation outcomes then allow for identification of \u201cgeneral conditions of successful implementation\u201d and reverse engineering of causal pathways that led to the emissions reductions. Our dataset and replicable, scalable ML-framework can subsequently provide a first step towards disentangling which specific measures, or none at all, led to the observed emissions reductions. Since we were limited to data on cities\u2019 population, GDP, and fossil-fuel CO\n \n 2\n \n emissions, our analysis cannot account for other underlying structural differences (e.g., variation in governance institutions, etc.) that may further elucidate differences in emissions outcomes, since climate change action and policies are \u201cdeeply entwined with other policy agendas. \u201d\n \n \n 27\n \n \n
\n\n \n Future Research\n \n
\n\n Since the availability of self-reported emissions inventory data at the subnational level is primarily constrained to Europe, future studies must broaden the search for relevant datasets and proxies that can fill this gap, particularly for capacity- and resource-constrained entities in the Global South.\n \n 28\u201331\n \n Actors in these countries face limitations (e.g., expertise, lack of clearly designated roles in relevant government agencies for producing inventories, insufficient documentation and archival systems) and technical issues (e.g., incomplete or non-existent activity data or lack of experimental data for developing countries or technology-specific emission factors) for producing emissions inventories\n \n 10,32\n \n . Our next step is to expand our approach to a set of subnational jurisdictions outside of Europe to produce a global dataset for cities participating in transnational climate initiatives, as recorded in Hsu et al.\u2019s (2020)\n \n 10\n \n dataset of more than 12,000 cities and regional governments.\u00a0We have produced a scalable, reproducible framework and methodology for identifying spatial boundaries of cities and are able to match these boundaries to globally-gridded datasets, and then to utilize self-reported emissions and other data to predict and validate a machine-learning model. We find compelling evidence that large-scale, geospatial datasets can be applied to estimate city-level carbon dioxide emissions, even for small city actors that comprise the majority of participants in the EUCoM. Our method bridges the gap between these globally available, remote-sensing derived geospatial datasets to city-scale actors, a shortcoming Pan et al. (2021)\n \n 33\n \n note in fossil-fuel CO2 datasets like the ODIAC inventory, which primarily distributes national fossil-fuel CO\n \n 2\n \n emissions spatially based on satellite measurements of light-output intensity, and which may not correctly attribute emissions to subnational actors.\n
\n\n This research is a first step towards addressing the \u201clack of systematic knowledge on global contributions of cities to the Paris Agreement,\u201d\n \n \n 34\n \n \n which acknowledges the role of \u201call levels of government\u201d\n \n \n 35\n \n \n and seeks specific information regarding their impacts.\n \n \n 36\n \n \n Few city actors participating in transnational climate initiatives report monitoring and inventory data, and even major cities claiming global climate leadership are absent from reporting.\n \n \n 9\n \n ,\n \n 10\n \n ,\n \n 34\n \n ,\n \n 37\n \n \n Our study provides the most consistent approach and time series data to date, providing quantitative evidence of cities\u2019 participating in transnational climate governance mitigation performance, with potential for broadening the scope to areas outside of Europe.\n
\n\n Consistent, comparable, and widespread emissions data are essential to support the Paris Agreement\u2019s \u201cfacilitative and catalytic\u201d\n \n \n 38\n \n \n mode and its \u201cpledge and review and ratchet\u201d mechanism designed to continuously evaluate national and subnational actors\u2019 progress and contributions to global mitigation efforts.\n \n \n 39\n \n \n For virtuous, catalytic cycles supporting this process to occur, emissions data are needed to assess which actions are effective in driving mitigation.\n
\n\n Data for cities participating in the EUCoM were collected from two sources: Kona et al. (2021), which provides a \u201cverified and harmonized version\u201d of the EUCoM data for 6,200 member cities as of the end of 2019. The Kona et al. (2021) dataset for EUCoM cities includes self-reported emissions data (e.g., baseline or monitoring emissions inventories), as well as other characteristic data of the cities from the European Statistical Agency. We supplemented this dataset with more recent data for cities from the EUCoM website, which was scraped using the Beautiful Soup Python package (Richardson, 2007) in February 2021. We primarily collected information on each cities\u2019 adhesion date to the EUCoM initiative, baseline emissions year, baseline emissions (in tons of carbon dioxide emissions or tCO\n \n 2\n \n ), emissions reduction target, target year, and any reported inventory emissions (i.e., emissions data reported at a later year than a defined baseline year, from each city\u2019s Progress page). We also derived information regarding the cities\u2019 population and geographic coordinates (latitude/longitude) from the EUCoM website if available. Since Kona et al. (2021) apply a series of statistical techniques to validate their dataset, we prioritized self-reported emissions data from this source if there were data available for a city both in Kona et al. (2021) and the EUCoM website. Supplementary Fig.\u00a01 shows a scatterplot of the logged emissions data from both the EUCoM website and Kona et al. (2021), which illustrates a strong correlation (r\n \n 2\n \n =\u20090.986). In total, our dataset contained names of 8,242 cities participating in the EUCoM initiative, with 6,309 reporting any emissions information. We also imputed a 20 percent emissions reduction target by 2020 if no specific emissions reduction target was reported in Kona et al. (2021) or on the EUCoM website for the purposes of the tracking progress analysis described in our previous study (Hsu et al., 2020).\n
\n\n An important first step in building our predictive emissions model was determining a set of underlying predictors of city-level carbon emissions that would be universally available for all EUCoM cities and LAUs in Europe. We evaluated several predictors of urban greenhouse gas emissions to include as predictors in our model, based on existing literature regarding major sources and drivers of cities\u2019 emission profiles (Seto et al., 2014; Marcotullio et al., 2013; Dodman, 2009; Rosa and Dietz, 2012). In terms of emission sources, the energy sector, specifically conversion of energy to electricity, is the largest source of urban greenhouse gas emissions, comprising around half upwards to 65 percent of total urban emissions, followed by the transportation sector (15 to 20 percent) (Marcotullio et al., 2013). Since stationary sources do not explain city greenhouse gas emissions in their entirety, we also investigated other proxies for major emissions sources, including heating and cooling demand, and fine particulate air pollution, which in cities results primarily from transport (~\u200925 percent; Karagulian et al., 2015). We also included population and gross domestic product (GDP) as relevant socioeconomic drivers of urban climate emissions (Seto et al., 2014), and evaluated a few country-level predictors, based on our previous study (Hsu et al., 2020) that found national-level emissions reductions were predictors of city-level climate change performance, including country-level CO\n \n 2\n \n emissions trend (2000\u20132018) (WRI CAIT, 2020) and carbon intensity of electricity-generation for the European Union (EUROSTAT, 2021). Further details on the sources of these datasets and their processing are detailed in Methods.\n
\n\n Since high-resolution emissions data as a result of electricity production and consumption are not available for the vast majority of cities included in our analysis, we relied on the Open-Data Inventory for Anthropogenic Carbon Dioxide (ODIAC) database, which provides a globally-gridded, annual 1 km x 1 km spatial resolution data of carbon dioxide emissions from fossil fuel combustion, cement production, and gas flaring from 2000 to 2019.\n \n \n 13\n \n \n We selected the ODIAC dataset based on prior evaluation of its relevance for urban-level carbon emissions analysis, as described in Hsu et al. (2020).\n \n \n 10\n \n \n
\n\n As proxies for building energy consumption due to heating and cooling, we downloaded monthly-averaged, (0.5x0.625 degree or 55.5 x 69.375 km) spatial resolution land surface temperature data from the NASA MERRA-2 temperature product\n \n \n 40\n \n \n and then calculated heating and cooling degree days (HDD and CDD, respectively) based on the number of monthly-averaged measurements that deviate from a baseline temperature,\n \n \n \\({T}_{base}\\)\n \n \n , which were then multiplied according to the number of days in each respective month (i.e., assuming the same HDD or CDD for each day of the month) and then summed across a year, according to the Equations (1\u20132) below:\n
\n\n \n \n \\(HDD= {{\\Sigma }}_{m}\\left({T}_{base}-{T}_{i}\\right) \\times {{Days}_{m}}^{+}\\)\n \n \n (Eq.\u00a01)\n
\n\n \n \n \\(CDD= {{\\Sigma }}_{m}\\left({T}_{i}-{T}_{base}\\right) \\times {{Days}_{m}}^{+}\\)\n \n \n (Eq.\u00a02)\n
\n\n where\n \n \n \\({T}_{base}=\\)\n \n \n 15.5 degrees C for HDD and\n \n \n \\({T}_{base}\\)\n \n \n = 22 degrees C for CDD\n \n \n 41\n \n \n and\n \n \n \\(m\\)\n \n \n is the month. For the EU model, we excluded cooling degree days since 99 percent of European cities had 0 cdd.\n
\n\n We included an annual, gridded (~\u20091 km ) exposure to fine particulate matter pollution (PM\n \n 2.5\n \n ) for years 2001 to 2015\n \n 42\n \n , since PM\n \n 2.5\n \n pollution is generated from sources similar to carbon emissions in urban areas, mainly fossil fuel combustion from electricity generation and transportation.\n \n \n 43\n \n \n We also evaluated a few country-level predictors, based on a previous study\n \n \n 10\n \n \n that found national-level emissions reductions were predictors of city-level climate change performance, including country-level CO\n \n 2\n \n emissions trend (2000\u20132018)\n \n \n 17\n \n \n and carbon intensity of electricity-generation for the European Union,\n \n \n 44\n \n \n although our final model did not include these variables, since they did not contribute significantly to the feature importance for our model (Fig.\n \n 1\n \n b).\n
\n\n We further accounted for population and gross domestic product (GDP) as relevant socioeconomic drivers of urban climate emissions.\n \n \n 6\n \n \n For population, we used the Gridded Population of the World (GPW) dataset,\n \n \n 45\n \n \n which provides population estimates at a 1-km spatial resolution for five-year increments from 2000 to 2020. We calculated annual population estimates by linearly interpolating between these five-year increments. For GDP, we used a globally, annually gridded GDP per capita data at a 1-km spatial resolution from Kummu et al., 2018,\n \n \n 46\n \n \n which provides data from 1990 to 2015. We used a spline interpolation method using the na_interpolation function from the imputeTS package\n \n \n 47\n \n \n in R to impute GDP per capita values for cities from 2016 to 2018 to match the time series of the other spatial predictors.\n
\n\n Since the original format of these predictor variables (e.g., fossil-fuel CO\n \n 2\n \n emissions) are all gridded spatial data, we merged these datasets to each EUCoM city through spatial joins. We first collected the latitude and longitude of each city\u2019s centroid as provided by the various data sources. When the city centroids were not available from Kona et al. (2021),\n \n \n 48\n \n \n EU Covenant of Mayors\u2019 website, or we determined errors in the geographic coordinates from either of these sources, we extracted the city centroids through Wikipedia\u2019s GeoHack website (citation:\n \n \n https://www.mediawiki.org/wiki/GeoHack\n \n \n \n \n \n ).\n \n
\n\n To determine each city\u2019s spatial boundaries, we used distinct approaches described below. For most of the cities, we collected data for local administrative units (LAUs), which are defined as \u201clow-level administrative divisions of a country below that of a province, region or state,\u201d for all 28 European Union countries from the European Union\u2019s Statistical Agency.\n \n 49\n \n The LAU data was spatially joined to our EUCoM city data frame in Python using the geopandas\n \n 50\n \n package to associate each city with a LAU boundary for the purposes of matching additional predictor variables. We implemented a series of quality checks to ensure that the spatial joins were conducted correctly and to identify any issues in the geographic coordinates that may have been incorrectly specified on the EU Covenant website. These quality checks include 1) evaluating whether cities have the same geographic coordinates but are identified with distinct names; 2) comparing the reported population in the Kona et al. (2021)\n \n \n 48\n \n \n or EUCoM website for an individual actor and the interpolated population after the spatial join; 3) examining any city with self-reported per capita emissions less than 0.2 tons per person or greater than 40 tons per person; 4) compound annual growth rate in emissions is greater than \u2212\u200950 percent and less than 50 percent. These checks allowed us to determine whether there were any errors in the spatial join or underlying data collected for the EUCoM cities from either Kona et al. (2021)\n \n \n 48\n \n \n or the EUCoM website.\n
\n\n Where manual corrections to LAUs also did not result in correct spatial joins, we utilized OpenStreet Map (OSM)\n \n 51\n \n to get the correct boundary, particularly for large cities that may encompass more than one LAU. Supplementary Fig.\u00a02 illustrates a few examples of the incorrect spatial join results and the fixed boundaries with OSM. After we verified the cities\u2019 boundaries, we then applied zonal statistics using the Python package rasterstats version 0.15.0,\n \n 52\n \n where each predictor variable was summarized for each city using its spatial boundary. Based on the definition of the predictor variables, we calculated mean values, except population, where we calculated the sum of all pixels that intersect with each city or LAU boundary,\n
\n\n \n Model for predicting emissions and climate change performance\n \n
\n\n Cities participating in the EUCoM are required to submit a Sustainable Energy Action Plan that includes a baseline emissions monitoring inventory, and a monitoring inventory every two years after that. Yet, at the time of data collection in February 2021, out of the nearly 10,000 signatories listed on the website, only 6,114 actors had reported any emissions data, and only 1,400 had reported more than one year of emissions monitoring data. We only included cities\u2019 data with an interpolated population greater than the 5th percentile (374 inhabitants) of the cities\u2019 population distribution. In total, 329 cities had populations below this threshold and were not included in the training or the prediction datasets. Consistent with Hsu et al. (2020), we also filtered out datapoints that reported less than 0.2 tons CO2 per person or greater than 40 tons CO2 per person. The time period for self-reported emissions data ranged from 1990 to 2020, but we only used data greater than 2000 (5,621 unique actors with 7,007 emissions datapoints) for the model training since this is the time period available for the predictor variables.\n
\n\n We further split our data into three subsets: the first subset used as training data includes all EUCoM cities that have at least one year of emissions data reported, whether its baseline emissions or a later inventory-year of data reported (EUCoM, 2021); a second subset are cities participating in the EUCoM but have not reported any emissions data; the third subset are cities not participating in the EUCoM. The first subset of reported emissions data to the EUCoM are used as training data to predict emissions for the latter two subsets of data. We applied the model built with the first dataset to these cities and predict their likely emission of a given year. Supplementary Fig.\u00a03 provides a flow diagram of the processing steps described above. Our training and test datasets were generated based on a standard 80/20 split of the data while preserving the underlying country representation (i.e., slightly over half of the available training data are from cities in Italy (52 percent), followed by Spain (26 percent).\n
\n\n We evaluated several regression models including multilinear regression, random forest, SVM, and extreme gradient boosting (XGBoost). The multilinear model is from the R base library; random forest and SVM are from R package caret version 6.0-86\n \n 53\n \n ; and XGBoost from XGBoost R package version 1.3.2.1.\n \n 54\n \n We chose root mean square error (RMSE) and r\n \n \n 2\n \n \n as the model comparison matrix to examine how each model performs on both the training and test datasets. For random forest, SVM, and XGBoost models that are controlled by a set of hyperparameters, we applied grid search with 5-folder cross validation to the models to get the best parameters that result in the lowest RMSE. Supplementary Table\u00a03 shows the hyperparameters we used in these three models. Missing values in independent variables are a common issue in ML-based models, and the models we evaluated handle missing values in different ways. The XGBoost model is capable of handling missing values without any imputation. Therefore, after we trained an XGboost model with complete data in all independent variables (referred as XGBoost-w/o NA), we also trained the XGBoost model with the data may have NA values in the independent variables (referred as XGBoost-w/-NA in the following sections. Note that all NA values are dropped after we split the data into training and test sets, so that all train and test dataset are exactly the same for models besides XGBoost-w/ NA. Supplementary Table\u00a04 shows the train and test RMSE and r\n \n \n 2\n \n \n of the best tuned models. Both the random forest and XGBoost model are tree-based regression models, and our results suggest that the tree based models perform better than other models for our dataset (Supplementary Table\u00a04). Additionally, the XGBoost-w/ NA model is trained with 357 more data points with NA values in the independent variables and achieved: train RMSE\u2009=\u200924202.05, test RMSE\u2009=\u2009155865.63, train r\n \n 2\n \n =\u20090.99, test r\n \n 2\n \n =\u20090.90.\n
\n\n Based on the model training results and the capability of handling missing values, we decided to proceed with XGBoost. XGBoost stands for \u201cextreme gradient boosting\u201d and has gained popularity due to its high performance in machine-learning competitions such as Kaggle (Nielsen, 2016). Gradient boosting models like XGBoost perform supervised regression tasks through an iterative approach to predict a target variable (i.e., emissions), optimizing predictive performance by combining multiple \u201cweak\u201d trees to fit new models that are more accurate predictors of a response variable.\n \n 55,56\n \n The XGBoost gradient-boosting model has XGBoost has been widely used in air quality monitoring\n \n 57\u201359\n \n and greenhouse gas (GHG) emissions estimation\n \n 60\n \n for its high efficiency, flexibility, and portability. Si and Du (2020)\n \n 56\n \n further note additional advantages of XGBoost, which requires less data preprocessing and has fewer hyperparameters, parameters an ML model uses to control the learning process for tuning.\n \n 61\n \n
\n\n Our implementation of the XGBoost is determined by a set of hyperparameters, which are parameters the machine learning model uses to control the learning process.\n \n 61\n \n These included the maximum depth of the tree, the learning rate, the minimum sum of weight in a node, minimum loss reduction, and the percent of rows to use in each tree which are the standard hyperparameters included in the XGBoost implementation in R.\n \n 62\n \n To obtain the best hyperparameters set for the model and evaluate how the model performs, we first split our dataset into a training dataset and testing dataset with a 80/20 split sampling across countries, meaning we used 80 percent of the data as training data to predict the other 20 percent of the dataset.\n \n 56\n \n We then conducted a grid search (Supplementary Table\u00a02) on the hyperparameters with 5-folder cross-validation to determine the model with the lowest mean root mean squared error. Supplementary Table\u00a02 shows the hyperparameter ranges and the optimized values. Following the hyperparameter grid search, we trained the model with the training dataset with the best result from the hyperparameter grid search. We then tested the model using the test data.\n
\n\n The final model was built with the optimal parameter set from the grid search, which is the process of building models with all the possible parameter combinations and finding the best parameter set with which the model performs the best on training samples. As Supplementary Table\u00a02 describes, the optimum result for the model is achieved when max depth\u2009=\u20095, minimum child weight\u2009=\u20091, eta (learning rate)\u2009=\u20090.1, gamma\u2009=\u20090.5, and trains the model with 999 rounds. The best model performance obtained was RMSE\u2009=\u200926859.36 tons emissions r2\u2009=\u20090.99, MAE\u2009=\u20099173.89. Supplementary Fig.\u00a04 shows scatter plots of the self-reported and predicted emissions for the training and test datasets. We used the XGBoost R package\u2019s built-in function\n \n xgb.importance\n \n to determine the final model\u2019s feature importance (i.e., which predictors have the greatest predictive or explanatory power).\n \n 54,62\n \n
\n\n After building the final model with optimal parameters and evaluation, we applied our model to 1) EUCoM cities that do not report emissions; and 2) all LAUs in Europe that do not participate in the EUCoM. We bootstrapped 1,000 predicted emissions intervals for each year for each actor to ensure robust median estimates. In addition to the optimum parameters from the grid search, we used the \u201csubsample\u201d parameter to introduce randomness into the model. This parameter determines the percent of rows in our dataset to use in each tree. We set this value to 0.90 and, so the model is built with 90% of the total dataset. We then calculated the 5th percentile, 95th percentile, mean, and median value for each predicted emissions estimates for each actor and year.\n
\n\n We calculated several performance metrics (e.g., linear trend in predicted emissions between 2001 and 2018, annual percentage change in emissions, and annualized percentage reduction in per capita emissions) using the predicted emissions data for each actor and evaluated them before utilizing the annualized percentage reduction in per capita emissions (annual per capita emissions trend) as our main evaluation metric, consistent with Hsu et al. (2020),\n \n \n 10\n \n \n as described in Eq.\u00a03.\n
\n\n \n \n \\({reduction}_{c}= -100\\times \\frac{{predemissions}_{\\text{m}\\text{i}\\text{n}\\left(year\\right)}-{predemissions}_{\\text{m}\\text{a}\\text{x}\\left(year\\right)}}{{predemissions}_{\\text{m}\\text{i}\\text{n}\\left(year\\right)}}\\times \\frac{1}{\\text{max}\\left(year\\right)-\\text{m}\\text{i}\\text{n}\\left(year\\right)}\\)\n \n \n (Eq.\u00a03)\n
\n\n Consistent with Hsu et al. (2020), we determined whether a city is \u2018on track\u2019 to achieving their stated emission reduction goal or not, we calculated the ratio of actual (i.e., achieved) per capita emissions reduction in the inventory year to the targeted per capita emissions reduction in the inventory year, both in comparison to the baseline year, assuming that emissions reduction between the baseline year and the target year are pro-rated linearly (i.e., constant emissions reduction from one year to the next). More specifically, we define\n \n \n \\(\\rho\\)\n \n \n through the following Equations (4\u20137):\n
\n\n \n \n \\({Reduction}_{achieved}={Predemissions}_{min\\left(year\\right)}- {Predemissions}_{max\\left(year\\right)}\\)\n \n \n (Eq.\u00a04)\n
\n\n where:\n
\n\n \n \n \\({Predemissions}_{min\\left(year\\right)}\\)\n \n \n is predicted emissions per capita of the city in the minimum year for which predictor data are available. For most cities this was the year 2001;\n
\n\n \n \n \\({Predemissions}_{max\\left(year\\right)}\\)\n \n \n is the predicted emissions per capita of the city in the maxmum year for which predictor data are available. For most cities this was the year 2018;\n
\n\n \n \n \\(Timelapsed=\\left({Year}_{max}-{Year}_{min}\\right)\u00f7\\left({Year}_{target}-{Yea{r}_{min}}_{}\\right)\\)\n \n \n (Eq.\u00a05)\n
\n\n Where:\n
\n\n \n \n \\({Year}_{min}\\)\n \n \n is the minimum year for which predicted emissions data are available\n
\n\n \n \n \\({Year}_{max}\\)\n \n \n is the maximum year for which predicted emissions are available\n
\n\n \n \n \\({Year}_{target}\\)\n \n \n is the year by which committed emissions reductions are to be achieved\n
\n\n \n \n \\({Reduction}_{required}={Predemissions}_{\\text{m}\\text{i}\\text{n}\\left(year\\right)}\\times Target \\times Timelapsed\\)\n \n \n (Eq.\u00a06)\n
\n\n where:\n
\n\n \n \n \\(Target\\)\n \n \n is the committed emissions reduction of the city (percentage).\n
\n\n \n \n \\(\\rho =\\frac{{Reduction}_{achieved}}{{Reduction}_{required}}\\)\n \n \n (Eq.\u00a07)\n
\n\n To investigate whether participation in the EUCoM is associated with a change in a cities\u2019 emissions, we employed an interrupted time series (ITS) modeling approach\n \n \n 18\n \n \n to compare trends in EUCoM cities\u2019 annual per capita emissions prior to and following their adhesion year. ITS designs evaluate an outcome for a population sample exposed to an intervention before and after, using repeated observations at regular intervals.\n \n 63,64\n \n Although there is strong internal validity of an ITS design, there are limitations in terms of potential weak external validity in that the results may not be generalizable to other groups due to the fact that ITS cannot rule out the possibility of unmeasurable or uncontrolled factors leading to a change in the outcome variable.\n
\n\n We estimate annual percent changes in per capita emissions reductions (\n \n \n \\(pct.chg)\\)\n \n \n from 2001 to 2018 for each city (\n \n \n \\(i\\)\n \n \n ) in country (\n \n \n \\(c\\)\n \n \n ) for each year (\n \n \n \\(t\\)\n \n \n ) with the following Eq.\u00a0(8):\n
\n\n \n \n \\({pct.chg}_{i,c,t}= {\\alpha }_{i}+ {\\beta }_{1}Time+{\\beta }_{2}Joined+ {{\\beta }_{3}TSJ+{{\\gamma }}_{C}+ \\text{l}\\text{o}\\text{g}\\left(GDP\\right)}_{i,c,t}+\\text{log}\\left(pop densit{y}_{i,t,c}\\right)+pre{dicted emissions}_{i,t,c,}+ {\u03f5}_{i,c,t}\\)\n \n \n (Eq.\u00a08)\n
\n\n where\n \n \n \\(Time\\)\n \n \n is a variable that indicates the number of years since a city adhered to the EUCoM initiative;\n \n \n \\(Joined\\)\n \n \n is a dummy variable that indicates whether the observation refers to before (0) or after (1) the city adhered;\n \n \n \\(TSJ\\)\n \n \n is the time elapsed since a city joined the EUCoM in years. We also control for differences between cities\u2019 population density, GDP per capita, and emissions per capita predicted by our machine learning model. We also include country dummies (\n \n \n \\({{\\gamma }}_{C})\\)\n \n \n to control for unobserved, time-invariant factors common to cities within a country.\n
\n\n This study is certainly not without its limitations. There are a few areas of uncertainty that could affect the validity of our predictions and results. First, we assume that the self-reported emissions inventories from the EUCoM actors are a valid source of data to train our model and predict others\u2019 emissions. We used the \u201cverified\u201d dataset of self-reported emissions data for 6,200 cities that have reported emissions inventory data evaluated by the European Commission\u2019s Joint Research Centre.\n \n \n 48\n \n \n Although Kona et al. (2021) applied a series of statistical checks to validate these reported emissions inventories, they note several limitations. Since the focus of the EUCoM is on greenhouse gas emissions relate to sectors where a local authority has power to influence through sectoral and policy measures, participating cities only report emissions from selected sources (e.g., energy consumption for buildings, transport and local energy generation, industrial sources not already covered by the EU Emissions Trading Scheme, and waste/wastewater) (EUCoM, 2016).\n \n 65\n \n Kona et al. (2021)\n \n \n 48\n \n \n acknowledge that the EUCoM inventories were \u201cnever meant to be a method to create exhaustive inventories of all emission sources in the territory or to deal with emissions already included in national-scale control initiatives, such as the EU Emissions Trading System (ETS) mechanisms.\u201d Therefore a second limitation is that there are emissions sources and sectors inherently missing from EUCoM cities\u2019 inventories, including supply chain or consumption-based emissions sources. Third, the use of different emissions factors, estimation methodologies, and reporting boundaries may add uncertainty. Fourth, we assume that the spatial boundaries of EUCoM cities and LAUs remained static over the time period, while these may have changed over time. Last, while we observed significant differences in EUCoM cities\u2019 emissions reduction trends compared to cities that do not participate, there may be some fundamental differences between these groups of cities. To evaluate our model\u2019s sensitivity to this potential factor, we included a dummy variable to designate whether an EUCoM city has committed to an ambitious emissions reduction target (greater than 20 percent) or if it has simply adopted the minimum EU target of 20 percent, which other non-EUCoM cities are presumably subject as part of national and regional climate targets. As shown in Supplementary Fig.\u00a05, we found our model was able to predict emissions from both sets of actors with similar accuracy (r\n \n 2\n \n =\u20090.94 for both groups). As Fig.\n \n 1\n \n b illustrates, the dummy variable \u2018ambitious 2020 target\u2019 did not contribute to the model\u2019s predictive gain values. Ideally we would be able to include self-reported emissions data from non-EUCoM cities in our training dataset, but these data are not available.\n
\n\n Data scraping and geospatial data processing were conducted using python (version 3.68) and the R statistical programming environment (version 3.6.2). The machine learning model was developed and conducted in R using the XGBoost package.\n \n 54\n \n Figures were made using ggplot2\n \n 66\n \n data visualization package and maps were made in QGIS (version 3.16)\n
\n\n 1.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Hsu, A.\n \n et al.\n \n ClimActor, harmonized transnational data on climate network participation by city and regional governments.\n \n Sci. Data\n \n (2020) doi:10.1038/s41597-020-00682-0.\n
\n\n 2.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Reckien, D.\n \n et al.\n \n How are cities planning to respond to climate change? Assessment of local climate plans from 885 cities in the EU-28.\n \n J. Clean. Prod.\n \n (2018) doi:10.1016/j.jclepro.2018.03.220.\n
\n\n 3.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Hsu, A. & Rauber, R. Diverse climate actors show limited coordination in a large-scale text analysis of strategy documents.\n \n Commun. Earth Environ. 2021 21\n \n \n 2\n \n , 1\u201312 (2021).\n
\n\n 4.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Hsu, A., Cheng, Y., Weinfurter, A., Xu, K. & Yick, C. Track climate pledges of cities and companies.\n \n Nature\n \n \n 532\n \n , 303\u2013306 (2016).\n
\n\n 5.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Milojevic-Dupont, N. & Creutzig, F. Machine learning for geographically differentiated climate change mitigation in urban areas.\n \n Sustain. Cities Soc.\n \n \n 64\n \n , 102526 (2021).\n
\n\n 6.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Intergovernmental Panel on Climate Change & Intergovernmental Panel on Climate Change. Human Settlements, Infrastructure, and Spatial Planning. in\n \n Climate Change 2014 Mitigation of Climate Change\n \n (2015). doi:10.1017/cbo9781107415416.018.\n
\n\n 7.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Stehle, F., Hickmann, T., Lederer, M. & H\u00f6hne, C. Urban Climate Politics in Emerging Economies: A Multi-Level Governance Perspective:\n \n https://doi-org.libproxy.lib.unc.edu/10.1177/2455747120913185\n \n 245574712091318 (2020) doi:10.1177/2455747120913185.\n
\n\n 8.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Boehnke, R., Hoppe, T., Brezet, H., production, K. B.-J. of cleaner & 2019, \u00a0undefined. Good practices in local climate mitigation action by small and medium-sized cities; exploring meaning, implementation and linkage to actual lowering of.\n \n Elsevier\n \n .\n
\n\n 9.\u00a0 \u00a0 \u00a0 \u00a0\u00a0Domorenok, E., Acconcia, G., Bendlin, L. & Campillo, X. R. Experiments in EU Climate Governance: The Unfulfilled Potential of the Covenant of Mayors.\n \n Glob. Environ. Polit.\n \n \n 20\n \n , 122\u2013142 (2020).\n
\n\n 10.\u00a0 \u00a0 \u00a0\u00a0Hsu, A.\n \n et al.\n \n Performance determinants show European cities are delivering on climate mitigation.\n \n Nat. Clim. Chang.\n \n (2020) doi:10.1038/s41558-020-0879-9.\n
\n\n 11.\u00a0 \u00a0 \u00a0\u00a0Ibrahim, N., Sugar, L., Hoornweg, D. & Kennedy, C. Greenhouse gas emissions from cities: Comparison of international inventory frameworks.\n \n Local Environ.\n \n (2012) doi:10.1080/13549839.2012.660909.\n
\n\n 12.\u00a0 \u00a0 \u00a0\u00a0Rolnick, D.\n \n et al.\n \n Tackling Climate Change with Machine Learning. (2019).\n
\n\n 13.\u00a0 \u00a0 \u00a0\u00a0Oda, T.\n \n et al.\n \n Errors and uncertainties in a gridded carbon dioxide emissions inventory.\n \n Mitig. Adapt. Strateg. Glob. Chang.\n \n \n 24\n \n , 1007\u20131050 (2019).\n
\n\n 14.\u00a0 \u00a0 \u00a0\u00a0Raupach, M. R., Rayner, P. J. & Paget, M. Regional variations in spatial structure of nightlights, population density and fossil-fuel CO2 emissions.\n \n Energy Policy\n \n \n 38\n \n , 4756\u20134764 (2010).\n
\n\n 15.\u00a0 \u00a0 \u00a0\u00a0Moran, D.\n \n et al.\n \n Estimating CO&lt;sub&gt;2&lt;/sub&gt; emissions for 108\u2009000 European cities.\n \n Earth Syst. Sci. Data\n \n \n 14\n \n , 845\u2013864 (2022).\n
\n\n 16.\u00a0 \u00a0 \u00a0\u00a0Nangini, C.\n \n et al.\n \n A global dataset of CO2 emissions and ancillary data related to emissions for 343 cities.\n \n nature.com\n \n .\n
\n\n 17.\u00a0 \u00a0 \u00a0\u00a0World Resources Institute (WRI). Climate Watch 2020. (2020).\n
\n\n 18.\u00a0 \u00a0 \u00a0\u00a0Shadish, W. R., Cook, T. D. & Campbell, D. T.\n \n Experimental and quasi-experimental designs for generalized causal inference\n \n . (2002).\n
\n\n 19.\u00a0 \u00a0 \u00a0\u00a0van der Heijden, J. Studying urban climate governance: Where to begin, what to look for, and how to make a meaningful contribution to scholarship and practice.\n \n Earth Syst. Gov.\n \n \n 1\n \n , 100005 (2019).\n
\n\n 20.\u00a0 \u00a0 \u00a0\u00a0Gordon, D., sustainability, C. J.-C. opinion in environmental & 2018, \u00a0undefined. City-networks, global climate governance, and the road to 1.5 C.\n \n Elsevier\n \n .\n
\n\n 21.\u00a0 \u00a0 \u00a0\u00a0Armstrong, J. H. Modeling effective local government climate policies that exceed state targets.\n \n Energy Policy\n \n \n 132\n \n , 15\u201326 (2019).\n
\n\n 22.\u00a0 \u00a0 \u00a0\u00a0Kousky, C., policy, S. S.-C. & 2003, \u00a0 undefined. Global climate policy: will cities lead the way?\n \n Taylor Fr.\n \n \n 3\n \n , 359\u2013372 (2003).\n
\n\n 23.\u00a0 \u00a0 \u00a0\u00a0Broto, V. C. Energy landscapes and urban trajectories towards sustainability.\n \n Energy Policy\n \n \n 108\n \n , 755\u2013764 (2017).\n
\n\n 24.\u00a0 \u00a0 \u00a0\u00a0Seto, K. C.\n \n et al.\n \n From Low- To Net-Zero Carbon Cities- To Next Global Agenda.\n \n Annu. Rev. Environ. Resour.\n \n \n 46\n \n , 377\u2013415 (2021).\n
\n\n 25.\u00a0 \u00a0 \u00a0\u00a0Aldy, J. E. The crucial role of policy surveillance in international climate policy.\n \n Clim. Change\n \n \n 126\n \n , 279\u2013292 (2014).\n
\n\n 26.\u00a0 \u00a0 \u00a0\u00a0Hale, T. N.\n \n et al.\n \n Sub- and non-state climate action: a framework to assess progress, implementation and impact.\n \n Clim. Policy\n \n \n 21\n \n , 406\u2013420 (2021).\n
\n\n 27.\u00a0 \u00a0 \u00a0\u00a0Bulkeley, H. Can cities realise their climate potential? Reflections on COP21 Paris and beyond.\n \n Local Environ.\n \n \n 20\n \n , 1405\u20131409 (2015).\n
\n\n 28.\u00a0 \u00a0 \u00a0\u00a0Bulkeley, H.\n \n et al.\n \n Governing climate change transnationally: Assessing the evidence from a database of sixty initiatives.\n \n Environ. Plan. C Gov. Policy\n \n \n 30\n \n , 591\u2013612 (2012).\n
\n\n 29.\u00a0 \u00a0 \u00a0\u00a0Chan, S., Falkner, R., Goldberg, M. & van Asselt, H. Effective and geographically balanced? An output-based assessment of non-state climate actions.\n \n Clim. Policy\n \n \n 18\n \n , 24\u201335 (2018).\n
\n\n 30.\u00a0 \u00a0 \u00a0\u00a0Kuramochi, T.\n \n et al.\n \n Beyond national climate action: the impact of region, city, and business commitments on global greenhouse gas emissions.\n \n Clim. Policy\n \n (2020) doi:10.1080/14693062.2020.1740150.\n
\n\n 31.\u00a0 \u00a0 \u00a0\u00a0Pattberg, P., Biermann, F., Chan, C. & Mert, A. Public-private partnerships for sustainable development: Emergence, influence and legitimacy.\n \n Public-Private Partnerships Sustain. Dev. Emergence, Influ. Legitimacy\n \n (2012) doi:10.4337/9781849809313.\n
\n\n 32.\u00a0 \u00a0 \u00a0\u00a0Ogle, S. M.\n \n et al.\n \n Advancing national greenhouse gas inventories for agriculture in developing countries: Improving activity data, emission factors and software technology.\n \n Environ. Res. Lett.\n \n \n 8\n \n , (2013).\n
\n\n 33.\u00a0 \u00a0 \u00a0\u00a0Pan, G., Xu, Y. & Ma, J. The potential of CO2 satellite monitoring for climate governance: A review.\n \n J. Environ. Manage.\n \n \n 277\n \n , 111423 (2021).\n
\n\n 34.\u00a0 \u00a0 \u00a0\u00a0Boehnke, R. F., Hoppe, T., Brezet, H. & Blok, K. Good practices in local climate mitigation action by small and medium-sized cities; exploring meaning, implementation and linkage to actual lowering of carbon emissions in thirteen municipalities in The Netherlands.\n \n J. Clean. Prod.\n \n \n 207\n \n , 630\u2013644 (2019).\n
\n\n 35.\u00a0 \u00a0 \u00a0\u00a0United Nations Framework Convention on Climate Change (UNFCCC).\n \n Adoption of the Paris Agreement, 21st Conference of the Parties\n \n . https://unfccc.int/sites/default/files/english_paris_agreement.pdf (2015).\n
\n\n 36.\u00a0 \u00a0 \u00a0\u00a0REVISED NON-PAPER FOR INFORMAL CONSULTATIONS 1 PREPARING FOR THE FIRST GLOBAL STOCKTAKE REVISED NON-PAPER BY THE CHAIRS OF THE SBSTA AND SBI.\n
\n\n 37.\u00a0 \u00a0 \u00a0\u00a0No, Cities Are Not Actually Leading on Climate. Enough With the Mindless Cheerleading | Greentech Media. https://www.greentechmedia.com/articles/read/hard-truths-about-city-failures-with-clean-energy.\n
\n\n 38.\u00a0 \u00a0 \u00a0\u00a0politics, T. H.-G. environmental & 2016, \u00a0 undefined. \u201cAll hands on deck\u201d: The Paris agreement and nonstate climate action.\n \n direct.mit.edu\n \n (2016) doi:10.1162/GLEP_a_00362.\n
\n\n 39.\u00a0 \u00a0 \u00a0\u00a0Hale, T. Catalytic cooperation.\n \n Glob. Environ. Polit.\n \n \n 20\n \n , 73\u201398 (2020).\n
\n\n 40.\u00a0 \u00a0 \u00a0\u00a0Bosilovich, M. G., R. Lucchesi, \u00a0and M. S.\n \n MERRA-2: File Specification\n \n . (2015).\n
\n\n 41.\u00a0 \u00a0 \u00a0\u00a0Spinoni, J.\n \n et al.\n \n Changes of heating and cooling degree-days in Europe from 1981 to 2100.\n \n Int. J. Climatol.\n \n \n 38\n \n , e191\u2013e208 (2018).\n
\n\n 42.\u00a0 \u00a0 \u00a0\u00a0Van Donkelaar, A., Martin, R. V., Li, C. & Burnett, R. T. Regional Estimates of Chemical Composition of Fine Particulate Matter Using a Combined Geoscience-Statistical Method with Information from Satellites, Models, and Monitors.\n \n Environ. Sci. Technol.\n \n \n 53\n \n , 2595\u20132611 (2019).\n
\n\n 43.\u00a0 \u00a0 \u00a0\u00a0Engel-Cox, J., Kim Oanh, N. T., van Donkelaar, A., Martin, R. V. & Zell, E. Toward the next generation of air quality monitoring: Particulate Matter.\n \n Atmos. Environ.\n \n \n 80\n \n , 584\u2013590 (2013).\n
\n\n 44.\u00a0 \u00a0 \u00a0\u00a0Agency, E. E. Greenhouse gas emissions by aggregated sector.\n \n Eurostat\n \n (2019).\n
\n\n 45.\u00a0 \u00a0 \u00a0\u00a0CIESIN. Gridded Population of the World, Version 4. http://dx.doi.org/10.7927/H49C6VBN (2016).\n
\n\n 46.\u00a0 \u00a0 \u00a0\u00a0Kummu, M., Taka, M. & Guillaume, J. H. A. Gridded global datasets for Gross Domestic Product and Human Development Index over 1990-2015.\n \n Sci. Data\n \n \n 5\n \n , (2018).\n
\n\n 47.\u00a0 \u00a0 \u00a0\u00a0Moritz, S., J., T. B.-B.-R. & 2017, \u00a0 undefined. imputeTS: time series missing value imputation in R.\n \n cran.microsoft.com\n \n .\n
\n\n 48.\u00a0 \u00a0 \u00a0\u00a0Kona, A.\n \n et al.\n \n Global Covenant of Mayors, a dataset of greenhouse gas emissions for 6200 cities in Europe and the Southern Mediterranean countries.\n \n Earth Syst. Sci. Data\n \n \n 13\n \n , 3551\u20133564 (2021).\n
\n\n 49.\u00a0 \u00a0 \u00a0\u00a0Glossary:Local administrative unit (LAU) - Statistics Explained. https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Glossary:Local_administrative_unit_(LAU).\n
\n\n 50.\u00a0 \u00a0 \u00a0\u00a0Jordahl, K. geopandas: Python tools for geographic data. (2014).\n
\n\n 51.\u00a0 \u00a0 \u00a0\u00a0Haklay, M., computing, P. W.-I. P. & 2008, \u00a0 undefined. Openstreetmap: User-generated street maps.\n \n ieeexplore.ieee.org\n \n (2008).\n
\n\n 52.\u00a0 \u00a0 \u00a0\u00a0Perry, M. rasterstats 0.16.0. (2021).\n
\n\n 53.\u00a0 \u00a0 \u00a0\u00a0Kuhn, M. Building Predictive Models in R Using the caret Package.\n \n J. Stat. Softw.\n \n \n 28\n \n , 1\u201326 (2008).\n
\n\n 54.\u00a0 \u00a0 \u00a0\u00a0Chen, T. & Guestrin, C. XGBoost: A Scalable Tree Boosting System.\n \n Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discov. Data Min.\n \n doi:10.1145/2939672.\n
\n\n 55.\u00a0 \u00a0 \u00a0\u00a0Seyedzadeh, S., Pour Rahimian, F., Oliver, S., Rodriguez, S. & Glesk, I. Machine learning modelling for predicting non-domestic buildings energy performance: A model to support deep energy retrofit decision-making.\n \n Appl. Energy\n \n \n 279\n \n , 115908 (2020).\n
\n\n 56.\u00a0 \u00a0 \u00a0\u00a0Si, M. & Du, K. Development of a predictive emissions model using a gradient boosting machine learning method.\n \n Environ. Technol. Innov.\n \n \n 20\n \n , (2020).\n
\n\n 57.\u00a0 \u00a0 \u00a0\u00a0Joharestani, M. Z., Cao, C., Ni, X., Bashir, B. & Talebiesfandarani, S. PM2.5 Prediction Based on Random Forest, XGBoost, and Deep Learning Using Multisource Remote Sensing Data.\n \n Atmos. 2019, Vol. 10, Page 373\n \n \n 10\n \n , 373 (2019).\n
\n\n 58.\u00a0 \u00a0 \u00a0\u00a0Pan, B. Application of XGBoost algorithm in hourly PM2.5 concentration prediction.\n \n IOP Conf. Ser. Earth Environ. Sci.\n \n \n 113\n \n , 012127 (2018).\n
\n\n 59.\u00a0 \u00a0 \u00a0\u00a0Si, M. & Du, K. Development of a predictive emissions model using a gradient boosting machine learning method.\n \n Environ. Technol. Innov.\n \n \n 20\n \n , 101028 (2020).\n
\n\n 60.\u00a0 \u00a0 \u00a0\u00a0Li, Y. & Sun, Y. Modeling and predicting city-level CO 2 emissions using open access data and machine learning.\n \n Environ. Sci. Pollut. Res. 2021 2815\n \n \n 28\n \n , 19260\u201319271 (2021).\n
\n\n 61.\u00a0 \u00a0 \u00a0\u00a0Yang, L. & Shami, A. On hyperparameter optimization of machine learning algorithms: Theory and practice.\n \n Neurocomputing\n \n \n 415\n \n , 295\u2013316 (2020).\n
\n\n 62.\u00a0 \u00a0 \u00a0\u00a0Chen, T.\n \n et al.\n \n xgboost: Extreme Gradient Boosting. (2021) doi:10.1145/2939672.2939785.\n
\n\n 63.\u00a0 \u00a0 \u00a0\u00a0Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted time series studies of public health interventions.\n \n Int. J. Epidemiol.\n \n \n 47\n \n , 2082\u20132093 (2018).\n
\n\n 64.\u00a0 \u00a0 \u00a0\u00a0Kleck, G. & Patterson, E. B. The impact of gun control and gun ownership levels on violence rates.\n \n J. Quant. Criminol.\n \n \n 9\n \n , 249\u2013287 (1993).\n
\n\n 65.\u00a0 \u00a0 \u00a0\u00a0(EUCoM), E. C. of M.\n \n The Covenant of Mayors for Climate and Energy Reporting Guidelines\n \n . https://www.covenantofmayors.eu/IMG/pdf/Covenant_ReportingGuidelines.pdf (2016).\n
\n\n 66.\u00a0 \u00a0 \u00a0\u00a0Wickham, H.\n \n ggplot2 Elegant Graphics for Data Analysis\n \n .\n \n Journal of the Royal Statistical Society: Series A (Statistics in Society)\n \n (2016). doi:10.1007/978-3-319-24277-4.\n
\n\n
\n| \n \n Statistic\n \n | \n \n \n N\n \n | \n \n \n Mean\n \n | \n \n \n St. Dev.\n \n | \n \n \n Min\n \n | \n \n \n Pctl(25)\n \n | \n \n \n Pctl(75)\n \n | \n \n \n Max\n \n | \n
|---|---|---|---|---|---|---|---|
| \n \n (1) EUCoM Cities reporting emissions data\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n GDP per capita\n \n | \n \n \n 5,472\n \n | \n \n \n 34,270.330\n \n | \n \n \n 10,500.000\n \n | \n \n \n 2,898.161\n \n | \n \n \n 24,907.560\n \n | \n \n \n 42,653.010\n \n | \n \n \n 65,779.700\n \n | \n
| \n \n Heating degree days\n \n | \n \n \n 5,479\n \n | \n \n \n 2,346.058\n \n | \n \n \n 1,506.650\n \n | \n \n \n 0.000\n \n | \n \n \n 1,012.622\n \n | \n \n \n 3,351.190\n \n | \n \n \n 8,894.994\n \n | \n
| \n \n Population density\n \n | \n \n \n 5,479\n \n | \n \n \n 577.841\n \n | \n \n \n 1,416.094\n \n | \n \n \n 0.238\n \n | \n \n \n 46.062\n \n | \n \n \n 512.802\n \n | \n \n \n 22,114.780\n \n | \n
| \n \n Population\n \n | \n \n \n 5,479\n \n | \n \n \n 33,775.060\n \n | \n \n \n 199,844.500\n \n | \n \n \n 35.308\n \n | \n \n \n 1,683.855\n \n | \n \n \n 15,272.490\n \n | \n \n \n 8,965,276.000\n \n | \n
| \n \n Fossil-fuel CO2 emissions\n \n | \n \n \n 5,479\n \n | \n \n \n 41,870.170\n \n | \n \n \n 222,138.600\n \n | \n \n \n 0.000\n \n | \n \n \n 2,345.647\n \n | \n \n \n 17,175.120\n \n | \n \n \n 5,957,429.000\n \n | \n
| \n \n Fossil-fuel CO2 emissions per capita\n \n | \n \n \n 5,479\n \n | \n \n \n 2.229\n \n | \n \n \n 13.378\n \n | \n \n \n 0.000\n \n | \n \n \n 0.804\n \n | \n \n \n 1.967\n \n | \n \n \n 775.742\n \n | \n
| \n \n Fine particulate air pollution (PM2.5)\n \n | \n \n \n 5,423\n \n | \n \n \n 11.164\n \n | \n \n \n 4.877\n \n | \n \n \n 2.151\n \n | \n \n \n 7.065\n \n | \n \n \n 14.410\n \n | \n \n \n 29.181\n \n | \n
| \n \n (2) EUCoM Cities not reporting emissions data\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n GDP per capita\n \n | \n \n \n 1,685\n \n | \n \n \n 31,615.630\n \n | \n \n \n 10,489.700\n \n | \n \n \n 2,575.388\n \n | \n \n \n 23,853.250\n \n | \n \n \n 40,499.790\n \n | \n \n \n 84,746.950\n \n | \n
| \n \n Heating degree days\n \n | \n \n \n 1,685\n \n | \n \n \n 2,589.232\n \n | \n \n \n 1,630.994\n \n | \n \n \n 0.000\n \n | \n \n \n 1,142.226\n \n | \n \n \n 3,557.862\n \n | \n \n \n 7,956.604\n \n | \n
| \n \n Population density\n \n | \n \n \n 1,685\n \n | \n \n \n 512.940\n \n | \n \n \n 1,707.357\n \n | \n \n \n 2.344\n \n | \n \n \n 45.674\n \n | \n \n \n 352.777\n \n | \n \n \n 45,852.780\n \n | \n
| \n \n Population\n \n | \n \n \n 1,685\n \n | \n \n \n 34,692.720\n \n | \n \n \n 161,566.900\n \n | \n \n \n 60.567\n \n | \n \n \n 1,481.973\n \n | \n \n \n 12,578.350\n \n | \n \n \n 2,672,199.000\n \n | \n
| \n \n Fossil-fuel CO2 emissions\n \n | \n \n \n 1,685\n \n | \n \n \n 48,690.320\n \n | \n \n \n 237,230.000\n \n | \n \n \n 0.000\n \n | \n \n \n 2,361.174\n \n | \n \n \n 16,103.160\n \n | \n \n \n 4,420,102.000\n \n | \n
| \n \n Fossil-fuel CO2 emissions per capita\n \n | \n \n \n 1,685\n \n | \n \n \n 2.037\n \n | \n \n \n 4.678\n \n | \n \n \n 0.000\n \n | \n \n \n 0.901\n \n | \n \n \n 2.207\n \n | \n \n \n 120.650\n \n | \n
| \n \n Fine particulate air pollution (PM2.5)\n \n | \n \n \n 1,641\n \n | \n \n \n 10.508\n \n | \n \n \n 4.053\n \n | \n \n \n 2.924\n \n | \n \n \n 7.318\n \n | \n \n \n 13.390\n \n | \n \n \n 30.334\n \n | \n
| \n \n (3) All other LAUS\n \n | \n \n | \n\n | \n\n | \n\n | \n\n | \n\n | \n\n | \n
| \n \n GDP per capita\n \n | \n \n \n 39,742\n \n | \n \n \n 33,670.690\n \n | \n \n \n 12,293.380\n \n | \n \n \n 4,552.309\n \n | \n \n \n 25,150.820\n \n | \n \n \n 40,107.690\n \n | \n \n \n 100,726.400\n \n | \n
| \n \n Heating degree days\n \n | \n \n \n 39,817\n \n | \n \n \n 3,797.706\n \n | \n \n \n 1,467.800\n \n | \n \n \n 0.000\n \n | \n \n \n 3,108.901\n \n | \n \n \n 4,698.598\n \n | \n \n \n 9,271.064\n \n | \n
| \n \n Population density\n \n | \n \n \n 39,817\n \n | \n \n \n 335.148\n \n | \n \n \n 939.363\n \n | \n \n \n 0.329\n \n | \n \n \n 57.365\n \n | \n \n \n 262.214\n \n | \n \n \n 29,391.080\n \n | \n
| \n \n Population\n \n | \n \n \n 39,817\n \n | \n \n \n 8,348.308\n \n | \n \n \n 22,851.110\n \n | \n \n \n 1,000.194\n \n | \n \n \n 1,663.810\n \n | \n \n \n 6,567.054\n \n | \n \n \n 866,314.800\n \n | \n
| \n \n Fossil-fuel CO2 emissions\n \n | \n \n \n 39,817\n \n | \n \n \n 12,973.500\n \n | \n \n \n 70,225.040\n \n | \n \n \n 202.095\n \n | \n \n \n 2,338.124\n \n | \n \n \n 9,386.230\n \n | \n \n \n 4,706,230.000\n \n | \n
| \n \n Fossil-fuel CO2 emissions per capita\n \n | \n \n \n 39,817\n \n | \n \n \n 1.637\n \n | \n \n \n 1.507\n \n | \n \n \n 0.200\n \n | \n \n \n 0.907\n \n | \n \n \n 1.953\n \n | \n \n \n 39.518\n \n | \n
| \n \n Fine particulate air pollution (PM2.5)\n \n | \n \n \n 39,817\n \n | \n \n \n 10.815\n \n | \n \n \n 4.303\n \n | \n \n \n 1.915\n \n | \n \n \n 7.465\n \n | \n \n \n 13.245\n \n | \n \n \n 35.632\n \n | \n
\n
\n\n
\n| \n | \n\n \n Mean\u2009\u00b1\u2009sd trend (1)\n \n | \n \n \n Mean\u2009\u00b1\u2009sd trend (2)\n \n | \n \n \n Mean difference\n \n | \n \n \n Standard error\n \n | \n
|---|---|---|---|---|
| \n \n (1) EUCoM cities vs. (2) All LAUs\n \n | \n \n \n -1.22\u2009\u00b1\u20092.00\n \n | \n \n \n 5.21\u2009\u00b1\u200911.03\n \n | \n \n \n 6.43***\n \n | \n \n \n 0.0004\n \n | \n
| \n \n (1) EUCoM cities reporting inventories vs. (2) EUCoM cities not reporting inventories\n \n | \n \n \n -1.6\u2009\u00b1\u20092.0\n \n | \n \n \n -0.08\u2009\u00b1\u20091.5\n \n | \n \n \n 1.52***\n \n | \n \n \n 0.002\n \n | \n
| \n \n (1) Ambitious EUCoM cities versus (2) unambitious EUCoM cities\n \n | \n \n \n -1.53\u2009\u00b1\u20092.7\n \n | \n \n \n -0.47\u2009\u00b1\u20092.4\n \n | \n \n \n 1.06****\n \n | \n \n \n 0.001\n \n | \n
\n
\n\n \n Note\n \n
\n\n \n *\n \n p\u2009<\u20090.1;\n \n **\n \n p\u2009<\u20090.05;\n \n ***\n \n p\u2009<\u20090.01\n
\n\n
\n\n
\n| \n \n country\n \n | \n \n \n Number of EUCoM cities evaluated\n \n | \n \n \n population\n \n | \n \n \n Share of national emissions (%)\n \n | \n \n \n Share of national population (%)\n \n | \n \n \n On track (%)\n \n | \n \n \n Reporting emissions (%)\n \n | \n \n \n Percentage of cities reducing emissions (%)\n \n | \n
|---|---|---|---|---|---|---|---|
| \n \n Austria\n \n | \n \n \n 25\n \n | \n \n \n 78952\u2009\u00b1\u2009355736\n \n | \n \n \n 10\n \n | \n \n \n 22\n \n | \n \n \n 8\n \n | \n \n \n 36\n \n | \n \n \n 32\n \n | \n
| \n \n Belarus\n \n | \n \n \n 10\n \n | \n \n \n 153961\u2009\u00b1\u2009131148\n \n | \n \n \n 7\n \n | \n \n \n 16\n \n | \n \n \n 0\n \n | \n \n \n 30\n \n | \n \n \n 30\n \n | \n
| \n \n Belgium\n \n | \n \n \n 454\n \n | \n \n \n 20615\u2009\u00b1\u200932880\n \n | \n \n \n 49\n \n | \n \n \n 81\n \n | \n \n \n 37\n \n | \n \n \n 65\n \n | \n \n \n 92\n \n | \n
| \n \n Bulgaria\n \n | \n \n \n 25\n \n | \n \n \n 102572\u2009\u00b1\u2009251793\n \n | \n \n \n 35\n \n | \n \n \n 37\n \n | \n \n \n 20\n \n | \n \n \n 88\n \n | \n \n \n 44\n \n | \n
| \n \n Croatia\n \n | \n \n \n 62\n \n | \n \n \n 17771\u2009\u00b1\u200922203\n \n | \n \n \n 29\n \n | \n \n \n 27\n \n | \n \n \n 13\n \n | \n \n \n 89\n \n | \n \n \n 56\n \n | \n
| \n \n Cyprus\n \n | \n \n \n 21\n \n | \n \n \n 23011\u2009\u00b1\u200927514\n \n | \n \n \n 32\n \n | \n \n \n 38\n \n | \n \n \n 52\n \n | \n \n \n 95\n \n | \n \n \n 81\n \n | \n
| \n \n Czech Republic\n \n | \n \n \n 16\n \n | \n \n \n 144544\u2009\u00b1\u2009331087\n \n | \n \n \n 7.6\n \n | \n \n \n 22\n \n | \n \n \n 19\n \n | \n \n \n 25\n \n | \n \n \n 75\n \n | \n
| \n \n Denmark\n \n | \n \n \n 38\n \n | \n \n \n 81207\u2009\u00b1\u2009106028\n \n | \n \n \n 39\n \n | \n \n \n 52\n \n | \n \n \n 87\n \n | \n \n \n 89\n \n | \n \n \n 92\n \n | \n
| \n \n Estonia\n \n | \n \n \n 5\n \n | \n \n \n 100626\u2009\u00b1\u2009159730\n \n | \n \n \n 13\n \n | \n \n \n 38\n \n | \n \n \n 40\n \n | \n \n \n 100\n \n | \n \n \n 40\n \n | \n
| \n \n Finland\n \n | \n \n \n 12\n \n | \n \n \n 185181\u2009\u00b1\u2009154967\n \n | \n \n \n 18\n \n | \n \n \n 40\n \n | \n \n \n 67\n \n | \n \n \n 67\n \n | \n \n \n 92\n \n | \n
| \n \n France\n \n | \n \n \n 107\n \n | \n \n \n 113776\u2009\u00b1\u2009275561\n \n | \n \n \n 16\n \n | \n \n \n 18\n \n | \n \n \n 40\n \n | \n \n \n 68\n \n | \n \n \n 75\n \n | \n
| \n \n Germany\n \n | \n \n \n 75\n \n | \n \n \n 236089\u2009\u00b1\u2009462209\n \n | \n \n \n 15\n \n | \n \n \n 21\n \n | \n \n \n 43\n \n | \n \n \n 57\n \n | \n \n \n 79\n \n | \n
| \n \n Greece\n \n | \n \n \n 150\n \n | \n \n \n 39904\u2009\u00b1\u2009110196\n \n | \n \n \n 40\n \n | \n \n \n 55\n \n | \n \n \n 24\n \n | \n \n \n 83\n \n | \n \n \n 54\n \n | \n
| \n \n Hungary\n \n | \n \n \n 146\n \n | \n \n \n 40601\u2009\u00b1\u2009211103\n \n | \n \n \n 34\n \n | \n \n \n 58\n \n | \n \n \n 8\n \n | \n \n \n 34\n \n | \n \n \n 33\n \n | \n
| \n \n Ireland\n \n | \n \n \n 13\n \n | \n \n \n 135365\u2009\u00b1\u2009184416\n \n | \n \n \n 16\n \n | \n \n \n 30\n \n | \n \n \n 54\n \n | \n \n \n 62\n \n | \n \n \n 69\n \n | \n
| \n \n Italy\n \n | \n \n \n 3854\n \n | \n \n \n 13702\u2009\u00b1\u200982716\n \n | \n \n \n 60\n \n | \n \n \n 87\n \n | \n \n \n 56\n \n | \n \n \n 77\n \n | \n \n \n 88\n \n | \n
| \n \n Latvia\n \n | \n \n \n 20\n \n | \n \n \n 65246\u2009\u00b1\u2009161737\n \n | \n \n \n 50\n \n | \n \n \n 65\n \n | \n \n \n 40\n \n | \n \n \n 95\n \n | \n \n \n 55\n \n | \n
| \n \n Lithuania\n \n | \n \n \n 15\n \n | \n \n \n 71618\u2009\u00b1\u2009102545\n \n | \n \n \n 24\n \n | \n \n \n 39\n \n | \n \n \n 27\n \n | \n \n \n 80\n \n | \n \n \n 33\n \n | \n
| \n \n Luxembourg\n \n | \n \n \n 9\n \n | \n \n \n 2943\u2009\u00b1\u20091427\n \n | \n \n \n 2.1\n \n | \n \n \n 4.3\n \n | \n \n \n 44\n \n | \n \n \n 11\n \n | \n \n \n 89\n \n | \n
| \n \n Malta\n \n | \n \n \n 23\n \n | \n \n \n 5710\u2009\u00b1\u20094610\n \n | \n \n \n 39\n \n | \n \n \n 25\n \n | \n \n \n 57\n \n | \n \n \n 83\n \n | \n \n \n 87\n \n | \n
| \n \n Netherlands\n \n | \n \n \n 29\n \n | \n \n \n 169151\u2009\u00b1\u2009184105\n \n | \n \n \n 19\n \n | \n \n \n 28\n \n | \n \n \n 31\n \n | \n \n \n 52\n \n | \n \n \n 79\n \n | \n
| \n \n Norway\n \n | \n \n \n 6\n \n | \n \n \n 190069\u2009\u00b1\u2009231656\n \n | \n \n \n 16\n \n | \n \n \n 21\n \n | \n \n \n 67\n \n | \n \n \n 33\n \n | \n \n \n 100\n \n | \n
| \n \n Poland\n \n | \n \n \n 41\n \n | \n \n \n 147621\u2009\u00b1\u2009309957\n \n | \n \n \n 8.3\n \n | \n \n \n 16\n \n | \n \n \n 17\n \n | \n \n \n 80\n \n | \n \n \n 68\n \n | \n
| \n \n Portugal\n \n | \n \n \n 136\n \n | \n \n \n 25919\u2009\u00b1\u200960168\n \n | \n \n \n 35\n \n | \n \n \n 34\n \n | \n \n \n 46\n \n | \n \n \n 78\n \n | \n \n \n 79\n \n | \n
| \n \n Romania\n \n | \n \n \n 97\n \n | \n \n \n 163639\u2009\u00b1\u2009447539\n \n | \n \n \n 44\n \n | \n \n \n 79\n \n | \n \n \n 25\n \n | \n \n \n 64\n \n | \n \n \n 55\n \n | \n
| \n \n Slovakia\n \n | \n \n \n 29\n \n | \n \n \n 20169\u2009\u00b1\u200949424\n \n | \n \n \n 8\n \n | \n \n \n 11\n \n | \n \n \n 7\n \n | \n \n \n 10\n \n | \n \n \n 69\n \n | \n
| \n \n Slovenia\n \n | \n \n \n 28\n \n | \n \n \n 27513\u2009\u00b1\u200959356\n \n | \n \n \n 23\n \n | \n \n \n 37\n \n | \n \n \n 29\n \n | \n \n \n 96\n \n | \n \n \n 64\n \n | \n
| \n \n Spain\n \n | \n \n \n 1885\n \n | \n \n \n 19804\u2009\u00b1\u2009111796\n \n | \n \n \n 44\n \n | \n \n \n 79\n \n | \n \n \n 74\n \n | \n \n \n 78\n \n | \n \n \n 92\n \n | \n
| \n \n Sweden\n \n | \n \n \n 59\n \n | \n \n \n 82516\u2009\u00b1\u2009144215\n \n | \n \n \n 67\n \n | \n \n \n 47\n \n | \n \n \n 51\n \n | \n \n \n 54\n \n | \n \n \n 80\n \n | \n
| \n \n Switzerland\n \n | \n \n \n 7\n \n | \n \n \n 109682\u2009\u00b1\u2009150017\n \n | \n \n \n 8.4\n \n | \n \n \n 7.7\n \n | \n \n \n 57\n \n | \n \n \n 86\n \n | \n \n \n 100\n \n | \n
| \n \n Turkey\n \n | \n \n \n 18\n \n | \n \n \n 932090\u2009\u00b1\u20091167289\n \n | \n \n \n 13.7\n \n | \n \n \n 20\n \n | \n \n \n 6\n \n | \n \n \n 39\n \n | \n \n \n 17\n \n | \n
| \n \n Ukraine\n \n | \n \n \n 25\n \n | \n \n \n 672451\u2009\u00b1\u2009674177\n \n | \n \n \n 17\n \n | \n \n \n 36\n \n | \n \n \n 32\n \n | \n \n \n 76\n \n | \n \n \n 48\n \n | \n
| \n \n United Kingdom\n \n | \n \n \n 44\n \n | \n \n \n 565707\u2009\u00b1\u20091366590\n \n | \n \n \n 24\n \n | \n \n \n 37\n \n | \n \n \n 41\n \n | \n \n \n 75\n \n | \n \n \n 75\n \n | \n
| \n \n Note: Table only includes countries with more than 5 city actors.\n \n | \n
\n
\n\n
\n| \n | \n\n \n \n Dependent variable\n \n :\n \n | \n
|---|---|
| \n | \n\n \n Annual Percentage Change\n \n | \n
| \n \n Time\n \n | \n \n \n 0.158\n \n ***\n \n \n | \n
| \n | \n\n \n (0.012)\n \n | \n
| \n \n Joined\n \n | \n \n \n -1.638\n \n ***\n \n \n | \n
| \n | \n\n \n (0.130)\n \n | \n
| \n \n Time Since Joining EUCoM\n \n | \n \n \n 0.320\n \n ***\n \n \n | \n
| \n | \n\n \n (0.024)\n \n | \n
| \n \n log(Population density)\n \n | \n \n \n 1.033\n \n ***\n \n \n | \n
| \n | \n\n \n (0.026)\n \n | \n
| \n \n log(GDP per capita)\n \n | \n \n \n 3.484\n \n ***\n \n \n | \n
| \n | \n\n \n (0.150)\n \n | \n
| \n \n Predicted emissions per capita\n \n | \n \n \n 0.339\n \n ***\n \n \n | \n
| \n | \n\n \n (0.010)\n \n | \n
| \n \n Constant\n \n | \n \n \n -52.116\n \n ***\n \n \n | \n
| \n | \n\n \n (3.621)\n \n | \n
| \n \n \n Note\n \n : Standard errors are in parentheses.\n \n\n The regressions include country fixed effects.\n \n *\n \n p\u2009<\u20090.1;\n \n **\n \n p\u2009<\u20090.05;\n \n ***\n \n p\u2009<\u20090.01\n \n | \n
\n
\n\n
\n\n
\n\n
\n\n
\n\n
\n\n
\n\n Supplementary Information\n
\n \n\n Ultra-low temperature resistant adhesive is highly desired yet scarce for material adhesion for the potential usage in Arctic/Antarctic or outer space exploration. Here we develop a solvent-free processed low-temperature tolerant adhesive with excellent adhesion strength and organic solvent stability, wide tolerable temperature range (i.e. -196 to 55\u00b0C), long-lasting adhesion effect (>\u200960 days, -196\u00b0C) that exceeds the classic commercial hot melt adhesives. Notably, manufacturing at scale can be easily achieved by the facile scale-up solvent-free processing, showing much potential towards practical application in Arctic/Antarctic or planetary exploration.\n
\n\n \n One Sentence Summary\n \n : We have designed a kind of solvent-free adhesive with excellent low temperature resistance up to -196\u00b0C and can be readily scale-up manufactured on a kilogram scale through a solvent-free heat-assisted process.\n
\n\n Adhesive plays a critical role in numerous fields such as construction, textile, electronic products and aerospace, etc\n \n \n 1\n \n \u2013\n \n 3\n \n \n . The ever-growing practical demands and sustainable development for society and industry within a wide temperature range, for example, research at the poles of the Earth (e.g., South Pole, 20.7 to -94.2 \u2103) and human space exploration (e.g., Moon, 127 to -183 \u2103; Mars, 20 to -140 \u2103; Saturn, -130 to -191 \u2103; Neptune, -210 to -218 \u2103), call for high-strength adhesive at low temperature\n \n \n 4\n \n \u2013\n \n 8\n \n \n . Up to now, a majority of traditional adhesives are based on polymer as the main component\n \n \n 9\n \n \u2013\n \n 11\n \n \n , for example, commercially available hot melt adhesives include ethylene-vinyl acetate copolymer (EVA), polyamide (PA) and polyether sulfone (PES), etc. Despite their widespread use in daily life, they still have some bottlenecks\n \n \n 12\n \n \u2013\n \n 17\n \n \n , especially at low temperature: 1) high crosslinking density and low surface energy lead to difficult bonding and easy debonding between substrate surface and adhesive; 2) poor interfacial infiltration effect with easily formed thicker adhering layer, resulting in undesired residual stress; 3) the traditional polymer molecules tend to be frozen at low temperature, leading to volumetric contraction, enhanced fragileness, weakened mechanical force transmission across the substrates, and reduced resistance to crack propagation and 4) the long-term stability in low temperature is generally unmet, and the adhesion mechanism especially that under low temperature has been less investigated. Although some strategies, such as adding plasticizer/crosslinking agents or non-polar substituents, can elevate the temperature tolerance range of adhesives to some extent\n \n \n 15\n \n ,\n \n 17\n \n ,\n \n 18\n \n \n , the lowest temperature resistance for most of commercial hot melt adhesives is above \u2212\u200950\u00b0C. Therefore, the novel functional polymer-based adhesives that can be used at ultra-low temperature are still demanded yet largely unmet for specific scenarios like Arctic/Antarctic or outer space exploration.\n
\n\n Polyoxometalates (POMs) have triggered a flurry of attentions to the fields of chemistry and materials science owing to their unique physic-chemical properties\n \n \n 19\n \n \u2013\n \n 22\n \n \n . Actually, POMs are frequently deemed as desired building blocks for constructing adhesive due to the following advantages\n \n \n 23\n \n \u2013\n \n 26\n \n \n : 1) favorable interface adhesion might be promoted by regulating the crosslink density of polymer using POMs to reduce the residual stress; 2) cohesion strength of adhesive would be markedly enhanced by the interaction with POMs carrying multiple hydrogen protons and oxygen-rich surface, resulting in the strong interaction energy and large energy dissipation\n \n \n 27\n \n \u2013\n \n 29\n \n \n and 3) POMs with well-defined structures and compositions can act ideal templates for the theoretical calculations\n \n \n 30\n \n ,\n \n 31\n \n \n . Nevertheless, the currently reported POMs-based adhesives usually rely on solvents (e.g., organic or aqueous agents)\n \n \n 32\n \n \u2013\n \n 37\n \n \n , since it can break intermolecular forces of POMs and other counterparts, and subsequently assemble into viscous materials\n \n \n 38\n \n ,\n \n 39\n \n \n . However, low temperature would lead to phase transition of solvents and media, which makes adhesive become brittle or deformed to seriously weaken their functionality or applicability\n \n \n 40\n \n \n . Up to now, it has been reported that POMs based adhesive can act as low temperature adhesion by removing the solvent\n \n \n 41\n \n \n , yet the presence of solvents would result in many inconveniences in practical applications including storage, transportation, or processing processes, as well as the possible performance suppression caused by the residue of solvents. Thus, based on the above considerations and inspired by the pioneering works, the exploration of solvent-free method to facilely prepare low-temperature tolerant adhesive would be an intriguing target for practical usage like Arctic/Antarctic or outer space exploration, yet related research works have been rarely reported as far we know.\n
\n\n As a proof-of-concept, a type of H\n \n 4\n \n SiW\n \n 12\n \n O\n \n 40\n \n (SiW\n \n 12\n \n ) based solvent-free polymer (SSFP) adhesive has been successfully designed and prepared (Scheme\n \n 1\n \n ). The SSFP adhesive exhibits high adhesion strength, favorable interfacial adhesion ability, excellent organic solvent stability and ultra-low temperature tolerance, which is superior to commercially available hot melt adhesives. Moreover, theoretical calculations prove that the strong interaction energy between SiW\n \n 12\n \n and polyethylene glycol (PEG) though abundant hydrogen bonds endows SSFP adhesive with favorable adhesion performance under a wide temperature range. This work may promote the development of solvent-free adhesives for potential applications of Arctic/Antarctic or planetary exploration.\n
\n\n
\n\n A white solvent-free SSFP adhesive is facilely prepared on a kilogram-scale through a heat-assisted process (Supplementary Fig.\n \n S1\n \n , detail see Methods). Fourier-transform infrared (FT-IR) spectra (Fig.\n \n 1\n \n a and Supplementary Fig. S2) show that four typical characteristic peaks of SiW\n \n 12\n \n deriving from the stretching vibration bands of W\u2009=\u2009O\n \n d\n \n , Si\u2013O\n \n a\n \n , W\u2013O\n \n b\n \n \u2013W and W\u2013O\n \n c\n \n \u2013W, respectively, are still clearly discernible in the SSFP adhesive, indicating the retained structure of SiW\n \n 12\n \n within the SSFP adhesive matrix. Simultaneously, the stretching vibration of etheric oxygen groups (\u2013C\u2013O\u2013C\u2013) in PEG at 1113 cm\n \n \u2212\u20091\n \n is slightly shifted to 1108 cm\n \n \u2212\u20091\n \n after forming SSFP adhesive\n \n \n 30\n \n \n , which is ascribed to the possible hydrogen-bonding interaction. Besides, the powder X-ray diffraction (PXRD) certifies the amorphous nature of SSFP adhesive (Fig.\n \n 1\n \n b), which is different from PEG and SiW\n \n 12\n \n , or even their physical mixture (Supplementary Fig. S3). These results suggest that SiW\n \n 12\n \n is dispersed in SSFP adhesive matrix, and the crystallization of PEG is obviously inhibited after hybridizing with SiW\n \n 12\n \n . Additionally, their structures, chemical compositions and states of PEG and SiW\n \n 12\n \n in adhesive have been confirmed with X-ray photoelectron spectroscopy (XPS) (Supplementary Figs. S4 and S5),\n \n \n 32\n \n \n Si NMR (Fig.\n \n 1\n \n c) spectra\n \n \n 42\n \n \n ,\n \n \n 1\n \n \n H NMR spectra (Fig.\n \n 1\n \n d), solid-state and liquid-state\n \n \n 13\n \n \n C NMR tests (Figs.\n \n 1\n \n , e and f, and Supplementary Fig. S6). The low field nuclear magnetic resonance (LF-\n \n \n 1\n \n \n H NMR) reveals that a\u2009~\u200945 times decrease of crosslink density (from 69.93 \u00d7 10\n \n \u2212\u20094\n \n to 1.55 \u00d7 10\n \n \u2212\u20094\n \n mol mL\n \n \u2212\u20091\n \n ) can be detected in SSFP adhesive when compared with PEG (Fig.\n \n 1\n \n h), indicating that SiW\n \n 12\n \n would be hybridized with PEG and occupies a certain space to decrease the number of cross-linked bonds in PEG. Moreover, the results are further supported by the decaying proton transverse relaxation curves (Fig.\n \n 1\n \n g)\n \n \n 43\n \n \n . Besides, the scanning electron microscopy (SEM) tests show that SSFP adhesive has a denser and flatter surface than that of PEG treated under similar heating process (Fig.\n \n 2\n \n a and Supplementary Fig. S7), proving the vital role of SiW\n \n 12\n \n in decreasing the residual stress. Furthermore, energy-dispersive X-ray spectroscopy (EDS) element mapping analyses indicate the uniform dispersion of SiW\n \n 12\n \n in SSFP adhesive (Fig.\n \n 2\n \n a).\n
\n\n
\n\n The adhesion effect plays a vital role in the application of adhesive materials\n \n \n 44\n \n \n . Satisfyingly, this obtained SSFP adhesive (Fig.\n \n 2\n \n b) exerts favorable adhesion capability on different types of artificial and natural materials (Supplementary Fig. S8). Subsequently, the quantitative tests of adhesion strength of SSFP adhesive (Fig.\n \n 2\n \n e) are measured. Strong adhesion forces are presented on high surface energy substrates, such as stainless steel (SS, 3.7 MPa), glass (3.1 MPa), and copper (Cu, 2.7 MPa), owing to the existence of strong chemical bonds and mechanical interlocking\n \n \n 45\n \n \n . Meanwhile, relatively weaker adhesion forces are shown on low surface energy substrates, such as polycarbonate (PC), polypropylene (PP) and polytetrafluoroethylene (PTFE) (Fig.\n \n 2\n \n e). Notably, this SSFP adhesive with excellent adhesion performance is superior to the almost all of the POMs based adhesives that have been reported so far (Supplementary Fig. S9 and Table\u00a01)\n \n \n 30\n \n ,\n \n 33\n \n \u2013\n \n 36\n \n ,\n \n 38\n \n ,\n \n 42\n \n \n . Besides, the SS substrates joined with SSFP adhesive can easily tolerate a weight of ~\u200950 kg (Fig.\n \n 2\n \n c), indicating the remarkable adhesion capabilities of SSFP adhesive. Furthermore, it can be observed that the adhesion strength of SSFP adhesive on SS still maintains at about 3.7 MPa after multi-recyclable adhesion and deadhesion recycling (Supplementary Fig. S10). The distribution of SSFP adhesive on SS after detachment implies that the adhesion failure mainly occurs in interfacial adhesion between adhesive and substrates, indicating the high cohesion interaction of SSFP adhesive (Supplementary Fig. S11)\n \n \n 46\n \n ,\n \n 47\n \n \n .\n
\n\n
\n\n Based on the above results, SS substrate is selected as a model substrate to further monitor the adhesion performance and elucidate the interaction mechanism of SSFP adhesive. Interestingly, the adhesion strength of SSFP adhesive on SS is positively proportional to SiW\n \n 12\n \n content (mass ratio, 2 : 5 to 2 : 3) (Supplementary Figs. S12 to S16). Nevertheless, the adhesive will transform into a very hard and brittle material with negligible viscosity when adding excessive SiW\n \n 12\n \n . Furthermore, the molecular weight of PEG has been screened from ~\u20092000 to ~\u200920000 based on its viscosity in macroscopic level, and the optimal viscosity can be achieved for PEG (~\u200910000). Therefore, the SSFP adhesive with the optimized composition (PEG, ~\u200910000; mass ratio\u2009=\u20092 : 5) is selected as a model sample for subsequent in-depth study. It is noteworthy that the adhesion strength of SSFP adhesive is far superior to the SiW\n \n 12\n \n based solvent-assisted polymer (SSAP) adhesive\n \n \n 30\n \n \n (approximately 65 times, Supplementary Figs. S17 and S18). As expected, a similar H\n \n 3\n \n PW\n \n 12\n \n O\n \n 40\n \n (PW\n \n 12\n \n ) based adhesive (PSFP) can also be produced by the same solvent-free method (Supplementary Fig.\n \n S1\n \n ), and has been confirmed by various characterizations (Supplementary Figs. S19 to S21). Whereas, the adhesion strength of PSFP adhesive is weaker than that of SSFP adhesive, which testifies that SiW\n \n 12\n \n with more H protons is more conducive to the formation of high-strength adhesives (Supplementary Fig.\n \n S1\n \n 6). Additionally, no adhesive is observed by heat-assisted process after replacing PW\n \n 12\n \n with Na\n \n 3\n \n PW\n \n 12\n \n O\n \n 40\n \n (Supplementary Figs. S22 to S24), confirming the vital role of H protons for the preparation of solvent-free adhesives. In addition, we further investigate the influence of polymer on the formation of adhesive by replacing PEG with polycaprolactone (PCL) carrying less hydrogen bond acceptors. The result displays that similar adhesive is still formed, yet its adhesion strength is much weaker than that of SSFP adhesive (Supplementary Figs. S25 and S26). Besides, no adhesive is formed when replacing PEG with polyethylene (PE) and polyvinylidene difluoride (PVDF) (Supplementary Fig. S25), manifesting the crucial role of hydrogen bond acceptor for the formation of solvent-free adhesives. Beyond that, the PEG analogues, PEGME bearing the methyl group and hydroxyl group at each terminus and PEGdME bearing the methyl groups at both termini, can also crosslink with SiW\n \n 12\n \n to form adhesives (Supplementary Fig. S27), and the rheology and lap-shear adhesion test results show that these adhesives possess similar shear strength (Supplementary Fig. S28) and viscosity (Supplementary Fig. S29), suggesting that the formation and behavior of adhesive are primarily attributable to hydrogen bond interaction between H protons of SiW\n \n 12\n \n and etheric oxygen groups of PEG rather than the terminal groups.\n
\n\n As is well-known, most of adhesives based on polymers tend to be dissolved or swelled in organic solvents, resulting in markedly weakened adhesion performances and seriously hindered practical applications\n \n \n 48\n \n \n . Hence, the maintenance of robust adhesion strength in organic solvents is an eye-catching trait for adhesives. Interestingly, the SSFP adhesive is insoluble in some organic solvents, such as mesitylene (TMB), dioxane (Diox), octanoic acid (OA), 1,4-dibromobutane (DiBrb), petroleum ether (PE) and N-hexane (N-hex). For example, no separation or displacement phenomenon has been observed in a long-term adhesion test for at least 14 days (Supplementary Fig. S30). In addition, the adhesion strength remains relatively stable for 14 days of immersion in different organic solvents (Fig.\n \n 2\n \n f). Moreover, SSFP adhesive is capable of performing rapid adhesion (Supplementary Fig. S31 and video S1) and preventing an emergency leakage for organic solvents (Fig.\n \n 2\n \n d and video S2). In sharp contrast, ethyl vinyl acetate (EVA), a kind of commercially available hot melt adhesive, shows negligible adhesion strength after soaking in mesitylene for only 7 days when compared to that of SSFP adhesive (Supplementary Fig. S32).\n
\n\n Based on above-mentioned high adhesion strength of SSFP adhesive, its viscosity, energy storage and loss modulus have been further traced using rheology tests. Compared to the PEG, the SSFP adhesive exhibits higher viscosity and lower liquidity in rheological characterization (Supplementary Fig. S33). Rheology measurements verify that the modulus and viscosity are both inversely proportional to the operating temperature (Figs.\n \n 3\n \n , a and b). The SSFP adhesive remains gel-like or solid-like state in macroscopic behaviors at below ~\u200950\u00b0C. Specifically, loss modulus of the SSFP adhesive (Fig.\n \n 3\n \n a) exceeds storage modulus (G\u2033 > G\u2032) at above ~\u200950\u00b0C, resulting in a viscosity-dominated viscoelasticity state that can accelerate the interfacial bonding. Moreover, the reversibility of storage modulus (G\u2032), loss modulus (G\u2033), and complex viscosity (\u03b7*) can be realized in cycling tests under circulating temperatures, which might be attributed to the solvent-free phase and invertible hydrogen bond interaction that enable reversible temperature-induced rheological behaviors.\n
\n\n
\n\n Viscous materials with low-temperature resistance play an important role in exploration under extreme environments especially in a wide temperature-variable range, such as, research at the poles of the Earth (e.g., South Pole, 20.7 to -94.2 \u2103) and exploration of the outer planets (e.g., Mars, 20 to -140 \u2103). However, traditional polymer based adhesives tend to become brittle, and increase its residual stress as the temperature decreases\n \n \n 49\n \n \n . Here, the thermogravimetric analysis (TGA) and differential scanning calorimeter (DSC) tests display that the glass transition temperature (\n \n T\n \n \n g\n \n ) of SSFP adhesive is -31.1\u00b0C, which is lower than that of PEG (48.3\u00b0C) (Supplementary Figs. S34 and S35), implying that the SSFP adhesive has better flexibility at relatively low temperature. Impressively, the SSFP adhesive adhered between SS slices can easily tolerate a 2 kg weight under liquid nitrogen conditions, and still maintain the original state after returning to room temperature (Fig.\n \n 3\n \n c and video S3). In addition, the SSFP adhesive is not observed to have significant contraction or rupture after freezing with liquid nitrogen (Supplementary Fig. S36). In contrast, the obvious volumetric shrinkage and rupture for PEG and EVA adhesive occur at 25\u00b0C and \u2212\u2009196\u00b0C, respectively (Supplementary Figs. S37 and S38). The above results show that volumetric contraction of SSFP adhesive can be significantly inhibited at -196\u00b0C after crosslinking SiW\n \n 12\n \n with PEG. To support it, the adhesion strength of SSFP adhesive has been tested at different temperatures. Particularly, the adhesion strength is negligible for SSFP adhesive at 55\u00b0C, then gradually increases as the temperature decreases to 4\u00b0C (~\u20093.8 MPa) (Fig.\n \n 3\n \n d and Supplementary Fig. S39). After that, the adhesion strength gradually decreases when the temperature decreases to -196\u00b0C. More significantly, the SSFP adhesive can still maintain relatively high adhesion strength of 2.96 MPa after immersion in liquid nitrogen for 60 days (Supplementary Fig. S40), while the EVA adhesive is immediately frozen-cracked once it entered liquid nitrogen (Supplementary Fig. S41). These results demonstrate that the SSFP adhesive has admirable ultra-low temperature-resistant adhesion performance than that of commercial hot melt adhesive. Furthermore, the low-temperature rheological measurements (Fig.\n \n 3\n \n e) showed that SSFP adhesive has a stable storage modulus (G\u2032) and loss modulus (G\u2033) in the temperature range from \u2212\u2009120 to 25\u00b0C, which is consistent with its actual performance. Moreover, the temperature-dependent FT-IR spectra (Fig.\n \n 3\n \n f and Supplementary Fig. S42) indicate negligible change in the four signals of both SiW\n \n 12\n \n and PEG, manifesting that the interactions in SSFP adhesive remains almost intact under wide temperature range\n \n \n 50\n \n \n .\n
\n\n
\n\n Based on above excellent adhesion properties of SSFP adhesive, the adhesion mechanism has been further investigated. In general, cohesion and interfacial adhesion are two main factors affecting adhesion performance. Nevertheless, most of the theoretical calculation of adhesives focus on the simulation between adhesives and substrates, and it is still very scarce for the study of interaction contributing to cohesion. Thus, we have applied the density functional theory (DFT) calculations to investigate the SSFP adhesive at molecular level\n \n \n 31\n \n ,\n \n 41\n \n \n . The results reveal that the interaction energy between them is -229.69 kJ/mol for one PEG fragment, -507.14 kJ/mol for two PEG fragments, and \u2212\u2009764.81 kJ/mol for three PEG fragments, respectively (Fig.\n \n 4\n \n a). Obviously, the binding stability between POMs and PEG can be significantly enhanced via strong hydrogen bond interaction. Hence, SiW\n \n 12\n \n as a cross-linking agent will be interweaved and anchored into PEG networks, resulting in the formation of stable and durable adhesive. In addition, the molecular dynamics (MD) simulation is further performed to evaluate the temperature-dependent interaction energy and hydrogen bonds. At 25\u00b0C, the interaction energy and hydrogen bond percentage between PEG and SiW\n \n 12\n \n are average \u2212\u20091168 kJ/mol and 39.13% in this model system at 2 ns, respectively (Fig.\n \n 4\n \n b, Supplementary Fig. S43 and video S4). At 55\u00b0C, the fluctuation of interaction energy is more obvious, and significantly reduces to -980 kJ/mol (Supplementary Figs. S44 and S45, video S5). Moreover, the percentage of hydrogen bonds (40.0%) at 55\u00b0C remains almost the same with that at 25\u00b0C (Supplementary Fig. S46). Remarkably, at -196\u00b0C, the interaction energy quickly reaches equilibrium and remain stable within 50 ps (Fig.\n \n 4\n \n c and video S6). In addition, the dramatically increased interaction energy remains at average \u2212\u20091250 kJ/mol (Fig.\n \n 4\n \n d), meanwhile the percentage of formative hydrogen bonds (35.71%) is slightly affected by low temperature (Supplementary Fig. S43). High interaction energy at low temperature will elicit large energy dissipation of SSFP when dragging the adhesive, in which the synergistic interaction of them might be the dominating reasons for the SSFP adhesive to exhibit excellent adhesion performance at ultra-low temperature. As a contrast, when replacing SiW\n \n 12\n \n with PW\n \n 12\n \n , both the interaction energy and the percentage of formative hydrogen bonds are significantly reduced to -817 kJ/mol and 29.27% (Supplementary Figs. S47 to S49), owing to the fewer H protons that can be provided by PW\n \n 12\n \n to cross-link with PEG under the same conditions.\n
\n\n In summary, we have prepared a kind of POMs based solvent-free polymer adhesive on a kilogram scale through a heat-assisted process. It is worth noting that the achieved SSFP displays excellent interfacial adhesion ability on different substrates, high adhesion strength and organic solvent stability, wide tolerable temperature range (i.e. -196 to 55\u00b0C) and long-lasting adhesion effects (>\u200960 days) at -196\u00b0C. The high-performance of SSFP exceeds that of commercial hot melt adhesives. Furthermore, combined experimental results with theoretical calculations, the strong interaction energy between POMs and PEG is the main factor for the high adhesion performance at low temperature, possessing enhanced cohesion strength, suppressed polymer crystallization and volumetric contraction. This work enriches the types of low-temperature resistance adhesives, and would shed light on the development of advanced solvent-free adhesives for Arctic/Antarctic or planetary exploration.\n
\n\n Scheme 1 is available in the Supplementary Files section.\n
\n\n Supplementary Video S1. The SSFP adhesive performing rapid adhesion in organic solvent\n
\n \n\n Supplementary Video S2. The SSFP adhesive preventing an emergency leakage for organic solvent.\n
\n \n\n Supplementary Video S3. The SSFP adhesive performing low-temperature resistance.\n
\n \n\n Supplementary Video S4. MD of the SSFP adhesive at 25 \u2103.\n
\n \n\n Supplementary Video S5. MD of the SSFP adhesive at 55 \u2103.\n
\n \n\n Supplementary Video S6. MD of the SSFP adhesive at -196 \u2103.\n
\n \n\n Supporting Information\n
\n \n\n Scheme 1. Schematic illustration of the low temperature effect on adhesion behaviors. a, The lowest temperatures of the South Pole and representative outer space planets. b, The schematic illustration of the low temperature effect on adhesion behaviors for the solvent-assisted and solvent-free POMs based adhesives.\n
\n \n\n Despite the important advances in azaaryl C\u2212H activation/functionalization, reductive functionalization of ubiquitously distributed but weakly reactive azaarenes remains to date a challenge. Herein, by a strategy incorporating a tandem coupling sequence into the reduction of azaarenes, we present a dearomative annulation of azaarenes into promising fused\n \n syn\n \n -N-heterocycles by combination with a large variety of aniline derivatives and paraformaldehyde under ruthenium(II) reductive catalysis, proceeding with excellent selectivity, mild conditions, and broad substrate and functional group compatibility. Mechanistic studies reveal that the products are formed via hydride transfer-initiated\n \n \u03b2\n \n -aminomethylation and\n \n \u03b1\n \n -arylation of the pyridyl core in azaarenes, paraformaldehyde serves as both the C1-building block and reductant precursor, and the use of Mg(OMe)\n \n 2\n \n base plays a critical role in determining the reaction chemo-selectivity by lowering the hydrogen transfer rate. The present work opens a door to further develop valuable reductive functionalization of unsaturated systems by taking profit of formaldehyde-endowed two functions.\n
\n\n \n \n Catalysis\n \n \n
\n\n \n syn-N-heterocycles\n \n
\n\n \n azaarenes\n \n
\n\n \n paraformaldehyde\n \n
\n\n Azaarenes constitute a class of ubiquitously distributed substances applied in numerous fields of science and technology.\n \n 1\u20132\n \n The development of new strategies enabling efficient and selective transformation of weakly reactive azaarenes into functional frameworks is of important significance, as they not only pave the avenues to access novel functional products, but also enrich the synthetic connotation of the azaarenes. To date, except the well-established electrophilic substitution utilizing azaarenes as the nucleophiles under harsh conditions,\n \n 3\u20134\n \n the recently emerged C\u2212H activation/functionalization has offered many desirable ways for structural modification of the azaarenes.\n \n 5\u20137\n \n In comparison with these aromaticity-retaining transformations, only a handful examples focused on dearomative coupling of active indole derivatives,\n \n 8\u201311\n \n whereas dearomative functionalization of inert pyridine-fused azaarenes (e.g., quinolines, isoquinolines, naphthyridines, phenanthroline, etc.)\n \n 12\u201314\n \n has been scarcely explored.\n
\n\n In recent years, hydrogen transfer-mediated coupling reactions have emerged as appealing tools in the production of various functional products, since there is no need for high pressurized H\n \n 2\n \n and elaborate experimental setups. For instance, in addition to the well-known reductive amination applied for amine syntheses,\n \n 15\u201316\n \n several groups such as Beller,\n \n 17\u201318\n \n Kempe,\n \n 19\u201320\n \n Kirchner,\n \n 21\u201322\n \n Liu,\n \n 23\u201325\n \n and others\n \n 26\u201329\n \n have applied borrowing-hydrogen strategy to alkylate amines and the\n \n \u03b1\n \n -site of carbonyl compounds with alcohols. Krische has demonstrated elegant contributions on the linkage of alcohols/carbonyls with unsaturated C\u2212C bonds.\n \n 30\u201332\n \n Bruneau et al have achieved\n \n \u03b2\n \n -C(sp\n \n 3\n \n )\u2212H alkylation of N-alkyl cyclic amines.\n \n 33\u201334\n \n The Li group has converted phenols into synthetically useful amines.\n \n 35\u201336\n \n Our group has reported a reductive quinolyl\n \n \u03b2\n \n -C\u2212H alkylation with a low-active heterogeneous cobalt catalyst.\n \n 37\n \n Later, Donohoe et al have demonstrated interesting examples on the\n \n \u03b2\n \n -functionalization of azaarenes.\n \n 38\u201340\n \n Despite these important advances, the strategy incorporating a tandem coupling sequence into the reduction of azaarenes remains to date a challenge due to the difficulty in controlling the reaction selectivity: on one hand, the azaarenes tends to undergo direct hydrogenation to form non-coupled cyclic amines under catalytic reduction conditions, on the other hand, it is hard to selectively transfer hydrogen only to one specific sites among different substrates.\n
\n\n Here, we conceived that, through an initial pretreatment of azaarenes\n \n A\u2019\n \n with bromoalkanes to form azaarenium salts\n \n 41\u201342\n \n , it would offer a solution to achieve the desired synthetic purpose: (\u2170) the combination of a suitable metal catalyst (M) and hydrogen donor (HD) forms reductive metal hydride species [HM\n \n n\n \n X]\n \n in-situ\n \n , which allows hydride transfer (TH) to the azaarenium salts\n \n A\n \n and generates allylic amine\n \n int-1\n \n and its tautomer N-alkyl enamine\n \n int-2\n \n (Fig.\n \n 1\n \n a). Such an enamine (\n \n int-2\n \n ) has higher\n \n \u03b2\n \n -reactivity in trapping electrophiles than its -NH counterpart and lowers the formation of non-coupled cyclic amine\n \n A\u2019\u2019\n \n . (\u2171) It is relatively difficult to reduce electron-rich enamine\n \n int-2\n \n to the undesired cyclic amine\n \n A''\n \n . Based on such an idea, we wish herein to report, for the first time, a dearomative annulation reaction of azaarenium salts\n \n A\n \n with aniline derivatives\n \n B\n \n and paraformaldehyde (Fig.\n \n 1\n \n b) under ruthenium(II) reductive catalysis, which offers a general way for diastereoselective construction of fused\n \n syn\n \n -N-heterocycles\n \n C\n \n featuring promising structural motifs of teterahydroquinoline and hexahydro-1,6-naphthyridine that are frequently found in natural alkaloids\n \n 43\u201344\n \n and biomedical molecules,\n \n 45\u201347\n \n as exemplified by the leading anesthesia drug taripiprazole\n \n 1\n \n , active composition\n \n 2\n \n used for treating EP1 receptor-mediated diseases,\n \n 45\n \n PXR agonist\n \n 3\n \n ,\n \n 46\n \n and anticancer agents\n \n 4\n \n (Fig.\n \n 1\n \n c)\n \n 47\n \n .\n
\n\n
\n\n \n Investigation of reaction conditions.\n \n We commenced our studies by performing the reaction of N-benzyl quinolinium bromide\n \n A\n \n 1\n \n \n , N-ethyl aniline\n \n B\n \n 1\n \n \n , paraformaldehyde, and base in MeOH at 65\n \n o\n \n C for 18 h by employing\u00a0[RuCl\n \n 2\n \n (\n \n p\n \n -cymene)]\n \n 2\n \n as the catalyst. Among various bases tested, Mg(OMe)\n \n 2\n \n exhibited the best chemo-selectivity since there is no formation of by-product N-benzyl tetrahydroquinoline\n \n A\n \n 1\n \n \n \u2019\u2019 (Table 1, entries 1-4). The absence of catalyst or base failed to yield product\n \n C\n \n 1\n \n \n (entries 5-6), showing that both of them are indispensable for the product formation. Then, we screened several other metal catalysts applied frequently in hydrogen transfer reactions (see Table S1 in the Supplementary Information (SI)). The results showed that Ir(I) or Ir(III) catalysts\u00a0were also applicable, but the base metal catalysts (Co, Fe, Mn, and Ni) were totally ineffective for the transformation\n \n \n \n \n (entries 7-8 and Table S1).\u00a0Here, we chose the cost-effective\u00a0[Ru(\n \n p\n \n -cymene)Cl\n \n 2\n \n ]\n \n 2\n \n \n \n as the preferred catalyst to further evaluate the solvents and temperatures, it showed that methanol and 55\n \n o\n \n C were more preferable (entries 9-10). Decrease of the base or (CH\n \n 2\n \n O)\n \n n\n \n amount diminished the product yields (entries 11-12). Thus, the optimal yield of product\n \n C\n \n 1\n \n \n was obtained when the reaction in methanol was performed at 55\n \n o\n \n C for 18 h by using the combination of\u00a0[Ru(\n \n p\n \n -cymene)Cl\n \n 2\n \n ]\n \n 2\n \n and Mg(OMe)\n \n 2\n \n (entry 10). Interestingly, the use of Mg(OMe)2 base always resulted in excellent selectivity in affording product\n \n C\n \n 1\n \n \n (entries 3, and 7-12).\n
\n\n \n Substrate scope.\n \n With the availability of the optimal reaction conditions (Table 1, entry 10), we then assessed the substrate scope of the newly developed synthetic protocol. As shown in Fig. 2, various quinolinium salts\n \n A\n \n (\n \n A\n \n 1\n \n \n \u2212\n \n A\n \n 21\n \n \n , see Scheme S1 for their structures) in combination with N-ethylaniline\n \n B\n \n 1\n \n \n and paraformaldehyde were evaluated. Gratifyingly, all the reactions underwent smooth reductive annulation and furnished the desired fused N-heterocycles in reasonable to excellent isolated yields with excellent\n \n syn\n \n -diastereoselectivity (\n \n C\n \n 1\n \n \n \u2212\n \n C\n \n 21,\n \n \n d.r. > 20 : 1). The structure of compound\n \n C\n \n 1\n \n \n was confirmed by X-ray crystallography diffraction and NOESY spectrum (Fig. S1-Fig.S3). As expected, the application of 1,5-dibromopentane generated the alkyl-linked product\n \n C\n \n 21\n \n \n in good yield. Noteworthy, a variety of functionalities (e.g., \u2212Me, \u2212OMe, \u2212SPh, amido, \u2212F, \u2212Cl, \u2212Br, ester, \u2212CF\n \n 3\n \n , \u2212NO\n \n 2\n \n , alkenyl, and alkyl) on the quinolinium salts were well tolerated, and their electronic properties affected the product formation to some extent. Interestingly, no reduction of the nitro and alkenyl groups was observed (\n \n C\n \n 17\n \n \n and\n \n C\n \n 18\n \n \n ), and the halo-substrates also did not undergo hydrodehalogenation, indicating that the reaction proceeds in a chemoselective manner. In general, quinolines bearing an electron-donating group (\n \n C\n \n 2\n \n \n \u2212\n \n C\n \n 7\n \n \n , and\n \n C\n \n 13\n \n \n \u2212\n \n C\n \n 14\n \n \n ) afforded relatively higher product yields than those having an electron-withdrawing group (\n \n C\n \n 8\n \n \n \u2212\n \n C\n \n 10,\n \n \n and\n \n C\n \n 12\n \n \n ), presumably because the electron-rich quinolinium salts can result in more reactive enamine intermediates that are beneficial to the electrophilic coupling process (Fig. 1a). The retention of these functionalities offers the potential for post-functionalization of the obtained products.\n
\n\n Next, we turned our attention to the synthesis of structurally diversified products by variation of both azaarenes\n \n A'\n \n and anilines\n \n B\n \n . First, a series of N-alkyl anilines (\n \n B\n \n 2\n \n \n \u2212\n \n B\n \n 18\n \n \n , see Scheme S2 for their structures) in combination with quinolinium salt\n \n A\n \n 1\n \n \n were tested. As illustrated in Fig. 3, all the reactions efficiently afforded the desired product in moderate to excellent isolated yields with exclusive\n \n syn\n \n -selectivity (\n \n C\n \n 22\n \n \n \u2212\n \n C\n \n 36\n \n \n , d.r. > 20\u00a0:\u00a01). The electronic properties of the substituents on the benzene ring of the anilines significantly affected the product formation. Especially, anilines containing electron-donating groups (\n \n C\n \n 22\n \n \n \u2212\n \n C\n \n 23\n \n \n ,\n \n \n \n C\n \n 27\n \n \n and\n \n \n \n C\n \n 35\n \n \n ) gave much higher yields than those with electron-withdrawing groups (\n \n C\n \n 24\n \n \n \u2212\n \n C\n \n 25\n \n \n ). This observation is attributed to electron-rich anilines favoring the electrophilic coupling process during the formation of the products. In addition to N-alkyl anilines, diarylamine\n \n B\n \n 16\n \n \n also served as an effective coupling partner, affording the N-aryl product\n \n C\n \n 33\n \n \n in moderate yield. As expected, primary anilines were not applicable for the transformation, as they easily reacted with formaldehyde to form aminals. Interestingly, tetrahydroquinolines (\n \n B\n \n 8\n \n \n and\n \n B\n \n 9\n \n \n ) and 2,3,4,5-tetrahydro-1H-benzo[b]azepine (\n \n B\n \n 10\n \n \n ), two specific aniline derivatives, also underwent efficient multicomponent annulation to afford the polycyclic products (\n \n C\n \n 34\n \n \n \u2212\n \n C\n \n 36\n \n \n ,\n \n \n \n C\n \n 38\n \n \n and\n \n \n \n C\n \n 41\n \n \n ). In addition to quinolines, other azaarenes, such as 1,8-naphthyridines (\n \n A\n \n 22\n \n \n \u2212\n \n A\n \n 25\n \n \n ), thieno[3,2-b]pyridine\n \n A\n \n 26\n \n \n , 1,7-phenanthroline\n \n A\n \n 27\n \n \n , 1,10-phenanthroline\n \n A\n \n 28\n \n \n , 5-substituted isoquinolines, and more challenging pyridine were also amenable to the transformation, delivering the desired products in an efficient manner (\n \n C\n \n 37\n \n \n \u2212\n \n C\n \n 50\n \n \n , d.r. > 20\u00a0\uff1a1), these examples demonstrate the capability of the developed chemistry in the functionalization of pyridine-containing azaarenes including the N-bidentate ligands (\n \n C\n \n 37\n \n \n \u2212\n \n C\n \n 42\n \n \n ,\n \n C\n \n 47\n \n \n ).\n
\n\n Noteworthy, 5-substituted isoquinolines afforded the desired annulation products (Fig. 3,\n \n C\n \n 48\n \n \n and\n \n C\n \n 49\n \n \n ), whereas 5-nonsubstituted isoquinolines generated structurally novel products\n \n D\n \n by installing an additional methyl group at the\n \n \u03b2\n \n -site of the N-heteroaryl reactants, and all the products are produced with exclusive\n \n syn\n \n -diastereoselectivity (d.r. > 20 :1, Fig. S4). As shown in Fig. 4, N-benzyl isoquinolinium salts were firstly employed to couple with paraformaldehyde and N-ethyl aniline\n \n B\n \n 1\n \n \n . All the reactions gave rise to the desired annulation products in moderate to excellent yields upon isolation (\n \n D\n \n 1\n \n \n \u2212\n \n D\n \n 13\n \n \n ). Then, the transformation of secondary anilines including the N-alkyl and N-aryl ones was evaluated. Gratifyingly, all these anilines smoothly coupled with N-benzyl quinolium salt\n \n A\n \n 1\n \n \n and paraformaldehyde, delivering the annulation products in reasonable to high yields (\n \n D\n \n 14\n \n \n \u2212\n \n D\n \n 26\n \n \n ). Similar to the results described in Fig. 2 and 3, various functionalities on both isoquinolium salts and anilines are well tolerated (\u2212Bn, \u2212Et, \u2212Me, \u2212F, \u2212Cl, \u2212Br, boronic ester, \u2212SO\n \n 2\n \n Me, \u2212\n \n n\n \n -hexyl, \u2212OMe, \u2212CF\n \n 3\n \n , \u2212CO\n \n 2\n \n Me, alkenyl, cyclohexyl, and\n \n i\n \n -propyl). The substituents on the aryl ring of the isoquinoline salts have little influence on the product formation, whereas the substituents of the anilines significantly affected the product yields. Generally, aniline bearing an electron-donating group afforded higher yields (e.g.,\n \n D\n \n 14\n \n \n \u2212\n \n D\n \n 16\n \n \n and\n \n D\n \n 20\n \n \n \u2212\n \n D\n \n 23\n \n \n ) than those of anilines with an electron-withdrawing group (e.g.,\n \n D\n \n 17\n \n \n \u2212\n \n D\n \n 19\n \n \n and\n \n D\n \n 24\n \n \n ), suggesting that the reaction involves an electrophilic coupling process. Benzocyclic amines (1,2,3,4-tetrahydroquinoxaline, 1,2,3,4-tetrahydroquinoline, and 2,3,4,5-tetrahydro-1H-benzo[b]azepine) and N1-isopropyl-N4-phenylbenzene-1,4-diamine also served as effective coupling partners, affording the polycyclic products in moderate to high yields (\n \n D\n \n 27\n \n \n \u2212\n \n D\n \n 30\n \n \n ). These examples show the practicality of the developed chemistry in the construction of structurally complex polycyclic N-heterocycles.\n
\n\n \n Synthetic applications\n \n \n .\n \n Further, we explored the synthetic applications of the developed chemistry. As shown in Fig. 5a, 6-ester quinolinium salts, arising from initial esterification of 6-carboxylic quinoline and subsequent pretreatment with benzyl bromide, efficiently reacted with aniline\n \n B\n \n 1\n \n \n and paraformaldehyde to afford products\n \n C\n \n 51\n \n \n \n \n and\n \n C\n \n 52\n \n \n , which are the analogues of analgesic\n \n 48\n \n and the agents used for antioxidation and antiproliferation\n \n 49\n \n , respectively. Through successive amidation and formation of N-benzyl heteroarenium salt, 6-amino quinoline was efficiently transformed in combination with aniline\n \n B\n \n 1\n \n \n into camphanic amide\n \n C\n \n 53\n \n \n (Fig. 5b), an agent capable of stereoisomeric separation.\n \n 50\n \n Further, the gram-scale synthesis of product\n \n D\n \n 1\n \n \n was successfully achieved by scaling up the reactants to 10 mmol, which still gave a desirable yield (Fig. 5c). Interestingly, representative compounds\n \n C\n \n 29\n \n \n and\n \n D\n \n 1\n \n \n underwent efficient debenzylation to afford N-unmasked products\n \n C\n \n 54\n \n \n and\n \n D\n \n 31\n \n \n in the presence of a Pd/HCOONH\n \n 4\n \n system in methanol (Fig. 5d), which demonstrates the practicality of the developed chemistry in further preparation of fused heterocycles containing a useful -NH motif.\n
\n\n \n Mechanistic investigations.\n \n To gain mechanistic insights into the reaction, we conducted several control experiments (Fig. 6). First, the model reaction was interrupted after 6 hours to analyze the product system. Except for the formation of product\n \n C\n \n 1\n \n \n in 23% yield, a dihydroquinoline\n \n int-1\n \n was isolated in 5% yield (Fig. 6a). Subjection of compound\n \n int-1\n \n (Fig. 6a and Fig. S6) with aniline\n \n B\n \n 1\n \n \n under the standard conditions resulted in product\n \n C\n \n 1\n \n \n in high isolated yield (Fig. 6b), showing that\n \n int-1\n \n is a key reaction intermediate. However, removal of Ru-catalyst from the standard conditions failed to produce\n \n C\n \n 1\n \n \n and the \u03b1-arylated product\n \n C\n \n 1\n \n '\n \n (Fig. 6c), revealing that the reaction initiates with Ru-catalyzed hydrogen transfer, instead of nucleophilic arylation of substrate\n \n A\n \n 1\n \n \n with aniline\n \n B\n \n 1\n \n \n . Further, the model reaction using deuterated methanol solvent yielded product\n \n C\n \n 1\n \n \n without any D-incorporation (Fig. 6d). In sharp contrast, the same reaction by replacing paraformaldehyde with the fully deuterated one gave product\n \n C\n \n 1\n \n -\n \n d\n \n \n n\n \n \n with 35% and 27% D-ratios at the\n \n \u03b1\n \n and\n \n \u03b3\n \n -sites and more than 99% D-ratio at the newly formed aminomethyl group (Fig. 6e and Fig. S8). These two crucial experiments show that the formaldehyde serves as both the source of the reductant and C1-building block for the formation of the newly formed\n \n \u03b2\n \n -methylene group, and there is a tautomerism between\n \n int-1\n \n and enamine\n \n int-2\n \n (Fig. 1a). In parallel, we conducted the control experiments in terms of the generation of product\n \n D\n \n 1\n \n \n \n \n (Scheme S3). The results also support that dihydroquinoline\n \n int-6\n \n (Scheme S3b) and\n \n \u03b2\n \n -methyl dihydroquinoline\n \n int-9\n \n (Fig S7) are the reaction intermediates, and formaldehyde serves as the reductant source and C1-building block in the construction of the product (Scheme S3d, S3e and Fig. S9).\n
\n\n Based on the above findings, the plausible pathways toward the formation products\n \n C\n \n 1\n \n \n and\n \n D\n \n 1\n \n \n are depicted in Fig. 7. Initially, the metal hydride species [Ru\n \n II\n \n HX] is generated via Mg(OMe)\n \n 2\n \n addition to formaldehyde (\n \n E\n \n 1\n \n \n ) followed by transmetallation (\n \n E\n \n 2\n \n \n ) with [Ru\n \n II\n \n X\n \n 2\n \n ] and\n \n \u03b2\n \n -hydride elimination and release of formate ester (detected by GC-MS analysis, Fig. S5). Then, the hydride transfer from [Ru\n \n II\n \n HX] to quinolinium salt\n \n A\n \n 1\n \n \n forms dihydroquinoline\n \n int-1\n \n and its enamine tautomer\n \n int-2\n \n along with regeneration of the catalyst precursor\n \n \n [Ru\n \n II\n \n X\n \n 2\n \n ]. Meanwhile, the condensation between aniline\n \n B\n \n 1\n \n \n and formaldehyde affords iminium\n \n B\n \n 1\n \n '\n \n . Then, the\n \n \u03b2\n \n -nucleophilic addition of\n \n int-2\n \n to\n \n B\n \n 1\n \n '\n \n gives the\n \n \u03b2\n \n -aminoalkyl iminium\n \n int-3\n \n . Further, the electron-rich benzene ring of\n \n int-3\n \n attacks the iminium motif from both the same (\n \n int-3b\n \n ) and opposite (\n \n int-3a\n \n ) sides of the H-atom at the\n \n \u03b2\n \n -site. In comparison, the form of\n \n int-3a\n \n (opposite side) is more favorable due to the less steric hindrance, thus affording product\n \n C\n \n 1\n \n \n with\n \n syn\n \n -selectivity after deprotonation of the coupling adduct\n \n int-4\n \n (path a of Fig 7b, namely electrophilic aryl C\u2212H aminoalkylation). Alternatively, the [4+2] cycloaddition of\n \n int-2\n \n and\n \n B\n \n 1\n \n '\n \n \n \n via\n \n endo\n \n or\n \n exo\n \n \u03c0-\u03c0 stacking also rationalizes the formation of\n \n int-4\n \n and product\n \n C\n \n 1\n \n \n (path b of Fig. 7b, via\n \n int-5\n \n and\n \n int -4\n \n ). Similarly, the generation of product\n \n D\n \n 1\n \n \n from isoquinoline is shown in Fig. 6c. The hydride transfer from [Ru\n \n II\n \n HX] to isoquinolinium salt\n \n A\n \n 32\n \n \n initially forms enamine\n \n int-6\n \n (Scheme S3a, S3b and Fig. S6).\n \n \n Then, the\n \n \u03b2\n \n -capture of formaldehyde by\n \n int-6\n \n followed by\n \n \n based-facilitated dehydration of\n \n int-7\n \n and hydride transfer to alkenyl iminium salt\n \n int-8\n \n forms\n \n \n \n \u03b2\n \n -methyl enamine\n \n int-9\n \n (Scheme S3a and Fig. S7). Subsequently, the\n \n \u03b2\n \n -capture of\n \n B1\u2019\n \n by\n \n int-9\n \n followed by intramolecular attack of the electron-rich phenyl ring to the iminium motif of\n \n int-10\n \n from the sterically less hindered back side of the methyl group, or the [4+2] cycloaddition of\n \n int-9\n \n and\n \n B\n \n 1\n \n '\n \n \n \n via \u03c0-\u03c0 stacking gives intermediate\n \n int-11\n \n . Finally, the deprotonation of\n \n int-11\n \n generates product\n \n D\n \n 1\n \n \n with\n \n syn\n \n -diastereoselectivity (Fig 7c).\n
\n\n To better reveal the product formation including the unique\n \n cis\n \n -selectivity, computational study was preformed using the density functional theory (see details in SI). First, the participation of Mg(OMe)\n \n 2\n \n , KOMe, and\n \n t\n \n -BuOK as the bases in the generation of [Ru\n \n II\n \n HCl] was calculated. The barrier for the transmetallation step with Mg(OMe)\n \n 2\n \n (14 kcal\u00b7mol\n \n -1\n \n , Fig. S10) is significantly higher than the other two bases (6.6 kcal\u00b7mol\n \n -1\n \n for\n \n t\n \n -BuOK and 3.8\u00a0kcal\u00b7mol\n \n -1\n \n for\n \n t\n \n -MeOK, Fig. S11 and Fig. 12) in the potential energy surfaces. This trend is in accordance with the fact that the higher charge of Mg\n \n 2+\n \n increases the stability of adduct\n \n E\n \n 1\n \n \n (Scheme 7a) and makes the dissociation of -MgOMe and transmetallation more difficult, thus leading to a slow forming rate of [Ru\n \n II\n \n HCl]. Correspondingly, a slow generation of enamine\n \n int-2\n \n via hydride transfer from [Ru\n \n II\n \n HCl] to the azaarenium salt\n \n A1\n \n is beneficial to the capture of\n \n int-2\n \n by\n \n B1\n \n ', and effectively avoids the formation of undesired N-benzyl tetrahydroquinoline\n \n A\u2019\u2019\n \n (table 1).\n
\n\n Then, the potential energy profile computed for the conversion of\n \n int-2\n \n and\n \n B\n \n 1\n \n '\n \n to\n \n C\n \n 1\n \n \n is shown in Fig. 8a, and all of the structures were optimized in CH\n \n 3\n \n OH solution. The formation of intermediate\n \n int-3\n \n via\n \n \u03b2\n \n -nucleophilic addition of\n \n int-2\n \n to\n \n B\n \n 1\n \n '\n \n \n \n has an energy barrier of 14.0\u00a0kcal mol\n \n -1\n \n (\n \n TS4\n \n ), which represents an exergonic process, as\n \n int-3a\n \n is 6.3 kcal\u00b7mol\n \n -1\n \n higher than\n \n int-2\n \n . In comparison, the formation of\n \n int-3b\n \n has a similar energy barrier of 14.5 kcal mol\n \n -1\n \n (\n \n TS4\n \n \n '\n \n ), the torsion of C3-N2 bond of\n \n int-3a\n \n to form\n \n int-3b\n \n has a barrier of 8.7 kcal mol\n \n -1\n \n (\n \n TS5\n \n ). However, the attack of the aniline benzene ring to the iminium motif of\n \n int-3b\n \n from the same side of the pyridyl\n \n \u03b2\n \n -H is less favored, which is due to the high stereoscopic hindrance of the N-ethyl group and pyridyl\n \n \u03b2\n \n -H as well as the long distance (~5.7 \u00c5) between the pyridyl\n \n \u03b1\n \n -carbon (C1) and aniline\n \n ortho\n \n -carbon (C5). Thus,\n \n int-3a\n \n becomes a favorable intermediate. Starting with\n \n int-3a\n \n , the attack of aryl C5-atom on the C1-atom to form\n \n int-4\n \n with a barrier of 15.5 kcal mol\n \n -1\n \n (\n \n TS6\n \n ) represents an exergonic reaction, as\n \n int-4\n \n is 10.8 kcal mol\n \n -1\n \n higher than\n \n int-3a\n \n . Finally, the formation of product\n \n C\n \n 1\n \n \n ,\n \n \n via deprotonative aromatization of the coupling adduct\n \n int-4\n \n has no energy barrier, is favored from a thermodynamic point of view (\u0394\n \n G\n \n = -54.1 kcal mol\n \n -1\n \n ). In terms of the [4+2] cycloaddition of\n \n B\n \n 1\n \n '\n \n and\n \n int-2\n \n ,\n \n \n the manner of\n \n endo\n \n \u03c0-\u03c0 stacking encountered commonly in the Diels\u2013Alder reactions has a significant energy barrier of 39.0 kcal mol\n \n -1\n \n (\n \n TS7\n \n ). So, this pathway is disfavored. Meanwhile, we also found that it is difficult to form the transition state of\n \n exo\n \n \u03c0-\u03c0 stacking due to the higher steric hindrance and long interaction distance. Based on the computational studies, path a shown in Fig. 7b is believed to be a favorable way in generating product\n \n C\n \n 1\n \n \n .\n
\n\n As for the formation of requisite intermediate\n \n int-9\n \n , it involves four main steps (Fig. 8b): the\n \n \u03b2\n \n -addition of\n \n int-6\n \n to HCHO (\n \n int-6\n \n \u2192\n \n int-6'\n \n ), proton transfer from the methanol (\n \n int-6'\n \n \u2192\n \n int-7\n \n ), Mg(OMe)\n \n 2\n \n -induced proton abstraction and dissociation of OH\n \n -\n \n (\n \n int-7\n \n \u2192\n \n int-7'\n \n \u2192\n \n int-8\n \n ), and hydride transfer (\n \n int-8\n \n \u2192\n \n int-9\n \n ). Noteworthy, the formation of\n \n int-8\n \n clearly proceeds under the assistance of Mg\n \n 2+\n \n , and the hydride transfer (\n \n int-8\n \n \u2192\n \n int-9\n \n ) by the [RuHCl] complex requires to overcome an energy barrier of 11.6 kcal\u00b7mol\n \n -1\n \n (\n \n TS11\n \n ), and the reaction is endothermic by 6.1 kcal mol\n \n -1\n \n . In comparison with this process, other parts can easily take place with a maximum barrier of 8.8 kcal mol\n \n -1\n \n (\n \n TS8\n \n ). Once\n \n int-9\n \n has formed, the formation of\n \n D\n \n 1\n \n \n from\n \n int-9\n \n undergoes a similar way of\n \n C\n \n 1\n \n \n generation from\n \n int-3\n \n (Fig. 8a):\n \n \u03b2\n \n -addition of\n \n int-9\n \n to\n \n B\n \n 1\n \n '\n \n , intramolecular cyclization via C1-C5 bond formation, and based-promoted deprotonation to yield\n \n \n product\n \n D\n \n 1\n \n \n . The calculations show that the steps from\n \n int-9\n \n to\n \n D\n \n 1\n \n \n have a slightly higher barrier than the corresponding transition state from\n \n int-3\n \n to\n \n C\n \n 1\n \n \n (17.5 kcal\u00b7mol\n \n -1\n \n for\n \n TS13\n \n \n vs\n \n 15.5 kcal\u00b7mol\n \n -1\n \n for\n \n TS6\n \n ).\n
\n\n In summary, by a strategy incorporating a tandem coupling sequence into the reduction of azaarenium salts, we have developed an unprecedented intermolecular\n \n syn\n \n -diastereoselective annulation reaction by reductive ruthenium(II) catalysis. A variety of azaarenes were efficiently transformed in combination with a large variety of aniline derivatives into fused N-heterocycles by employing paraformaldehyde as both a crucial agent to generate active ruthenium(II)-hydride species and a C1-building block, proceeding with readily available feedstocks, excellent selectivity, mild conditions, and broad substrate and functional group compatibility. The present work has established a practical platform for the transformation of ubiquitously distributed but weakly reactive azaarenes into functional organic frameworks that are difficult accessible with the existing approaches, and further discovery of bioactive and drug-relevant molecules due to the promising potentials of the obtained compounds featuring the teterahydroquinolyl and hexahydro-1,6-naphthyridyl motifs. Mechanistic studies reveal that the products are formed via hydride transfer-initiated\n \n \u03b2\n \n -aminomethylation and\n \n \u03b1\n \n -arylation of the azaarenium salts, and the use of Mg(OMe)\n \n 2\n \n as a base plays a critical role in determining the reaction chemo-selectivity by lowering the hydrogen transfer rate. The work presented fills an important gap in the capabilities of utilizing azaarenes as the synthons, and opens a door to further develop valuable reductive functionalization of inert unsaturated systems by taking profit of formaldehyde-endowed two functions.\n
\n\n \n Typical procedure I for the synthesis of product C\n \n 1\n \n \n
\n\n Under N\n \n 2\n \n atmosphere,\u00a0[Ru(\n \n p\n \n -cymene)Cl\n \n 2\n \n ]\n \n 2\n \n (1 mol %), 1-benzylquinolin-1-ium bromide\n \n A\n \n 1\n \n \n (0.2 mmol), N-ethylaniline\n \n B\n \n 1\n \n \n (0.2 mmol), Mg(OMe)\n \n 2\n \n (0.75 eq, 12.9 mg)\uff0c(CH\n \n 2\n \n O)\n \n n\n \n (10.0 eq , 60 mg) and methanol (1 mL)\u00a0were introduced in a Schlenk tube, successively.\u00a0Then the Schlenk tube was closed and the resulting mixture was stirred at 55\n \n o\n \n C for 18 h. After cooling down to room temperature, the mixture was extracted with ethyl acetate, washed with 5% Na\n \n 2\n \n CO\n \n 3\n \n solution, dried with anhydrous sodium sulfate, and then concentrated by removing the solvent under vacuum. Finally, the residue was purified by preparative TLC on silica to give 12-benzyl-5-ethyl-5,6,6a,7,12,12a-hexahydrodibenzo[\n \n b,h\n \n ][1,6]naphthyridine\n \n C\n \n 1\n \n \n .\n
\n\n \n Data availability\n \n
\n\n The authors declare that all relevant data supporting the findings of this study are available within the paper and its supplementary information files.\n
\n\n Table 1 is in the supplementary files section.\n
\n\n Table 1\n
\n \n\n Supplementary Information-data\n
\n \n\n Supplementary Information-NMR spectra\n
\n \n\n Epigenetic reprogramming occurs during reproduction to reset the genome for early development. In flowering plants, mechanistic details of parental methylation remodeling in zygote remain elusive. Analysis of allelic-specific DNA methylation in rice hybrid zygotes and during early embryo development indicates that paternal DNA methylation is predominantly remodeled to match maternal allelic levels upon fertilization, which persists after the first zygotic division. The DMA methylation remodeling pattern supports the predominantly maternal-biased gene expression during zygotic genome activation (ZGA) in rice. However, parental allelic-specific methylations are reestablished at the globular embryo stage and associate with allelic-specific histone modification patterns in hybrids. These results reveal a maternal-controlled paternal DNA methylation remodeling pattern for zygotic genome reprograming and suggest existence of a chromatin memory allowing parental allelic-specific methylation to be maintained in the hybrid.\n
\n\n The zygotic transition, from a fertilized egg to an embryo, is central to animal and plant reproduction. In animals, embryo development depends on maternally provided factors until zygotic genome activation (ZGA) that takes place after one to several cell divisions depending on the species\n \n \n 1\n \n \n . In flowering plants, ZGA is rapidly initiated and occurs before the first zygotic division\n \n \n 2\n \n \u2013\n \n 5\n \n \n . Studies in plants have found that a large number of genes are expressed\n \n de novo\n \n which are required for zygotic division\n \n \n 6\n \n \u2013\n \n 9\n \n \n . However, parental contribution to the zygotic transcriptome in plants is under debate\n \n \n 5\n \n ,\n \n 10\n \n \n . Studies in Arabidopsis revealed that maternal transcripts dominated the transcriptomes of embryos at the 2\u20134 cell and globular stages\n \n \n 11\n \n \n , whereas other results indicated that maternal and paternal genomes contributed equally to the transcriptome of early embryos, even at the 1\u20132 cell stage or elongated zygotes\n \n \n 6\n \n ,\n \n 7\n \n ,\n \n 12\n \n \n . In rice, analysis of allele-specific transcriptome in the zygote revealed that the transcription of the zygotic genome is mainly from the maternal alleles, which results in a maternally dominated transcriptome\n \n \n 9\n \n \n . ZGA is a gradual process that relies on large-scale chromatin reprogramming leading to an increasing number of zygotically expressed genes\n \n \n 1\n \n \n , which may involve crosstalk between the parental epigenomes to control zygote and early development.\n
\n\n In mammals, it is generally assumed that two distinct phases of epigenetic reprogramming serve to prevent inheritance of ancestral epigenetic signatures. This reprogramming process comprises the erasure of DNA methylation marks from the previous generation followed by a reestablishment of DNA methylation\n \n \n 1\n \n 3\n \n . Unlike in mammals, plant DNA methylation is found to be only partially remodeled or reconfigured in the gametes and the unicellular zygote\n \n \n 14\n \n \u2013\n \n 1\n \n 7\n \n . The partial epigenetic reprogramming of DNA methylation may contribute to stable epigenetic inheritance relatively frequently observed in plants\n \n \n 1\n \n 3\n \n . In the meantime, the DNA methylation remodeling is also essential for plant reproduction, as perturbation of DNA methylation by mutation of DNA demethylase genes affected function of the gametes and impaired the development of zygote and embryo as well as endosperm in rice\n \n \n 17\n \n \u2013\n \n 1\n \n 9\n \n . In plants, knowledge on epigenetic basis and dynamics of the parental contributions during fertilization and early embryogenesis is limited, despite its importance in understanding epigenetic inheritance and the effects of parental genome interactions in the context of nonself pollination in plants.\n
\n\n To investigate the parental epigenome dynamics in the zygote, we first analyzed the egg, sperm and zygote (at 6.5 hours after pollination, HAP, after the gamete nuclear fusion\n \n \n 9\n \n \n ) DNA methylation patterns of elite hybrid rice \u201cSY63\u201d parental lines (MH63 and ZS97), using a bisulfite sequencing (BS-seq) protocol developed for small numbers of cells\n \n \n 17\n \n ,\n \n 20\n \n ,\n \n 21\n \n \n . DNA methylomes data were obtained from 25 eggs or zygotes and 150 sperm cells, two biological replicates were performed with a sequencing depth of about 24.7\u201375.4 \u00d7 genome coverage (\n \n Supplementary Table\u00a01\n \n ). Principal component analysis revealed a high reproducibility of the replicates, a cell-type-specific distribution pattern (with the sperm methylome more distal from that of egg or zygote), and a clear difference between the two parental lines (\n \n Extended Data\n \n Fig.\n \n 1\n \n a). Boxplots indicated that sperm cells showed globally higher CG methylation (mCG) but lower CHG methylation (mCHG) than egg cells (\n \n Extended Data\n \n Fig.\n \n 2\n \n a). Unlike in Arabidopsis sperm where CHH methylation (mCHH) is lost\n \n \n 14\n \n \n , the rice sperm mCHH was higher than the egg level (\n \n Extended Data\n \n Fig.\n \n 1\n \n b, c;\n \n Extended Data\n \n Fig.\n \n 2\n \n a), which may be due to a different landscape and higher levels of mCHH in the rice genome\n \n \n 22\n \n ,\n \n 23\n \n \n . In the zygote mCG and mCHH levels were lower than in the sperm, while the mCHG was at the intermediate levels of the egg and sperm cells (\n \n Extended Data\n \n Fig.\n \n 2\n \n a). Density plots revealed higher methylation variations between zygote and sperm than between zygote and egg (\n \n Extended Data\n \n Fig.\n \n 2\n \n b). The analysis confirmed that the parental DNA methylation was rapidly remodeled upon fertilization in rice\n \n \n 17\n \n \n , and suggested a predominant remodeling of the male methylome in the zygote. Scanning differentially methylated regions (DMRs, within 50-bp windows with the cutoff of methylation difference at CG\u2009>\u20090.5, CHG\u2009>\u20090.3, and CHH\u2009>\u20090.1,\n \n P\n \n <\u20090.05) between the gametes and zygotes revealed that more than half of the DMRs concerned non-transposable element (non-TE) regions (\n \n Extended Data\n \n Fig.\n \n 2\n \n c). Comparisons between MH63 egg and ZS97 sperm or between ZS97 egg and MH63 sperm, as used in the reciprocal crosses, revealed higher DNA methylation variations than between egg and sperm within the inbred lines (\n \n Extended Data\n \n Fig.\n \n 1\n \n c, d).\n
\n\n
\n\n
\n\n Next, we analyzed the methylomes of the reciprocal hybrid zygotes (at 6.5 HAP) and globular embryos (GE, at 72 HAP) of SY63 (ZS97 as female, MH63 as male) and MZ (MH63 as female, ZS97 as male) (\n \n Extended Data\n \n Fig.\n \n 1\n \n a;\n \n Supplementary Table\u00a01\n \n ). In the hybrid zygotes, the methylation levels appeared higher than in the male and female gametes, particularly at CG and CHG sites (Fig.\n \n 1\n \n ), consistent with previous observations of enhanced DNA methylation in hybrid vegetative tissues\n \n \n 24\n \n ,\n \n 25\n \n \n . Although such overall reinforcement was not observed in the inbred zygotes (\n \n Extended Data\n \n Fig.\n \n 2\n \n a), substantial portions (about 30\u201366%) of CG and CHG hyper DMRs of the inbred zygote versus sperm (Z \u2013 S) or egg (Z \u2013 E) (\n \n Extended Data\n \n Fig.\n \n 2\n \n c) overlapped with those found in the reciprocal hybrid zygotes (\n \n Extended Data\n \n Fig.\n \n 3\n \n a, b), suggesting that DNA methylation at a number of specific loci (\n \n Extended Data\n \n Fig.\n \n 3\n \n c;\n \n Supplementary Dataset 1\n \n ) tended to be reinforced upon fertilization. In the hybrid globular embryos (GE), genic methylation levels were maintained or even augmented compared to the zygote levels, while TE methylation (mainly mCG and mCHG) was lower than in the zygote but close to the gametes or seedling levels (Fig.\n \n 1\n \n a, b)\n \n \n 26\n \n \n , indicating that DNA methylation continued to be remodeled during early embryogenesis.\n
\n\n
\n\n To follow up the egg versus sperm (E \u2013 S) DMRs in the zygote, we analyzed their methylation levels in both the inbred and hybrid zygotes. In the hybrid zygotes the overall methylation levels of the CG and CHG DMRs were close to the egg levels, while those of the CHH DMRs paralleled the lower parental levels (Fig.\n \n 2\n \n a, b). Similar profiles were also observed in the inbred zygotes (\n \n Extended Data\n \n Fig.\n \n 4\n \n ). To confirm the observation, we separated the parental allele-specific reads from the hybrid zygote BS-seq data by using the 1,351,242 single nucleotide polymorphisms (SNPs) between MH63 and ZS97 genomes\n \n \n 27\n \n \n . The allele-specific methylation reads in the hybrid zygotes were about 10.6%-14.1% of the total reads, similar to those observed in DNA methylomes of hybrid rice vegetative tissues\n \n \n 26\n \n ,\n \n 28\n \n \n . In the hybrid zygotes, the methylation levels of both the maternal and paternal alleles of the E \u2013 S CG and CHG DMRs (both hypo and hyper) were close to the egg but distinct from the sperm levels (Fig.\n \n 2\n \n a-b, d-e). These observations suggested that the paternal alleles of the E \u2013 S CG and CHG DMRs were remodeled to match the methylation levels of the maternal alleles in the zygote. To further confirm the results, we crossed the ZH11 variety with MH63 and obtained methylation data from 2-cell embryos (harvested at 12 HAP) (\n \n Supplementary Table\u00a01\n \n ). Analysis of parental allele-specific methylation in the 2-cell embryos by using the SNPs between the MH63 and ZH11 genomes\n \n \n 29\n \n \n , obtained a similar result (Fig.\n \n 2\n \n c), indicating that the maternal-controlled remodeling of paternal allele-specific methylation in the zygote persisted till at least the 2-cell embryo stage.\n
\n\n
\n\n To investigate whether the zygotic remodeling of paternal methylation was maintained during embryogenesis, we analyzed the methylation levels of the E \u2013 S DMRs in the GEs of the reciprocal crosses. In the GEs, methylations of the CG and CHG DMRs were at the intermediate levels of the gametes (Fig.\n \n 3\n \n \n )\n \n . However, the levels of CHH DMRs remained to parallel the lower parental levels (Fig.\n \n 3\n \n \n )\n \n , consistent the observation of mCHH loss during embryogenesis in Arabidopsis\n \n \n 30\n \n \n . However, the paternal allelic methylations of the CG and CHG DMRs were close to the sperm levels, whereas those of maternal alleles were close to the egg levels (Fig.\n \n 3\n \n ), suggesting that the parental allelic-specific methylation, which had been observed in seedling and panicle tissues of SY63 and MZ\n \n \n 26\n \n \n , and detected in the reciprocal hybrids between NIP and 9311 varieties (\n \n Extended Data\n \n Fig.\n \n 5\n \n ), were reestablished in the GE (Fig.\n \n 3\n \n ;\n \n Extended Data Fig.\u00a06c, d\n \n ).\n
\n\n
\n\n To study whether the reestablishment of parental allelic-specific DNA methylation in the hybrid embryos was related to specific chromatin signatures, we analyzed histone modification marks including H3K27ac, H3K4me3 and H3K9me2 in the E (ZS97) \u2013 S (MH63) DMRs using the ChIP-seq data obtained from MH63 and ZS97 seedling tissues\n \n \n 31\n \n \n . In the CG and CHG hyper DMRs, the active histone marks H3K27ac and H3K4me3 were absent from ZS97, but present at very high levels in MH63 alleles. By contrast, the H3K9me2 (a repressive mark that tightly associates with mCG and mCHG in plants) levels of the DMRs were high in ZS97, but absent from MH63 alleles (Fig.\n \n 4\n \n a, b). In the hypo DMRs, opposite histone modification profiles were observed (Fig.\n \n 4\n \n a, b). Similar observations were made for the E (MH63) \u2013 S (ZS97) DMRs (\n \n Extended Data Fig.\u00a06a, b\n \n ). Thus, methylation differences between the male and female gametes appeared to associate with distinct histone marks in vegetative tissues of the respective parental lines. To study whether the association could be detected in the hybrid cells, we performed H3K4me3 ChIP-seq of the hybrid SY63 seedling tissues, and analyzed the parental allele-specific H3K4me3 by using SNPs between MH63 and ZS97. The analysis revealed that, in E (ZS97) \u2013 S (MH63) hyper DMRs, H3K4me3 was depleted from the maternal (ZS97), but present at very high levels in the paternal (MH63) alleles. In the hypo DMRs, a reverse situation was observed (Fig.\n \n 4\n \n c;\n \n Extended Data Fig.\u00a06c, d\n \n ). Analysis of chromatin modification data of the reciprocal hybrids between NIP and 9311 varieties\n \n \n 32\n \n \n , revealed a similar result (\n \n Extended Data Fig.\u00a07\n \n ). Together, these data indicated that parental allelic-specific methylation associated with parental allelic-specific histone marks, which may be underlying the reestablishment of parental allelic-specific DNA methylations during early embryogenesis and maintenance during development.\n
\n\n To study whether the parental methylation remodeling pattern was associated with gene expression in the zygote, using RNA-seq we analyzed transcriptomes of sperm, egg, zygote (6.5 HAP) and GE (72 HAP) of the reciprocal crosses between MH63 and ZS97 (\n \n Supplementary Table\u00a02\n \n ), with 3 biological replicates (r\u2009=\u20090.94\u2009~\u20091.0) (\n \n Extended Data Fig.\u00a08a\n \n ). Principal component analysis indicated that the sperm transcriptomes were distal from those of egg, zygote, and GEs (\n \n Extended Data Fig.\u00a08b\n \n ). Comparison of the hybrid zygotes with the respective egg cells revealed more than 2000 up- and downregulated genes (|log2 (Fold Change)| > 1, Q-value\u2009<\u20090.01) in the reciprocal hybrid zygotes, among which 601 genes were commonly up-regulated (\n \n Extended Data Fig.\u00a08c, d\n \n ). These genes are enriched in DNA replication, ethylene signaling, mitotic cell cycle, and calcium signaling (\n \n Extended Data Fig.\u00a08d\n \n ), and showed overlaps with previously reported zygotic transcriptomes of different rice varieties (\n \n Extended Data Fig.\u00a09a\n \n )\n \n \n 9\n \n ,\n \n 17\n \n ,\n \n 33\n \n \n . These genes displayed higher transcription levels in zygote than egg in the different rice varieties and could be clustered based on their expression in egg or sperm cells (\n \n Extended Data Fig.\u00a09b)\n \n . Many were previously reported to associate with ZGA, including\n \n WUSCHEL-related homeobox 5\n \n (\n \n WOX5\n \n ),\n \n MINICHROMOSOME MAINTENANCE 6\n \n (\n \n MCM6\n \n ),\n \n MCM7/10\n \n ,\n \n CYCB2;2\n \n ,\n \n Kip-related proteins 1\n \n (\n \n KRP1\n \n ),\n \n Rapid alkalinization factor 3\n \n (\n \n RALF3\n \n ), and\n \n Anaphase-Promoting Complex 10\n \n (\n \n APC10\n \n )\n \n 9,33\u221236\n \n . In addition, DNA replication such as\n \n POLA4\n \n ,\n \n POLD1/4, OsRPA1/3\n \n (\n \n Replication protein A\n \n ) and 17 histone encoding genes were found in the rice hybrid zygotes (\n \n Extended Data Fig.\u00a09b; Supplementary Dataset 2\n \n ).\n
\n\n To study the parental contribution to the zygotic gene expression, we analyzed parental allelic-specific reads from the reciprocal hybrid zygote transcriptomes and found that most of the reads were maternal allelic-specific and about 1.5%~4.1% of the reads was paternal allelic-specific (\n \n Extended Data Fig.\u00a010a\n \n ). This was consistent with previous results that in rice ZGA occurs in the zygote, with unequal parental contribution where most genes are expressed primarily from the maternal genome\n \n \n 9\n \n \n . This corroborated with the observation that paternal DNA methylation was remodeled to match the maternal levels. However, egg-produced mRNAs might persist in the early zygote, as observed in Arabidopsis\n \n \n 6\n \n ,\n \n 37\n \n \n . From the genes with allelic-specific expression in the reciprocal hybrid zygotes, we identified 1063 and 28 genes as maternal- and paternal specifically expressed genes (Fig.\n \n 5\n \n a\n \n )\n \n , indicating that gene imprinting occurred in the rice zygote. Among the 28 paternal specifically expressed genes (PEGs) in the zygote, only one was found as endosperm-expressed PEGs in rice\n \n \n 38\n \n \n , indicating a different gene imprinting program between zygote and endosperm in rice. Most of the 28 zygotic PEGs were already highly expressed in the sperm (Fig.\n \n 5\n \n b). Several genes such as plasma membrane protein gene\n \n GEX1\n \n , RALF-like secreted peptide RALF3, and\n \n Arabinogalactan protein 7\n \n (\n \n AGP7\n \n ) were shown to function in male gametophyte development and during early embryogenesis\n \n \n 39\n \n \u2013\n \n 41\n \n \n . Nearly all (26/28) of the PEGs showed a low expression in the egg cells (Fig.\n \n 5\n \n b), nine of which hypo DNA methylation in the sperm cells or at the paternal alleles in the zygotes (Fig.\n \n 5\n \n c), suggesting that PEG might have escaped the zygotic remodeling, as observed in mammals\n \n \n 42\n \n \n . The maternal alleles of the PEGs could be repressed by other chromatin signatures, such as PRC2-H3K27me3 in Arabidopsis\n \n \n 43\n \n \n . Most of the 1063 maternal-specifically expressed genes (MEGs) showed expression in the egg cells (\n \n Extended Data Fig.\u00a010b\n \n ), and displayed lower mCHH in egg than in sperm (\n \n Extended Data Fig.\u00a010c, d\n \n ), consistent with the overall higher mCHH in sperm than egg (\n \n Extended Data\n \n Fig.\n \n 1\n \n b-d). In the zygotes, the paternal alleles of the MEGs also showed higher mCHH than the maternal alleles (\n \n Extended Data Fig.\u00a010e\n \n ), suggesting that mCHH may be involved in the repression of paternal alleles that likely had also escaped the remodeling process in the zygote.\n
\n\n Analysis of parental allelic-specific reads from the hybrid GE transcriptomes revealed comparable numbers of genes with maternal and paternal allelic-specific expression (\n \n Extended Data Fig.\u00a011a, b\n \n ), indicating an increased paternal contribution to gene expression in GE, as observation in Arabidopsis\n \n \n 11\n \n \n , which was consistent with the reestablishment of the parental allelic-specific DNA methylome in GE. Analysis of monoallelic gene expression in the reciprocal hybrid GEs identified 102 PEGs and 350 MEGs (\n \n Extended Data Fig.\u00a011c\n \n ), suggesting that gene imprinting persisted till the globular embryo stage, which was, however, not detected in the rice mature embryos\n \n \n 44\n \n \n . The GE imprinted genes were different from those detected in the zygote, except 10% of the zygotic MEGs were remained in GE. Interesting, 36 zygotic MEGs became PEG in GE (\n \n Extended Data Fig.\u00a011d\n \n ). The data supported a dynamic reprogramming of gene expression during embryogenesis.\n
\n\n Epigenetic reprograming is essential for gametogenesis and zygotic development. Sperm cell chromatin is highly condensed but becomes loose after fusion with the egg nucleus allowing transcription of the paternal genome to initiate in the zygote. The predominant paternal DNA methylation remodeling in the zygote may be part of the process. The finding that DNA methylation at a specific set of loci was enhanced in both inbred and hybrid zygotes relative to the gametes, suggests that the parental methylation remodeling is non-stochastic. The finding that maternal methylation-based remodeling of paternal alleles in the rice zygote and in 2-cell embryos corroborates with the model that maternal epigenetic pathways control paternal contributions to early embryogenesis in Arabidopsis\n \n \n 11\n \n \n , but contrasts with the findings in zebrafish that after fertilization the maternal genome is reprogrammed to match the paternal methylation pattern that is inherited during early embryogenesis\n \n \n 45\n \n ,\n \n 46\n \n \n .\n
\n\n Although the mechanistic details are unclear, maternal epigenetic information and/or regulators inherited from the egg cell may be involved in the process. This is supported by the observations in Arabidopsis that paternal alleles are initially downregulated by the maternal histone H3K9me2 methyltransferase KYP and DNA methyltransferases CMT3 and DRM2\n \n 11\n \n . Genes of these enzymes as well as other methylation regulators were found to be expressed at high levels in the rice egg and zygote\n \n \n 47\n \n \n (\n \n Extended Data Fig.\u00a011e, f\n \n ). The maternal-based remodeling of paternal allelic DNA methylation shown in this work likely associated with the zygotic transition to which parental genomes unequally contribute, with most genes expressed primarily from the maternal genome\n \n \n 9\n \n ,\n \n 37\n \n \n (\n \n Extended Data Fig.\u00a010a;\n \n Fig.\n \n 5\n \n a).\n
\n\n The reestablishment of paternal allelic-specific methylation observed in rice globular embryos is reminiscent of the re-methylation process in post-implantation embryos in mammals\n \n \n 48\n \n \n , and suggests existence of a memory for parental allelic-specific methylation. Possibly, interplay between parental allelic-specific DNA methylation and histone modifications, which may depend on the associate DNA sequences, may elicit a chromatin memory that facilitates the reestablishment and/or maintenance of parental allelic-specific epigenetic signatures in the next generation. This, together with the partial DNA methylation remodeling in the gametes and zygote which may spare not only imprinted genes but also other loci, would facilitate transgenerational inheritance of inherent and acquired epigenetic information in plants. Elucidation of mechanisms underlying the setup of parental allelic-specific methylation memory during plant embryogenesis and its maintenance during development would lead to new strategies for crop improvement.\n
\n\n \n Rice sperm, egg cell, zygote and GEs isolation\n \n
\n\n Rice (\n \n Oryza sativa\n \n spp.) varieties Zhenshan 97 (ZS97,\n \n indica/xian\n \n ), Minghui 63 (MH63,\n \n indica/xian\n \n ), and Zhonghua 11 (ZH11,\n \n japonica/geng\n \n ) were used in this study. The three inbred lines were grown in paddy field under normal agricultural conditions in Wuhan, China. To collect hybrid and/or isogenic zygotes, the female lines were hand emasculated and pollinated with the indicated male lines\u2019 pollen, and the hybrid zygotes were isolated at 6.5 HAP (for unicellular zygote) and 12 HAP (for two-cell stage zygote), the GEs were isolated at 72 HAP. Egg cell and zygote were isolated from ovaries of rice as previously reported\n \n 17,20\n \n . Briefly, ovaries of unpollinated and pollinated florets were manually dissected under dissection microscope. Then the dissociated ovule was transferred into 0.53 M mannitol solution (Sigma) and broken to release egg cell or zygote. The isolated cells were stained with fluorescein diacetate (Invitrogen, Cat. # F1303) and collected by a micromanipulator system (Eppendorf, TransferMan 4r). Twenty-five egg or zygote\u00a0cells were pooled for each replicate for BS-seq or RNA-seq libraries, each cell-type or each genotype with three biological replicates. Sperm cells were collected as previously reported method\n \n 49,50\n \n with minor modifications. Briefly, about 30 anthers were collected from mature florets before anthesis in a plastic dishes with 3 mL of 12% sucrose, then broken with forceps to release pollen. Sperm cells were released by gentle shaking for 30 min and filtered through 20 \u03bcm and then 10 \u03bcm nylon bolting clothes. Subsequent steps were performed as described\n \n 49,50\n \n .\n
\n\n \n RNA-seq and BS-seq library construction and sequencing\n \n
\n\n For RNA-seq library construction, mRNAs were extracted from the collected rice gamete and zygote cells, then reverse transcribed and amplified by using a Single Cell Full Length mRNA Amplification Kit (Vazyme, Cat.# N712) according to manufacturer\u2019s instruction. cDNAs were purified with VAHTS DNA Clean Beads (Vazyme, Cat.# N411) and fragmented into 200~500 bp lengths, then used for PCR amplification, adapter/index ligation, and DNA purification with a TruePrep\u00ae DNA Library Prep Kit V2 for Illumina (Vazyme, Cat.# TD502). BS-seq libraries were constructed using a previously reported protocol\n \n 21\n \n with modified primer adapter 2 oligos and iPCRtag primers\n \n 17\n \n . RNA-seq and BS-seq libraries were sequenced by an Illumina NovaSeq 6000 platform (Annoroad Gene Technology, China)\u00a0with the PE150 (paired-end 150 nucleotides) method.\n
\n\n \n RNA-seq data analysis\n \n
\n\n RNA-seq raw reads were filtered by fastp\n \n 51\n \n (v.0.20.1) to remove low-quality reads and adapter. Clean reads were aligned to the MH63 reference genome (MH63RS3, Rice Information GateWay [RIGW],\n \n http://rice.hzau.edu.cn/rice_rs3/\n \n ) by HISAT2\n \n 52\n \n (v.2.2.1). To improve alignment of ZS97 RNA-seq data, a pseudogenome was constructed by using the MH63 genome as backbone and replacing the SNPs (between MH63 and ZS97) with ZS97 genotype to map ZS97 sequencing reads. The unique mapping reads were retained for further analysis. StringTie\n \n 53\n \n (v.2.1.4) was used for transcripts assembly and gene quantitation. DESeq2\n \n 54\n \n package was used for gene differential expression analysis. Genes with TPM (transcripts per million) \u2265 1 (at least in one sample\u00a0in the comparisons) and with |log2 (fold change)| \u2265 2 and adjusted\n \n P\n \n < 0.01 were considered as differentially expressed genes (DEGs).\n
\n\n For allele-specific expression (ASE) analysis of the hybrids, the SNPs between MH63 and ZS97 were masked with N by using SNPsplit\n \n 55\n \n (v.0.3.4). Clean reads were aligned on the N-masked MH63 genome by HISAT2 (v.2.2.1) and the unique mapping reads were retained. The parental allele-specific reads were separated from the hybrids data by using SNPsplit program. The separated reads were normalized for allelic-specific expression level calculation. Allele-specific expression genes were identified with the cut-offs |log2 (fold change)|\u00a0> 1 and adjusted\n \n P\n \n < 0.01 between two parental alleles by DESeq2 package.\n
\n\n \n BS-seq data analysis\n \n
\n\n BS-seq low-quality reads were filtered out from the raw data by Trim_Galore (v.0.6.6; http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). Clean reads were aligned on the MH63 genome by Bismark (v.0.23.1)\n \n 56\n \n using default parameters. ZS97 BS-seq reads were aligned on the SNP-N-masked MH63 genome. Unique mapping reads were retained for further analysis. PCR duplications were removed by command of deduplicate_bismark and DNA methylation sites were extracted by command of bismark_methylation_extractor from Bismark software (v.0.23.1). Individual cytosines with more than three reads were retained for DNA methylation level calculation.\n
\n\n For allele-specific methylation analysis, the SNPs between MH63 and ZS97 were masked with N by SNPsplit (v.0.3.4). The cleaned high-quality reads were mapped to the N-masked MH63 genome by Bismark. After removing duplications, the allele-specific reads were separated from the hybrids by the SNPsplit. Individual cytosines that were covered by at least three allele-specific reads were considered for allele-specific methylation level calculation.\n
\n\n To identify differential methylated regions (DMRs), the whole genome was divided into 50-bp bins. Bins that contained at least five cytosines each and every cytosine with at least a three-fold coverage were retained. Bins with methylation differences greater than 0.5, 0.3, and 0.1 respectively at CG, CHG, and CHH contexts with false discovery rate (FDR) < 0.05 between comparisons were considered as DMRs. The FDR was generated from an adjusted\n \n P\n \n -value (Fisher\u2019s exact test) using the Benjamini-Hochberg method.\n
\n\n Density plots were generated by comparing the average cytosine methylation levels within 50 bp bins between two samples. Only the bins contained at least 20 informative sequenced cytosines (i. e., the sum of the sequence depth of each cytosine multiplied by the number of cytosines within 50 bp bins in the CG, CHG, or CHH context) in both samples and 0.5 CG, 0.3 CHG, or 0.1 CHH methylation ratios in either sample were retained as previously described\n \n 17,18\n \n . The frequency distribution of fractional methylation differences between comparisons was shown by density plots.\n
\n\n \n Chromatin immunoprecipitation-sequencing (ChIP-seq) and data analysis\n \n
\n\n Chromatin immunoprecipitated experiments were conducted as previously described\n \n 57\n \n . Briefly, about 2 g of rice seedling leaves were crosslinked by 1% (v/v) formaldehyde for 30 min and used for chromatin extraction. Chromatin was fragmented to around 200 bp by sonication using a Bioruptor Plus System (Diagenode), and then incubated with antibody-conjugated beads (anti-H3K4me3, Abcam Cat.# ab8580) overnight. After washing three times, immunoprecipitated chromatin was de-crosslinked and DNA was purified, non-precipitated chromatin was used as input. DNA isolated from chromatin immunoprecipitation was used for sequencing libraries construction according to the protocol of Illumina TruSeq ChIP Sample Prep Set A and sequenced on Illumina HiSeq2500 platform.\n
\n\n Fastp (v.0.20.1) was used for remove low-quality reads and adapter from the ChIP-seq raw data. Clean reads were mapped to the MH63RS3 genome by Bowtie2 (v.2.2.8). ZS97 sequencing reads were mapped to the SNP-N-masked MH63 genome. Duplications were removed using Picard (v.2.1.1). The bigwig files were generated by using a command of bamCoverage from deepTools (v.3.3.0). The ChIP-seq data of the hybrids were aligned to the SNP-N-masked MH63 genome by Bowtie2 (v.2.2.8). SNPsplit (v.0.3.4) software was used to separate the parental allele-specific modification reads.\n
\n\n 51\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Abiko, M.\n \n et al.\n \n Identification of proteins enriched in rice egg or sperm cells by single-cell proteomics.\n \n PLoS One\n \n \n 8\n \n , e69578, doi:10.1371/journal.pone.0069578 (2013).\n
\n\n 52\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Li, C., Xu, H., Russell, S. D. & Sundaresan, V. Step-by-step protocols for rice gamete isolation.\n \n Plant Reprod\n \n \n 32\n \n , 5-13, doi:10.1007/s00497-019-00363-y (2019).\n
\n\n 53\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Chen, S., Zhou, Y., Chen, Y. & Gu, J. fastp: an ultra-fast all-in-one FASTQ preprocessor.\n \n Bioinformatics\n \n \n 34\n \n , i884-i890, doi:10.1093/bioinformatics/bty560 (2018).\n
\n\n 54\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype.\n \n Nat Biotechnol\n \n \n 37\n \n , 907-915, doi:10.1038/s41587-019-0201-4 (2019).\n
\n\n 55\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Pertea, M.\n \n et al.\n \n StringTie enables improved reconstruction of a transcriptome from RNA-seq reads.\n \n Nat Biotechnol\n \n \n 33\n \n , 290-295, doi:10.1038/nbt.3122 (2015).\n
\n\n 56\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2.\n \n Genome Biol\n \n \n 15\n \n , 550, doi:10.1186/s13059-014-0550-8 (2014).\n
\n\n 57\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Krueger, F. & Andrews, S. R. SNPsplit: Allele-specific splitting of alignments between genomes with known SNP genotypes.\n \n F1000Res\n \n \n 5\n \n , 1479, doi:10.12688/f1000research.9037.2 (2016).\n
\n\n 58\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Krueger, F. & Andrews, S. R. Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications.\n \n Bioinformatics\n \n \n 27\n \n , 1571-1572, doi:10.1093/bioinformatics/btr167 (2011).\n
\n\n 59\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Ma, X.\n \n et al.\n \n An enhanced network of energy metabolism, lysine acetylation, and growth-promoting protein accumulation is associated with heterosis in elite hybrid rice.\n \n Plant Commun\n \n , 100560, doi:10.1016/j.xplc.2023.100560 (2023).\n
\n\n Supplementary Datasets are not available with this version\n
\n\n \n Supplementary Dataset 1.\n \n The differentially methylated genes (DMGs) commonly found in inbred and hybrid zygotes relative to gametes.\n
\n\n \n Supplementary Dataset 2.\n \n The up-regulated and downregulated genes commonly detected in the reciprocal hybrid zygotes.\n
\n\n Supplementary Table 1-2\n
\n \n\n Extended Data Fig. 1. DNA methylomes of inbred and/or hybrid rice gametes, zygotes, and globular embryos.\nExtended Data Fig. 2. DNA methylation in the gametes and zygotes of the hybrid parental lines.\nExtended Data Fig. 3. Conserved DMRs between inbred and hybrid zygotes relative to the gametes.\nExtended Data Fig. 4. Zygotic methylation levels of the egg versus sperm DMRs in MH63 and ZS97.\nExtended Data Fig. 5. Maintenance of parental allelic-specific methylations in the reciprocal hybrids of NIP and 9311.\nExtended Data Fig. 6. Parental allelic-specific methylations associated with specific histone marks in the parental lines.\nExtended Data Fig. 7. H3K4me3 levels of the DMRs between NIP and 9311 in the parental lines and the reciprocal hybrids.\nExtended Data Fig. 8. Transcriptomic analysis of the reciprocal hybrid zygotes.\nExtended Data Fig. 9. Analysis of hybrid zygote transcriptomes.\nExtended Data Fig. 10. DNA methylation levels of the maternal expressed genes (MEG) in the reciprocal hybrid zygotes.\nExtended Data Fig. 11. Identification of PEGs and MEGs in the reciprocal hybrid globular embryos.\n
\n \n\n Cellular oxidative stress resistance and bioactivities showed great significance for long-term survival and cardiac regeneration. Cardiosphere-derived cells (CDCs) are favorable cell sources for myocardial infarction (MI) therapy, but effective culture systems for CDC spheroids, cardiospheres (CSps), cultivation and cell function enhancement are not well established. Here, a liquid crystal substrate, octyl hydroxypropyl cellulose ester (OPC), was developed for CSps production and preconditioning. With unique surface properties and mechanical responsiveness, significantly more size-controllable CSps were acquired using OPC substrate, and the OPC-CSps showed improved cell bioactivities and oxidative stress resistance under the stimulation of mechanical-induced pyroptosis. RNA sequencing and metabolism analysis demonstrated the increased metabolic level and improved mitochondrial function of OPC-CSps. In a rat MI model, OPC-CSps significantly improved long-term cardiac function, promoted angiogenesis, and reduced cardiac remodeling in the 3-month observation. Collectively, this study provides a promising and effective system for preparing massive functional CSps for myocardial infarction therapy.\n
\n\n The 30-day mortality following myocardial infarction (MI) was 13.6% on average\n \n \n 1\n \n \n . When MI occurs, myocardial ischemia causes a series of irreversible pathological processes, such as severe inflammation, massive cell death, and cardiac fibrosis, which ultimately lead to heart failure\n \n \n 2\n \n ,\n \n 3\n \n \n . To date, many clinical and animal studies have shown cell-based therapies as promising approaches to reverse or slow MI disease progression\n \n \n 4\n \n ,\n \n 5\n \n ,\n \n 6\n \n ,\n \n 7\n \n \n .\n
\n\n To pursue satisfactory therapeutic outcomes, many effective cell processing methods have been extensively developed to improve cell bioactivities. Hanging drops, spinner flasks, and three-dimensional (3D) bioprinting have been used to prepare spheroids or organoids\n \n \n 8\n \n ,\n \n 9\n \n ,\n \n 10\n \n \n . By providing mechanical cues, extracellular matrix (ECM), and soluble factors in native niches, the 3D spheroids could promote pluripotency marker expression (\n \n Nanog\n \n ,\n \n Oct4\n \n ,\n \n Sox2\n \n ), cardiac lineage differentiation, paracrine secretion, anti-inflammation, and antisenescence\n \n \n 11\n \n ,\n \n 12\n \n ,\n \n 13\n \n ,\n \n 14\n \n \n . To further improve cell survival in hostile environments and their therapeutic potential, many researchers have suggested that simulating the inflammatory environment with preconditioning strategies could enhance cell resistance to adverse effects. Hypoxia and low-concentration inflammatory factor treatment are widely used preconditioning strategies\n \n \n 15\n \n ,\n \n 16\n \n \n . Following preconditioning treatment, the phenotype of pretreated cells shifted in therapeutically desirable directions, and their abilities to resist inflammation were greatly enhanced\n \n \n 17\n \n ,\n \n 18\n \n ,\n \n 19\n \n \n . These studies demonstrated the feasibility and effectiveness of preconditioning treatment, and they highlighted the significance of enhancing cellular bioactivities and inflammation resistance in damaged tissue regeneration.\n
\n\n Cardiosphere-derived cells (CDCs) are of endogenous cardiac origin\n \n \n 20\n \n \n and possess the ability to form 3D spherical clones, cardiospheres (CSps), in vitro\n \n \n 21\n \n \n . Compared to monolayer CDCs, CSps possessed improved growth factor secretion and cardiac regeneration potential, making them a good cell source for MI therapy\n \n \n 22\n \n ,\n \n 23\n \n \n . Previously, preconditioning CSps with pericardial fluid obtained from myocardial infarction was prepared by our colleagues Zhang et al., and the paracrine function and survival rate of the pericardial fluid-pretreated CSps dramatically increased, exhibiting significant improvement of MI cardiac function\n \n \n 24\n \n \n . Moreover, Zhang et al. reported that pericardial application could serve as a new and effective route for CSps transplantation, and this therapeutic strategy also showed favorable potential for further clinical application\n \n \n 25\n \n \n .\n
\n\n Cellular physiological activities could be manipulated by the properties of the contacted substrate\n \n \n 26\n \n ,\n \n 27\n \n ,\n \n 28\n \n \n . Liquid crystal patterns could directly introduce cells into a 3D environment and form cell structures in situ\n \n \n 29\n \n \n , which may be beneficial for mass spheroid production during large-scale clinical applications. In addition, the alignment of monolayer support cells could form a nematic liquid crystal pattern to induce cell death at the stress localization\n \n \n 30\n \n \n . Given this, the phenomenon might be applied to develop a local inflammatory milieu by turning this theoretical model into cell product preparation methods. Different from the artificial stimulus inducer, these dynamic inflammatory secretomes released by dead cells could be more complex and comprehensive\n \n \n 31\n \n \n , which may be beneficial for stimulating CSps to acquire enhanced inflammation resistance. Therefore, using a liquid crystal substrate for pretreated CSps might be a stable, convenient, and effective strategy to achieve mass production and cell function improvement.\n
\n\n Previously, a new kind of cholesteric liquid crystal substrate, octyl hydroxypropyl cellulose ester (OPC), was developed by our group\n \n \n 32\n \n \n . By using the properties of liquid crystals to promote 3D spheroid formation and induce cell death, the internal cells in the spheroid could be activated by inflammatory factors secreted from the external cells. Therefore, the goal of this work was to prepare a comprehensive optimized 3D culture platform using OPC for effective CSps production and preconditioning. The bioactivities, metabolism, and function of OPC-CSps were analyzed, and their therapeutic effects on heart function, angiogenesis, inflammatory infiltration, and ventricular remodeling were evaluated in a rat MI model.\n
\n\n The synthesis schematic diagram of OPC is shown in Figure. 1a. The polarized light microscopic images revealed that the OPC displayed the characteristics of liquid crystals, including birefringence, fissures, and fingerprint-like texture (Fig.\n \n 1\n \n b). The atomic force microscope results showed that the surface of the OPC substrate was a nonflat profile with wavy bulges. The height of the grains ranged from 20\u201335 nm, and the roughness (root mean square height, Sq) was 2.33\u2009\u00b1\u20090.29 nm (Fig.\n \n 1\n \n c). The static contact angle of the OPC substrate was 106.49\u2009\u00b1\u20092.36\u00b0, indicating that it had a hydrophobic surface (Fig.\n \n 1\n \n d). The effect of shear force on the OPC substrate surface was examined. As the X-ray diffraction (XRD) results showed, there were two diffraction peaks before shearing, and peaks at approximately 2\u03b8 of 20\u201322\u00b0 were enhanced following the application of shear force, indicating the rearrangement of the liquid crystal unit (Fig.\n \n 1\n \n e). In addition, the average crystallization rate and the grain size of the vertical (002) crystal plane were calculated according to the XRD results, and the average crystallization rate increased from 17.91\u201321.48%, and the grain size of the vertical (002) crystal plane increased from 0.69 nm to 1.14 nm. The viscoelasticity of the OPC substrate was examined by a stress-controlled rheometer, and the phase transition was observed. At 1\u201310 rad, the loss modulus (G\") was higher than the energy storage modulus (G'), and OPC preferred viscous deformation behavior. In contrast, at 10\u2013100 rad, the energy storage modulus (G') was higher than the loss modulus (G\"), indicating an elastic deformation tendency (Fig.\n \n 1\n \n f). The OPC substrate was nontoxic in cell culture (Fig.\n \n 1\n \n g).\n
\n\n
\n\n The formation of CSps on the polystyrene (PS) substrate, the ultralow attachment (ULA) substrate and the OPC substrate was observed, and different shapes of CSps were acquired (Fig.\n \n 2\n \n a). The adherent CDCs on the PS substrate converged and formed regular spherical cloning, and fibroblast-like supporting cells were in the lowest of the PS-CSps. The ULA-CSps were formed by suspended single-cell stacking, and they developed into noncircular, oval, and irregular shapes. On the OPC substrate, CDCs initially attached to the substrate and gradually aggregated to form regular and circular spheroids, and no supporting cells were observed around the OPC-CSps. During the formation of CSps, the OPC-CSps showed a stable tendency compared with the PS-CSps and the ULA-CSps (Fig.\n \n 2\n \n b). Following a 3-day cultivation, the sizes of PS-CSps, ULA-CSps, and OPC-CSps were 280.96\u2009\u00b1\u200940.56 \u00b5m, 203.34\u2009\u00b1\u200969.36 \u00b5m, and 120.29\u2009\u00b1\u200915.34 \u00b5m, respectively. In addition, the spheroid density on the OPC substrate was significantly higher than that on the other two substrates (Fig.\n \n 2\n \n c). The expression level of CSps surface markers was analyzed. Compared to the PS group and the ULA group, the OPC group exhibited the highest expression of KDR and Sca-1(\n \n P\n \n <\u20090.05). In addition, the ULA group and the OPC group showed an increase in the expression of CD31 and CD34 (\n \n P\n \n <\u20090.05), along with a decrease in CD90 (\n \n P\n \n <\u20090.05) and no significant difference in CD105 compared to the PS group (Fig.\n \n 2\n \n d).\n
\n\n
\n\n The expression of\n \n caspase-1\n \n was observed in the peripheral cells of the OPC-CSps, while no positive signals were observed in the PS group and the ULA group (Fig.\n \n 3\n \n a). The transcription levels and protein expression levels of the pyroptosis-related key factors\n \n caspase-1\n \n and\n \n IL-1\u03b2\n \n in the OPC group were both significantly upregulated compared with those in the PS group and the ULA group (\n \n P\n \n <\u20090.05) (Fig.\n \n 3\n \n b-\n \n 3\n \n d). As the transmission electron microscope (TEM) results showed, normal structures of the nucleus, mitochondria, and rough endoplasmic reticulum were observed in the PS group. However, endoplasmic reticulum dilatation, degranulation, and swollen mitochondria were observed in the cells from the ULA group. Moreover, cells in the OPC-CSps showed normal nuclear structures and abundant normal mitochondria. In contrast to the PS group and the ULA group, many microvesicles could be observed on the membrane surface (Fig.\n \n 3\n \n e). Following a 3-day cultivation, the survival rate of the ULA group was markedly lower than that of the PS group. The survival rate of the OPC group was significantly higher than that of the ULA group (\n \n P\n \n <\u20090.05), and it showed no significant difference from the PS group (Fig.\n \n 3\n \n f). The proliferation ability of the cells from OPC-CSps was significantly higher than that from ULA-CSps (\n \n P\n \n <\u20090.05), and it showed no significant difference from the PS-CSps (Fig.\n \n 3\n \n g). In addition, the transcription levels of the cell pluripotency markers\n \n Oct4\n \n ,\n \n Nanog\n \n , and\n \n Sox2\n \n (Fig.\n \n 3\n \n h) and the paracrine-related genes\n \n VEGF\n \n ,\n \n HGF\n \n ,\n \n IGF-1\n \n , and\n \n bFGF\n \n dramatically increased following OPC culture compared with the PS and ULA groups (\n \n P\n \n <\u20090.05) (Fig.\n \n 3\n \n i).\n
\n\n
\n\n The results of RNA-sequencing (RNA-Seq) analysis showed that there were 1251 differentially expressed genes (DEGs) between the ULA group and the OPC group (Fig.\n \n 4\n \n a), and 232 DEGs were not among the DEGs of PS vs. ULA or PS vs. OPC (Fig.\n \n 4\n \n b). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis showed that the DEGs between the OPC-CSps and the ULA-CSps were enriched in the glycolytic/glycogenic pathway and the HIF-1 signaling pathway (Fig.\n \n 4\n \n c). Moreover, gene set enrichment analysis (GSEA) also revealed the downregulation of the hypoxic and glycolytic components in the OPC-CSps compared to the ULA-CSps (Fig.\n \n 4\n \n d). The oxidative phosphorylation genes in the OPC group, including\n \n CS\n \n ,\n \n COXII\n \n ,\n \n IDH2\n \n ,\n \n SDHA\n \n , and\n \n MDH2\n \n , were notably upregulated compared with those in the PS and ULA groups (\n \n P\n \n <\u20090.05). Compared to the ULA group, the key genes of the glycolytic pathway,\n \n HK2\n \n ,\n \n LDHA\n \n , and\n \n PFKL\n \n , dramatically decreased in the OPC group (\n \n P\n \n <\u20090.05) (Fig.\n \n 4\n \n e). In addition, the transcription levels of these three genes showed no difference between the PS group and the OPC group. Compared to the PS-CSps and the ULA-CSps, the 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl) amino)-2-deoxyglucose (2-NBDG) uptake level of the OPC-CSps significantly decreased (\n \n P\n \n <\u20090.05) (Fig.\n \n 4\n \n f). The lactate production by the OPC-CSps was markedly lower than that by the ULA-CSps, and it was significantly higher than that by the PS-CSps (\n \n P\n \n <\u20090.05) (Fig.\n \n 4\n \n g). Among these three groups, the OPC-CSps had the highest ATP production level (Fig.\n \n 4\n \n h).\n
\n\n The TEM results showed that the mitochondria in the ULA-CSps exhibited obvious swelling, vacuolization, and cristae breakage, while the mitochondria in the OPC-CSps and the PS-CSps maintained normal cristae morphology. Furthermore, the density of mitochondria was significantly increased in the OPC group compared to the PS group and the ULA group (\n \n P\n \n <\u20090.05) (Fig.\n \n 4\n \n i). Mitochondrial membrane potential levels were evaluated by immunofluorescence staining of rhodamine 123 fluorescence intensity, and the membrane potential level was significantly enhanced in the OPC group compared to the PS group and the ULA group (\n \n P\n \n <\u20090.05) (Fig.\n \n 4\n \n j).\n
\n\n
\n\n H\n \n 2\n \n O\n \n 2\n \n stimulation was used to test cellular oxidative stress resistance. As shown in Fig.\n \n 5\n \n a and\n \n 5\n \n b, after exposure to H\n \n 2\n \n O\n \n 2\n \n for 24 h, the fluorescence intensity of the ULA group was significantly lower than that of the PS group(\n \n P\n \n <\u20090.05). Moreover, the fluorescence intensity of the OPC group further decreased compared to the PS group and the ULA group, suggesting that OPC-CSps generated the least reactive oxygen species (ROS) and superoxide under an oxidative stress environment. Following 12 h of H\n \n 2\n \n O\n \n 2\n \n stimulation, the percentage of viable cells in the OPC group (73.7% \u00b1 1.4%) was significantly higher than that in the PS group (59.7% \u00b1 7.5%) and the ULA group (59.2% \u00b1 5.4%) (\n \n P\n \n <\u20090.05) (Fig.\n \n 5\n \n c). There was no difference in the cell survival rate between the PS-CSps and the ULA-CSps. Following 24 h of H\n \n 2\n \n O\n \n 2\n \n stimulation, a similar trend in the survival rate was observed among the three groups, and the survival rates of the PS, ULA, and OPC groups were 1.5% \u00b1 0.7%, 5.4% \u00b1 2.1%, and 31.7% \u00b1 4.7%, respectively (Fig.\n \n 5\n \n d).\n
\n\n
\n\n In vivo live imaging was performed to detect the survival rate of transplanted OPC-CSps within the infarct area. The survival rate of transplanted OPC-CSps was 53.83% \u00b1 9.01% at week 2 and 16.18% \u00b1 3.68% at week 4 (Fig.\n \n 6\n \n a&\n \n 6\n \n b). According to the echocardiography results, serious motor dysfunction of the anterior ventricular wall was observed in the vehicle group following ligation. However, motor function was maintained in the OPC-CSps group compared to the vehicle group. The results also revealed that the OPC-CSps group significantly improved cardiac function starting at week 4, and this tendency was sustained throughout the 12-week observation (Fig.\n \n 6\n \n c). Compared to the vehicle group, the OPC-CSps group showed a remarkable increase in left ventricular fractional shortening (LVFS) and left ventricular ejection fraction (LVEF) (\n \n P\n \n <\u20090.05) (Fig.\n \n 6\n \n d&\n \n 6\n \n f). The LVEF and LVFS at week 12 relative to week 4 of the vehicle group decreased by 2.74\u2009\u00b1\u20092.33% and 2.51\u2009\u00b1\u20091.24%, respectively. In contrast, the OPC-CSps group showed a 10.12\u2009\u00b1\u20092.57% improvement in LVEF and 4.54\u2009\u00b1\u20092.33% in LVFS at week 12 relative to week 4 (Fig.\n \n 6\n \n e&\n \n 6\n \n g). In addition, compared to the vehicle group, the OPC-CSps group showed a significant reduction in left ventricular internal diameters both in systole (LVIDs) and diastole (LVIDd) at week 8 and week 12 (\n \n P\n \n <\u20090.05) (Fig.\n \n 6\n \n h-k).\n
\n\n
\n\n Compared to the vehicle group, the OPC-CSps group exhibited a significant cardioprotective effect (Fig.\n \n 7\n \n a&\n \n 7\n \n b). Clear vascular structures at the infarct region border were observed in the vehicle group and the OPC-CSps group. However, compared to the vehicle group, there were fewer perivascular collagens in the OPC-CSps group (Fig.\n \n 7\n \n a). Compared to the sham group, an increase in the size of the infarct area (29.25% \u00b1 3.63%) and a decrease in myocardial tissue retention (23.66% \u00b1 3.52%) were observed in the vehicle group. Compared to the vehicle group, the OPC-CSps group showed a significant decrease in infarct area size (18.54% \u00b1 3.19%) and an increase in retained myocardial tissue (41.69% \u00b1 7.99%) (Fig.\n \n 7\n \n c&\n \n 7\n \n d). Compared to the vehicle group (0.60\u2009\u00b1\u20090.06 mm), the thickness of the left ventricular wall was higher in the OPC-CSps group (1.27\u2009\u00b1\u20090.19 mm), which reached 46% of the normal left ventricle wall thickness (2.71\u2009\u00b1\u20090.28 mm) (Fig.\n \n 7\n \n e).\n
\n\n For cardiac inflammation evaluation, CD68\n \n +\n \n macrophages were calculated. In the sham group, macrophage infiltration was scarcely observed, and the number of CD68\n \n +\n \n macrophages in the vehicle group significantly increased compared to that in the sham group (\n \n P\n \n <\u20090.05). In the OPC-CSps group, the number of CD68\n \n +\n \n macrophages significantly decreased compared to that in the vehicle group (\n \n P\n \n <\u20090.05) (Fig.\n \n 7\n \n b&\n \n 7\n \n f). The structure and distribution of the vessels in the LV wall were observed by \u03b1-SMA immunofluorescence. The vessel lumen could be obviously observed in the sham, vehicle, and OPC-CSps groups. Compared to the sham group, the vessel density of the vehicle group and the OPC-CSps group both significantly increased (\n \n P\n \n <\u20090.05). Compared to the vehicle group, a significantly higher vascular density (\n \n P\n \n <\u20090.05) and mature large-diameter blood vessels (>\u2009100 m) were observed in the OPC-CSps groups (Fig.\n \n 7\n \n b&\n \n 7\n \n g). Terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) staining results showed that the percentage of apoptotic cells in the vehicle group (62.71% \u00b1 10.17%) was significantly higher than that in the sham group, and a significant decrease was observed in the OPC-CSps group (29.11% \u00b1 9.54%) (\n \n P\n \n <\u20090.05) (Fig.\n \n 7\n \n b&\n \n 7\n \n h).\n
\n\n As the wheat germ lectin (WGA) staining results showed, the average cardiomyocyte cross-sectional area was significantly higher in the infarct zone, border zone, and remote zone in the vehicle group than in the sham group (\n \n P\n \n <\u20090.05). However, in the border zone and the remote zone, the OPC-CSps group exhibited smaller cardiomyocyte sizes than the vehicle group (\n \n P\n \n <\u20090.05), and no significant differences were observed between the sham group and the OPC-CSps group in all areas measured (Fig.\n \n 7\n \n i-l).\n
\n\n
\n\n Substantial progress in MI cell therapy has been widely reported, and improving transplanted cell survival and therapeutic outcomes remain the key issues to be addressed. Optimizing the culture system had great significance in obtaining abundant transplanted cells with favorable bioactivities. This study aimed to improve the therapeutic potential of CSps by optimizing their culture substrate. The prepared OPC substrate was a kind of cholesteric liquid crystal obtained by the esterification between HPC and OC. When cultured on the OPC substrate, CDCs could spontaneously form homogenous 3D spheroids at a high density, which was beneficial for quickly acquiring sufficient CSps in clinical applications. Compared with the PS substrate and the ULA substrate, CSps cultured on the OPC substrate could be activated and exhibited a superior paracrine effect, enhanced metabolic state, and improved oxidative stress resistance in the unique pyroptosis microenvironment. In a rat MI model, CSps prepared by OPCs showed a long-term cardioprotective effect within 12 weeks. A decrease in host cell apoptosis, improvement in angiogenesis, and reduction in ventricular remodeling were observed following OPC-CSps transplantation.\n
\n\n For producing highly functional CSps, preparing the proper cell culture substrate is the first and vital step. In this study, liquid-crystal OPC was synthesized using HPC as the rigid chain and OC as the flexible chain (Fig.\n \n 1\n \n ). An optical texture formed by lattice defects was observed under the polarizing microscope, which was due to the characterization of entropy-induced phase transitions of OPC. After being subjected to an external shear force, the rigid chains of liquid crystal materials maintain molecular orderliness, while the flexible chains adjust their orientation in response to mechanical variations. The change in orientation of flexible chains not only improved the orderliness of materials but also drove the change in molecular position ordering of rigid chains when such changes accumulated to a certain extent. These alterations eventually led to the formation of lattice defects inside the material and further resulted in texture changes. The increase in the crystallinity index following shear force application implied an improvement in the orderliness of the materials, and the change in grain size was related to the change in the molecular position of the rigid chain.\n
\n\n In addition, the complex phase behavior of OPC is determined by the structural mosaic of rigid chains and flexible chains. Within the detection range of rheological characterization, a phase transition was observed in OPC. In the 1\u201310 rad region, the stress hysteresis of the flexible chain resulted in the tendency of viscous deformation. In the 10\u2013100 rad region, the rigid chain dominated, and the stress hysteresis of the flexible chain attenuated, leading to a higher tendency of elastic deformation in OPC. The OPC substrate exhibited the highest CSps production efficiency among the three groups (Fig.\n \n 2\n \n c). The density of CSps in the OPC group was 500 times greater than that in the PS group, while it was 10 times higher than that in the ULA group. These findings showed that OPC could rearrange the liquid crystal units in response to external stress, and the characteristics of mechanical responsiveness could promote CSps formation, which could satisfy the demand for large-scale CSps production in clinical applications.\n
\n\n The sizes of 3D spheroids have a significant influence on cell viabilities. Spheroids within 200 \u00b5m in diameter could grow under sufficient nutrition and oxygen supply, which was beneficial for delaying the formation of hypoxic cores and maintaining the viability of their internal cell populations\n \n \n 33\n \n ,\n \n 34\n \n \n . In this study, different sizes of CSps were observed in the PS, ULA, and OPC groups. CSps cultured on the OPC substrate were observed to form 70\u2013140 \u00b5m in diameter within five days (Fig.\n \n 2\n \n b), and the internal cells in the OPC-CSps maintained normal cell ultrastructure (Fig.\n \n 3\n \n e). In contrast, the ULA spheroids were 100\u2013470 \u00b5m in diameter and beyond the size of effective nutrient and oxygen transportation\n \n \n 35\n \n \n , so the ULA-CSps showed serious cell damage in the core of ULA-CSps with a higher portion of apoptosis (Fig.\n \n 3\n \n e&\n \n 3\n \n f). Owing to the mechanical cues from OPC, massive CSps with controllable size and favorable bioactivity could be obtained effectively.\n
\n\n In addition, cell\u2012cell and cell-ECM contacts also greatly affect the phenotypes of the cells in CSps. Compared to the PS group, the expression of CD31 and CD34 significantly increased in the ULA group and the OPC group (Fig.\n \n 2\n \n d), which could be related to the microenvironmental cues in the ECM according to previous studies\n \n \n 36\n \n \n . The different portions of CD90\n \n +\n \n cells could be observed when cultured on the PS, ULA, and OPC substrates, so it is reasonable to assume that the expression level of CD90 could be regulated by the physical and chemical environments provided by each substrate. In addition, several studies have shown that a decrease in mesenchymal markers, such as CD90, could lead to an improvement in the pluripotency of spheroid cells\n \n \n 37\n \n ,\n \n 38\n \n \n . In this study, compared with the PS group and the ULA group, the OPC group showed the lowest expression level of CD90 and the highest expression levels of pluripotency markers, including\n \n Nanog\n \n ,\n \n Sox2\n \n , and\n \n Oct4\n \n (Fig.\n \n 3\n \n h). Furthermore, the highest expression of the cardiovascular progenitor marker KDR and the cardiac fibro-adipogenic progenitor marker Sca-1 was observed in the OPC group (Fig.\n \n 2\n \n d). In conclusion, culturing CDCs on the OPC substrate could not only obtain more progenitors in the CDC population but also facilitate their differentiation into the endothelial lineage.\n
\n\n In addition to favorable cell bioactivities, improving cellular resistance to oxidative stress and the inflammatory environment is another key issue in cell therapy. Following acute MI, the degradation of the extracellular matrix and the cytokines released by dead cardiomyocytes lead to serious inflammation, and massive host cells induce pyroptosis\n \n \n 39\n \n \n .\n \n Caspase-1\n \n and\n \n IL-1\u03b2\n \n are the classical factors of the pyroptosis signaling pathway. In this study, pyroptosis of the external cells of CSps was induced by mechanical cues from OPC, with the activation of\n \n caspase-1\n \n and an increase in the release of\n \n IL-1\u03b2\n \n (Fig.\n \n 3\n \n a-d). Meanwhile, by receiving the proper stimulation from external pyroptosis, the internal cells with higher bioactivities exhibited an improved paracrine effect, and improved\n \n VEGF\n \n ,\n \n HGF\n \n ,\n \n IGF-1\n \n , and\n \n bFGF\n \n were observed (Fig.\n \n 3\n \n i). In addition, compared with the hypoxic microenvironment in CSps from the ULA group, the cellular proliferation activity of OPC-CSps was maintained (Fig.\n \n 3\n \n g). Therefore, these results demonstrated that the OPC substrate could provide proper stimulation for CSps to improve cellular paracrine effects and maintain their proliferation ability.\n
\n\n Cellular inflammation is directly related to oxidative stress, and improving antioxidative stress ability is vital for cell survival in MI cell therapy. When exposed to oxidative stress, the generated hydroxyl radicals can react with all biological macromolecules, causing DNA, protein, membrane damage, and ultimately cell death. Mitochondria are the center of energy metabolism, and they control many signals in cell fate programs\n \n \n 40\n \n \n . It was reported that enhancing mitochondrial respiration and function could reduce the damage caused by oxidative stress\n \n \n 41\n \n \n . In this study, compared to the PS group and the ULA group, the OPC-CSps exhibited higher mitochondrial density and membrane potential levels (Fig.\n \n 4\n \n i&\n \n 4\n \n j). It was reported that cells in hypoxia would lead to mitochondrial damage\n \n \n 42\n \n \n , and with a compact core in the CSps, the ULA group showed lower oxidative stress resistance. In contrast, under the proper stimulation induced by OPC, the CSps in the OPC group could acquire the ability to resist oxidative stress before transplantation (Fig.\n \n 5\n \n ). Taking these results together, CSps with improved oxidative stress resistance could be obtained using OPC as the culture substrate.\n
\n\n Furthermore, RNA-seq analysis was employed to investigate the underlying mechanism of differences in cell bioactivity and oxidative stress resistance among the three groups (Fig.\n \n 4\n \n a). The metabolic level and bioenergetic state are highly related to the availability of oxygen and nutrients\n \n \n 43\n \n \n . In this study, DEGs between the OPC group and the ULA group were significantly enriched in the glycolysis/gluconeogenesis pathway and HIF-1 signaling pathway, but these pathways were not in the top 7 enriched pathways between the OPC group and the PS group (Fig.\n \n 4\n \n c&\n \n 4\n \n d). These results further proved that the structure of the OPC-CSps could satisfy the demand for internal CDC metabolism. It was reported that the enhancement of oxidative phosphorylation could surmount mitochondrial fission and functional failure\n \n \n 44\n \n \n , while glycolysis was associated with mitochondrial dysfunction\n \n \n 45\n \n \n . In this study, the oxidative phosphorylation of OPC-CSps was enhanced, while the ULA-CSps altered their energy production toward glycolysis (Fig.\n \n 4\n \n e). Therefore, the highest ATP level was observed in the OPC group (Fig.\n \n 4\n \n h), and the OPC-CSps showed improved mitochondrial function and effective protection against cell damage in oxidative stress. In addition, the OPC group showed a significant decrease in glucose uptake levels (Fig.\n \n 4\n \n f), suggesting that the OPC-CSps may survive longer in nutrient-limited conditions. In conclusion, these results illustrated that CSps could switch toward a highly metabolically active state when cultured on the OPC substrate, which is beneficial for OPC-CSps\u2019 long-term survival in the hostile microenvironment.\n
\n\n With favorable cell viabilities, superior antioxidative stress, and long-term cellular survival ability, the therapeutic effect of OPC-CSps on MI was evaluated. As the results showed, the OPC-CSps group significantly improved MI cardiac function. Compared with the overtime decline tendency of the vehicle group, a consistent increase in left ventricular systolic and diastolic function was observed in the OPC-CSps group during the 12-week observation (Fig.\n \n 6\n \n ). To pursue favorable outcomes in MI therapy, reducing myocardial inflammation and rebuilding the vessel network are crucial issues for protecting cardiac structure and function. Previous studies have reported that CSps could regulate the inflammation of infarcted myocardium through immunomodulatory effects\n \n \n 46\n \n ,\n \n 47\n \n ,\n \n 48\n \n \n . In addition, CSps can secrete various growth factors and bioactive molecules that are involved in the vessel network rebuilding process\n \n \n 49\n \n ,\n \n 50\n \n \n . In this study, owing to the effective preconditioning treatment in vitro, the OPC-CSps acquired enhanced inflammation resistance and improved regenerative function. The transplanted CSps were tracked by Dir labeling, and the results showed that 16.18% \u00b1 3.68% of CSps survived 4 weeks following transplantation (Fig.\n \n 6\n \n a&\n \n 6\n \n b). Meanwhile, a significant decrease in CD68\n \n +\n \n macrophages in the border zone was observed in the OPC-CSps group (Fig.\n \n 7\n \n b&\n \n 7\n \n f). Moreover, effective angiogenesis within the infarcted myocardium was widely observed at 12 weeks following MI (Fig.\n \n 7\n \n b&\n \n 7\n \n g). Therefore, these in vivo results demonstrated that transplanting OPC-CSps could achieve satisfactory long-term cardiac function recovery by effectively reducing myocardial inflammation and promoting angiogenesis.\n
\n\n Furthermore, protecting normal cardiac structure was a key issue for maintaining MI cardiac function. Owing to the severe inflammation and the hostile environment following MI, excessive degradation or impaired synthesis of ECM after MI was considered to accelerate ventricular remodeling, including myocardial fibrosis and cardiac hypertrophy, and ultimately lead to heart failure\n \n \n 51\n \n \n . Cardiac fibrosis greatly reduces cardiac function by replacing necrotic myocardial tissue with enlarged scars. In addition, massive cell death, a decrease in contractile activity in the affected zone, and increased hemodynamic burden were assumed to be the main causes of cardiac hypertrophy\n \n \n 52\n \n \n . The cytokines secreted by CSps were reported to inhibit the proliferation of cardiac fibroblasts and reconstitute the cardiac vascular network, which were beneficial for reducing ventricular adverse remodeling and hypertrophy\n \n \n 24\n \n ,\n \n 49\n \n ,\n \n 53\n \n ,\n \n 54\n \n \n . Transplantation of OPC-CSps significantly decreased the infarct area, and an increase in viable cardiac tissue was observed (Fig.\n \n 7\n \n a, c, d), showing a beneficial effect in preventing cardiac hypertrophy (Fig.\n \n 7\n \n i-l). These results supported that transplanting OPC-CSps could greatly protect MI cardiac function by reducing ventricular remodeling.\n
\n\n In summary, to improve the MI therapeutic potential of CSps, a novel cell culture substrate, liquid crystal OPC, was prepared for CSps culture and preconditioning. The OPC substrate served as a special mechanical cue to effectively promote the formation of CSps of controllable size. Furthermore, the OPC-CSps exhibited significant enhancement of biological function and antioxidative stress abilities. In the rat MI model, OPC-CSps not only showed great cell retention and survival in the infarct area but also significantly improved cardiac wall thickness, angiogenesis, and long-term cardiac function. In conclusion, using the OPC substrate could satisfy the demand for large-scale CSps production with excellent cardiac regeneration abilities for MI therapy.\n
\n\n Rats were housed in specific pathogen-free conditions with 12 h day/light cycles. Rats were healthy and had free access to water and food. All animal studies were performed in accordance with the ethical guidelines of the National Guide for the Care and Use of Laboratory Animals and approved by Jinan University Animal Care and Use Committee (Approval numbers: I ACUC-20210113-06). 4-week-old male Sprague Dawley rats (Guangdong Medical Laboratory Animal Center) were used for isolating primary CDCs, and 3-month-old female Sprague Dawley (Vital River) rats were used to establish myocardial infarction animal models. Every attempt was made to minimize the use of animals and pain.\n
\n\n OPC was prepared via esterification between hydroxypropyl cellulose (HPC) (Sigma\u2012Aldrich, Mw\u2009=\u2009100,000 g/mol) and octanoyl chloride (OC) (Sigma\u2012Aldrich). Briefly, 5.0 g HPC was dissolved in 30 mL dehydrated acetone with mild stirring. 7 ml OC was added to the solution when HPC dissolved completely, and the reaction was kept at 55\u00b0C for 4 h. Then, 300 ml distilled water was added to the reaction mixture, and a cream color sticky mass was obtained after removing the liquid phase. The cream-colored sticky mass was dissolved in acetone and precipitated by adding water to the solution. This step was repeated 6 times. After dissolving in ethanol and dialyzing in distilled water 15 times to remove the residual OC, OPC could be obtained after precipitation. Finally, the OPC product was dried in a vacuum at 55\u00b0C for 48 h.\n
\n\n For the preparation of the OPC substrate, a 3% OPC concentration mixture was obtained by stirring OPC and ethanol for 1 h at 20\u00b0C. It was cast onto clean culture dishes. After the solvents evaporated at room temperature, the dishes were washed with distilled water 10 times for 4 h each time. Then, the dishes were sterilized by Co60 irradiation (15 kGy).\n
\n\n The surface characteristics of OPC were observed by polarized optical microscope (Carl Zeiss Axioskop40). An atomic force microscope (BENYUAN) was used to analyze the surface roughness in on-contact mode. The measurement of the water contact angle on the OPC substrate was tested at room temperature by a contact angle meter (Kruss DSA100) with ultrapure water as the testing liquid and a humidity of 80%.\n
\n\n The changes in the OPC structure when subjected to a shear force were tested by X-ray diffraction (XRD, Dmax1200). Briefly, 3% OPC solution was added to the glass surface and then covered with the magnesium sheet. Next, the samples were dried by placing them in a vacuum at 55\u00b0C for 48 h. Then, the magnesium sheet was slid by weights of equal mass with a distance of 3 mm. The XRD patterns were recorded from 5\u00b0 to 40\u00b0 at a step width of 0.02\u00b0 and scanning speed of 8\u00b0/min.\n
\n\n The crystallinity index (\n \n Cr. I\n \n ) of OPC was determined according to the Segal method and calculated using Eq.\u00a0(1)\n \n 55\n \n
\n\n
\n\n where\n \n I\n \n \n \n 002\n \n \n is the maximum intensity of the main diffraction, and\n \n I\n \n \n \n amorph\n \n \n is the intensity of the amorphous background scatter measured at 2\n \n \u03b8\n \n =\u200918\u00b0 where the intensity is minimum.\n
\n\n The crystallite diameter (\n \n D\n \n \n \n 002\n \n \n ) perpendicular to the (\n \n 002\n \n ) plane was calculated from the Scherrer Eq.\u00a0(2)\n \n 56\n \n
\n\n
\n\n where\n \n K\n \n is a Scherrer constant that equals 0.9,\n \n \u03bb\n \n is the wavelength of the radiation (1.54 \u00c5 for CuK\u03b1),\n \n \u03b2\n \n is the width of the peak at half maximum, and\n \n \u03b8\n \n is the angle of incidence.\n
\n\n Finally, the rheological properties of OPC were measured by a DHR-2 stress-controlled rheometer (TA Instruments). The oscillation-frequency mode at 37\u00b0C was incorporated for rheological tests with a strain of 1% and \u03c9\u2009=\u20091\u2013100 rad s\n \n \u2212\u20091\n \n .\n
\n\n For the OPC toxicology test, 2\u00d710\n \n 3\n \n CDCs were seeded in 96-well plates and cultured in 100 \u00b5l of CSps culture medium or OPC immersion culture medium. Ten microliters of CCK8 (Dojindo) was added to each well and incubated for 2 h every 24 h for five consecutive days, and the optical density values were recorded at 450 nm wavelengths.\n
\n\n Cardiac tissue specimens from the septum of the left ventricle were minced and digested with collagenase IV (Sigma) for 20 min at 37\u00b0C. These tissues were plated on poly-d-lysine (Sigma)-coated dishes in CSps culture medium, which consisted of 500 ml Iscove DMEM (Corning), 1% L-glutamine (Corning), 1% penicillin\u2012streptomycin solution (Corning), 10% fetal bovine serum (BD Bioscience) and 0.1 mmol/L \u03b2-mercaptoethanol (Gibco). After 1\u20132 weeks, a monolayer of adherent cells that grew out from these tissues was harvested by 0.05% trypsin and passaged on poly-d-lysine-coated dishes. Following 3\u20137 days of cultivation, the CSps were collected and plated onto fibronectin-coated dishes and expanded as monolayer CDCs. All cultures were cultured in 5% CO\n \n 2\n \n at 37\u00b0C.\n
\n\n Polystyrene (PS) substrate, ultralow attachment (ULA) substrate, and OPC substrate were used to culture CDCs and obtain CSps. Cells were seeded onto three different substrates at a density of 7 \u00d7 10\n \n 4\n \n cells/cm\n \n 2\n \n . The formation of CSps in each group was observed by an inverted microscope (Olympus IX71) for 5 consecutive days. The diameter and the total number of CSps at the same time point were measured using ImageJ software, and at least 30 CSps from each group were randomly chosen.\n
\n\n The expression levels of the CSps surface markers CD31 (1:200, GB11063-3, Servicebio), CD34 (1:200, ab81289, Abcam), CD90 (1:200, ab225, Abcam), CD105 (1:200, ab156756, Abcam), Sca-1 (1:200, ab51317, Abcam), and KDR (1:200, sc6251, Santa Cruz) were determined by flow cytometry. After a 3-day cultivation, cells and CSps from different substrates were obtained and digested into single cells. After incubation with the primary antibodies for 1 h and the corresponding secondary antibodies for 30 min, Alexa Fluor 488 goat anti-mouse IgG (1:200, ab150113, Abcam) or Alexa Fluor 488 goat anti-rabbit IgG (1:200, ab150077, Abcam) was used. The staining results were analyzed by flow cytometry (BD FACSCanto), and a negative isotypic control was used during the analysis.\n
\n\n The cells and CSps from each substrate were harvested and fixed with 2.5% glutaraldehyde overnight at 4\u00b0C. Following dehydration with a series of graded ethanol solutions, samples were immersed in 1% osmium tetroxide for 2 h. Next, the samples were embedded in resin and cut to 60 nm thickness. Then, the cell ultrastructure was visualized and photographed with transmission electron microscope (TEM, JEOL).\n
\n\n Cell proliferation was determined by CCK8 (Dojindo) after being cultivated for 3 days on each substrate. Cell apoptosis rates were determined by using the Alexa Fluor\u00ae 488 Annexin-V/Dead Cell Apoptosis Kit (Elabsicence) according to the manufacturer's instructions. The results were analyzed and recorded by flow cytometry (BD FACSCanto). Survival cells in the Q4 gate were quantified and analyzed. Following 3 days of cultivation, the culture supernatants from each group were harvested, and the\n \n IL-1\u03b2\n \n concentration was measured by rat\n \n IL-1\u03b2\n \n ELISA kits (MEIMIAN).\n
\n\n Total RNA was extracted using TRIzol reagent. High-Capacity cDNA Reverse Transcription Kits (Thermo Fisher Scientific) were used for cDNA synthesis according to the manufacturer\u2019s instructions. qRT-PCRs were performed on a Mini Cycler PCR instrument with SYBR Green reagent (Toyobo). Gene expression was normalized to\n \n GAPDH\n \n mRNA, and relative expression was calculated by 2\n \n \u2212\u0394\u0394CT\n \n . The primer sequences of the detected genes are listed in Supplementary Table\u00a01.\n
\n\n The cells and CSps cultured on the PS, ULA, and OPC substrates were lysed in RIPA buffer (Solarbio) for protein extraction. The protein concentrations of the samples were adjusted by bicinchoninic acid (BCA) protein assay (Thermo Fisher Scientific) according to the manufacturer's protocol. Forty micrograms of protein was loaded into each lane of a 10% sodium dodecyl sulfate\u2012polyacrylamide gel electrophoresis (Millipore) gel and transferred to polyvinylidene fluoride membranes (Millipore). After being blocked with 5% bovine serum albumin in Tris buffered saline Tween (Biosharp) at room temperature for 1 h, the membranes were incubated with the primary monoclonal antibodies against\n \n caspase-1\n \n (1:500, ab1872, Abcam) and\n \n \u03b2-actin\n \n (1:1000, ab8226, Abcam) in TBST overnight at 4\u00b0C, after which HRP-labeled Anti-Rabbit IgG antibody (1:2000, 7074, Cell Signaling Technology) and HRP-labeled Anti-mouse IgG antibody (1:2000, 7076, Cell Signaling Technology) was added and incubated for another 1 h. The results were visualized via enzyme-linked chemiluminescence by an ELC kit (Thermo Fisher Scientific).\n \n \u03b2-Actin\n \n was used as an internal control.\n
\n\n The glucose uptake ability of CDCs was evaluated by using the fluorescent glucose 2-NBDG (Cayman). After 3 days of cultivation of CDCs on different substrates, all the culture medium was removed and replaced with glucose-free DMEM containing 50 \u00b5M 2-NBDG for 30 min. Consequently, the fluorescence intensity of the cells was measured by flow cytometry (BD FACSCanto). Following a 3-day cultivation, the culture medium of each group was collected to test the extracellular lactate contents, and the reagent kit was the Lactate Colorimetric Assay Kit (Nanjing JianCheng). After CDCs were cultured on different substrates for 3 days, cells and CSps were harvested and digested into single cells. For the intracellular ATP content assay, cells were seeded in a 96-well plate at a density of 1\u00d710\n \n 5\n \n , and then the intracellular ATP content was determined by using the CellTiter-Glo Luminescent Cell Viability Assay (Promega). For the mitochondrial membrane potential measurement, digested cells were washed twice with PBS and incubated with 2 \u00b5M Rho123 (Beyotime) for 20 min in a dark environment at 37\u00b0C. Fluorescence images were taken under a fluorescence microscope (Olympus, FV3000), and the fluorescence intensity of the cells was analyzed with ImageJ software.\n
\n\n According to the manufacturer's protocol, a total RNA Kit I (Omega) was used to extract sample RNA. Sequencing was performed on the Illumina platform. The raw RNA-seq reads were aligned to the Rattus norvegicus genome (rn6) by hisat2 (version:2.1.0). Mapped reads were counted by featureCounts (v.1.6.2), and gene expression was calculated by R and the DESeq2 package. Significant differentially expressed genes (DEGs) among cells cultured on the PS, ULA, and OPC substrates were evaluated using DESeq (1.28.0), and genes with a |log2FC|>=1 and an adjusted\n \n P\n \n value\u2009<\u2009=\u20090.05 were selected for further analysis. Hierarchical clustering was performed for DEGs using a heatmap. Kobas (3.0) was used for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. Additionally, gene set enrichment analysis (GSEA) was carried out using GSEA v.4.2.3.\n
\n\n Following a 3-day cultivation, cells from each group were subjected to H\n \n 2\n \n O\n \n 2\n \n stimulation for 24 h. First, intracellular ROS and superoxide production was measured using the cellular ROS/superoxide detection assay kit (Abcam). In short, H\n \n 2\n \n O\n \n 2\n \n -stimulated CDCs were incubated for ROS/superoxide detection for 30 min at 37\u00b0C. The results were observed and recorded by a laser confocal scanning microscope (Olympus, FV3000). In addition, following 12 h or 24 h of exposure to H\n \n 2\n \n O\n \n 2\n \n , all the cells were harvested to measure the survival rate using the Alexa Fluor\u00ae 488 Annexin-V/Dead Cell Apoptosis Kit (Invitrogen). The results were analyzed by flow cytometry (BD FACSCanto).\n
\n\n The establishment of the MI model and transplantation methods were carried out according to our previous study. In brief, female rats aged 3 months (220\u2009\u00b1\u200920 g) were anesthetized with 3% isoflurane and ventilated through endotracheal intubation. Mechanical ventilation was provided with room air at 60 to 70 breaths min\n \n \u2013\n \n 1\n \n \n using a rodent respirator (Taimeng Company). Subcutaneous stainless-steel electrodes were used to record the standard electrocardiogram. After shaving the chest, a left thoracotomy was performed to expose the heart at the fifth intercostal space. The left anterior descending coronary artery (LAD) was ligated using a 6\u2009\u2212\u20090 silk suture, and ischemia was confirmed by observing ST segment (or J point) elevation and the occurrence of cardiac cyanosis. After stabilizing for 15 minutes, rats in the vehicle group were transplanted with blank Matrigel, and rats in the OPC-CSps group were transplanted with OPC-CSps Matrigel suspensions (0.1 mL). OPC-CSps were transplanted into the OPC-CSps group with a total number of 5 \u00d7 10\n \n 6\n \n cells. For the sham group, rats were subjected to the same procedure without ligation and Matrigel injection. Finally, the animals were closed at the chest and monitored, and antibiotics and 0.9% normal saline solution were administered. Three rats from each group were killed at week 2 and cardiac samples were collected to assess macrophage infiltration. Continuous cardiac cycles were collected for 8 rats from each group.\n
\n\n Prior to transplantation, CSps were labeled with 3.5 \u00b5g/ml 1,1-dioctadecyl-3,3,3,3-tetramethylindotricarbocyanine iodide (DiR, Invitrogen) following the manufacturer's instructions. The survival status and localization of Dir-labeled CSps were observed using a Bruker In Vivo Xtreme II Imager (Bruker). The Bruker MI SE and ImageJ software were applied for imaging processing and data analysis.\n
\n\n Cardiac function and the movement of the left ventricular wall were measured using a Vevo 2100 ultrasonic system before surgery and at 4, 8, and 12 weeks postsurgery. After the rats were anesthetized with isoflurane, a parasternal long-axis view was obtained by two-dimensional echocardiography, and the m-mode was adjusted perpendicular to the basal segment for delineation of the segmental motion curve. The left ventricular internal diameter at end diastole (LVIDd), left ventricular internal diameter at end systole (LVIDs), left ventricular fractional shortening (LVFS), and left ventricular ejection fraction (LVEF) were measured and recorded.\n
\n\n For cell sample preparation, CSps from each group were harvested, fixed in a 75% ethanol solution, and embedded in OCT compound, and 5 \u00b5m sections were used. For tissue sample preparation, after rats were euthanized, the heart tissues were harvested, fixed in 4% paraformaldehyde, embedded in paraffin, and serially sectioned at 5 \u00b5m thickness. Masson trichrome staining was performed on tissue sections. For immunofluorescence staining, the sections were blocked with 5% goat serum and then incubated with primary antibodies, including anti-caspase-1(1:100, ab1872, Abcam), anti-CD68 (1:200, 360018, Zhengneng), anti-cardiac troponin T (cTnT, 1:500, ab209813, Abcam) or anti-smooth muscle alpha-actin (\u03b1-SMA, 1:400, ab1878, Abcam) at 4\u00b0C overnight. After washing with PBS 3 times, 488 nm goat anti-rabbit (1:400, ab150077, Abcam) or 594 nm goat anti-rabbit (1:400, ab150080, Abcam) secondary antibodies were added and incubated for 2 h at room temperature. The dilutions were 1:200 for the primary antibody and 1:400 for the secondary antibody. Apoptosis and the cross-sectional area of myocardial cells were assessed by terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) (Promega) and FITC-labeled wheat germ lectin (WGA) (1:500, Thermo Fisher Scientific) staining, respectively. DAPI showed the nuclei. All stained images were observed and photographed by laser confocal scanning microscope (Olympus, FV3000).\n
\n\n All analyses were performed using GraphPad Prism 9.0 (GraphPad Software Inc). For comparisons of two groups, a two-tailed Student\u2019s t test was used. Comparisons of multiple groups were made using one- or two-way ANOVA. All quantitative results were expressed as the mean\u2009\u00b1\u2009standard deviation, and differences with a\n \n P\n \n value\u2009<\u20090.05 were considered statistically significant.\n
\n\n Reporting Summary\n
\n \n\n Catalysts based on palladium are among the most effective in the complete oxidation of methane. Despite extensive studies, the nature of their catalytically active species and conceivable structural dynamics remains elusive. Here, we combine\n \n operando\n \n transmission electron microscopy (TEM) with\n \n \n near-ambient pressure\n \n \n X-ray photoelectron spectroscopy (NAP-XPS) and density functional theory (DFT) calculations to investigate the active state and catalytic function of Pd nanoparticles (NPs) under methane oxidation conditions. By direct imaging we show how the particle size, phase composition and dynamics respond to changes of the gas-phase chemical potential and how Pd catalysts transform from a static state to a highly dynamic, catalytically active state that is characterized by phase coexistence and oscillatory phase transition in a reactive atmosphere. Aided by DFT calculations, we rationalize the origin for the observed redox dynamics and provide atomistic insights into the active structures and the underlying reaction mechanism.\n
\n\n Natural gas engines have become a promising alternative to traditional petrol and diesel engines owing to the high energy density of CH\n \n 4\n \n and reduced NO\n \n x\n \n and CO\n \n 2\n \n emissions.(\n \n 1\n \n \u2013\n \n 3\n \n ) However, the lean-burn operation of natural gas engines typically leads to incomplete oxidation of CH\n \n 4\n \n and yields unburned CH\n \n 4\n \n in the exhaust.(\n \n 2\n \n ,\n \n 4\n \n ) This is unwanted since CH\n \n 4\n \n is a more potent greenhouse gas than CO\n \n 2\n \n .(\n \n 5\n \n \u2013\n \n 7\n \n ) In order to minimize the CH\n \n 4\n \n emission, catalytic conversion of unburned CH\n \n 4\n \n to CO\n \n 2\n \n and H\n \n 2\n \n O is required. Among various materials, Pd-based catalysts have been recognized as the most effective in the complete combustion of CH\n \n 4\n \n .(\n \n 1\n \n ,\n \n 3\n \n ,\n \n 5\n \n ,\n \n 8\n \n ,\n \n 9\n \n ) However, while significant research efforts have been devoted to this catalytic system, our understanding of the working state of Pd catalysts is still insufficient for a rational development of improved catalysts.(\n \n 10\n \n \u2013\n \n 13\n \n ) In particular, there is debate over the nature of the active surface. Some reports suggest that metallic Pd is more active than PdO in methane combustion,(\n \n 14\n \n \u2013\n \n 17\n \n ) however, most recent studies attribute the catalytic activity to PdO\n \n x\n \n or the presence of a metal/oxide interface.(\n \n 4\n \n ,\n \n 18\n \n \u2013\n \n 21\n \n ) These divergent conclusions may be linked to a dynamic co-existence of Pd and PdO under reaction conditions, making the assignment of distinct active structures and establishment of structure-activity relationship challenging.\n
\n\n Recent advances in the application of\n \n in situ\n \n and\n \n operando\n \n techniques in heterogeneous catalysis have enabled detailed insights into the working state of various catalysts.(\n \n 22\n \n \u2013\n \n 25\n \n ) Among these techniques,\n \n in situ\n \n transmission electron microscopy (TEM) is a particularly powerful tool for studying the atomic structure and dynamic behavior of materials as it offers a real-time and real-space imaging of catalysts with high temporal and spatial resolution under external stimuli.(\n \n 26\n \n \u2013\n \n 33\n \n ) In particular, the combined use of online mass spectrometry (MS) with\n \n in situ\n \n TEM has demonstrated a great potential in improving our understanding of the structure-performance relationships in catalytic processes, for instance, H\n \n 2\n \n or CO oxidation.(\n \n 34\n \n \u2013\n \n 37\n \n ) Yet, the majority of previous\n \n in situ\n \n /\n \n operando\n \n studies of Pd-based methane oxidation catalysis have used spectroscopic techniques with only a limited spatial resolution (e.g., X-ray absorption spectroscopy and X-ray photoemission spectroscopy).(\n \n 38\n \n ,\n \n 39\n \n ) Although those methods provide element-specific information about the oxidation state and local coordination environment, including either mostly bulk or (sub)surface sites when using XAS or XPS, respectively, this information is integral (i.e., averaged over micron-size specimen areas). Consequently, if active species comprise only a small fraction of the specimen, as is typically the case with industrial catalysts, their elucidation becomes challenging.(\n \n 40\n \n ) Furthermore, the co-existence of multiple phases complicates the search for structure-activity correlations. In this context, studies using\n \n operando\n \n TEM experiments can address these challenges by attaining a sufficient spatial resolution to link directly the nanoscale dynamics, typical for redox reactions with metal nanoparticle catalysts, to the catalytic performance (activity, selectivity and stability).(\n \n 34\n \n ,\n \n 36\n \n )\n
\n\n Recent\n \n operando\n \n TEM studies have indicated that Pd catalysts engage in oscillatory phase transformations at nanoscale (reshaping, particle splitting) under methane oxidation conditions.(\n \n 34\n \n ) The Pd and PdO phases co-exist and form phase boundaries within a single particle, consistent with earlier studies by Xiong and Huang et al.(\n \n 12\n \n ,\n \n 21\n \n ) Zhang and Bychkov et al. studied collective phase oscillations that are linked to oscillations of catalytic activity and investigated how amplitude and frequency of the oscillations depend on gas composition and temperature.(\n \n 17\n \n ,\n \n 41\n \n ) While insightful, these findings are still insufficient to unambiguously identify the active surface state, the origin of phase oscillations and influence of phase oscillations on catalytic activity. A deeper understanding of these key issues can be achieved via temporally and spatially-resolved atomic-level direct observation of the working state during methane oxidation conditions, including also the investigation of dynamic changes of Pd NPs as a function of the chemical potential of the gas phase. Herein, we utilize\n \n operando\n \n TEM, that is, real-time electron microscopy imaging coupled with online MS, complemented with surface studies using NAP-XPS to probe the active state, and with the aid of DFT calculations, to derive structure-performance relationships that govern methane oxidation on Pd NPs. We show how size, phase composition and structural dynamics of Pd NPs respond to changes of the gas-phase chemical potential. We reveal the catalytically active state (phase coexistence and oscillations) and structures down to the atomic level and offer insights into the underlying reaction mechanism as well as the origin of phase oscillations under methane oxidation conditions.\n
\n\n Oxidative treatment of Pd particles\n
\n\n Figure\n \n S1\n \n a shows a TEM image of the as obtained Pd NPs (Sigma Aldrich). The NPs feature an elongated shape with sizes ranging from 20 to 150 nm (Figure\n \n S1\n \n a). Selected-area electron diffraction (SAED) and high-resolution TEM (HRTEM) images reveal that the Pd particles are metallic with a\n \n fcc\n \n structure (Figure\n \n S1\n \n b, c). Prior to their use in methane oxidation, Pd NPs were calcined using a Micro-Electro-Mechanical System (MEMS)-based\n \n in situ\n \n nanoreactor in 20% O\n \n 2\n \n in He (\n \n p\n \n (O\n \n 2\n \n )\u2009=\u200936 mbar) at 300\u00b0C for 10 h to remove possible carbonaceous deposits and contaminants. Although the calcined Pd NPs show no obvious morphological changes (Fig.\n \n 1\n \n a), SAED analysis reveals the co-existence of both Pd and PdO phases (Fig.\n \n 1\n \n b), suggesting that Pd is partially oxidized during the calcination pretreatment. HRTEM imaging under 36 mbar O\n \n 2\n \n reveals a core-shell microstructure of the calcined NPs, with a metallic core encapsulated by an oxide shell that is ca. 2\u20135 nm thin (Fig.\n \n 1\n \n c).\n
\n\n
\n\n Influence of temperature and gas phase composition on particle dynamics, shape, and size\n
\n\n Starting from the calcined Pd NPs (Pd/PdO core/shell structures), we switched the gas phase from 20% O\n \n 2\n \n in He (\n \n p\n \n (O\n \n 2\n \n )\u2009=\u200936 mbar) to the reactive atmosphere containing 22% CH\n \n 4\n \n and 4.9% O\n \n 2\n \n (\n \n p\n \n (CH\n \n 4\n \n )\u2009=\u200939.5 mbar,\n \n p\n \n (O\n \n 2\n \n )\u2009=\u20098.8 mbar) in He at 350\u00b0C (\n \n p\n \n (total)\u2009=\u2009180 mbar).\n \n In situ\n \n imaging shows that there are no obvious changes of the Pd particles both during the gas switching and after stabilization of the gas phase. Next, we increased the temperature from 350 to 800\u00b0C within 10 min (Fig.\n \n 1\n \n d-g, Movie S1) to study how the particle shape and size respond to the change of temperature.\n \n In situ\n \n observation during this temperature increase shows no changes of the calcined NPs up to 460\u00b0C (Fig.\n \n 1\n \n d and Figure S2a, b). However, when the temperature is increased from 460 to 590\u00b0C, hillocks of reduced Pd start to appear and grow on the surface of the particles (Figure S2b-f and Fig.\n \n 1\n \n e). This process continues and results in a surface reconstruction and, eventually, fragmentation of particles. As the temperature increases further from 590 to 800\u00b0C, the particles begin to sinter (Figs.\n \n 1\n \n f, g and Figure S2g-i), finally leading to formation of metallic Pd at 800\u00b0C (Figure S3). Interestingly, the particle sintering induced by high temperature is reversible,\n \n i.e.\n \n large metallic particles split into smaller particles when the temperature is decreased from 800 to 550\u00b0C (Fig.\n \n 1\n \n h-k, Figure S4 and Movie S2). Having observed the effect of temperature, we further investigated how the gas phase composition influence the size and shape of Pd particles.\n \n In situ\n \n TEM observation while adding CH\n \n 4\n \n into the O\n \n 2\n \n /He flow at 550\u00b0C reveals a gradual fragmentation of the particles (Fig.\n \n 1\n \n l-o, Movie S3). SAED study reveals an increased Pd to PdO ratio after the CH\n \n 4\n \n addition (Figure S5). A similar fragmentation has been observed in the case of Cu particles in a redox atmosphere containing O\n \n 2\n \n and H\n \n 2\n \n ,(\n \n 34\n \n ) and was explained by oxidation and subsequent reduction of particles that occur repeatedly due to the co-presence of both reducing and oxidizing species at a comparable chemical potential. Continuous\n \n in situ\n \n observation further shows that the particles do not split into ever smaller particles but rather their size stabilizes around a certain range (ca. 5\u201345 nm, Figure S6) under the conditions applied, indicating that the particle size is a function of temperature and gas composition.\n
\n\n To summarize, the\n \n in situ\n \n observations made during the heating and subsequent cooling, as well as during the gas switching, demonstrate directly that the particle size and structural dynamics of Pd are dictated by the chemical potential of the gas phase. Since the particles of Pd adapt their shape, phase composition (Pd and PdO) and surface structure to the surrounding environment, the state of Pd observed\n \n ex situ\n \n , after passing through a temperature drop and change in atmosphere, does not represent the active state under reactive conditions.\n
\n\n Identification of the phase composition and detection of catalytic activity\n
\n\n To gain insights into the active state of Pd NPs during methane oxidation and uncover the structure-performance relationships, we turned to experiments that combine\n \n in situ\n \n TEM observations with online MS analysis of the effluent gas (\n \n operando\n \n TEM). As shown in Figs.\n \n 2\n \n a and\n \n 2\n \n b and Movie S4, the particles display no obvious structural dynamics at 350\u00b0C. This is evident from only a minor difference observed when comparing images taken at different times (Figs.\n \n 2\n \n a and\n \n 2\n \n b), as shown in Fig.\n \n 2\n \n c by green contrast (this contrast was obtained by comparing Figs.\n \n 2\n \n a and\n \n 2\n \n b, see SI for details). However, at 550\u00b0C the particles indeed show a dynamic behavior (Figs.\n \n 2\n \n e and\n \n 2\n \n f, Movie S5) that leads to constant morphological changes and migration of particles (Fig.\n \n 2\n \n g, green contrast).\n
\n\n \n In situ\n \n electron diffraction was used to identify phases present under these reaction conditions. Analyses of the SAED patterns and the corresponding radial intensity profiles indicate that Pd and PdO co-exist both at 350\u00b0C and 550\u00b0C (Figs.\n \n 2\n \n d,\n \n 2\n \n h). The diffraction spots show almost no changes with time at 350\u00b0C, implying no dynamic changes (Figure S7, Movie S6). In contrast, the diffraction spots are changing over time at 550\u00b0C (Figure S8, Movie S7), implying the presence of structural dynamics, in line with the TEM imaging. A comparison of PdO(101) to Pd(111) peak intensity ratio at 350 and 550\u00b0C reveals a higher relative fraction of Pd\n \n 0\n \n at 550\u00b0C (Figure S9), consistent with the decreasing oxygen chemical potential with temperature.(\n \n 42\n \n ,\n \n 43\n \n )\n \n In situ\n \n observation at medium magnifications further reveals that structural dynamics involve particle reshaping, sintering, outgrowth, and splitting (Fig.\n \n 2\n \n i-l, Movie S8). The presence of these dynamics is a consequence of the competing oxidizing and reducing processes near the Pd/PdO phase boundary. Note that the electron dose rates used for aforementioned\n \n in situ\n \n observations are considerably low,\n \n i.e.\n \n , merely 250\u2013700 e\u00b7nm\n \n \u2212\u20092\n \n s\n \n \u2212\u20091\n \n . Under such applied dose rates, no influence of the electron beam could be detected. This is further supported by a control experiment in which the electron beam was cut off between shots to minimize the extent of electron irradiation. This experiment shows significant changes in the particle shape and relative location, similar to these shown in Fig.\n \n 2\n \n e, f and Fig.\n \n 2\n \n i-l, confirming that the dynamics are not beam-induced (Figure S10).\n
\n\n Turning now to analysis of the gas composition by a mass spectrometer connected to the outlet of the\n \n in situ\n \n TEM nanoreactor (Fig.\n \n 2\n \n m), the collected MS data shows a sharp increase of the CO\n \n 2\n \n signal intensity and, simultaneously, a decrease of the CH\n \n 4\n \n and O\n \n 2\n \n signal intensity that coincide with the onset of redox dynamics at 350\u2013550\u00b0C. This data suggests clearly that the observed structural dynamics at 550\u00b0C are linked to catalytic activity. The CO signal is also present in the MS data (Figure S11), yet the intensity of the detected CO signal is consistent with that expected from the fragmentation of CO\n \n 2\n \n by the electron impact ionization method of the MS.(\n \n 44\n \n ) In other words, only CO\n \n 2\n \n is produced in the catalytic reaction even under the O\n \n 2\n \n -lean conditions used in this work, which agrees well with the previous studies that have been conducted under similar experimental conditions.(\n \n 41\n \n ,\n \n 45\n \n ) Yet, since the redox dynamics of individual NPs are mutually decoupled, MS data shows only integral (averaged) signal and therefore oscillations reported in previous studies(\n \n 41\n \n ,\n \n 45\n \n ) are not seen in the MS data of this work.\n
\n\n
\n\n High-resolution observations of redox dynamics and interfacial structures\n
\n\n Having demonstrated the catalytic activity of Pd NPs and its correlation with structural dynamics, we performed\n \n in situ\n \n high-resolution observations to obtain atomic-scale information about the transient structures of the catalyst under reaction conditions. It should be mentioned that high-resolution imaging typically requires high electron dose rates that may induce beam effects, such as beam-induced reduction.(\n \n 46\n \n ) We compared the structural dynamics recorded under high and low dose rates, and found qualitatively similar results (see Fig.\n \n 3\n \n and Figure S12), indicating that the applied dose rates for atomic-level imaging do not have a significant impact on the observed phenomena. Even though the electron irradiation might cause a change in the local chemical potential, it can be effectively compensated by varying either the partial pressure of gas phase or the temperature.\n \n In situ\n \n observations at atomic-scale not only confirm the presence of Pd and PdO phases and their oscillatory phase transition in individual particles (Fig.\n \n 3\n \n ), but reveal further that the interconversion takes place on the surfaces of both, metal and metal oxide crystallites, as discussed in detail below.\n
\n\n Figure\n \n 3\n \n a-c shows snapshots from Movie S9 exhibiting oscillatory phase transition at a metallic surface. At 550\u00b0C in the gas mixture of CH\n \n 4\n \n , O\n \n 2\n \n and He (\n \n p\n \n (CH\n \n 4\n \n )\u2009=\u200936.7 mbar,\n \n p\n \n (O\n \n 2\n \n )\u2009=\u200916.7 mbar and\n \n p\n \n (He)\u2009=\u2009126.6 mbar), we observe a periodic emergence and disappearance of a small oxide domain (green highlight) on the metal surface (red highlight). The transiently formed PdO domain shows semi-coherence with the underlying metallic particle, given by PdO(110) // Pd(200) interface. This phase epitaxy agrees with the previous report on the oxidation (or reduction) of Pd in undiluted oxygen (or hydrogen) and CO oxidation.(\n \n 47\n \n \u2013\n \n 50\n \n ) Due to the lattice mismatch between PdO and Pd, a slight tilting (2.5-3\u00b0) of the PdO(110) to Pd(200) is observed, which evidences the interfacial strain. The presence of lattice strain between metal and metal oxide leads to a shrinkage of PdO(110)\n \n d\n \n -spacing at the interface region from 2.25 to 2.12\u20132.13 \u00c5. The compressed lattice\n \n d-\n \n spacing is better visualized by the inverse Fourier transform image of PdO and the corresponding line profiles, as shown in Fig.\n \n 3\n \n d,e and\n \n 3\n \n f,g.\n
\n\n
\n\n Figure\n \n 3\n \n . Oscillatory phase transition between Pd and PdO on the metallic surface (\n \n a\n \n -\n \n c\n \n ) and the oxide surface (\n \n h-j\n \n ) at 550\u00b0C in the gas mixture of CH\n \n 4\n \n , O\n \n 2\n \n and He (\n \n p\n \n (CH\n \n 4\n \n )\u2009=\u200936.7 mbar,\n \n p\n \n (O\n \n 2\n \n )\u2009=\u200916.7 mbar and\n \n p\n \n (He)\u2009=\u2009126.6 mbar). Insets show FFTs of the corresponding HRTEM images. (\n \n d\n \n ,\n \n e\n \n ) Inverse Fourier transform image and line profiles of regions close to and away from the PdO/Pd interface. (\n \n f\n \n ,\n \n g\n \n ) Inverse Fourier transform image and line profiles of regions close to and away from the PdO/Pd interface. (\n \n k\n \n ) HRTEM image and (\n \n l\n \n ) enlarged HRTEM images from the dashed rectangle in (\n \n k\n \n ), revealing the presence of a monolayer PdO\n \n x\n \n on Pd. The inset of (\n \n l\n \n ) illustrates the atomic model. (\n \n m\n \n ) Lattice\n \n d\n \n -spacing analysis. Electron dose rates for (\n \n a\n \n -\n \n c\n \n ), (\n \n h\n \n -\n \n j\n \n ) and (\n \n k\n \n ) were 2.5\u00d710\n \n 5\n \n , 1.6\u00d710\n \n 5\n \n and 2.5\u00d710\n \n 5\n \n e\u00b7nm\n \n \u2212\u20092\n \n s\n \n \u2212\u20091\n \n , respectively.\n
\n\n In addition, the formation of a single layer PdO\n \n x\n \n on Pd is also identified (Fig.\n \n 3\n \n k). Structural analysis reveals that the lattice distance between the topmost layer and the second one is about 3.1 \u00c5, which is notably larger than lattice\n \n d\n \n -spacings of metallic Pd, suggesting that it is oxidic (Fig.\n \n 3\n \n l,m). The underlying Pd shows lattice fringes with a\n \n d\n \n -spacing of 2.25 \u00c5, corresponding to (111) planes of\n \n fcc\n \n structured Pd.\n
\n\n Overall,\n \n in situ\n \n high-resolution observations suggest that an active Pd catalyst is composed both of metal and metal oxide phases that interconvert dynamically under reaction conditions, in line with the results collected at a low magnification discussed above. In addition, surfaces of NPs contain no obvious carbonaceous deposits, indicating that the catalyst is efficient in transforming CH\n \n 4\n \n to CO\n \n 2\n \n . Associated with the oscillatory phase transition is the on-going formation of interfaces between metal and oxide domains. The strained coherent Pd/PdO interfaces might play a role in the catalytic process and will be discussed in more detail in the theoretical section of this work (\n \n vide infra\n \n ).\n
\n\n Surface composition and the electronic state of Pd NPs studied by NAP-XPS\n
\n\n To investigate the surface composition and the electronic state of Pd, NAP-XPS experiments were performed (experimental details are provided in the Supporting Information file). XPS data collected at 350\u00b0C in 1 mbar O\n \n 2\n \n (Fig.\n \n 4\n \n a) shows that the Pd\n \n 3d\n \n region contains peaks that can be fitted with two major components at a binding energy (BE) of 334.9 eV and 336.1 eV, assigned to Pd\n \n 0\n \n and Pd\n \n 2+\n \n electronic states, respectively.(\n \n 51\n \n ,\n \n 52\n \n ) The Pd\n \n 2+\n \n electronic state is likely due to the PdO phase observed in our TEM study described above. The incomplete oxidation of Pd to PdO (ca. 1:1 ratio of Pd\n \n 0\n \n to Pd\n \n 2+\n \n according to the fittings results shown in Table\n \n S1\n \n ) implies that under 1 mbar O\n \n 2\n \n and at 350\u00b0C the oxidation kinetics are slow (\n \n vide infra\n \n ).\n
\n\n
\n\n Increasing the temperature to 550\u00b0C in 1 mbar O\n \n 2\n \n leads to the evolution of peaks in the Pd\n \n 3d\n \n region such that only Pd\n \n 2+\n \n peak remains, explained by the oxidation of Pd\n \n 0\n \n and formation of PdO (Fig.\n \n 4\n \n b). The co-feeding of CH\n \n 4\n \n to the O\n \n 2\n \n flow (CH\n \n 4\n \n : 2.25 ml/min CH\n \n 4\n \n ; O\n \n 2\n \n : 0.5 ml/min) at 550\u00b0C leads to notable changes in the Pd\n \n 3d\n \n region (Fig.\n \n 4\n \n c). These changes are associated with the appearance of Pd\n \n 0\n \n state and the disappearance of Pd\n \n 2+\n \n state, explained by the reduction of PdO to metallic Pd. In addition, a feature with the intermediate (between Pd\n \n 2+\n \n and Pd\n \n 0\n \n ) BE energy appears at 335.3 eV, denoted Pd\n \n \u03b4+\n \n (dark-red line). A peak at this BE has been previously ascribed to a surface oxide PdO\n \n x\n \n (\n \n 52\n \n ,\n \n 53\n \n ). This assignment is in line with the\n \n in situ\n \n observation of the formation of a PdO\n \n x\n \n monolayer on the Pd metal seen in Fig.\n \n 3\n \n k. However, a partially reduced PdO surface may also contribute to the presence of this peak. Decreasing the temperature from 550\u00b0C to 350\u00b0C in the CH\n \n 4\n \n /O\n \n 2\n \n mixture results in a notable decrease in the intensity of the Pd\n \n 0\n \n peak, accompanied with the reappearance of the Pd\n \n 2+\n \n peak along with an increase of the relative intensity of the Pd\n \n \u03b4+\n \n peak (Fig.\n \n 4\n \n d). Those results correlate with a higher chemical potential of oxygen at 350\u00b0C relative to 550\u00b0C. The subsequent increase of temperature from 350\u00b0C to 550\u00b0C, without changing the gas-phase composition, leads to the disappearance of the Pd\n \n 2+\n \n peak and restores the ratio between Pd\n \n \u03b4+\n \n and Pd\n \n 0\n \n peaks (1:4.4) seen prior to the initial lowering the temperature from 550\u00b0C to 350\u00b0C, indicating a high reversibility of changes in the electronic states of Pd (Fig.\n \n 4\n \n c,e).\n
\n\n Turning now to the analysis of the O\n \n 1s\n \n region (Figure S13, Table S2), experiments with methane oxidation at 550\u00b0C display peaks with a BE of 536.4 eV and 535.0 eV, ascribed to H\n \n 2\n \n O and CO\n \n 2\n \n respectively, due to the presence of these gases near the specimen surface.(\n \n 54\n \n \u2013\n \n 56\n \n ) Consistent with the MS data of the\n \n operando\n \n TEM experiment, no XPS peak of gaseous CO is detected, suggesting the complete oxidation of CH\n \n 4\n \n . Notably, the peaks of the gas-phase CO\n \n 2\n \n and H\n \n 2\n \n O are almost invisible in the O\n \n 1s\n \n XPS region when measured at 350\u00b0C, confirming a lower catalytic activity at 350\u00b0C as compared to 550\u00b0C (Figure S13). Expectedly, when the temperature is increased from 350\u00b0C to 550\u00b0C, the contribution from methane combustion products, H\n \n 2\n \n O and CO\n \n 2\n \n , in the gas-phase increases (Figure S13), which correlates with the re-establishing of Pd\n \n \u03b4+\n \n and Pd\n \n 0\n \n with the ratio of ca. 1:4.4 (Fig.\n \n 4\n \n e).\n
\n\n To summarize, NAP-XPS results show clearly that the surface chemical state of the Pd NP catalyst is highly sensitive to the gas-phase composition and temperature (Fig.\n \n 4\n \n and Figure S13), underlying thereby the relevance of\n \n in situ\n \n characterization methods for the understanding of its active state. The formation of CO\n \n 2\n \n and H\n \n 2\n \n O at 550\u00b0C and at a lower rate at 350\u00b0C is consistent with the MS data collected during\n \n operando\n \n TEM study, demonstrating that the catalyst is active in the dynamic state, displaying the Pd\n \n \u03b4+\n \n : Pd\n \n 0\n \n ratio of 1:4.4 while producing merely the full oxidation products. Additional discussion on XPS data and results (Tables S3 and S4) based on first principles are provided in Supporting Information.\n
\n\n DFT calculations\n
\n\n DFT calculations were carried out to understand the nature of the phase transitions and help identify structures active in methane oxidation. These simulations were performed using the Quantum ESPRESSO package(\n \n 57\n \n ) with the GGA-PBE exchange and correlation potential, full computational details are given in the supporting information. Following the experimental observations, we computed the stability of different Pd(100)/O surface terminations as a function of the gas-phase chemical potential by way of\n \n ab initio\n \n atomistic thermodynamics. This thermodynamic analysis shows that PdO is the stable phase up to 635\u00b0C (or up to ca. 745\u00b0C with entropic corrections for the solid(\n \n 58\n \n )) at the experimental O\n \n 2\n \n pressure, while the single layer PdO\n \n x\n \n is only metastable (Figure S14). Therefore, the experimental observation of metallic Pd and single layer PdO\n \n x\n \n is driven by the kinetics of the methane oxidation reaction.\n
\n\n
\n\n To gain insight into the kinetically driven phase transitions, we computed the reaction energetics of methane oxidation on different possible surfaces revealed by\n \n in situ\n \n observations, i.e., clean Pd, strained PdO, unstrained PdO, and Pd/PdO (a PdO monolayer on Pd), see Supporting Information for details. While the majority of previous efforts have focused on the first C-H abstraction step owing to the difficulty of activating methane\u2014gas-phase initiation of the reaction is activated by 2.5 eV(\n \n 59\n \n )\u2014we simulate each step involved in methane oxidation to gain a better understanding of the complete path. We found the reaction proceeds by a sequential H-transfer from the adsorbed CH\n \n 4\n \n to adsorbed O on both the metal and oxide. On the Pd(100) surface, we found the first and second dehydrogenation steps of the adsorbed CH\n \n 4\n \n to be of similar energy and rate-limiting (Figure S15), i.e. the initial activation of methane is slow, as is expected by comparison to gas-phase energies. Estimating the barriers from a universal Br\u00f8nsted-Evans-Polyani relationship (BEP)(\n \n 60\n \n \u2013\n \n 63\n \n ) suggests an activation energy of about 1.0 eV for these steps (Table S5), which would make the metal surface an excellent catalyst when considering that the gas-phase barrier is 2.5 eV. However, dehydrogenation of methane on the metallic surface is predicted to be slow compared to surface oxidation through the dissociation of gas-phase O\n \n 2\n \n , where the barrier predicted from BEP is 0.7 eV (Table S5). Thus, the metallic surface phase is predicted to oxidize towards the thermodynamically favored oxide phase or a metastable surface oxide.\n
\n\n On the unstrained PdO surface, the reaction mechanism is qualitatively different from that predicted on Pd\n \n 0\n \n in that C-O bonds are formed with the H transfer steps. That is, a C\u2013O bond is formed as the first H is stripped from the adsorbed CH\n \n 4\n \n resulting in OH and O-CH\n \n 3\n \n (see Fig.\n \n 5\n \n and details in SI). Once O-CH\n \n 3\n \n formed, a second hydrogen atom is transferred to a second oxygen on the surface and the remaining O-CH\n \n 2\n \n fragment is further oxidized to form the OCH\n \n 2\n \n O species shown in Fig.\n \n 5\n \n . This OCH\n \n 2\n \n O fragment transfers a hydrogen atom to a surface oxygen to form a formate-like OCHO species. In the final step, hydrogen is transferred from formate to surface oxygen to yield CO\n \n 2\n \n . Our DFT calculations predict C-O bond formation significantly reduces the barriers associated with methane activation, which has two important consequences. Rather than the first step being rate-limiting, we find the third dehydrogenation (i.e., hydrogen transfer from OCH\n \n 2\n \n O to surface oxygen) is rate limiting on the oxide. And because both C-O bonds are formed through rapid steps at the start of methane oxidation before the rate limiting steps, complete oxidation should be observed over the oxide, as our MS data shows. Moreover, with a BEP estimated barrier of 0.2 eV (Table S6), the combustion step on the oxide is predicted to occur more rapidly than on the metallic surface. Because this small barrier is associated with surface reduction, it implies that the oxide surface may be reduced during methane oxidation, which could result in the formation of metallic Pd and, due to re-oxidation by oxygen, in oscillatory phase transitions. Under such conditions, PdO is strained due to the lattice mismatch with metallic Pd (see Fig.\n \n 3\n \n ). This strain may be expected to influence the catalytic performance. Our DFT calculations suggest that such strained PdO, whether multilayer (blue) or single-layer (red), will follow the same mechanism as unstrained PdO (green) but has more favorable energetics for methane combustion, giving an activation energy estimated from BEP to be near zero. Reoxidation of the strained PdO\n \n x\n \n surface, however, is predicted to be slower than its reduction, with barriers of 0.6 and 0.8 eV for oxygen dissociation on the strained bulk and single-layer PdO, respectively (Table S7). Thus, while the strained material is more active in combustion, re-oxidation remains a limiting factor, which will result in the presence of metallic phase. Thermodynamics will, however, push the catalyst back to PdO, thereby setting up redox dynamics and phase coexistence/oscillations. Given that the phase coexistence has been observed in many redox active metal catalysts, the kinetic hindrance of reoxidation of partially reduced oxide could be a general mechanism for the simultaneous presence of both metal and metal oxide as well as their dynamic interconversion in redox reactions.\n
\n\n Real-time observations during temperature or gas phase changes clearly reveal that the particle size, shape and surface structures of Pd NPs are a function of chemical potential of gas phase. Under the co-presence of CH\n \n 4\n \n and O\n \n 2\n \n at 550\u00b0C, neither metallic Pd nor PdO is present as a static species and a highly dynamic state characterized by phase coexistence and oscillatory transitions between Pd and PdO is observed. While phase coexistence and oscillations have been known from earlier studies,(\n \n 17\n \n ,\n \n 41\n \n ,\n \n 64\n \n ) the real-time and space information of the associated dynamics are not well documented despite those might be the key to understand the catalytic function. In particular, both our\n \n operando\n \n TEM and these earlier studies demonstrate that such redox dynamics are correlated with the catalytic activity. The observation of the PdO \u2192 Pd phase transition points directly towards a Mars-Van Krevelen-like (MvK-like) mechanism, as the lattice O in PdO is consumed by methane.(\n \n 65\n \n ) Once the lattice O in PdO is depleted, it transforms into metallic Pd. Subsequently, the dissociative adsorption of O\n \n 2\n \n on Pd leads to the reformation of PdO, through which the activity can be regenerated. Considering that PdO is the thermodynamically stable phase under these conditions, the presence of metallic Pd demonstrates the key role of reaction kinetics in determining the chemical state of the Pd catalysts.\n
\n\n High-resolution imaging further reveals the occurrence of oscillatory phase transition on the surface of both, metal and oxide particles, with the transient formation of strained and coherent interface between Pd and PdO (Fig.\n \n 3\n \n ). Building on these atomic details, we have constructed models for use in first principles to understand their catalytic function.\n \n Ab initio\n \n simulations reveal that while PdO is thermodynamically stable, oscillatory phase transition occurs because the oxide is more effective in activating the C-H bonds of methane than the O-O bond of O\n \n 2\n \n gas (Figure S15 and Fig.\n \n 5\n \n ). The C-H activation and the subsequent combustion step can lead to the reduction of oxide to metal. Conversely, the metal is ineffective at activating C-H bonds but strongly activates O\n \n 2\n \n , causing oxidation of metal. Thus, the preferential activation of reductant on the oxide and oxidant on the metal possibly induce the oscillatory phase transitions between the two states.(\n \n 66\n \n ) The appearance of strained PdO at the Pd/PdO interfaces during these phase oscillations can further enhance C-H bond activation to improve the catalytic performance. Our results thus suggest that one phase is not equally good at activating both reductant and oxidant in the gas phase, and an improved catalytic activity may be accessible if the system can be driven and stabilized at a dynamic state characterized by phase coexistence and cooperation.(\n \n 10\n \n ,\n \n 17\n \n ,\n \n 19\n \n ) In combination with our previous works on copper under different redox reactions,(\n \n 34\n \n ,\n \n 67\n \n ,\n \n 68\n \n ) we can conclude that the emergence of catalytic activity is related to the dynamic interplay between coexisting phases, which plays a general mechanism for redox-active metal catalysts.\n
\n\n In summary, this work provides insights into the active structures of Pd catalysts and explains the origin of phase coexistence and oscillations in methane oxidation, which are of fundamental significance in deepening our understanding of Pd-based methane combustion system and other metals-based redox catalytic systems. The dynamic picture of the constantly generated active sites/structures revealed by\n \n operando\n \n TEM, however, cannot be obtained by\n \n ex situ\n \n and\n \n post mortem\n \n characterizations or ensemble-average\n \n in situ\n \n techniques, and thus emphasizes the importance of\n \n operando\n \n TEM in uncovering the dynamic nature and catalytically relevant processes of catalysts. This work also highlights the importance of complementary use of other\n \n in situ\n \n tools in catalysis research for advancing our understanding.\n
\n\n Particle dynamics during temperature increase in CH4/O2 mixture\n
\n \n\n Particle dynamics during temperature decrease in CH4/O2 mixture\n
\n \n\n Particle dynamics during addition of CH4 into O2 flow\n
\n \n\n Particle dynamics at 350 \u00b0C in CH4/O2 mixture\n
\n \n\n Particle dynamics at 550 \u00b0C in CH4/O2 mixture\n
\n \n\n In situ electron diffractions at 350 \u00b0C in CH4/O2 mixture\n
\n \n\n In situ electron diffractions at 550 \u00b0C in CH4/O2 mixture\n
\n \n\n Particle recontruction, sintering and splitting (outgrowth) dynamics at 550 \u00b0C in CH4/O2 mixture\n
\n \n\n Atomic-scale observation of surface redox dynamics on a metallic particle\n
\n \n\n Atomic-scale observation of surface redox dynamics on an oxide particle\n
\n \n\n Supplementary file containing methods, additional disscussion and results\n
\n \n\n Some specific chemotherapeutic drugs are able to enhance tumor immunogenicity and facilitate antitumor immunity by inducing immunogenic cell death (ICD). However, tumor immunosuppression induced by the adenosine pathway hampers this effect. In this study, E-selectin-modified thermal-sensitive micelles were designed to co-deliver a chemotherapeutic drug (doxorubicin, DOX) and an A2A adenosine receptor antagonist (SCH 58261), which simultaneously exhibited chemo-immunotherapeutic effects when applied with microwave irradiation. After intravenous injection, the fabricated micelles, ES-DSM, effectively adhered to the surface of leukocytes in peripheral blood mediated by E-selectin, and thereby hitchhiking with leukocytes to achieve a higher accumulation at the tumor site. Further, local microwave irradiation was applied to induce hyperthermia and accelerated the release rate of drugs from micelles. Rapidly released DOX induced tumor ICD and elicited tumor-specific immunity, while SCH 58261 alleviated immunosuppression caused by the adenosine pathway, further enhancing DOX-induced antitumor immunity. In conclusion, this study presents a strategy to increase the tumor accumulation of drugs by hitchhiking with leukocytes, and the synergistic strategy of chemo-immunotherapy not only effectively arrested primary tumor growth, but also exhibited superior effects in terms of antimetastasis, antirecurrence and antirechallenge.\n
\n\n \n \n Oncology\n \n \n
\n\n \n thermal-sensitive micelles\n \n
\n\n \n leukocytes\n \n
\n\n \n immunogenic cell death\n \n
\n\n \n adenosine\n \n
\n\n \n immunosuppression\n \n
\n\n Several chemotherapeutic drugs, especially anthracyclines, have been repurposed to provoke antitumor immune responses by inducing immunogenic cell death (ICD) in addition to direct tumor killing effects.\n \n 1\n \n Tumor ICD is accompanied by the release of damage-associated molecular patterns (DAMPs), including the exposure of calreticulin (CRT), secretion of adenosine triphosphate (ATP), and release of high mobility group protein B1 (HMGB1).\n \n 2-6\n \n These DAMPs have been identified to facilitate dendritic cell (DC) maturation and antigen presentation to na\u00efve T cells.\n \n 7, 8\n \n Subsequently, the activation of T cells leads to the recruitment of cytotoxic T cells (CTLs) to the tumor site, thereby promoting tumor-specific cellular immunity, which can further enhance antitumor effects of chemotherapeutic agents.\n \n 9, 10\n \n
\n\n Despite the ICD induction and immune response initiation of these select chemotherapeutic drugs, there remain challenges. Tumor cells can release large amounts of ATP during ICD induced by chemotherapeutic drugs, which is subsequently metabolized to adenosine (ADO, a potent immunosuppressor) by ectonucleotidases, such as CD39 and CD73.\n \n 11\n \n The engagement of ADO and A2A ADO receptors (A2AR, an immune checkpoint) on various immune cell surfaces hampers the immune reaction toward tumor cells, further exacerbating tumor immunosuppression.\n \n 12-14\n \n Therefore, the paradoxes between ICD-induced antitumor immunity and ADO-mediated immunosuppression remain a formidable challenge. Fortunately, preclinical studies targeting the adenosinergic pathway have gained much attention for their clinical potential in overcoming tumor-induced immunosuppression. Blockade of the ectonucleotidases that generate ADO, or the A2AR that mediates adenosinergic signals in immune cells, will greatly contribute to restraining tumor growth and metastasis.\n \n 15-19\n \n This suggests the possible benefits of utilizing ADO-related therapeutic approaches in combination with chemotherapeutic drugs with ICD induction ability. In particular, antagonists of A2AR are just occurring to be deployed into oncology, which can block the interaction between ADO and A2AR, thereby alleviating tumor immunosuppression and facilitating the antitumor immune response.\n \n 20, 21\n \n It is worth noting that A2AR is widely distributed on a variety of immune cells and is a ubiquitous immune checkpoint, which holds promise for addressing the low response rate of PD-1/PD-L1 blockade therapies.\n \n 19\n \n Therefore, the combined application of chemotherapeutic drugs and A2AR antagonists may amplify antitumor efficacy.\n
\n\n However, both chemotherapeutic drugs and A2AR antagonists have limited tumor targeting ability after intravenous administration, which often induces undesirable adverse effects and unsatisfactory efficacy. Smart nanoparticle drug delivery system is an effective way to alter biodistribution of drugs and achieve spatiotemporally controlled drug release, which is beneficial for improving treatment safety and efficacy.\n \n 9, 22-24\n \n Significantly, thermal-sensitive drug delivery system has attracted much attention; hyperthermia stimuli at the tumor site can accelerate the drug release from nanoparticles to achieve precise therapy, and on the other hand, hyperthermia itself can also suppress tumor growth.\n \n 25, 26\n \n Despite these advantages, delivering nanoparticle platforms in patients with advanced forms of cancer remains a challenge. Only a fraction of all drug-loaded nanoparticles can reach the tumor site, while the vast majority of nanoparticles are cleared by the reticuloendothelial system (RES), and the clinical translation of the EPR effect from animal models to humans has been proven to be challenging.\n \n 27\n \n Additionally, elevated fluid pressures and the lack of well-defined vasculature also hinder the application of nanoparticles in tumor therapy.\n \n 28-30\n \n
\n\n A strategy that potentially addresses the challenges listed above and optimizes biodistribution in a highly specific manner involves the use of circulating cells to mediate the transport of drug-loaded nanoparticles.\n \n 31-35\n \n Specifically, leukocytes, which share similar migration patterns to tumor cells in blood and tissues,\n \n 36\n \n can also be utilized to carry drug-loaded nanoparticles and pass challenging biological barriers to accumulate in tumor sites.\n \n 37, 38\n \n
\n\n Inspired by the natural tumor targeting capacity of leukocytes, we herein fabricated E-selectin-modified thermal-sensitive micelles (ES-DSM), which were co-loaded with the chemotherapeutic drug doxorubicin (DOX) and the A2AR antagonist SCH 58261 (hereafter referred to as SCH). After intravenous administration, the ES-DSM could hitch a ride on leukocytes mediated by E-selectin to across biological barriers and achieve increased tumor accumulation. Subsequently, local microwave stimulation was applied to induce hyperthermia and accelerated the release rate of drugs from nanoparticles. Rapidly released DOX not only directly killed tumor cells but also improved tumor immunogenicity by inducing ICD. The maturation and antigen presentation of DCs were facilitated, and further tumor-specific T cell immunity was elicited. On the other hand, released SCH prevented the engagement of ADO with A2AR on the surface of various immune cells, which relieved the immunosuppression phenomenon and further enhanced DOX-induced tumor-specific cellular immunity (\n \n Scheme 1\n \n ). Consequently, considerably enhanced antitumor efficacy might be achieved via the synergistic effect of chemo-immunotherapy.\n
\n\n (see Scheme 1 in the Supplementary Files)\n
\n\n \n Characterization of NTA-PEG-p-(AAm-co-AN)\n \n
\n\n First, the amphiphilic polymer NTA-PEG-p-(AAm-co-AN) (\n \n Figure 1a\n \n ) was synthesized according to\n \n Scheme S1\n \n . The chemical structure of the polymers was confirmed by\n \n 1\n \n H NMR spectra as shown in\n \n Figure 1b and S1\n \n . The molecular weights of p-(AAm-co-AN) and PEG-p-(AAm-co-AN) were measured as 10.9 kDa and 14.3 kDa respectively. To evaluate the thermal sensitivity of the polymer, turbidity measurements were performed to determine the upper critical solution temperature (UCST) of p-(AAm-co-AN). As shown in\n \n Figure 1c\n \n , the transmittance of the polymer solution increased from 4\u00b0C to 43\u00b0C and became constant above 43 \u00b0C, which confirmed that the UCST value of the polymer was 43\u00b0C. Further, synthesized NTA-PEG-p-(AAm-co-AN) was found to self-assemble into micelles in aqueous solution at ambient temperature, and the critical micelle concentration (CMC) was determined to be 33.2 \u03bcg/mL (\n \n Figure 1d\n \n ). Importantly, blank micelles that self-assembled from NTA-PEG-p-(AAm-co-AN) were proven to be thermal-sensitive. As exhibited in\n \n Figure 1e\n \n , blank micelles presented regular and uniform spherical morphologies at both 25\u00b0C and 37 \u00b0C, but irregular shapes at 43\u00b0C and 50\u00b0C, supporting the stability of blank micelles at physiological temperature (37\u00b0C) as well as their destruction under hyperthermic condition (43\u00b0C). NTA in the polymer was used to chelate Ni\n \n 2+\n \n to afford Ni-NTA, which could further efficiently bind to the His-tag of recombinant E-selectin, thereby introducing E-selectin onto the surface of micelles. The chelating ability of NTA-PEG-p-(AAm-co-AN) to Ni\n \n 2+\n \n was demonstrated by ICP-MS, and the result showed that 0.96 mol of Ni\n \n 2+\n \n could be chelated per mole of the polymer.\n
\n\n \n Characterization of E-selectin-modified, DOX and SCH co-loaded micelles (ES-DSM)\n \n
\n\n Subsequently, DOX and SCH co-loaded micelles (DSM) were prepared with feed ratios of DOX and SCH of 4% and 1%, respectively. The encapsulation efficiency and drug loading of DOX were 92.9\u00b10.61% and 2.7\u00b10.01%, respectively, while those of SCH were 41.8\u00b10.97% and 0.41\u00b10.005%, respectively. Further, E-selectin was introduced onto the micelle surface to obtain ES-DSM. As shown in\n \n Figure S2\n \n , as E-selectin modifications increased, the particle size of ES-DSM increased, while the potential decreased. ES-DSM applied in this study was prepared by adding 2 \u03bcg/mL E-selectin into a solution of 1 mg/mL polymer.\n \n Figure 1f\n \n showed that the particle size and potential of DSM were 164.0\u00b17.0 nm and 3.93\u00b10.05 mV, respectively. However, when E-selectin was introduced onto micelles to form ES-DSM, the particle size increased to 247.7\u00b115.6 nm while the potential decreased to -1.2\u00b10.09 mV, which further proved that the preparation of ES-DSM was successful. The spherical morphology of ES-DSM was also observed by TEM (\n \n Figure 1g\n \n ).\n
\n\n Further, the thermal sensitivity of micelles was investigated by determining particle sizes at different temperatures. As presented in\n \n Figure 1h\n \n , the size of blank micelles remained below 100 nm at 5 to 37\u00b0C, while it was almost undetectable at 43\u00b0C and above, which was consistent with the TEM results in\n \n Figure 1e\n \n . Importantly, the sizes of DSM and ES-DSM increased to more than 1000 nm when detected at 43\u00b0C and above, which was due to the dissolution of the micelles under thermal conditions, and the insoluble drugs DOX and SCH were released immediately to form precipitates. Afterwards, the thermal-sensitive in vitro drug release behavior of ES-DSM was evaluated by the dialysis method at 37 and 43\u00b0C. As shown in\n \n Figure 1j and k\n \n , under physiological condition (37\u00b0C), the drug release rates were relatively slow, and approximately 40% and 50% of SCH and DOX were released, respectively, within 48 hours. However, under thermal condition (43\u00b0C), the release rates of SCH and DOX were considerably accelerated and were similar to the profile of free drugs. The rapid drug release behavior of ES-DSM at 43\u00b0C was the result of micelle disintegration.\n
\n\n Subsequently, the specific recognition ability of ES-DSM to leukocytes was evaluated. Both DSM and ES-DSM were demonstrated to be biocompatible with leukocytes and had no significant impact on cell viability or penetration ability (\n \n Figure S3\n \n ). At different times after the intravenous injection of DSM or ES-DSM, leukocytes were isolated, and the fluorescence intensity of DOX was detected by flow cytometry.\n \n Figure 1l and S4\n \n showed that the fluorescence intensity of leukocytes exhibited a negligible change within 24 hours after DSM injection but was significantly enhanced after ES-DSM injection, and approximately 30% of leukocytes were DOX positive at 24 h post-injection. In addition, leukocytes were isolated 24 h after injection and observed by confocal microscopy, which demonstrated that ES-DSM adhered to the surface of leukocytes (\n \n Figure 1i\n \n ). Taken together, in contrast to DSM, ES-DSM presented an efficient leukocyte targeting ability and adhered to the surface of leukocytes, further emphasizing the important role of E-selectin in the hitchhiking of micelles to leukocytes.\n
\n\n \n Cellular drug release, cytotoxicity and ICD induction ability of ES-DSM supplemented with hyperthermia\n \n
\n\n Next, the thermal-sensitive drug release behavior at the cellular level was investigated by confocal microscopy. First, Nile red was used as the model drug to prepare Nile red-loaded micelles. When 4T1 cells were exposed to Nile red-loaded micelles and treated with hyperthermia (+), Nile red was released rapidly and bound with the intracellular lipid membrane, and fluorescence was observed, which was similar to free Nile red. However, cells without hyperthermia (-) exhibited weaker fluorescence intensity because the drug was not released (\n \n Figure 2a and S5\n \n ). In addition, when 4T1 cells were exposed to DOX-loaded micelles, after being treated with hyperthermia (+), DOX was liberated and obviously entered the nucleus, which was similar to free DOX. When treated without hyperthermia (-), DOX resided in micelles and was therefore mainly distributed in the cytoplasm (\n \n Figure 2b\n \n ). These results indicated the thermal-sensitive nature of drug-loaded micelles at the cellular level.\n
\n\n Then, the cytotoxicity of free DOX and SCH (DS), DSM and ES-DSM was assessed. Initially, the biocompatibility of blank micelles was confirmed, and hyperthermia treatment did not affect 4T1 cell viability (\n \n Figure S6\n \n ). After exposure to DS, DSM or ES-DSM with or without hyperthermia, 4T1 cell viability was measured by MTT assay. In\n \n Figure 2c and d\n \n , there was no significant difference in cytotoxicity between the groups of DS supplemented with or without hyperthermia (IC\n \n 50\n \n values were 8.50 and 8.45 \u03bcM, respectively). However, compared to the DSM and ES-DSM treated groups (IC\n \n 50\n \n values were 30.70 and 29.35 \u03bcM, respectively), the hyperthermia treated groups exhibited higher cytotoxicity (IC\n \n 50\n \n values were 11.25 and 10.50 \u03bcM, respectively), which was similar to the toxicity of free drugs (DS). The reason for this difference was that the drugs could be released immediately from micelles under the thermal condition to execute their tumor cell killing function. Importantly, the modification of E-selectin exhibited negligible interference on the cytotoxicity of drug-loaded micelles. In addition, 4T1 cell apoptosis induced by different treatments was detected by flow cytometry. As displayed in\n \n Figure 2e and f\n \n , the DSM and ES-DSM treated groups supplemented with hyperthermia presented more severe early and late apoptosis than the unheated groups. All of these results indicated that the drug-loaded micelles applied with hyperthermia exhibited more effective antitumor effect than the unheated groups, which was attributed to the thermal-sensitive release behavior of drugs from micelles.\n
\n\n In addition, the ICD induction ability of drug-loaded micelles was analyzed. DOX can efficiently induce ICD in tumors, which is accompanied by the exposure of CRT, secretion of ATP, and release of HMGB1 (\n \n Figure 2g\n \n ). Therefore, we tested whether enhanced CRT, ATP and HMGB1 were observed when 4T1 cells were incubated with different agents with or without hyperthermia.\n \n Figures 2h-k\n \n showed that hyperthermia promoted the exposure of ICD biomarkers induced by DSM and ES-DSM. The levels of CRT, ATP and HMGB1 increased when the drug-loaded micelles were combined with hyperthermia, which was similar to the free drugs.\n
\n\n \n Maturation of DCs in the binary co-incubation system\n \n
\n\n During the ICD process of tumor cells, CRT is overexpressed and provides an \u201ceat-me\u201d signal for dendritic cell uptake,\n \n 4, 5\n \n while released HMGB1 and ATP serve as adjuvant stimuli for dendritic cell maturation (\n \n Figure 3a\n \n ).\n \n 39\n \n Therefore, after 4T1 cells were exposed to different agents with or without hyperthermia and incubated for 24 h, immature DCs were added to co-incubate for another 48 h, and biomarkers of mature DCs (CD80, CD86 and MHC \u2161) were analyzed by flow cytometry. As shown in\n \n Figure 3b-c, f-g and S7\n \n , when 4T1 cells were pretreated with DSM or ES-DSM and hyperthermia, they promoted the maturation of DCs. The expression of CD80, CD86 and MHC \u2161 was similar to that in the free drug (DS) treated groups but significantly higher than that in the unheated DSM or ES-DSM treated groups. Moreover, immunologic factors secreted by DCs were monitored by ELISA kits.\n \n Figure 3h-j\n \n demonstrated that levels of IL-12p70 (a DC-secreted immune-related cytokine) and IL-6 in the suspension of the co-incubation system increased while IL-10 decreased when DSM or ES-DSM were applied in combination with hyperthermia, which was consistent with the DS treated groups. These results further supported the thermal-sensitive property of the drug-loaded micelles and that the ICD of tumor cells facilitated DC maturation.\n
\n\n It is worth noting that ADO in the tumor environment can bind to A2AR on the DC surface, thereby inhibiting DC maturation and antigen presentation. SCH serves as an antagonist to block the interaction between ADO and A2AR at the DC surface, further relieving the immunosuppression of DCs (\n \n Figure 3a\n \n ). To verify the effect of SCH on the immune response, 1 \u03bcM of NECA (an analog of ADO) was added to the co-incubation system to simulate the tumor microenvironment,\n \n 40\n \n and then DC maturation was evaluated. As displayed in\n \n Figure d-e, k-l and S8\n \n , when only DOX (groups of D, DM and ES-DM with hyperthermia) was in the co-incubation system, the expression of CD80, CD86 and MHC \u2161 was lower than that of the groups containing both DOX and SCH (groups of DS, DSM and ES-DSM with hyperthermia), which also exhibited more secretion of IL-12p70 and IL-6 but less IL-10 (\n \n Figure 3m-o\n \n ). These results showed that the presence of NECA arrested the maturation of DCs, but SCH relieved this phenomenon by blocking the interaction between NECA and A2AR.\n
\n\n \n Activation of T cells in the ternary co-incubation system\n \n
\n\n Mature DCs facilitated by tumor ICD can present antigens to na\u00efve T cells, further promote their differentiation into cytotoxic T cells (CTLs) or regulatory T cells (Tregs), and finally elicit T cell immune responses (\n \n Figure 4a\n \n ). Therefore, a ternary co-incubation system of 4T1 cells (which had been pretreated with different agents with or without hyperthermia), immature DCs, and splenic lymphocytes was constructed and cultured for 48 h. Subsequently, the proliferation of CD3\n \n +\n \n CD4\n \n +\n \n and CD3\n \n +\n \n CD8\n \n +\n \n T cells was analyzed. As exhibited in\n \n Figure 4b and d-e\n \n , when 4T1 cells were pretreated with DSM or ES-DSM in combination with hyperthermia, both CD3\n \n +\n \n CD4\n \n +\n \n and CD3\n \n +\n \n CD8\n \n +\n \n T cells in the co-incubation system proliferated significantly and were more abundant than those in unheated groups. The negligible difference between the drug-loaded micelles with hyperthermia and free drugs treated groups suggested that the thermal-sensitive drug release behavior enabled micelles to execute the efficient antitumor effect. Further, CD4\n \n +\n \n Foxp3\n \n +\n \n T cells, known as regulatory T cells (Tregs), which can hamper effective antitumor immunity, were obviously decreased when DSM and ES-DSM were applied with hyperthermia, suggesting that tumor ICD effectively stimulated T cell immunity and weakened the immunosuppressive effect of Tregs (\n \n Figure S9\n \n ). Besides, the cytokines (TNF-\u03b1, IL-2 and IFN-\u03b3) secreted by lymphocytes in the co-incubation system treated with drug-loaded micelles with hyperthermia exhibited a trend similar to that of the free drug groups (\n \n Figure 4f-h\n \n ). These results proved that the 4T1 cell ICD induced by thermal-sensitive drug-loaded micelles facilitated the antigen presenting ability of DCs to na\u00efve T cells, further promoting their differentiation into CTLs rather than Tregs.\n
\n\n Importantly, ADO can interact with A2AR on the surface of T cells to inhibit the antitumor effect of CTLs and facilitate the immunosuppressive impact of Tregs. Fortunately, SCH can block the interaction between ADO and A2AR on the T cell surface, thereby reversing the undesired immunosuppressive phenomenon (\n \n Figure 4a\n \n ). To verify this effect, 1 \u03bcM of NECA was added to the ternary co-incubation system, and the percentages of CD3\n \n +\n \n CD4\n \n +\n \n , CD3\n \n +\n \n CD8\n \n +\n \n and CD4\n \n +\n \n Foxp3\n \n +\n \n T cells were detected.\n \n Figure 4c, i-j and S10\n \n showed that the application of SCH (groups of DS, DSM and ES-DSM with hyperthermia) liberated T cells from the negative impact of NECA and promoted the proliferation of antitumor T cells. In addition, the levels of secreted cytokines (TNF-\u03b1, IL-2 and IFN-\u03b3) also demonstrated the anti-immunosuppressive effect of SCH (\n \n Figure 4k-m\n \n ).\n
\n\n \n In vivo antitumor efficacy of ES-DSM with microwave radiation (MW)\n \n
\n\n Next, the biodistribution of drug-loaded micelles was investigated in 4T1 tumor-bearing mice, and ICG was used as the model drug. ICG-loaded micelles with or without E-selectin modification were intravenously injected. As shown in\n \n Figure 5a and S11\n \n , ICG-loaded micelles with or without E-selectin modification accumulated at the tumor site. However, E-selectin-modified micelles exhibited less liver accumulation and more tumor targeting. Further, CD45 (a biomarker of leukocytes) in tumor sections was labeled and observed. As displayed in\n \n Figure 5b\n \n , the fluorescence of ICG (red) and CD45 (green) overlapped obviously after the injection of E-selectin-modified ICG-loaded micelles, indicating that the increase of micelles in tumors was benefited from hitching a ride on leukocytes.\n
\n\n Thereafter, the antitumor efficacy of DSM and ES-DSM was explored and the treatment regimen was displayed in\n \n Figure 5c\n \n . Mice were intravenously (i.v.) injected with different agents every 3 days, and in situ microwave thermotherapy was performed 24 h after i.v. injection, for 4 consecutive doses. In addition, to examine the effect of CD8\n \n +\n \n T cells on the antitumor immune response, an anti-CD8 antibody was intraperitoneally (i.p.) injected every 3 days to deplete CD8\n \n +\n \n T cells starting on day -3. The body weight of the free drug treated group (DS+MW) decreased significantly compared to that of the other drug-loaded micelle groups, suggesting that the micelles reduced the side effects of free drugs (\n \n Figure 5d\n \n ). Changesin tumor volume were shown in\n \n Figure 5e-f and S12\n \n , and the photograph of tumor tissues at the end of the observation period was displayed in\n \n Figure 5h\n \n . Compared with the group treated with saline (Saline), the application of microwave radiation (Saline+MW) exhibited negligible efficacy, and the tumor inhibition rate was approximately 7.2%. Mice treated with free drugs and microwave hyperthermia (DS+MW) showed a tumor inhibition rate of about 47.8%. Importantly, drug-loaded micelles plus microwave hyperthermia (DSM+MW) exhibited a better efficacy (approximately 73.5%). It is worth noting that, in comparison with the DSM+MW group, the E-selectin-modified drug-loaded micelles combined with microwave hyperthermia (ES-DSM+MW) group presented a better tumor inhibition effect (about 87.7%), which was due to the satisfactory tumor targeting efficiency of ES-DSM mediated by leukocytes. In addition, when applied without microwave radiation, ES-DSM treated mice exhibited a poor antitumor effect with an inhibition rate of about 33.6%, which was because the drugs were trapped in the micelles without hyperthermia stimulation and could not be released to execute their function. Further, the E-selectin-modified DOX-loaded micelles supplemented with microwave radiation (ES-DM+MW) group exhibited an approximately 49.8% tumor inhibition rate, which was not as effective as that of ES-DSM+MW group, suggesting the important role of SCH in antitumor efficiacy. Moreover, there was a negligible antitumor effect when CD8\n \n +\n \n T cells of mice were depleted (ES-DSM+MW+anti-CD8), indicating that CD8\n \n +\n \n T cells were indispensable for the antitumor efficacy. Furthermore, the survival time of mice in the ES-DSM+MW group was significantly prolonged compared to that of the other groups (\n \n Figure 5g\n \n ). Further, tumor tissues of different groups were collected and used for pathological study. TUNEL (\n \n Figure 5j\n \n ) and H&E (\n \n Figure S13\n \n ) staining of tumor tissues definitely proved that ES-DSM+MW led to a large amount of cell apoptosis and necrosis compared to that in the other groups.\n
\n\n Metastasis is one of the most important reasons for high mortality in cancer patients. Therefore, pulmonary metastasis in each group of mice was evaluated. At the end of the observation period, lung tissues were collected for the observation of metastatic tumor nodules.\n \n Figure 5i and k\n \n suggested that ES-DSM applied with microwave hyperthermia remarkably suppressed pulmonary metastasis compared to other treatments. This conclusion was further verified by the H&E staining of lung tissues (\n \n Figure 5l\n \n ). All of these results indicated that ES-DSM+MW efficiently prevented pulmonary metastasis in tumor-bearing mice.\n
\n\n \n Immune response elicited by ES-DSM with microwave radiation (MW)\n \n
\n\n Further, the in vivo immune response elicited by ES-DSM+MW was investigated. First, mature DCs in tumors and sentinel lymph nodes (SLNs) were analyzed by flow cytometry. As exhibited in\n \n Figure S14 and S15\n \n , biomarkers of mature DCs (CD80\n \n +\n \n and CD86\n \n +\n \n ) in the ES-DSM+MW group were significantly higher than those in the other groups. Since primary CTLs (CD8\n \n +\n \n T cells) responses are important in suppressing tumor growth and helper T cells (CD4\n \n +\n \n T cells) play important roles in the regulation of adaptive immunity, they are considered critical effectors for cancer immunotherapy.\n \n 41\n \n Therefore, at the end of the observation period, PBMCs, spleens (\n \n Figure S16\n \n ) and tumors (\n \n Figure 6a and S17a-b\n \n ) were obtained from each group and T cells were measured by flow cytometry. In comparison to the other groups, the ratios of CD3\n \n +\n \n CD4\n \n +\n \n and CD3\n \n +\n \n CD8\n \n +\n \n T cells were considerably increased in the ES-DSM+MW group. In contrast, CD4\n \n +\n \n Foxp3\n \n +\n \n T cells, known as regulatory T cells (Tregs), which can hamper effective antitumor immunity, were significantly decreased in the tumor tissue of the ES-DSM+MW treated group (\n \n Figure 6b and S17c\n \n ). Further, tumor-specific memory T cells (TMEs) were analyzed by detecting the ratio of CD8\n \n +\n \n CD44\n \n +\n \n T cells. A remarkable increase in the percentage of TEMs in both spleens\n \n (Figure 6c and e)\n \n and tumors\n \n (Figure 6d and f)\n \n was observed, suggesting strong immune surveillance in mice after ES-DSM+MW treatment. Subsequently, antitumor cytokine levels (TNF-\u03b1, IFN-\u03b3 and IL-2) in the serum, spleen and tumor of mice were measured and displayed in\n \n Figures 6g-i\n \n . The results suggested that cytokine levels of mice in the ES-DSM+MW group were the highest, indicating the best antitumor immune response. Taken together, the immune response in the ES-DSM+MW group was stronger than that of the DSM+MW and ES-DM+MW groups, which was due to the better tumor targeting ability mediated by E-selectin and the anti-immunosuppressive effect of SCH. Moreover, when ES-DSM were applied without MW, the immune response in mice was unsatisfactory because the drugs were difficult to be released from the micelles to execute antitumor functions.\n
\n\n The exposure of DAMPs during tumor ICD was an important factor in eliciting antitumor immunity; therefore, levels of CRT and HMGB1 in tumor tissues after different treatments were examined. As\n \n Figure 6j-k\n \n displayed, ES-DSM+MW treatment induced dramatic increases in CRT and HMGB1 in tumor tissues, supporting the remarkable ICD induction ability of this strategy. Tumor-infiltrating CD8\n \n +\n \n T cells (\n \n Figure 6l\n \n ), CD69\n \n +\n \n T cells (\n \n Figure S18a\n \n ) and perforin (\n \n Figure S18b\n \n ) were also increased after ES-DSM+MW treatment. In contrast, the biomarker of Tregs, Foxp3, was significantly reduced (\n \n Figure 6m\n \n ). Altogether, these results demonstrated that the combination of ES-DSM and microwave thermotherapy induced strong ICD and generate a robust immune response at the tumor site.\n
\n\n \n Antimetastasis, antirecurrence and antirechallenge efficacy of ES-DSM with microwave radiation (MW)\n \n
\n\n To further confirm the treatment efficacy of ES-DSM+MW on the inhibition of pulmonary metastasis, a 4T1 pulmonary metastatic tumor model was established by injecting Luc-4T1 cells into mice via the tail vein, followed by different treatments (\n \n Figure 7a\n \n ). Pulmonary metastatic tumors of mice in each group were monitored by the bioluminescence signal at days 5, 10 and 20, and the lungs were isolated for bioluminescence imaging at day 20. Representative images were displayed in\n \n Figure 7b-c\n \n , and treatment with ES-DSM+MW showed the strongest antitumor efficacy against pulmonary metastatic tumors. However, the ES-DM+MW group exhibited a poor antimetastatic effec because immunosuppression could not be alleviated and the antitumor immune response cannot be activated effectively in the absence of SCH.\n
\n\n Moreover, a recurrent and rechallenged tumor model was established and treated as shown in\n \n Figure 7d\n \n . After different treatments, 90% of the primary tumor was removed surgically on day 12. The residual tumor bed was further monitored and the growth of recurrent tumor was displayed in\n \n Figure 7e\n \n , which suggested that ES-DSM+MW treatment significantly inhibited the recurrence of tumor after surgery, followed by the DSM+MW group. Meanwhile, a second tumor was inoculated on the other side of mice on day 12 and the growth of the rechallenged tumor was shown in\n \n Figure 7f\n \n . Similarly, the growth of the rechallenged tumor in the ES-DSM+MW group was the most inhibited, but treatment with ES-DM+MW did not arrest the growth of rechallenged tumor. The growth of recurrent and rechallenged tumors depended on the level of immune memory after different treatments. As the remarkably increase in the TEM percentage was demonstrated in mice treated with ES-DSM+MW (\n \n Figure 6c-f\n \n ), the residual tumor bed and the second inoculated tumor could be recognized and killed immediately by TEMs. In addition, the infiltrating CD8\n \n +\n \n T cells in rechallenged tumor were remarkably increased in the ES-DSM+MW group, while Foxp3\n \n +\n \n T cells (Tregs) were greatly reduced (\n \n Figure 7g\n \n ), further emphasizing the importance of the immune response in the antitumor process.\n
\n\n \n Biocompatibility\n \n
\n\n Equally important, the biocompatibility of the various treatments was also verified by hemolysis assay and H&E staining. There was no hemolysis caused by the drug-loaded micelles (\n \n Figure S19\n \n ). In comparison to the cardiotoxicity of free drugs, the major organs of mice in the drug-loaded micelles treated groups appeared to be normal, without obvious histopathological abnormalities, degeneration, or lesions, indicating that no cellular or tissue damage occurred (\n \n Figure S20\n \n ).\n
\n\n In summary, we developed E-selectin-modified thermal-sensitive micelles to co-deliver a chemotherapy agent (DOX) and an immune checkpoint inhibitor (SCH 58261). After intravenous administration, the fabricated ES-DSM can hitchhike with leukocytes mediated by E-selectin to achieve a higher accumulation of drugs at the tumor site. Then, local microwave irradiation can be applied to induce hyperthermia and accelerate the release rate of drugs. Rapidly released DOX can not only directly kill tumor cells but can also improve the immunogenicity of tumors by inducing ICD. Released DAMPs facilitate the maturation and antigen presentation of DCs, further eliciting tumor-specific T cell immunity. On the other hand, the released SCH can prevent the engagement of ADO with A2AR on the surface of various immune cells, which can liberate the antitumor responses of DCs and CTLs while hampering the activity of Tregs. Consequently, tumor immunosuppression is relieved, and DOX-induced tumor-specific cellular immunity is enhanced. Ultimately, considerably enhanced antitumor efficiency will be achieved via the synergistic effect of chemo-immunotherapy.\n
\n\n \n Materials.\n \n Acrylonitrile (AN) was purchased from Qinghongfu Technology Co., Ltd. (Beijing, China) and purified by atmospheric distillation before use. Acrylamide (AAm), 4,4'-azobis (4-cyanovaleric acid) (ACVA), dimethyl sulfoxide (DMSO) and azelaic acid were provided by Aladdin (Shanghai, China). The amino polyethylene glycol amine (H\n \n 2\n \n N-PEG-NH\n \n 2\n \n ) (Mw=5kDa) was purchased from ToYongBio Tech.Inc. (Shanghai, China). N\u03b1,N\u03b1-Bis (carboxymethyl)-L-lysine (NTA) was obtained from Energy Chemical (Shanghai, China). Doxorubicin hydrochloride and indocyanine green (ICG) were brought from Meilun Biotechnology Co., Ltd. (Dalian, China). SCH 58261 was purchased from TCI (Tokyo, Japan). Nile red was obtained from Aladdin (Shanghai, China). Recombinant mouse E-selectin Fc chimera (ES) was from R&D Systems (Minneapolis, USA). 5\u2032-(N-ethylcarboxamido)adenosine (NECA) was bought from ApexBio Technology LLC (Houston, USA). RPMI 1640 medium and fetal bovine serum (FBS) obtained from Sigma (St. Louis, MO, USA) and Sijiqing Biological Engineering Materials Co. Ltd. (Hangzhou, China), respectively. The ELISA kits were all purchased from Meimian industrial Co., Ltd. (Jiangsu, China).\n
\n\n \n Cell culture and animals.\n \n The murine 4T1 breast cancer cells and Luc-4T1 (luciferase-expressing mouse breast carcinoma) cells were cultured in RPMI 1640 medium supplemented with 10% (v/v) FBS and penicillin/streptomycin (100 U/mL of each) and maintained in the cell incubator (37\u2103 and 5% CO\n \n 2\n \n ). The cells were regularly split using trypsin/EDTA. For the hyperthermia treated groups, the cells were placed in the cell incubator (43\u2103 and 5% CO\n \n 2\n \n , 30min) immediately after adding the test agents, followed by incubation at 37\u2103 for pre-set time period.\n
\n\n Balb/c mice (female, 6 to 8 weeks old, 18-20 g) were purchased from Slack Laboratory Animal Co., Ltd (Shanghai, China). All animal experiments were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals with the approval of the Scientific Investigation Board of Zhejiang University, Hangzhou, China.\n
\n\n \n Synthesis and characterization of NTA-PEG-p-(AAm-co-AN).\n \n Firstly, p-(AAm-co-AN) with a UCST of 43\u2103 was synthesized by solution copolymerization of AN and AAm initiated by ACVA. Briefly, 10.95g (150mmol) of AAm was weighed into a 500-mL three-necked flask and dissolved in 170mL of anhydrous DMSO. Subsequently, 2.55g (50mmol) of AN was added. Nitrogen was pumped for 1 h to remove the oxygen from the system. After that, 30mL of separately degassed anhydrous DMSO containing 0.519g (1.853mmol) of ACVA was dropped into the system through a constant pressure dropping funnel. Then placed the flask into a water bath which had been preheated to 65\u00b0C. The reaction mixture was subsequently polymerized for 8h under nitrogen protection and rapidly cooled to room temperature in an ice bath. The product was precipitated in 10-fold excess volume of methanol. The precipitate was then washed thrice with methanol and dried in a vacuum oven at 70\u00b0C for 24h.\n
\n\n Next the H\n \n 2\n \n N-PEG-NH\n \n 2\n \n was introduced to p-(AAm-co-AN) through the chemical reaction between one of the amine groups in H\n \n 2\n \n N-PEG-NH\n \n 2\n \n and the carboxyl groups of p-(AAm-co-AN). Briefly, 500mg (0.1mmol) of p-(AAm-co-AN) was weighed into a 50-mL flask and dissolved in 10mL of DMSO, to which 95mg (0.5mmol) of EDC and 57mg (0.5mmol) of NHS was added and stirred at room temperature for 4h. Subsequently, the mixture solution was added dropwise to 10mL DMSO containing 500mg (0.2mmol) of H\n \n 2\n \n N-PEG-NH\n \n 2\n \n (Mw=5kDa) at 50\u00b0C. The reaction mixture was stirred for 48h and then dialysis against deionized water with a dialysis membrane (MWCO: 8~14kDa) for 48h, followed by lyophilization and the PEG-p-(AAm-co-AN) was obtained.\n
\n\n Then the NTA was grafted onto PEG-p-(AAm-co-AN) with azelaic acid as the linker. Briefly, 19mg (100\u03bcmol) of azelaic acid was dissolved in 10mL of DMSO, to which 20mg (100\u03bcmol) of EDC and 11.5mg (100\u03bcmol) of NHS was added and stirred at room temperature for 10h to activate one of the carboxyl groups of azelaic acid. Subsequently, 500mg (33.5\u03bcmol) of PEG-p-(AAm-co-AN) was dissolved in 10mL of DMSO and added dropwise into above mixture solution, 67\u03bcmol of triethylamine was also supplemented. The reaction mixture was stirred for 17h at room temperature and then dialysis against deionized water with a dialysis membrane (MWCO: 3.5kDa) for 48h, followed by lyophilization to afford the carboxyl-containing PEG-p-(AAm-co-AN). Next, 420mg (28\u03bcmol) of carboxyl-containing PEG-p-(AAm-co-AN) was dissolved in 10mL of DMSO, 54mg (280\u03bcmol) of EDC and 32.5mg (280\u03bcmol) of NHS was added and stirred at room temperature for 4h. Then 147mg (560\u03bcmol) of NTA and 1.12mmol of triethylamine were dissolved in 10mL of DMSO/H\n \n 2\n \n O mixed solution (DMSO:H\n \n 2\n \n O=3:2), added dropwise into above solution and reacted at room temperature for 24h. After dialysis against deionized water with a dialysis membrane (MWCO: 3.5kDa) for 48h and lyophilization, the final product NTA-PEG-p-(AAm-co-AN) was afforded.\n
\n\n The\n \n 1\n \n H-NMR spectra of the polymers were obtained using an NMR spectrometer (AC-80, BrukerBioSpin, Germany). p-(AAm-co-AN), PEG, PEG-p-(AAm-co-AN) and NTA-PEG-p-(AAm-co-AN) were dissolved in DMSO-\n \n d6\n \n at concentrations of 20mg/mL. The molecular weights of p-(AAm-co-AN) and PEG-p-(AAm-co-AN) were analyzed using gel permeation chromatography (GPC) with DMSO as an eluent. PLgel MIXED-C columns (particle size: 5mm; dimensions: 7.5mm \u00d7 300mm) that had been calibrated with narrow dextran monodisperse standards were employed with a differential refractive index detector. The flow rate was 0.6mL/min. Dispersed the polymers in water at a concentration of 2mg/mL to facilitate the determination of UCST value, the optical transmittance of polymer solutions at different temperature was measured at 637nm using an ultraviolet-visible spectrophotometer (UV-2401, Shimadzu, Japan). The UCST value of p-(AAm-co-AN) was determined at the temperature when the optical transmittance became constant. The critical micelle concentration (CMC) of NTA-PEG-p-(AAm-co-AN) was determined using fluorescence spectroscopy and pyrene as a probe. Pyrene was first dissolved in acetone at a concentration of 0.0012mg/mL and added into tubes. Following evaporation of the acetone at 50\u00b0C, 5 mL of polymer solutions at different concentrations ranging from 2 to 1000\u03bcg/mL were added. After the solution was treated with water bath ultrasonication for 30 min, the emission spectra were recorded on a fluorescence spectrophotometer (F-2500, Hitachi High-Technologies Co., Japan) at room temperature. The excitation wavelength was 336 nm, and the slit widths were set at 10 nm (excitation) and 2.5 nm (emission). The pyrene emission was monitored over a wavelength range of 360-450 nm. From the pyrene emission spectra, the intensity ratio of the first peak (I\n \n 1\n \n , 374 nm) to the third peak (I\n \n 3\n \n , 384 nm) was analysed and used to calculate the CMC.\n
\n\n \n Thermal sensitivity of blank micelles.\n \n The NTA-PEG-p-(AAm-co-AN) was dispersed in water at a concentration of 0.5mg/mL, followed by 30 rounds of probe-type ultrasonic treatment (pulsed every 2s for a 3s duration, 400W). After stirring at 25\u00b0C for 0.5h, the blank micelles solution was obtained. The blank micelles solution was quartered and incubated at different temperature (25, 37, 43, 50\u00b0C) for 0.5h, dropped onto the preheated copper grids and dry at corresponding temperature. Subsequently, the morphologies of blank micelles at different temperature were observed by TEM.\n
\n\n \n Preparation and characterization of E-selectin modified DOX/SCH co-loaded micelles (ES-DSM).\n \n The DOX used in the preparation of drug-loaded micelles was obtained by the reaction between DOX\u00b7HCl and two molar equivalents of triethylamine in DMSO for 24 h. Dialysis against water to precipitate the insoluble DOX, followed by centrifuging and lyophilizing to obtain DOX powder for further use. 20mg of NTA-PEG-p-(AAm-co-AN) was dispersed in 3mL of water and treated by probe ultrasound for 30 rounds, stirring at 25\u00b0C for 0.5h to form the stable blank micelles. DOX and SCH 58261 (SCH) were dissolved together in DMSO at the final concentrations of 0.8mg/mL and 0.2mg/mL, respectively. Then 1mL of DMSO solution of DOX/SCH was added dropwise to the micelles solution with constant stirring (DOX: SCH: polymer= 4:1:100). Subsequently, 3mg of NiCl\n \n 2\n \n \u00b7H\n \n 2\n \n O was added and the mixture was stirred at 25\u00b0C for anther 2h, followed by dialyzing against water (MWCO: 3.5 kDa) for 24h and centrifuging at 4000rpm for 10min to eliminate aggregates of non-encapsulated DOX/SCH. Ultimately, the solution of DOX/SCH co-loaded micelles (DSM) was lyophilized and stored at 4\u00b0C.\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 E-selectin could be introduced onto the surface of DSM between the interaction of His-tag of E-selectin and Ni-NTA of polymer. Briefly, different concentrations of E-selectin (0, 0.1, 0.2, 0.5, 1, 2, 3\u03bcg/mL) were added to the DSM solution (at a polymer concentration of 1mg/mL) respectively, incubated at 37\u00b0C for 1h and further in 4\u00b0C overnight to afford the E-selectin modified DSM (ES-DSM). The preparation of DOX loaded micelles (DM and ES-DM) were the same as above, except the absence of SCH. The particle sizes and zeta potentials of DSM and ES-DSM were recorded by dynamic light scattering (DLS) (Zetasizer, 3000HS, 66 Malvern Instruments Ltd.). The morphology of ES-DSM was observed by transmission electron microscopy (TEM) (JEOL JEM-1230, Japan). The encapsulation efficiency (EE) and drug loading (DL) were determined by fluorospectro photometer (DOX: Ex=480nm, Em=560nm, Slit width=5nm; SCH: Ex=320nm, Em=385nm, Slit width=5nm). Briefly, the drug-loaded micelles were disrupted by DMSO and the total DOX and SCH contents were quantified. EE% and DL% were calculated by the following formulas:\n
\n\n [IMAGE_METHODS_1]\n
\n\n \n Thermal-triggered size changes of micelles.\n \n The size changes of micelles in response to temperature were monitored by DLS. The sizes of blank micelles, DSM and ES-DSM in different temperatures (5, 25, 25, 37, 43, 50\u00b0C) were measured. The samples (at a polymer concentration of 1mg/mL) were incubated at the corresponding temperature for 5 minutes before measurement. There are three repeat groups for each sample.\n
\n\n \n Thermal-sensitive in vitro drug release behavior of ES-DSM.\n \n The DOX and SCH release profiles of ES-DSM in different temperatures were tested by dialysis method. The dialysis bags (MWCO: 3.5 kDa) containing 1mL of free DOX and SCH (DS), and ES-DSM (concentrations of DOX and SCH were 90\u03bcg/mL and 15\u03bcg/mL, respectively) were immersed into falcon tubes containing 30mL PBS (pH 7.4).The tubes were put into incubator shakers (37\u2103 and 43\u2103, respectively) and horizontally shaken at 60rpm/min. At each pre-set time point, the release media were collected and replaced with fresh PBS. The DOX and SCH contents in the release media were detected by fluorospectro photometer. Each time point was performed trice.\n
\n\n \n Leukocyte-adhering ability of ES-DSM.\n \n 200\u03bcL of DSM or ES-DSM (concentrations of DOX and SCH were 300\u03bcg/mL and 50\u03bcg/mL, respectively) was injected into the mice via the tail vein, and at 2, 8 and 24h after injection, the leukocytes of treated mice were isolated. The DOX fluorescence on the obtained leukocytes was analyzed by flow cytometry (ACEA NovoCyte, USA) and confocal laser scanning microscope (CLSM) (Leica SP8, Germany).\n
\n\n \n Thermal-sensitive drug release behavior of micelles at cellular level.\n \n Firstly, Nile red was loaded into the micelles. The preparation of Nile red-loaded micelles was the same as DSM, excepted the model drug used was Nile red instead of DOX/SCH. 4T1 cells were suspended in RPMI 1640 medium and seeded in 12-well plate at a density of 1\u00d710\n \n 5\n \n cells per well and allowed to attach overnight. Subsequently, the cells were treated with free Nile red or Nile red-loaded micelles (at a final Nile red concentration of 0.1\u03bcg/mL) and the hyperthermia treated groups were placed in the cell incubator (43\u2103 and 5% CO\n \n 2\n \n , 30min) immediately, followed by incubation at 37\u2103 for 6h. After washed trice with PBS, the cells were harvested and fluorescence intensity was detected by flow cytometry. Besides, the cell fluorescence was also observed by CLSM. After incubation and washed trice with PBS, the cells were fixed and the nuclei were stained by DAPI, followed by CLSM observation.\n
\n\n Then, DOX was loaded into the micelles. The free DOX and DOX-loaded micelles were added to 4T1 cells at a final DOX concentration of 4.5\u03bcg/mL. After treated with hyperthermia and 6h incubation, the cells were washed trice with PBS and fixed. After staining by DAPI, the cells were observed by CLSM.\n
\n\n \n Cytotoxicity and apoptosis.\n \n Firstly, the cytotoxicity of blank micelles was measured by MTT assay. 4T1 cells were suspended in RPMI 1640 medium and seeded in 96-well plate at a density of 1\u00d710\n \n 4\n \n cells per well and allowed to attach overnight. Then the cells were exposed to blank micelles at a series of concentrations (0, 100, 200, 400, 600, 800, 1000\u03bcg/mL) for 48 hours. The hyperthermia treated groups were placed in the 43\u2103 cell incubator for 30min, followed by incubation at 37\u2103 until 48h. Subsequently, 20\u03bcL of 3-(4,5-Dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT) solution (5mg/mL) was added to each well for an additional 4 hours incubation at 37\u2103. After that, the medium was replaced with 100\u03bcL of DMSO to dissolve the purple formazan crystals in the bottom of the well. The plate was shaken for 30min, and the absorbance of the solution in each well was measured by microplate reader at 570nm. Cell viability was calculated in reference to negative cells without exposure to test agents. All of experiments were repeated thrice.\n
\n\n Subsequently, the cytotoxicity of free DOX/SCH (DS), DSM and ES-DSM combined with or without hyperthermia were determined by MTT assay. 4T1 cells were suspended in RPMI 1640 medium and seeded in 96-well plate at a density of 1\u00d710\n \n 4\n \n cells per well and allowed to attach overnight. Then the cells were exposed to DS, DSM or ES-DSM at different drug concentrations for 48hours (the concentration ratio of DOX and SCH is 6:1). The hyperthermia treated groups were immediately placed in the cell incubator which had pre-set to 43\u2103 for 30min after exposing to the test agents, followed by incubation at 37\u2103 until 48h. Cell viability was measured as described above.\n
\n\n Cell apoptosis induced by DS, DSM and ES-DSM combined with or without hyperthermia were investigated by flow cytometry. 4T1 cells were suspended in RPMI 1640 medium and seeded in 12-well plate at a density of 1\u00d710\n \n 5\n \n cells per well and allowed to attach overnight. Subsequently, the cells were exposed to DS, DSM or ES-DSM (concentrations of DOX and SCH were 4.5\u03bcg/mL and 0.75\u03bcg/mL, respectively) and treated with or without hyperthermia. After a 24h-incubation, cells were harvested and stained by the Annexin V-FITC/PI apoptosis detection kit (Beyotime Biotech, China) according to the manufacturer\u2019s instructions, followed by flow cytometer analysis.\n
\n\n \n Detection of the ICD biomarkers.\n \n The exposure of DAMPs (CRT, HMGB1 and ATP) of tumor cells after different treatment were detected. Briefly, 4T1 cells were treated with DS, DSM or ES-DSM (concentrations of DOX and SCH were 4.5\u03bcg/mL and 0.75\u03bcg/mL, respectively) with or without hyperthermia. The expression of CRT was observed by their immunofluorescence\n \n via\n \n CLSM at the time of 12h (Calreticulin Rabbit Monoclonal Antibody, 1:500, Beyotime, China). Semi-quantitative analysis was performed using Image J software. After incubating for 48h, the cell culture supernatant was collected and the contents of ATP and HMGB1 were detected by corresponding ELISA kits.\n
\n\n \n Co-incubation of tumor cells and bone-marrow-derived DCs.\n \n The murine bone-marrow-derived DCs (SMDCs) were isolated from 6-week old Balb/c female mice according to the established protocols.\n \n 1, 2\n \n Briefly, the bone marrow of mice was collected via flushing the femurs and tibias with PBS, and red blood cells were lysed. The remaining cells were washed twice with PBS and cultured in the complete RPMI 1640 medium containing recombinant murine GM-CSF (20ng/mL) (MedChemExpress, USA) for 6 days to acquire the immature DCs. On day 7, the immature DCs were co-incubated with 4T1 cells which had been previously treated with PBS, DS, DSM or ES-DSM (concentrations of DOX and SCH were 4.5\u03bcg/mL and 0.75\u03bcg/mL, respectively) (supplemented with or without hyperthermia) 24h ago. After a 48h-co-incubation, DCs were stained with the indicated antibodies including PE-CD80, APC-CD86 (BioLegend, USA) and PE-MHC \u2161 (ThermoFisher, USA), analyzed by flow cytometry. In addition, the cytokine levels in the supernatant of the co-incubation system including IL-12p70, IL-6 and IL-10 were detected using ELISA kits.\n
\n\n Besides, the immature DCs were co-incubated with 4T1 cells which had been previously treated with D (DOX alone), DS, DM, DSM, ES-DM or ES-DSM (concentrations of DOX and SCH were 4.5\u03bcg/mL and 0.75\u03bcg/mL, respectively) and supplemented with hyperthermia 24h ago. After a 48h-co-incubation with the presence of 1\u03bcM (a dose that mimics the concentration of adenosine found in the tumor microenvironment) of NECA (adenosine analog),\n \n 3\n \n DCs were stained with the indicated antibodies including PE-CD80, APC-CD86 (BioLegend, USA) and PE-MHC \u2161 (ThermoFisher, USA), analyzed by flow cytometry. In addition, the cytokine levels in the supernatant of the co-incubation system including IL-12p70, IL-6 and IL-10 were detected using ELISA kits.\n
\n\n \n Co-incubation of tumor cells, bone-marrow-derived DCs and\n \n \n spleen lymphocytes\n \n \n .\n \n Spleen lymphocytes were extracted from the spleens of Balb/c mice using lymphocyte density gradient centrifugation with Ficoll-paque PREMIUM. The immature DCs and lymphocytes were co-incubated with 4T1 cells which had been previously treated with PBS, DS, DSM or ES-DSM (concentrations of DOX and SCH were 4.5\u03bcg/mL and 0.75\u03bcg/mL, respectively) (supplemented with or without hyperthermia) 24h ago. After a 48h-co-incubation, lymphocytes were stained with the indicated antibodies including FITC-CD3, APC-CD8, PE-CD4 and Percific Blue-Foxp3 (BioLegend, USA), analyzed by flow cytometry. In addition, the cytokine levels in the supernatant of the co-incubation system including TNF-\u03b1, IL-2 and IFN-\u03b3 were detected using ELISA kits.\n
\n\n Besides, the immature DCs and lymphocytes were co-incubated with 4T1 cells which had been previously treated with D, DS, DM, DSM, ES-DM or ES-DSM (concentrations of DOX and SCH were 4.5\u03bcg/mL and 0.75\u03bcg/mL, respectively) and supplemented with hyperthermia 24h ago. After a 48h-co-incubation with the presence of 1\u03bcM of NECA, lymphocytes were stained with the indicated antibodies including FITC-CD3, APC-CD8, PE-CD4 and Percific Blue-Foxp3 (BioLegend, USA), analyzed by flow cytometry. In addition, the cytokine levels in the supernatant of the co-incubation system including TNF-\u03b1, IL-2 and IFN-\u03b3 were detected using ELISA kits.\n
\n\n \n Biodistribution of DSM and ES-DSM.\n \n The orthotopic tumor models were established by subcutaneous injection of 4T1 cells (5\u00d710\n \n 5\n \n ) dispersed in serum-free RPMI 1640 medium into the third breast pad of Balb/c mice. Treatment began when the tumor volume reached 500 mm\n \n 3\n \n . For the observation and imaging of the micelles biodistribution, ICG-loaded micelles were prepared the same as DSM, and the modification of E-selectin was the same as ES-DSM. 200\u03bcL of ICG-loaded micelles or ES-ICG-loaded micelles was injected into the mice via the tail vein and at 2, 6, 12, 24h after injection, the treated mice were anesthetized and the fluorescence images were acquired by Maestro\n \n in vivo\n \n imaging system. 24h after injection, the mice were sacrificed to harvest the main organs (heart, liver, spleen, lung, kidneys, and tumor). Fluorescence images were acquired, and the fluorescence intensity of these organs was measured\n \n ex\n \n \n vivo\n \n using an\n \n in vivo\n \n imaging system. The fluorescence of ICG and CD45 in tumors were analyzed by immunofluorescence.\n
\n\n \n In vivo antitumor study.\n \n 5\u00d710\n \n 5\n \n of 4T1 cells were orthotopically injected into one of the breast pads of Balb/c mice. After one week, the mice were randomly sorted into 8 groups (6 mice per group) to respectively receive one of the following treatments once every 3 days: Saline, Saline+MW, DS+MW, DSM+MW, ES-DSM, ES-DM+MW, ES-DSM+MW, ES-DSM+MW+anti-CD8, for 4 times of treatment. 3mg/kg DOX and 0.5mg/kg SCH per dose was used in the treatment and at 24h post-\n \n i.v.\n \n injection of the test agents, the mild microwave (MW) was applied locally for 30min. The microwave probe was positioned 1cm away from the fixed animal and oriented towards the tumor. The anti-CD8 antibody (BioXcell, USA) was intraperitoneal (i.p.) injected to deplete the CD8\n \n +\n \n T cells on the days of -3 and treated every 3 days until the end of monitoring. The body weight and tumor volume were monitored every 2 days and the survival time was monitored. The tumor volume was calculated using the formula: a\n \n 2\n \n \u00d7b/2, in which a and b represent the smallest and largest diameters of the corresponding tumor, respectively.\n
\n\n At the end of monitoring on day 23, the mice were sacrificed and main organs (heart, liver, spleen, lung, kidney, and tumor) were harvested and fixed in 4% paraformaldehyde, embedded in paraffin, cut into 5\u03bcm slices and stained with H&E, then examined under a light microscope. The apoptosis of tumor tissue also be studied by immunofluorescence of TUNEL staining. To demonstrate the ICD of tumor tissues, CRT and HMGB1 levels were studied by immunohistochemistry. To examine the immune response, the infiltration of CD8\n \n +\n \n T cells and Tregs (Foxp3) in tumors were analyzed by immunofluorescence, while the infiltration of active T cells (CD69) and perforin were studied by immunohistochemistry. T cells (CD3\n \n +\n \n , CD8\n \n +\n \n and CD4\n \n +\n \n ) in PBMC, spleen and tumor were isolated and analyzed using flow cytometry. The CD3\n \n +\n \n CD4\n \n +\n \n Foxp3\n \n +\n \n T cells in tumor and CD3\n \n +\n \n CD8\n \n +\n \n CD44\n \n +\n \n T cells in spleen and tumor were evaluated by flow cytometry. Levels of TNF-\u03b1, IFN-\u03b3 and IL-2 in serum, spleen and tumor were examined using the ELISA kits. DCs (CD11c\n \n +\n \n , CD80\n \n +\n \n and CD86\n \n +\n \n ) isolated from tumor and sentinel lymph node (SLN) were also analyzed by flow cytometry.\n
\n\n A lung metastatic model of breast cancer was also stablished to further investigate the treatment efficacy on metastatic cancer. Initially, the orthotopical breast tumor bearing mice was established by injecting 5\u00d710\n \n 5\n \n of 4T1 cells. 6 days later, 1\u00d710\n \n 5\n \n of Luc-4T1 cells were injected intravenously. Then, the mice were randomly sorted into 5 groups (3 mice per group) to respectively receive one of the following treatments once every 3 days: Saline+MW, DS+MW, DSM+MW, ES-DM+MW, ES-DSM+MW, for 4 times of treatment. 3mg/kg DOX and 0.5mg/kg SCH per dose was used in the treatment and the MW was applied at 24h post-\n \n i.v.\n \n injection of the test agent. The growth of pulmonary metastasis tumors was monitored by IVIS Spectrum imaging system (PerkinElmer, USA) after intraperitoneal injection of D-luciferin (15mg/mL, 200 \u03bcL). At the end of monitoring on day 20, the mice were sacrificed and the fluorescence images of lungs were acquired.\n
\n\n Tumor recurrence and re-challenge study were further invested. The orthotopical breast tumor bearing mice was established as mentioned above and received different treatments. After 4 times of treatment, 90% of the primary tumor was removed surgically on day 12, and the tumor bed was further monitored and the volume of recurrence tumor was calculated every 2 days. Simultaneously, 5\u00d710\n \n 5\n \n of 4T1 cells were inoculated into the breast pads on the other side of mice on day12. The re-challenged tumor was also monitored every 2 days. At the end of monitoring on day 30, the mice were sacrificed and re-challenged tumor was collected to analyze the infiltration of CD8\n \n +\n \n T cells and Tregs (Foxp3) by immunofluorescence.\n
\n\n \n Statistical Analysis.\n \n Statistical calculations were performed using Prism 7 software (GraphPad). Data were expressed as the mean and SD. The abbreviation ns means no significant difference. Differences were statistically evaluated by Student\u2019s\n \n t\n \n test. The differences were considered to be statistically significant for a p value of <0.05 (*p<0.05, **p<0.01, ***p<0.001). To analyze the survival time of mice, Kaplan-Meier survival curves were generated, and Log-rank Mantel\u2013Cox tests were performed. P values of < 0.05 were considered significant (*p<0.05, **p<0.01, ***p<0.001).\n
\n\n Scheme 1. Schematic depiction of the fabrication of ES-DSM and the synergistic effect of chemo-immuno-microwave hyperthermia therapy of ES-DSM delivered by leukocytes.\n
\n \n\n Supplementary Information\n
\n \n\n 3D orthogonal woven composites are receiving increasing attention with the ever-growing market of composites industries. New challenge what we face to is how the damage tolerance improve in such composites with orthogonal and layer-to-layer structure under both mechanical and extreme environment. In this paper, a novel impact damage suppression strategy is proposed by combining structural and electromagnetic properties to realize advanced functionalities. An integrated experimental platform is designed with a power system, a drop-testing machine and data acquisition devices to investigate the synergistic effects of pulse current and impact force on composites. Experimental results exhibit that pulse current can effectively reduce delamination damage and residual deformation. A multi-field coupled damage model is developed to analyze the evolutions of temperature, current and damage. The microcrack formation and extrusion deformation in yarns causes the local current redistribution in carbon fibers, and its interaction with the self-field produces an obvious anti-impact effect. The obtained results reveal the mechanism of damage suppression and provide a potential orientation for improving damage tolerance of these composites.\n
\n\n \n 3D orthogonal woven composites\n \n
\n\n \n Synergistic effects\n \n
\n\n \n Impact force\n \n
\n\n \n Pulse current\n \n
\n\n \n Dynamic damage analysis\n \n
\n\n \n Damage suppression\n \n
\n\n Carbon fiber reinforced composites are extensively used in aerospace, aeronautics and national defense fields due to its high specific stiffness, strength and designability [\n \n 1\n \n \u2013\n \n 4\n \n ]. However, they are sensitive to low-velocity impact events from dropped tools, debris or birds during its manufacture and service, which can cause delamination failure and result in a significant reduction of load-bearing capacity [\n \n 5\n \n ,\n \n 6\n \n ]. Numerous studies have therefore been carried out to improve low-velocity impact response of composites from the structural design such as ply-stacking sequence [\n \n 7\n \n ,\n \n 8\n \n ], ply thickness design [\n \n 9\n \n ,\n \n 10\n \n ], 3D structure design [\n \n 11\n \n ,\n \n 12\n \n ]. Compared to 2D laminated composites, 3D orthogonal woven composites (3DOWCs) can improve the delamination behavior due to the presence of through-thickness Z-binder yarns [\n \n 13\n \n \u2013\n \n 16\n \n ]. In addition, fiber hybridization is also an effective method of improving impact resistance, as reported in Refs. [\n \n 17\n \n ,\n \n 18\n \n ]. Further, by taking advantage of the conductive carbon fibers due to their internal near-graphite structures, a certain level of electromagnetic field can increase the strength and interlaminar resistance of carbon fiber/epoxy laminates exposed therein [\n \n 19\n \n \u2013\n \n 22\n \n ]. As shown in Fig.\n \n 1\n \n a, the three electrons in the carbon atom form three stable\n \n \u03b4\n \n bonds with other three electrons in the surrounding carbon atoms, while the remaining unbonded electron in the carbon atom forms a large free-moving\n \n \u03c0\n \n -conjugated electron cloud between the planes [\n \n 23\n \n \u2013\n \n 26\n \n ]. The deformation and separation of the hexagonal carbon rings require high energy, providing the strength of the carbon fiber at macro level, while the free electrons in electron cloud make it a good electrical conductor. Hence, it is a new perspective to improve the performance of composite materials to make full use of the multifunctional property of materials, such as a key physical effect in conductive fibers whereby matter becomes compressed under the effect of an electric field.\n
\n\n
\n\n Although electromagnetic environments can improve the mechanical properties of conductive laminates, its effect and mechanism on more complex conductive structures such as the 3DOWCs have not been reported. As shown in Fig.\n \n 1\n \n b,\n \n 3\n \n DOWCs are comprised of conductive carbon fiber/epoxy prepreg with complex structure and insulating epoxy resin filled between prepreg. The conductive warp and weft yarns are interwoven vertically in the plane direction. The conductive Z-binder yarns undulate along warp direction and are applied to bind weft and warp yarns in thickness direction. At this point, the yarns are equivalent to bare wires when pulse current is applied. The currents pass through the interwoven strands of the yarn, forming a complex conductive network as shown in Fig.\n \n 1\n \n c. The impact force is applied to 3DOWCs on the basis of electrification, which is a typical real-time coupling problem among the electromagnetic, thermal and mechanical fields, as shown in Fig.\n \n 1\n \n d. In this case, the Lorentz force due to the current-field interaction between current and its self-field as well as the corresponding electrothermal stress affect the mechanical response. In turn, deformation and damage induced by impact force change the interlaced structure of the 3DOWCs plate, resulting in the redistributions of electromagnetic and thermal fields.\n
\n\n In this paper, the effects of impact force and pulse current on the 3DOWCs plate and the interaction mechanism of multi-physics field are studied. An integrated experimental platform, which is composed of current supply, drop-testing machine and data acquisition devices, is designed to realize the synergistic effects of pulse current and impact force on the 3DOWCs plate via wireless telecommunication technology. The dynamic numerical model of synergistic responses for 3D yarn-level orthogonal woven composites with pulse current is developed based on the real-time sequential coupling of electromagnetic, thermal, and mechanical fields. We find that the pulse current introduced into the 3DOWCs plate can effectively suppress impact damage. Our experimental observations and simulation results further reveal the damage suppression mechanism for the 3DOWCs under pulse current and impact load.\n
\n\n The experiments are carried out on a self-developed pulse current-impact and data collection integrated experimental platform at ambient temperature. As shown in Fig.\n \n 2\n \n a, the platform includes a DIT152 drop-testing machine, a function signal generator (Puyuan DG4062), a current supply (Agilent 6692A), a Tektronix MDO3054 oscilloscope, a current probe (TCP0150), a voltage probe (DP1650A), a set of data acquisition devices and a self-developed computer-controlled trigger system. This system allows the drop falling time, current action time and data collection time to be controlled arbitrarily. It is noted that Agilent 6692A 6600-watt power supply provides a steady current, while a time-varying pulse current is required in the experiment to reduce the adverse effect of the current-induced Joule heating on the plate. To this end, a function signal generator (Puyuan DG4062) is connected to the current supply to generate arbitrary time-varying voltage waveforms. Once the two devices are connected, current supply trigger and the generated waveform are controlled by the function generator. The input voltage\n \n \n \\({U_{in}}\\)\n \n \n on the function generator and the output current\n \n \n \\({I_{out}}\\)\n \n \n on the power supply can be calculated via the relationship\n \n \n \\({U_{in}}/5={I_{out}}/110\\)\n \n \n .\n
\n\n In addition, a fixture must be electrically insulated and meet the impact requirements in accordance with ASTM Standards D5379 to characterize composite material correctly during measurement. As shown in the upper left corner of Fig.\n \n 2\n \n a, a fixture is customized and constructed according to the standard. Non-conductive wood is utilized to construct the main body. Copper bars are used to construct the positive and negative electrodes, ensuring that the current is conducted along the composite in the experiment. The 3DOWCs plates with dimensions of 148 mm \u00d7 148 mm \u00d7 4.5 mm are made of F-46 epoxy resin and T700 carbon fiber with a fiber volume fraction of 82.43%, as shown in Fig.\n \n 2\n \n b. Composite prepreg contains 6 layers of warp tows, 7 layers of weft tows and the binding yarns, each of which is composed of 12K carbon fibers. A 3DOWCs plate is a part of the circuit and the contact resistances between the plate and copper bus bars are negligibly small by evenly coating a thin layer of conductive silver paint (SPI#05002-AB) at their interfaces.\n
\n\n
\n\n Before starting the collaborative experiment of impact force and pulse current on the 3DOWCs plate, all the devices need to be connected and adjusted to the required parameters, as shown in Fig.\n \n 2\n \n c. The impact module is relatively integrated in the drop-testing machine and thus its connection is ignored. In order to realize signal conversion, the function generator\u2019s positive output wire and negative output wire are connected to the VP connector and IP connector located on the back of the power supply, respectively. The 3DOWCs plate is centered in the customized fixture to ensure that the impact point is exactly centered on its upper surface. Two lead wires of the copper electrodes from the power supply are in contact with the two sides of the 3DOWCs plate. As can be seen from Fig.\n \n 2\n \n c, the devices are connected to red lines to form a series circuit. The composite resistance is obtained by Ohm\u2019s law. The voltage probe is fixed on two copper electrodes to measure the voltage across the 3DOWCs plate. The current probe is held on the cable, which connects the power supply and copper electrode to measure the current passing through the 3DOWCs plate. The voltage and current data are recorded by the oscilloscope. Taking into account the anisotropy of the 3DOWCs plate, two heat-flow sensors are attached to the plate along the warp and weft directions to detect the heat flow and temperature responses. As shown in Fig.\n \n 2\n \n a, eight thermal resistors are also pasted around impact point of the plate to detect temperature changes around the impact point. The heat flow data are collected by a heat flow meter and the thermal resistance data are collected by a 20-channel data acquisition instrument. Once all the devices are connected, the required pulse waveform and amplitude, as shown in Fig.\n \n 2\n \n d, are set through the mechanical buttons on the function generator. Finally, the trigger time of impact module, current module and data acquisition module in the self-developed control program are modified according to the experimental requirements. Note that the duration of the impact force is more than three orders of magnitude shorter than that of the pulse current. Thus, in order to make sure that the impactor just contacts with the specimen when the pulse current reaches its amplitude value, the impact module is controlled to delay triggering for a time of\n \n \n \\({t_ * } - {t_f}\\)\n \n \n .\n \n \n \\({t_ * }\\)\n \n \n is the time when the current reaches its amplitude.\n \n \n \\({t_f}=\\sqrt {{{2h} \\mathord{\\left/ {\\vphantom {{2h} g}} \\right. \\kern-0pt} g}}\\)\n \n \n is the free-fall time of the impactor, where\n \n \n denotes the height of the gravity center of the impactor from the upper surface of the plate, and\n \n \n denotes the gravitational acceleration. In this step, the coordination of experiment and acquisition devices using wireless communication technology is required, which is key to the success of the integrated experimental platform. After confirmation, the run button in the program is triggered to make each device start working.\n
\n\n The time histories of force, deformation and impact energy corresponding to impact energy of 50 J and different pulse current amplitudes (i.e.,\n \n \n \\({I_{am}}\\)\n \n \n =\u20090 A, 30 A, 70 A and 110 A) are plotted in Fig.\n \n 3\n \n a, Fig.\n \n 3\n \n b and Fig.\n \n 3\n \n c, respectively. In the initial loading stage (Stage I), the impactor begins to contact with the plate and then leads to local deformation near the impact point. Meanwhile, the kinetic energy of the impactor is gradually converted into the elastic energy of the plate. As small elastic deformation hardly affects the current distribution, the force, deformation or energy curves corresponding to different\n \n \n \\({I_{am}}\\)\n \n \n almost overlap in this stage. Then, the impact force increases rapidly and some oscillations occur due to multiple factors, such as large difference in mass or stiffness between the impactor and the plate. As inelastic energy accumulates, the plate gradually enters the maximum loading stage (Stage II). Large deformation is observed near the impact point, and the plate is strongly squeezed locally, resulting in densification of the current channels, which in turn resists the impact through the Lorentz force generated by the current interacting with its self-field. In this stage, the force curves are separated, and so do the deformation and energy curves. When the plate reaches the maximum deformation, the kinetic energy of the impactor is utterly converted into the elastic and inelastic energies of the plate. At the same time, the plate is subjected to the maximum impact force. It is found that as\n \n \n \\({I_{am}}\\)\n \n \n increases, the maximum impact force increases gradually. Besides, the maximum energy in the plate appears at\n \n \n \\(t=\\)\n \n \n 3.312 ms, 3.224 ms, 3.196 ms and 3.188 ms for the cases of\n \n \n \\({I_{am}}\\)\n \n \n = 0 A, 30 A, 70 A and 110 A, respectively, and the corresponding maximum deformations are also observed at the same instants. That is, the plate will arrive early at the most dangerous stage of damage with increasing\n \n \n \\({I_{am}}\\)\n \n \n . In the rebounding stage (Stage III), the impactor is rebounded by the plate until the two are completely separated. As the impact force decreases, the plate gradually recovers its deformation due to the release of elastic energy, while the residual deformation is remained due to the dissipation of inelastic energy. Moreover, the inelastic energy or residual deformation decreases with the increase of\n \n \n \\({I_{am}}\\)\n \n \n , which can be observed more intuitively from Fig.\n \n 3\n \n d. The inelastic energy and residual deformation of\n \n \n \\({I_{am}}\\)\n \n \n = 110 A are respectively reduced by 35.81% and 47.64% compared with those of\n \n \n \\({I_{am}}\\)\n \n \n = 0 A, implying that the Lorentz force generated by the pulse current in the warp direction can effectively resist the impact force on the plate.\n
\n\n
\n\n An optical microscope and ultrasonic C-scan device are adopted to characterize the effect of pulse currents on the impact damage suppression of the plates. The damage morphology for the cross section of the plate under impact energy of 50 J is shown in Fig.\n \n 3\n \n e. The resin matrix around the impact point is debonding from the top and bottom surfaces of the plate. This is because the resin matrix is in an unconstrained state and its performance is inferior to that of the yarns. Delamination among various types of yarns is found on the bottom surface of the plate due to debonding or even shedding of the resin matrix. Additionally, resin cracks among yarns are found under the impactor, which are blocked by the interwoven yarns to effectively suppress the propagation of cracks. The delamination depth and area after impact are obtained to quantify the damage suppression efficiency of pulse current on the plate, which is based on the conventional B-scan and ply-by-ply C-scan analysis in the ultrasonic C-scan device. The gross damage images with a resolution of 1500 \u00d7 1500 pixels are shown in Figs.\n \n 3\n \n f-i for the plates under impact energy of 50 J with\n \n \n \\({I_{am}}\\)\n \n \n = 0 A, 30 A, 70 A and 110 A. The depth of damage is illustrated by color, ranging from red (shallow) to green (deep). At the cross sections of A-A and B-B, the green circular area with damage depth greater than 1 mm accounts for 0.057% of the full-scale 3DOWCs plate for the case of\n \n \n \\({I_{am}}\\)\n \n \n = 0 A. As\n \n \n \\({I_{am}}\\)\n \n \n increases, the green area gradually shrinks, indicating a significant decrease of the impact damage. Since few experimental tools are available to directly observe the multi-field coupling evolution inside composite structures, the suppression mechanism of pulse current on impact damage will be further discussed in detail in the following sections by means of numerical models.\n
\n\n As schematically shown in Fig.\n \n 4\n \n a, the synergistic responses of 3DOWCs plate are studied sequentially by a three-scale modeling strategy including microscale, mesoscale and macroscale models. The mesoscale and macroscale structures are modeled based on the performance parameters obtained from a microscale representative volume element (RVE). According to the theory of composite material in Refs. [\n \n 27\n \n \u2013\n \n 29\n \n ], fiber yarns can be regarded as unidirectional fiber-reinforced composites composed of transversely isotropic carbon fibers and isotropic resin matrix. The mixture rule [\n \n 30\n \n ] and microscale classical hexagon model are adopted to obtain accurate mechanical and thermal parameters of yarns. Material parameters of carbon fiber provided by the manufacturer are listed in Table\n \n 1\n \n . The experimental data of resin matrix are shown in Figs.\n \n 4\n \n b-d. The comparative mechanical results of yarns are listed in Table\n \n 2\n \n using the mixture rule and finite element method. The current conduction in yarns exhibits strong anisotropy and is temperature-independent below 343 \u2103, which have been validated by experiments [\n \n 31\n \n \u2013\n \n 34\n \n ]. It is worth noting that fibers in yarns are in contact with each other and form a contact network in transverse direction due to slight inclinations or ripples. In this sense, the hexagon model applied to calculate mechanical and thermal parameters is unsuitable for calculating transverse conductivity. Therefore, the mixture rule and Kirchhoff\u2019s current law in Refs. [\n \n 35\n \n ] and [\n \n 36\n \n ] are used to calculate the longitudinal and transverse conductivities, respectively. The electrical parameters of carbon fiber are given in Table\n \n 3\n \n and the obtained conductive parameters of yarn are listed in Table\n \n 4\n \n with a fiber volume fraction of 82.43%. Carbon fiber and epoxy resin are non-magnetic materials so that the permeability of yarns is 1\n \n \u00b5\n \n \n \n 0\n \n \n . In the mesoscale, a single RVE is established based on the geometric structure of the 3DOWCs plate and the microscopic photographs of cross sections as shown in Fig.\n \n 4\n \n e and Fig.\n \n 4\n \n f. The mesoscale single RVE contains regularly arranged yarns and resin matrix filled among the yarns. The Hashin criterion and elastoplastic criterion are developed to investigate the damages of yarns and resin matrix, respectively. The cross sections of all yarns are assumed to be rectangular to guarantee the numerical convergence. Both the weft and warp yarns are straight and perpendicular to each other with interlacement. Furthermore, Z-binder yarns undulate along the warp direction with a useful feature to bind weft and warp yarns in the thickness direction. Then, a macroscale model with the size of 148 mm \u00d7 74 mm \u00d7 4.5 mm is established by duplicating the mesoscale single RVE.\n
\n\n
\n| \n \n T700-12K Carbon Fibers\n \n | \n \n \n Value\n \n | \n
|---|---|
| \n \n Longitudinal elastic modulus\n \n E\n \n \n \n f,1\n \n \n (GPa)\n \n | \n \n \n 230\n \n | \n
| \n \n Transverse elastic modulus\n \n E\n \n \n \n f,2\n \n \n (GPa)\n \n | \n \n \n 14\n \n | \n
| \n \n Poisson\u2019s ratio\n \n \u03c5\n \n \n \n f,12\n \n \n \n | \n \n \n 0.25\n \n | \n
| \n \n Poisson\u2019s ratio\n \n \u03c5\n \n \n \n f,23\n \n \n \n | \n \n \n 0.3\n \n | \n
| \n \n Shear modulus\n \n G\n \n \n \n f,12\n \n \n /\n \n G\n \n \n \n f,1\n \n 3\n \n (GPa)\n \n | \n \n \n 9\n \n | \n
| \n \n Shear modulus\n \n G\n \n \n \n f,23\n \n \n (GPa)\n \n | \n \n \n 5\n \n | \n
| \n \n Ultimate tensile strength\n \n X\n \n \n \n f,t\n \n \n (MPa)\n \n | \n \n \n 4900\n \n | \n
| \n \n Ultimate compressive strength\n \n X\n \n \n \n f,c\n \n \n (MPa)\n \n | \n \n \n 2470\n \n | \n
| \n \n Density\n \n \u03c1\n \n \n \n f\n \n \n (kg/m\n \n 3\n \n )\n \n | \n \n \n 2400\n \n | \n
| \n \n Longitudinal coefficient of thermal expansion\n \n \u03b1\n \n \n \n f\n \n 1\n \n (10\n \n \u2212\u20096\n \n /\u2103) (kg/m\n \n 3\n \n )\n \n | \n \n \n -0.5\n \n | \n
| \n \n Transverse coefficient of thermal expansion\n \n \u03b1\n \n \n \n f\n \n 2\n \n (10\n \n \u2212\u20096\n \n /\u2103)\n \n | \n \n \n 10.2\n \n | \n
| \n \n Longitudinal thermal conductivity\n \n \u03bb\n \n \n \n f1\n \n \n (W/m.K)\n \n | \n \n \n 11\n \n | \n
| \n \n Transverse thermal conductivity\n \n \u03bb\n \n \n \n f2\n \n \n (W/m.K)\n \n | \n \n \n 1.3\n \n | \n
| \n \n Specific heat capacity\n \n C\n \n \n \n p\n \n \n (J/kg.K)\n \n | \n \n \n 750\n \n | \n
\n
\n\n
\n| \n | \n\n \n \n E\n \n \n \n 11\n \n \n \n\n (GPa)\n \n | \n \n \n \n E\n \n \n \n 22\n \n \n \n\n (GPa)\n \n | \n \n \n \n \u03c5\n \n \n \n 12\n \n \n \n | \n \n \n \n \u03c5\n \n \n \n 23\n \n \n \n | \n \n \n \n G\n \n \n \n 12\n \n \n \n\n (GPa)\n \n | \n \n \n \n G\n \n \n \n 23\n \n \n \n\n (GPa)\n \n | \n \n \n \n \u03b1\n \n \n \n y1\n \n \n \n\n (10\n \n \u2212\u20096\n \n /\u2103)\n \n | \n \n \n \n \u03b1\n \n \n \n y2\n \n \n \n\n (10\n \n \u2212\u20096\n \n /\u2103)\n \n | \n
|---|---|---|---|---|---|---|---|---|
| \n \n Rule of mixture\n \n | \n \n \n 190.137\n \n | \n \n \n 10.597\n \n | \n \n \n 0.268\n \n | \n \n \n 0.309\n \n | \n \n \n 5.547\n \n | \n \n \n 3.820\n \n | \n \n \n -\n \n | \n \n \n -\n \n | \n
| \n \n FEM\n \n | \n \n \n 192.029\n \n | \n \n \n 11.644\n \n | \n \n \n 0.235\n \n | \n \n \n 0.359\n \n | \n \n \n 5.4312\n \n | \n \n \n 3.7012\n \n | \n \n \n 0.074\n \n | \n \n \n 33.452\n \n | \n
\n
\n\n
\n| \n \n Fiber diameter\n \n d\n \n \n \n f\n \n \n (\u00b5m)\n \n | \n \n \n Fiber conductivity\n \n\n \n \u03c3\n \n \n \n f\n \n \n (S/m)\n \n | \n \n \n Fiber modulus\n \n\n \n E\n \n (GPa)\n \n | \n \n \n Processing pressure\n \n\n \n P\n \n (MPa)\n \n | \n
|---|---|---|---|
| \n \n 7\n \n | \n \n \n 6.25\u00d710\n \n 4\n \n \n | \n \n \n 230\n \n | \n \n \n 0.8\n \n | \n
\n
\n\n
\n| \n \n Electrical conductivity\n \n | \n \n | \n\n \n Equivalent permeability\n \n | \n ||
|---|---|---|---|---|
| \n \n \n \u03c3\n \n \n \n 11\n \n \n (S/m)\n \n | \n \n \n \n \u03c3\n \n \n \n 22\n \n \n ,\n \n \u03c3\n \n \n \n 33\n \n \n (S/m)\n \n | \n \n | \n\n \n \n \u00b5\n \n \n \n 11\n \n \n \n | \n \n \n \n \u00b5\n \n \n \n 22\n \n \n ,\n \n \u00b5\n \n \n \n 33\n \n \n \n | \n
| \n \n 2.75\u00d710\n \n 4\n \n \n | \n \n \n 25.595\n \n | \n \n | \n\n \n 1\n \n \u00b5\n \n \n \n 0\n \n \n \n | \n \n \n 1\n \n \u00b5\n \n \n \n 0\n \n \n \n | \n
\n
\n\n
\n\n In order to form a conductive path across the model, pulse current is imposed to one side of the model and the potential on the opposite side is set to 0 V. Heat radiation with a coefficient of 0.9 is applied to the top and bottom surfaces exposed to the air. Meanwhile, the parallel magnetic flux boundary is applied to the top and bottom surfaces and the opposite side of symmetry plane for the 3DOWCs plate. The vertical magnetic flux boundary is applied to the symmetry plane of the 3DOWCs plate. The ambient temperature is set to 25\u00b0C. Additionally, the simply supported boundary and fixed boundary of the plate are consistent with those in the experiment. The hemispherical impactor with a diameter of 16 mm and a mass of 2.75 kg is modeled as a rigid body. Moreover, all degrees of freedom of the impactor except axial direction are fixed to avoid rotating or shifting. The initial velocity is defined to the impactor reference point according to the initial impact energy requirement. Furthermore, the cyclic sequential coupling method is adopted to solve the interaction of electromagnetic, thermal and mechanical fields on the 3DOWCs plate. Electromagnetic and thermal processes are conducted using implicit solving methods, while the low-velocity impact process is carried out using explicit solving strategies. In order to verify the multi-physics field coupling model, the numerical force curves for impact energy of 50 J with\n \n \n \\({I_{am}}\\)\n \n \n = 0 A, 30 A, 70 A and 110 A are compared with the experimental force curves. As shown in Figs.\n \n 4\n \n g-j, these two sets of curves exhibit nearly the same trend. Additionally, the maximum impact forces\n \n \n \\({F_{\\hbox{max} }}\\)\n \n \n obtained by numerical and experimental methods are listed in Table\n \n 5\n \n , in which the two types of data are in good agreement with a mean error of 16.745%.\n
\n\n
\n| \n \n Model\n \n | \n \n \n Experimental results (N)\n \n | \n \n \n Finite element analysis results (N)\n \n | \n \n \n Error (%)\n \n | \n
|---|---|---|---|
| \n \n 50 J-0 A\n \n | \n \n \n 12514\n \n | \n \n \n 9898\n \n | \n \n \n 20.90\n \n | \n
| \n \n 50 J-30 A\n \n | \n \n \n 10130\n \n | \n \n \n 10713\n \n | \n \n \n 5.76\n \n | \n
| \n \n 50 J-50 A\n \n | \n \n \n 12589\n \n | \n \n \n 9880\n \n | \n \n \n 21.52\n \n | \n
| \n \n 50 J-110 A\n \n | \n \n \n 12514\n \n | \n \n \n 10161\n \n | \n \n \n 18.80\n \n | \n
\n
\n\n To further investigate the mechanism of damage suppression, the evolutions of multi-physical fields and impact damage for the plate under impact energy of 50 J and\n \n \n \\({I_{am}}\\)\n \n \n = 30 A are presented in Fig.\n \n 5\n \n and Fig.\n \n 6\n \n , respectively. Before impact, the yarns in the plate are spatially interlaced horizontally and vertically to form an overlapping 3D network with interface resistances. When the two ends of the plate are connected to a circuit, the electric potential decreases uniformly along the warp direction, and the current density through the YZ cross section is evenly distributed at the macro level. The interaction of the current with its self-field causes the plate to be subjected to a compressive electromagnet force. At the same time, the resulting effective thermal strain induced by the temperature variation is evenly distributed on the plate (see Fig.\n \n 5\n \n ). In the initial loading stage, the kinetic energy of the impactor begins to convert into the elastic energy of the plate, and the deformation of the plate increases with the impact force. At the moment, the potential difference around the impact point increases dramatically due to the impact deformation (see Fig.\n \n 5\n \n ). The yarns around the impact point are squeezed and the current channels become denser, resulting in a higher current density than elsewhere. Meanwhile, the enhanced magnetic flux density in the shape of Y-axis symmetric ginkgo biloba leaf around the impact point can be observed in the XY plane, resulting in stronger Lorentz force in this region. In addition, the maximum heat flux density is presented at the impact point compared to other regions, indicating that the compressive deformation caused by impact force leads to an increase in thermal conductivity. The effective thermal strain of up to 0.113e-6 is distributed on the plate in the shape of a flashlight beam along the negative Y-axis originated from the impact point. However, it still accounts for a small proportion in the total effective strain of 0.468e-6. This feature is quite different from that for the direct or alternating current case, for which the electrothermal effect caused by the transport current can lead to severe damage inside laminates, as shown in Ref. [\n \n 21\n \n ].\n
\n\n In the maximum loading stage, impact force and deformation continue to increase until the energy of the impactor is fully transferred to the elastic and inelastic energy of the plate (see Fig.\n \n 6\n \n ). Such behavior leads to resin matrix cracking, debonding and yarn breakage around the impact point because the stress induced by impact deformation of the plate reaches the damage tolerances of the resin and yarns. At this point, large deformation and damage lead to changes in the current channels. However, the overall current is still concentrated in the deformed warps, although the interface resistances become smaller due to the extrusion among the components. The Lorentz force generated by the redistributed current and its self-field continuously resists the impact force. The physical mechanism can be revealed by the Lorentz force distribution. As shown in the side view of Fig.\n \n 7\n \n b, when the impact point of the plate is moved downward by impact force, the local upper part along the thickness gradually moves below the mid-plane, and the direction of the electromagnet force changes from downward to upward, thus producing a macroscopic resistance just opposite to the impact force. As shown in the top view of Fig.\n \n 7\n \n c, the upper and lower parts of the plate along the width are subjected to vertical downward and upward electromagnetic forces, respectively. Under the impact force, the upper part of the warp yarns around the impact point gradually bend upwards, while the lower part of the warp yarns bend downwards, which results in the increase of the density of local warp yarns. The warp yarns are then subjected to a large electromagnetic force directed towards the impactor, thereby resisting the impact force. It should be noted that when the warp yarns are partially cracked or even broken due to excessive impact (see Fig.\n \n 7\n \n a), the current will form a new circuit in the intricate yarns to continuously resist the impact force. At this time, the maximum effective thermal strain value increases, but still contributes less than the compressive electromagnetic strain generated by Lorentz force.\n
\n\n In the rebounding stage, the plate cannot recover to its original state because the damage and plastic deformation has dissipated part of the kinetic energy of the impactor. At this time, the extrusion effect among yarns is weakened, reducing the deformation of current-carrying yarns and further the magnitude of pulse current density. However, the direction of the current density in this stage remains consistent with that in the maximum loading stage. Meanwhile, magnetic flux intensity region of ginkgo biloba leaf distribution observed in the XY plane decreases by about one third compared with that in the maximum loading stage. As a result, the magnitude of the total Lorentz force decreases while its direction coincides with the rebound direction of the impactor. In this sense, the total Lorentz force at this stage can accelerate the rebound of the plate. Moreover, the contribution of the Lorentz force is much larger than that of the effective thermal stress, further indicating that the pulse current can suppress the impact damage of composite plates.\n
\n\n
\n\n
\n\n
\n\n This work provides a novel strategy to suppress impact damage in 3D orthogonal woven composites by combining the structural property of woven fabrics and electromagnetic property of carbon fibers. An integrated experimental platform is designed to study the synergistic effects of pulse current and impact force by means of wireless telecommunication technology. The experimental results show that the introduction of pulse current to 3D orthogonal woven composites can significantly reduce the depth and area of impact damage. To be specific, an increase in pulse current amplitude from 0 A to 110 A results in a decrease in inelastic energy from 27.59 J to 17.71 J, as well as a reduction in residual deformation from 2.75 mm to 1.44 mm. The mechanism of damage suppression is revealed by using a multi-field coupled model on the basis of the theories of damage mechanics extended to account for classical electromagnetism and heat transfer. Under impact loading, the local extrusion among yarns changes the direction of pulse current flowing in carbon fibers, and its interaction with the self-field just provides a compressive electromagnetic force that resists the impact force. The current will form a new circuit in the intricate yarns to continuously resist the impact even if the impact damage causes a partial circuit break. In addition, the pulse current can keep the energy loss of composites at a low level, which effectively avoids damage caused by thermal expansion. Based on the results, we provide some general guidelines on how to take full advantage of the electromagnetic property of carbon fibers within composites to achieve active impact damage suppression.\n
\n\n Commercial Cu-exchanged small-pore SSZ-13 (Cu-SSZ-13) zeolite catalysts are highly active for the selective catalytic reduction (SCR) of NO\n \n \n x\n \n \n with NH\n \n 3\n \n , but distinct from other catalyst systems, their activity is unexpectedly inhibited in the presence of NO\n \n 2\n \n . Here, we combined kinetic experiments,\n \n in-situ/operando\n \n X-ray absorption spectroscopy, and density functional theory (DFT) calculations to obtain direct evidence that under reaction conditions, strong oxidation by NO\n \n 2\n \n forces Cu ions to exist mainly as fixed framework Cu\n \n 2+\n \n species (fw-Cu\n \n 2+\n \n ), which impede the formation of dynamic binuclear Cu\n \n +\n \n species that serve as the main active sites for the standard SCR (SSCR) reaction. As a result, the SSCR reaction is significantly inhibited by NO\n \n 2\n \n in the zeolite system, and the NO\n \n 2\n \n -involved SCR reaction occurs with an energy barrier higher than that of the SSCR reaction on dynamic binuclear sites. Moreover, the NO\n \n 2\n \n -involved SCR reaction tends to occur at the Br\u00f8nsted acid sites (BAS) rather than the fw-Cu\n \n 2+\n \n sites. This work clearly explains the strikingly distinctive selective catalytic behavior in the zeolite system.\n
\n\n \n \n Catalysis\n \n \n
\n\n \n Cu-SSZ-13\n \n
\n\n \n Fast SCR reaction\n \n
\n\n \n Kinetics\n \n
\n\n \n In-situ/Operando XAFS\n \n
\n\n \n DFT calculation\n \n
\n\n Increasingly stringent mobile source emission regulations have been pursued around the world to tackle environmental pollution. Nitrogen oxides (NO\n \n \n x\n \n \n ) are inevitable gaseous pollutants emitted from internal combustion engines. Selective catalytic reduction of NO\n \n \n x\n \n \n with NH\n \n 3\n \n (NH\n \n 3\n \n -SCR) is the most widely adopted technology to eliminate NO\n \n \n x\n \n \n .\n \n 1,2\n \n The successful commercialization of Cu-SSZ-13 as an NH\n \n 3\n \n -SCR catalyst is a significant achievement for diesel engine exhaust post treatment.\n \n 3\n \n In the past decade, numerous studies have endeavored to uncover the standard SCR (SSCR) reaction mechanism\n \n 4-7\n \n , hydrothermal deactivation mechanism\n \n 8-11\n \n , and SO\n \n 2\n \n poisoning deactivation mechanism\n \n 12-14\n \n , and to develop economic and sustainable synthesis methods for Cu-SSZ-13\n \n 15-18\n \n , bringing about continuous optimization of Cu-SSZ-13 for commercial SCR catalysts.\n
\n\n In actual application, a diesel oxidation catalyst (DOC) is utilized to oxidize carbon monoxide (CO) and hydrocarbons (HCs), accompanied by partial oxidation of NO to NO\n \n 2\n \n . The formed NO\n \n 2\n \n can participate in the NH\n \n 3\n \n -SCR process through the so-called \u201cfast SCR\u201d reaction (FSCR, reaction 1, consisting of reactions 2 and 3). It is generally believed that the deNO\n \n \n x\n \n \n efficiency of the FSCR reaction should be higher than that of SSCR (reaction 4) due to bypassing NO oxidation, which is usually the rate-limiting step in the SSCR reaction\n \n 19,20\n \n .\n
\n| \n \n \n | \n \n \n NO + NO\n \n 2\n \n + 2NH\n \n 3\n \n \u2192 2N\n \n 2\n \n + 3H\n \n 2\n \n O\n \n | \n \n \n \uff081\uff09\n \n | \n
| \n \n \n | \n \n \n 2NO\n \n 2\n \n + 2NH\n \n 3\n \n \u2192 NH\n \n 4\n \n NO\n \n 3\n \n + N\n \n 2\n \n + H\n \n 2\n \n O\n \n | \n \n \n \uff082\uff09\n \n | \n
| \n \n \n | \n \n \n NO + NH\n \n 4\n \n NO\n \n 3\n \n \u2192 N\n \n 2\n \n + NO\n \n 2\n \n + 2H\n \n 2\n \n O\n \n | \n \n \n \uff083\uff09\n \n | \n
| \n \n \n | \n \n \n 4NO + 4NH\n \n 3\n \n + O\n \n 2\n \n \u2192 4N\n \n 2\n \n +6H\n \n 2\n \n O\n \n | \n \n \n \uff084\uff09\n \n | \n
\n However,\u00a0there have been few studies reporting that NO\n \n 2\n \n measurably promotes the NH\n \n 3\n \n -SCR efficiency over Cu-SSZ-13 catalytic systems. On the contrary, inhibition of NO conversion by NO\n \n 2\n \n was found over Al-rich Cu-SSZ-13 catalysts due to NH\n \n 4\n \n NO\n \n 3\n \n formation, which is the so-called \u201cabnormal fast NH\n \n 3\n \n -SCR reaction\u201d.\n \n 21\n \n In our recent study, we found that the inhibiting effect of NO\n \n 2\n \n was closely related to Br\u00f8nsted acid sites (BASs) and can be alleviated by hydrothermal aging due to the decrease in the number of BASs in Cu-SSZ-13.\n \n 22\n \n Therefore, we speculated that NO\n \n 2\n \n reduction probably occurs at BASs. Also, we previously observed the reaction between NO and NH\n \n 4\n \n NO\n \n 3\n \n occurring at BASs over the H-SSZ-13 catalyst.\n \n 23\n \n Furthermore, Kubota et al. found that NO reacts with NH\n \n 4\n \n NO\n \n 3\n \n more rapidly than NH\n \n 4\n \n NO\n \n 3\n \n decomposition over H-AFX and H-CHA zeolites.\n \n 24,25\n \n However, the situation in Cu-containing zeolites is more complicated. More recently, Liu et al. investigated the FSCR mechanism over the Cu-OH site on Cu-CHA zeolite and showed the important role of BASs in the FSCR reaction.\n \n 26\n \n Currently, little is known about the state and coordination of Cu species under\n \n in-situ/operando\n \n conditions, which is crucial to uncover\u00a0the SCR reaction mechanism over Cu-based zeolites. Researchers have conducted numerous experimental and theoretical studies to explore the SSCR reaction mechanism in the past decade. Thus, the SSCR mechanism has been relatively clear, in which dynamic binuclear Cu\n \n +\n \n species are the primary active sites.\n \n 4,5,27\n \n However, the influence of NO\n \n 2\n \n on the active Cu sites and the mechanism of the NO\n \n 2\n \n -involved SCR reaction are barely discussed, and are worth exploring since NO and NO\n \n 2\n \n always coexist in actual applications.\n
\n\n In this study, the SCR reaction over the Cu-SSZ-13 catalyst in the presence of both NO and NO\n \n 2\n \n was studied by kinetic measurements.\n \n In-situ\n \n /\n \n operando\n \n X-ray absorption fine structure (XAFS) measurements were applied to reveal the state of copper species under SSCR (only NO as NO\n \n \n x\n \n \n ), FSCR (mixture of equal NO and NO\n \n 2\n \n as NO\n \n \n x\n \n \n ) and NO\n \n 2\n \n -SCR (only NO\n \n 2\n \n as NO\n \n \n x\n \n \n ) reaction conditions. Density functional theory (DFT) calculations were conducted to identify the NO\n \n 2\n \n -involved SCR reaction pathways. These results provide new insights into the role of NO\n \n 2\n \n in the NH\n \n 3\n \n -SCR reaction and shed light on the actual application of Cu-SSZ-13 catalysts in the presence of both NO and NO\n \n 2\n \n .\n
\n\n \n Kinetic studies of NO\n \n \n \n x\n \n \n \n conversion under SSCR, FSCR and NO\n \n \n \n 2\n \n \n \n -SCR conditions.\n \n We first carried out kinetic studies on the SSCR reaction, with the results shown in Fig.\n \n 1\n \n . The SSCR rate increases linearly with the square of Cu loading when the Cu loading is below 1.7 wt.% (magnified in Fig.\n \n 1\n \n b) due to the participation of Cu dimers in O\n \n 2\n \n activation, which is consistent with previous studies\n \n \n 4\n \n ,\n \n 5\n \n \n . The increase trend slows down with further rise in the Cu loading. The turnover frequency (TOF) shows a volcano-type tendency, with a maximum at Cu loading of 1.7 wt.% (Fig.\n \n 1\n \n c). The increase in TOF at low Cu loading is attributed to the quadratic increase in the SSCR rate. The underutilization of active Cu sites due to mass-transfer limitations results in a decline in TOF at high Cu loading.\n \n \n 28\n \n \n The activation energy (Ea) and pre-exponential factor (A) both increase with the increase in Cu loading, which was also observed by Gao et al.\n \n \n 28\n \n \n These results can be explained by the transformation of the SSCR from a locally homogeneous reaction to a heterogeneous reaction with increasing Cu loading, in which Cu dimer and Cu monomer species serve as primary active centers, respectively.\n \n \n 4\n \n ,\n \n 28\n \n \n
\n\n Figure S1 shows the NO\n \n \n x\n \n \n , NO and NO\n \n 2\n \n conversion levels over Cu-SSZ-13 with different Cu loadings under steady-state FSCR conditions. We normalized the NO and NO\n \n 2\n \n reaction rates by catalyst weight as a function of Cu loading, with the results shown in Fig.\n \n 2\n \n a and Fig.\n \n 2\n \n b, respectively. NO reduction was severely suppressed at low temperatures. The extremely low NO conversion at low temperatures was generally thought to be resulted from zeolite pore blocking by the formation of stable NH\n \n 4\n \n NO\n \n 3\n \n .\n \n \n 21\n \n ,\n \n 23\n \n \n . Interestingly, the NO\n \n 2\n \n reduction markedly decreased with the increase in Cu loading, while it increased as the number of BASs rose at low temperatures (Fig.\n \n 2\n \n b,\n \n 2\n \n c\n \n and Fig. S2\n \n ). This demonstrated that the BASs participated in the reduction of NO\n \n 2\n \n , which was also observed in the NO\n \n 2\n \n -SCR reaction (\n \n Fig. S3\n \n ). Moreover, the turnover frequency (TOF) of NO\n \n 2\n \n on BASs hardly changed as a function of BAS amounts.\n \n Fig. S4\n \n presents the NO\n \n 2\n \n reaction rate as a function of Cu loading and BASs under NO\n \n 2\n \n -SCR conditions, which showed the same trend as that with the co-existence of NO and NO\n \n 2\n \n . The above results indicated that NO\n \n 2\n \n primarily reacted at BASs while NO was difficult to be reduced in the presence of NO\n \n 2\n \n . It is generally known that NO can be effectively reduced at Cu sites. However, the formation of NH\n \n 4\n \n NO\n \n 3\n \n impedes NO access to the active Cu sites. Instead, NO reacts with NH\n \n 4\n \n NO\n \n 3\n \n at BASs to form N\n \n 2\n \n through reaction (3). As a result, the NO conversion was relatively low, while NO\n \n 2\n \n showed high conversion through reaction with NH\n \n 3\n \n to form NH\n \n 4\n \n NO\n \n 3\n \n at low temperatures.\n
\n\n \n Wavelet transform analysis of\n \n \n in-situ/operando\n \n \n EXAFS measurements.\n \n Further, we conducted\n \n in-situ/operando\n \n XAFS experiments on Cu-SSZ-13 samples to uncover the valence state and coordination of copper species under different conditions. Wavelet transform (WT) analysis of extended X-ray absorption fine structure (EXAFS) spectra is a powerful technique to resolve overlapping contributions from different neighbor atoms at close distances around the absorber. As shown in Fig.\n \n 3\n \n a, the pretreated sample shows a distinct first shell peak at (4.5 \u00c5\n \n \u22121\n \n , 1.3 \u00c5), which is associated with contributions from framework oxygen atoms. This result suggested that the copper species mainly exist as fw-Cu\n \n 2+\n \n species, which have high coordination numbers.\n \n \n 27\n \n \n For the second shell sphere (R(\u00c5) > 2 \u00c5), two lobes, at (3.5 \u00c5\n \n \u22121\n \n ,2.8 \u00c5) and (6.5 \u00c5\n \n \u22121\n \n 3.3 \u00c5), are well-resolved due to the different backscattering properties of various atoms, which strongly depend on the atomic number. The first lobe is assigned to the second-shell oxygen atom due to the low k value of oxygen atoms. The latter one is attributed to the signals from the Si or Al atoms of the framework. Although some studies attributed the latter lobe to the Cu-Cu contributions in oxygen-bridged Cu dimers,\n \n \n 29\n \n ,\n \n 30\n \n \n we scarcely observed CuO\n \n \n x\n \n \n species in X-ray absorption near edge structure (XANES) profiles (\n \n Fig. S5a\n \n ) and did not carry out the procedure of introducing O\n \n 2\n \n to NH\n \n 3\n \n -treated Cu-SSZ-13, to form oxygen-bridged Cu dimers with four NH\n \n 3\n \n ligands. Therefore, we deduced that the lobe at 6.5 \u00c5\n \n \u22121\n \n is primarily from the framework Si or Al atoms in the second shell in this work. In fact, the copper species in Cu-SSZ-13 are initially in the solvated state as [Cu(H\n \n 2\n \n O)n]\n \n 2+\n \n under ambient conditions, which weakens the interaction between copper species and the zeolite framework.\n \n \n 27\n \n ,\n \n 31\n \n \n High-temperature treatment in O\n \n 2\n \n /N\n \n 2\n \n removes the coordinated water molecules and oxidizes copper species to Cu\n \n 2+\n \n . As a result, the copper species are in a high valence state and strongly interact with the zeolite framework through electrostatic forces.\n
\n\n After NO adsorption, Cu\n \n 2+\n \n ions are partially reduced, resulting in a slight decrease in the coordination numbers (CNs) of the first shell, denoted by the decrease and weakening of the colored area (Fig.\n \n 3\n \n b). The lobes resulting from the contributions of the second shell stretched to (3.5 \u00c5\n \n -\n \n 1\n \n \n , 3.1 \u00c5) and (6.5 \u00c5\n \n -\n \n 1\n \n \n , 3.7 \u00c5), respectively. When the pretreated sample was exposed to an NH\n \n 3\n \n or NO+NH\n \n 3\n \n atmosphere, the signal of the first shell sharply decreased (Fig.\n \n 3\n \n c and\n \n 3\n \n d), suggesting that the CNs of the Cu ions significantly declined due to their reduction. Moreover, the two lobes are not well-resolved in the spectra, indicating a decrease in the scattering from the second shell. This is consistent with the formation of dynamic [Cu(NH\n \n 3\n \n )\n \n 2\n \n ]\n \n +\n \n species, which is supported by the appearance of feature B in\n \n Fig. S5a\n \n after NH\n \n 3\n \n or NO+NH\n \n 3\n \n adsorption. After oxidation by O\n \n 2\n \n and NO\n \n 2\n \n , the CNs of the first shell increased to a level similar to that of the pretreated sample, accompanied by the formation of two well-resolved lobes at the second shell (Fig.\n \n 3\n \n e and\n \n 3\n \n f). This demonstrated that Cu(NH\n \n 3\n \n )\n \n 2\n \n \n +\n \n species are oxidized into Cu\n \n 2+\n \n ions and that the interaction between the Cu\n \n 2+\n \n ions and the zeolite framework is recovered. Compared with oxidation by O\n \n 2\n \n , oxidation by NO\n \n 2\n \n resulted in a higher signal for the lobe at ~6.5 \u00c5\n \n -\n \n 1\n \n \n , indicating that more Cu\n \n +\n \n species are oxidized into framework-bonded Cu\n \n 2+\n \n ions (fw-Cu\n \n 2+\n \n ) during the reaction with NO\n \n 2\n \n .\n
\n\n Figure\n \n 4\n \n g-\n \n 4\n \n i depicts 2D plots of the WT EXAFS spectra under SSCR, FSCR and NO\n \n 2\n \n -SCR conditions. Under SSCR conditions, the WT EXAFS spectra resemble the ones in Fig.\n \n 3\n \n c and\n \n 3\n \n d. The first shell peak weakened, indicating a decrease in the CNs of Cu species under SSCR conditions. The absence of the lobes at the second shell suggests the easy mobility of the copper complex due to the NH\n \n 3\n \n solvation effect. The results proved the existence of large amounts of dynamic Cu(NH\n \n 3\n \n )\n \n 2\n \n \n +\n \n species during the SSCR reaction. In the presence of NO\n \n 2\n \n , however, the CNs of the first shell significantly increased, indicating the oxidation of copper species. Moreover, two well-resolved lobes at the second shell are observed, suggesting that oxidization leads to the copper species becoming closely coordinated with the zeolite framework, which limits their mobility during the SCR reaction. The WT EXAFS spectra are consistent with the Fourier-transformed (FT) EXAFS results (\n \n Fig. S6 and Table S1\n \n ), which are discussed in detail in the Supporting Information.\n
\n\n \n DFT calculation.\n \n Next, we used the DFT method to calculate various possible FSCR reaction pathways. There are two possible reaction mechanisms when FSCR occurs at active Cu sites, corresponding to NH\n \n 2\n \n NO and NH\n \n 4\n \n NO\n \n 3\n \n reaction pathways, respectively. The NH\n \n 2\n \n NO reaction pathways over isolated [Cu(NH\n \n 3\n \n )\n \n 2\n \n ]\n \n +\n \n and dimer Cu ions are shown in\n \n Fig. S7 and S8\n \n , respectively. The reduction process resembles what takes place in the SSCR reaction, in which Cu\n \n II\n \n OH(NH\n \n 3\n \n )\n \n 2\n \n species are reduced to [Cu(NH\n \n 3\n \n )\n \n 2\n \n ]\n \n +\n \n by NH\n \n 3\n \n and NO with an energy barrier of 0.32 eV, accompanied by the formation of two H\n \n 2\n \n O and one N\n \n 2\n \n .\n \n \n 4\n \n ,\n \n 5\n \n \n By contrast, the reduced [Cu(NH\n \n 3\n \n )\n \n 2\n \n ]\n \n +\n \n species are oxidized by NO\n \n 2\n \n rather than O\n \n 2\n \n . The rate-determining steps involve the reduction of Cu\n \n 2+\n \n (NH\n \n 3\n \n )\n \n 2\n \n NO\n \n 2\n \n , with high energy barriers of 1.65 and 1.58 eV for Cu monomer and dimer, respectively. This result is comparable with the value reported by Janssens et al.\n \n \n 32\n \n \n More details regarding the reaction pathway of intermediate NH\n \n 2\n \n NO decomposition are shown in\n \n Fig. S9\n \n .\n
\n\n Given the experimental result that NH\n \n 4\n \n NO\n \n 3\n \n forms easily during the FSCR reaction, we next show the results of the NH\n \n 4\n \n NO\n \n 3\n \n reaction pathway on Cu active sites and BASs in Fig.\n \n 4\n \n and\n \n 5\n \n , respectively. The fw-Cu\n \n 2+\n \n OH structure is used as the active site since fw-Cu\n \n 2+\n \n OH dominates during the FSCR and NO\n \n 2\n \n -SCR reactions, as indicated by the operando XAFS results (\n \n Fig. S6\n \n ). As shown in Fig.\n \n 4\n \n , fw-Cu\n \n 2+\n \n OH first adsorbs an NH\n \n 3\n \n molecule to reach a coordinatively saturated state, which interacts with NO\n \n 2\n \n to form an HNO\n \n 3\n \n molecule without any energy barrier. The B species is actually considered to be NH\n \n 4\n \n NO\n \n 3\n \n adsorbed on Cu sites. Next, the adsorbed HNO\n \n 3\n \n reacts with NO from the gas phase with an energy barrier of 0.87 eV, resulting in the formation of adsorbed HNO\n \n 2\n \n and the release of an NO\n \n 2\n \n molecule (C\u2192D). Then, the adsorbed HNO\n \n 2\n \n reacts with the NH\n \n 3\n \n ligand to generate NH\n \n 2\n \n NO and H\n \n 2\n \n O. NH\n \n 2\n \n NO is easily decomposed into N\n \n 2\n \n and H\n \n 2\n \n O through a series of H-migration and isomerization processes (\n \n Fig. S9\n \n ).\n \n \n 33\n \n \n As the desorption of N\n \n 2\n \n and H\n \n 2\n \n O molecules occurs, NH\n \n 3\n \n and NO\n \n 2\n \n are adsorbed at the Cu site and react to generate NH\n \n 2\n \n NO and -OH groups. With the decomposition of NH\n \n 2\n \n NO into N\n \n 2\n \n and H\n \n 2\n \n O, the fw-Cu\n \n 2+\n \n OH site is regenerated. The rate-determining step of the FSCR cycle over the fw-Cu\n \n 2+\n \n OH site corresponds to the reaction of adsorbed HNO\n \n 2\n \n with NH\n \n 3\n \n ligand to produce NH\n \n 2\n \n NO and H\n \n 2\n \n O (E\u2192F), with an energy barrier of 1.58 eV.\n
\n\n The FSCR reaction pathway at BASs is displayed in Fig.\n \n 5\n \n . NH\n \n 3\n \n is adsorbed on the BASs to form NH\n \n 4\n \n \n +\n \n species. Two NO\n \n 2\n \n molecules interact with the NH\n \n 4\n \n \n +\n \n species to form NH\n \n 4\n \n NO\n \n 3\n \n species and release an NO molecule without any energy barrier (B\u2192C). The release of NO was also observed in our previous studies during NO\n \n 2\n \n adsorption on H-SSZ-13 zeolite.\n \n \n 34\n \n \n Then, NO interacts with an NH\n \n 3\n \n from the gas phase to form an NH\n \n 3\n \n \u2219\u2219\u2219NO complex, which further reacts with NH\n \n 4\n \n NO\n \n 3\n \n to form NH\n \n 4\n \n \n +\n \n , HNO\n \n 3\n \n , and NH\n \n 2\n \n NO via an H-migration process (C\u2192D). NH\n \n 2\n \n NO is decomposed into N\n \n 2\n \n and H\n \n 2\n \n O. Subsequently, HNO\n \n 3\n \n reacts with NO from the gas phase, resulting in the formation of HNO\n \n 2\n \n and the release of an NO\n \n 2\n \n molecule (G\u2192H). Reaction between HNO\n \n 2\n \n and NH\n \n 4\n \n \n +\n \n species leads to the formation of an NH\n \n 3\n \n \u2219\u2219\u2219NO complex and an H\n \n 2\n \n O molecule. The NH\n \n 3\n \n \u2219\u2219\u2219NO complex transfers an H atom to regenerate the BAS and changes into NH\n \n 2\n \n NO, which then decomposes into N\n \n 2\n \n and H\n \n 2\n \n O. The whole catalytic cycle is completed. The overall energy barrier of the FSCR over the Br\u00f8nsted acid site is 1.27 eV, which corresponds to the reaction between NH\n \n 4\n \n NO\n \n 3\n \n and the NH\n \n 3\n \n \u2219\u2219\u2219NO complex, much lower than that over various Cu sites. The DFT-calculated results indicate that the FSCR process in the SSZ-13 zeolite system tends to occur at BASs.\n
\n\n In summary, we found that NO\n \n 2\n \n leads a deep oxidation of copper species as fw-Cu\n \n 2+\n \n , which significantly inhibits the locally homogeneous SSCR over dynamic dimer Cu sites by combining analysis of\n \n in situ/operando\n \n spectroscopic measurements with DFT calculations demonstrates. As a result, the FSCR reaction occurs primarily at the BASs even though it has a higher energy barrier (1.27 eV) than the locally homogeneous SSCR reaction (about 1.0 eV\n \n \n 32\n \n \n ). This work reveals the origin of the abnormal NH\n \n 3\n \n -SCR behavior over the commercial Cu-SSZ-13 catalyst in the presence of NO\n \n 2\n \n .\n
\n\n \n Sample preparation.\n \n The initial Cu-SSZ-13 zeolite was\n \n in-situ\n \n synthesized by a one-pot method.\n \n \n 15\n \n \n Due to the excess Cu in the initial product, aftertreatments were required to optimize the Cu contents and distribution. In detail, the as-synthesized Cu-SSZ-13 was post-treated with 0.1 mol/L HNO\n \n 3\n \n at 80\u00b0C for 12 h to remove CuO\n \n \n x\n \n \n species. After calcination at 600\u00b0C, the sample was stirred in NH\n \n 4\n \n NO\n \n 3\n \n solution (0.01 ~ 0.2 mol/L) at 40\u00b0C for the second post-treatment process, followed by filtration, washing, drying and calcination at 600\u00b0C. The obtained Cu-SSZ-13 catalysts were Al-rich zeolites with Si/Al of ~5 and various Cu loading from 0.4 to 3.8 wt.% (\n \n Table S2\n \n ).\n
\n\n \n Catalyst evaluation.\n \n The standard SCR (SSCR), fast SCR (FSCR) and slow SCR (NO\n \n 2\n \n -SCR) reactions were carried out in a fixed-bed flow reactor system with an online Nicolet Is50 spectrometer, which was used to detect the concentration of reactants and products. The SSCR conditions included 500 ppm NO and 500 ppm NH\n \n 3\n \n ; the FSCR conditions included 250 ppm NO, 250 ppm NO\n \n 2\n \n and 500 ppm NH\n \n 3\n \n ; the NO\n \n 2\n \n -SCR conditions included 300 ppm NO\n \n 2\n \n and 500 ppm NH\n \n 3\n \n . All the conditions included 5% H\n \n 2\n \n O, 5%O\n \n 2\n \n and N\n \n 2\n \n balance. The total flow rate was 500 mL/min. The NO\n \n \n x\n \n \n (NO and NO\n \n 2\n \n ) conversion was calculated at steady state:\n
\n| \n | \n\n \n \n \n \\(NO conversion=\\left(1- \\frac{{\\left[NO\\right]}_{out}}{{\\left[NO\\right]}_{in}}\\right) \\times 100\\text{\\%}\\)\n \n \n \n | \n \n \n (1)\n \n | \n
|---|---|---|
| \n | \n\n \n \n \n \\({NO}_{2} conversion=\\left(1- \\frac{{\\left[{NO}_{2}\\right]}_{out}}{{\\left[{NO}_{2}\\right]}_{in}}\\right) \\times 100\\text{\\%}\\)\n \n \n \n | \n \n \n (2)\n \n | \n
| \n | \n\n \n \n \n \\({NO}_{x} conversion=\\left(1- \\frac{{\\left[\\text{N}\\text{O}\\right]}_{\\text{o}\\text{u}\\text{t}}+{\\left[{\\text{N}\\text{O}}_{2}\\right]}_{\\text{o}\\text{u}\\text{t}} }{{\\left[\\text{N}\\text{O}\\right]}_{\\text{i}\\text{n}}+ {\\left[{\\text{N}\\text{O}}_{2}\\right]}_{\\text{i}\\text{n}}}\\right) \\times 100\\text{\\%}\\)\n \n \n \n | \n \n \n (3)\n \n | \n
\n To conduct the kinetic studies, the high gas hourly space velocity (GHSV) was controlled by adjusting the catalyst weight. The GHSV of SSCR, FSCR and NO\n \n 2\n \n -SCR were about ~800,000 h\n \n -1\n \n , ~1,000,0000 h\n \n -1\n \n and ~2,000,000 h\n \n -1\n \n , respectively. The reaction rates (r) in this study are normalized by catalyst weight based on equation (1). The activation energies (Ea) were calculated by the Arrhenius equation (2).\n
\n| \n | \n\n \n \n \n \\(r=\\frac{{F}_{{NO}_{x}}\\bullet {X}_{{NO}_{x}}}{{W}_{cat}}\\)\n \n \n \n | \n \n \n (1)\n \n | \n
|---|---|---|
| \n | \n\n \n \n \n \\(r={\\left[{NO}_{x}\\right]}_{0}A{e}^{\\left(-\\frac{Ea}{RT}\\right)}\\)\n \n \n \n | \n \n \n (2)\n \n | \n
\n where F\n \n \n NOx\n \n \n represents the NO\n \n \n x\n \n \n flow rate (mol/s), X\n \n \n NOx\n \n \n represents the NO\n \n \n x\n \n \n conversion, W\n \n cat\n \n is the mass of the catalyst (g), and [NO\n \n \n x\n \n \n ]\n \n 0\n \n is inlet concentration of NO\n \n \n x\n \n \n . NOx represents the NO, NO\n \n 2\n \n or both.\n
\n\n \n Characterizations.\n \n The elemental composition of the catalysts was measured by inductively coupled plasma atomic emission spectroscopy (ICP-AES). N\n \n 2\n \n adsorption-desorption analysis of the samples was conducted on a Micromeritics ASAP 2020 instrument. The acid site distribution and contents were measured by NH\n \n 3\n \n temperature-programmed desorption (NH\n \n 3\n \n -TPD) using the NH\n \n 3\n \n -SCR activity measurement instrument. Samples of about 30 mg were used and pretreated in 10% O\n \n 2\n \n /N\n \n 2\n \n at 500\u00a0\u00b0C\u00a0for 30 min before cooling down to 120\u00a0\u00b0C. Then, the gas was changed to 500 ppm NH\n \n 3\n \n /N\n \n 2\n \n for 60 min, followed by N\n \n 2\n \n purging for 60min. Finally, the temperature was raised to 700\u00a0\u00b0C\u00a0at a rate of 10\u00a0\u00b0C/min.\n
\n\n The\n \n in situ/operando\n \n X-ray absorption fine structure\n \n (in situ/operando\n \n XAFS) experiments were performed in the 1W1B beamline of Beijing Synchrotron Radiation Facility (BSRF). The absorption data from -200 eV to 800 eV of the Cu K-edge (8979 eV) were collected. The sample was first pretreated in O\n \n 2\n \n /He at 500\u00a0\u00b0C\u00a0for 30 min before decreasing the temperature to 200\u00a0\u00b0C, after which the Pre. spectra were collected. Then, the sample was exposed to 500 ppm NH\n \n 3\n \n /He, 500 NO/ He and 500 ppm NH\n \n 3\n \n / He + 500 ppm NO/ He for 60 min, respectively, and spectra were collected. After reduction by (NO+NH\n \n 3\n \n )/He, the sample was exposed to 5% O\n \n 2\n \n /N\n \n 2\n \n and 500 ppm NO\n \n 2\n \n /N\n \n 2\n \n for 60 min, respectively, to obtain the absorption data for the oxidized sample. Moreover, the operando absorption data were collected after the pretreated samples were exposed to SSCR, FSCR and NO\n \n 2\n \n -SCR atmospheres for 60\u00a0min. The X-ray absorption near-edge structure (XANES) data were background-corrected and normalized using the Athena module implemented in the IFFEFIT software package.\n \n 35\n \n Extended X-ray absorption fine structure (EXAFS) data were analyzed and fitted using Athena and Artemis (3.0 < k < 13.0\u00a0\u00c5\n \n -1\n \n ). An amplitude reduction factor (S\n \n 0\n \n \n 2\n \n ) of 0.85 was used for all the fitted data sets. Wavelet transform (WT) analysis of the EXAFS was performed to precisely investigate the local coordination environment of copper species.\n
\n\n \n Computational details.\n \n Spin-polarized periodic DFT calculations were carried out with the Vienna\n \n ab initio\n \n simulation package (VASP 5.4.4)\n \n 36\n \n The Perdew\u2212Burke\u2212Ernzerhof (PBE) generalized gradient approximation was adopted with van der Waals correction proposed by Grimme\n \n \n (i.e., DFT-D3 method)\n \n 37\n \n . The Kohn-Sham orbitals were expanded with a plane-wave basis set with a cutoff energy of 500 eV, and the plane augmented wave (PAW) method was used to describe the interaction between the valence electrons and the cores\n \n 38\n \n . The DFT + U method was applied to Cu 3d states with the values of U\n \n eff\n \n = 6.0 eV to describe the on-site Coulomb interactions\n \n 33,39\n \n . During geometrical optimization, the self-consistent-field electronic energies were converged to 1\u00d710\n \n -5\n \n eV and all other atoms were fully relaxed until the maximum force on the atoms was less than 2\u00d710\n \n -2\n \n eV/\u00c5. The Brillouin zone was sampled with a Monkhorst-Pack k-point grid of 1\u00d72\u00d72. The Gaussian smearing method was utilized, with a smearing width of 0.2 eV. The transition states of elementary steps were located using the climbing image nudged elastic band (CI-NEB) method with several intermediate images between initial and final states\n \n 40,41\n \n . Thermodynamic data were processed with the VASPKIT code\n \n 42\n \n and the Gibbs free energies are calculated at 200 \u00b0C.\u00a0The SSZ-13 zeolite structure was modelled using two rhombohedral unit cells (24 tetrahedral atoms) with size of 18.84 \u00c5\u00d79.42 \u00c5\u00d79.42 \u00c5 (\n \n Figure S10\n \n ). One Si atom was replaced by one Al atom in each double 6-membered ring, resulting in a model with Si/Al ratio of 11. One H atom was introduced onto one of the O atoms connected with each Al atom to keep the structure charge-neutral. Based on previous studies\n \n 4,33\n \n , the present computational settings and models were reliable for investigating the NH\n \n 3\n \n -SCR mechanism over Cu-SSZ-13 zeolites.\n
\n\n Table of Contents Graphic\n
\n \n\n Strikingly Distinctive NH3-SCR Behavior over Cu-SSZ-13 in the Presence of NO2\n
\n \n\n This study presents a copper-catalyzed, substrate-controlled regio- and enantioselective intermolecular hydrosilylation method capable of accommodating a broad scope of alkenes and prochiral silanes. The approach offers an efficient and versatile pathway to generate enantioenriched linear and branched alkyl-substituted Si-stereogenic silanes. Key features of this reaction include mild reaction conditions, simple catalytic systems, compatibility with diverse substrates, high yields, and enantioselectivities.\n
\n\n Silicon, like carbon, belongs to Group IVA in the periodic table and is the second most abundant element in the Earth's crust.\n \n 1\n \n Comparatively, while chiral molecules containing a stereogenic carbon atom have been extensively studied and practically utilized, there has also been substantial interest in synthesizing organic silicon molecules with Si-stereogenic centers.\n \n 2\n \n These molecules play a pivotal role in various fields such as organic synthesis, functional materials, and biomedicines.\n \n 3\n \n Traditionally, the creation of silicon-centered chirality has heavily relied on stoichiometric chiral reagents or auxiliaries.4 To enhance efficiency and reduce waste, several novel strategies for synthesizing Si-stereogenic silanes through the catalytic desymmetrization of prochiral organosilanes have been developed (Scheme 1a).\n \n 5\n \n Additionally, significant advancements have been made recently in constructing silicon-stereogenic silanes from racemic starting materials through dynamic kinetic asymmetric transformation (DYKAT)\n \n 6a\u2013c\n \n and metal-catalyzed kinetic resolution (KR)\n \n 6d\n \n . Among various Si-stereogenic organosilanes, monohydrosilanes have attracted considerable attention due to their unique reactivity and promising applications.\n \n 3d\n \n
\n\n Within the realm of strategies developed for the synthesis of chiral monohydrosilanes, metal-catalyzed asymmetric intermolecular hydrosilylation of C\u2013C unsaturated bonds emerges as an efficient approach (Scheme 1b).\n \n 5, 7\u201310\n \n It is noteworthy that while enantioselective hydrosilylation of terminal alkenes has been extensively studied for creating a C-stereogenic center with a C-Si bond,\n \n 3, 5, 11\n \n only limited progress has been made in constructing a Si-stereogenic center via this strategy. To date, there have only been two reported instances of hydrosilylation of terminal olefins for the preparation of linear alkyl-substituted Si-stereogenic silanes.\n \n 9\n \n In 2018, Hou and colleagues reported a groundbreaking, catalytic, and enantioselective method for synthesizing Si-stereogenic silanes through Sc-catalyzed intermolecular hydrosilylation of terminal alkenes.\n \n 9a\n \n Subsequently, He and co-workers demonstrated an example using Rh-catalysis, yielding Si-stereogenic compounds with moderate enantioselectivities.\n \n 9b\n \n In contrast, only a few intermolecular reaction protocols have successfully enabled the simultaneous creation of both a C- and a Si-stereocenter through noble transition metal-catalyzed enantioselective Si\u2013H bond insertion of carbene species\n \n 12\n \n or Co-catalyzed intermolecular hydrosilylation of 1,3-dienes\n \n 8\n \n . However, the catalytic asymmetric synthesis of chiral monohydrosilanes with both a C- and a Si-stereogenic center, using readily available alkenes as substrates through the hydrosilylation process, remains unexplored. Therefore, the\n
\n\n
\n\n development of highly efficient and robust methodologies catalyzed by abundant metals to access chiral hydrosilanes with a Si-stereogenic center or with both a C- and a Si-stereogenic center is highly desirable. Building upon previous work on copper-catalyzed asymmetric hydrosilylation of unsaturated C\u2013C bonds,\n \n 7e, 11h, 13\n \n we present a Cu-catalyzed intermolecular regio- and enantioselective hydrosilylation of alkenes with prochiral silanes for the synthesis of diverse enantioenriched hydrosilane products (Scheme 1c).\n
\n\n [IMAGE_INTRODUCTION_1]\n
\n\n \n \n a\n \n \n Conditions:\n \n 1a\n \n (0.2 mmol),\n \n 2\n \n (0.6 mmol), copper catalyst (4 mol %) and ligand (8 mol %) were stirred at 40\u00b0C for 2 d under N\n \n 2\n \n atmosphere.\n \n \n b\n \n \n Yields were determined by\n \n 1\n \n H NMR using 1,1,2,2-tetrachloroethane as an internal standard.\n \n \n c\n \n \n The er values were determined by chiral HPLC analysis.\n \n \n d\n \n \n 8 mol% catalyst, 8.8 mol% ligand and 8.8 mol% secondary ligand.\n \n \n e\n \n \n 10 mol% catalyst, 11 mol% ligand and 11 mol% secondary ligand.\n \n \n f\n \n \n Isolated yield.\n
\n\n \n Evaluation of reaction conditions for the copper-catalyzed hydrosilylation of allylbenzene.\n \n Using allylbenzene (\n \n 1a\n \n ) as the model substrate, the Cu-catalyzed hydrosilylation of alkenes was first investigated by assessing the steric effects of different dihydrosilanes, including PhMeSiH\n \n 2\n \n (\n \n 2a\n \n ), mesityl(methyl)silane (\n \n 2b\n \n ), and Ph(\n \n \n t\n \n \n Bu)SiH\n \n 2\n \n (\n \n 2c\n \n ). Various copper precursors and ligands were also examined for the model reaction. The results of selected experiments are summarized in Table\u00a01. The reactions between allylbenzene and the dihydrosilanes R\n \n 1\n \n R\n \n 2\n \n SiH\n \n 2\n \n were conducted using 4 mol% Cu(OAc)\n \n 2\n \n and 8 mol% (\n \n R\n \n ,\n \n R\n \n )-Ph-BPE (\n \n L\n \n \n \n 1\n \n \n ) at 40\u00b0C (entries 1\u2009\u2212\u20093). Notably, it was observed that the reaction of allylbenzene with mesityl(methyl)silane (\n \n 2b\n \n ) produced the\n \n anti\n \n -Markovnikov product at 40\u00b0C with moderate conversion and a 94:6 enantiomeric ratio (entry 2). Despite PhMeSiH\n \n 2\n \n (\n \n 2a\n \n ) exhibiting a higher conversion rate compared to mesit-yl(methyl)silane (\n \n 2b\n \n ) and Ph(\n \n \n t\n \n \n Bu)SiH\n \n 2\n \n (\n \n 2c\n \n ), the latter demonstrated better enantioselectivities. A slight enhancement in enantioselectivity was noted upon transitioning from Cu(OAc)\n \n 2\n \n to CuOAc as the copper precursor (entry 4). Conversely, the use of CuCl as the catalyst did not yield any reaction (entry 5). By increasing the catalyst loading to 8 mol%, the yield of\n \n 3ab\n \n could be elevated to 58% (entry 6). Encouragingly, incorporating a secondary ligand proved advantageous for this reaction (entries 7\u2009\u2212\u200910),\n \n 14\n \n with CyJohnPhos delivering the optimal outcome (95:5 enanti-omeric ratio; entry 10). Ultimately, the best result was obtained when conducting the reaction of\n \n 1a\n \n with\n \n 2b\n \n (3.0 equiv) in the presence of CuOAc (10 mol%), (\n \n R\n \n ,\n \n R\n \n )-Ph-BPE (\n \n L\n \n \n \n 1\n \n \n , 11 mol%), and CyJohnPhos (11 mol%) at 40\u00b0C for 2 days, resulting in a 75% yield and a 95:5 enantiomeric ratio (entry 11).\n
\n\n \n Scope of linear-selective hydrosilylation of alkenes.\n \n Following the identification of an active catalyst and optimized conditions for the\n \n anti\n \n -Markovnikov hydrosilylation of allylbenzene (entry 11 in Table 1), we proceeded to explore the substrate scope. Key findings from this investigation are outlined in Scheme 2. Initially, we examined the hydrosilylation of various alkenes using\n \n 2b\n \n or\n \n 2c\n \n . Allylbenzene derivatives containing both electron-donating and electron-withdrawing groups could efficiently undergo reactions with mesityl(methyl)silane (\n \n 2b\n \n ) to yield the desired chiral linear products, typically achieving moderate to excellent yields with good enantioselectivities (\n \n 3ab\n \n \u2013\n \n 3fb\n \n ).\n
\n\n In general, electron-withdrawing allylbenzene proved to be a superior substrate (\n \n 3db\n \n ) that yielded the hydrosilylation product with higher efficiency. The efficiency and enantioselectivity of the desired products were slightly influenced as the carbon chain prolonged, as observed in\n \n 3hb\n \n and\n \n 3ib\n \n . Additionally, heteroaryl-substituted alkenes served as suitable substrates, producing chiral silanes with high efficiency and enantioselectivity (\n \n 3jb\n \n \u2013\n \n 3lb\n \n ). Various functional groups such as amino (\n \n 3mb\n \n ), phenoxy (\n \n 3nb\n \n ), thioether (\n \n 3ob\n \n ), silyloxy (\n \n 3pb\n \n ), halogens (\n \n 3qb\n \n ) were all compatible. These reactions proceeded smoothly, yielding the corresponding tertiary silane products with good yields (57\u201391%) and high enantioselectivities (91:9 to 96:4 er). In cases where the substrate contained both terminal and internal olefin units, the reaction selectively occurred at the less sterically hindered terminal olefin, leaving the internal olefin moiety intact (\n \n 3rb\n \n ). When\n \n tert\n \n -butyl group-substituted silane was utilized, the desired products were obtained with improved efficiency and enantioselectivity (\n \n 3cc\n \n \u2013\n \n 3uc\n \n ). The absolute configuration of 3tc was determined through X-ray diffraction analysis (CCDC: 2358701). Furthermore, 1,4-diallylbenzene was selectively hydrosilylated, resulting in the bis-silane product\n \n 3vc\n \n with a yield of 71%, an enantiomeric ratio of 98:2 er, and a diastereomeric ratio of 88:12 dr. Prochiral silanes were examined under identical conditions (Scheme 2). The efficiency and enantioselectivity of the product were significantly influenced by the steric hindrance of the silane. Reactions involving silanes bearing various bulky aryl or alkyl groups exhibited high enantioselectivity, albeit with a slight decrease in efficiency (\n \n 3ac\n \n \u2013\n \n 3af\n \n ). A variety of readily available alkylphenylsilanes proved to be suitable substrates (\n \n 3ah\n \n \u2013\n \n 3aj\n \n ). Although the desired products were obtained when the arylmethylsilane contained electron-donating or electron-withdrawing groups (\n \n 3ag\n \n ,\n \n 3ak\n \n ), their efficiency and enantioselectivity were negatively affected.\n
\n\n \n Evaluation of reaction conditions for the copper-catalyzed hydrosilylation of styrene.\n \n Based on the aforementioned investigation, we aimed to expand the substrate scope by incorporating aryl alkenes (Table 2). Despite compound (\n \n R\n \n ,\n \n R\n \n )-\n \n 5ak\n \n demonstrating excellent efficiency and a favorable enantiomeric ratio, the diastereomeric ratio achieved under the optimized conditions (entry 11, Table 1) was only moderately satisfactory (entry 1). It is worth noting that introducing a second ligand in this reaction didn\u2019t have a positive effect (entry 2). When switching from CuOAc to Cu(OAc)\n \n 2\n \n as the copper precursor (entry 3), a slight increase in diastereoselectivity was observed. Although different chiral ligands were tested, the desired outcomes were not achieved (entries 4\u20136). En-couragingly, increasing the amounts of the chiral ligand yielded positive results (entries 7\u20138). Reducing the reaction time to 36 hours did not significantly alter the results, resulting in the target product (\n \n R\n \n ,\n \n R\n \n )-\n \n 5ak\n \n being obtained in a 91% isolated yield, with a 98:2 er and up to a 95:5 dr ratio (entry 9).\n
\n\n \n Table\n \n \n 2\n \n \n . Optimization of reaction conditions for the copper-catalyzed hydrosilylation of allylbenzene\n \n \n .\n \n \n \n \n a\n \n \n \n \n \n \n , b, c, d\n \n \n \n
\n\n \n \n \n [IMAGE_RESULTS_AND_DISCUSSION_1]\n \n \n \n
\n
\n \n \n a\n \n \n Conditions:\n \n 4a\n \n (0.2 mmol),\n \n 2k\n \n (0.6 mmol), copper catalyst (4 mol%) and ligand were stirred at 40\u00b0C for 2 d under N\n \n 2\n \n atmosphere.\n \n \n b\n \n \n Yields were determined by\n \n 1\n \n H NMR using 1,1,2,2-tetrachloroethane as an internal standard.\n \n \n c\n \n \n The dr values were determined by GC analysis of the crude reaction mixture.\n \n \n d\n \n \n The er values were determined by chiral HPLC analysis.\n \n \n e\n \n \n 10 mol% CuOAc.\n \n \n f\n \n \n 36 h.\n \n \n g\n \n \n 24 h.\n \n \n h\n \n \n Isolated yield.\n
\n\n \n Scope of branched-selective hydrosilylation of alkenes.\n \n Under the optimized conditions, we explored the substrate scope as illustrated in Scheme 3. Enantioenriched branched silanes were successfully synthesized, incorporating halo-genated (\n \n 5bk\n \n ), electron-rich (\n \n 5ck\n \n ), electron-deficient (\n \n 5fk\n \n ) aryl groups, or a fused aromatic ring (\n \n 5gk\n \n ). Yields varied between 70% and 94%, with diastereomeric ratios ranging from 86:14 to 95:5, and enantiomeric ratios reaching up to 98:2. Notably, the reaction did not accommodate aryl bromide and iodide substrates. A styrene derivative carrying a methylthio group (\n \n 5ek\n \n ) was obtained in moderate yield but displayed poor diastereo- and enantioselectivity. Of significance was the selective transformation of aryl alkenes bearing trisubstituted olefin moieties with the silane reagent resulting in high efficiency and stereoselectivity (\n \n 5dk\n \n ). Furthermore, a wide array of heteroaryl-substituted alkenes proved to be suitable substrates for the production of chiral silanes exhibiting both C- and Si-stereogenic cen-ters efficiently, enantioselectively, and with moderate diastereoselectivity (\n \n 5ik\n \n \u2013\n \n 5lp\n \n ). The absolute configuration of the chiral branched alkylsilane product was conclusively determined through X-ray crystallographic analysis of compound\n \n 5lp\n \n (CCDC: 2358711).\n
\n\n Next, the investigation focused on the scope of prochiral silanes. Reactions with readily available arylmethylsilanes exhibited remarkable efficiency, diastereoselectivity, and enantioselectivity (\n \n 5ca\n \n ,\n \n 5cl\n \n \u2013\n \n 5cp\n \n ). Another alkylarylsilane (\n \n 5ci\n \n ) was converted to the target products with a yield of 92%, a diastereomeric ratio (dr) of 94:6, and an enantiomeric ratio (er) of 97:3. However, diarylsilanes and sterically hindered silanes such as mesityl(methyl)silane (\n \n 2b\n \n ) and Ph(\n \n \n t\n \n \n Bu)SiH\n \n 2\n \n (\n \n 2c\n \n ) were found to be unsuitable substrates for these reactions.\n
\n\n \n Mechanistic Investigation.\n \n Deuterium labeling experiments were conducted as part of the investigation into the reaction mechanism. Initially, the isotopically labeled substrate methyl(phenyl)silane-\n \n d\n \n \n \n 2\n \n \n (\n \n 2a-\n \n \n d\n \n \n \n 2\n \n \n ) was subjected to standard conditions, leading to the formation of deuterated products\n \n 3aa-\n \n \n d\n \n \n \n 2\n \n \n and\n \n 5aa-\n \n \n d\n \n \n \n n\n \n \n , illustrated in Scheme 5a. To enhance our comprehension of the reversibility factors that influence the reaction steps, multiple control experiments were performed. When styrene (\n \n 4a\n \n ),\n \n 2a-\n \n \n d\n \n \n \n 2\n \n \n , and (4-methoxyphenyl)(methyl)silane (\n \n 2m\n \n ) were simultaneously employed under standard conditions, the integration of the Si- H/D peaks of\n \n 5aa-\n \n \n d\n \n \n \n m\n \n \n and\n \n 5am-\n \n \n d\n \n \n \n m\n \n \n gives a ratio of approximately 1:1. The obtained result provides further evidence for the reversibility of migratory insertion and\n \n \u03b2\n \n -hydride elimination.\n \n 15\n \n Another crossover experiment was performed using 1.0 equiv\n \n 2a-\n \n \n d\n \n \n \n 2\n \n \n and\n \n 2m\n \n reacting without an alkene under standard conditions. The presence of deuterium crossover was confirmed through\n \n 1\n \n H NMR analysis. These findings strengthen our conclusion that the generation of copper hydride species, migratory insertion, and\n \n \u03b2\n \n -hydride elimination display reversibility (Scheme 5b). Based on this data, we propose a potential reaction mechanism for the copper-catalyzed intermolecular regiodivergent and stereoselective hydrosilylation of alkenes (Scheme 5c).\n
\n\n \n Computational studies.\n \n To elucidate the origin of the regio- and stereoselectivities observed in the reaction, we resorted to DFT studies on the hydride-insertion and the subsequent metathesis steps (See SI for the details). The potential energy surface leading to the linear products was explored with 1-butene as the model substrate (Scheme 6a). The migratory insertion step from\n \n S1\n \n to\n \n TS1\n \n has a barrier of 10.9 kcal/mol, generating the linear alkyl Cu(I)\n \n P1\n \n , with an energy downhill of 23.1 kcal/mol. Subsequent silane association leads to further energy downhill of 3.3 kcal/mol. From\n \n S2\n \n , the conformation space of the metathesis step was mapped similarly. Our calculation shows that the most energetically favored pathway proceeds through transition structure\n \n TS2\n \n _conf\n \n B\n \n _\n \n R\n \n (in favor of the\n \n R\n \n -product) with a barrier of only 13.4 kcal/mol, which is much more favored (by 13.0 kcal/mol) than the\n \n \u03b2\n \n -H elimination backward pathway (26.4 kcal/mol), in congruent with the absence of H/D scrambling observed experimentally (Scheme 5a). The second lowest-energy TS is\n \n TS2\n \n _conf\n \n A\n \n _\n \n S\n \n which is 3.8 kcal/mol higher, in good agreement of the sense and degree of enantiocontrol observed experimentally.\n
\n\n For the reaction with styrene as starting material, we systematically sampled the conformational space of the hydrocupration and metathesis transition states in the formation of the branched product. It was found that the relative energy of the conformational of the\n \n R\n \n configuration was lower than that of the\n \n S\n \n configuration, where\n \n TS1\u2019\n \n \n _\n \n conf\n \n 2\n \n \n _\n \n \n re\n \n was the conformation with the lowest energy (Scheme 5b). The results show that the Cu\u2013H insertion TS forming terminal C\u2013Cu bond (\n \n TS1\u2019\n \n _r.r.) is 8.8 kcal/mol higher than\n \n TS1\u2019\n \n _conf\n \n 2\n \n _\n \n re\n \n , which is consistent with the exclusive regioselectivity observed experimentally. The most stable conformation\n \n TS1\u2019\n \n _conf\n \n 2\n \n _\n \n re\n \n leads to the lowest energy benzyl-Cu(I) intermediate\n \n P1\u2019\n \n _conf\n \n 2\n \n _\n \n re\n \n , which then associates with the phenyl silane to give the \u03c3-complex\n \n S2\u2019\n \n with a 3.8 kcal/mol energy drop. Subsequently, the conformational space of the metathesis with Ph(Me)SiH\n \n 2\n \n was also mapped. Of these transition structures,\n \n TS2\u2019\n \n _conf\n \n b\n \n \n _\n \n \n R\n \n ,\n \n R\n \n and\n \n TS2\u2019\n \n \n _\n \n conf\n \n d\n \n \n _\n \n \n R\n \n ,\n \n R\n \n converged to the same structure which was found to be lowest in energy. Notably, there is only a small energy difference of 1.9 kcal/mol between the\n \n \u03b2\n \n -H elimination from the benzylic Cu(I) and the metathesis, suggesting of a partially reversible hydrometallation step before the rate-determining metathesis. This can explain the isotope scrambling observed in the mechanistic experiments (Scheme 5a, 5b). According to distortion interaction analysis (DIAS) of these TSs, it can be concluded that for both substrates, the configuration established at the silicon atom are a result of differences of the distortion of the silane moiety.\n
\n\n In summary, we have successfully developed a copper-catalyzed, highly selective hydrosilylation method for alkenes. Various monosubstituted alkenes with aromatic and aliphatic groups efficiently reacted with hydrosilanes to produce enantiomerically enriched alkyl-substituted silanes with a Si-stereogenic center or Si/C two stereogenic centers under substrate influence. Control experiments suggested that metathesis likely plays a crucial role as the rate-determining step. The outstanding regioselectivity and enantioselectivity were further confirmed through DFT calculations. Additionally, the unreacted Si\u2009\u2212\u2009H bond in the chiral silane products provides opportunities for further derivatization.\n
\n\n Schemes 1 to 5 are available in the Supplementary Files section\n
\n\n Scheme 1. Synthesis of Silicon-stereogenic monohydrosilanes.\n
\n \n\n Scheme 2. Scope of the borylation reaction of aryl and aliphatic substituted allylsilanes.\n \n \n a, b, c\n \n \n \n \n a\n \n \n \n \n Conditions: 1 (0.2 mmol, 1.0 equiv), 2 (3.0 equiv), CuOAc (10 mol %), (\n \n R\n \n ,\n \n R\n \n )-Ph-BPE (11 mol %) and CyJohnPhos (11 mol %) were stirred at 40 \u00b0C for 2 d under N\n \n 2\n \n atmosphere. (See Supporting Information for the detailed experimental procedures).\n \n \n b\n \n \n Isolated yields.\n \n \n c\n \n \n The er val-ues were determined by chiral HPLC analysis.\n \n \n d\n \n \n 4 d.\n \n \n e\n \n \n 2c (6.0 equiv) was added.\n \n \n f\n \n \n The dr value was determined by chiral HPLC analysis.\n
\n \n\n Scheme 3. Scope of branched-selective hydrosilylation of alkenes.\n \n \n a, b, c, d\n \n \n \n \n a\n \n \n Conditions: 4 (0.2 mmol, 1.0 equiv), 2 (3.0 equiv), Cu(OAc)\n \n 2\n \n (4.0 mol%) and (\n \n R\n \n ,\n \n R\n \n )-Ph-BPE (8.0 mol%) were stirred at 40 \u00b0C for 36 h under N\n \n 2\n \n atmosphere. (See Supporting Information for the detailed experimental procedures).\n \n \n b\n \n \n Isolated yields.\n \n \n c\n \n \n The er values were determined by chiral HPLC analysis.\n \n \n d\n \n \n The dr values were determined by GC analysis or\n \n 1\n \n H NMR of the crude reaction mixture.\n \n \n e\n \n \n 72 h.\n \n \n f\n \n \n In extra dry cyclohexane (2.0 M).\n
\n \n\n Scheme 4. Gram-scale synthesis and functionalization.\n
\n \n\n Scheme 5. Mechanistic studies and proposed reaction pathway.\n
\n \n\n Wearable ultraviolet (UV) detectors have attracted considerable interest in the military and civilian realms. However, semiconductor-based UV detectors are easily interfered by elongation due to the elastic modulus incompatibility between rigid semiconductors and polymer matrix. Polymer detectors containing UV responsive moieties seriously suffer from slow response time. Herein, a novel UV illuminance-mechanical stress-electric signal conversion has been proposed based on well-defined ionic liquid (IL)-containing liquid crystalline polymer (ILCP) and highly elastic polyurethane (TPU) composite fabrics, to achieve a robust UV monitoring and shielding device with a fast response time of 5 s. Due to the electrostatic interactions and hydrogen bonds between ILs and LC networks, the ILCP-based device can effectively prevent the exudation of ILs and maintain stable performance upon stretching, bending, washing and 1000 testing cycles upon 365 nm UV irradiation. This work provides a generalizable approach toward the development of full polymer-based wearable electronics and soft robots.\n
\n\n \n wearable ultraviolet (UV) detectors\n \n
\n\n \n polymer science\n \n
\n\n \n liquid crystalline composites\n \n
\n\n For the past decade, wearable electronics have been rapidly developed for envisioned applications in a wide range of areas such as healthcare devices,\n \n \n 1\n \n \n environmental monitors,\n \n \n 2\n \n \n power supplies,\n \n \n 3\n \n \n and Internet of Things technologies,\n \n \n 4\n \n \n because of their multiple sensing capabilities, unique flexibility, and portability.\n \n \n 5\n \n \n Among the versatile wearable electronics, ultraviolet (UV) detectors have garnered extensive attention in recent years for their significant applications in the military and civilian realms, such as defense warning systems,\n \n \n 6\n \n \n photochemical synthesis,\n \n \n 7\n \n \n medical treatments,\n \n \n 8\n \n \n security communication,\n \n \n 9\n \n \n etc. However, due to the strong penetrating ability, excess UV exposure can reach deep skin layers, leading to serious health hazards such as skin ageing and skin cancer.\n \n \n 10\n \n \n In this regard, it is highly desired to achieve a wearable all-in-one UV monitoring and shielding device for daily life and industrial production under UV irradiation.\n \n \n 11\n \n \n Previous works have reported wearable UV detectors by geometrically patterning rigid photoelectric semiconductors (such as zinc oxide) on a flexible polymer matrix, featuring superior sensitivity.\n \n \n 12\n \n ,\n \n 13\n \n \n However, the devices are prone to occur delamination under elongation because of the elastic modulus incompatibility between rigid semiconductors and flexible matrix, resulting in a serious decrease in performance. In addition, due to the poor conformability and processing capability of rigid semiconductors, more suitable UV shielding materials are desired for wearable devices.\n
\n\n Based on polymer chemistry, functional polymer materials have realized the integration of skin-inspired properties in electronics, such as stretchability, conformability, adhesion, self-healing and degradability.\n \n \n 14\n \n ,\n \n 15\n \n \n To fabricate polymer-based UV detectors, photoisomerizable moieties (e.g., azobenzene, spiropyran and diarylethene) are generally introduced into polymers.\n \n \n 16\n \n \n Under UV irradiation, the photoisomerizable moieties will induce physical and chemical property changes of polymers, such as color or polarity. To date, reported flexible polymer-based UV detectors mainly exhibit visual color changes without electric signal output, which hinders the integration and application in sensor systems.\n \n \n 17\n \n ,\n \n 18\n \n \n The UV induced physical and chemical property changes can be converted to electric (capacitance or current) signals via incorporating conductive fillers into the polymer matrix, such as carbon nanotubes, silver nanowires and ionic liquids (ILs).\n \n \n 19\n \n ,\n \n 20\n \n \n Among these, ILs are attracting growing attention because of their high ionic conductivity, chemical and thermal stability, nonvolatility, and low flammability.\n \n \n 21\n \n ,\n \n 22\n \n \n For instance, Watanabe and coworkers reported a reversible ion-conducting switch by mixing azobenzene molecules into poly(N-isopropyl acrylamide) (PNIPAm) IL gels.\n \n \n 19\n \n \n Due to the\n \n trans\n \n -to-\n \n cis\n \n isomerization under UV light, the increased polarity of azobenzene triggers the macroscopic sol-gel transition of IL gels, resulting in the increase of ionic conductivity. However, physically crosslinked PNIPAm IL gels suffer from poor mechanical properties, exudation of ILs, and a slow response time of 6 min. Alaniz and coworkers designed poly(ethylene oxide) films covalently grafted diarylethene-containing ILs, using for the light-modulation of ionic conductivity.\n \n \n 20\n \n \n Diarylethene rings reversibly switch between open-close state under alternative UV/visible light, leading to the charge differences and thus the change of ionic conductivity. Nevertheless, the films exhibited a long response time of 15 min. Till now, the reported works mainly concentrated on the polarity modulation of photoisomerizable moieties via illumination to control the ionic conductivity. Generally, it will take some time for photoisomerizable moiety-containing polymers to reach the photostationary state upon illumination.\n \n \n 19\n \n ,\n \n 20\n \n ,\n \n 23\n \n \n The response time is substantially decided by the polymer selection and composite structures. Therefore, an elaborate polymer composite of UV responsive polymers effectively integrated with ILs is highly desired for sensitive UV detectors.\n
\n\n Liquid crystalline elastomers (LCEs), or crosslinked liquid crystalline polymers (CLCPs), are highly sensitive to environmental stimuli owing to the synergistic effect of liquid crystalline (LC) alignment and polymer networks.\n \n \n 24\n \n ,\n \n 25\n \n \n Azobenzene moieties are always functionalized into LC networks to achieve light-driven deformation of CLCPs. Based on the\n \n trans-cis\n \n isomerization of azobenzene mesogens under alternative UV and visible light or heating, the azobenzene-containing (Azo) CLCPs undergo a reversible phase transition between an LC phase and an isotropic state, generating internal stress and leading to reversible shape changes.\n \n \n 26\n \n \n It is reported that based on reasonably designed structures, as long as 1 mol% of azobenzene mesogens reach to the photostationary state upon illumination, the generated energy can lead to the phase transition of the whole systems, featuring a fast response time of several seconds for Azo-CLCPs.\n \n \n 27\n \n ,\n \n 28\n \n \n If integrated with ILs by feasible structural designs, the stress induced by phase transition has the potential to directly facilitate the ion migration of ILs, and rapidly transform to current or resistance signals. Thus, CLCPs can be utilized as wearable electronics, benefiting from the rapid response ability and robust mechanical properties. However, CLCPs are conventionally fabricated by one-pot copolymerization of LC monomers and crosslinkers in LC cells.\n \n \n 24\n \n \n Thus, CLCPs generally exhibit poor processability because of the insoluble and infusible crosslinked networks. To date, CLCPs used as wearable UV detectors have not been reported yet.\n
\n\n We previously reported a solution-processable CLCPs consisting of azobenzene-containing block copolymers (Azo-BCPs) and branched polyethyleneimine (PEI), as shown in Fig.\n \n 1\n \n a.\n \n \n 29\n \n \n CLCPs can be fabricated by a two-step post-crosslinking method, which involves the molding of non-crosslinked LCPs and the subsequent crosslinking in PEI solution. Herein, we demonstrate novel IL-containing liquid crystalline polymers (ILCPs) via introducing hydrophobic ILs into the aforementioned Azo-CLCP matrix for intrinsically flexible and highly sensitive UV detectors. To improve the mechanical property, thermoplastic polyurethane (TPU) is selected to mix with Azo-BCPs to fabricate LCP-TPU fabrics through electrospinning (Fig.\n \n 1\n \n b), benefiting from their solution-processabilities. Then, the CLCP-TPU fabrics are obtained by the post-crosslinking of LCP-TPU fabrics in PEI solution, capable of portability and air permeability.\n \n \n 30\n \n \n The high porosity and large surface area of fabrics also accelerate the penetration of ILs for preparing ILCP fabrics. Benefiting from the cooperative effects of Azo-LC alignment and polymer networks, the ILCP-based device constructs a rapid conversion of UV illuminance-mechanical stress-electric signals, which substantially facilitates the sensitivity of UV detectors (Fig.\n \n 1\n \n c). In addition, the abundant amino groups in PEI and urethane groups in TPU make the device easily adhere to skins by multiple hydrogen bonds (Movie S1). Moreover, the exudation of ILs is effectively suppressed by the strong electrostatic interactions between amino groups and ions. Thus, the ILCP fabrics can be utilized as a flexible, stretchable, and washable UV monitoring and shielding materials. The ILCP-based device demonstrates an illuminance detection range of 10\u2009~\u2009270 mW cm\n \n -\n \n 2\n \n \n , stability of 1000 testing cycles and especially a response time of 5 s under 365 nm UV light. Owing to the electric signal output, we have designed the wearable ILCP-based on-demand information encoding electronics for UV security unlock and input via Internet of Things technologies.\n
\n\n \n Fabrication of ILCP fabrics\n \n
\n\n In the fabrication process, firstly, the LCP-TPU fabrics were prepared by electrospinning mixture solution of Azo-BCPs and TPU onto an aluminum foil collector in large scale (Figure 1d). Secondly, the obtained LCP-TPU fabrics were immersed in an ethanol solution of PEI for 6 h to complete the post-crosslinking reaction to form CLCP-TPU fabrics, and then dried in vacuum at 50\n \n o\n \n C overnight. The average diameter of obtained CLCP-TPU fibers is ~ 1.5 \u03bcm (Figure 1e). Subsequently, the CLCP-TPU fabrics were swollen in pure ILs for 12 h. Herein, the ILs are hydrophobic 1-ethyl-3-methylimidazolium bis-(trifluoromethylsulfonyl)imide ([EMIm][TFSI]), which possess high voltage stability, humidity resistance and strong self-adhesion stability.\n \n 31\n \n After being dried in vacuum at room temperature for 6 h, the free-standing ILCP fabrics were finally obtained with the fibers swollen up to ~ 2.0 \u03bcm in diameter (Figure 1f). The ILCP fabrics exhibit the mechanical properties of 240% elongation and 5.2 MPa tensile strength (Figure S2), profiting from the elasticity of TPU and plasticization of ILs, fulfilling the needs of electronic skins.\n \n 32\n \n
\n\n \n Performance of the ILCP-based device\n \n
\n\n The IL content and ionic conductivity of ILCPs are affected by the immersion time (Figure 2a). After immersion in ILs for 12 hours, the IL content increased to 69% and the conductivity reached to 0.855 mS cm\n \n -1\n \n , meeting the needs of electronics.\n \n 33\n \n In general, flowable ILs within the polymer matrix are liable to exude under heating or vacuum, seriously losing the performance of devices.\n \n 34\n \n Figure 2b compares the changes of IL content of CLCP-TPU fabrics and non-crosslinked LCP-TPU fabrics with time. The samples were treated under vacuum at room temperature for 24 h and subsequently at 50\n \n o\n \n C for another 24 h. It is clear that the IL content of CLCP-TPU fabrics slightly decreased from 76% to 64% in the initial 3 hours and then maintained at the platform. In contrast, for non-crosslinked LCP-TPU fabrics, the IL content was severely falling to 22% without a platform. Therefore, the crosslinked networks and PEI complexation effectively restrain the exudation of ILs. The response to UV illuminance of ILCP-based device is illustrated in Figure 2c, d. The device exhibited an increased current signal in response to UV light (Figure S3). When the weight ratio of Azo-BCP/TPU was 1:3, the fitted relationship between relative current changes and UV power density was calculated by linear regression method, as shown below:\n
\n\n [IMAGE_RESULTS_1]\n
\n\n Where \u0394I/I\n \n 0\n \n is the relative current change and P represents UV power density. The slope of the fitting curve stands for the responsivity of ILCP-based device.\n
\n\n The device demonstrated an illuminance detection range of 10 ~ 270 mW cm\n \n -2\n \n , a response time of 5 s, and a linear correlation coefficient of 0.998. The responsivity of ILCP-based device reduced with the decrease of Azo-BCP content. When the weight ratio of Azo-BCP/TPU increased to 1:1, the linear detection range reduced because of macroscopical bending deformation of ILCPs under strong UV illuminance. Upon 90 mW cm\n \n -2\n \n UV light, the relative current change (\u0394I/I\n \n 0\n \n ) was constantly stable in static testing, undergoing 1000 testing cycles, 30% strain stretching and 90\n \n o\n \n bending, due to the intrinsical flexibility of ILCPs (Figure 2e, f). Especially, the device can sustain 50 dipping cycles in water because of the great humidity resistance of ILs. The performance almost remained unchanged after 30 days at room temperature (Figure S4). Compared with other reported wearable photodetectors, the comprehensive performance (such as flexibility and response time) of the ILCP-based device is excellent (Table S1). Furthermore, the device exhibited a decreased current signal during dynamic stretching and bending, because of the increase in resistance (Figure S5). The negative current shift can be facilely distinguished from the positive ones upon UV response, extending the practical applications of the device.\n
\n\n \n Working mechanism of the ILCP-based device\n \n
\n\n To further investigate the working mechanism of the ILCP-based device, the optical microscopic (OM), polarizing optical microscopic (POM) and mechanical testing were conducted to illustrate the\n \n in-situ\n \n characterizations of ILCP fibers upon UV light (Figure 3a-c and Figure S6, 7). The POM images reflect the orientation of LC alignments through electrostatic fields of electrospinning. Because of contraction of the single fiber upon UV light, the fiber networks deform microscopically and generate contraction stress macroscopically (Movie S2). When turning off UV light, the contraction stress is immediately released due to the excellent elasticity of TPU. To analyze the effect of elasticity, we compared the response and recovery time of TPU-based ILCPs with those of low-elastic polyacrylonitrile (PAN)-based ILCPs (Figure S8). The response and recovery time of TPU-based ILCPs are 5 s and 4 s, respectively, much faster than those of PAN-based ILCPs (11 s and 22 s, respectively). It is reasonable that the elastic TPU matrix can effectively transfer the UV induced contraction stress to ILs and then accelerate ions migrating, converting to electric signals rapidly. On the other hand, it is reported that the mechanical stretch tends to align the LC mesogens along the stretching direction, referred as mechano-alignment.\n \n 35\n \n Here, the recovery stress of TPU can dominate the mechano-alignment of CLCPs, endowing a fast cyclic operation of the device without visible light or heating. The combination of CLCPs, ILs and TPU constructs an effective conversion mechanism of UV illuminance-mechanical stress-electric signals, bringing in rapid response time for ILCP-based devices. Additionally, the increased polarity of\n \n cis\n \n -azobenzene may contribute to the light-modulated ionic conductivity alternations.\n \n 19\n \n
\n\n \n UV shielding ability of composite fabrics\n \n
\n\n The LCP-TPU, CLCP-TPU and ILCP fabrics have a similar UV shielding ability, because they have almost the same UV transmittance spectrum. To simplify the fabrication process, the as-electrospun LCP-TPU fabrics can be directly utilized as recyclable UV shielding materials (Figure 3d). A handheld electrospinning device is used to prepare conformal UV shielding fabrics facilely (Figure S9).\n \n 36\n \n Moreover, the LCP-TPU fabrics can redissolve in DMF/THF solutions for the recycling fabrication. To evaluate the UV shielding property, the ultraviolet protection factor (UPF) was calculated based on the GB/T 18830-2009 evaluation method, detailed in Section S3 (Supplementary Information). When UPF > 50 and UV-A light transmittance < 5%, the sample can be identified as UV protective textiles. The calculated UPF of LCP-TPU fabrics (20 \u03bcm thick) was initially 25040. Considering of the practical working conditions, the UPF decreased to 13851 at 100% strain and was almost unchanged upon washing. The UV shielding property and stability of LCP-TPU fabrics is much better than those of commercial products of UV protection clothing, nitrile gloves and previously reported UV shielding materials (Figure 3e, f).\n \n 11, 37\n \n However, the calculation parameters of UPF are based on solar UV irradiation (up to ~ 3 mW cm\n \n -2\n \n ) in daily life. Under some conditions such as lithography and photocuring, the UV illumination can reach to or higher than 20 mW cm\n \n -2\n \n . Thus, spiropyran was selected as a sensitive photochromic indicator to evaluate the UV shielding property of LCP-TPU fabrics via spectrophotometry (Figure 3g and Figure S10). Upon 20 mW cm\n \n -2\n \n UV light, the LCP-TPU fabrics exhibited the long-time shielding ability of 120 min, while the nitrile gloves lost efficacy within 1 min. Furthermore, the LCP-TPU fabrics can be reused after irradiation with 520 nm visible light (200 mW cm\n \n -2\n \n ) for 10 min. For practical applications, the LCP-TPU fabrics can be applied to prevent the UV ageing of nitrile gloves or skins (Figure S11).\n
\n\n \n Applications of UV security encoding\n \n
\n\n As discussed above, the electric signal output of the ILCP-based device allows it for the facile integration into the Internet of Things system. The ILCP-based device can not only record the real-time UV illuminance on liquid crystal display (LCD) screen, but also transfer data to smartphones through Bluetooth for remote monitoring (Figure S12, 13 and Movie S3). Furthermore, UV light can avoid being intercepted and interfered because of its short wavelength, which can be used for security communication. Under various extreme environments in the military field, it is highly desired for a portable and robust UV detector to guarantee communication quality. Thus, we have designed wearable information encoding and encryption applications based on ILCP fabrics, such as UV unlock and UV input. As illustrated in Figure 4a, the UV coding system is consisted of an ILCP-based encoding part and an Arduino singlechip-based decoding part.\n
\n\n The physical connection diagram and encoding logic of UV security unlock are displayed in Figure 4b, c. If the ILCP array composed of three ILCP-based devices is irradiated with the UV illuminance of \u201c20, 30, 40\u201d mW cm\n \n -2\n \n in right sequence, the green LED will be turned on, indicating a unlock command. Otherwise, the red LED will be turned on, indicating a lock command (Figure 4f and Movie S4). We also conducted the application of UV security input (Figure 4d, e). When the alphabets of A~Z are critically corresponded to the determined UV illuminance, the UV illuminance information can be recognized and read out by a smartphone, resulting in the input of \u201cBUAA\u201d on the screen (Figure 4g and Movie S5). Moreover, the information can be reprogrammed and transformed by simply adjusting the code. The present work opens up a new paradigm for UV security information storage and communication based on wearable and robust UV responsive moiety-containing polymer detectors.\n
\n\n In summary, we demonstrate a robust ILCP-based UV monitoring and shielding device via the combination of electrospun CLCP-TPU fabrics and hydrophobic ILs. The novel signal conversion, that is UV illuminance-mechanical stress-electric signals, significantly improves the sensitivity of flexible UV detectors, benefiting from the cooperative effects of Azo-LC alignment and polymer networks. The excellent elasticity of TPU facilitates the fast cyclic operation of the device without visible light or heating. The device exhibits a wide UV illuminance detection range of 10\u2009~\u2009270 mW cm\n \n -\n \n 2\n \n \n , a response time of 5 s and a recovery time of 4 s. In addition, the abundant amino groups in PEI and urethane groups in TPU make the device easily adhere to skins by multiple hydrogen bonds. The strong electrostatic interactions between amino groups and ions effectively restrain the exudation of ILs. Thus, the device maintains stable upon 1000 UV on/off testing cycles, 30% strain stretching, 90\n \n o\n \n bending, and especially 50 dipping cycles in water. Furthermore, the as-electrospun recyclable LCP-TPU fabrics have the long-time shielding ability of 120 min upon 20 mW cm\n \n -\n \n 2\n \n \n UV light, with a high UPF of 25040. The LCP fabrics can be reused after irradiation with 520 nm visible light (200 mW cm\n \n -\n \n 2\n \n \n ) for 10 min, applying to prevent the UV ageing of skins or other materials. For practical applications in various extreme environments, demonstrations of information encoding and encryption based on wearable ILCPs are successfully realized for UV unlock and UV input via the Internet of Things technologies, benefiting from the portability and robustness of ILCPs. The performance of ILCPs is expected to be further enhanced through the improvement of the device structure and testing circuit. The well-defined composites of ILCPs and novel signal conversion mechanism provide brand-new opportunities for functional polymer-based sensors to fabricate various sensitive wearable electronics.\n
\n\n \n Materials\n \n
\n\n Tetrahydrofuran (THF, 99.5%, Aladdin), N, N-dimethylformamide (DMF, 99.9%, Aladdin), Ethanol (EtOH, 99.8%, Aladdin), Branched polyethyleneimine (PEI,\n \n M\n \n \n \n w\n \n \n =1\u00d710\n \n 4\n \n g/mol, 99%, Aladdin), Thermoplastic polyurethane (TPU, 1190A, Elastollan), and 1-ethyl-3-methylimidazolium bis-(trifluoromethylsulfonyl)imide ([EMIm][TFSI], 99%, Adamas) were used as received. The block copolymers (PEO\n \n 45\n \n -\n \n b\n \n -PMA(Az)\n \n 28\n \n ,\n \n M\n \n \n \n n\n \n \n =1.82\u00d710\n \n 4\n \n g/mol,\n \n M\n \n \n \n w\n \n \n \n /M\n \n \n \n n\n \n \n =1.13), were synthesized by atom transfer radical polymerization (ATRP), detailed in Section S1 and Figure S1 (Supplementary Information).\n \n \n 29\n \n ,\n \n 38\n \n ,\n \n 39\n \n \n
\n\n \n Fabrication of the ILCP-based device\n \n
\n\n Firstly, Azo-BCPs (80 mg) and TPU (240 mg) were dissolved in DMF and THF mixture (1 mL, volume ratio 1:1) through stirring for 12 h at room temperature. A self-assembly electrospinning device was used to produce fibers with a positive DC voltage of 15 kV and a feed rate of 1 mL/h. The distance between the extrusion nozzle and the roller was set at 15 cm. The LCP-TPU fabrics of 20 \u00b5m thick were collected on aluminum foil wrapped around a rotating roller. The LCP-TPU fabrics were immersed in an ethanol solution of PEI (2 mg/mL) for 6 h and then dried in vacuum overnight. Subsequently, the obtained CLCP-TPU fabrics were swollen in pure ILs for 12 h. The fabrics were sandwiched in filter papers, clamped by two glass sheets, and then dried in vacuum (0.02 Mpa) at room temperature for 6 h. The free-standing ILCP fabrics were finally obtained. For the UV response testing of ILCP fabrics (5 \u00d7 15 mm\n \n \n 2\n \n \n ), copper wires were attached at the opposite sides of the ILCP fabrics by silver paste.\n
\n\n \n Characterization and Measurements\n \n
\n\n Scanning electron microscope (SEM) images were obtained on a TESCAN VEGA3 microscope. UV light at 365 nm was obtained from Omron (ZUV-C30H) LED irradiator. Visible light at 520 nm was obtained from CCS (HLV-22GR-3W) LED irradiator. The IL content is defined as the ratio of (\n \n w\n \n -\n \n w\n \n \n \n 0\n \n \n )/\n \n w\n \n , where\n \n w\n \n \n \n 0\n \n \n and\n \n w\n \n denote the measured weights of CLCPs before and after immersion in ILs, respectively. The weights of CLCPs were measured by an analytical balance (METTLER TOLEDO ME104E). The current response was monitored by an electrometer (Keithley 6517B) with an applied voltage of 0.1 V. The tensile properties were measured in a stretch mode at a strain rate of 3 mm/min at room temperature on Xieqiang CTM2050 mechanical analyser. Optical microscopic (OM) and polarizing optical microscopic (POM) experiments were conducted by Shang Guang 59XF microscope. The recyclable UV shielding fabrics were prepared by a handheld electrospinning device (Junada MPEG-1). The UV shielding properties were evaluated by UV-vis transmittance spectra (Shimadzu UV-2600 spectrophotometer). The commercial UV protection clothing was purchased from Decathlon (90 \u00b5m). The Android application of UV encoding was made by MIT App Inventor 2. All the photos and movies were recorded on a SONY HDR-CX450 digital video camera.\n
\n\n 1\u00a0\u00a0\u00a0 Lee, G. H.\n \n et al.\n \n Multifunctional materials for implantable and wearable photonic healthcare devices.\n \n Nat. Rev. Mater.\n \n \n 5\n \n , 149 (2020).\n
\n\n 2\u00a0\u00a0 Hua, Q.\n \n et al.\n \n Skin-inspired highly stretchable and conformable matrix networks for multifunctional sensing.\n \n Nat. Commun.\n \n \n 9\n \n , 244 (2018).\n
\n\n 3\u00a0\u00a0\u00a0 Zamarayeva, A. M.\n \n et al.\n \n Flexible and stretchable power sources for wearable electronics.\n \n Sci. Adv.\n \n \n 3\n \n , e1602051 (2017).\n
\n\n 4\u00a0\u00a0\u00a0 Sarycheva, A., Polemi, A., Liu, Y., Dandekar, K., Anasori, B. & Gogotsi, Y. 2D titanium carbide (MXene) for wireless communication.\n \n Sci. Adv.\n \n \n 4\n \n , eaau0920 (2018).\n
\n\n 5\u00a0\u00a0\u00a0 Lim, Y. W., Jin, J. & Bae, B. S. Optically Transparent Multiscale Composite Films for Flexible and Wearable Electronics.\n \n Adv. Mater.\n \n , 1907143 (2020).\n
\n\n 6\u00a0\u00a0\u00a0 Chen, H., Liu, K., Hu, L., Al-Ghamdi, A. A. & Fang, X. New concept ultraviolet photodetectors.\n \n Mater. Today\n \n \n 18\n \n , 493 (2015).\n
\n\n 7\u00a0\u00a0\u00a0 Kotikian, A., Truby, R. L., Boley, J. W., White, T. J. & Lewis, J. A. 3D Printing of Liquid Crystal Elastomeric Actuators with Spatially Programed Nematic Order.\n \n Adv. Mater.\n \n \n 30\n \n , 1706164 (2018).\n
\n\n 8\u00a0\u00a0 Premi, S.\n \n et al.\n \n Chemiexcitation of melanin derivatives induces DNA photoproducts long after UV exposure.\n \n Science\n \n \n 347\n \n , 842 (2015).\n
\n\n 9\u00a0\u00a0\u00a0 Deng, W.\n \n et al.\n \n All\u2010Sprayed\u2010Processable, Large\u2010Area, and Flexible Perovskite/MXene\u2010Based Photodetector Arrays for Photocommunication.\n \n Adv. Opt. Mater.\n \n \n 7\n \n , 1801521 (2019).\n
\n\n 10\u00a0 Saraiya, M.\n \n et al.\n \n Interventions to prevent skin cancer by reducing exposure to ultraviolet radiation: a systematic review.\n \n Am J Prev Med\n \n \n 27\n \n , 422 (2004).\n
\n\n 11\u00a0 Liang, F.\n \n et al.\n \n Layer-by-Layer Assembly of Nanofiber/Nanoparticle Artificial Skin for Strain-Insensitive UV Shielding and Visualized UV Detection.\n \n Adv. Mater. Technol.\n \n \n 5\n \n , 1900976 (2020).\n
\n\n 12\u00a0 Liu, X., Gu, L., Zhang, Q., Wu, J., Long, Y. & Fan, Z. All-printable band-edge modulated ZnO nanowire photodetectors with ultra-high detectivity.\n \n Nat. Commun.\n \n \n 5\n \n , 4007 (2014).\n
\n\n 13\u00a0 Cai, S., Xu, X., Yang, W., Chen, J. & Fang, X. Materials and Designs for Wearable Photodetectors.\n \n Adv. Mater.\n \n \n 31\n \n , 1808138 (2019).\n
\n\n 14\u00a0 Yang, J. C., Mun, J., Kwon, S. Y., Park, S., Bao, Z. & Park, S. Electronic Skin: Recent Progress and Future Prospects for Skin-Attachable Devices for Health Monitoring, Robotics, and Prosthetics.\n \n Adv. Mater.\n \n \n 31\n \n , 1904765 (2019).\n
\n\n 15\u00a0 Tran, H., Feig, V. R., Liu, K., Zheng, Y. & Bao, Z. Polymer Chemistries Underpinning Materials for Skin-Inspired Electronics.\n \n Macromolecules\n \n \n 52\n \n , 3965 (2019).\n
\n\n 16\u00a0 Bisoyi, H. K. & Li, Q. Light-Driven Liquid Crystalline Materials: From Photo-Induced Phase Transitions and Property Modulations to Applications.\n \n Chem. Rev.\n \n \n 116\n \n , 15089-15166 (2016).\n
\n\n 17\u00a0 Zou, W., Sastry, M., Gooding, J. J., Ramanathan, R. & Bansal, V. Recent Advances and a Roadmap to Wearable UV Sensor Technologies.\n \n Adv. Mater. Technol.\n \n , 1901036 (2020).\n
\n\n 18\u00a0 Wei, S.\n \n et al.\n \n Multicolor Fluorescent Polymeric Hydrogels.\n \n Angew. Chem. Int. Ed.\n \n ,\n \n 59,\n \n 2 (2020).\n
\n\n 19\u00a0 Wang, C.\n \n et al.\n \n Reversible Ion-Conducting Switch by Azobenzene Molecule with Light-Controlled Sol-Gel Transitions of the PNIPAm Ion Gel.\n \n ACS Appl. Mater. Interfaces\n \n \n 12\n \n , 42202 (2020).\n
\n\n 20\u00a0 Nie, H.\n \n et al.\n \n Light-Controllable Ionic Conductivity in a Polymeric Ionic Liquid.\n \n Angew. Chem. Int. Ed.\n \n \n 59\n \n , 5123 (2020).\n
\n\n 21\u00a0 Zhang, S., Zhang, Q., Zhang, Y., Chen, Z., Watanabe, M. & Deng, Y. Beyond solvents and electrolytes: Ionic liquids-based advanced functional materials.\n \n Prog. Mater. Sci.\n \n \n 77\n \n , 80 (2016).\n
\n\n 22\u00a0 Cui, J., Li, Y., Chen, D., Zhan, T. G. & Zhang, K. D. Ionic Liquid\u2010Based Stimuli\u2010responsive Functional Materials.\n \n Adv. Funct. Mater.\n \n , 2005522 (2020).\n
\n\n 23\u00a0 Li, P., Tan, S., Wu, Y., Wang, C. & Watanabe, M. Azobenzene-Based Ionic Liquid Switches Phase Separation of Poly(N-isopropylacrylamide) Aqueous Solutions as a Molecular Trigger, Leading to UV Shutdown of Ionic Transport.\n \n ACS Macro Lett.\n \n \n 9\n \n , 825-829 (2020).\n
\n\n 24\u00a0 Yu, Y., Nakano, M. & Ikeda, T. Photomechanics: directed bending of a polymer film by light.\n \n Nature\n \n \n 425\n \n , 145 (2003).\n
\n\n 25\u00a0 White, T. J. & Broer, D. J. Programmable and adaptive mechanics with liquid crystal polymer networks and elastomers.\n \n Nat. Mater.\n \n \n 14\n \n , 1087 (2015).\n
\n\n 26\u00a0 Ikeda, T., Mamiya, J. & Yu, Y. Photomechanics of liquid-crystalline elastomers and other polymers.\n \n Angew. Chem. Int. Ed.\n \n \n 46\n \n , 506 (2007).\n
\n\n 27\u00a0 Cheng, Y., Lu, H., Lee, X., Zeng, H. & Priimagi, A. Kirigami-Based Light-Induced Shape-Morphing and Locomotion.\n \n Adv. Mater.\n \n \n 32\n \n , 1906233 (2020).\n
\n\n 28\u00a0 Pang, X., Lv, J. A., Zhu, C., Qin, L. & Yu, Y. Photodeformable Azobenzene-Containing Liquid Crystal Polymers and Soft Actuators.\n \n Adv. Mater.\n \n \n 31\n \n , 1904224 (2019).\n
\n\n 29\u00a0 Zheng, X.\n \n et al.\n \n A Cut-and-Weld Process to 3D Architectures from Multiresponsive Crosslinked Liquid Crystalline Polymers.\n \n Small\n \n \n 15\n \n , 1900110 (2019).\n
\n\n 30\u00a0 Dong, K., Peng, X. & Wang, Z. L. Fiber/Fabric-Based Piezoelectric and Triboelectric Nanogenerators for Flexible/Stretchable and Wearable Electronics and Artificial Intelligence.\n \n Adv. Mater.\n \n \n 32\n \n , 1902549 (2020).\n
\n\n 31\u00a0 Cao, Z., Liu, H. & Jiang, L. Transparent, mechanically robust, and ultrastable ionogels enabled by hydrogen bonding between elastomers and ionic liquids.\n \n Mater. Horizons\n \n \n 7\n \n , 912 (2020).\n
\n\n 32\u00a0 Chortos, A., Liu, J. & Bao, Z. Pursuing prosthetic electronic skin.\n \n Nat. Mater.\n \n \n 15\n \n , 937-950 (2016).\n
\n\n 33\u00a0 Kim, Y. M. & Moon, H. C. Ionoskins: Nonvolatile, Highly Transparent, Ultrastretchable Ionic Sensory Platforms for Wearable Electronics.\n \n Adv. Funct. Mater.\n \n \n 30\n \n , 1907290 (2020).\n
\n\n 34\u00a0 Wang, Z., Si, Y., Zhao, C., Yu, D., Wang, W. & Sun, G. Flexible and Washable Poly(Ionic Liquid) Nanofibrous Membrane with Moisture Proof Pressure Sensing for Real-Life Wearable Electronics.\n \n ACS Appl. Mater. Interfaces\n \n \n 11\n \n , 27200 (2019).\n
\n\n 35\u00a0 Bai, R. & Bhattacharya, K. Photomechanical coupling in photoactive nematic elastomers.\n \n J Mech Phys Solids\n \n \n 144\n \n , 104115 (2020).\n
\n\n 36\u00a0 Haik, J., Kornhaber, R., Blal, B. & Harats, M. The Feasibility of a Handheld Electrospinning Device for the Application of Nanofibrous Wound Dressings.\n \n Adv. Wound Care\n \n \n 6\n \n , 166 (2017).\n
\n\n 37\u00a0 Liu, A.\n \n et al.\n \n The synergetic modification of surface micro-dissolution and cationization for fabricating cotton fabrics with high UV resistance and conductivity by enriched GO coating.\n \n Cellulose\n \n \n 27\n \n , 10489 (2020).\n
\n\n 38\u00a0 Zheng, X., Zhao, Y. & Chen, A. Self-assembly of liquid-crystalline block copolymers in thin films: control of microdomain orientation.\n \n Polym J\n \n \n 50\n \n , 671 (2018).\n
\n\n 39\u00a0 Zheng, X.\n \n et al.\n \n Polydimethylsiloxane-assisted alignment transition from perpendicular to parallel of cylindrical microdomains in block copolymer films.\n \n RSC Adv.\n \n \n 6\n \n , 93298 (2016).\n
\n\n Supplementary MOVIE S1\n
\n \n\n Supplementary MOVIE S2\n
\n \n\n Supplementary MOVIE S3\n
\n \n\n Supplementary MOVIE S4\n
\n \n\n Supplementary MOVIE S5\n
\n \n\n Thiamine and pyridoxine are essential B vitamins that serve as enzymatic cofactors in energy metabolism, protein and nucleic acid biosynthesis, and neurotransmitter production. In humans, thiamine transporters SLC19A2 and SLC19A3 primarily regulate cellular uptake of both vitamins. Genetic mutations in these transporters, which cause thiamine and pyridoxine deficiency, have been implicated in severe neurometabolic diseases. Additionally, various prescribed medicines, including metformin and fedratinib, manipulate thiamine transporters, complicating the therapeutic effect. Despite their physiological and pharmacological significance, the molecular underpinnings of substrate and drug recognition remain unknown. Here we present ten cryo-EM structures of human thiamine transporters SLC19A3 and SLC19A2 in outward- and inward-facing conformations, complexed with thiamine, pyridoxine, metformin, fedratinib, and amprolium. These structural insights, combined with functional characterizations, illuminate the translocation mechanism of diverse chemical entities, and enhance our understanding drug-nutrient interactions mediated by thiamine transporters.\n
\n\n B vitamins, including thiamine (vitamin B1) and pyridoxine (vitamin B6), are a group of water-soluble, chemically varied compounds that perform important roles in bodily functions including normal growth and development\n \n \n 1\n \n \n . Dietary intake of these vitamins is indispensable, as they cannot be synthesized\n \n de novo\n \n in humans and other mammals\n \n \n 2\n \n \n . Thiamine is absorbed in the small intestine and rapidly converted to its active form, thiamine pyrophosphate (TPP), which constitutes the primary thiamine store and acts as a key coenzyme in the release of energy from carbohydrates, RNA and DNA synthesis, and nerve activity\n \n \n 3\n \n \n . Likewise, the metabolically active form of pyridoxine after cellular absorption, pyridoxal 5\u2019-phosphate (PLP), acts as an essential cofactor in numerous enzymatic reactions, primarily in amino acid metabolism including the biosynthesis of neurotransmitters\n \n \n 4\n \n \n .\n
\n\n Two solute carriers SLC19A2 and SLC19A3, identified as high-affinity thiamine transporters, have been demonstrated largely responsible for moving cationic thiamine and pyridoxine across the plasma membrane\n \n \n 5\n \n \u2013\n \n 7\n \n \n . SLC19A2 is widely distributed in human tissues but is highly enriched in the skeletal muscle, while SLC19A3 is most abundant in placenta followed by liver, kidney and heart\n \n \n 8\n \n \n . Genetic mutations of SLC19A2 cause a thiamine-responsive megaloblastic anemia syndrome (TRMA), an autosomal recessive disorder featuring diabetes mellitus, megaloblastic anemia and sensorineural deafness\n \n \n 9\n \n \n , which has been phenocopied by targeted disruption of the equivalent SLC19A2 gene in mice\n \n \n 10\n \n \n . Mutations in SLC19A3 are associated with Wernicke\u2019s-like encephalopathy\n \n \n 11\n \n \n and biotin- and thiamine-responsive basal ganglia disease (BTBGD)\n \n \n 12\n \n \u2013\n \n 14\n \n \n , which may reflect the critical role of SLC19A3 in maintaining thiamine levels in the blood and brain\n \n \n 15\n \n ,\n \n 16\n \n \n . Notably, TRMA patients often do not exhibit other neurological or cardiac symptoms of thiamine deficiency that are seen in SLC19A3-related diseases\n \n \n 17\n \n \n . Despite the potentially fatal consequences, some symptoms can be alleviated by receiving high dosages of thiamine supplements\n \n \n 2\n \n \n , potentially via alternate absorption routes, such as the low-affinity, high-capacity nonspecific organic cation transporter 1 (OCT1)\n \n \n 18\n \n \n . Aside from thiamine and pyridoxine import, thiamine transporters are also influenced by several cationic medicines, including the antidiabetic metformin, the antidepressant amitriptyline, the antineoplastic fedratinib, and the antibiotics amprolium\n \n \n 19\n \n \u2013\n \n 22\n \n \n . Caution should be exercised when using these drugs, since transporter-mediated drug-nutrient interactions would predispose the patients to thiamine and pyridoxine deficiencies\n \n \n 20\n \n ,\n \n 23\n \n ,\n \n 24\n \n \n .\n
\n\n SLC19A2 and SLC19A3, together with the homologous folate transporter SLC19A1, constitute the vitamin transporting SLC19 subfamily, which belongs to the Major Facilitator Superfamily (MFS)\n \n \n 25\n \n ,\n \n 26\n \n \n . Despite extensive functional characterization of the transport activity, drug-nutrient interactions, and genetic mutation mapping, the precise molecular basis of substrate transport and drug/inhibitor recognition by SLC19A2 and SLC19A3 has yet to be fully explored. Recent structural advancements in SLC19A1 have provided great insight into folate transportation\n \n \n 27\n \n \u2013\n \n 29\n \n \n . However, unlike SLC19A1 that distributes the anionic folate, SLC19A2 and SLC19A3 shuttle cationic thiamine and pyridoxine under physiological pH conditions. Thus, an understanding of SLC19A1 transport mechanism may not be directly applicable to SLC19A2 and SLC19A3.\n
\n\n In this study, we determined ten cryo-EM structures of SLC19A3 and SLC19A2 with a variety of substrates and drugs, in the outward- and inward-facing conformations. Complemented by biochemical and cellular analysis, these conformational snapshots revealed shared features as well as unique elements of both transporters for vitamin transport and drug recognition.\n
\n\n Thiamine transporters (~\u200955 kDa) lack discernable extramembrane domains apart from the 12 transmembrane helices (TMs). To facilitate the cryo-EM analysis, we immunized mice with a shorter version of human SLC19A3 construct (residues 6-472) that lacks the disordered but highly immunogenic N- and C-termini (SLC19A3\n \n cryo\n \n , Extended Data Fig.\n \n 1\n \n a). We isolated a high-affinity fragment antigen-binding region (Fab) against SLC19A3, and successfully determined the cryo-EM structures of SLC19A3-Fab in the apo state and in complex with thiamine, pyridoxine, fedratinib, amprolium, and metformin (Fig.\n \n 1\n \n and Extended Data Figs.\n \n 1\n \n \u2013\n \n 5\n \n ). The truncated construct (SLC19A3\n \n cryo\n \n ) showed the radiolabeled thiamine (\n \n \n 3\n \n \n H-thiamine) uptake activity similar to wild-type in stably transfected HEK293T cells (Extended Data Fig.\n \n 1\n \n d), therefore it was still referred to as SLC19A3 hereafter. Notably, all of these SLC19A3 structures were captured in the outward-facing state by the intracellular side Fab binder, implying a conformation-specific antibody generated by the antigen vaccination strategy.\n
\n\n To aid in the structural identification of SLC19A3, we also employed a different strategy by adding a helical MPER peptide prior to the amino-end of TM1 helix (SLC19A3\n \n MPER\n \n ) and assembling a stable complex with its high-affinity antibody (Fab_10E8v4, Extended Data Fig.\n \n 1\n \n b)\n \n \n 30\n \n \n . SLC19A3\n \n MPER\n \n retains robust\n \n \n 3\n \n \n H-thiamine uptake in HEK293T cells, with activity levels approximately half those of wild-type SLC19A3, possibly due to perturbed surface localization (Extended Data Fig.\n \n 1\n \n d); therefore, this MPER-fusion construct was also denoted as SLC19A3 for simplicity. Interestingly, such an approach enabled the capture of SLC19A3 in a distinct inward-facing conformation, either in the presence of thiamine or the antineoplastic drug fedratinib (Fig.\n \n 1\n \n and Extended Data Figs.\n \n 2\n \n \u2013\n \n 5\n \n ).\n
\n\n We also used the same MPER-fusion approach for human SLC19A2. In accordance, the N-terminal 30 residues of SLC19A2 were removed to design the contiguous helix formation of MPER segment with TM1 (SLC19A2\n \n MPER\n \n , Extended Data Fig.\n \n 1\n \n c). The SLC19A2\n \n MPER\n \n protein accumulated moderately less\n \n \n 3\n \n \n H-thiamine compared to wild-type in HEK293 cells (Extended Data Fig.\n \n 1\n \n f), likely due to the decreased surface expression, as shown by Said and colleagues that the N-terminal sequence (residues 19\u201329) is important for cell surface localization of SLC19A2\n \n 31\n \n . For simplicity, the fusion construct is still referred to as SLC19A2. By using the same MPER-Fab binder, we obtained the inward-facing conformation of SLC19A2 in complex with either thiamine or pyridoxine (Fig.\n \n 1\n \n and Extended Data Figs.\n \n 3\n \n \u2013\n \n 5\n \n ). Notably, the relative orientation of this MPER/Fab differs in the two closely related transporters (Extended Data Fig.\u00a06a), as the MPER segment failed to form a seamless helix with SLC19A2 TM1, probably because of the variation in junction residues (Phe30/Leu31 in hSLC19A2 vs Ile13/Tyr14 in hSLC19A3, Extended Data Fig.\u00a07).\n
\n\n As expected, apo SLC19A3 adopts the canonical MFS fold, with the translocation passage formed between two pseudo-symmetrically related domains: N-domain TMs 1\u20136, and C-domain TMs 7\u201312. Two helical bundles are connected by a long intracellular linker (Lys194 to Lys276) between TM6 and TM7 (Figs.\n \n 2\n \n a-b). A well-resolved density for an amphipathic helical stretch (Phe262-Cys272) in the outward-facing map is embedded parallelly in the membrane. Compared to the 3.15-\u00c5 resolution apo SLC19A3 map, both the 3.0-\u00c5 outward-facing and the 3.36-\u00c5 inward-facing maps with thiamine supplemented exhibit an additional density that fits well for a thiamine molecule in the translocation funnel (Figs.\n \n 2\n \n c-f). The electropositive thiamine sits snugly in the overall electronegative cavity, which is positioned close to the extracellular side of SLC19A3 TMD (Figs.\n \n 2\n \n c and\n \n 2\n \n e). Such a superficial location of substrate binding pocket is reminiscent of the homologous SLC19A1 bound by folate\n \n \n 27\n \n \u2013\n \n 29\n \n \n .\n
\n\n Comparison of the thiamine-bound outward- and inward-facing SLC19A3 structures reveals that the transporter adopts a similar rocker-switch movement as seen in other MFS members. Notably, SLC19A3 pivots at one-third of the funnel axis, close to the extracellular side, whereas other MFS transporters typically rock around the central site\n \n \n 25\n \n ,\n \n 26\n \n \n . A closer look at the thiamine binding pocket reveals both similarities and differences between the outward- and inward-facing states. In the outward-facing conformation, thiamine is mainly embraced by residues from the N-domain (Fig.\n \n 2\n \n d). Specifically, the aminopyrimidine ring of thiamine wedges deeply into the N-domain helical bundle along the horizontal membrane plane, and stacks against Tyr113 on TM4 and, to a lesser extent, Trp59 on TM2 via \u03c0-\u03c0 interactions. The primary amine and the adjacent ring-nitrogen fully engage with Glu110 on TM4 through hydrogen-bonding. The methyl group on the aminopyrimidine moiety points to a hydrophobic cage lined by Val109 on TM4, and Thr93 and Leu97 on TM3. Linked to the aminopyrimidine by a methylene bridge, the thiazolium ring on the other side of thiamine is bent nearly perpendicular to the aminopyrimidine ring, and faces the ample translocation funnel that establishes \u03c0 stacking against Phe56 on TM2. Glu32 on the substantially unwound segment of TM1 is in close vicinity of the second ring-nitrogen of aminopyrimidine (3.4 \u00c5) and the positively charged thiazolium nitrogen (5.5 \u00c5), which may provide additional electrostatic attraction and selectivity for cationic thiamine. The hydroxyethyl tail of thiamine is approaching the backbone carbonyl oxygen of Asn297 on TM7, the only contact with the C-domain bundle in the outward-facing conformation.\n
\n\n Along with the conformational transition of SLC19A3 from outward-facing to inward-facing state, thiamine exhibits a substantial rearrangement. In the inward-facing SLC19A3 structure, the thiamine molecule adopts a more extended conformation, compared to the bent posture in the outward-facing state (Fig.\n \n 2\n \n e). While the aminopyrimidine moiety of thiamine remains accommodated by the similar set of residues on N-domain, the thiazolium ring swings away from Phe56 toward the interior of translocation funnel. This substantial movement establishes the primary amine on aminopyrimidine ring bonding with Asn297, reorients the thiazolium ring sandwiched between TM1 and TM7, moves the thiazolium nitrogen closer to Glu32 (4.8 \u00c5), and approaches the hydroxyethyl tail to Glu320 on TM8. Moreover, additional interactions between thiamine and Tyr151, Leu296, and Gln300 are also established (Fig.\n \n 2\n \n f). Thus, the thiamine is fully coordinated by both N-domain and C-domain when SLC19A3 transits from outward- to inward-facing state. The interaction network is further validated by our mutagenesis analysis on the cellular uptake of\n \n \n 3\n \n \n H-thiamine (Fig.\n \n 2\n \n g and Extended Data Fig.\n \n 1\n \n e).\n
\n\n Human SLC19A2 is the first identified high-affinity thiamine transporter\n \n \n 5\n \n \n , which shares\u2009~\u200948% sequence identity with its close homolog SLC19A3 (Extended Data Fig.\u00a07). Both transporters can transport thiamine efficiently, while SLC19A2 has a slightly larger\n \n K\n \n m and higher import\n \n V\n \n max values than SLC19A3, and their transport profiles can be altered differently by pH conditions, suggesting different mechanisms underlying SLC19A2 and SLC19A3-mediated thiamine absorption\n \n \n 5\n \n \u2013\n \n 7\n \n \n . To address this issue, we first measured the thiamine binding affinity with purified SLC19A2 or SLC19A3 in different pH buffers via a microscale thermophoresis (MST) assay (Fig.\n \n 3\n \n ). At pH 7.5, thiamine exhibits a comparable affinity with both SLC19A2 (\n \n K\n \n d\u2009~\u200985.9 \u00b5M) and SLC19A3 (\n \n K\n \n d\u2009~\u200966.4 \u00b5M, Fig.\n \n 3\n \n a), consistent with the reported\n \n K\n \n m difference\n \n \n 7\n \n \n . Surprisingly, thiamine binds more strongly to SLC19A2 (\n \n K\n \n d\u2009~\u20091.2 \u00b5M), and even tighter to SLC19A3 (\n \n K\n \n d\u2009~\u20090.05 \u00b5M) at pH 6.0 (Fig.\n \n 3\n \n b).\n
\n\n To gain a deeper understanding of the different behavior, we prepared thiamine-bound SLC19A2 sample under the same condition as thiamine-bound SLC19A3\n \n MPER\n \n , and determined a 3.28-\u00c5 inward-facing structure at pH 6.0 (Extended Data Fig.\n \n 5\n \n ). As expected, the overall structure of SLC19A2 is similar to that of SLC19A3, with the main chain C\u03b1 root mean standard deviation (RMSD) of 0.8 \u00c5 (Extended Data Fig.\u00a06b). Consistently, thiamine occupies the cavity of similar interaction elements on SLC19A2 as described above for SLC19A3 (Figs.\n \n 3\n \n e and\n \n 3\n \n f), which is consistent with alterations in cellular uptake capacity of\n \n \n 3\n \n \n H-thiamine upon alanine substitution of pocket residues (Fig.\n \n 3\n \n k). However, closer inspection into the substrate pocket did reveal some unique features. First, residues Tyr74, Leu127, Phe169 and Val313 on SLC19A2 are replaced by Phe56, Val109, Tyr151 and Leu296 at equivalent positions on SLC19A3 (Extended Data Fig.\u00a07). In the inward-facing SLC19A3, the thiazolium ring of thiamine is approached by Tyr151 hydroxyl group at its sulfur on one side, and by the hydrophobic Leu296 on the other side (Fig.\n \n 2\n \n f). Instead, SLC19A2 Phe169 lacks the hydroxyl group, while Val313 has a shorter side chain. Second, Asn297 establishes a hydrogen bond with the primary amine group of thiamine in SLC19A3, while the counterpart Asn314 of SLC19A2 orients away from thiamine (Fig.\n \n 3\n \n f). Therefore, these minor but significant variations may contribute to a slightly lower affinity of thiamine for SLC19A2 than for SLC19A3, resulting in divergent kinetics for the two transporters.\n
\n\n Thiamine transporters SLC19A2/A3 have been recently identified as the long-seeking carrier for pyridoxine (vitamin B\n \n 6\n \n ) absorption, a protonophore-sensitive process that favors acidic conditions over neutral to basic conditions\n \n \n 7\n \n \n . Our MST measurements revealed that pyridoxine binds relatively weaker to SLC19A2 (\n \n K\n \n d\u2009~\u2009161.4 \u00b5M) than to SLC19A3 (\n \n K\n \n d\u2009~\u200988.8 \u00b5M) at pH 6.0 (Fig.\n \n 3\n \n c), consistent with previous cellular uptake\n \n K\n \n m values\n \n \n 7\n \n \n . Interestingly, both transporters showed substantially increased affinity for pyridoxine at pH 7.5 (Fig.\n \n 3\n \n d).\n
\n\n To understand the molecular mechanism for pyridoxine recognition and transportation, we further determined the structures of pyridoxine in complex with SLC19A3 and SLC19A2 (Extended Data Figs.\n \n 4\n \n and\n \n 5\n \n ). The outward-facing pyridoxine-bound SLC19A3 structure is nearly identical to thiamine-bound SLC19A3 (C\u03b1 RMSD 0.43 \u00c5, Extended Data Fig.\u00a06c), with pyridoxine inserted at the similar cavity to thiamine and embraced by almost the same set of residues exclusively on the N-domain (Figs.\n \n 3\n \n g and\n \n 3\n \n h). Specifically, the pyridine ring is clamped by Phe56 and Tyr113 through \u03c0-\u03c0 stacking, contacted by Glu32 and Glu110 via hydrogen bonding, and buttressed by Trp59, Thr93, Trp94, Leu97 and Val109 upon hydrophobic interaction (Fig.\n \n 3\n \n h). Likewise, the inward-facing pyridoxine-bound SLC19A2 structure is also similar to thiamine-bound SLC19A2 (C\u03b1 RMSD 1.04 \u00c5, Extended Data Fig.\u00a06d), with pyridoxine occupying the same cluster of hydrophilic or hydrophobic residues in thiamine binding site (Figs.\n \n 3\n \n i and\n \n 3\n \n j). These observations thus corroborate the notion that pyridoxine is a competitive substrate for thiamine transporters\n \n \n 7\n \n \n .\n
\n\n Fedratinib (Inrebic\u00ae) is a newly FDA-approved selective inhibitor of Janus kinase 2 (JAK-2) to treat myeloproliferative diseases including myelofibrosis\n \n \n 32\n \n \n , with a boxed warning regarding the risk of potentially fatal encephalopathy. The clinical development of fedratinib was halted in 2013, when several cases consistent with Wernicke\u2019s encephalopathy were reported in some participants\n \n \n 20\n \n \n . We confirmed the inhibitory effect of fedratinib on both SLC19A2- and SLC19A3-mediated thiamine absorption in HEK293T cells (Fig.\n \n 4\n \n a), and assessed the direct binding of fedratinib to purified SLC19A3 (\n \n K\n \n d\u2009~\u20090.54 \u00b5M), and to a lesser extent to SLC19A2 (\n \n K\n \n d\u2009~\u20096.78 \u00b5M), by the MST assay (Fig.\n \n 4\n \n b). The 10-fold difference in\n \n in vitro\n \n binding affinity likely underpins the mechanism that fedratinib inhibits thiamine uptake by SLC19A3 slightly stronger than by SLC19A2 (IC50: 1.09 \u00b5M for SLC19A3 vs 10.7 \u00b5M for SLC19A2, respectively)\n \n \n 33\n \n \n . We then determined the structures of SLC19A3 with fedratinib in the outward- and inward-facing conformations at 3.1-\u00c5 and 3.0-\u00c5 resolution, respectively (Extended Data Figs.\n \n 4\n \n and\n \n 5\n \n ).\n
\n\n Both fedratinib-bound structures share an overall similar architecture with the corresponding thiamine-bound outward-facing (C\u03b1 RMSD 0.37 \u00c5) and inward-facing SLC19A3 (C\u03b1 RMSD 0.73 \u00c5), respectively (Extended Data Figs.\u00a06e and 6f). In the outward-facing structure, fedratinib adopts a bent conformation with its two semi-equal length branches kinking around the 2,4-diaminopyrimidine moiety (Fig.\n \n 4\n \n c). Interestingly, this 2,4-diaminopyrimidine group occupies the same position as the aminopyridine ring of thiamine bound in outward-facing SLC19A3, which allows the establishment of \u03c0-\u03c0 stacking against Tyr113, and hydrogen bonding with two acidic residues Glu32 and Glu110 via its two amine nitrogens. In addition, the benzene ring on the pyrrolidine branch also \u03c0-stacks with Phe56 on TM5, a mimic of the thiamine thiazolium ring, and the terminal pyrrolidine ring approaches Asn297, Tyr298 and Ile301 on TM7. On the opposite sulfonamide branch, the sulfonyl group H-bonds with Tyr113 and is proximal to Arg29 on TM1, and the distal hydrophobic tert-butyl group is close to Tyr151 and Leu296 (Fig.\n \n 4\n \n d).\n
\n\n In the inward-facing state, however, fedratinib adopts an even more compact conformation (Fig.\n \n 4\n \n e). Although the diaminopyrimidine group and the sulfonamide branch remain in nearly the same position as that in the outward-facing state, the pyrrolidine branch swings away from Phe56 and bends toward the intracellular exit, with the benzene ring T-stacking against Trp59 indole group and the pyrrolidine ring facing its sulfonamide group (Fig.\n \n 4\n \n f). This conformational rearrangement of fedratinib is similar to that of the thiamine transitions from outward- to inward-facing state. These features further support the concept that fedratinib inhibits thiamine transporters by structurally mimicking thiamine\n \n \n 20\n \n \n .\n
\n\n Thiamine-like drugs and other structurally unrelated cationic compounds have been demonstrated interaction with thiamine transporters\n \n \n 21\n \n ,\n \n 33\n \n \n . Our cellular\n \n \n 3\n \n \n H-thiamine uptake assays confirmed the inhibitory effects of thiamine analogues (amprolium, oxythiamine, trimethoprium, pyrimethamine), tyrosine kinase inhibitors (fedratinib, momelotinib, imatinib), antidepressant sertraline, as well as metformin. We further expanded the inhibitor list to include CDKs inhibitor abemaciclib and reverse transcriptase inhibitor etravirine (Extended Data Fig.\u00a08a). Most, if not all, of the drugs have an aminopyrimidine core, a typical characteristic suitable for recognition by thiamine transporters\n \n \n 21\n \n ,\n \n 33\n \n \n . To further elucidate the molecular basis of these compounds in addition to fedratinib, we determined SLC19A3 structures in complex with coccidiostat amprolium and antidiabetic metformin, both in outward-facing conformation at 3.1-\u00c5 resolution (Extended Data Figs.\n \n 4\n \n and\n \n 5\n \n ).\n
\n\n In contrast to the bent conformation of thiamine, amprolium adopts an extended pose in the similar binding pocket on SLC19A3 N-domain (Fig.\n \n 5\n \n a). The aminopyrimidine ring of amprolium overlaps with that of thiamine and is engaged by the same cluster of residues, as anticipated. The propyl chain adorned on pyrimidine ring extends to the hydrophobic cage composed of Thr93, Trp94, Leu97 and Val109. The pyridine ring, a substitute for thiamine\u2019s thiazolium ring, stacks nearly face-to-face with Trp59 and edge-to-face against Phe56 (Fig.\n \n 5\n \n b-c). The semi-conserved interaction network thus maintains a tight contact for amprolium with SLC19A3 and SLC19A2 (\n \n K\n \n d\u2009~\u20090.45 \u00b5M and ~\u20093.05, respectively) at pH 6.0, with comparable binding affinities to thiamine (Extended Data Fig.\u00a08b).\n
\n\n The metformin-SLC19A3 structure demonstrates a similar coordination network for the biguanide to that of thiamine, albeit without an aminopyrimidine ring (Fig.\n \n 5\n \n d). Specifically, metformin is clamped by Phe56, Trp59 and Tyr113 via cation-\u03c0 interactions and balanced by flanking hydrogen bonds with Glu32 and Glu110. The dimethyl substituent inserts into the same hydrophobic cage as described above (Figs.\n \n 5\n \n e-f). This interaction pattern differs from that of organic cation transporter 1 (OCT1), a well-known carrier for metformin\n \n \n 34\n \n \n , which has a similar millimolar affinity to SLC19A3 (Extended Data Fig.\u00a09). Despite cation-\u03c0 stacking against neighboring aromatic residues, metformin did not interact with the acidic residues Glu386 or Asp474 in the inward-facing OCT1 structure\n \n \n 35\n \n \n . Notably, in the same study, the thiamine was also distant from either Glu386 or Asp474 on OCT1 (Extended Data Fig.\u00a09). Nevertheless, our data support the notion that metformin is a substrate and inhibitor of SLC19A3\n \n 19\n \n .\n
\n\n Membrane transporters play a crucial role in the absorption and distribution of nutrients and drugs across biogenic membrane barriers. Significant progress has been achieved in characterizing the functional aspects of cellular thiamine and pyridoxine uptake through high-affinity transporters SLC19A3/THTR-2 and SLC19A2/THTR-1. These transporters, responsible for thiamine and pyridoxine uptake, are potential targets for drug-drug and drug-nutrient interactions. By determining the cryo-EM structures of SLC19A3 and SLC19A2 at different transport states, coupled with the structure-based mutagenesis analysis, here we highlight critical determinants governing the recognition and transport of substrate B vitamins (thiamine and pyridoxine) and therapeutic drugs (metformin, fedratinib, and amprolium) by thiamine transporters.\n
\n\n It has been proposed that the proton gradient across membrane acts as a driving force for thiamine transporters-mediated ligand movement\n \n \n 7\n \n ,\n \n 19\n \n ,\n \n 36\n \n \n . The pH conditions impact SLC19A2- and SLC19A3-mediated thiamine and pyridoxine absorption differently. Our structural and biochemical analyses shed light on this phenomenon. Several acidic residues, including Glu32, Glu110, and Glu320, participate in thiamine and pyridoxine recognition. Notably, genetic mutations of Glu320 have been associated with Wernicke\u2019s encephalopathy (E320Q)\n \n \n 11\n \n \n and BTBGD (E320K)\n \n \n 37\n \n \n , underscoring the significance of this residue. We speculate that interactions between protons and these key residues play pivotal roles in substrate translocation. Thiamine exhibits tighter binding with both transporters at pH 6.0, while pyridoxine binds more strongly at pH 7.5 (Fig.\n \n 3\n \n ). This increased ligand binding affinity under certain pH conditions may prolong the duration of ligand-transporter association, thereby influencing the transport cycling rate. This would explain previous observations that thiamine uptake activity peaks around pH 7.4, while pyridoxine transport peaks around pH 5.5\n \n 7\n \n .\n
\n\n Fedratinib, a recently licensed antineoplastic treatment, selectively targets intracellular JAK2 kinase\n \n \n 20\n \n \n . Despite extensive pharmacological profiling, its specific absorption mechanism remains unknown. The resemblance between fedratinib and thiamine, both possessing the aminopyrimidine chemical core, along with substantial conformational rearrangements observed in the outward- and inward-facing SLC19A3 structures, suggests that fedratinib inhibits SLC19A3 or SLC19A2-mediated thiamine absorption by competing for the same molecular components required for thiamine translocation. This also implies that fedratinib may be imported by thiamine transporters, warranting further investigation.\n
\n\n Cationic medicines like metformin and fedratinib may predispose patients to thiamine and pyridoxine deficiency, even without known risk factors. Regular assessment of plasma thiamine levels is recommended during treatment with these medications. Adequate thiamine supplementation may mitigate this overlooked disorder. Diabetic patients frequently experiencing severe lactic acidosis concurrent with thiamine deficiency while taking biguanide drugs (e.g. buformin and metformin), have shown recovery following intravenous infusion of high-dose thiamine\n \n \n 24\n \n ,\n \n 38\n \n \n . Alternatively, the molecular insights into SLC19A3 inhibition by prescribed drugs elucidated in the current study, applicable to SLC19A2 as well, may aid in the future improvement of these medicines through structure-based rational design.\n
\n\n During the preparation of our manuscript, a preprint by Gabriel\n \n et al\n \n . presented cryo-EM structures of SLC19A3 with thiamine in outward- and inward-facing states, and several drugs including fedratinib and amprolium in inward-facing state\n \n \n 39\n \n \n . Meanwhile, Dang\n \n et al\n \n . reported SLC19A3 bound by thiamine, pyridoxine and fedratinib in the inward-facing conformation\n \n \n 40\n \n \n . These structures, obtained via different strategies, thus cross-validate and complement our findings regarding SLC19A3 and SLC19A2.\n
\n\n In conclusion, our elucidation of thiamine transporters structures in association with vitamins B\n \n 1\n \n and B\n \n 6\n \n , and several clinical drugs, alongside biochemical evidence, offers insights into the mechanism underlying the transport of cationic vitamins and medicines. This also provides a framework for developing targeted pharmaceutical strategies.\n
\n\n The human SLC19A3 (residues 6\u2013472) gene (UniProt accession code: Q9BZV2) was subcloned and inserted into a pFastBac-dual vector (Invitrogen), followed by a TEV site and a C-terminal 11\u00d7His tag. SLC19A3 was expressed in Spodoptera frugiperda Sf9 cells. Cells were infected at a density of 2.5-3 \u00d7 10\n \n 6\n \n cells per ml. After growing for 72 h at 27\u00b0C, cells were collected and resuspended in buffer containing 50 mM Tris-HCl pH 7.5, 300 mM NaCl, 20 mM imidazole. Then cells were disrupted by sonication and solubilized with 1% (w/v) n-dodecyl-b-d-maltoside (DDM, Anatrace) at 4\u00b0C for 1.5 h. After centrifugation (16,000 rpm, 30 min, 4\u00b0C), the supernatant was incubated with nickel affinity resin at 4\u00b0C for 1 h. The resin was then washed with buffer containing 50 mM Tris-HCl pH 7.5, 300 mM NaCl, 40 mM imidazole and 0.025% (w/v) n-dodecyl-b-D-maltoside (DDM, Anatrace). The human SLC19A3 protein was eluted with 50 mM Tris pH 7.5, 300 mM NaCl, 500 mM imidazole, 0.025%(w/v) DDM. Then the protein was concentrated and loaded onto a Superdex 200 Increase 10/300 GL size-exclusion column (GE Healthcare) in the presence of 20 mM HEPES pH 7.5, 150 mM NaCl, 0.025% (w/v) DDM. The peak fractions were concentrated to about 10 mg/ml.\n
\n\n Human SLC19A3 lacking the N-terminal 12 residues was modified with an N-terminal MPER(LWNWFDITNWLWYIKSL) and cloned into the pcDNA3.1 vector. SLC19A3 was expressed in HEK293 cells. Cells were infected at a density of 2 \u00d7 10\n \n 6\n \n cells per ml and 10 mM sodium butyrate was added. Cells overexpressed SLC19A3 were collected 58 h after infection and were re-suspended in buffer A (50 mM HEPES, pH 7.5, 150 mM NaCl, 5% glycerol, and 1 \u00d7 protease inhibitor cocktail). The re-suspended cells were lysed mechanically with a Dounce tissue grinder and agitated at 4\u00b0C for 3 h in lysis buffer containing 1% LMNG, and 0.1% CHS. After agitation, the supernatant was collected after centrifugation at 1,2000 rpm at 4\u00b0C for 40 min and incubated with anti-Strep affinity resin by agitation for 3 h. Then the resin was collected on a gravity column and the supernatant was incubated with new anti-Strep affinity resin by agitation for 3 h. Then, the resin was washed with 10 column volume (CV) of buffer B (buffer A supplemented with 0.1% LMNG, 0.01% CHS), buffer C (buffer A supplemented with 0.01% LMNG, 0.001% CHS), buffer D (buffer A supplemented with 0.001% LMNG, 0.0001% CHS). SLC19A3 was eluted with buffer E (buffer A supplemented with 2.5 mM D-Desthiobiotin). The elution was added to Fab_10E8v4 in a ratio of 1:10 and concentrated then further purified by size-exclusion chromatography on a Superose6 10/300 GL column (GE Healthcare) in buffer containing 0.001% LMNG, 0.0001% CHS, 50 mM HEPES, pH 7.5, and 150 mM NaCl. For the Thiamine sample, Thiamine (Sigma Aldrich) was added at a concentration of 2.5 mM to the SEC buffer (50 mM MES, pH 6.0, and 150 mM NaCl, 0.001% LMNG, 0.0001% CHS). The peak fractions were concentrated to 14.5 mg/ml for grid preparation.\n
\n\n Human SLC19A2 lacking the N-terminal 30 residues was modified with an N-terminal MPER tag and cloned into the pcDNA3.1 vector. The expression and purification of SLC19A2 were carried out similarly as described above with SLC19A3.\n
\n\n The BALB/c mice were immunized monthly with human SLC19A3 (residues 6-472) plus Mn2\u2009+\u2009adjuvant (MnStarter Biotechnology) and CpG1826 (Thermo Fisher Scientific). Sera from the immunized mice were analyzed for reactivity towards human SLC19A3-expressing HEK293T cells using flow cytometry. Hybridoma formation was conducted following our previously established protocols\n \n \n 41\n \n \n . Briefly, three days post final boost immunization (without adjuvant), spleen and lymph node cells from the immunized mice were isolated and fused with SP2/0 myeloma cells using 50% polyethylene glycol (PEG) 4000 (Sigma-Aldrich). Subsequently, the cells were plated in 96-well plates containing Hypoxanthine/Aminopterin/Thymidine (HAT) medium (Sigma-Aldrich) with feeder cells (peritoneal macrophages) that had been seeded 24 hours earlier. The hybridoma culture supernatants were screened using flow cytometry with human SLC19A3 and human SLC19A3 (6/7 loop replaced with GSGSGSGSGS) expressing HEK293T cells on an Attune NxT Flow Cytometer (Thermo Fisher Scientific). Double positive hybridomas were subcloned, after 7 days, the subcloned supernatants were analyzed as previously described. The double positive clones were then collected, barcoded using the 10x Chromium Single Cell platform (10x Genomics), and 17 recombinant monoclonal antibodies against SLC19A3 were synthesized using established methods\n \n \n 27\n \n \n . BCR sequences were assembled and evaluated using the V(D)J tool of Cell Ranger suit (v.6.0.1) against the GRCm38 reference genome with specific parameters. Contigs labeled as low-confidence, non-productive or with UMIs\u2009<\u20092 were excluded. Only cells containing at least one light chain (IGL) and one heavy chain (IGH) were remained. In each filtered B cell, light and heavy chains were paired and ranked based on their mean clone levels. The selected V(D)J sequences of IGL and IGH were synthesized and cloned into vectors containing the myc-tagged heavy chain constant regions 1 (CH1) of mouse IgG2a or \u03ba light chain respectively. The antibodies were produced in 293F Human Embryonic Kidney cells and purified with a protein A agarose prepacked column. The binding of the purified antibodies to SLC19A3 and its truncations was confirmed by FACS.\n
\n\n The DNA sequences of the heavy and light chains of MPER - bound Fab_10E8v4 (McIlwain BC, et al 2021,\n \n J Mol Biol.\n \n ) were cloned into pDEC vectors separately. 2 mg plasmid of Fab (including 1 mg heavy and 1 mg light chains) and 4 mg of PEI were mixed in 100 mL of medium for 15 min at RT before being added into 1 L of cell culture at a density of 2 \u00d7 10\n \n 6\n \n ml\n \n \u2212\u20091\n \n , containing 10 mM sodium butyrate. After 48 h, the cell culture media was centrifuged at 2000 g for 15 min, and then the supernatant was filtered and exchanged into buffer F (50 mM Tris, pH 8.0, 150 mM NaCl, 5% glycerol). Then, the supernatant was incubated with Ni beads by agitation for 3 h, and the beads were washed with 20 CV of buffer F containing 20 mM imidazole. Fab_10E8v4 was eluted with buffer F with the addition of 300 mM imidazole. The elution was concentrated to high concentration and stored at -80\u00b0C. SLC19A3-bound Fabs were expressed and purified similarly as Fab_10E8v4.\n
\n\n For the SLC19A3-Fab complex in detergent micells, SLC19A3 was incubated with fab at a molar ratio of 1:1.1 for 1h and the mixture was loaded onto the Superdex 200 Increase 10/300 GL size-exclusion column and eluted with buffer containing 20 mM HEPES pH 7.5, 150 mM NaCl and 0.025%(w/v) DDM. Peak fractions containing SLC19A3\u2013Fab complexes were concentrated to about 12 mg/ml. For nanodisc reconstitution, SLC19A3 was mixed with membrane scaffold protein MSP1D1 and POPG (Avanti) at a molar ratio of 1:1.9:84 and incubated at 4\u00b0C with constant rotation for 1 h. Subsequently, Bio-beads were added to the mixture at 100 mg/ml to remove detergent at 4\u00b0C overnight with gentle agitation. The Bio-beads were removed and the nanodisc reconstitution mixture was incubated with fab at a molar ratio of 1:1.1 for 1h. Then the mixture was loaded onto the Superdex 200 Increase 10/300 GL size-exclusion column and eluted with buffer containing 20 mM HEPES pH 7.5 and 150 mM NaCl. Peak fractions corresponding to nanodisc-reconstituted SLC19A3\u2013Fab complexes were concentrated to about 12mg/ml.\n
\n\n Quantifoil Au 1.2/1.3 (300 mesh) grids were glow-discharged 10 mA for 50 s in an PELCO easiGlo instrument. 2.5 \u00b5L protein samples were deposited on the grids and blotted for 4 s with filter paper at 4 \u2103 and 100% humidity using Vitrobot (FEI) equipment and vitrified in liquid ethane at liquid nitrogen temperature. The frozen grids were transferred under cryogenic conditions and stored in liquid nitrogen for subsequent screening and cryo-EM data collection. To prepare the substrate-bound SLC19A3 samples, 100 \u00b5M fedratinib was incubated with the MPER-SLC19A3/Fab_10E8v4 complex on ice for 1h. For the cryo-EM sample of SLC19A2 bonded pyridoxine, 200 \u00b5M pyridoxine was incubated with the MPER-SLC19A2/Fab_10E8v4 complex on ice for 1h. To improve particle distribution, 0.035 mM fluorinated octyl maltoside was added to all cryo-EM samples. All datasets were collected on Titan Krios G4 cryo-electron microscope operated at 300 kV, equipped with a Falcon G4i direct electron detector with a Selectris X imaging filter (ThermoFisher Scientific), operated with a 20-eV slit size. Movie stacks were acquired using the EPU software (ThermoFisher Scientific) in super-resolution mode with a defocus range of -1.2 to -2.0 \u00b5m and a final calibrated pixel size of 0.932 \u00c5. The total dose per EER (electron event representation) movie was 50 e\n \n \u2013\n \n /\u00c5\n \n 2\n \n .\n
\n\n For the SLC19A3-thiamine/pyridoxine/metformin samples, the purified SLC19A3\u2013Fab complexes in detergent micells were concentrated to about 12 mg/ml and separately incubated with 5 mM thiamine (Sigma-Aldrich), 5 mM pyridoxine (Sigma-Aldrich) or 5 mM metformin (Sigma-Aldrich) for 1 h before being applied to the grids. For fedratinib/amprolium-bound samples, the purified nanodisc-reconstituted SLC19A3\u2013Fab complexes were separately incubated with 1 mM fedratinib (Sigma-Aldrich) or 5 mM amprolium (Sigma-Aldrich) for 1 h before cryo-EM sample preparation. In brief, 3 \u00b5l of the purified SLC19A3-ligand complexes in detergent micelles or nanodiscs was added to glow-discharged holey grids (Au R1.2/1.3, 300 mesh Quantifoil). The grids were blotted for 3-4s at 4\u00b0C with 100% humidity, and then plunge-frozen into liquid ethane. The cryo-EM data for SLC19A3-amprolium sample were collected using a Titan Krios electron microscope (Thermo Fisher Scientific) equipped with a BioQuantum GIF energy filter with a K2 summit direct detector (Gatan). Other cryo-EM datasets were collected using SerialEM\n \n \n 42\n \n \n on the Talos Arctica 200 kV FEG (Thermo Fisher Scientific) with a K2 summit direct electron elector (Gatan) and a GIF quantum energy filter (Gatan). All movie stacks were automatically acquired at a magnification of 130,000\u00d7 under superresolution mode. The slit width was set to 20 eV. The total dose was 60 e \u00c5\n \n \u22122\n \n with a dose rate of 9.2 e\n \n \u2212\n \n \u00c5\n \n \u22122\n \n s. Each video was fractionated into 32 frames. The defocus range was set between \u2212\u20091.2 and \u2212\u20091.5 \u00b5m. The pixel size was calibrated at 0.5 \u00c5 (\u00d7130,000) under super-resolution mode. Images were recorded using beam\u2013image shift data collection methods\n \n \n 4\n \n \n .\n
\n\n For the outward-open SLC19A3-apo/thiamine/pyridoxine/fedratinib/amprolium/metformin structure dataset, 914, 2465, 1445, 2227, and 2152 super-resolution movie stacks were aligned, summed and dose-weighted using the program MotionCor2\n \n 43\n \n , and then imported into cryoSPARC\n \n \n 44\n \n \n . The processing of the outward-open SLC19A3-apo/ thiamine/ pyridoxine/ fedratinib/ amprolium/ metformin structure analysis adopted a similar scheme of classification and refinement; therefore, the detailed procedures were introduced with SLC19A3-thiamine dataset processing as example (Extended Data Fig.\n \n 3\n \n ). The processing of the other datasets was illustrated in flowchart (Extended Data Figs.\n \n 4\n \n \u2013\n \n 5\n \n ). All datasets were similarly processed in cryoSPARC (v.4.2.1) and RELION (v.3.1.4)\n \n \n 45\n \n \n .\n
\n\n For the inward-open SLC19A2 and SLC19A3 structures, all datasets were similarly processed in cryoSPARC (v.3.3.2) and RELION (v.3.1.4). Briefly, each 1080-frame EER movie was divided into 40 subgroups, and beam-induced motion was corrected using a MotionCor2-like algorithm implemented in RELION. Exposure-weighted micrographs were then imported into cryoSPARC for CTF (contrast transfer function) estimation by patch CTF. Particles were blob-picked and extracted and multiple rounds of 2D classification were performed. Multiple rounds of heterogeneous refinement (3D classification) were performed using ab initio reference maps reconstructed with good 2D averages. The good particles were then converted to Bayesian polishing in RELION and imported back into cryoSPARC. Final maps were obtained by local refinement on the transmembrane domain of SLC19A3. The resolution of these maps was estimated internally in cryoSPARC by gold standard Fourier shell correlation using the 0.143 criterion.\n
\n\n For the atomic model of apo SLC19A3, the structure of SLC19A3 (ID: AF-Q9BZV2-F1) predicted by AlphaFold\n \n \n 46\n \n \n , as the initial model, was manual fitted in UCSF Chimera\n \n \n 47\n \n \n and checked in COOT\n \n \n 48\n \n \n . The corrected model was further refined by real space refinement in PHENIX\n \n \n 49\n \n \n . CIF files for ligands were generated in PHENIX using eLBOW\n \n \n 50\n \n \n . In COOT and PHENIX, with the apo SLC19A3 as the initial model, the atomic model of ligand-bound SLC19A3 was generated by several rounds of real space refinement. Thiamine, or pyridoxine, or fedratinib, or amprolium, or metformin was fitted into the density using COOT. The resulting model was then manually rebuilt in COOT and further refined by real space refinement in PHENIX. The model stereochemistry was evaluated using the comprehensive validation (cryo-EM) utility in PHENIX. The final refinement statistics are provided in Extended Data Table\u00a01. All figures were prepared with UCSF ChimeraX\n \n \n 51\n \n \n or Pymol (PyMOL Molecular Graphics SYtem, v.2.3.4, Schr\u00f6dinger) (\n \n \n https://pymol.org/2/\n \n \n \n \n ).\n
\n\n The DNA sequences encoding human SLC19A2/SLC19A3 and responsive mutants were cloned into a lentiviral plasmid. This lentiviral plasmid co-expressed the reporter gene mcherry through a P2A sequence controlled by the human EEF1A1 promoter. For lentiviral gene transduction, HEK293T cells were transfected with the respective lentiviral vectors and packaging plasmids \u03c3NRF and vesicular stomatitis virus G (an envelope plasmid) using standard calcium phosphate techniques. After 48 hours, culture supernatants were collected, filtered through 0.45-\u00b5m polyethersulfone filters (Merck Millipore) and supplemented with 8 \u00b5g/ml polybrene (Sigma-Aldrich). Cells were infected by spinfection (1,500 rpm, 180 min, room temperature). Following 72 hours of culture, lentiviral-infected cells expressing similar levels of mcherry were isolated using a BD FACSAria III cell sorter (BD Biosciences).\n
\n\n Stably expressing either wild type or mutated SLC19A3 or SLC19A2 293T cells were seeded into poly-lysine-coated 24-well plates at 1 x 10\n \n 5\n \n cells per well and grown for 12 hours. Cells were firstly washed once with 0.5 ml HBSS buffer at pH 7.4 and then incubated in HBSS buffer at 37\u2103 for 10 minutes. Subsequently, 0.2 ml HBSS buffer containing 5 nM [\n \n \n 3\n \n \n H]-thiamine (American Radiolabeled Chemicals) was used to replace the cell medium to initiate the uptake assay. After 3 minutes, cells were washed twice with 0.5 ml ice-cold HBSS buffer, and then lysed with 0.2 ml 0.2 M NaOH for 5 minutes. The amount of [\n \n \n 3\n \n \n H]-thiamine was calculated by and indicated concentration of inhibitors was used to initiate the uptake. All datasets were analyzed using GraphPad Prism 9 (\n \n \n https://www.graphpad.com/\n \n \n \n \n ).\n
\n\n Porcine reproductive and respiratory syndrome virus (PRRSV) sublineage 8.7 has been estimated as one of the most devastating and longest-circulating lineages in PRRSV, especially the emergence and prevalence of highly pathogenic PRRSV in 2006. Despite a rapid increase in sublineage 8.7 virus epidemic outbreaks in Asian countries over recent years, very little is known about the patterns of virus evolution, spread, and the spatial, demographic, and ecological factors influencing PRRSV spread. Relying on a national PRRSV surveillance project established over 20 years ago, we expanded the genomic dataset outbreak in China and deployed a series of phylogeographic extension of this dataset that enables formal testing the contribution of a range of predictor variables to the geographic spread of PRRSV. We revealed the principal role of Guangdong as a central source in Asia, with rural swine activities and provincial distance contributing to spatial spread. Independent recombination analysis of interlineage and intralineage with its temporal dynamics captured a peak wave spanning 2014 to 2016. Noted that several HP-PRRSV modified live vaccines (MLVs) were hastily approved for use on a remarkably emergency basis in China since the epidemic whereas few studies focused on its potential impact on the field spanning a long temporal vaccination, we sequenced all available three MLVs and genomic analysis suggested a key leaky period spanning 2011 to 2017, with two concurrent amino acid mutations located in ORF1a 957 and ORF2 250. Overall, our study provides a phylodynamic framework to showcase a full-scale knowledge of PRRSV sublineage 8.7 evolution, transmission dynamics, and potential leaky evidence of HP-PRRSV MLVs, providing critical insights into new MLV development under\n \n Nidovirale\n \n order.\n
\n\n Porcine reproductive and respiratory syndrome (PRRS), one of the most devastating diseases globally spreading around the pork industry, especially in China and the United States, causes continuous economic losses over decades\n \n \n 1\n \n \n . The etiologic agent, porcine reproductive and respiratory syndrome virus (PRRSV), is a single positive-strand enveloped RNA virus, which possesses over 15kbs genome, containing at least ten open reading frames (ORFs): ORF1a, ORF1b, ORF2a, ORF2b, ORF3, ORF4, ORF5, ORF6, and ORF7. PRRSV belongs to the order\n \n Nidovirales\n \n and family\n \n Arteriviridae\n \n , emerging almost simultaneously as two genotypes (\n \n Betaarterivirus suid\n \n 1 and\n \n Betaarterivirus suid\n \n 2), with almost 50%-70% nucleotides homogeneity\n \n \n 2\n \n \n . Besides,\n \n Betaarterivirus suid\n \n 2 can be further divided into 9 disparate lineages based on ORF5 gene (Shi et al., 2010). To date, lineage 1, lineage 3, lineage 5, and sublineage 8.7 strains have been circulated in mainland of China\n \n \n 3\n \n ,\n \n 4\n \n \n .\n
\n\n PRRSV Lineage 8 strains have a long evolutionary history around the world. In 1995, an \u201cabortion storm\u201d swept over Iowa and spilled over the whole United States, for which lineage 8 PRRSVs were part of the emerging strains\n \n \n 5\n \n \n . Subsequently, sublineage 8.7 strain CH-1a was first detected in China and subsequently induced the widely viral dissemination on the farms without a district biosafety measure. Thereafter, sublineage 8.7 strains have been frequently detected and ascertained as an endemic cluster in China. However, the outbreak of highly pathogenic PRRSV (HP-PRRSV) with much higher virulence reported in 2006 further exacerbated the evolutionary history of sublineage 8.7 and resulted in more economic losses as the global swine trade grew at an amazing rate\n \n \n 6\n \n \n . Further evolutionary analysis suggested HP-PRRSV isolates belong to sublineage 8.7. In the following years, a series of mutation conducts such as recombination and adaptive evolution of HP-PRRSV also enlarged the complexity of the genetic diversities\n \n \n 7\n \n ,\n \n 8\n \n \n . Given the increased harm from HP-PRRSV, three modified live vaccines (MLV) with the attenuated strains JXA1-R, HuN4-F112, and TJM-F92 were imminently licensed for emergency control of HP-PRRSV, which had been widely used in China until recent years\n \n \n 9\n \n \u2013\n \n 11\n \n \n , and another HP-PRRSV derived vaccine GDr180 was also licensed in 2015. Due to lacking the 3\u2019-5\u2019 exonuclease proofreading ability during replication, the MLV is characterized by low replication fidelity and high mutability (recombination and reversion to virulence), that is to say, HP-PRRSV-derived MLV acquires more chances to escape from host immune system\n \n \n 12\n \n \n . However, although several\n \n in vivo\n \n and\n \n in vitro\n \n experiments have confirmed the potential that HP-PRRSV-derived MLV vaccines can easily regain its virulence, we still lacked a study to comprehensively assess the impact underlying the HP-PRRSV MLV vaccines noted it has been widely used in China over fifteen years\n \n \n 10\n \n ,\n \n 13\n \n ,\n \n 14\n \n \n .\n
\n\n When confronted with low diversity and sampling bias, evolutionary reconstructions may greatly benefit from integrating additional sources of information. Bayesian phylodynamic approaches are particularly adept for this purpose, and phylogeographic methods in particular have been extended to take advantage of human transportation data as proxies of population-level connectivity between locations. This approach has been utilized in a wide range of applications, including the identification of the key drivers of Ebola virus spread in West Africa and Omicron BA.1 in the United Kingdom\n \n \n 15\n \n ,\n \n 16\n \n \n . Additionally, recent attempt of Bayesian phylogeographic inference with GLM extension successfully proved the role of anthropogenic activities (i.e., swine trade) in the transmission of the Porcine Epidemic Diarrhea Virus\n \n \n 17\n \n \n .\n
\n\n Under our national epidemiological surveillance project of PRRSV, China is constantly experiencing the key genotypic shift from lineage 8 to lineage 1\n \n 3,18\n \n . However, as for the devastating lineage, i.e., sublineage 8.7, we still failed to systematically investigate how the dominant sublineage 8.7 spread globally and locally. To fulfill the gaps mentioned above and carry out the investigation as a part of the national surveillance project of PRRSV, we assembled the largest genomic dataset regarding PRRSV-2 sublineage 8.7, including 242 novel sublineage 8.7 ORF5 sequences collected spanning 2005\u20132022. Importantly, we also sequenced all available MLV vaccines derived from sublineage 8.7 HP-PRRSV. With this expanded genomic dataset, we performed a series of genomic analyses to answer important questions on the emergence and spread of sublineage 8.7, including: 1) How did sublineage 8.7 emerge and spread globally? 2) As the most devastated country, what factors may affect the spread of the virus in China? 3) Is the recombination of PRRSV showing any exceptional preference dynamics? 4) Did the vaccine strains contribute to the clinical sublineage 8.7 spread in China? Overall, these findings filled the key knowledge of PRRSV evolution. These works highlighted the importance of necessary and continuous genomic surveillance to drop the genomic diversity of PRRSV globally and provide \u201cleaky\u201d evidence of HP-PRRSV MLVs.\n
\n\n In our national surveillance project on PRRSV, over 6000 clinical specimens were collected from the suspected PRRSV-positive farms in most pig-raising regions of China from 2005 to 2022. Specimens were ground by a freezing grinder (JXFSTPRP-CLN-48, Shanghai Jingxin Industrial Development Co., Ltd., China) and the viral genomes were extracted by RNA fast200 kit (Fastagen, Shanghai, China) following the instruction of the manufacturer. Collectively, 242 newly ORF5 sequences and 42 new complete genomes belonging to sublineage 8.7 were obtained in the China mainland (Supplementary Table\u00a01). Furthermore, we downloaded all\n \n Betaarterivirus suid 2\n \n ORF5 and complete genome sequences as of 2022 (update to Mar, 2022) from the GenBank database. Furthermore, ORF5 dataset were filtered to exclude sequences that: 1) lack of collection date or location (accurate to province units for Chinese isolates), 2) vaccine strains, 3) unverified sequence, 4) virus serially passaged in cells, 5) ambiguous and deleted residues. Subsequently, the left database of ORF5 sequences was aligned using MAFFT algorithm, truncating all the nucleotide sites except the ORF5 in MEGA7 software manually\n \n \n 19\n \n ,\n \n 20\n \n \n . For the complete-genomes database, we implemented MAFFT to align the sequences then removed ambiguous regions using the TrimAL algorithm\n \n \n 21\n \n \n .\n
\n\n Subsequently, multiple rounds of maximum-likelihood analyses were deployed to form the final lineage 8 database using IQ-TREE\n \n \n 22\n \n \n . Totally, 3708 ORF5 sequences belonging to lineage 8 were extracted. Furthermore, 2340 ORF5 sequences (including 242 sequences from our lab) were identified as sublineage 8.7 and extracted from the global lineage 8 phylogeny to generate the final sublineage 8.7 genomic database. Of note, as the phylogeography analysis of sublineage 8.7 is part of our main part, we further extract the geographical information of viruses in China accurately to the provincial level.\n
\n\n To reconstruct the evolutionary relationship, we firstly utilized RDP4 (Martin et al., 2015) to detect recombination events among our dataset. Detection methods were manipulated including RDP\n \n \n 23\n \n \n , Chimaera\n \n \n 24\n \n \n , 3SEQ\n \n \n 25\n \n \n , GENECONV\n \n \n 26\n \n \n , and MaxChi\n \n \n 27\n \n \n . Furthermore, BootScan\n \n \n 28\n \n \n and SiScan\n \n \n 29\n \n \n were employed as secondary detection with highest acceptable\n \n p-\n \n value threshold of 0.05. Other parameters were carried out by default setting. The sequence would be excluded when three or more methods judge it as recombinant\n \n \n 30\n \n \n .\n
\n\n In the Nextstrain pipeline, we leveraged the maximum likelihood analyses to infer ancestral nodes and the phylogeny and the dispersal history of sublineage 8.7 implemented in the built-in python base framework TreeTime\n \n \n 31\n \n \n . Maximum likelihood ancestral nodes of discrete traits such as country or region of isolation allows identification of estimated transmission events given the sampled database, together with inferred probability distributions of ancestral state at each node. To be specific, distinct sample node colors indicate ancestral state and shifts are drawn as links between demes on the map\n \n \n 32\n \n \n . We firstly employed\n \n align\n \n augur command\n \n \n 20\n \n \n to match sequences into a qualified layout. Next, we employed the\n \n tree building\n \n augur command with built-in algorithm IQ-TREE 2\n \n 33\n \n with General Time-Reversible (GTR) model to build the preliminary raw Maximum Likelihood tree without any time and ancestral node annotation. Here we refined the raw tree with TreeTime via augur to infer a time-resolved phylogeny tree\n \n \n 31\n \n \n . Then we matched all the sequence spatiotemporal properties to the raw tree via a call to augur. Finally, we employed TreeTime again to jointly estimate the ancestral, dispersal track combined with phylogeny.\n
\n\n The magnitude and redundancy of the sublineage 8.7 dataset prohibits a fully Bayesian inference approach; hence, we followed a subsampling strategy that was deployed in HIV study to enable robust phylogeographic analyses\n \n \n 34\n \n \n . The subsampling therefore consisted of removing sequences such that monophyletic clusters that entirely consist of samples from the same geographical units are represented by a single sample. This is justified by the province-level geographic resolution of our data. Monophyletic clusters consisting solely of sequences from the same state do not bring any additional information on the between-state diffusion process we aim to infer. We discarded all but one randomly selected sequence per monophyletic cluster to provide a systematic way to significantly reduce the initial dataset, and kept only those sequences that pertained to our question of interest. In practice, this was done by first estimating a maximum-likelihood tree using FastTree 2.1 and removing potential hypermutants (or sequences with mislabeled sampling dates) by performing a root-to-tip regression using TempEst. We then constructed a new tree without the outliers, parsed the tree using the \u2018ape\u2019 R package to identify the state-specific clusters, and removed the redundant sequences from their corresponding clades. Since the main objective of this step was to greatly reduce the number of sequences, we did not take into account branch support values when selecting the clusters from which we subsampled. This also allowed us to avoid setting arbitrary threshold values in the clustering step. The resulting dataset consists of 1371 sequences from the original database of sublineage 8.7 (1782 sequences), which is practical for fully Bayesian phylodynamic inference (Supplementary table\n \n 1\n \n ).\n
\n\n Our aim was to run a phylogeographic model with a GLM extension in order to determine which factors were associated with virus movement between different locations in sublineage 8.7, we considered all possible explanatory predictors that were collected by the Chinese Bureau of Statistics. Identified predictors included climatological, ecological, physical (e.g., altitude), anthropogenic factors (e.g., gross population) and sample size. In Total, we collected a total of 20 potential predictors for the phylogeographic reconstruction of PRRSV within China in province-specific measures (Supplementary table 2\u20133). For all predictors, a separate origin and destination predictor is included. For a given variable, we wish to test the influence on a location when that location is treated as the origin or the destination of a viral transmission. This brings the total number of predictors to 40. Preliminary results of the analysis with all the predictors showed that only sample size had strong support and all other predictors had extremely low support, which suggested that the sample size predictors were obscuring all the other predictors. The goal of the sample size predictor was not to explicitly check if it was significant, but rather to serve as a sanity check. If other variables are found to be significant, we can assume they were not included in the model simply because of sampling bias since this is already explicitly accounted for by the sample size predictor. Hence, we separated these predictors as two independent analyses: 1) accommodating all predictors including sampling size; 2) accommodating all predictors except sampling size. For each geographical location, we calculated and visualized the distance between pairs of centroids (Supplementary Fig.\u00a08)\n \n \n 35\n \n \n .\n
\n\n In order to decrease computation time and to deal with the large number of separate locations in our dataset, we split the estimation of the phylogeny and the dispersal history into two separate analyses: we first ran a phylogenetic analysis in order to obtain a subset of around 1000 phylogenies, then ran an empirical tree distribution model using this subset of trees. This second analysis included the estimation of the geographic spread of the virus, as well as an GLM extension to test the influence of various parameters on this spread. Our overall goal was to reconstruct the detailed spread and origin of sublineage 8.7.\n
\n\n We generated this tree set using the HKY\u2009+\u2009\u0393\n \n 4\n \n and employing skygrid as the tree-generative process. We opted for an uncorrelated relaxed clock with an underlying lognormal distribution. We assumed a CTMC reference prior on the (mean) clock rate parameter. Hamiltonian Monte Carlo transition kernel with 82 parameters and 82 as the last transition point was employed to achieve efficient estimation of skygrid model parameters. We deployed an empirical tree using IQTREE\u2009+\u2009Treetime as a starting tree. distribution which required a total of 800\u00a0million iterations to achieve adequate statistics parameters. All molecular clock phylogenetic analyses were completed using BEAST v1.10 using the BEAGLE library v3.2.0 to improve computational speed\n \n \n 36\n \n ,\n \n 37\n \n \n . The geographic reconstruction on the empirical trees set was run for 30\u00a0million iterations. We combined at least two independent chains to confirm convergence at the same point and removed the first 10% of each chain as burn-in in each BEAST run. We computed maximum clade credibility (MCC) trees using TreeAnnotator. Tree visualizations were constructed using FigTree (\n \n \n http://tree.bio.ed.ac.uk/software/figtree/\n \n \n \n \n \n )\n \n , and geographical visualizations using SpreaD3\n \n 38\n \n .\n
\n\n To reconstruct the process of spatial dispersal, we modeled the immediate transition rates between distinct states through a continuous-time Markov chain (CTMC) approach. This framework involves representing movement between distinct geographic units (i.e., 33 provinces) using a K\u00d7K infinitesimal rate matrix, denoted as \u039b. Each element \u039b\n \n ij\n \n within this matrix signifies the instantaneous relative transition rate from location I to j. These transition rates were parameterized based on a set of potential explanatory predictors (P) within a Generalized Linear Model (GLM) structure\n \n \n 35\n \n \n . The relative transition rates (\u039b\n \n ij\n \n ) were determined by a log-linear function involving the P predictors. Each predictor, denoted by coefficient\n \n \u03b2\n \n \n \n p\n \n \n for p\u2009=\u20091, ..., P, quantified its contribution to \u039b. Additionally, an indicator variable\n \n \u03b4\n \n \n \n p\n \n \n determined the predictor\u2019s inclusion or exclusion in the model. To explore various combinations of predictors within the GLM, and to assess the relative impact of each predictor on the spatial spread of PRRSV sublineage 8.7, the posterior probability for the indicator variables through this process was calculated by Bayesian stochastic search variable selection (BSSVS). Bernoulli prior probability distributions were assigned to\n \n \u03b4\n \n \n \n p\n \n \n to ensure that 50% of the prior probability mass accounted for the absence of predictors. We assumed a priori that all coefficients were independent and normally distributed, with a mean of 0 and a standard deviation of 2. This approach enabled simultaneous evaluation of a potentially substantial number of predictors. The threshold of Bayes factor followed by that Bayes factor exceeding 150 indicates very strong support, a factor over 20 indicates strong support, and a factor beyond 3 indicates positive support for a predictor\u2019s influence. Variables selected for use in phylogeographic GLM are shown in Supplementary Table\u00a03. We deployed a strategy that encompasses Markov jump estimates of transition histories as done by previous study, averaging them over the entire Bayesian posterior\n \n \n 39\n \n \n . Our focus lies in studying the ancestral transition history of specific taxa in the phylogeny by summarizing Markov jump estimates as time-based trajectories including as a Sankey plot to depict posterior expected Markov jump estimates among all relevant provinces.\n
\n\n Firstly, we provided a full-genome overview of recombination patterns of lineage 8 intralineage and distinct interlineage respectively. Full-length sequences in our database were further employed and distinct reference lineage strains were incorporated as reference sequences (L1: NADC30, L2: XW008, L3: GM2, L4: EDRD-1, L5: VR-2332, L6: P129, L7: SP, L8: CH-1a, HP-PRRSV: JXA1, L9: MN30100) as well. We used Splitstree5 to assess the recombination events scheme of interlineage and intralineage with the genomic distances using a GTR mode and we visualized the splits as EqualAngle transformation using 1000 bootstrap replicates. The remaining parameters are referred to by default\n \n \n 40\n \n \n .\n
\n\n Second, we have calculated the high-frequency recombination regions to more accurately understand the recombination characteristic. As for interlineage recombination detection, we used RDP4 among our dataset with reference strains respectively using algorithms described before. Also, we have deployed Simplot v3.5.1 to further examine each independent recombination event, the result of which was considered as a verification to RDP. As for intralineage recombination, we constructed all the sublineage 8.7 strains to do a full exploratory recombination scan using all methods mentioned in the interlineage recombination. Both of each single recombination event with multiple off-springs were excluded for reducing repetition. All the parameters setting in this part were parallel to the former.\n
\n\n In terms of the HP-PRRSV vaccines used in China, there have been four legally approved vaccines including JXA1-R, TJM-F92, HuN4-F112, and GDr180 since the emergence of HP-PRRSV in 2006. We have secured them and obtained the full genomes using meta-transcriptome as described previously\n \n \n 41\n \n \n . Briefly, we constructed four libraries of vaccine sequences then sequenced on the MGISEQ-200 RS sequencer platform with pair-end of 150bp. Then we trimmed the adaptor of short reads by Trimmomatic\n \n \n 42\n \n \n and removed all low-quality reads (QC\u2009<\u200920). The refined reads were then assembled by MEGAHIT\n \n \n 43\n \n \n . The assembled contigs were mapped with nr database using DIAMOND. Samtools\n \n \n 44\n \n \n and iVar\n \n \n 45\n \n \n were finally run to obtain consensus sequences with criteria of sequencing depth\u2009>\u2009100 and minimum threshold 10 times, or to be written with N.\n
\n\n Together with the lineage 8 complete genome database, we have aligned the sequences using MAFFT7, then constructed the maximum likelihood tree using IQ-TREE2 based on the best-fit nucleotides substitution model of GTR\u2009+\u2009F\u2009+\u2009I\u2009+\u2009\u0393\n \n 4\n \n according to Bayesian Information Criterion and 1000 bootstrap of replicates to pick up vaccine homology clusters. For each selected cluster, we have built stochastic subsets from the PRRSV-2 complete genome database to re-estimate its homology with each vaccine isolate. These clusters were further called to build haplotype analysis using RdRp. We analyzed the concurrent amino acid mutation motifs existing in each cluster with disparity of ancestral strains by R v4.1.3 (ggtree and ggmsa packages)\n \n \n 46\n \n \n .\n
\n\n Here we quantified the distribution according to genotypic and spatiotemporal information in our genomic database (Fig.\n \n 1\n \n ). From the timeline, it was estimated that from 1994 to 1999, only a very few sequences were obtained, whereas an extremely pronounced surge presented from 2000 to 2005. Then lineage 8 strains have undergone rapid population expansion from 2006 to 2016, most of which were clustered into sublineage 8.7 and located in China. Besides, the proportions of paraphyletic clusters have steadily reduced except in the period of 2000\u20132005 when a robust prevalence was seen in the United States. Also, we have additionally added the geographical information detailed to provincial level in China on account of an unprecedented proportion of sublineage 8.7 in which China has possessed to better characterize the geographic distribution of lineage 8 isolates. Generally, in our dataset, the sequences included all provinces in China, and southern China (Guangdong) played a staple role in population amounts. However, it may be attributable to passive sampling or active sampling or the underlying impact of both.\n
\n\n The phylogenetic estimation of sublineage 8.7 indicated that after a short period of local prevalence of classical CH-1a-like isolates, the total population has transmitted toward the HP-PRRSV cluster with major genetic distance. In the classical CH-1a-like cluster, we could identify that sublineage 8.7 cluster have spread substantially in eastern China (i.e., Zhejiang, Fujian, Shandong, and Jiangsu) and Northern China (i.e., Beijing and Henan) in the early stage (Fig.\n \n 2\n \n ). Furthermore, with the transition of virus from classical cluster to HP-PRRSV cluster, South China has harbored more transmission, typically in Guangdong. Since then, HP-PRRSV have developed into an endemic cluster and were distributed over 30 provinces and autonomous regions of China and several countries in Asia, of which offsets were overbranch from Guangdong backbone, indicative of potential origin of these epizootics (Fig.\n \n 2\n \n ). Although Chinese strains have yet developed into geographically specific clades, isolates located in Guangdong were distributed in all main branches, indicative of that Guangdong played a crucial epicenter in the dispersal to all other parts.\n
\n\n As mentioned, we employed two independent phylogeographic models with including and excluding sample size which enables checking the pitfall of sampling bias to our model. In our time-resolved maximum clade credibility (MCC) tree excluding sampling size, we identified an origin of this lineage in Henan (Fig.\n \n 3\n \n a). Shortly after this estimated first occurrence, we identified the trunk of all major clusters as being from Guangdong except a single Zhejiang cluster branching off from Guangdong trunk in approximately 2000 (Fig.\n \n 3\n \n a). Subsequently, it is estimated that there were multiple introductions from Guangdong to other provinces, such as Hubei, Henan, Shandong, Jiangsu and Guangxi (Fig.\n \n 3\n \n a and Supplementary Fig.\u00a06a). In addition to examining node-based spatial transmission patterns, we have also undertaken a comprehensive analysis of dispersal dynamics focusing on overall sequences by incorporating pairwise Markov jumps (Fig.\n \n 3\n \n c). Guangdong exhibited an exceptional frequency of introductions to other destinations. Particularly, Guangxi, Shandong, Henan, Hubei, Jiangsu and Zhejiang are top-prioritized as recipients. Within these, Shandong Henan, and Jiangsu further acted as second hubs to significantly affect viral dissemination. As we took into account the sampling size in the GLM model, Guangxi exhibited a higher frequency of dissemination to provinces located in central and southern China, Guangdong in particular. Accordingly, the strong role of Jiangsu as a secondary hub to disseminate virus was greatly diminished compared with non-sampling size mode.\n
\n\n Another key element of this study is to assess and quantify the contribution of various demographic, epidemiological and mobility-related factors in shaping the dissemination of sublineage 8.7 in China. We considered and incorporated ecological, anthropocentric, economic, and geographical variables at provincial level, which may contribute to the process of viral spread using a discrete phylogeographic generalized linear model (GLM) approach. Before our\n \n bona fide\n \n GLM estimation, we firstly fitted a preliminary GLM analysis incorporating all variables with sampling size to check the credibility of other variables. Specifically, by doing this, we were not to explicitly check if the predictor of sampling size was significant, but rather to serve as a sanity check. If other variables were found to be significant, it was suggested not to include the corresponding variables in our final GLM model simply because of sampling bias since this is already explicitly accounted for by the sample size predictor. The results of our sanity check showed only sample size exhibited strong support (i.e., posterior estimates) whereas other predictive variables remained extremely unnoticeable, explaining that the sample bias would not absorb other variables in the GLM model (Supplementary Fig.\u00a07). As such, we accommodated all variables in our final GLM model. Figure\n \n 4\n \n showed the posterior estimates of the inclusion probabilities and conditional effect sizes (on a log scale) of the covariates to quantify the contribution of predictor variables to the among-province lineage transition rates. The distance matrix between region centroids was recovered as a strong predictor of lineage movement (inclusion probability\u2009>\u20090.7, Bayes factor [BF]\u2009>\u2009200), indicative of that virus lineage transmission occurred more frequently between nearby provinces, such as the strong Markov jump between Guangdong and Guangxi (Fig.\n \n 3\n \n c and Supplementary Fig.\u00a06c). In addition, we identified that demographic estimates including gross population at origin and rural pork consumption at destination supported the spread of PRRSV. However, such support was not observed in corresponding averaged data, i.e., population density and\n \n per capita\n \n meat consumption. Adding to the fact that several key variables in the swine industry including breeding stock, slaughter amounts were not associated with virus lineage movement, we assumed that the PRRSV transmission was highly correlated with integrated human activities and regionally (provincially) heterogenetic distribution of swine industry not limited to the activities of the process of intensive swine breeding system but rather the rural small-scale farming system.\n
\n\n We employed two different approaches to identify potential recombination events. On the one hand, we constructed a phylogenetic network to detect the recombination distribution among inter- and intra- of lineage 8 (Fig.\n \n 5\n \n a). Using the pairwise homology index of the neighbor-net method, we have identified an extremely significant recombination signal (p\u2009<\u20090.001). On the other hand, a more detailed investigation of the recombination pattern of inter- and intra-lineage 8 recombination revealed divergent recombination preference.\n
\n\n We tracked the recombinant history of lineage 8 with other lineages as well as intralineage and quantified them in the temporal dimension. Specifically, in terms of interlineage recombination, the first recombinant event can be traced back to 2007, with relatively fewer recombination events detected during 2007\u20132013. However, since 2014, the interlineage recombinant events have been detected exponentially, in which lineage 1 and lineage 3 contributed frequently as minor parents, particularly in 2014\u20132018 (Fig.\n \n 5\n \n b, d, and e). Accordingly, the region of ORF1ab and ORF3-ORF5 were the hottest regions of recombination in structural protein regions since 2010, which encodes GP3, GP4, GP5, as well as a series of non-structural proteins (nsp) (Fig.\n \n 5\n \n e). Specific to ORF1a, frequent recombination events were detected in the nsp2 region (40.0%), in which most of these events were totally associated with lineage 1 (91.7%) in 2014\u20132016. In ORF1b, these events focused on the nsp12 region (38.5%) contributed by lineage 1, 3, and 5. Comparatively, recombination distribution on structural regions was about 55% higher than that of non-structural regions, although the genomic length of non-structural regions was relatively longer than that of structural regions.\n
\n\n Intralineage recombination events were detected more frequently compared with interlineage recombination. Similarly, the onset of intralineage recombination dated back to 2007, with fewer events between 2008\u20132009. However, we observed a surge in recombination events during the period from 2010 to 2015, followed by significant fluctuations in the frequency of these events. Unlike interlineage, the hottest genomic region of intralineage recombination underlied in the nsp region, i.e., ORF1ab. Specifically, events in ORF1a showed a relatively uniform distribution regardless of the genomic length of a specific region. Comparatively, nsp9 were detected with higher frequencies in ORF1b. Except for ORF1ab, ORF4 exhibited a relatively high frequency among structural protein regions. Overall, both the interlineage recombination likelihood and the genomic hotspot region differed with the status in intralineage 8.\n
\n\n Several studies have claimed the PRRS MLV as a leaky status, whereas rare study has tested the hypothesis\n \n \n 12\n \n \n . We sequenced all HP-PRRSV-related vaccines approved for clinical use in China to obtain four HP-PRRS MLV complete genomes, i.e., JXA1-R, HuN4-F112, GDr180, and TJM-F92. Then, by deploying several evolutionary approaches as well as temporal analysis, we identified several amino acid \u201cleaky\u201d markers relating to HP-PRRS MLV.\n
\n\n We firstly applied an ML phylogeny on our complete genome dataset as well as the vaccine strains to identify corresponding monophyletic clusters within each vaccine strain i.e., vaccine clusters (Supplementary Fig.\u00a01). Then by using ClusterPicker, a total of 41 clinical strains associated with corresponding vaccines were selected with the fairly robust bootstrapping support (bootstrap value\u2009>\u200985%). Specifically, we constructed a haplotype using the nsp9 gene (encoding RdRp) to coalescently identify the homogeneous relationship between field isolates and vaccine strains. Except for the GDr180 cluster, field strains fell into clusters rooted by homogenous vaccine strains, suggestive of that these field strains were likely to be homogeneous with MLV vaccines (Fig.\n \n 6\n \n ). To be specific, in JXA1-R haplotype spectrum, serial passage vaccine candidates of JXA1 from JXA1/P10 to JXA1/P70 were observed with evident convergence to the progenitor node JXA1-R and several field variants were further branch off the JXA1-R node, indicative of conceivable hereditary relationship among the field strains and JXA1-R. Of particular interest, NT2/2015 isolate, a reported reversion case of JXA1-R, were also embedded as a relative pivot node of evolutionarily ancestral relationship. Strikingly, a single strain located in Vietnam was identified in JXA1-R cluster, with a key position of branches stretched out subsequently. Besides, MLV vaccines TJM-F92 and HuN4-F112 were also embedded in a more key position, and that indicates a more key hub of viral dissemination among TJM F92 and HuN4-F112 related field strains. As a counterexample, GDr180, the latest approved vaccine in 2015 with less market share, shared with less homogeneous relationship with field strains in that all strains in this cluster were embedded at the terminal of our haplotype, which suggested a less likely homogeneous evolutionary relationship (Fig.\n \n 6\n \n A). We further analyzed its connection from the perspective of temporal signal (Fig.\n \n 6\n \n B). In JXA1, TJM, HuN4 clusters, yearly reported cases of field strains remained zero until these vaccines were approved for clinical use (i.e., 2011). Specifically, since the MLVs (including JXA1-R, TJM-F92, and HuN4-F112) were widely used, each vaccine cluster has increased remarkably for six years from 2011 to 2017, during which the new lineage (lineage 1) has been introduced into Asia. The sharp rise of the clinical strains in six continuous years reflects the clinical impact of MLV vaccination. Cases declined since 2017 as the dominating lineage jumped from lineage 8 to lineage 1 in China. In the GDr180 cluster, however, we observed an abnormal surge from 2006 to 2009, during which the GDr180 was not developed yet. In fact, the GDr180 was not developed until 2015, and we did not identify any strain related to the GDr180 cluster in such a short period.\n
\n\n We further identified the potential amino acid markers associated to MLV reversion cases clinically followed by the algorithm: i. The amino acid site substituted between parental strain and MLV strain, such as JXA1 and JXA1-R; ii. the potential amino acid site mutated consistently in the field strains (at least 50%); iii. The amino acid mutation sites in the field strains are consistent with those in MLV strains. In terms of the above analyses, the GDr180 cluster was further excluded. Specifically, TJM-F92 cluster isolates have been identified with 35 concurrent amino acid mutations distributed on ORF1ab, ORF3, and ORF5 (Supplementary Fig.\u00a02); JXA1-R have been identified with 32 concurrent amino acid mutations distributed among ORF1ab, ORF2, ORF3, ORF4, and ORF5 (Supplementary Fig.\u00a03); HuN4-F112 cluster isolates have been identified with 13 concurrent amino acid mutations distributed among ORF1ab, ORF2, and ORF5 (Supplementary Fig.\u00a04). We identified that JXA1-R and HuN4-F112 clusters shared an identical amino acid substitution (JXA1:F250S\u3001HuN4: T250I) in ORF2. In analogy to this pattern, JXA1-R and TJM-F92 clusters shared T225A in ORF3 and an identical amino acid substitution in ORF1a (JXA1-R: E957G TJM-F92:T957S).\n
\n\n Despite a rapid increase in the number of sublineage 8.7 virus infections in Asian countries over recent years, very little was known about the patterns of virus emergence and spread. Relying on a national long-term PRRSV surveillance project, we collected over 6,000 suspected positive samples and obtained 242 newly ORF5 sequences and 42 complete genomes belonging to sublineage 8.7 in approximately two decades and integrated them with public genomic data to form the largest collection of available PRRSV sublineage 8.7 sequences to answer the question of how sublineage 8.7 emerged, evolved, transmitted, and recombined (intra- and interlineage) in the nearly two decades\n \n \n 3\n \n ,\n \n 4\n \n ,\n \n 41\n \n ,\n \n 47\n \n \n . More pragmatically, we confirmed the high weight of rural swine activities and provincial distance contributing to the sublineage 8.7 spatial spread. Note that several HP-PRRSV MLVs were hastily approved for use on a remarkably emergency basis in China at that time whereas few studies focused on its potential impact on the field spanning a long temporal vaccination, we further sequenced all HP-PRRSV MLVs. As such, we found strong leaky evidence of HP-PRRSV MLVs based on a multivariate perspective, which may have restored its virulence in clinical farming.\n
\n\n In our nextstrain analysis of the total sublineage 8.7 clusters, although several offsets were detected in the USA and Russia, nearly the whole phylogeny trunk was located in Asia, suggesting sporadic transmission events from China to other countries and without any outbreak events identified in other regions. In addition, the classical sublineage 8.7 cluster and transition cluster have exhibited, with absolute resolution, longer branch length, indicative of comprehensive genetic divergence than the subsequently emerging HP-PRRSV cluster (Fig.\n \n 2\n \n ). Nonetheless, it also explained that a huge mutation spectrum presented prior to the emergence of sublineage 8.7 HP-PRRSV, termed by \u201ctransition lineage\u201d, suggesting the virus under greater host innate immune pressure and adaptative evolution during the early invasion period. This observation coincides with a study suggesting that the ongoing convergence of SARS-CoV-2 lineages includes multiple mutations that can enhance the persistence of diverse virus lineages during host immune recognition\n \n \n 48\n \n \n . In the dispersal history, the nextstrain result formulated a major hypothesis on how PRRSV sublineage 8.7 may be maintained in strict transmission foci. The dissemination pattern of sublineage 8.7 constitutes a connected network of Asian regions; that is, South China serves as a principal province of PRRSV maintaining and spreading not only for the Chinese population but also for the other neighboring Asian regions such as Vietnam and Thailand. As such, Thailand and Vietnam could act as second hubs of spreading sources and diffusing viral sources to neighboring countries (Laos and Cambodia). We deployed a subsampling approach in which the sampling counts of each specific region were strictly limited up to 60 to re-estimate sublineage 8.7 phylogeographic analysis to test this hypothesis, which supported the major results, indicative of the likely accurate estimation of regional spread in Southeast Asia (Supplementary Fig.\u00a05). As well, the epicenter role of Guangdong was also corroborated by our GLM analyses in China. In our ensuing Bayesian discrete phylogeographic result, we accurately estimated the early transmission link from Guangdong to nearby provinces (Guangxi) and central China, such as Henan and Hubei, with strong Markov jump support. Similarly, He et al also has proved the epicenter role of Guangdong with another important porcine virus i.e., PEDV using Bayesian discrete analysis, with less weight compared with that in PRRSV\n \n \n 17\n \n \n . This study also successfully linked the swine industry trade and pork consumption with PEDV spread in China in their GLM extensions. In our GLM model, we found strong support for provincial distance as well as demographic factors such as population amount at origin, pork sale at rural, and per capita disposable income at destination to PRRSV spread in China. We estimated this variation may be attributable to the host infectivity heterogeneity (PEDV: piglets; PRRSV: boar and pregnant sow) and different transmission capabilities between PRRSV and PEDV.\n
\n\n Recombination occurred with ubiquity as a result of virulence enhancement, host shifting, and adaptability strengthening. Likewise, PRRSV recombination is also significant and pervasive in that it largely enhances genetic diversities and reduces the cross-protection of vaccines. In this study, we systematically analyzed the intra- and interlineage recombination of PRRSV lineage 8 with its temporal dynamics and found a principal recombination wave spanning 2014 to 2016. As we know, frequent homogeneous RNA viral recombination is the major result of random template conversion during replication and is thought to be deployed by the \u201ccopy-choice\u201d mechanism of RdRp. In such a long-term evolution, any probability distributions may turn to be tendency events if such stochastic recombination could be conducive to viral survival. Although high-level recombination existed among intra- and interlineage, we found that interlineage recombination was more targeted to structural protein regions (GP3-GP5), whereas intralineage recombination was more concentrated on non-structural protein regions (ORF1a), with specifically the case for breakpoint at nsp2-nsp5, which mainly involved into antagonizing with host innate immune systems such as deubiquitin, IFN antagonist and membrane modification\n \n \n 1\n \n \n . Besides, such a great difference among the number of inter- and intralineage recombinations may be involved with the flush vaccination of lineage 8 MLVs. Until now, lineage 8 possessed the largest amounts of approved MLV vaccines of PRRSV in China. Since all PRRSV MLV could continue to replicate in the host, the \u201ccopy-choice\u201d characteristic of RNA polymerase offers a possibility to recombine with field strains in host. On the other hand, it should be noted that China currently possesses only L5 lineage vaccines, derived from the VR2332 lineage, as well as L8 lineage vaccine strains. The extensive use of L8 lineage vaccines significantly outweighs that of L5 lineage vaccines, thereby elevating the probability of genetic recombination occurrences. Hence, lineage 8 MLV vaccines may gain more possibilities to recombine with field strains. However, it is under an intricate field that we may not interpret, with unilateralism, the erratic phenomenon just in terms of one aspect.\n
\n\n Our study provided the first exploration of quantifying how the MLVs are likely to affect immunized herds under the field. By multiple independent phylogenetic reconstructions and recombination elimination, we have identified four MLV groups characterized with evolutionarily homogeneity. We inspected the temporal signal of potential descendants within each group. Each strain in the JXA1-R, TJM-F92, and HuN4-F112 groups coincidentally supported our scenario whose prerequisite was that the time of vaccines approved is prior to the prevalence of associated field isolates. However, the temporal signal and the haplotype analysis of GDr180 cluster was on the contrary, showing temporal irrelevance between GDr180 and field isolates. We hypothesized that it is partly due to that GDr180 is the latest approved MLV vaccines (2015), as such, it has been vaccinated with relatively low frequency in the field. It is less impacted then in the clinic and would still continue to be monitored. In the other three HP-PRRSV MLV vaccines, JXA1-R, the most frequently administered HP-PRRSV vaccine and mandatory immunized before 2017 in China, were also the vaccines that pose relevance with the most numerous field strains. It, in turn, reflected the impact that MLV vaccines brought to the field, and vice versa. Given that the JXA1-R-homogeneity strain, KU842720/Hanvet1/Vietnam, was detected under the context of without approved JXA1-R vaccine importing in Vietnam, the consistently spatial spreading scale with our estimation of sublineage 8.7 national transmission further highlighted the significance of continuous monitoring and restrict quarantine underlying cross-regional livestock trading. Although multiple approaches such as infectious clones and challenge experiments have been attempted previously, the common characteristic of these results related to reversion sites was merely appropriate for specific cases whereas we are still in the paucity of amino acid markers from MLV supported by comprehensive clinical whole genome data. Our results firstly showed several common amino acid substitution positions spanning whole-genome scale, which may be associated with HP-PRRSV MLV reversion markers albeit specific molecular markers varied with different vaccine clusters. These results should be invaluable hints and be helpful when it comes to potential vaccine reversion cases and potential vaccine escape mutants and other potentially problematic variants.\n
\n\n In summary, we constituted the largest dataset to reconstruct sublineage 8.7 spatial dynamics, its associated ecological, demographic, and swine-farming practices implication, and potential leaky evidence of HP-PRRSV MLVs. Given that the phylogeographic studies in temporospatial dynamics of porcine-associated infectious diseases increase exponentially, further studies that integrate the spread patterns of different pathogens and investigate how and why they may vary against different objects such as PEDV and PRRSV are invaluable for effective targeted control of swine pathogens. Importantly, as PRRSV and SARS-CoV-2 are two important members of\n \n Nidovirales\n \n order, the long-term clinical data of PRRSV MLVs immunization can be a good reference for SARS-CoV-2 vaccine safety evaluation and future RNA virus MLV development.\n
\n\n \n Data availability\n \n
\n\n The 242 new ORF5 sequences, 42 new complete genomes, and four vaccine sequences have been deposited in the China National GeneBank DataBase (CNGBdb) under the accession number CNP0004735. Any remaining data could be acquired in the supplement files or requested from the corresponding authors. All anonymized data, code, and analysis files are available on GitHub: https://github.com/jobsxing/GLM-LINEAGE8.7.\n
\n\n \n Acknowledgment\n \n
\n\n National Natural Science Foundation of China [grant number 32102704], GB acknowledges support from the Research Foundation - Flanders (\u201cFonds voor Wetenschappelijk Onderzoek - Vlaanderen,\u201d G0E1420N, G098321N) and from the Internal Funds KU Leuven (Grant No. C14/18/094). This study is supported by the Key-Area Research and Development Program of Guangdong Province [grant number 2019B020211003], China Agriculture Research System of MOF and MARA.\n
\n\n We would gratefully thank all the researchers and laboratories for their generous genomic uploaded data in NCBI that we have used in this study. Furthermore, we would thank Xiaoqin Xu, Yuli Luo, and Qian Kuang for the large-scale sampling and routine monitoring relying on national surveillance of PRRSV in China.\n
\n\n dataset 1-3\n
\n \n\n Supplementary figures\n
\n \n\n Advances in proteomics and mass spectrometry enable the study of limited cell populations, where high-mass accuracy instruments are typically required. While triple quadrupoles offer fast and sensitive low-mass accuracy measurements, these instruments are effectively restricted to targeted proteomics. Linear ion traps (LITs) offer a versatile, cost-effective alternative capable of both targeted and global proteomics. Here, we describe a workflow using a new hybrid quadrupole-LIT instrument that rapidly develops targeted proteomics assays from global data-independent acquisition (DIA) measurements without needing high-mass accuracy. Using an automated software approach for scheduling parallel reaction monitoring assays (PRM), we show consistent quantification across three orders of magnitude in a matched-matrix background. We demonstrate measuring low-level proteins such as transcription factors and cytokines with quantitative linearity below two orders of magnitude in a 1 ng background proteome without requiring stable isotope-labeled standards. From a 1 ng sample, we found clear consistency between proteins in subsets of CD4\n \n +\n \n and CD8\n \n +\n \n T cells measured using high dimensional flow cytometry and LIT-based proteomics. Based on these results, we believe hybrid quadrupole-LIT instruments represent an economical solution to democratizing mass spectrometry in a wide variety of laboratory settings.\n
\n\n Systems biology is the study of interactions within and between cells, where the goal is to learn how those interactions give rise to the complex behavior seen in an entire system.\n \n 1\n \n One challenge is that many complex biological processes, such as adaptive immunity, are built from small populations of distinct cell types acting in concert.\n \n 2,3\n \n Improvements in proteomics methods and mass spectrometry (MS) instrumentation have paved the way for low-input and single-cell proteomics, which make it possible to study how limited cell populations contribute to the whole. While the majority of single-cell methods use tandem mass tags (TMT)\n \n 4\n \n to increase signal (and thus consistency) with data-dependent acquisition (DDA),\n \n 5,6\n \n several groups have demonstrated that data-independent acquisition (DIA) is an effective solution to measuring low-input samples.\n \n 7\u20139\n \n However, high-mass accuracy instruments are required in nearly all cases.\n
\n\n While single-cell and low-input global proteomics is typically acquired using high-mass accuracy instruments, nominal-mass instruments, such as triple quadrupoles, lead in quantitative sensitivity using targeted selected reaction monitoring (SRM).\n \n 10\n \n With SRM, peptides are detected based on monitoring multiple fragment ion signals produced by each selected precursor ion. Transitions (diagnostic precursor/fragment ion pairs) in a pre-specified schedule must be provided to the instrument for monitoring at specific times within the chromatographic gradient.\n \n 11\n \n While triple quadrupoles are extremely quick instruments capable of rapidly switching between ion pairs, they can only monitor a single\n \n m/z\n \n at a time. As such, triple quadrupoles are limited to targeted experiments, which require a high-mass resolution instrument to select and schedule targeted peptides and transitions before migrating to a nominal-mass instrument for high-throughput monitoring.\n
\n\n An alternative targeted method to SRM is parallel reaction monitoring (PRM), which uses a quadrupole-equipped high-resolution mass spectrometer where the third quadrupole is replaced with an Orbitrap\u2122 (Q-Orbitrap, also known as a Q-Exactive\u2122) or a time-of-flight analyzer (Q-ToF). Rather than measure precursor/fragment transitions, all precursor-specific fragment ions are collected in a full tandem mass spectrum with PRM.\n \n 12\n \n A major advantage of PRM is that diagnostic fragment ions are selected after the experiment is performed, which can vastly simplify the assay development process. PRM has provided meaningful biological insight into several diseases, including systemic autoimmune diseases,\n \n 13\n \n multiple sclerosis,\n \n 14\n \n and colorectal cancer.\n \n 15\n \n When coupled with global proteomics, PRM is a powerful tool for interrogating system-wide interactions between cells.\n
\n\n Linear ion traps (LITs) are another versatile, fast, nominal-mass analyzer comparable in resolution and complexity to triple quadrupoles. Modern Thermo Scientific\u2122 Tribrid\u2122 instruments have incorporated LITs as a tertiary analyzer, coupled with an Orbitrap.\n \n 16\n \n Using a Tribrid instrument, Heil et al\n \n 17\n \n showed that the benefit of PRM lies within its ability to monitor multiple product ions produced within a selected precursor\n \n m/z\n \n range and that the LIT in Tribrids was an effective readout for targeted proteomics. A LIT measures ions trapped in an electric field by adjusting RF and DC voltages to selectively eject ions based on their\n \n m/z\n \n to collect MS/MS spectra. Unlike triple quadrupoles, which have to \u201cdwell\u201d at each increment of m/z to form a spectrum, LITs acquire full scan MSn data quickly and sensitively,\n \n 18\n \n making them also viable for global proteomics using DDA or DIA.\n \n 19\n \n As a result, a hybrid quadrupole-LIT (Q-LIT) could act as an \u201call-in-one\u201d nominal-mass instrument capable of both targeted and global proteomics.\n
\n\n As with triple quadrupoles, LITs are extremely sensitive, ion-efficient mass analyzers apt for low-input proteomics.\n \n 20\n \n In some circumstances, LITs can be more effective than high-resolution mass analyzers for low-input samples (\u2264\u200910 ng)\n \n 21\n \n and can measure single cells without multiplexing reagents.\n \n 22\n \n At higher sample input (\u2265\u2009100 ng), the lack of high mass accuracy overshadows the increased sensitivity of LITs. There exist other compelling reasons to consider LIT-based instruments in high-throughput applications. In particular, LITs operate at high pressure (10\n \n \u2212\u20093\n \n mTorr) in comparison to ToF analyzers (10\n \n \u2212\u20096\n \n mTorr), where ions have to travel uninterrupted for meters, or Orbitrap analyzers (10\n \n \u2212\u200910\n \n mTorr), where ions can travel for more than a kilometer. Lower vacuum pump requirements allow LITs to be built more affordably, robustly, and housed in smaller instrument footprints.\n
\n\n Here, we present a workflow using a hybrid quadrupole-LIT (Q-LIT) instrument from Thermo Scientific as a single instrument for rapidly generating targeted assays for low-input experiments. With the Q-LIT, we demonstrate how to build nominal-mass targeted transition libraries using both DDA and gas-phase fractionated (GPF) DIA libraries. We then show the quantitative accuracy of targeted PRMs with a Q-LIT using matched-matrix calibration curves collected with 1, 10, and 100 ng total protein to model low abundant immune cell populations. To facilitate this, we developed an open-source software tool that directly schedule-optimized PRM assays from DDA and DIA libraries. Finally, we show quantitative consistency measuring low-level biological targets in cytokine-stimulated CD4\n \n +\n \n and CD8\n \n +\n \n T cells with as little as 1 ng on column. These results suggest that Q-LITs can perform as inexpensive stand-alone instruments for quantitative proteomics, capable of a wide range of measurements without needing high-resolution mass spectrometry.\n
\n\n Linear ion traps (LITs) are robust, sensitive, and fast mass analyzers, yet these instruments have limited mass resolution. Previously, our lab demonstrated that LITs could be used effectively as stand-alone mass analyzers to measure low-input samples using an Orbitrap Eclipse\u2122 Tribrid mass spectrometer.\n \n 22\n \n In that work, we detected approximately 400 proteins from single cells using data-independent acquisition coupled with chromatogram libraries to help make detections.\n \n 23\n \n While our Eclipse instrument configuration ignored the high-resolution Orbitrap mass analyzer, we performed those experiments in the context of a high-end Tribrid instrument. Furthermore, the 400 proteins we measured were the easiest to observe but not necessarily the most biologically useful to monitor. While reduced representation approaches\n \n 24,25\n \n that quantify a limited panel of easily observed proteins can help infer biological states, significant hurdles must be overcome to predict the expression patterns of unmeasured proteins. As such, directly measuring panels of proteins of interest in low-input samples using targeted proteomics may be preferable to global proteomics.\n
\n\n In this work, we sought to answer three remaining questions. First, by eliminating the Orbitrap, could an affordable quadrupole-LIT (Q-LIT) mass spectrometer perform at a high level as a stand-alone instrument for both library generation and targeted proteomics measurement? Second, can a Q-LIT mass spectrometer quantify peptides at and below the level of single cells? Third, can quantitative experiments measure low-level biologically relevant proteins like cytokines and transcription factors at or below 1 ng? To this end, we assessed several parameters of the Stellar\u2122 mass spectrometer, a new hybrid Q-LIT design produced by Thermo Scientific. First, we tested proteome-wide library generation; then, we assessed quantitative linearity using targeted PRM experiments with 100, 10, and 1 ng sample inputs. Finally, we tested sensitivity and measurement consistency in a biological context.\n
\n\n \n A Q-LIT workflow for generating PRM assays using DDA and DIA data\n \n
\n\n The Stellar MS is a hybrid Q-LIT mass spectrometer with improved ion transmission features capable of performing rapid scans up to 200 kDa/s (\n \n Figure 1A\n \n ). The instrument shares many of the same design components as existing Orbitrap-based instruments.\n \n 26\u201328\n \n Tribrid instruments accumulate fragment ions in the collision cell (also known as the ion routing multipole, IRM) before transfer to the mass analyzer. Analogously, the Stellar accumulates fragment ions in the collision cell (Q2, also known as the ion concentrating routing multipole, ICRM) before transfer to the LIT for mass analysis. These arrangements produce high scanning speeds by performing fragment accumulation in parallel with mass analysis in the low-pressure LIT.\n \n 29,30\n \n
\n\n An advantage of the Q-LIT geometry is that it is suitable for both global discovery proteomics as well as targeted proteomics. To leverage this, we implemented a workflow to generate high-quality peptide libraries using off-line fractionated DDA or GPF-DIA, and software to build on-the-fly PRM assays for the same instrument (\n \n Figure 1B\n \n ). Briefly, the software takes a DDA-based spectrum library or a DIA-based chromatogram library as input and compares it with potential targeted proteins. The target list contains a list of critical accession numbers along with other optionally desired entries in a selected FASTA database. The assay can be modified using both a peptide inclusion and exclusion list. Assays can be adjusted depending on instrument settings, where the maximum assay density and a retention time scheduling window width must be selected. A recent single-injection DIA run also ensures that the retention time schedule matches the current LC column conditions.\n
\n\n Peptides are selected using this software tool based on the highest signals recorded in the library. For DDA, this signal is based on precursor intensity if available. For GPF-DIA, this signal is based on the intensity of the third largest fragment ion per peptide following common SRM/PRM conventions of requiring at least three transitions.\n \n 29\n \n The algorithm chooses peptides using a greedy approach, where the most abundant peptides are scheduled first. After the algorithm chooses a specified number of peptides for a given protein (typically 3-5), no additional peptides from that protein are considered. Additionally, peptides cannot be added to a retention time region if any time point in that region has already reached the maximum assay density. Once the algorithm iterates through all possible peptides, the software tool produces a scheduling report and a target inclusion list for the Thermo method editor. While the tool focuses on simplifying scheduling for Thermo instruments, it is analyzer and vendor agnostic, supporting scheduling for both Orbitrap and ToF instruments. This software workflow has an accessible graphical user interface built into the EncyclopeDIA code base (see\n \n Supplemental Note\n \n for more details).\n
\n\n \n Developing a comprehensive target library\n \n
\n\n PRM assays are commonly generated from various sources, including public repositories that store targeted proteomics data such as the PeptideAtlas,\n \n 31\n \n CPTAC,\n \n 32\n \n or Panorama.\n \n 33\n \n Additionally, assays can be built from global data, with targets selected from empirical measurements on the biological matrix of interest. For this work, we wanted to use methods that could be fully acquired on the Q-LIT but still be capable of detecting low-abundant peptides. One advantage of this approach is that targets are tuned for the instrument from the context of retention time scheduling and optimal transition selection.\n
\n\n To generate a low-input PRM assay on the Q-LIT, we tested two standard methods of building libraries: a chromatogram library using GPF-DIA and a spectral library using fractionated DDA. We collected libraries from a pool of IL-2 and IL-15-stimulated T cell proteomes. To build the chromatogram library, 6x gas-phase fractions were used with 2\n \n m/z\n \n wide isolation windows across mass ranges of 100\n \n m/z\n \n per injection. Since the background proteome matrix is not chemically altered or diluted, this approach produces retention times that closely match the quantitative PRM experiments. In contrast, we offline fractionated the DDA library samples using high-pH reverse phase separations to yield a total of 10 fractions, which were analyzed in separate injections. Consequently, each fraction has a simplified matrix background, which may not reflect retention times as consistently in unfractionated quantitative samples. The DDA and DIA methods produced libraries that were similar in size but of surprisingly distinct populations of peptides (\n \n Figure 2A\n \n ), presumably due to the different fractionation methods and matrix backgrounds used to generate each library.\n
\n\n In addition to producing slightly more peptide detections, peptide-centric extraction\n \n 34\n \n of DIA datasets is more akin to fragment-level quantification using targeted methods than DDA measurements.\n \n 35\n \n As such, peptides detected using GPF-DIA are more likely to produce robust, targeted assays since the mode of discovery uses similar methodologies to the final quantitative measurements. However, some sample types, such as enriched phosphopeptides, may be better suited to library generation with DDA since they can take advantage of stochastic sampling to detect more peptides and proteins over technical replicates.\n \n 36\n \n For this work, we chose to proceed with the GPF-DIA library for assay development, but the scheduling software produced for this work functions with either library source.\n
\n\n For DIA injections, search results from EncyclopeDIA and CHIMERYS were combined for downstream work (\n \n Figure 2B\n \n ). CHIMERYS is a spectrum-centric search engine that builds on INFERYS to provide spectra and retention-time predictions for peptides in a given FASTA database.\n \n 37\n \n In comparison, we searched a Prosit-predicted spectral library\n \n 38,39\n \n with the peptide-centric search engine, EncyclopeDIA, which was adapted for analyzing ion trap data. Consequently, EncyclopeDIA was limited to searching +2 and +3 peptides to maintain a reasonable search space, while CHIMERYS was configured to consider modifications and higher charge states as well. More peptides were detected from CHIMERYS compared to EncyclopeDIA in each gas-phase fraction (\n \n Supplemental Figure S1A\n \n ), but considering the superset of detections increased the total number of potential targets (\n \n Supplemental Figure S1B\n \n ) and both search engines produced an equal number of viable peptide targets that could be used in downstream PRM experiments (\n \n Supplemental Figure S1C\n \n ). In all cases, the retention times from CHIMERYS-detected peptides were re-peak picked using EncyclopeDIA to identify candidate target transitions for PRM measurement in a combined DIA library.\n
\n\n \n Assessing Q-LIT PRM quantitative accuracy at low input\n \n
\n\n With low-input global proteomics, we preferentially measure only the most abundant proteins. We stress-tested the quantitative accuracy of the Q-LIT system using PRMs by measuring biologically relevant proteins that tend to occur at a range of levels in the proteome. To accomplish this, we first functionally annotated candidate peptides in the combined DIA library using the PANTHER database.\n \n 40\n \n We selected target proteins based on GO-terms and Reactome pathways for T cell differentiation, immune biology, T cell activation, cytokines, and transcription factors with a focus on choosing proteins associated with the dynamics of memory T cells. Using the PRM scheduling algorithm, we constructed three assays using the same bank of proteins, where each assay was suited to a different input level: up to 50 peptides/cycle for 100 ng of material, 20 peptides/cycle for 10 ng of material, and 10 peptides/cycle for 1 ng of material. Ultimately, the 100 ng assay monitored 481 peptides, the 10 ng assay monitored 151, and the 1 ng assay monitored 61. To maintain a 2-second cycle time using 1 ng of material, the maximum ion injection time (maxIIT) was set to 200 ms. Similarly, at 10 ng of material, the maxIIT was set to 95 ms (slightly below 100 ms to accommodate the additional time required to route ions in the mass spectrometer). At 100 ng of material, the ion injection time was set to 50 ms; however, each scan rarely met that length of time.\n
\n\n We performed matrix-matched calibration curves\n \n 41\n \n at 100 ng, 10 ng, and 1 ng levels to assess the quantitative accuracy of the Q-LIT over several orders of magnitude. Dilutions in a buffer background are useful to assess instrument sensitivity, but because background noise decreases at the same rate as target peptides, quantitative linearity will always appear to be more accurate than in a real background matrix. Matrix-matched calibration curves are more effective at assessing linearity in real-world scenarios since the background signal does not change with dilution. To accomplish this, we had to build a suitable background matrix of similar composition to our target T cell proteome. Our approach used dimethyl labeling to modify the foreground T cell proteome, which kept the same composition while also producing different precursor and fragment masses. Dimethyl labeling was first introduced as a multiplexing method where multiple samples would be labeled and mixed prior to mass spectrometry.\n \n 42\n \n In our approach, only the background is modified, where free amines are mass-shifted by two methyl groups (28 Da). This shifts any labeled precursors (even incomplete reactions with a single methyl group) outside of the precursor isolation window used by PRM measurements, ensuring that any foreground signals will not be confused with background signals. Additionally, dimethyl labeling is affordable, easy, and quick, as peptides are labeled to 99.9% completion within a 1-hour reaction.\n
\n\n While ion traps generally have a more limited dynamic range compared to Orbitrap-based mass spectrometers, the sensitivity of ion traps allows for superior detection and quantification of low-input samples.\n \n 17\n \n Additionally, Orbitraps have slower scanning speeds compared to the Q-LIT, limiting the number of peptides that can be targeted within a cycle. Here, we found that reasonable quantitative accuracy can be achieved with the Q-LIT at low input while targeting a similar number of peptides as with higher-input Orbitrap-based PRM assays. At 100 ng, the quantitative accuracy of most peptides acquired with PRM remains consistent for nearly two orders of magnitude (\n \n Figure 3A\n \n ), where the median lower limit of detection (LoD) was 0.83:100 and the median lower limit of quantification (LoQ) was 2.8:100 (\n \n Figure 3B\n \n ), where only 0.6% of peptides could not be assigned a LoQ. Quantification was slightly worse at the 10 and 1 ng levels, where 4.6% and 20% of peptides could not be assigned a LoQ. Unsurprisingly, at the 100 ng level signal is more easily distinguishable from noise and the LoD distribution is generally higher than at 10 ng or 1 ng (\n \n Supplementary Data\n \n ).\n
\n\n Single cells typically produce between 0.1 and 0.3 ng of peptides, depending on the cell type. Considering the 1 ng sample, the median measured peptide produced a linear signal in this range (0.198:1). Several peptides showed a linear response below 0.1 ng. For example, the peptide ECESYFK from Granzyme B was found to have a LoQ of 0.043:1, equating to a proteome fraction consisting of 43 pg in a background of 1 ng, and was still measurable above background at the 18 pg level (\n \n Figure 4A and 4B\n \n ). Two other Granzyme B peptides, VAAGIVSYGYK and TQQVIPMVK, produced even lower LoDs (below 10 pg equivalents). Granzyme B is a serine protease implicated in multiple autoimmune diseases.\n \n 43\n \n All told, 61 peptides with estimated LoQs in the 1 ng assay corresponded to 30 quantified proteins. At least 6 points across the peak were sampled for all peptides in the assay achieving 8-10 points across the peak on average.\n
\n\n \n Validated cell populations for quantitative testing\n \n
\n\n In addition to showing quantitative accuracy in a controlled matched matrix, we wanted to validate measurement precision in low-input biological experiments. The interleukins (IL) family of proteins is a class of cytokines expressed by many cells, including immune cells, which bind to specific receptors that elicit pro- and anti-inflammatory roles.\n \n 44\n \n Certain cytokines, such as IL-2 and IL-15, bind to receptors on the surface of T cells in specific biological events, such as activation and differentiation. Both of these molecules have been successfully used as part of immunotherapies to combat cancer.\n \n 45\u201348\n \n Interestingly, IL-2 and IL-15 are structurally similar in homology and activate T cells through the same receptor subunits (IL-2/IL-15R\u03b2\u03b3),\n \n 49,50\n \n mediating largely similar biological effects on T cells.\n \n 51,52\n \n However, possibly related to expression of the private IL-2R\u03b1 and IL-15R\u03b1 chains, IL-2 induces an effector-like phenotype (with low CD62L expression) while IL-15 induces a memory-like phenotype (with higher CD62L).\n \n 53\n \n We generated activated T cells cultured in IL-2 or IL-15 to replicate an effector-like and memory-like phenotype for CD4\n \n +\n \n and CD8\n \n +\n \n cells (\n \n Supplemental Figure S2A\n \n ). We selected this model system to showcase the ability to generate LIT-PRM assays using well-studied biology at inputs below 1 ng. Additionally, flow cytometry was used as an orthogonal technique to validate the cell populations present in IL-2 and IL-15 treated T cells on days 5, 6, and 10 exhibited an effector-like and memory-like phenotype (\n \n Supplemental Figure S2B and S2C\n \n ).\n
\n\n At day 10, flow cytometry identified that each culture was predominantly composed of T cells, with CD8\n \n +\n \n T cells being the majority subset in both IL-2 (83.8%) and IL-15 (92.2%) cultures (\n \n Figure 5\n \n ). Correspondingly, we found that CD4\n \n +\n \n T cells composed 14.6% of the cells stimulated with IL-2 and 6.9% of cells stimulated by IL-15. This was reflected in our targeted proteomics data, as the CD4 protein was the second most downregulated protein in IL-15-stimulated T cells compared to IL-2-stimulated cells (\n \n Figure 6A\n \n ). We note that this protein was not technically quantified at 1 ng, as the one peptide for CD4 (VVQVVAPETGLWQCLLSEGDKVK) lacked linearity in signal as estimated by the calibration curve. We measured the same peptide at the 10 ng level, where we calculated the LoD to be 0.96:10 (ratio of foreground to background) with an LoQ of 8.3:10 (\n \n Supplemental Figure S3\n \n ). This indicated that at the 1 ng level, CD4 should be above the LoD but below the LoQ, and our results match these calculations.\n
\n\n IL-2 stimulation is known to push activated T cells into an effector-like population, reflected by the paired flow cytometry data on day 10. Granzyme B, from which we estimated the most responsive peptide (TQQVIPMVK) was quantitative to 0.025:1 (ratio of foreground to background), is an effector molecule secreted by cytotoxic CD8\n \n +\n \n T cells. We found peptides associated with this protein were 1.55x lower in IL-15 than IL-2 stimulated cells using targeted proteomics (\n \n Figure 6A\n \n ), which matches flow cytometry data indicating that the number of effector CD8\n \n +\n \n T cells (T\n \n EFF\n \n ) are lower when stimulated with IL-15 than IL-2. While both IL-2 and IL-15 resulted in the activation of T cells, IL-15 stimulation led to the differentiation of memory-like T cells, as demonstrated in the flow cytometry data. The CD44 receptor antigen is a cell surface receptor that helps cells facilitate cell-cell interaction and response to the tissue microenvironment. Interestingly, we found that the expression of CD44 is slightly higher in IL-2 compared to IL-15, indicating that IL-2 stimulated cells had a higher population of activated cells, with a 1.6x median fold change in abundance. T cells that express CD62L have an increased population of memory T cells after IL-15 stimulation.\n \n 50\n \n Flow cytometry data indicated that we had a higher population of CD62L\n \n +\n \n cells in the IL-15 stimulated condition compared to T cells stimulated with IL-2 (\n \n Supplemental Figure S2\n \n ), indicating a higher population of memory-like T cells. Compared to the IL-15 stimulated cells, IL-2 stimulated T cells expressed IL-2R\u03b2/IL-15R\u03b2 at a higher ratio (\n \n Figure 6A\n \n ), which is associated with a memory phenotype. In general, we observe high analytical precision using PRM with a Q-LIT platform, even in 1 ng assays. Most peptides were measured with less than a 20% coefficient of variation between 3 technical replicates (\n \n Figure 6B\n \n ).\n
\n\n Ultimately, we detected 100% of the proteins monitored with flow cytometry using global proteomics during library generation. While some of these proteins were hard to observe at low input (1 ng), we were able to quantify 75% above an estimated LoQ with targeted proteomics. This overlap indiates complementary benefits of using flow cytometry in tandem with targeted proteomics to fully capture immune cell state. While single-cell proteomics using mass spectrometry continues to develop, flow cytometry is the best method for measuring a small number of proteins (6-12) on thousands of individual cells within a single day. On the other hand, targeted mass spectrometry on 1-10 T cells (equivalent to around 0.1 and 1 ng) can monitor tens to hundreds of proteins, including cytokines and transcription factors, which cannot be easily monitored using flow cytometry.\n
\n\n Here, we demonstrate a complete workflow for acquiring global libraries and rapidly building\n \n de novo\n \n PRM assays using only a Q-LIT mass spectrometer. While the Q-LIT is capable of DDA, we found that equivalently large libraries could be quickly generated using GPF-DIA. We also found that PRM assays using the Q-LIT were linearly quantitative even at low input, enabling us to accurately measure difficult to measure cytokines, transcription factors, and immune proteins. From a broader perspective, high-resolution mass spectrometry is expensive in terms of instrument costs and requiring greater technical experience to operate successfully. In contrast, Q-LIT mass analyzers are easier to maintain and more cost-effective to operate in part because they have less stringent vacuum requirements compared to Orbitrap-based mass spectrometers, making them an appealing option for low-input proteomics. These factors are especially important in single-cell proteomics, where each biological sample needs thousands of injections. Our results suggest that Q-LITs provide laboratories with competent low-input proteomics analysis in situations where high resolution is impractical. We believe these instruments offer a high value-to-expense ratio, potentially democratizing mass spectrometry on a broader array of laboratory settings. This democratization is particularly impactful in immuno-oncology, where proteome-based analysis of immune cell populations can uncover crucial biomarkers to guide clinical decisions.\n
\n\n \n T cell cultures\n \n
\n\n Splenocytes from C57BL/6 mice were stimulated with plate-bound anti-CD3 mAb (145-2C11 clone) on day 0 in complete media, as described previously.\n \n 54\n \n On day 2, cells were washed and re-plated with human (h) IL-2 or hIL-15 at 200 ng/mL. Cells were washed and split on days 4 and 6. Flow cytometry was performed on days 5, 6, and 10, and cells were washed three times with DPBS, centrifuged at 500 RCF for 5 minutes, pelleted, and stored at -80\u00b0C on days 6 and 10 for mass spectrometry. A third condition was maintained without stimulation as a control for flow cytometry as well.\n
\n\n \n Flow cytometry\n \n
\n\n Flow cytometry was performed as previously described.\n \n 54\n \n Briefly, cells collected on days 5, 6, and 10 were stained with live/dead fixable blue dead-cell stain (Invitrogen #L23105), and antibodies for B220, CD4, CD8, CD25, CD44, CD62L, CD69, and TCRb (see antibody details in\n \n Supplemental Table 1\n \n ). Stained cells were acquired with a Cytek Biosciences Aurora\u2122 5-laser flow cytometer and analyzed using BD Biosciences FlowJo\u2122 software.\n
\n\n \n Proteomics sample preparation\n \n
\n\n Frozen cell pellets were lysed in a 5% SDS buffer with 50 mM TEAB, 1x HALT, and 2 mM MgCl\n \n 2\n \n . DNA was sheared with a Bioruptor\u00ae Pico by sonicating at 14\u2103 for 30 seconds, followed by 30 seconds of rest, a total of 10 times. Sheared cells were then spun down at 13,000 RCF for 10 minutes, and the protein supernatant was retained. Protein quantities were estimated using a Pierce\u2122 bicinchoninic acid (BCA) Protein Assay Kit. Proteins were reduced with 40 mM dithiothreitol (DTT), alkylated with 40 mM iodoacetamide, and quenched with 20 mM DTT. Acidification was done with 2.5% phosphoric acid, and protein was loaded onto suspension trap (s-trap) micros (Protifi LLC). Digestion was performed with trypsin at a 1:20 ratio of enzyme to protein at 47\u2103 for 2 hours, then eluted. Peptides were dried down and stored at -80\u2103.\n
\n\n According to the kit protocol, an aliquot of dried peptides was separated according to basicity using a Pierce High pH Reverse-Phase Fractionation Kit. Briefly, 50 \u03bcg of peptides were resuspended in 0.1% trifluoroacetic acid in HPLC-grade water. The separation mini-columns from the kit were centrifuged at 5000 RCF for 2 minutes to remove any liquid and pack the resin. The mini-columns were then equilibrated with 100% acetonitrile and washed 3 times with water. Resuspended peptides were loaded, and the flow through was collected as the first fraction. The mini-columns were washed with water, and the eluent was collected as the second fraction. The elution buffers specified from the kit were then used to produce the following 8 fractions. Fractionated peptides were then dried down and stored at -80\u2103 until mass spectrometry-based analysis for DDA-based library generation.\n
\n\n A separate aliquot of the eluted peptides was dimethyl labeled using an in-solution amine-labeling reaction published by Boresema et al.\n \n 42\n \n Digested peptides were resuspended in 100 mM TEAB (pH = 8.5). Formaldehyde (4%) was added to the resuspended peptides and mixed. Sodium cyanoborohydride (0.6 M) was then added to catalyze the dimethyl labeling reaction for 90 minutes at 22\u2103 while mixing vigorously. The reaction was quenched with 1% ammonia and 5% formic acid. All peptides were resuspended in 2% acetonitrile with 0.1% formic acid. Calibration curves were generated by mixing labeled and unlabeled peptides at different concentrations. In these mixtures, unlabeled peptides were diluted in a dimethyl-labeled background over 4 orders of magnitude (\n \n Supplemental Table 2\n \n ) and aliquoted at different concentrations prior to mass spectrometry analysis.\n
\n\n \n
\n LC-MS settings\n \n
\n Data was acquired on a Thermo Scientific\u2122 Stellar\u2122 MS coupled to a Vanquish\u2122 Neo UHPLC system. Solvent A consisted of 100% water with 0.1% formic acid, and solvent B contained 80% acetonitrile with 0.1% formic acid. An Easy-Spray\u2122 source was used for ionization at 2000 V, and the ion transfer tube was set to 275\u2103. Peptides were separated on a 25 cm C18 analytical Easy-Spray column, packed with 2 \u03bcm beads along a 50-minute linear gradient as follows: from 0-4 minutes, 2% B, 4-8 minutes increased to 8% B, 8 to 58 minutes increased with 28% B, 58 to 65 minutes increased to 44% B, followed by a 10-minute wash at 100% B. The flow for the entire gradient was set to 250 nL/min. The instrument was configured to expect chromatography of approximately 15 seconds and fragment peptides with a default charge state of 2 and a collision cell gas pressure of 8 mTorr.\n
\n\n \n DDA on an ion trap instrument\n \n
\n\n For DDA experiments, the RF lens was set to 30%. Precursor spectra were collected ranging from 350-1250\n \n m/z\n \n at a scan rate of 67 kDA/s. The automatic gain control (AGC) target was set to \u201cStandard\u201d with an absolute AGC target of 3e4. The maximum ion injection time (maxIIT) was set to 100 ms and spectra were collected using centroiding in positive mode. MS2 scans were collected only on peptides with a charge state greater than 1, excluding undetermined charge states. An intensity threshold of 5E2 was used to trigger an MS2 scanand an HCD NCE of 30%. Following the MS2 measurement, the peptide m/zs were placed on a dynamic exclusion list for fragmentation for 2 seconds using a precursor mass tolerance of +/- 0.5\n \n m/z\n \n . Twenty DDA scans were taken in each cycle with a 1.6 m/z isolation window around the precursor of interest. Fragment ions were scanned at 125 kDa/second scan rate from 200-1500\n \n m/z\n \n using an AGC target of 1E4 and a maxIIT of 50 ms.\n
\n\n
\n \n DIA and PRM on an ion trap instrument\n \n
\n For both DIA and PRM experiments, the precursor range was set to 350-1250\n \n m/z\n \n and measured at a rate of 67 kDa/second. The AGC target was set to \u201cStandard,\u201d which is equivalent to 1e4, and the maxIIT was set to 100 ms. The loop control was set to \u201call.\u201d Peptides were fragmented with HCD with NCE set to 30%, and fragments were scanned at 67 kDa/second over a range of 200-1500\n \n m/z\n \n . Precursor isolation windows for DIA were consistently 8\n \n m/z\n \n wide, where margins were set to forbidden zone locations. Six gas phase fractions were used to collect chromatogram libraries over 400-500, 500-600, 600-700, 700-800, 800-900 and 900-1000\n \n m/z\n \n . The majority of settings were the same as a wide-window DIA scan, with the exception of 2\n \n m/z\n \n wide isolation windows over the adjusted precursor\n \n m/z\n \n range for each method.\n
\n\n All settings for PRM scans were the same except for the isolation window width and maxIIT. For PRM assays at 10, 20, and 50 peptides per cycle, the maxIIT was set to 200, 95, and 50 ms, respectively. Precursor isolation windows were set to 2\n \n m/z\n \n ,where MS/MS were collected over 200-1500\n \n m/z\n \n .\n
\n\n \n Data analysis\n \n
\n\n Global data was first converted to the universal mzML format using peak picking. DIA data was analyzed with EncyclopeDIA v.3.0.0-SNAPSHOT using the Ion Trap/Ion Trap mode. Mass tolerance was set to 0.4 Da where a minimum of 3, but a maximum of 5 ions were used for quantification. The chromatogram library was generated by searching 6 gas phase fractions against a Prosit\n \n 38\n \n predicted library. The Prosit library contained spectrum predictions of all +2 and +3 ions from a mouse FASTA from UniProt, which was accessed on October 22, 2019. The predicted library allowed for up to 1 missed cleavage, with a default charge state of 3, and default NCE of 33 over 396.4-1002.7\n \n m/z\n \n . Wide-window (8\n \n m/z)\n \n injections were searched against the chromatogram library using the same search settings. Global DDA data was searched in Scribe using the Ion Trap/Ion Trap instrument mode for b and y tryptic peptides, with a library mass tolerance of 0.4 Da. The 10 high-pH fractionated injections were searched against the same Prosit predicted library used to generate the DIA library.\n
\n\n Global DIA data was searched using CHIMERYS\n \n 55\n \n intelligent search algorithm (MSAID GmbH) in Thermo Scientific\u2122 Proteome Discoverer\u2122 3.1, using analogous settings for the EncyclopeDIA search. A predicted spectrum library was generated from the mouse fasta database by INFERYS\u2122 deep learning framework (MSAID GmbH) for all tryptic +2, +3, and +4 peptides between 7-30 amino acids in length. For processing, spectrum files were selected using the ion trap MS setting, with a signal-to-noise peak threshold of 1.5. The top 24 peaks were selected in each window with a fragment mass tolerance set to 0.4 Da. Fixed carbamidomethyl modifications and a maximum of 2 oxidized methionines were allowed per peptide. The retention times from CHIMERYS were extracted for all detections and combined with EncyclopeDIA\u2019s detections. For peptides detected by both software tools, the EncyclopeDIA retention times were preferred. The combined detections and retention times were used to select peaks within EncyclopeDIA and run against a 1% FDR to obtain a combined search engine library. The fractionated injections were also searched in Proteome Discoverer, using a mass tolerance of 0.4 Da for all +2, +3, and +4 peptides, and the same settings used for the CHIMERYS search,\n
\n\n Skyline\n \n 56,57\n \n version 23.1.0.455 was used for targeted analysis. For analyzing the calibration curves and other PRM injections of IL-2 and IL-15 replicates, the chromatogram library was first imported to serve as a reference point for integrating low-input PRMs. With the imported DIA results, transition settings were altered, and the PRM samples were imported. For both imports, the settings peptide settings were set to Trypsin [KR\\P], with a maximum of 2 missed cleavages, and the mouse fasta used to generate the Prosit library was used to generate a background proteome. Retention time window predictions were set to 5 minutes; however, measured retention times were used when present. Peptides between 7 and 40 amino acids in length were used, and \u201cauto-select all matching peptides\u201d was left checked. Only carbamidomethylation modifications were considered for cysteine. For transition settings, peptides of +2, +3, and +4 precursor charges, along with +1 and +2 fragment ion charges were considered for b and y ion types. For product ion selection, we considered the third ion to the second to last ion. DIA precursor windows were used for exclusion when importing the chromatogram library. The ion match tolerance for the library was set to 0.4 Da, and 6-9 product ions were used from filtered product ions. For the instrument parameters, a 200-1500\n \n m/z\n \n range was considered, with a method match tolerance of 0.4 Da, and \u201cdynamic min product\n \n m/z\n \n \u201d and \u201ctriggered chromatogram acquisition\u201d were checked. For the full-scan parameters, DIA was used when importing the chromatogram library file as a reference point for integrating calibration curves. The gas-phase fractionated isolation windowing scheme was imported from the files for a QIT mass analyzer, with a resolution of 0.4 Da and retention time filtering within 5 minutes of MS/MS IDs. For importing PRM injections, the \u201cPRM\u201d acquisition method was used rather than the DIA method. Finally, lower limits of detection (LoD) and quantification (LoQ) were estimated using EncyclopeDIA, and calculated quantities of IL-2 and IL-15 replicates were determined using calibration curves.\n
\n\n Dataset 1\nSupplemental Data: Figures of Merit reports and calibration curves made from EncyclopeDIA.\n
\n \n\n Reporting Summary\n
\n \n\n Supplemental Table S1: Flow cytometry panel\nSupplemental Table S2: Dilutions used for calibration curves at 100 ng, 10 ng, and 1 ng of total material\nSupplemental Figure S1: Comparison of Chimerys and EncyclopeDIA detections for chromatogram libraries\nSupplemental Figure S2: Additional flow cytometry validation data which contributed to Figure 5.\nSupplemental Figure S3: The calibration curve for a peptide from the CD4 antigen protein.\nSupplemental Note: Tutorial for using PRM Scheduler embedded in EncyclopedDIA\n
\n \n\n Despite accumulating evidence that the development of Alzheimer's disease (AD) is highly associated with the aggregation of A\u03b2 peptides. Still, FDA has approved only one anti-aggregation-based therapy over the past several decades. Here, we report the discovery of an A\u03b2 peptide aggregation inhibitor: an ultra-small nanodot called C\n \n 3\n \n N. C\n \n 3\n \n N nanodots alleviate aggregation-induced neuron cytotoxicity, rescue neuronal death, and prevent neurite damage\n \n in vitro\n \n . Importantly, they reduce the global cerebral A\u03b2 peptides levels, particularly in fibrillar amyloid plaques, and restore synaptic loss in AD mice. Consequently, these C\n \n 3\n \n N nanodots significantly ameliorate behavioral deficits of APP/PS1 double transgenic AD mice. Moreover, analysis of critical tissues (e.g., heart, liver, spleen, lung, and kidney) display no obvious pathological damage, suggesting C\n \n 3\n \n N nanodots are biologically safe. Finally, molecular dynamics simulations also reveal the inhibitory mechanisms of C\n \n 3\n \n N nanodots in A\u03b2 peptides aggregation and its potential application against AD.\n
\n\n \n Alzheimer's Disease\n \n
\n\n \n C3N Nanodots\n \n
\n\n \n A\u03b2 Peptides Aggregation\n \n
\n\n \n Restore Synaptic Loss\n \n
\n\n \n Ameliorate Behavioral Deficits\n \n
\n\n Alois Alzheimer reported the first case of Alzheimer's disease (AD) in 1906\n \n 1, 2\n \n . Now, more than one century later, AD remains an unresolved public health problem worldwide\n \n \n 3\n \n \n . AD is a progressive neurodegenerative disease associated with insidious onset and slow progression of behavioral and cognitive dysfunction. The severity of the AD from early stage\n \n \n 4\n \n \n advances to obvious symptoms which further aggravates the need to utilize immediate remedies against the progression of the disease. Moreover, the incidence of AD also increases with the increasing age reflected by the increasing rate of ~\u200927.6% in 65\u201374 year-old people to ~\u200936.4% in people over 80 years old\n \n \n 5\n \n \n . This significant increase with age also poses a worldwide threat of acquiring AD among elderly population. This also urges the need of developing novel and effective AD management therapies for clinical purposes.\n
\n\n Growing evidence suggests the aggregation of A\u03b2 peptides is highly related with synaptic dysfunction, neuroinflammation, oxidative stress damage, neurotoxicity mediated by the triggered hyperphosphorylation of downstream Tau protein, as well as the ultimate cell death\n \n \n 6\n \n ,\n \n 7\n \n \n . In contrast, suppression of the A\u03b2 peptides aggregation process also offers a suitable therapeutic strategy against AD. However, the successful implementation of this concept remains a huge challenge despite decades of effort along this direction. On the other hand, the lack of effective drugs against AD, with the only available FDA approved drug\n \n aducanumab\n \n \n \n 8\n \n \n , raises high demand for alternate therapeutic options. Besides, other anti-AD agents (including peptides\n \n \n 9\n \n ,\n \n 10\n \n \n , polymers\n \n \n 11\n \n ,\n \n 12\n \n \n , small drug molecules\n \n \n 13\n \n ,\n \n 14\n \n ,\n \n 15\n \n ,\n \n 16\n \n \n , and metal oxides\n \n \n 17\n \n \n ) show only a very mild inhibition effect on A\u03b2 peptides aggregation. Recently, nanomaterials (e.g., graphene oxide\n \n \n 18\n \n \n , fullerenes\n \n \n 19\n \n ,\n \n 20\n \n \n , quantum dots\n \n \n 21\n \n \n , carbon nanotube\n \n \n 22\n \n \n , and g-C\n \n 3\n \n N\n \n 4\n \n \n 23, 24\n \n ) have been reported to inhibit, directly or indirectly, the aggregation of A\u03b2 peptides, including both the inhibition of oligomer fibrillization and disaggregation of mature fiber\n \n in vitro\n \n , but very few of them can still work\n \n in vivo\n \n . Interestingly, graphene quantum dots were also found to inhibit \u03b1-synuclein aggregation, disassociate mature fibrils, and penetrate the blood-brain barrier (BBB) leading to ultimate protection of dopamine neurons\n \n \n 25\n \n \n . Therefore, the use of nanomaterials may offer valuable alternate source as therapeutic agents for protein conformational diseases (e.g., AD, Parkinson's disease, Huntington's disease, Type 2 diabetes).\n
\n\n In this study, we demonstrate that C\n \n 3\n \n N nanodots can significantly inhibit A\u03b2 peptides aggregation, relieve aggregation-induced neuron cytotoxicity, rescue neuronal death, protect neurites from damage, and exhibit only mild cytotoxicity both\n \n in vitro\n \n and\n \n in vivo\n \n . Moreover, the intraperitoneal administration of C\n \n 3\n \n N nanodots for 6 months significantly improves the learning and spatial memory abilities of APP/PS1 in double transgenic AD mice. Additionally, the underlying molecular mechanism of A\u03b2 peptide aggregation inhibition by C\n \n 3\n \n N nanodots has also been explored using all-atom molecular dynamics (MD) simulations. Thus, we believe our current study provides novel insights into the anti-A\u03b2 peptides aggregation capability of C\n \n 3\n \n N nanodots and its potential application against AD.\n
\n\n C\n \n 3\n \n N nanodots were synthesized by polymerization of 2,3-diaminophenazine using hydrothermal synthesis following a previous report\n \n \n 26\n \n \n . The synthesized nanodots had an average lateral size of 4.5\u2009\u00b1\u20090.4 nm (Fig.\n \n 1\n \n a) with a lattice spacing of 0.21 nm, which corresponds to the (100) plane of graphite. Initially, the identification and characterization of C\n \n 3\n \n N nanodots were performed using several spectroscopic techniques including UV\u2013visible (UV\u2013vis) absorption spectroscopy, Fourier transform infrared (FTIR) spectroscopy and X-ray photoelectron spectroscopy (XPS). (\n \n Figure S1\n \n ).\n
\n\n We first studied the role of C\n \n 3\n \n N nanodots towards the aggregation behavior of A\u03b2\n \n 42\n \n peptides, which were shown to have more implications than A\u03b2\n \n 40\n \n in forming neurotoxic assemblies and causing AD pathogenesis\n \n \n 27\n \n ,\n \n 28\n \n \n . In the absence of C\n \n 3\n \n N nanodots, A\u03b2\n \n 42\n \n peptides aggregated into mature amyloid fibers, as demonstrated by various experimental procedures. This included the utility of ThT fluorescence, dot blot assay, atomic force microscope (AFM), transmission electron microscope (TEM), and CD spectroscopy. During these investigations, C\n \n 3\n \n N nanodots effectively inhibited the aggregation of A\u03b2\n \n 42\n \n peptides (Fig.\n \n 1\n \n ). It was evident from delayed aggregation kinetics and reduced ThT fluorescence intensity (after convergence of aggregation process) following C\n \n 3\n \n N nanodots treatment. The inhibition strength was found positively correlated with C\n \n 3\n \n N nanodots treatment concentration (Fig.\n \n 1\n \n b). The final peptide self-assembly samples were also examined through dot blotting using an amyloid fiber conformation-specific antibody (mOC87)\n \n \n 29\n \n \n . Notably, amyloid fiber content decreased with the increasing concentration of C\n \n 3\n \n N nanodots during treatment (Fig.\n \n 1\n \n c). This confirmed the inhibition function of C\n \n 3\n \n N nanodots against peptides aggregation. Morphologically, A\u03b2\n \n 42\n \n peptides aggregated to long and well-defined mature fibers after 24 hours in PBS\n \n without\n \n C\n \n 3\n \n N nanodots, as demonstrated through AFM and TEM imaging (Fig.\n \n 1\n \n d and S2). In contrast, incubation\n \n with\n \n C\n \n 3\n \n N nanodots for 24 hours resulted in a gradual morphologic change of A\u03b2\n \n 42\n \n peptides self-assembly samples from long mature fibers to diffused punctiform structures. Overall, these results suggested that C\n \n 3\n \n N nanodots effectively inhibit the aggregation of A\u03b2\n \n 42\n \n peptides.\n
\n\n To further unveil the regulating process and underlying molecular mechanisms of C\n \n 3\n \n N nanodots towards inhibiting aggregation of these peptides, we then performed all-atom molecular dynamics (MD) simulations. In the absence of C\n \n 3\n \n N, two A\u03b2\n \n 42\n \n peptides self-assembled into a partially ordered structure (containing \u03b2-sheets). However, C\n \n 3\n \n N nanodot application significantly inhibited the formation of any \u03b2-sheets. For instance, in two out of three trajectories (run 1 and run 3), very rare \u03b2-sheet contents were formed (i.e. in run 2, \u03b2-sheet appeared at t\u2009=\u200980 ns, then disappeared at t\u2009=\u2009340 ns) (Fig.\n \n 1\n \n e and S3). Convergence of the simulations (>\u2009900 ns) demonstrated an overall decrease in the \u03b2-sheet content of 10.6\u2009\u00b1\u20091.5% without C\n \n 3\n \n N to 0.2\u2009\u00b1\u20090.6% with C\n \n 3\n \n N. Simultaneously, the random-coiled and bend components increased from 37.0\u2009\u00b1\u20092.4% to 40.6\u2009\u00b1\u20091.7%, and 13.5\u201320.3% (Fig.\n \n 1\n \n f), respectively. These findings suggested that C\n \n 3\n \n N nanodot effectively redirects A\u03b2\n \n 42\n \n peptides self-assembly to disordered structures. Moreover, CD spectroscopy confirmed that C\n \n 3\n \n N nanodots redirected the secondary structure of A\u03b2\n \n 42\n \n peptides (at t\u2009=\u200924 hours) from the \u03b2-sheet-rich to disordered random-coiled conformations (Fig.\n \n 1\n \n g). These results sufficiently demonstrate the structural modulating role of C\n \n 3\n \n N nanodot in impeding the aggregation of A\u03b2\n \n 42\n \n peptides.\n
\n\n The detailed interaction energies including both van der Waals (vdW) and electrostatic (elec) interactions between C\n \n 3\n \n N and peptides were also explored (Fig.\n \n 1\n \n h). This was performed by analyzing the key binding configurations in a typical trajectory to better illustrate the binding mechanisms. Driven by vdW and hydrophobic interactions, one peptide was adsorbed onto the surface of C\n \n 3\n \n N (t\u2009=\u20091 ns) and strengthened by \u03c0\u2012\u03c0 stacking interactions (F4 and F20) (t\u2009=\u200910 ns). At t\u2009=\u200913 ns, another peptide was adsorbed onto the edge of C\n \n 3\n \n N by electrostatic attractions between E17, D7, and E3 residues with \u2012NH\n \n 3\n \n \n +\n \n groups at the edge of C\n \n 3\n \n N nanodot. At t\u2009=\u200933 ns, this peptide was fully adsorbed onto the other side of C\n \n 3\n \n N nanodot via vdW and \u03c0\u2012\u03c0 stacking interactions. After 96 ns, the adsorption process converged. At this state, most hydrophobic and aromatic residues were adsorbed onto the C\n \n 3\n \n N nanodot surface. Meanwhile, some charged or polar residues formed salt-bridge or hydrogen bonds with edge groups (e.g., \u2012COO\n \n \u2012\n \n and \u2012NH\n \n 3\n \n \n +\n \n ) of C\n \n 3\n \n N nanodot while suppressing subsequent aggregation of peptides. Hence, the strong adsorption between peptides and C\n \n 3\n \n N nanodot was collectively driven by a combination of vdW and electrostatic, hydrophobic, hydrogen bonding, and \u03c0\u2012\u03c0 stacking interactions, with the vdW interaction dominating (Fig.\n \n 1\n \n h), to induce disruption in peptides self-assembly and form disordered structures.\n
\n\n As shown above, C\n \n 3\n \n N nanodots exhibited an effective inhibiting function against A\u03b2\n \n 42\n \n peptides aggregation at molecular level. At this stage, it was logical to examine whether C\n \n 3\n \n N nanodots alleviate aggregation-induced neuron cytotoxicity (Fig.\n \n 2\n \n ). Herein, we analyzed primary neuron cells viability and toxicity under different conditions using cell counting kit 8 (CCK-8), Lactate Dehydrogenase (LDH), and Live/Dead assays. The CCK-8 assay results demonstrated that A\u03b2\n \n 42\n \n peptides aggregation causes severe toxicity in neurons. This was found after neuronal cells incubation with 50 \u00b5M A\u03b2\n \n 42\n \n peptides for 24 hours which resulted in a survival rate of only 29.89\u2009\u00b1\u20093.98%. However, increased treatment concentration with C\n \n 3\n \n N nanodots resulted in improved cell survival rate: ~44.83\u2009\u00b1\u20096.90% (100 \u00b5g/ml) to 65.52\u2009\u00b1\u20099.12% (500 \u00b5g/ml) (Fig.\n \n 2\n \n a). Hence, C\n \n 3\n \n N nanodots dose-dependently relieved A\u03b2\n \n 42\n \n peptides aggregation-induced neuron cytotoxicity, which was further confirmed by LDH (Fig.\n \n 2\n \n b) and Live/Dead experimental (Fig.\n \n 2\n \n c\n \n & d\n \n ) results. In addition, the cytotoxicity of C\n \n 3\n \n N nanodots was found very mild with C\n \n 3\n \n N nanodots administered at 500 \u00b5g/mL resulting a neuronal survival rate of ~\u200988.51\u2009\u00b1\u20092.0%. We further investigated the morphologies of neurons under different conditions using scanning electron microscope (SEM) technology (Fig.\n \n 2\n \n e). Normal neurons presented in a plump-pear shape with many dendrites. However, A\u03b2\n \n 42\n \n aggregation-induced significant deformations of neurons, e.g., the cellular body shrunk notably and was accompanied by severe dendrites loss. In contrast, the treatment with C\n \n 3\n \n N nanodots resulted in well maintained dense dendrites suggesting inverse effect against the toxicity caused by A\u03b2\n \n 42\n \n peptides aggregation in neurons. It also distinguished the mild influence of C\n \n 3\n \n N nanodots on the shape of neurons.\n
\n\n In addition, the cytotoxicity of C\n \n 3\n \n N nanodots in several cell lines was also examined, including red blood cells (RBC), rat adrenal chromaffin cell tumor cells (PC12), primary neuron (Neuron), primary astrocytes (Astrocyte), human umbilical vein endothelial cells (HUVCE), and human neuroblastoma cells (SH-SY5Y). The results showed that C\n \n 3\n \n N nanodots possess decent cytocompatibility among all tested cell lines\n \n (Figure S4 and S5)\n \n . This revealed that C\n \n 3\n \n N nanodots alleviate neuron cytotoxicity, reduce cell death, protect A\u03b2\n \n 42\n \n aggregation-induced axonal and dendritic damages and demonstrate remarkable cytocompatibility.\n
\n\n Following the encouraging\n \n in vitro\n \n findings, we sought to determine whether C\n \n 3\n \n N nanodots have neuroprotective functions towards AD mice via inhibition of A\u03b2 peptides aggregation. For this purpose, we used APP/PS1 double transgenic mice as the model AD organism. This\n \n in-vivo\n \n model overexpresses A\u03b2 peptides in the brain by inducing amyloid plaque formation which eventually leads to the occurrence of AD symptoms\n \n \n 31\n \n ,\n \n 32\n \n \n . The expression of A\u03b2 peptides in APP/PS1 mice begins at 3\u20124 months of age. Thus, we treated the APP/PS1 mice with C\n \n 3\n \n N nanodots-saline solution per day from 3 to 9 months via intraperitoneal injection. APP/PS1 mice received saline only were set as the positive control group, and wildtype mice with non-intervention were set as the negative control. After six months of C\n \n 3\n \n N nanodots injection vs. no injection, the cognitive function of APP/PS1 mice were examined using the Morris water maze and novel object recognition tests (Fig.\n \n 3\n \n ).\n
\n\n We first refined the optimal C\n \n 3\n \n N nanodots administration dose from the assessment of the escape latency. In the Morris water maze test during the 5-day learning phase, the latency time for APP/PS1 mice to find the survival platform (initially placed in the third quadrant) in the saline group underwent a very mild decrease as shown previously\n \n \n 33\n \n \n . Treatment with C\n \n 3\n \n N nanodots significantly shortened the latency time, indicating a remarkably improved learning capacity of AD mice (Fig.\n \n 3\n \n a). We also noted that treatment with 1 mg/kg/d dose obtained better therapeutic effect than that treated with 5 mg/kg/d dose (Fig.\n \n 3\n \n a), suggesting that 1 mg/kg/d may be the optimal dose. The potential contribution of the swimming capability (swimming speed) to the learning effects was excluded because there was no distinct difference in the average swimming speed between two C\n \n 3\n \n N nanodots treated groups and the wildtype mice (Fig.\n \n 3\n \n b). Overall, these results demonstrated the efficacy of C\n \n 3\n \n N nanodots in the treatment and improving the learning capacity of AD mice, with an optimal dose of ~\u20091 mg/kg/d. Hence, 1 mg/kg/d was used in the following\n \n in vivo\n \n experiments.\n
\n\n To further measure the spatial memory capability, the third quadrant residence time of mice was accumulated during 60s swimming after retrieval of the survival platform on day 6. C\n \n 3\n \n N nanodots-treated AD mice spent significantly more time in the third quadrant and crossed this target quandrant more often compared to control APP/PS1 mice (Fig.\n \n 3\n \n c, d & f). In addition, the time to explore the new object among APP/PS1 mice was significantly reduced as compared to that of the wild-type. However, treatment with C\n \n 3\n \n N nanodots can remarkably prolong the time of APP/PS1 mice to explore the new object, which resulted in the recognition index (RI) of APP/PS1 mice (treated with C\n \n 3\n \n N nanodots) was remarkably improved to level comparable with wild-type mice (Fig.\n \n 3\n \n e & g). These results were indicative that C\n \n 3\n \n N nanodots treatment could partially rescue these defects in APP/PS1 mice and may offer utility against AD.\n
\n\n Furthermore, the body weights among both C\n \n 3\n \n N nanodots treated and untreated AD mice increased steadily during the entire administration period (\n \n Figure S6\n \n ) suggesting the higher biocompatibility of C\n \n 3\n \n N nanodots in animals. Moreover, the H&E staining in heart, liver, spleen, lung, and kidney tissue showed no distinct lesions indicating C\n \n 3\n \n N nanodots related limited biotoxicity\n \n in vivo\n \n (\n \n Figure S7\n \n ).\n
\n\n Next, we detected the level of cerebral fibrillar amyloid plaques as hallmark of AD\n \n \n 34\n \n \n in wild-type and APP/PS1 mice untreated/treated with C\n \n 3\n \n N nanodots. The 6E10 anti-A\u03b2 antibody was used because of its specific binding capability with residues 1 to 16 of the A\u03b2 peptide. Notably, massive amyloid plaques accumulated in both the cerebral cortex and hippocampus of APP/PS1 mice treated with saline (1.22\u2009\u00b1\u20090.29%) (Fig.\n \n 4\n \n a). However, the amyloid plaques deposition levels remarkably decreased after treatment with 1 mg/kg/d (0.33\u2009\u00b1\u20090.18%; a 74.4% decrease) C\n \n 3\n \n N nanodots treatment (Fig.\n \n 4\n \n b). These results were also confirmed by counting the number of amyloid plaques (Fig.\n \n 4\n \n c).\n
\n\n Considering that A\u03b2\n \n 40\n \n and A\u03b2\n \n 42\n \n peptides are the dominant component of the plaques in the brains of AD patients\n \n \n 27\n \n \n . We then used enzyme-linked immunosorbent assay (ELISA) to quantify the level of intra-cephalic A\u03b2\n \n 42\n \n /A\u03b2\n \n 40\n \n peptides. This involved using Tris-buffered saline (TBS) \u2012, sodium dodecyl sulfate (SDS)\u2012, and formic acid (FA)\u2012soluble A\u03b2 forms corresponding to the soluble, partially soluble (non-dense plaque), and completely insoluble (dense plaque) A\u03b2 forms, respectively. These analyses showed that treatment with C\n \n 3\n \n N nanodots decreased the level of A\u03b2\n \n 42\n \n / A\u03b2\n \n 40\n \n peptides by 37% / 33%, respectively. The relative FA\u2012soluble A\u03b2\n \n 42\n \n / A\u03b2\n \n 40\n \n species levels was reduced most significantly by 84% / 81% (Fig.\n \n 4\n \n d & e), which suggested that C\n \n 3\n \n N nanodots effectively inhibit A\u03b2 peptides aggregating into completely insoluble dense plaques. Overall, C\n \n 3\n \n N nanodots possessed the strong ability to delay or obstruct A\u03b2 peptide aggregation pathogenesis\n \n in vivo\n \n .\n
\n\n Synaptic dysfunction is another important pathological feature of AD\n \n \n 35\n \n ,\n \n 36\n \n \n having a strong impact in nerves development and neurotransmitters release (including dopamine and glutamate). The SNAP25 and VAMP2 proteins are the two main synaptic proteins which protects synaptic integrality\n \n \n 37\n \n ,\n \n 38\n \n \n . Therefore, we assessed the changes in expression levels of these two proteins using western blot and immunohistochemistry fluorescence assays. Western blot results demonstrated that the content of two proteins was up-regulated (Fig.\n \n 5\n \n a) after treatment with C\n \n 3\n \n N nanodots. The quantification of the SNAP25 and VAMP2 protein levels was performed using grey density analyses by utilizing Image J software. The results showed that expression levels of the two proteins were increased by 43% & 22% respectively, after treatment with C\n \n 3\n \n N nanodots (Fig.\n \n 5\n \n b).\n
\n\n In addition, we also examined the synaptic damage by double-staining brain tissue with an antibody against microtubule-associated protein 2 (MAP2; a neuronal marker) and SNAP25 (a synaptic marker) (n\u2009=\u20093/group). The co-localization of SNAP25 and MAP2 signals reflected the expression level of synaptic proteins in neurons. In APP/PS1 mice treated with saline only, SNAP25/MAP2 co-localization yellow pixel intensity decreased remarkably, which indicated serious dysfunction of the neural network as compared to wild-type mice. In contrast, treatment with C\n \n 3\n \n N nanodots for six months resulted in significantly elevated SNAP25/MAP2 expression levels (Fig.\n \n 5\n \n c). These results demonstrated that C\n \n 3\n \n N nanodots maintains an effective protective function in the synapse.\n
\n\n In this study, an effective A\u03b2 peptides aggregation nano-inhibitor called C\n \n 3\n \n N nanodots has been explored against AD. This novel inhibitor redirects peptide self-assembly to disordered off-pathway species and reduces aggregation-induced neuron cytotoxicity\n \n in vitro\n \n and\n \n in vivo\n \n . Several experimental analyses including ThT fluorescence, dot blot assays and CD spectra collectively demonstrate that C\n \n 3\n \n N nanodots guide A\u03b2 peptides self-assembly to disordered structures rather than \u03b2-sheet-rich structures. Similarly, morphological observations using AFM and TEM imaging showed that after treatment with C\n \n 3\n \n N nanodots these A\u03b2 peptides form small diffused oligomeric structures, in contrast to the long and well-defined mature amyloid fibers formed in the absence of C\n \n 3\n \n N nanodots. The results from CCK-8, LDH, and Live/Dead assay revealed that C\n \n 3\n \n N nanodots relieve neuron toxicity induced by A\u03b2 aggregation and rescue neuronal death. SEM images further helped to depict that C\n \n 3\n \n N nanodots protect normal neuronal morphology from A\u03b2 aggregation-induced destruction. Furthermore, MD simulations demonstrated that both the non-specific hydrophobic and electrostatic interactions, and the specific \u03c0\u2012\u03c0 stacking and hydrogen bonding interactions between C\n \n 3\n \n N nanodots and A\u03b2 peptides synergistically obstruct the aggregation process of A\u03b2 peptides.\n
\n\n The cognitive abilities among the studied mice were also restored following C\n \n 3\n \n N nanodots treatment. After C\n \n 3\n \n N nanodots treatment, the cognitive ability of APP/PS1 mice significantly improved to levels comparable with wild-type mice. Several immunological experiments including immunohistochemistry fluorescence, western blot, and ELISA assays demonstrated that APP/PS1 mice experience a decrease in cerebral fibrillar amyloid plaque levels and an increase in SNAP25 and VAMP2 with C\n \n 3\n \n N nanodots treatments. Conclusively, this study not only provides useful experimental and theoretical basis for the application of C\n \n 3\n \n N nanodots in neuronal protection, but also offers the groundwork for subsequent optimal designs of nanomaterials targeting A\u03b2 peptides aggregation in AD.\n
\n\n The synthesis of C\n \n 3\n \n N nanodots was based on the method reported by our group\n \n \n 26\n \n \n . Briefly, the aqueous solution of 2,3-Diaminophenazine (80 mL, 1.4mM) was heated and kept at320\u00b0C for 36 hours in a 100 mLpoly (p-phenylene)-lined stainless-steel autoclave. The products were filtered by 0.02 \u00b5m alumina microporous membrane to obtain the raw C\n \n 3\n \n N nanodots. Then, the raw C\n \n 3\n \n N nanodots were treated with H\n \n 2\n \n O\n \n 2\n \n (5 M, 80\u00b0C for 6 h) for further oxidization. Finally, the sample was purified via membrane dialysis with the molecular weight cutoff of 500\u20131000 Da for 5 days, and the oxygen-modified C\n \n 3\n \n N nanodots were obtained.\n
\n\n The transmission electron microscopy (TEM) and high-resolution TEM (HRTEM) images were obtained using transmission electron microscope with the accelerating voltage of 200 kV (Tecnai G2 F20, FEI Corporation, American). The Fourier transform infrared (FT-IR) spectra of C\n \n 3\n \n N nanodots were characterized using fourier transform infrared spectrometer (Hyperion, Bruker Corporation, Germany). The UV-vis spectra analysis utilized a UV-vis spectrophotometer (Lambda 750, PerkinElmer, American), and the X-ray photoelectron spectra (XPS) were obtained using a X-ray photoelectron spectrometer (Axis ultra DLD, Kratos, Britain).\n
\n\n A\u03b2\n \n 42\n \n (NH2-DAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVVIV-COOH, purity\u2009\u2265\u200998%) peptides were purchased from APeptide Co., Ltd (Shanghai, China) and prepared according to protocols previous described\n \n \n 39\n \n ,\n \n 40\n \n \n . Briefly, A\u03b2\n \n 42\n \n peptides were first dissolved in hexafluoroisopropanol (HFIP, 10522, Sigma Aldrich) and sonicated for 10 minutes. The A\u03b2\n \n 42\n \n \u2012HFIP solution was then incubated at room temperature for 1 hour to ensure the monomerization and structural randomization of peptides, and placed into a fume hood to completely evaporate HFIP. The obtained peptide film was stored at \u201280\u00b0C. Immediately before use, the peptide film was resuspended to 5 mM in dimethyl sulfoxide (DMSO, D2650, Sigma Aldrich) and diluted to a final concentration of 100 \u00b5M in phosphate-buffered solution (PBS, 0.1 M). The solution was then centrifuged at 16000 g for 10 min at 4\u00b0C to remove the pre-formed fibers. In the aggregation experiment, A\u03b2\n \n 42\n \n (100 \u00b5M) was mixed with C\n \n 3\n \n N nanodots at various concentrations or PBS solution to a final concentration of 50 \u00b5M and then incubated at 37\u00b0C with constant agitation at 300 rpm for 24 h.\n
\n\n Fluorescence with Thioflavin T (ThT) was used to detect aggregated A\u03b2 containing \u03b2-sheets, as previously described\n \n \n 41\n \n \n . A 50 \u00b5L sample was mixed with 150 \u00b5L ThT (20 \u00b5M, T3516, Sigma Aldrich) in a 96-well plate. The resulting fluorescence intensity was detected immediately after mixing with a fluorescence plate reader (BioTek, USA) at excitation and emission wavelengths of 450 nm and 485 nm, respectively. Fluorescence values of C\n \n 3\n \n N nanodots and ThT were subtracted from that of the mixed solution. Error bars (\u00b1\u2009s.d.) of triplicate samples are shown for selected data points.\n
\n\n Dot blot assays were carried out as described previously\n \n \n 42\n \n \n to probe the formation level of A\u03b2\n \n 42\n \n amyloid mature fibers under different conditions. Briefly, 5 \u00b5L aliquots of the sample were dropped onto nitrocellulose membranes (1060002, GE Healthcare). Once the membranes dried, they were blocked for 1 hour with 3% nonfat milk in tris-buffered saline (TBS) solution and then incubated with Anti-Amyloid Fibril antibody (mOC87) (1:8000, ab201062, Abcam) overnight at 4\u00b0C. The membranes were washed 3 times in TBST for 5 minutes and then incubated with the horseradish peroxidase (HRP)-conjugated donkey anti-rabbit secondary antibody (1:5000, 711-035-152, Jackson ImunoResearch) for 2 hours at room temperature. Finally, the membranes were developed by chemiluminescence using ECL Plus (P0018S, Beyotime).\n
\n\n Here, 10 \u00b5L of each sample was dispersed on freshly cleaved mica sheets. After air-drying, samples were scanned and analyzed using the tapping mode of AFM (Bruker, Germany), and the height of the sample was recorded.\n
\n\n Ten microliters of each sample were dispersed on a copper grid (carbon and formvar coated 300 mesh, Zhongjing Technology Co., Ltd., China) for 2 min at room temperature. Then, they were washed twice with ultrapure water and negatively stained with 1% uranyl acetate for 2 min. After air-drying, images of peptides were observed using a Tecnai G2 spirit BioTwin TEM at 120 kV.\n
\n\n All samples were diluted six times under PBS conditions. Spectra were detected using a Jasco J-815 circular dichroism spectropolarimeter (1 mm path length cuvette) at 25\u00b0C. The spectrum of PBS was set as the baseline. Each sample was scanned three times and the average value was adopted. Raw data, after subtracting the buffer spectra, were smoothed according to the manufacturer\u2019s instructions.\n
\n\n Mouse primary cortical neurons were obtained from embryonic day 18 C57BL/6 mice as reported previously\n \n \n 25\n \n \n with minor changes. All animal procedures followed the policies of the Soochow University Animal Care and Use Committee (SUACUC). In brief, dissociated neurons were plated onto dishes coated with poly-D-lysine (P6407, Sigma Aldrich) then suspended in culture medium (Neurobasal Media (21103-049, Invitrogen) containing 2% B-27 (17504-044, Invitrogen), 1% penicillin/streptomycin (15140122, P/S, Gibco), 1% L-glutamine and 0.25% GlutaMax\u2122 (35050, Invitrogen)). Next, the plating medium was substituted with feeding medium (Neurobasal medium supplemented with 2% B27, 1% P/S, and 1% L-glutamine) on the second day after cell plating. The medium was replaced twice a week and the cultures were incubated in a 5% CO\n \n 2\n \n incubator at 37\u00b0C. Cells were used for experimentation 8 days after seeding.\n
\n\n Primary astrocyte cultures were extracted from the cerebral cortex of 1-3-d-old rats (Sprague-Dawley) following prior methods\n \n \n 43\n \n \n . In brief, dissociated cortical cells were suspended in DMEM media (sh30022.01b, Hyclone) containing 1% P/S (Gibco) and 10% Fetal bovine serum (10099141, Gibco) and plated on PDL-coated 75 cm\n \n \n 2\n \n \n flasks at a density of 6 \u00d7 10\n \n 5\n \n cells/cm\n \n 2\n \n . Monolayers of type 1 astrocytes were harvested 12\u201314 days after plating. Non-astrocytic cells were separated and removed from the flasks by shaking and changing the medium. Astrocytes were dissociated through trypsinization and reseeded on uncoated 96-well plates. The cells grew to 80\u201390% confluence before exposure to C\n \n 3\n \n N nanodots.\n
\n\n Cell viability was evaluated using CCK-8 kit (ck04, Dojindo), LDH cytotoxicity assay kit (K311-400, Biovision), and Live/Dead kit (l3224, Invitrogen). Before experimentation, the neuron culture medium was used to dilute 5 mM A\u03b2\n \n 42\n \n peptide stock solution and C\n \n 3\n \n N nanodots solution to achieve a mixture of 50 \u00b5M A\u03b2\n \n 42\n \n and C\n \n 3\n \n N nanodots at various concentrations (e.g., 100, 200, 300, 400, and 500 \u00b5g/mL). A control group with medium solution and experimental groups with 50 \u00b5M A\u03b2\n \n 42\n \n peptide solution and 500 \u00b5g/mL C\n \n 3\n \n N nanodots solution were analyzed. The culture solutions were incubated at 4\u00b0C for 24 hours and then added to cells for another 24 hours at 5% CO\n \n 2\n \n humidified environment 37\u00b0C.\n
\n\n For the CCK-8 assay, the diluted CCK-8 solution was added to the cells and incubated in 5% CO\n \n 2\n \n at 37\u00b0C for 30 minutes. The optical density was measured at 450 nm on a microplate reader (BioTek, USA). Cell viability of the control group was set to 100%, and cell viability of other groups was calculated by comparison to the control group.\n
\n\n The LDH assay was performed according to LDH cytotoxicity assay kit instructions. A group of cells treated with 1% Triton X-100 was added as a positive control; the cell-free group was the negative control. Optical density at 490 nm was measured on a microplate reader and the cytotoxicity of each group was calculated according to:\n
\n\n Cytotoxicity (%) = (Test Sample - Negative Control) / (Positive Control - Negative Control) \u00d7 100% (1)\n
\n\n For the Live/Dead assay, the prepared dye was incubated with cells for 15 minutes according to the Live/Dead kit instructions. Cells were then photographed under a fluorescence microscope (Leica, Germany), and live vs. dead cells were counted using Image J software.\n
\n\n Cells were planted on cell culture slides, washed twice with PBS, and fixed overnight with 2.5% glutaraldehyde at 4\u00b0C for morphological observation. Twenty-four hours later, they were washed 3x with ultrapure water for 5 minutes each. Then, 30%, 50%, 70%, 80%, 90%, 95%, and 100% ethanol dehydration occurred in sequence for 10 minutes. Gold was then sprayed on the surface of the sample, and cell morphology was observed using a scanning electron microscope (SEM, Zeiss, Germany).\n
\n\n APP/PS1 [B6C3-Tg (APPswePSEN1dE9)/Nju] double transgenic AD mice and C57BL/6 mice were used in this study (Nanjing Model Animal Research Center, Nanjing, China). All experiments were reviewed and approved by the Animal Ethics Committee of Soochow University. APP/PS1 mice were produced and maintained on a C57BL/6 hybrid background with free access to chow and drinking water under a 12 hour light/dark cycle under constant temperature (22\u2009\u00b1\u20091\u00b0C) and humidity (40\u201270%).\n
\n\n APP/PS1 mice were randomly divided into three groups. The positive control group was intraperitoneally injected with vehicle (saline; APP/PS1 group). The other two groups were injected intraperitoneally with either 1 mg/kg or 5 mg/kg C\n \n 3\n \n N nanodots solution. Littermate wild-type mice treated with saline solution were used as negative controls (WT group). Drugs were given once per day from 3 months of age for six months.\n
\n\n After behavioral tests, each group of mice was subdivided into two additional groups. In the first group, mice were subjected to cardiac perfusion under deep anesthesia and perfused with PBS and 4% paraformaldehyde (PFA, 158127, Sigma Aldrich) dehydrated with sucrose. In the second group, mouse brains were harvested by decapitation, then quickly placed in \u201280\u00b0C for the extraction of brain proteins.\n
\n\n Tissues at \u201280\u00b0C were homogenized in cold lysis buffer (P0013C, Beyotime) containing protease inhibitor cocktail (4693116001, Roche). The supernatants were incubated at 100\u00b0C for 10 min. Each protein (15 \u00b5g) was separated by electrophoresis using a 12% SDS-PAGE gel (P0692, Beyotime) and transferred onto a PVDF membrane (ipvh00010, Millipore). The membranes were blocked by incubation with 5% non-fat milk (wt/vol) in Tris-buffered saline containing 0.1% Tween-20 (vol/vol) (TBST) for 60 minutes at 25\u00b0C. The membranes were then incubated with primary antibody to synaptosomal-associated 25 KD protein (SNAP25, 1:2000, 111-002, Synaptic Systems), antibody to vesicle-associated membrane protein 2 (VAMP2, 1:1000, ab3347, Abcam), and antibody to \u03b2-actin (1:5 000, bs1002, Bioworld technology) overnight at 4\u00b0C. The membranes were washed thrice in TBST for 5 minutes and incubated with HRP\u2013conjugated IgG secondary antibody (1:5,000, Jackson ImunoResearch) for 2 hours at room temperature. The membranes were washed in TBST (3 \u00d7 5 minutes) before a 2-hour incubation with HRP-linked secondary antibodies to rabbit or mouse accordingly at room temperature. The membranes were then visualized using chemiluminescence on ECL Plus, and the densitometric quantifications were analyzed by Image J software.\n
\n\n Spatial learning and memory performance were tested using the MWM task and the novel object recognition test. The Morris water maze was conducted with minor adjustments as previously described\n \n \n 44\n \n \n , which was conducted in a circular pool (120 cm diameter) divided into four quadrants. In the center of the third quadrant (i.e., the target quadrant), a circular platform (i.e., survival platform) with a diameter of 10 cm was placed just below the water surface (1 centimeter). Mice were trained four times a day for the first five days, with quadrant one as the water entry point. The time for mice to find the survival platform within 60 seconds was recorded. On the sixth day, the survival platform was removed, and the time spent in each quadrant and locomotion of the mice were recorded.\n
\n\n For the novel object recognition test\n \n \n 45\n \n \n , a cube (side length of 50 cm) was used, and two identical objects (i.e., old objects) were placed symmetrically at a position 10 cm from the sidewall. Mice were placed with their backs to the objects from the perpendicular bisector of the two objects, and the exploration time of the mice was recorded for 7 minutes. Before placing the next mice, the chamber was cleaned with 75% ethanol. The mice were trained for three days. On the fourth day, one of the old objects was replaced with a novel object and the exploration time and path were recorded. The results are represented by the novel object recognition index (RI), which was calculated as follows:\n
\n\n RI = (time to explore the new object)/ (time to explore the new object\u2009+\u2009time to explore the old object) \u00d7 100% (2)\n
\n\n Data acquisition utilized detection and analysis software of Shanghai Xinsoft Information Technology Co., Ltd.\n
\n\n Immunohistochemistry was performed as previously described\n \n \n 46\n \n ,\n \n 47\n \n \n . After sucrose dehydration, brain tissue was embedded with optimal cutting temperature compound (OTC, 4583, SAKURA) and sliced into 15 \u00b5m sections (CM1950, Leica, Germany). Anti-6E10 (1:500, Covance) was used to examine the extracellular A\u03b2 deposits, anti-MAP2 (1:1000, Millipore) and anti-SNAP25 (1:500, Synaptic Systems) were used to detect dysfunction in neuronal networks. Brian sections were stained with primary antibodies overnight at 4\u00b0C in a humid chamber, after being washed in PBS, followed by 2 hours of incubation of Cy3-conjugated (Jackson ImmunoResearch) or/and 488-conjugated (Jackson ImmunoResearch) secondary antibodies in the dark at room temperature. Fluorescent images were acquired using a fluorescence microscope (Leica, Germany) or a confocal microscope (FV1200, Olympus, Japan) following coverslipping. The number and the area of senile plaques were quantitatively analyzed by Image J software. For histopathology of major organs, the heart, liver, kidney, spleen, lung, and kidney were isolated and stained with an H&E staining kit (ab245880, Abcam).\n
\n\n A\u03b2\n \n \n 40\n \n \n /A\u03b2\n \n \n 42\n \n \n content was measured using enzyme-linked immunosorbent assay (ELISA) according to the previous reports\n \n \n 11\n \n \n . The right hemisphere was weighed and homogenized in TBS (pH 7.4, 1:12, w/v) containing a complete protease inhibitor cocktail and centrifuged. Afterward, the precipitation was centrifuged in 2% SDS and 70% formic acid. The FA-soluble fraction was neutralized with 1 M Tris (pH 11.0) and then diluted with PBS. TBS-soluble and SDS-soluble fractions were directly diluted with PBS. Quantitation was performed according to the instructions using a Human A\u03b2\n \n \n 40\n \n \n /A\u03b2\n \n \n 42\n \n \n Elisa Kit (E-EL-H0542/ E-EL-M0068km, Elabscience Biotechnology). The optical density of the samples was measured with a microplate reader (BioTek, USA) at 450 nm wavelength, and the content of A\u03b2\n \n \n 40\n \n \n /A\u03b2\n \n \n 42\n \n \n in the brain was calculated as moles per gram of wet tissue.\n
\n\n All results are expressed as mean\u2009\u00b1\u2009standard deviation (SD) from at least three independent experiments. Statistical analyses conducted using GraphPad PRISM (version 9.0). Datasets with only two independent groups were analyzed for statistical significance using unpaired, two-tailed Student\u2019s t test or Mann-Whitney test. Datasets with more than two groups were analyzed using one-way ANOVA, followed by Bonferroni or Tukey post hoc test. Datasets with two independent factors were analyzed using two-way ANOVA, followed by Bonferroni post hoc test. All p values below or equal to 0.05 were considered significant. * p\uff1c0.05, ** p\uff1c0.01, and *** p\uff1c0.001.\n
\n\n The C\n \n \n 3\n \n \n N used in the simulations had a diameter of ~\u20094.5 nm corresponding to the average diameter of C\n \n \n 3\n \n \n N measured in the experiments (\n \n Figure S8\n \n ). The initial A\u03b2\n \n 42\n \n peptide crystal structure was taken from RCSB Protein Data Bank (PDB ID: 1Z0Q)\n \n \n 48\n \n \n (\n \n Figure S8\n \n ). To investigate the effect of C\n \n 3\n \n N on A\u03b2\n \n 42\n \n aggregation, two A\u03b2\n \n 42\n \n peptides were simulated in the absence or presence of C\n \n 3\n \n N. In the system without C\n \n 3\n \n N (control system), two peptides were solvated into a 9.6 nm \u00d7 9.1 nm \u00d7 6.5 nm water box containing 17,911 water molecules. The \u201cpeptides\u2009+\u2009C\n \n 3\n \n N\u201d system was derived from its counterpart, by randomly adding a C\n \n 3\n \n N with a minimum distance of 1.5 nm to any heavy atom of the peptide. Then, two A\u03b2\n \n 42\n \n peptides\u2009+\u2009C\n \n 3\n \n N were solvated into a water box (9.6 nm \u00d7 9.1 nm \u00d7 8.2 nm) containing 22,604 water molecules. Na\n \n +\n \n and Cl\n \n \u2012\n \n ions were added to the solvent to neutralize systems and mimic the physiological conditions of 0.15 mol/L NaCl. The detailed illustration of the initial system is shown in Figure S7.\n
\n\n The MD simulations were carried out using the GROMACS-4.6.6\n \n 49\n \n software package with AMBER99SB-ILDN force field\n \n \n 50\n \n \n . The VMD software was adopted to visualize the trajectories and configurations of the MD simulations\n \n \n 51\n \n \n . The TIP3P water model was adopted for solvent molecules\n \n \n 52\n \n \n . Long-range electrostatic interactions were conducted with the particle mesh Ewald method\n \n \n 53\n \n \n . The van der Waals (vdW) interactions were calculated with a smooth cutoff distance of 1.2 nm. Each solvated system was first minimized using the conjugate gradient method and succeeded by a 10 ns NPT relaxation at 300 K and 1 bar. During production runs, the simulation temperature and pressure were fixed at 300 K and 1 bar with the v-rescale thermostat and Parrinello\u2009\u2212\u2009Rahman coupling scheme\n \n \n 54\n \n ,\n \n 55\n \n \n , respectively. A time step of 2.0 fs was used, and coordinates were collected every 20 ps. For each system, three independent 1000 ns trajectories were collected for the analysis. Periodic boundary conditions were introduced in all directions. All solute bonds were constrained at their equilibrium values by employing the LINCS algorithm\n \n \n 56\n \n \n , and water geometry was constrained with the SETTLE algorithm\n \n \n 57\n \n \n .\n
\n\n Incorporating passive heating structures into personal thermal management technologies could effectively mitigate the escalating energy crisis. However, the current passive heating materials struggle to balance thickness and insulating capability, resulting in compromised comfort, space efficiency, and limited thermoregulatory performance. Here, a novel air-gelation strategy, is developed to directly synthesize ultrathin and self-sustainable heating metafabric with 3D dual-aerogel structural network during electrospinning. Controlling the interactions among polymer, solvent, and water enables the microphase separation of charged jets, while adjusting the distribution of carbon black nanoparticles within charged fluids to form fibrous networks composed of interlaced aerogel micro/nanofibers with heat storage capabilities. With an ultrathin thickness of 0.18 mm, the integrated metafabric exhibits exceptional thermal insulation performance (15.8 mW m\n \n \u22121\n \n K\n \n \u22121\n \n ), superhydrophobicity, enhanced mechanical properties, and high breathability while maintaining self-sustainable radiative heating ability (long-lasting warming of 8.8 \u2103). This strategy provides rich possibilities to develop advanced fibrous materials for smart textiles and thermal management.\n
\n\n The escalating energy crisis, amplified by high energy consumption in heating, highlights the urgency of personal thermal management strategies to reduce the energy\u00a0demand for indoor temperature regulation, by regulating the heat exchange between the human body and the environment\n \n 1,2\n \n .\u00a0Fibrous materials, with their unique accessibility and versatility, have gained prominence in the realm of thermal management materials, presenting innovative opportunities for enhancing energy efficiency and comfort in personal and industrial applications\n \n 3,4\n \n . However, common fibrous materials exhibit uncontrolled pore structures, large pore size (usually >50\u00a0\u03bcm), and limited porosity (typically <50%), which pose great challenges in achieving highly efficient thermal management performance\n \n 5,6\n \n . In contrast to conventional fibers, micro/nanofibers with smaller diameters demonstrate significant potential in effective thermal management applications owing to their reduced pore size (always >2 \u03bcm) and enhanced porosity that effectively trap still air while significantly restricting heat transfer\n \n 7\n \n \n -9\n \n . Recently,\u00a0the 3D micro/nanofibrous sponges prepared by freeze-drying technology or electrospinning method, achieving a fluffy structure with improved porosity and uniform pore structure, which extends the heat transfer path and enhances thermal insulation (thermal conductivity of\u00a0~28 mW m\n \n \u22121\n \n K\n \n \u22121\n \n )\n \n 10-13\n \n .\u00a0Nevertheless, the macropore network of these fibrous sponges is difficult to be refined,\u00a0coupled with the inherent non-porous structure within the fibers,\u00a0thereby limiting the effective suppression of air molecule heat transfer\n \n 14,15\n \n . Furthermore, their excessive thickness (>20 mm) compromises sweat transmission and joint mobility of the body, leading to poor wet comfort (usually\u00a0<2 kg m\n \n \u20132\n \n d\n \n \u20131\n \n ) and space utilization, indicating the necessity for design enhancements that meet practical performance requirements\n \n 16,17\n \n .\n
\n\n In comparison, aerogels, with their high porosity, and nanoscale pore size smaller than the mean free path of air molecules, have a thermal conductivity (~16 mW m\n \n \u22121\n \n K\n \n \u22121\n \n ) that is even lower than that of static air (~24 mW m\n \n \u22121\n \n K\n \n \u22121\n \n ), thus they are regarded as ideal material for thermal insulation\n \n 18-20\n \n . Despite their efficient heat insulation properties, the inherent brittleness and hygroscopic nature of zero-dimensional aerogel powders impose limitations on their practicality in wearable technology\n \n 21-23\n \n .\u00a0To address the challenge, several aerogel materials composed of micro/nanofibers have been developed in recent years\n \n 18,24-\n \n \n 26\n \n . Typically, synthesized through freeze-spinning technique, aerogel fibers exhibit aerogel-like porous structure while maintaining the flexibility of traditional fibers\n \n 27-2\n \n \n 9\n \n . Nevertheless, their suboptimal pore size (>500 nm) falls short in effectively impeding the movement of air molecules, a feat achieved by traditional silica aerogels with their significantly smaller pore size (<60 nm)\n \n 30-32\n \n . Additionally, their large diameters (>200 \u03bcm) restrict the arrangement possibilities within fabric construction, leading to significant gaps, uncontrollable porous structure between fibers, and the resulting uneven thermal insulation\n \n 33,\n \n \n 34\n \n .\u00a0These factors collectively contribute to moderate thermal conductivity and limited warmth retention, while the complex fabrication process further restricts their applications. Moreover, their inability to absorb solar and human body radiation leads to energy inefficiency, thereby limiting their broader application in thermal management as they can only impede heat transfer rather than store or regulate it\n \n 35,36\n \n . Therefore, significant efforts are required to devise a simple and practical method, capable of preserving the fine-pore structure of aerogels and the flexibility of fibers, while ensuring effective utilization and storage of radiant energy.\n
\n\n After careful observation of sunflower growth, we discovered that heliotropism of the flower disc and the Fibonacci sequence of seeds allowed for optimal absorption and storage of light and chemical energy\n \n 37\n \n . Inspired by these features, we ingeniously in situ introduced seed-like and size-matching CB nanoparticles into the nanopores of transparent PMMA fibers by the humidity-induced heterogeneous electrospinning, resulting in the direct construction of the metafabric in one step. The unique Knudsen effect of interconnected nanopores (20-60 nm) and the multi-scattering of nanoparticles facilitate heat energy storage in aerogel fibers, achieving nanoscale Anderson localization within the multi-porous regions centered around nanoparticles. As a result, the obtained metafabric exhibits excellent passive heat storage performance with approximately ~65% radiant energy retention, while maintaining an ultra-low thermal conductivity of 15.8\u00a0mW m\n \n \u20131\n \n K\n \n \u20131\n \n .\u00a0In addition, the physical interactions between nanoparticles and nanopores confer excellent mechanical properties upon the metafabric, enabling it to withstand 100 washing cycles and 1000 bucking cycles at a large strain of 50% without failure. Moreover, the metafabric also exhibits exceptional moisture permeability, as evidenced by its water vapor transmission (WVT) rate of\u00a03.6 kg m\n \n \u20132\n \n d\n \n \u20131\n \n , along with superhydrophobic properties demonstrated by water contact angle (WCA) of 150\u00b0. Remarkably, these characteristics are maintained while ensuring an ultrathin thickness (<0.2 mm). Consequently, this simple, scalable, and efficient metafabric for passive radiative heating not only significantly enhances clothing comfort in cold environments but also reduces energy consumption to help address the energy crisis.\n
\n\n \n Design and processing of metafabric.\n \n We developed the aerogel-structured micro/nanofiber metafabric, similar to a sunflower, which could gather the energy from sun and human body while storing heat around the human body (Fig.\n \n 1\n \n a and Supplementary Fig.\u00a01), achieving a wearable and long-lasting passive radiative heating. Our metafabric was designed based on three principles: (\n \n 1\n \n ) to maximize the collection of radiation, the metafabric must be able to absorb radiation in different wavelength bands simultaneously and efficiently; (\n \n 2\n \n ) to obtain heat storage in a limited space, the carrier of absorbers must be equipped with connected nanopores that locally suppress heat loss and reflect radiation from absorbers; (\n \n 3\n \n ) to be wearable comfortably and cope with different environments, the metafabric should be constructed using a fibrous network characterized by appropriate porous structure and hydrophobic external surface, while maintaining high porosity to facilitate efficient moisture transport. To satisfy the first requirement, as shown in Fig.\n \n 1\n \n a, CB nanoparticles (NPs) rich in \u03c0 electronic structure and carbon-carbon bonds was selected as the absorber of solar and human body radiation\n \n 37\n \n . Meanwhile, PMMA aerogel fibers composed of nanopores with size of 30\u201360 nm were set as the carrier of CB NPs (~\u200960 nm), which can use the transparent properties of PMMA to increase the irradiation depth of radiation, and inhibit the movement of air molecules by Knudsen effect\n \n 30\n \n . More importantly, the interconnective nanopores and uniformly embedded CB synergistically create an Anderson localization effect, as illustrated in Fig.\n \n 1\n \n a, effectively confining radiant heat within the fiber\n \n 38\n \n . The last requirement was satisfied by designing well-connected micro/nanofiber network with properly sized pore, which reflects human body radiation through Mie scattering while ensuring the its softness, continuity, and mechanical enhancement\n \n 39\n \n .\n
\n\n
\n\n The synthetic fabrication of metafabric involved four components: PMMA, CB, water-based fluoropolymer (WF), and DMAc, as presented in Supplementary Fig.\u00a02. Initially, the hydrophobic agent is blended into the PMMA solution, which is then heated and stirred in water bath at 60 \u2103. To enhance the dispersion of CB, the mixture is subjected to ultrasonic treatment. Finally, the prepared solution was used to synthesize metafabric in one step by our unique dual air-gelation technique. Manipulation of the phase inversion behavior of charged jets by electrospinning allowed for the direct creation of transparent and size-customized fibrous network composing of intertwined PMMA micro/nanofibers (Supplementary Fig.\u00a03). With the addition of WF, moisture from the air permeates the hydrophobic jets, triggering phase separation. The PMMA molecular chains align and aggregate around the solvent-enriched phase, intertwining physically. This gelation process forms a molecular network within the jets, generating a nanoporous structure inside the fiber (Fig.\n \n 1\n \n a). Figure\n \n 1\n \n b-c shows the ultra-thin thickness (~\u2009180 \u00b5m), scattering holes from fiber network, interconnective nanopores and evenly dispersed CB from aerogel fiber of the metafabric mentioned above. Further examination through transmission electron microscopy (TEM) uncovers the interconnected nature of these nanopores and the presence of CB nanoparticles deep within the aerogel fibers, as depicted in Fig.\n \n 1\n \n d. Complementing this, the optical and microscope photograph confirm the remarkable optical transparency of the PMMA fiber (Supplementary Fig.\u00a04). The elemental composition and spatial distribution within the membranes were analyzed through energy-dispersive X-ray spectroscopy (EDS) mapping. This technique confirmed the uniform attachment of WF in the membranes, with the fluorine element being completely surrounded by carbon and oxygen, as illustrated in Fig.\n \n 1\n \n e. Raman spectroscopic analysis of the metafabric provides further insight into the metafabric (Supplementary\n \n Fig.\u00a05\n \n ). The smaller ID (at 1350 cm\n \n \u2212\u20091\n \n ) to IG (at 1583 cm\n \n \u2212\u20091\n \n ) peak ratio indicates a higher degree of crystallinity and fewer structural defects\n \n 40\n \n . The innovative design of our metafabric with dual-aerogel networks provides an extended spectral response, covering a broad wavelength range from 0.1 to 15 \u00b5m (Fig.\n \n 1\n \n f). This enables the metafabric to absorb wavelengths from various sources, including the sun and the human body, across ultraviolet (UV), visible light (VL), near-infrared ray (NIR), and mid-infrared ray (MIR) spectra. Owing to these tailored structures and physical properties, the metafabric exhibits thermal insulation performance comparable to down feathers, yet with a thickness only 1% of the latter, demonstrating the incredible capacity for sustained heat generation. Moreover, the metafabric has a greater heating depth in the thickness direction and higher surface temperatures on both upper and lower layers compared to ordinary heating fabrics, as depicted in Fig.\n \n 1\n \n h. This indicates a more efficient absorption of light energy, showing the superior performance of our metafabric in heat management applications. In addition to its remarkable comprehensive performance, the metafabric can be produced on a large scale with dimensions of 0.6 \u00d7 1.5 m\n \n 2\n \n (Fig.\n \n 1\n \n i), based on our unique air-gelation technique.\n
\n\n \n Synthesis and nanostructures of metafabric\n \n . The synthesis of metafabric with the molecular networks and micro/nanofiber networks depended on water diffusion, molecular chain movement, and phase separation within the charged jets, as shown in Fig.\n \n 2\n \n a. Initially, in the absence of WF, which also acts as an anionic surfactant, results in the PMMA solution with higher surface tension. As water diffuses from the outside into the jet, the oxygen atoms in the ester groups (-COOCH\u2083) of PMMA exhibit some hydrophilicity, attracting water to the PMMA surface. This interaction, coupled with solvent evaporation, results in the solidification of the jet into solid fibers. Upon the addition of a hydrophobic agent, the ionization of CH\n \n 2\n \n =\u2009CHCOO\n \n \u2212\n \n in the solution leads to a decrease in surface tension, enhancing the mobility of the molecular chains\n \n 41\n \n . As water permeates, the fluorocarbon chains within the jet likely promote the aggregation of polymer molecules, leading to the formation of pores or microphase separation regions in the solution\n \n 42\n \n . These pores are typically enveloped by polymer, while the hydrophobic agent is repelled to the edges, thereby uniformly distributing water and solvent in the spherical arrangement within the jet. This distribution facilitates phase separation, and the lower surface tension allows this process to occur more rapidly\n \n 43\n \n . The stretching of the jet and the evaporation of the solvent culminate in the formation of a porous structure. The formation of an interconnected nanoscale porous structure composed of molecular chain networks is achieved following the disappearance of the solvent-rich phase and the solidification of the polymer-rich phase. This fabrication process is substantiated by SEM observations of the fiber cross-section, which show varying degrees of phase separation, as shown in Fig.\n \n 2\n \n b. With higher hydrophobic agent concentrations, the solution exhibits increased viscosity and electrical conductivity, alongside decreased surface tension. However, at the hydrophobic agent concentration of 9%, a sharp increase in both viscosity and conductivity is noted, which adversely affects the stability of the jet (Fig.\n \n 2\n \n c). Additionally, excessive surface tension can lead to overly rapid phase separation. This rapid separation prevents the timely formation of smaller pores, resulting in an increase in the diameter of the pores within the fibers, which is detrimental to the material inhibiting the movement of air molecules, ultimately leading to a decrease in thermal insulation performance\n \n 44\n \n .\n
\n\n
\n\n The construction mechanism of our material can be further elucidated through an examination of the phase separation behavior exhibited by various PMMA solutions. As depicted in Fig.\n \n 2\n \n d, the introduction of the hydrophobic agent accelerates the precipitation of PMMA, expanding the non-stable region of solution phase separation and quickening the phase separation process. This is evidenced by the binodal curve of the hydrophobic solution, which shows the closer proximity to the initial composition compared to the hydrophilic solution, indicating a faster initiation rate of phase separation\n \n 45\n \n . In order to further investigate the diverse pore structures of PMMA fibrous membranes, we conducted nitrogen physisorption analysis at a temperature of 77 K (Fig.\n \n 2\n \n e). The sorption behavior of PMMA aerogel micro/nanofiber membranes (AMMs) containing 6 wt% WF was characterized by a type II isotherm. Initial gradual N\n \n 2\n \n uptake at P/P\n \n 0\n \n <\u20090.9 indicated minimal interaction between nitrogen molecules and the fibers, reflecting a low micropore count in the AMMs. A marked N\n \n 2\n \n absorption increase at higher relative pressures (P/P\n \n 0\n \n >\u20090.9) highlighted the abundance of mesopores. Additionally, the distribution of water is more prevalent on the surface than in the interior of the jet, leading to the earlier solidification of the surface solution. This results in the formation of a porous structure in the inner layers, yielding fibers with a skin-core structure, consistent with the SEM observations, as shown in Fig.\n \n 2\n \n b. The study quantified the surface area and pore volume of AMMs, revealing a higher BET surface area of 68.38 m\n \n 2\n \n g\n \n \u2212\u20091\n \n compared to PMMA fibrous membranes (Supplementary\n \n Fig.\u00a06\n \n ). The pore size distribution (PSD) of the AMMs was determined using density functional theory (DFT) calculations. Interestingly, the PSD results align closely with the estimated size range observed in the SEM images (Fig.\n \n 2\n \n b). This alignment between theoretical calculations and empirical observations confirms the presence of abundant nanoscale pores, ranging from 30 to 60 nm, in the PMMA aerogel fibers.\n
\n\n \n Figure. 2f\n \n illustrates the impact of varying CB concentrations on the viscosity and surface tension of the PMMA mixture. On the other hand, achieving a uniform distribution of CB is crucial for obtaining fibers with energy storage properties\n \n 46\n \n . As shown in Fig.\n \n 2\n \n g, we propose the following hypotheses regarding the movement, force, and distribution of CB nanoparticles throughout the process of fiber formation in the solution. Initially, due to their high surface energy, CB nanoparticles tend to aggregate together. When introduced into the PMMA solution, the polar nature of DMAc (solvent) stabilizes the particles and prevents further aggregation. The addition of WF leads to a decrease in surface tension while heating, stirring, and ultrasonic treatment result in an even dispersion of particles throughout the solution. During electrospinning, water diffuses from the exterior to the interior of the jet, causing the hydrophobic nature of CB to interact with the hydrophobic components\n \n 47\n \n . The weakly negative charge on CB leads to mutual repulsion under an electric field, further dispersing the particles. As the particle size of CB is close to that of the pores, when the jet undergoes phase separation to form aerogel fibers, CB becomes embedded in some of the nanopores, forming nanoscale local closed-pore structures in tandem with the nanopores. Moreover, the PMMA aerogel fibers have fewer nanoparticles at lower concentrations of CB, preventing closed-pore structures and radiation absorption (Fig.\n \n 2\n \n h). However, at 3 wt% CB, nanopores are embedded with CB particles, forming effective nanoscale structures without particle aggregation. Increasing CB to 4.5 wt% leads to a viscosity exceeding 1.1 Pa s, surpassing the spinnable range for electrospinning solutions, preventing continuous fiber formation. As presented in Fig.\n \n 2\n \n i, the metafabric exhibits a fiber inter-pore distribution that closely matches the wavelengths of human body radiation, enabling it to absorb body radiation while reflecting part of the mid-infrared emissions through Mie scattering\n \n 39\n \n .\n
\n\n \n Mechanical performances of metafabric\n \n . Conventional supercritical drying methods yield silica aerogels and fibrous sponges obtained by freeze-drying that lack stretchability, failing to meet the mechanical requirements for human consumption, which is a crucial challenge that must be overcome in the development of aerogel fibers\n \n 48\n \n . The tensile stress-strain curves (Fig.\n \n 3\n \n a) demonstrate a notable negative correlation between tensile strength and WF content, but the positive correlation with the addition of CB. First, the hydrophobic agent causes the formation of porous structure inside PMMA fibers, resulting in a slight decrease in their tensile properties. Surprisingly, adding an optimal amount of CB (3 wt%) enhances the tensile fracture stress by approximately 54%, increasing it from 1.37 to 2.12 MPa, even surpassing the strength of solid PMMA fibers. Then, the metafabric displayed no visible plastic deformation and effectively maintained its initial maximum stress when subjected to 1000 bending deformation cycles at buckling strain of 50%, indicating excellent fatigue resistance against repetitive bending and buckling, as depicted in Fig.\n \n 3\n \n b. Based on these findings, we propose a particle mechanics enhancement mechanism (Fig.\n \n 3\n \n c). In conventional aerogel fibers under tensile stress, the stress initially concentrates on the polymer pore walls, leading to their fracture followed by the breakdown of the surface layer, culminating in total fiber failure. In contrast, for energy storage fibers, tensile stress initially transfers from the pore walls to the CB nanoparticles embedded within the pores. This dispersion of stress makes the pore walls more resistant to fracture, ultimately enhancing the overall mechanical properties of the fibers\n \n 46\n \n . The durability of the metafabric was rigorously tested through the 100-cycle washing process with a compressive strain (\u03b5) of 50%, as presented in Fig.\n \n 3\n \n d. Remarkably, the plastic deformation after these cycles was only 6.2%, which only exhibited a slight increase than the values without washing. Furthermore, the metafabric demonstrated outstanding stability, maintaining static stress levels between 100\u201391% and exhibiting nearly no plastic deformation (<\u20091%) throughout the cyclic washing process. Additionally, after 100 washing cycles, the metafabric showed almost no change in either size or weight (weight loss\u2009<\u20095%), indicating a robust integration of CB nanoparticles within the PMMA aerogel fibers, as shown in Fig.\n \n 3\n \n e. This stability is attributed to the hydrophobic interactions between CB, PMMA, and WF, resulting in a strong bond within the fiber structure\n \n 46\n \n . The BET surface area and pore volume of the metafabric also exhibited minimal changes post-washing, underscoring its exceptional reusability (Fig.\n \n 3\n \n f).\n
\n\n
\n\n Moreover, the metafabric exhibits the WVT rate of ~\u20093.6 kg m\n \n \u2212\u20092\n \n d\n \n \u2212\u20091\n \n , which is 4.5 times higher than that of 3M Thinsulate insulation materials and surpasses other thermal insulation textiles. Moreover, the metafabric exhibits good flexibility (Supplementary\n \n Fig.\u00a07\n \n ), and its breathability significantly exceeds that of other commercial fabrics, making it a superior choice in terms of air permeability (Fig.\n \n 3\n \n g). This suggests that the metafabric not only excels in thermal insulation but also meets the everyday comfort requirements of consumers. Compared to the traditional textile materials, as illustrated in Supplementary Fig.\u00a03, using PMMA as a carrier for heating particles significantly enhances radiation transmittance. Specifically, within the 200\u2013800 nm wavelength range, the average VIS emissivity of the metafabric reaches an impressive 0.9. This represents an increase of approximately five times and four times compared to EFM (0.18) and TH (0.24), respectively (Supplementary\n \n Fig.\u00a08\u20139\n \n ). Moreover, in comparison to commercially available high-end radiative heating fabrics, it has achieved a 50-fold reduction in thickness while enhancing its radiation absorption efficiency by 30%. These findings highlight the exceptional capability of metafabric to absorb solar radiation, demonstrating its superior performance in radiative heating. As shown in Supplementary\n \n Fig.\u00a010\u201311\n \n , under the synergistic hydrophobic effects of CB (carbon-carbon bonds) and WF (carbon-fluorine chains), this material exhibits an extraordinary superhydrophobic property, characterized by an impressive contact angle of 150\u00b0, significantly enhancing durability and providing exceptional water and stain resistance\n \n 49\n \n . The metafabric, in conjunction with its exceptional moisture permeability, offers comprehensive functional and comfortable protection for human activities (Supplementary\n \n Fig.\u00a012\n \n ).\n
\n\n \n Thermal regulation performance of metafabric\n \n . In view of the dual-aerogel network structure, exceptional radiation absorbency, and superior comfort, the metafabric exhibits promising potential for applications in smart textiles and personal thermal management. First, we studied the thermal management performance of the metafabric in a light-free environment. As illustrated in Fig.\n \n 4\n \n a and Supplementary\n \n Fig.\u00a013\n \n , the metafabric, with a mere thickness of 180 \u00b5m and high porosity (91.3%), exhibits an ultra-low thermal conductivity (15.8 mW m\n \n \u2212\u20091\n \n K\n \n \u2212\u20091\n \n ), significantly lower than other insulation materials and even below that of stationary air. The thermal conductivity of a material alone is insufficient to fully evaluate its heat transfer capabilities, as this measure could not show the effects of radiation and thermal convection on thermal management\n \n 50\n \n . Therefore, we set up a thermal resistance tester and xenon lamp experimental apparatus in a controlled temperature and humidity indoor environment to evaluate the comprehensive insulation capability of materials under different lighting conditions (Fig.\n \n 4\n \n b-c). The thermal resistance of AFM increased by 73% in the absence of sunlight due to its nanoporous structure. The metafabric exhibited a thermal resistance more than twice that of AFM, attributed to the absorption of mid-infrared radiation by CB and its synergistic effect with nanopores (Fig.\n \n 4\n \n b). Moreover, the metafabric's thermal resistance is an impressive 16 times greater than that of commercial thermal underwear, while its thickness is only one-third of the latter. This suggests that the nanopores inside the aerogel fibers efficiently prevents air molecules from moving and from transferring heat by the Knudsen effect (Supplementary\n \n Fig.\u00a014\n \n ). As demonstrated in Fig.\n \n 4\n \n c and Supplementary\n \n Fig.\u00a015\n \n , the metafabric dramatically increased to 0.588 K m\n \n 2\n \n W\n \n \u2212\u20091\n \n under simulated sunlight, owing to the absorption of visible and near-infrared light by the carbon black. At this stage, the thermal resistance of the metafabric was 178 times greater than a standard T-shirt and about 46% higher than a 3M fibrous sponge, while being only one percent as thick, highlighting its exceptional ultra-thin insulation and heating capabilities. Upon turning off the simulated sunlight, the thermal resistance of the metafabric decreased to 65% of its maximum value but remained 40% higher compared to the scenario with continuous absence of sunlight. This performance contrasts with that of ordinary heating fabrics, which showed a decrease to 29% of their original thermal resistance, nearly the same as their performance without sunlight exposure at the beginning.\n
\n\n Furthermore, we conducted direct thermal measurements under a Xenon lamp to evaluate the outdoor radiative heating performance of our metafabric. The temperature of each fabric sample was accurately monitored affixing three K-type thermocouples onto a copper plate, ensuring uniformity in thermal measurements (Fig.\n \n 4\n \n d). During the period from 9:00 to 10:30, after the simulated sunlight was activated, the temperature of the metafabric was consistently higher compared to other materials. As shown in Fig.\n \n 4\n \n e, it was approximately 2.3, 5.6, 5.8, 7.2, and 9.3 \u2103 higher than that of the Heating TU, TU, T-shirt, EFM, and bare skin simulators, respectively. During this phase, both the metafabric and the 3M fibrous sponges, which are 100 times thicker, maintained the highest temperatures. On the other hand, the 3M fibrous sponges maintained the simulated skin surface temperature primarily due to their high thickness, which provides a prolonged heat transfer path\n \n 51\n \n . From 10:30 to 12:00, after turning off the simulated sunlight, the temperature of the metafabric remained significantly higher compared to other materials. Specifically, it was about 3.8, 4.2, 4.9, 6.2, and 9.0 \u2103 higher than the Heating TU, TU, T-shirt, EFM, and bare skin simulators, respectively (Fig.\n \n 4\n \n e). Moreover, over the entire 3-hour period of the simulated sunlight exposure and shutdown, the temperature of the metafabric decreased by only 2.2 \u2103. In comparison to conventional radiative heating materials like HTU, which is six times thicker and experienced a temperature drop of 3.7 \u2103, the metafabric showed a more than 71% improvement in light energy storage efficiency, validating the energy storage characteristics of its unique structure. The translucency of PMMA, combined with the Anderson positioning induced by the small-sized nanopores and carbon black nanoparticles, facilitates radiation penetration into the fiber network and enables efficient absorption and storage of thermal radiation by the metafabric, as depicted in Fig.\n \n 4\n \n f.\n
\n\n Besides, we investigated the absorptive capability of metafabric for human-emitted radiation within the 5 to 14 \u00b5m wavelength spectrum (Fig.\n \n 4\n \n g). The test results showed that the metafabric had a higher infrared emissivity compared to ordinary heating fabrics. The rate of infrared temperature rise was over 73% higher than that of ordinary heating fabrics, demonstrating the superior mid-infrared absorption capability of the metafabric. To vividly demonstrate the heating performance of the metafabric, we utilized a heating plate and silicone pad to simulate human skin and employed an infrared thermal imaging camera to observe the surface temperature under different lighting conditions, as depicted in Fig.\n \n 4\n \n h. The thermal camera revealed a substantial temperature difference between the metafabric (38.1 \u2103) and TU (31.1 \u2103) after 30 seconds of illumination. After removing the materials, the simulated skin covered by the metafabric was 2.7 \u2103 warmer than that covered by TU and nearly 2 \u2103 warmer than that covered by the 3M fibrous sponge, suggesting that the metafabric could be an effective alternative to thicker insulating materials. As illustrated in Supplementary\n \n Fig.\u00a014\n \n , compared to conventional fabrics, porous fibers, and aerogel fibers, our energy storage fiber possesses both the ability to suppress heat at the molecular level and slow-release heating functionality, offering a distinct advantage in thermal management applications\n \n 52\n \n . We also compared our metafabric with commercial heating textiles in terms of adiabaticity, scalability, flexibility, penetrability, radiation absorption, lightness, thinness, and scalability\n \n 53\n \n . In comparison (Fig.\n \n 4\n \n i), our metafabric exhibits an exceptional hydrophobicity, more efficient radiation absorption, and better warmth retention performance than commercial thermal insulation materials at ultra-thin thickness.\n
\n\n
\n\n We have presented the facile methodology for the direct synthesis of self-sustainable radiative heating metafabric based on aerogel-structured micro/nanofibers using the unique dual air-gelation technique, which overcomes the longstanding challenge in practical application of fragile aerogel in thermal managament textiles. By regulating water diffusion, molecular chain movement, and phase separation within the charged jets, the dual-aerogel-structured metafabric was assembled. This metafabric was fabricated through the entanglement and interlacing of PMMA/CB aerogel micro/nanofibers, aiming to achieve a synergistic effect in terms of radiation absorption, insulation, and heating properties. With an ultrathin overall thickness of only 180 \u00b5m, our energy storage aerogel micro/nanofibers exhibit far lower thermal conductivity (15.8 mW m\n \n -1\n \n K\n \n -1\n \n ) and higher heating effect (8.8 \u2103) compared with the existing aerogel fibrous materials. Benefiting from the Anderson localization formed by nanopores (30\u201360 nm) and CB, the metafabric demonstrates self-sustainable radiative heating (>\u200965% solar heat retention rate). Together with its other outstanding features such as enhanced mechanical properties (plastic deformation of nearly 0% over 1000-cycle washing), super hydrophobicity (WCA of 150\u00b0), high moisture permeability (WVT of 3.6 kg m\n \n -2\n \n d\n \n -1\n \n ), and strong self-adaptability, with more optimization, we believe that the metafabric could demonstrate potential applications in various emerging applications such as smart textiles and personal thermal management.\n
\n\n \n Materials.\n \n PMMA powder (M\n \n w\n \n = 500000) was supplied by Shanghai yuanye Bio-Technology Co., Ltd. Fluorinated polyurethane (FPU, QF66) was obtained from Shanghai Taifu Chemical Co., Ltd. DMAc were provided by Shanghai Aladdin Chemistry Co., Ltd. Carbon black NPs (CB, d\u2009~\u200960 nm), were bought from Tianjin Zhengningxin Material Technology Co., Ltd.\n
\n\n \n Fabrication of the metafabric.\n \n PMMA powder was dissolved in DMAc to form a 30 wt% solution, followed by the addition of hydrophobic agents at 0, 1.5, 3, and 4.5 wt% concentrations. Then, CB particles were added at 0, 3, 6, and 9 wt% concentrations. After stirring at room temperature for 8 hours, the mixture underwent an hour of ultrasonic treatment. Afterwards, the aerogel fiber membranes were prepared using an electrospinning platform equipped with an humidity-controlled system, applying 30 kV voltage and extruding the solutions at 4 mL h\n \n \u20131\n \n . Electrospinning was conducted in a lab at a constant 23\u2009\u00b1\u20092 \u2103 and 55\u2009\u00b1\u20095% relative humidity.\n
\n\n \n Characterization.\n \n Microstructure analyzed using FE-SEM (Hitachi S-4800, Japan), TEM (FEI Tecnai F20, USA), and EDS (Bruker 610M, USA). Raman spectroscopy was performed using a LabRAM HR Evolution system at 532 nm excitation. Pore structure evaluated using a capillary flow porometer (CFP-1100AI, USA) and physisorption analyzer (Micromeritics ASAP 2460, USA). Porosity was determined through the equation: porosity = (\n \n \u03c1\n \n \n 0\n \n -\n \n \u03c1\n \n )/\n \n \u03c1\n \n \n 0\n \n \u00d7\u2009100%, where\n \n \u03c1\n \n represents the bulk density of the fibrous structures and\n \n \u03c1\n \n \n 0\n \n the density of polymer chips. Viscosity, conductivity, surface tension measured using a rotary viscometer (LVDV-1T), conductivity meter (FE30), and tensiometer (QBZY). Mechanical properties were assessed with a dynamic mechanical analyzer (Q850, TA Instruments, USA). Water vapor permeability tested using moisture permeability tester (YG601H, China), upright cup configuration. The WCA was measured with a goniometer (Kino SL200B, USA). UV-vis-NIR absorbance of the metafabric was recorded using a spectrophotometer (UV-3600, Shimadzu Ltd., Japan) with an integrating sphere. Solar desalination tests conducted using a solar simulator (PLS-SXE 300, Perfectlight Ltd., China), adjustable light intensity (3\u2009~\u200912 kW m\n \n \u2212\u20092\n \n ). Infrared images captured with an IR camera (Fluke-TiS75, USA). FTIR analyses were conducted on a Nicolet iS50 Spectrometer (Thermo Fisher Scientific Inc., USA), equipped with a Pike golden hemisphere integrating sphere, an MCT (Mercury Cadmium Telluride) detector cooled with liquid nitrogen, and a gold reference for background. The far-infrared emissivity and temperature increase measured using emissivity tester (DR915G, Wenzhou Darong Textile Instrument Co. Ltd., China) and temperature-rise tester (DR915W, Wenzhou Darong Textile Instrument Co. Ltd., China). Thermal conductivity and thermal resistance assessed using thermal constants analyzer (Hot Disk TPS2500, Sweden) and textile thermal transmittance tester (YG606E-\u2161, China).\n
\n\n Aqueous zinc-ion batteries (AZIBs) are increasingly recognized as a sustainable alternative to lithium-ion batteries (LIBs) due to their abundance, safety, and lower environmental impact. However, the hydrogen evolution reaction (HER) and uncontrolled diffusion of Zn\n \n 2+\n \n and SO\n \n 4\n \n \n 2-\n \n ions lead to the dendrite formation and side reactions, which hinder their practical application by forming a non-conductive layer on the Zn anode. This layer impedes the ion transport and electron flow, reducing the Coulombic efficiency (CE) for the Zn nucleation. Here, to simultaneously regulate the diffusion of H\n \n +\n \n , Zn\n \n 2+\n \n , and SO\n \n 4\n \n \n 2-\n \n in the electrolyte, an ion-sieving accelerating channel was constructed to unify the Zn deposition by introducing an eco-friendly layer-by-layer self-assembly of a flocculant poly(allylamine hydrochloride) (PAH) and its tautomer poly(acrylic acid) (PAA). The dual-ion channels, created by strong electrostatic interactions between carboxylate anions (COO\u207b) and ammonia cations (NH\u2083\u207a), promote the uniform Zn deposition along the (002) plane, exhibiting a CE of 99.8% after 1600 cycles in the Zn||Cu asymmetric cell. With the facile fabrication of the layer-by-layer self-assembled Zn anode, an Ah-level pouch cell (17.36 Ah) with a high mass loading (> 8 mg cm\u207b\u00b2) demonstrated exceptional performance, retaining a capacity of 93.6% for at least 250 cycles at 1.7 C. This research offers a universal strategy for optimizing electrode mechanisms and advancing the manufacturing process of eco-friendly, high-performance aqueous batteries.\n
\n\n Aqueous zinc-ion batteries (AZIBs) are regarded as one of the most promising alternatives to lithium-ion batteries for the grid-scale electrochemical energy storage (EES) systems due to their high volumetric capacity (5855 mAh cm\n \n \u2212\u20093\n \n ), low redox potential (-0.762 V vs standard hydrogen electrode (SHE)), and high safety.\n \n 1,2\n \n However, the hydrogen evolution reaction (HER) leads to a rapid rise in the local concentration of OH\n \n \u2212\n \n at the anode/electrolyte interface, which would further react with SO\n \n 4\n \n \n 2\u2212\n \n in the electrolyte to form the by-product (Zn\n \n 4\n \n SO\n \n 4\n \n (OH)\n \n 6\n \n \u22c5xH\n \n 2\n \n O, ZHS).\n \n 3\n \n Moreover, the generation of the inert by-product would reduce the active sites for Zn deposition, increase the nucleation barrier, and cause uncontrollable dendrite growth on the Zn anode. This results in the shortened cycle life and has hindered the commercial application of ZIBs.\n \n 4\n \n
\n\n Surface modification using inorganic and organic coatings could effectively inhibit the dendrite growth and side reactions at the Zn anode.\n \n 5\n \n Inorganic coatings including carbon-based materials, eutectic alloys, and metallic compounds (e.g., carbon dots (CDs),\n \n 6\n \n Zn-Cu,\n \n 7\n \n Zn-Sn,\n \n 8\n \n CaCO\n \n 3\n \n ,\n \n 9\n \n ZnF\n \n 2\n \n \n 10\n \n ) could be used as physical barriers to protect the Zn anode from corrosion and regulate Zn\n \n 2+\n \n diffusion to achieve uniform Zn deposition. However, the non-uniform physical barriers lead to a low Zn\n \n 2+\n \n conductivity and a significant volume change during plating/stripping, ultimately causing the cracking and peeling. In contrast, the flexible organic polyelectrolyte coatings such as polyamide,\n \n 11\n \n polyacrylamides,\n \n 12\n \n and poly(2-vinylpyridine)\n \n 13\n \n with 3D cross-linked polymer channels can provide active sites to facilitate Zn\n \n 2+\n \n transference and reduce the interface resistance.\n \n 14\n \n But the mono polyelectrolyte interface could not satisfactorily control the ion diffusion and offer a sufficient mechanical strength. For instance, although anionic polyelectrolytes could effectively regulate Zn\n \n 2+\n \n flux to homogenous deposition, it has a limited repelling effect on SO\n \n 4\n \n \n 2\u2212\n \n in the electrolyte, which would cause the by-products formation to a certain extent.\n \n 15,16\n \n Owing to the repulsion between anionic polyelectrolytes and the negative charged Zn anode, the adhesivity of the coatings is also not satisfactory. Based on this, the layer-by-layer (LbL) self-assembly of polyelectrolytes with the controllable composition and tunable properties, which allows sequential deposition of versatile polycations and polyanions on a charged substrate, is an attractive approach to enhance the overall performance of Zn anodes\n \n 17\u201319\n \n . The strong electrostatic interactions between polycations and polyanions could provide oppositely charged dual-ion channels to suppress the corrosion and passivation on the Zn anode, while also enhancing the mechanical strength (e.g., toughness, adhesion, self-healing).\n \n 20\u201322\n \n In addition, as summarized in Scheme\n \n 1a\n \n , the LbL self-assembly technique has outstanding advantages compared with other conventional surface modification techniques in the commercialization of AZIBs. The resource of polycations and polyanions with the characteristics of non-toxic, biocompatible, and biodegradable for the LbL self-assembled SEI layer are extremely abundant, which is not only conductive to the preparation of novel multifunctional SEI layers through the modification of polyelectrolytes, but also can promote the development of eco-friendly Zn-ion batteries. The LbL method also allows precise control of the thickness and composition of coatings, making it a sustainable method that is far superior to other surface modification techniques.\n \n 23,24\n \n Moreover, the LbL self-assembly technique is more cost-effective for the practical application due to its simple manufacturing process and low demand on equipment.\n \n 25\n \n However, to date, there have been limited studies on the use of LbL self-assembly technique for the interface engineering of Zn anodes. Identifying efficient and appropriate polyelectrolyte combinations for the LbL self-assembled layers remains a challenge, as is demonstrating their effectiveness in protecting Zn anodes and enhancing their applications.\n
\n\n
\n\n To overcome above mentioned challenges, we selected poly(allylamine hydrochloride) (PAH) and its tautomer poly(acrylic acid) (PAA) to prepare the LbL self-assembled PAH/PAA multilayers. Due to the tautomerization, the photon exchange would occur between the carboxylic acid group (-COOH) of PAA and the amine group (-RNH\n \n 2\n \n ) of PAH, leading to the negatively charged carboxylate (-COO\n \n \u2212\n \n ) and the positively charged ammonium (-RNH\n \n 3\n \n \n +\n \n ) with the strong electrostatic interactions.\n \n 26\n \n Based on this, the PAH/PAA multilayers could be seen as the dual-ion channels for SO\n \n 4\n \n \n 2\u2212\n \n and Zn\n \n 2+\n \n in the electrolyte, like an ionic sieve, the dual-ion channels sieve SO\n \n 4\n \n \n 2\u2212\n \n at the first shell and attract Zn\n \n 2+\n \n , which would regulate the mobility and dispersion of Zn\n \n 2+\n \n and suppress the side reactions, thereby improving the electrochemical performance of the Zn anode. Moreover, the strong electrostatic interactions between PAH and PAA could effectively improve the mechanical strength without affecting the ionic conductivity of the coating, and also simplify the preparation process.\n \n 19,27\n \n As illustrated in Scheme\n \n 1b\n \n , the preparation sequence of multilayers is to first coat the PAH layer and then the PAA layer (Anode\n \n \u2212\n \n \u2212 PAH\n \n +\n \n \u2212 PAA\n \n \u2212\n \n ), followed by a rinsing process after each coating to remove the weakly associated bound chains. Designing in the sequence: Anode\n \n \u2212\n \n \u2212 PAH\n \n +\n \n \u2212 PAA\n \n \u2212\n \n would enable a high adhesion to the negatively charged Zn anode and increase the zincophilicity of multilayers. PAA layer as the outer layer could first accelerate the desolvation process and regulate the diffusion of Zn\n \n 2+\n \n . Meanwhile, PAH layer would capture SO\n \n 4\n \n \n 2\u2212\n \n due to the low binding energy, resulting in the formation of ion-sieving accelerating channels to inhibit the HER and by-products. Remarkably, the LbL self-assembled PAH/PAA multilayers are favorable for the preferential nucleation and growth of Zn\n \n 2+\n \n along Zn(002) surface to form a smooth and dense deposition layer and suppress the dendrite formation (Scheme\n \n 1c\n \n ). Correspondingly, the PAH/PAA multilayers enable an excellent Coulombic efficiency of up to 99.8% after 1,600 cycles at 0.5 mA cm\n \n \u2212\u20092\n \n and 0.25 mAh cm\n \n \u2212\u20092\n \n for the Zn||Cu asymmetric cell. The Zn||MnO\n \n 2\n \n battery with the PAH/PAA coating layers displays an outstanding specific capacity of about 137 mAh g\n \n \u2212\u20091\n \n over 1,000 cycles with 91.3% capacity retention at 2 A g\n \n \u2212\u20091\n \n . Even in a Zn||VO\n \n 2\n \n pouch cell with a high loading mass, it exhibits an excellent discharge capacity of 17.36 Ah over 250 cycles at 1.7 C. This work provides a new insight for the surface modification of Zn anodes, the design of the LbL self-assembly of polyelectrolytes could not only effectively enhance the electrochemical performances and mechanical strengths of Zn anode, but also could be applied to other metal anode protection.\n
\n\n The LbL self-assembled PAH/PAA multilayers were prepared by the doctor-blading method.\n \n 28\u201330\n \n By optimizing the preparation process, three double PAH/PAA layers (Zn@PAH/PAA) with a thickness of ~\u2009280 nm could offer the most stable electrochemical performances (Figure\n \n S1\n \n and Figure S3). The composition of Zn@PAH/PAA was successfully confirmed by Fourier Transform Infrared (FTIR) spectroscopy, as shown in Figure S2. The bands at 1710 cm\n \n \u2212\u20091\n \n and 1247 cm\n \n \u2212\u20091\n \n are related to the C\u2009=\u2009O and C-O stretching vibration of carboxylic acid from the characterization bands of PAA, respectively.\n \n 31,32\n \n The bending of amine and amide group from the characterization bands of PAH are about 1633 cm\n \n \u2212\u20091\n \n and 1532 cm\n \n \u2212\u20091\n \n .\n \n 32,33\n \n Compared with the bands of pure Zn@PAA and Zn@PAH, the significant band shifts for Zn@PAH/PAA are observed, which are due to the electrostatic interactions between polyelectrolytes during the LbL self-assembly process. To investigate the effect of the LbL self-assembled PAH/PAA multilayers on the behavior of Zn\n \n 2+\n \n plating/stripping, the thermostability and Zn\n \n 2+\n \n transport kinetics of Zn anodes after coating were compared. The linear polarization curves indicate that the PAH/PAA multilayers could enhance the corrosion resistance of Zn anodes (Fig.\n \n 1\n \n a). The corrosion potential of Zn@PAH/PAA is increased from \u2212\u20090.1 V to -0.984 V, and the corrosion current is reduced to 1.109 mA, which is lower than that of the Bare Zn (2.99 mA). The higher corrosion potential and lower current mean more effective inhibition on the HER and by-products on Zn anodes.\n \n 34\n \n To further verify the corrosion resistance of Zn@PAH/PAA, the HER on both Bare Zn and Zn@PAH/PAA electrodes within the initial 20 min at 30 mA cm\n \n \u2212\u20092\n \n was observed, which is shown in Figure S4. After 10 min, small bubbles generate and trend to accumulate on the Bare Zn electrodes. In contrast, no obvious bubbling appears on the Zn@PAH/PAA electrode. Moreover, the XRD pattern of Bare Zn after 50 cycles exhibits the strong diffraction peaks of the by-product (Zn\n \n 4\n \n SO\n \n 4\n \n (OH)\n \n 6\n \n \u22c54H\n \n 2\n \n O, ZHS), which was not detected on the cycled Zn@PAH/PAA electrode (Fig.\n \n 1\n \n b). The results from SEM-EDS mapping in Figure S5 also show that there is large amount of ZHS on the Bare Zn electrode after 50 cycles, compared with the Zn@PAH/PAA electrode. These results indicate that the PAH/PAA multilayers could significantly suppress the HER and side reactions to enhance the thermostability of Zn anodes. In addition, ion-sieving accelerating channels could modulate interfacial kinetics of Zn\n \n 2+\n \n diffusion and deposition. As shown in Figure S6, the PAH/PAA multilayers offer an excellent hydrophilicity, of which the contact angle (58.6\u00b0) is smaller than that of Bare Zn (99.9\u00b0). The hydrophilicity of Zn@PAH/PAA could enable a lower interfacial energy barrier to regulate the diffusion of Zn\n \n 2+\n \n , which is demonstrated in the analysis of activation energy.\n \n 35\n \n Based on the EIS plots at different temperatures (Figure S7), the interfacial activation energy (\n \n E\n \n \n \n a\n \n \n ) was evaluated through the Arrhenius equation (Fig.\n \n 1\n \n c). The hydrophilic PAH/PAA multilayers could reduce\n \n E\n \n \n \n a\n \n \n from 17.34 kJ mol\n \n \u2212\u20091\n \n to 6.85 kJ mol\n \n \u2212\u20091\n \n , indicating that the PAH/PAA layers with a high zincophilicity could effectively regulate the Zn\n \n 2+\n \n solvation structure and accelerate the transference. The Zn transference numbers of Zn@PAH/PAA and Bare Zn were calculated and shown in Fig.\n \n 1\n \n d and Figure S8, where the high ionic conductivity of PAH/PAA multilayers could increase the Zn transference number from 0.284 to 0.481. To investigate the nucleation and growth behaviors of Zn\n \n 2+\n \n , the nucleation overpotential (\u03b7) on the Zn||Ti cell was evaluated (Fig.\n \n 1\n \n e). According to previous research,\n \n 11,36\n \n the critical nucleation radius (\u03b3\n \n crit\n \n ) and nucleation rate (\u03c9) could be described as below:\n
\n\n Where h is the height of Zn atom,\n \n \n \\(\\sigma\\)\n \n \n is the interface tension, A is the Zn atom mass,\n \n \n \\(\\rho\\)\n \n \n is the nucleus density, F is Faraday\u2019s constant, and L is Avogadro constant. As illustrated in Fig.\n \n 1\n \n e, \u03b7 is increased by 12.1 mV with the PAH/PAA multilayers, and the ratio of \u03b3\n \n crit\n \n for Zn@PAH/PAA and Bare Zn is 0.47, which is attributed to an increased nucleation rate. Hence, a high nucleation overpotential could be attributed to a uniform and dense Zn deposition. Moreover, the chronoamperometry (CA) test reflects that Zn\n \n 2+\n \n exhibits a 2D diffusion behavior on Zn@PAH/PAA, compared with a 3D diffusion on the Bare Zn (Fig.\n \n 1\n \n f). The current change with time indicates the increase of effective Zn nucleation sites in chronoamperograms. The current for Zn@PAH/PAA remains stable after 140 s, whereas due to the aggregation of Zn\n \n 2+\n \n , the current for the Bare Zn continues to decrease within 400 s. Combined with the result of \u03b7, the PAH/PAA multilayers could regulate the Zn nucleation sites and make a uniform deposition to efficiently inhibit dendrite growth.\n
\n\n Since PAH/PAA multilayers could enable the 2D diffusion and plating of Zn\n \n 2+\n \n , the texture evolution and morphology of Zn anodes during cycling were further studied. XRD patterns of Zn@PAH/PAA under different cycles in Fig.\n \n 2\n \n a and table\n \n S1\n \n reveal that the (002) peak increases significantly after cycling. The intensity ratio I\n \n (002)\n \n /I\n \n (101)\n \n becomes stronger, which is from 0.0611 (pristine Zn) to 0.1995 (15 cycles), 0.2298 (30 cycles), and 0.2973 (50 cycles). Density functional theory (DFT) calculations were carried out to analyze the adsorption energy of Zn\n \n 2+\n \n and PAH\n \n +\n \n /PAA\n \n \u2212\n \n on Zn(002). PAH\n \n +\n \n /PAA\n \n \u2212\n \n exhibits a stronger adsorption energy (-0.760 eV) than Zn\n \n 2+\n \n (-0.145 eV) on Zn(002), which is shown in Fig.\n \n 2\n \n b. Besides, the diffusion energy barrier of Zn\n \n 2+\n \n with the PAH/PAA multilayers coated on Zn(002) increases from 0.014 eV to 0.269 eV, suggesting that Zn@PAH/PAA could inhibit the aggregation of Zn\n \n 2+\n \n and lead to a 2D diffusion and parallel plating.\n \n 37\n \n These above results indicate that the PAH/PAA multilayers could induce the preferential nucleation and growth of Zn\n \n 2+\n \n along Zn(002). SEM images show that with the increase of cycles, more and more (002) textures are observed on the Zn@PAH/PAA electrode (Fig.\n \n 2\n \n d). There are many horizontal (002) textures stacked together on the Zn@PAH/PAA after 50 cycles, where the thickness of the deposition layer is about 8 \u00b5m (Figure S9). In sharp contrast, the morphology of Bare Zn after 50 cycles is extremely uneven with significant dendrite growth corresponding to a deposition layer of around 11 \u00b5m (Figure S10).\n \n In-situ\n \n optical images and 3D depth profiles also verify that the PAH/PAA layers could guide a smooth and dense plating (Figure S11 and Figure S12). The aggregation and uneven nucleation of Zn\n \n 2+\n \n on the Bare Zn electrode result in the dendrite formation and significantly roughens the surface after 5 min, while the surface of the Zn@PAH/PAA electrode remains smooth and homogenous for 20 min. A solid electrolyte interphase (SEI) layer with a thickness of about 19 nm for the 50 cycled Zn@PAH/PAA electrode was observed through TEM images, shown in Fig.\n \n 2\n \n e. Further zooming in the SEI layer, the (002) textures are the main plating orientation in SEI layer, of which the area is significantly larger than that of the (100) textures. Therefore, it could be known that the ion-sieving accelerating channels formed by the LbL self-assembled PAH/PAA layers could induce the Zn nucleation and deposition along the (002) lattice plane to form a smooth and dense Zn flake layer.\n
\n\n As the PAH/PAA multilayers could effectively enhance the thermostability and Zn\n \n 2+\n \n transport kinetics to make the uniform Zn(002) deposition, the stability of Zn@PAH/PAA was discussed in relation to the symmetric Zn cell. The Zn@PAH/PAA electrode exhibits an excellent stability around 1200 h at 1 mA cm\n \n \u2212\u20092\n \n and 1 mAh cm\n \n \u2212\u20092\n \n , whereas the Bare Zn electrode suffers short circuit around 76 h (Fig.\n \n 3\n \n a). With increasing the current and capacity densities to 5 mA cm\n \n \u2212\u20092\n \n and 5 mAh cm\n \n \u2212\u20092\n \n , the Zn@PAH/PAA electrode (340 h) still shows a longer cycling performance than the Bare Zn electrode (160 h), illustrated in Fig.\n \n 3\n \n b. Furthermore, the Zn@PAH/PAA electrode presents a high depth of discharge (DOD): 53.4% within around 170 h cycling at 8 mA cm\n \n \u2212\u20092\n \n and 4 mAh cm\n \n \u2212\u20092\n \n (Fig.\n \n 3\n \n c). As shown in Figure S13, the voltage profile for the Zn@PAH/PAA electrode at different current density and capacity exhibits a larger potential difference than that of the Bare Zn, which is due to the lower nucleation radius and higher nucleation rate of Zn@PAH/PAA as mentioned in Fig.\n \n 1\n \n e. In addition, a large voltage difference of the Zn@PAH/PAA electrode may also be caused by the PAH/PAA layers inducing the orientational plating of Zn\n \n 2+\n \n along Zn(002). Compared with the Bare Zn electrode, the rate performance of the Zn@PAH/PAA exhibits an outstanding Zn\n \n 2+\n \n plating/stripping stability at various current densities from 1 mA cm\n \n \u2212\u20092\n \n to 10 mA cm\n \n \u2212\u20092\n \n at 1 mAh cm\n \n \u2212\u20092\n \n (Figure S14). These GCD tests indicate that the PAH/PAA multilayers enable an excellent stability and reversibility for the Zn anode. The Coulombic efficiency (CE) was analysed by the asymmetric Zn||Cu cell. As shown in Fig.\n \n 3\n \n d and Fig.\n \n 3\n \n e, the initial CE of Cu@PAH/PAA (92.3%) is higher than that of bare Cu (86.7%), and the Cu@PAH/PAA electrode offers an outstanding CE around 99.8% after 1,600 h, while the bare Cu electrode suffers a rapid decline on CE after about 70 cycles. The CE performance indicates that the PAH/PAA multilayers could efficiently inhibit the side reactions and passivation of Zn anodes. Hence, the initial nucleation overpotential of Cu@PAH/PAA increases from 0.0746 V to 0.0961 V, which once again approves that PAH/PAA layers could enable a lower nucleation radius and a higher nucleation rate, thereby inducing the deposition of the homogenous and dense Zn(002) layer. The cumulative plated capacity (CPC) of the Cu@PAH/PAA electrode is 396 mAh cm\n \n \u2212\u20092\n \n , which is more competitive than most of recent research on the surface modification of Zn anodes (Fig.\n \n 3\n \n f and Table S2).\n
\n\n To investigate the mechanism of Zn\n \n 2+\n \n plating/stripping behaviours on Zn@PAH/PAA,\n \n in-situ\n \n Raman spectra were recorded during each cycle. As shown in Figure S15, the band assigned to the -CH\n \n 2\n \n stretching vibration is at 2928 cm\n \n \u2212\u20091\n \n for Zn@PAH/PAA, while it is at about 2930 cm\n \n \u2212\u20091\n \n for Zn@PAH.\n \n 38\n \n This difference is due to the electrostatic interactions between PAH and PAA polyelectrolytes as confirmed by FTIR. After immersing Zn@PAH/PAA to 2 M ZnSO\n \n 4\n \n for 15 min, the -CH\n \n 2\n \n stretching vibration band moves to 2931 cm\n \n \u2212\u20091\n \n , indicating the ionic interaction between -CH\n \n 2\n \n -NH\n \n 3\n \n \n +\n \n and SO\n \n 4\n \n \n 2\u2212\n \n . Moreover, the band shape between 1400 cm\n \n \u2212\u20091\n \n and 1450 cm\n \n \u2212\u20091\n \n is related to the -RNH\n \n 3\n \n \n +\n \n deformation and -CH\n \n 2\n \n bending, by which the shape change further confirms the interaction between -CH\n \n 2\n \n -NH\n \n 3\n \n \n +\n \n and SO\n \n 4\n \n \n 2\u2212\n \n .\n \n 39,40\n \n The band from 1750 cm\n \n \u2212\u20091\n \n to 1600 cm\n \n \u2212\u20091\n \n for Zn@PAH/PAA corresponds to the vibration of symmetric C-H of PAH and C\u2009=\u2009O in carboxylate groups of PAA, while it splits into two bands after immersing with ZnSO\n \n 4\n \n , owing to the ionic interaction between -COO\n \n \u2212\n \n and Zn\n \n 2+\n \n that enhances the intensity of C\u2009=\u2009O band.\n \n 41,42\n \n The periodic band changes could be observed in each plating/stripping cycle, as illustrated in Fig.\n \n 4\n \n a and Table S3. The ionic interaction between SO\n \n 4\n \n \n 2\u2212\n \n and -CH\n \n 2\n \n -NH\n \n 3\n \n \n +\n \n would make the band of -CH\n \n 2\n \n - vibration shift to a lower wavenumber during Zn\n \n 2+\n \n plating and to a higher wavenumber during Zn\n \n 2+\n \n stripping. Correspondingly, the relative band intensity of -CH\n \n 2\n \n bending and NH\n \n 3\n \n \n +\n \n deformation between 1400 cm\n \n \u2212\u20091\n \n and 1450 cm\n \n \u2212\u20091\n \n would change. Moreover, the coordination of Zn\n \n 2+\n \n and -COO\n \n \u2212\n \n would make the band of C\u2009=\u2009O vibration move to a lower wavenumber during plating, while the band would move to a higher wavenumber due to the escape of Zn\n \n 2+\n \n during stripping. These regular and reversible band changes indicate the formation of dual-ion channels between PAH and PAA polyelectrolytes. The binding energy was calculated to further discuss the interfacial mechanism. As shown in Fig.\n \n 4\n \n b, the binding energy of [PAA\n \n \u2212\n \n \u2212 Zn(H\n \n 2\n \n O)\n \n 4\n \n ]\n \n +\n \n decreases from \u2212\u200912.913 eV to -15.700 eV, combined with the activation energy calculation (Fig.\n \n 1\n \n c), which indicates that PAA\n \n \u2212\n \n would coordinate with Zn\n \n 2+\n \n to form the solvation structure of [PAA\n \n \u2212\n \n \u2212 Zn(H\n \n 2\n \n O)\n \n 4\n \n ]\n \n +\n \n , thereby regulating the Zn\n \n 2+\n \n diffusion. The interaction of SO\n \n 4\n \n \n 2\u2212\n \n and Zn@PAH/PAA was also investigated. The binding energy of [Zn(SO\n \n 4\n \n )\n \n 2\n \n ]\n \n 2\u2212\n \n is much larger than that of ZnSO\n \n 4\n \n (-10.068 eV and \u2212\u20093.785 eV, respectively), suggesting that Zn\n \n 2+\n \n is likely to bind with two SO\n \n 4\n \n \n 2\u2212\n \n to form [Zn(SO\n \n 4\n \n )\n \n 2\n \n ]\n \n 2\u2212\n \n (Figure S16). In addition, compared with [(PAH\n \n +\n \n )\n \n 3\n \n SO\n \n 4\n \n ]\n \n +\n \n and [(PAA\n \n \u2212\n \n )\n \n 1\n \n SO\n \n 4\n \n ]\n \n 3\u2212\n \n , [(PAH\n \n +\n \n )\n \n 3\n \n Zn(SO\n \n 4\n \n )\n \n 2\n \n ]\n \n +\n \n exhibits the lowest binding energy (-26.33 eV), indicating that SO\n \n 4\n \n \n 2\u2212\n \n would bind with PAH\n \n +\n \n to form the stable [(PAH\n \n +\n \n )\n \n 3\n \n Zn(SO\n \n 4\n \n )\n \n 2\n \n ]\n \n +\n \n coordination structure (Fig.\n \n 4\n \n c). Based on the above results, the interfacial mechanism during Zn\n \n 2+\n \n plating/stripping on the Zn@PAH/PAA electrode is illustrated in Fig.\n \n 4\n \n d. The ion-sieving accelerating channels in the structure of PAH\n \n +\n \n \u2212 SO\n \n 4\n \n \n 2\u2212\n \n \u2212 Zn(H\n \n 2\n \n O)\n \n 4\n \n \n 2+\n \n \u2212 PAA\n \n \u2212\n \n is constructed by the LbL self-assembled PAH/PAA multilayers, where PAA\n \n \u2212\n \n would regulate the Zn\n \n 2+\n \n solvation shell and accurate Zn\n \n 2+\n \n transport at the inner Helmholtz plane, and PAH\n \n +\n \n would bind with SO\n \n 4\n \n \n 2\u2212\n \n to inhibit the formation of ZHS. Indeed, the PAH/PAA multilayers would also induce the Zn nucleation and deposition along (002) texture to form the uniform and dense Zn flake layer, thereby suppressing dendrite formation.\n
\n\n A Zn||commercial MnO\n \n 2\n \n battery was assembled to investigate the effect of the PAH/PAA multilayers on the electrochemical performance of the full cell. The CV curves at a scan rate of 0.1 mV s\n \n \u2212\u20091\n \n are shown in Figure S17, where two redox peaks are related to the intercalation and de-intercalation of Zn\n \n 2+\n \n and H\n \n +\n \n , respectively.\n \n 43\n \n Because of the large nucleation overpotential and high nucleation rate of Zn@PAH/PAA mentioned earlier, the polarization for the Zn@PAH/PAA battery is larger than that of the Bare Zn battery. The rate performance for both electrodes from 0.1 A g\n \n \u2212\u20091\n \n to 5 A g\n \n \u2212\u20091\n \n is illustrated in Fig.\n \n 5\n \n a and Figure S18. Due to the improved interfacial kinetics and thermostability of Zn anodes, the Zn@PAH/PAA battery exhibits a higher specific capacity and reversibility than the Bare Zn battery at each current density (262 mAh g\n \n \u2212\u20091\n \n at 0.1 A g\n \n \u2212\u20091\n \n and 101 mAh g\n \n \u2212\u20091\n \n at 5 A g\n \n \u2212\u20091\n \n ). Furthermore, the Zn@PAH/PAA battery displays an outstanding specific capacity of ~\u2009137 mAh g\n \n \u2212\u20091\n \n over 1,000 cycles with 91.3% capacity retention at 2 A g\n \n \u2212\u20091\n \n , whereas the Bare Zn battery suffers a rapid capacity decline of 73 mAh g\n \n \u2212\u20091\n \n after 620 cycles (Fig.\n \n 5\n \n b). These results indicate that the LbL self-assembled PAH/PAA multilayers could significantly inhibit the side reactions and enhance the CE value, thereby improving the overall performance of the batteries. We also further investigate the LbL self-assembly technique in promoting the practical application of AZIBs. As shown in Fig.\n \n 5\n \n c, the Zn@PAH/PAA-commercial VO\n \n 2\n \n pouch cell with the high mass loading (>\u20098 mg cm\n \n \u2212\u20092\n \n ) was assembled, which exhibits an excellent Ah-level residual discharge capacity of 17.36 Ah after 250 cycles at 1.7 C, with a capacity retention of 96.3%. The capacity and C-rate achieved in this work are much higher than most previous work on Zn metal anodes, indicating the remarkable effect and huge potential of the LbL self-assembly technique to improve the anode stability in the practical application of Zn-ion batteries (Fig.\n \n 5\n \n d and Table S4).\n \n 44\u201348\n \n
\n\n In summary, the LbL self-assembled PAH/PAA multilayers with high mechanical strength and ionic conductivity could effectively enhance the reversibility and stability of the Zn anode. The ion-sieving accelerating channels constructed by the multilayers not only enable a high zincophilicity to regulate Zn\n \n 2+\n \n desolvation process, but also could capture SO\n \n 4\n \n \n 2\u2212\n \n to suppress the formation of by-products. Moreover, the PAH/PAA layers could induce Zn deposition along the (002) crystal plane to form a uniform and dense layer, inhibiting dendrite formation. Since the PAH/PAA multilayers remarkably enhance the interfacial Zn\n \n 2+\n \n transport kinetics and thermostability, the Zn||Zn symmetric cell achieves an ultra-long stability over 1200 h at 1 mA cm\n \n \u2212\u20092\n \n and 1 mAh cm\n \n \u2212\u20092\n \n , and the Zn||Cu asymmetric cell exhibits an outstanding Coulombic efficiency of 99.8% and a high CPC of 396 mAh cm\n \n \u2212\u20092\n \n after 1600 cycles at 0.5 mA cm\n \n \u2212\u20092\n \n and 0.25 mAh cm\n \n \u2212\u20092\n \n . Moreover, the PAA/PAH multilayers enable the Zn-MnO\n \n 2\n \n full cell an excellent capacity retention (91.3%) after 1,000 cycles at 2 A g\n \n \u2212\u20091\n \n , and the Zn@PAH/PAA-VO\n \n 2\n \n pouch cell retains a high discharge capacity of 17.36 Ah after 250 cycles at 1.7 C with a high mass loading. We anticipate that this work inspires a new strategy about the construction of ion-sieving accelerating channels through the LbL self-assembly of polyelectrolytes to protect the metal anode, promoting practical applications of aqueous rechargeable batteries.\n
\n\n Scheme 1 is available in the Supplementary Files section.\n
\n\n Chronological age offers an imperfect estimate of the molecular changes that occur with aging. Epigenetic age, which is derived from DNA methylation data, provides a more nuanced representation of aging-related biological processes. This study examines the bidirectional relationship between epigenetic age and the occurrence of brain health events (stroke, dementia, and late-life depression). Using data from the Health and Retirement Study, we analyzed blood samples from over 4,000 participants to determine how epigenetic age relates to past and future brain health events. Study participants with a prior brain health event prior to blood collection were 4% epigenetically older (beta 0.04, SE 0.01), suggesting that these conditions are associated with faster aging than that captured by chronological age. Furthermore, a one standard deviation increase in epigenetic age was associated with 70% higher odds of experiencing a brain health event in the next four years after blood collection (OR 1.70, 95%CI 1.16-2.50), indicating that epigenetic age is not just a consequence but also a predictor of poor brain health. Both results were replicated through Mendelian Randomization analyses, supporting their causal nature. Our findings support the utilization of epigenetic age as a useful biomarker to evaluate the role of interventions aimed at preventing and promoting recovery after a brain health event.\n
\n\n Age remains the principal risk factor for neurodegenerative conditions\n \n 1\n \n and the most substantial non-modifiable determinant for cerebrovascular disease, posing significant challenges to understanding the complex interplay of biological and molecular aging processes with disease risk\n \n 2\n \n . Despite chronological age serving as a conventional marker, recent advancements have introduced more sophisticated measures of aging. Central to these innovations are epigenetic clocks, a novel approach based on the analysis of DNA methylation patterns at CpG sites\n \n 3\n \n . This methylation process chemically alters DNA molecules, thereby modulating gene expression without changing the DNA sequence. In contrast to the DNA sequence, which remains largely unchanged throughout life, DNA methylation exhibits a degree of plasticity, allowing for changes in response to diverse lifestyle and environmental exposures, including established cardiovascular risk factors\n \n 4\n \n .\n
\n\n Epigenetic clocks, derived from weighted aggregation of methylation across select CpG sites, echo the principles of polygenic risk scores, offering a quantifiable measure of biological age\n \n 5\n \n . The selection of CpG sites and their integration into a singular biological age metric is informed by robust statistical models trained on specific outcomes, ranging from chronological age to more complex phenotypes associated with healthspan and lifespan. This approach has led to the development of various epigenetic clocks. Initially, these clocks were calibrated on chronological age\n \n 6\u201310\n \n , but subsequent iterations have focused on broader phenotypes, such as time-to-death\n \n 11\n \n or clinical parameters linked to morbidity and mortality\n \n 3\n \n . Notably, some epigenetic clocks, such as the PhenoAge\n \n 3\n \n , GrimAge\n \n 11\n \n , and Zhang\n \n 12\n \n clocks have demonstrated a superior ability to predict mortality and various health outcomes, significantly surpassing the predictive power of chronological age.\n
\n\n The pursuit of health and longevity is fundamentally tied to the preservation of a healthy brain. In the context of an aging global population, the imperative to sustain brain health becomes paramount, especially given the increased prevalence and incidence of neurological disorders, now the leading cause of disability-adjusted life years worldwide\n \n 13\n \n . Among aging-related brain diseases, stroke, dementia, and late-life depression have the highest prevalence and incidence\n \n 14\n \n , significantly impacting global brain health due to their disruptive effects on normal brain function. These conditions are closely related, sharing risk factors such as smoking, diet, physical activity, and socio-economic health determinants\n \n 15\u201319\n \n , which are also known to influence epigenetic clocks.\n \n 4\n \n Furthermore, stroke, dementia, and late-life depression can act as risk factors for each other, creating a complex web of interacting health problems\n \n 20,21\n \n . Finally, the occurrence of late-life depression has been shown to be associated with cerebral small vessel disease, aligning it with stroke and dementia from a pathophysiological perspective\n \n 22,23\n \n . This intricate relationship has given rise to the view that these conditions should not be treated as isolated outcomes, but as interconnected components of a broader aging process that requires a comprehensive approach\n \n 24,25\n \n . To promote healthy aging, it is thus necessary to deepen our understanding of the relationship between brain health and the systemic manifestations of the aging process.\n
\n\n Given the growing interest in understanding the aging process beyond chronological age and growing importance of brain health as a determinant of healthy aging, we tested the hypothesis that brain health events accelerate epigenetic aging, and conversely, that accelerated epigenetic aging increases the risk of brain health events. Given that the study of DNA methylation in brain health is still in its early stages, research in this field is limited and often involves small sample sizes. To address this, we conducted our analyses using the Health and Retirement Study, a large longitudinal study of older adults that is representative of the U.S. population. The collection of DNA methylation data in 2016 provided a unique opportunity to assess the impact of past brain health events as well as the future risk of such events in relation to epigenetic age. To evaluate the hypothesized bidirectional relationships, we used both traditional epidemiological associations and a genetic mendelian randomization (MR) framework. By leveraging genetic variants as instrumental variables, MR enabled us to support the causality of these associations with a higher level of evidence compared to observational analyses alone\n \n 26,27\n \n .\n
\n\n \n Cohort characteristics\n \n
\n\n The HRS enrolled 42,233 participants between 1992 and 2016. Of these, 4,018 provided blood samples in 2016 and were included in our analyses (Figure 1). Comparison of baseline characteristics between the complete HRS cohort and the subset with DNA methylation (DNAm) data can be found in Supplementary Table 1. The baseline characteristics of the studied population are presented in Table 1\u00a0(mean age: 70, 58% females). The average age at DNAm data acquisition was 70 years, 58% were females, 17% were Blacks, and 5% were Hispanics.\n
\n\n \n First stage: history of brain health events and epigenetic age\n \n
\n\n \n Observational analyses\n \n
\n\n Of the 4,018 participants included in this cross-sectional analysis at the time of blood sample collection in 2016, 342 (8.5%) had a stroke, 298 (7.4%) had dementia, and 322 (8.0%) already had a late-life major depressive episode prior to DNAm acquisition. This resulted in 806 (20.1%) participants with a history of at least one brain health event, including 127 (3.2%) with two events and 13 (0.3%) with all three events. In multivariable linear regression adjusting for age, sex, race/ethnicity, cardiovascular risk factors (BMI, smoking status) and comorbidities (hypertension, diabetes, heart attack, coronary artery disease, angina, congestive heart failure), brain health events were associated with a 4% increase (beta = 0.04, SD = 0.01, p=0.002) in mean normalized epigenetic age (Figure 4 and Table 2). This association was strengthened when only adjusting for age, sex and race/ethnicity, with an 8% increase (beta = 0.08, SD = 0.01, p<0.001) in mean epigenetic age.\n
\n\n In secondary analyses that considered each brain health event type separately, a history of stroke was associated with a 6% increase in epigenetic age (beta = 0.06, SD = 0.02, p=0.001 - Figure S2 and Table S7) after adjusting for demographics, risk factors, and comorbidities. Similarly, a history of dementia was associated with a 4% increase (beta = 0.04, SD = 0.02, p=0.035 - Figure S3 and Table S9). A history of late-life major depressive disorder was not associated with an increase in epigenetic age in the fully adjusted model (beta= 0.01, SD = 0.02, p=0.673 - Figure S4 and Table S11). Also, a history of either stroke or dementia was associated with a 4% increase in mean epigenetic age (beta= 0.04, SD = 0.01, p=0.003 - Figure S1 and Table S5).\n
\n\n \n Sensitivity analysis: late-life depression ascertained with a different age threshold\n \n
\n\n Given the existing variation in the age cutoff used to define late-life depression, in sensitivity analyses we considered an age threshold of 60 instead of 65 at the first major depressive episode. Out of 4,018 participants, 583 (14.5%) had a late-life depression prior to DNAm acquisition and 1,014 (25.2%) had a history of at least one brain health event.\u00a0In multivariable linear regression adjusting for age, sex and race/ethnicity, brain health events were associated with an 8% increase (beta = 0.08, SD = 0.01, p<0.001) in mean normalized epigenetic age.\u00a0After adjusting for cardiovascular risk factors and comorbidities as well, a history of brain health events was associated with\u00a0a 5% increase (beta = 0.05, SD = 0.01, p<0.001) in mean epigenetic age (Table S13).\n
\n\n \n Mendelian randomization analyses\n \n
\n\n Several different MR analyses (Figure 2) confirmed a positive association between genetically determined brain health events and accelerated epigenetic aging. In the primary analysis using 985 independent genetic instruments for brain health events and the inverse variance weighted MR method, genetically determined brain health events were associated with a 11% increase in mean epigenetic age (beta = 0.11, SD = 0.03, P < 0.001 \u2013 Table 3). The weighted median and MR-Egger methods, more conservative analytical approaches that are more robust to horizontal pleiotropy, yielded similar results, with genetically determined brain health events being associated, respectively, with 8% (beta = 0.8, SD = 0.04, P = 0.052) and 10% (beta = 0.1, SD = 0.04, P = 0.01) increases in epigenetic age. The MR-PRESSO global test and the MR-Egger Intercept did not suggest the presence of pleiotropy.\n
\n\n \n Second stage: epigenetic age and subsequent risk of brain health events\n \n
\n\n \n Observational analyses\n \n
\n\n Of the 4,018 participants with DNAm data, 806 (20.1%) had a history of brain health events before 2016 and 245 (6.1%) were missing data after the DNAm acquisition in 2016 (waves 14 and 15), including 116 (2.9%) who died and 129 (3.2%) who were lost to follow-up (Figure 1). Of the 2,967 participants included in the prospective analysis, 81 (2.7%) developed a stroke, 100 (3.4%) developed dementia and 95 (3.2%) developed a late-life major depressive disorder. This resulted in 261 (8.8%) participants developing at least one brain health event over the 4 years of follow-up, including 15 (0.5%) developing two. In multivariable logistic regression adjusting for demographics (age, sex and race/ethnicity), one SD increase in epigenetic age was associated with a 70% increase (OR = 1.70, 95%CI: 1.16-2.50) in the odds of brain health events (Figure 4 and Table 2). The inclusion of cardiovascular risk factors (BMI, smoking status) and comorbidities (hypertension, diabetes, heart attack, coronary artery disease, angina, and congestive heart failure) in this analysis is subject to debate. These factors are known to influence methylation changes and might be implicitly reflected in the baseline estimation of epigenetic age. Therefore, adjusting for these variables could potentially constitute an overadjustment. Nevertheless, a model that additionally accounted for these factors, alongside demographics, indicated that a one SD increase in epigenetic age was still associated with a 48% increase in the odds of brain health events (OR = 1.48, 95% CI: 0.99-2.21 \u2013 Table 2).\n
\n\n In secondary analyses, we observed that epigenetic age acceleration was associated with an increased likelihood of experiencing a combined outcome of stroke and dementia. This association was also observed when stroke and dementia were analyzed separately. However, no such association was found with late-life depression. Specifically, we found a 112% increase in the odds of developing either stroke or dementia (OR = 2.12, 95% CI: 1.35-3.32 \u2013 see Figure S1 and Table S6) for each one SD increase in epigenetic age, after adjusting for demographics. Similar results were obtained when considering stroke (OR = 2.12, 95% CI: 1.12-4.04 \u2013 see Figure S2 and Table S8) and dementia (OR = 1.98, 95% CI: 1.10-3.56 \u2013 see Figure S3 and Table S10) individually. However, for late-life depression, the association was entirely non-significant (OR = 0.80, 95% CI: 0.43-1.52 \u2013 see Figure S4 and Table S12).\n
\n\n \n Mendelian randomization analyses\n \n
\n\n Several different MR approaches (Figure 3) confirmed a positive association between genetically determined epigenetic age and higher odds of brain health events. In the primary analysis using 777 independent genetic instruments and the inverse variance weighted MR method, one SD increase in genetically determined epigenetic age was associated with 15% higher odds of brain health events (OR = 1.15, 95%CI: 1.06-1.25\u00a0\u2013 Table 3). The weighted median method yielded similar results (OR = 1.15, 95%CI: 1.00-1.31), as well as the MR Egger method (OR = 1.15, 95%CI: 1.00-1.31). The MR-PRESSO global test as well as the Egger intercept were not significant, indicating no substantial pleiotropy.\n
\n\n \n Sensitivity analysis: late-life depression ascertained with a different age threshold\n \n
\n\n We replicated the observational analyses with late-life depression ascertained using an age threshold of 60 instead of 65 at the first major depressive episode.\u00a0Out of the 2,779 participants included in the prospective analysis, 121 (4%) developed a late-life depressive disorder and 269 (10%) developed at least one brain health event over the 4 years of follow-up. In multivariable logistic models adjusting for demographics, one SD increase in epigenetic age was associated with a 57% increase (OR = 1.57, 95%CI: 1.07-2.31) in the odds of brain health events.\n
\n\n \n Sensitivity analysis: exclusion of people missing any follow-up waves\n \n
\n\n We replicated the observational analyses excluding those participants missing data for any of the waves 14 and 15, as opposed to only excluding participants missing data for both of the two waves. Of the 4,018 participants with DNAm data, 804 (20%) had a history of brain health event, 245 (6%) died and 394 (10%) were missing data for any of the waves 14 and 15, so this analysis included 2,573 participants. Of these, 79 (3%) developed a stroke, 75 developed dementia (3%), and 78 (3%) developed a late-life major depressive disorder. We observed a similar trend as in the primary analysis with a 1SD increase in epigenetic age leading to a 78% (OR = 1.78, 95%CI: 1.16 -2.72, Table S15) increase in the odds of brain health events after accounting for demographics.\n
\n\n In this two-stage epigenetic study within the Health and Retirement Study, we identified a significant bidirectional relationships between epigenetic aging and brain health events. In the first stage, the cross-sectional analysis revealed an association between a history of brain health events and accelerated epigenetic age. Specifically, patients with a prior history of stroke, dementia, or late-life depression exhibited a statistically significant increase in mean normalized epigenetic age, findings that remained robust after adjusting for a range of covariates. This association was further confirmed through Mendelian Randomization analyses, suggesting a causal linkage. In the second stage, the prospective cohort analysis revealed that individuals with an accelerated epigenetic age were at a substantially higher risk of developing brain health events. This association persisted after comprehensive adjustments for confounders and was also observed in Mendelian Randomization analyses, again providing evidence for a causal relationship. These findings underscore the reciprocal influence between accelerated aging and the manifestation of brain health events, enhancing our comprehension of this complex interplay.\n
\n\n Mounting evidence points to the importance of epigenetic age as a more accurate indicator of true biological aging compared to chronological age\n \n 3,28\n \n . Numerous studies have established that DNA methylation predicts all-cause mortality more accurately than chronological age alone\n \n 29\u201332\n \n . This predictive ability has been first studied using epigenetic data from specific tissues, where methylation patterns are closely linked to disease development. For instance, accelerated epigenetic aging in the dorsolateral prefrontal cortex is associated with increased amyloid accumulation and cognitive decline in Alzheimer\u2019s disease\n \n 33\n \n . Similarly, the progression of osteoarthritis and obesity is reflected in the accelerated methylation patterns of cartilage\n \n 34\n \n and liver tissues\n \n 35\n \n , respectively. Given the challenges and risks associated with tissue-specific sample collection, whole blood samples have become increasingly utilized for determining epigenetic age\n \n 28\n \n . This approach has been validated, showing a high correlation between epigenetic age derived from whole blood and that from specific tissues, making it a reliable proxy for general epigenetic age assessment\n \n 3\n \n . Subsequently, blood-derived epigenetic age acceleration has been linked to the occurrence of various conditions including cancer\n \n 36\u201339\n \n , cardiovascular and coronary heart diseases\n \n 3\n \n , Parkinson's disease\n \n 40\n \n and frailty\n \n 41,42\n \n . In addition, key risk factors such as high blood pressure\n \n 43\n \n , BMI\n \n 35\n \n , triglycerides\n \n 3\n \n , or glucose levels\n \n 3,43\n \n , as well as smoking\n \n 3\n \n and low physical activity\n \n 3,43\n \n have been shown to accelerate aging-related epigenetic modifications. These findings emphasize the influence of environmental factors and the dynamic nature of DNA methylation status. Finally, at a cellular level, DNA methylation clocks have been connected to three of the nine recognized hallmarks of aging\n \n 44\n \n : nutrient sensing, mitochondrial function, and stem cell composition, highlighting their integral role in characterizing the aging process\n \n 45\n \n .\n
\n\n This study adds important new evidence to epigenetic aging research by focusing on a broad observational outcome related to brain health. Stroke, dementia, and late-life depression, the most common aging-related brain conditions, are intricately linked. They share overlapping risk factors, including smoking, diet, physical activity, and socio-emotional health determinants, which contribute to the occurrence of all three\n \n 15\u201319\n \n and a common small vessel disease pathophysiology\n \n 22,23\n \n . Furthermore, the occurrence of one condition markedly increases the likelihood of developing the others: a history of depression heightens the risk of stroke\n \n 46\n \n and dementia\n \n 47\u201349\n \n ; stroke raises the chances of subsequent dementia\n \n 21\n \n or depression\n \n 50\n \n ; and dementia itself is a risk factor for both hemorrhagic stroke\n \n 51\n \n and depression\n \n 52\n \n . This intricate interplay has led to the perspective that these conditions should not be examined in isolation, but rather collectively, as distinct yet connected manifestations of a broader brain health aging process\n \n 24,25\n \n . Our findings lend substantial support to this viewpoint. We demonstrate that an acceleration in the body's epigenetic aging process significantly increases the risk of developing stroke or dementia, but not late-life depression. Because the pace of epigenetic aging can be slowed by lifestyle changes such as diet and exercise\n \n 43\n \n , our results suggest that taking care of our body as we get older is a potentially effective way of preventing brain health events. Moreover, our study reveals that stroke and dementia not only result from, but also contribute to, a general acceleration of epigenetic aging, as evidenced by blood-derived methylation changes. These results underscore the systemic nature of these conditions, suggesting that they should be considered comprehensively, rather than as pure neurological or psychiatric disorders.\n
\n\n Our study also provides important evidence suggesting that the association between epigenetic aging and brain health are causal,\u00a0as demonstrated by the results of our MR analyses.\u00a0MR is an epidemiological method that leverages DNA sequence variants as instrumental variables, offering a powerful means to deduce potential causal links between exposures and outcomes\n \n 26,27\n \n . By employing genetic variants that are randomly assigned during meiosis and remain constant throughout an individual's life, MR effectively acts as a form of natural randomization. This approach is particularly valuable as it helps to counteract confounding by environmental factors and reverse causation, which are prevalent sources of bias in observational studies. Consequently, MR serves as a valuable tool, complementing observational studies by adding a layer of evidence to suggest the causal nature of observed relationships\n \n 53\n \n . However, it is important to acknowledge that MR does not replace randomized controlled trials, which are still the gold standard for establishing causal associations. MR provides a crucial bridge in the hierarchy of scientific proof, particularly in scenarios where conducting trials is impractical or unethical.\n
\n\n Our findings pave the way for new research directions, particularly in exploring how epigenetic clocks can aid in the early detection of individuals at elevated risk of poor brain health. Currently, observational risk scores and polygenic risk scoring are widely recognized methods for categorizing individuals into different risk groups\n \n 54\n \n . Our study suggests that epigenetic clocks could fulfill a similar role and could potentially be integrated with other risk scores to enhance the precision in predicting those most susceptible to brain health events. This combined approach could significantly facilitate early intervention strategies. Furthermore, there is potential for therapeutic interventions focused on modulating the epigenetic aging process itself, with the goal of preventing aging-related observational events. Recent research in mice has shown that DNA methylation clocks can be reversed through epigenetic reprogramming, leading to notable increases in life expectancy\n \n 55\n \n . This underscores the profound influence of epigenetic modifications on the aging process as a whole. Such breakthroughs open possibilities for the development of targeted treatments that not only manage but also proactively mitigate the risks of aging-related neurological conditions by addressing their underlying epigenetic mechanisms.\n
\n\n The primary strength of our study is the utilization of the Health and Retirement Study, which is among the largest and best characterized cohorts with DNA methylation data. Acquiring DNA methylation data is often a costly endeavor, leading to smaller datasets that typically require integration with other datasets to reach sufficient power\n \n 11\n \n . The Health and Retirement Study\u2019s substantial size, combined with its demographic representativeness of the US population, significantly bolsters the generalizability of our findings to older Americans. Additionally, the application of MR analyses enabled us to strengthen our observational results, providing a more compelling argument for the causal nature of the relationships we identified. However, our study is not without limitations. First, although we adjusted for cardiovascular risk factors and comorbidities, we cannot rule out the possibility that unaccounted risk factors may be influencing the observed acceleration in epigenetic aging or the increased risk of brain health events. Second, our cross-sectional observational analysis is likely influenced by survival bias. It's reasonable to assume that survivors of brain health events are generally healthier and may demonstrate slower epigenetic aging compared to non-survivors. This factor could potentially skew our results towards the null hypothesis.\n
\n\n In conclusion, our findings using high quality data from the Health and Retirement Study cohort establish robust, bidirectional associations between epigenetic aging and brain health events. We have established that a history of stroke, dementia, or late-life depression is not only associated with accelerated epigenetic aging but also that an advanced epigenetic age increases the likelihood of these conditions. Through Mendelian Randomization analyses, we provide strong evidence supporting the causal nature of these relationships. Overall, our study makes a significant contribution to the understanding of aging-related brain health. It underscores the critical role of epigenetic factors and opens new pathways for future research and observational applications, particularly in early risk assessment and intervention strategies.\n
\n\n \n Study design\n \n
\n\n We conducted a 2-stage observational and genetic study nested within the HRS. Our goal was to investigate two different hypotheses: first, that persons who have survived brain health events, including stroke, dementia, and late-life depression, exhibit epigenetic age acceleration; and second, that those with accelerated epigenetic aging are at an elevated risk for subsequent brain health events. Both hypotheses were examined through a combination of observational and genetic analyses. To investigate the first hypothesis, we performed a nested cross-sectional analysis on HRS participants who had available DNA Methylation data. This allowed us to assess the association between survival from brain health events and epigenetic aging. To test the second hypothesis, we implemented a prospective cohort design using the same HRS group with available methylation data. This design enabled us to observe whether individuals with accelerated epigenetic aging were more likely to experience subsequent brain health events. The genetic analyses for both stages were conducted using one-sample Mendelian randomizations within the HRS cohort.\n
\n\n \n The Health and Retirement study\n \n
\n\n The HRS is an ongoing, longitudinal study that is nationally representative of older adults in the United States. Its primary aim is to provide a comprehensive understanding of the health and economic circumstances associated with aging at both individual and population levels. The HRS sample was assembled in several waves of enrollment and data collection. The HRS sample was compiled through multiple phases of recruitment and data collection. The inaugural cohort, enrolled in 1992, included individuals born between 1931 and 1941 (who were then aged 51-61), along with their spouses of any age. Subsequently, a distinct study named \"Asset and Health Dynamics Among the Oldest Old\" (AHEAD) was conducted, focusing on the cohort born between 1890 and 1923 (who were then aged 70 and above). In 1998, these two samples were merged and supplemented with the addition of two more cohorts: the \"Children of the Depression\" (CODA, born 1924-1930) and the \"War Babies\" (born 1942-1947). This was done to ensure the sample accurately represented the U.S. population over the age of 50. Later, the \"Early Baby Boomers\" (EBB, born 1948-1953) and the \"Mid Baby Boomers\" (MBB, born 1954-1959) were added in 2004 and 2010, respectively. The most recent addition was the \"Late Baby Boomers\" (LBB, born 1960-1965) in 2016\n \n 56\n \n . As of now, the HRS has successfully enrolled over 40,000 participants. Among these, nearly 20,000 have provided DNA samples, and DNA Methylation (DNAm) data has been obtained from 4,000 participants. The study conducts biennial interviews with participants, covering a broad range of variables such as income, employment, disability, physical health and functioning, and cognitive functioning. Further details about the HRS and its survey design can be found elsewhere\n \n 57\n \n . The study's protocol has received approval from the University of Michigan's institutional review board, and informed consent has been obtained from all participants.\n
\n\n \n Analytic sample\n \n
\n\n The present study utilized a subset of participants from the HRS who had available DNA Methylation (DNAm) data. DNAm assays were conducted on a non-random subsample of 4,018 individuals who took part in the Health and Retirement 2016 Venous Blood Study\n \n 58\n \n . The sample is predominantly female (54.3%) with a median age of 66 years, and ages ranging from 50 to 100 years. The sample exhibits racial diversity with 10.0% being non-Hispanic Black, 8.9% Hispanic and 81.1% non-Hispanic White and others. The sample is also socioeconomically diverse as indicated by the educational distribution: less than high school (14.0%), high school/GED (29.9%), some college (25.8%), and college+ (30.3%). More than a third of the sample is obese (44.5%), 11.0% are current smokers, and 44.2% are former smokers. The sample has been weighted to ensure it is representative of the broader U.S. population\n \n 58\n \n .\n
\n\n \n DNA methylation data\n \n
\n\n Detailed information on the 2016 Venous Blood Study is provided in the VBS 2016 Data Description\n \n 58\n \n .\u00a0Blood samples were obtained from willing respondents during in-home phlebotomy visits, ideally scheduled within four weeks of the 2016 HRS core interview. Although fasting was suggested, it was not required.\u00a0Methylation was assessed using the Infinium Methylation EPIC BeadChip. To ensure a balanced representation of key demographic variables (such as age, cohort, sex, education, and race/ethnicity), samples were randomized across plates, including 40 pairs of blinded duplicates. The correlation for all CpG sites was found to be greater than 0.97 when duplicate samples were analyzed. Data preprocessing and quality control were performed using the minfi package in R. A total of 3.4% of the methylation probes (equivalent to 29,431 out of 866,091) were excluded from the final dataset due to subpar performance, as determined by a detection p-value threshold of 0.01. Following the removal of these probes, samples that failed the detection p-value analysis were identified and removed using a 5% cut-off (minfi), resulting in the exclusion of 58 samples. Any samples that mismatched in sex and any controls (including cell lines and blinded duplicates) were also removed. High-quality methylation data were retained for 97.9% of the samples (n = 4,018). Any missing beta methylation values were replaced with the mean beta methylation value of the respective probe across all samples before the construction of DNAm age measures.\n
\n\n \n Epigenetic clocks\n \n
\n\n Thirteen epigenetic clocks have been constructed using the HRS DNAm data. These clocks are calculated as a weighted sum of aging-related CpGs, typically ranging from 100 to 500, with weights determined using a penalized regression model. These methylation clocks, which represent epigenetic age, are measured in epigenetic years, with the premise that each tick of the clock signifies aging. Among these thirteen clocks, nine are classified as first-generation clocks, calibrated based on age\n \n 6\u201310,39,59\u201361\n \n , while the remaining four are second-generation clocks, calibrated on health-related outcomes, namely Zhang\n \n 12\n \n , PhenoAge\n \n 3\n \n , GrimAge\n \n 11\n \n , and MPOA\n \n 62\n \n . These clocks exhibit significant variability in their mean values, ranges, and minimum and maximum ages. Some of the clocks, when expressed in years, have extremely high maximum ages (for example, Lin at 133 and Weidner at 148), while others have very low minimum ages (for example, Lin at 1.9). To create a composite value representing epigenetic age without any a priori selection of the clocks, we standardized them to approximate a normal distribution and took the average of these standardized clocks as our primary measure of epigenetic age. We also report results corresponding to each individual clock.\n
\n\n \n Genetic data\n \n
\n\n The genotyping for this study was carried out by the Center for Inherited Disease Research in the years 2011, 2012, and 2015. Detailed information regarding quality control can be accessed in the online Quality Control Report\n \n 63\n \n . Genotype data was collected from over 15,000 HRS participants using the Illumina HumanOmni2.5 BeadChips (HumanOmni2.5-4v1, HumanOmni2.5-8v1), which measures approximately 2.4 million SNPs. The Genetics Coordinating Center at the University of Washington, Seattle, WA, performed the genotyping quality control. Criteria for removal included individuals with missing call rates exceeding 2%, SNPs with call rates less than 98%, Hardy-Weinberg Equilibrium p-value less than 0.0001, chromosomal anomalies, and first-degree relatives in the HRS. Imputation to the 1000 Genomes Project Phase I v3 (released March 2012) was conducted using SHAPEIT2 and IMPUTE2. A worldwide reference panel consisting of all 1,092 samples from the Phase I integrated variant set was utilized. The Genetics Coordinating Center at the University of Washington, Seattle, WA, performed and documented these imputation analyses. All positions and names are aligned to the GRCh37/hg19 build.\n
\n\n \n Genetic instruments\n \n
\n\n We utilized genetic instruments derived from external genome-wide association studies (GWASes) to represent the exposure variables: brain health events for the first stage and epigenetic age for the second stage.\n
\n\n \n 1\n \n st\n \n stage\n \n
\n\n Our selection of genetic instruments involved the following sources for stroke, dementia and depression, respectively: the GIGASTROKE consortium's GWAS of all-cause stroke\n \n 64\n \n , the European Alzheimer & Dementia Biobank consortium\u2019s GWAS of Alzheimer\u2019s disease\n \n 65\n \n , and a meta-analysis of the three largest GWASes of depression\n \n 66\n \n . From each of these studies, we selected single nucleotide polymorphisms (SNPs) that were biallelic, common (minor allele frequency greater than 5%) and associated with the respective trait (p < 1e-5). To ensure the independence of these SNPs, we filtered out variants with an r2 (a measure of correlation between two genetic variants) greater than 0.1. This resulted in 382 SNPs for stroke, 256 for Alzheimer\u2019s disease, and 462 for depression. These SNPs were combined to yield 1100 instruments associated with either stroke, Alzheimer\u2019s disease, or depression. From this pool, 20 variants were excluded to ensure independence, 75 were not present in the imputed HRS genetic data, and 20 palindromic SNPs were excluded, resulting in a final list of 985 instruments. We then estimated the effect of the genetic instruments on the epigenetic age and on the brain health composite by conducting single-SNP association tests in HRS (Figure 2). The effect estimates corresponding to epigenetic age were obtained in HRS participants with DNAm and genetic data and the ones corresponding to brain health events were obtained in all HRS participants with genetic data (Figure 1).\n
\n\n \n 2\n \n nd\n \n stage\n \n
\n\n For the second stage, we selected genetic instruments by combining data from multi-ethnic GWASes\n \n 67\n \n of six epigenetic clocks: GrimAge\n \n 11\n \n , Hannum\n \n 8\n \n , PhenoAge\n \n 3\n \n , Horvath\n \n 9\n \n , PAI-1\n \n 11\n \n , and Gran\n \n 3,11,40\n \n . From each of these GWASes, we selected common SNPs (minor allele frequency >5%) associated with the respective epigenetic clock (p < 1e-5). To ensure the independence of these SNPs, we filtered out variants with an r2 greater than 0.1. This yielded 81 SNPs for the GrimAge clock, 84 for the Hannum clock, 104 for the PhenoAge clock, 103 for the Horvath clock, 75 for the PAI-1 clock, and 403 for the Gran clock. These SNPs were combined to obtain a pooled list of 850 SNPs associated with any of the six epigenetic clocks. From this pool, 52 variants were excluded to ensure independence, 6 were not present in the imputed HRS genetic data, and 15 palindromic SNPs were excluded, resulting in a final list of 777 instruments. We then estimated the effect of the genetic instruments on the epigenetic age and on the brain health composite by conducting single-SNP association tests in HRS (Figure 3).\n
\n\n \n Ascertainment of brain health events\n \n
\n\n \n Stroke\n \n
\n\n Stroke events were identified as the first instance of stroke in a dedicated variable evaluated throughout the study period (1992\u20132020), based on self-reported or proxy-reported doctor\u2019s diagnosis (Has a doctor ever told you that you had a stroke?). In cases where participants were unable to be directly interviewed (e.g., deceased), health care proxies were interviewed. Transient ischemic attacks were not systematically assessed and were not classified as strokes, and information on stroke subtype was not available. Previous studies using HRS data have demonstrated that associations between known risk factors and self-reported stroke incidence in the HRS align well with associations in studies using observationally verified strokes\n \n 68\n \n . Moreover, self-reported strokes in the HRS corresponded well with strokes coded according to the International Classification of Diseases in the Centers for Medicare and Medicaid Services records, with a sensitivity of 74% and a specificity of 93%\n \n 69\n \n .\n
\n\n \n Dementia\n \n
\n\n The ascertainment of all-cause dementia among self-respondents was carried out at each wave using the modified version of the Telephone Interview for Cognitive Status (TICS): a 27-point cognitive scale that encompasses immediate and delayed 10-noun free recall tests (each with a range of 0\u201310 points), a serial seven subtraction test (range: 0\u20135 points), and a backward count from 20 test (range: 0\u20132 points)\n \n 70,71\n \n . Based on their continuous score, we categorized cognitive status into two groups\u2014those with and without dementia\u2014using observationally verified cutpoints from the Aging, Demographics, and Memory Study (ADAMS). A supplemental study of the HRS, ADAMS involves in-home neuropsychological and observational assessments combined with expert clinician adjudication to obtain a gold-standard diagnosis of cognitive status\n \n 70,72\n \n . Respondents with scores ranging from 12 to 27 were classified as non-impaired; those with scores from 7 to 11 were identified as having cognitive impairment but no dementia; and those with scores from 0 to 6 were classified as having dementia. For the purposes of this paper, we focused solely on participants with dementia. A small percentage of respondents (0.8%\u20133.1%) declined to participate in tests of immediate and delayed recall and serial 7s. To address this, HRS has developed an imputation strategy for cognitive variables across all waves\n \n 73\n \n .\n
\n\n \n Late-life depression\n \n
\n\n Following a common definition from the literature\n \n 74\u201377\n \n , we defined late-life depression as a major depressive episode occurring after the age of 65 in an individual with no history of depressive episodes prior to this age. Depressive symptoms were evaluated using the validated, modified 8-item version of the Center for Epidemiologic Studies-Depression (CES-D) scale\n \n 78,79\n \n . During each biennial questionnaire, participants were asked to indicate (yes/no) whether they had experienced any of the 8 symptoms in the preceding week. A summary score (ranging from 0 to 8) was compiled by adding the number of affirmative responses across the 8 items, with two positively framed items being reverse-coded\n \n 78\n \n . Major depressive episodes were identified using dichotomized CES-D summary scores for each wave, with a cutoff of \u22654 symptoms. This threshold has been previously validated and is considered equivalent to the 16-symptom cut-off of the well-validated 20-item CES-D scale\n \n 76,78,80\n \n . In our sensitivity analyses, we explored an alternative definition of late-life depression found in the literature, characterized by a lower age cutoff of 60 years, instead of 65\n \n 81\u201383\n \n .\n
\n\n \n Covariates ascertainment\n \n
\n\n We collected self-reported demographic and socioeconomic variables at the onset of the Venous Blood Study\n \n 58\n \n , including age (continuous), sex (male or female), and race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic or other). Additionally, we gathered self-reported measures of health behaviors and health conditions at baseline, such as body mass index (continuous, kg/m2 derived from self-reported height and weight), and cigarette smoking status (nonsmoker, former smoker, current smoker). Health conditions were determined based on responses (yes/no) to the question \u201cHas a doctor ever told you that you had a (health condition)?\u201d for heart disease, diabetes, and hypertension. Previous studies using HRS data have shown that self-reported health conditions align substantially with medical records data, and that the self-reported health behavioral measures have strong external validity\n \n 84\u201388\n \n .\n
\n\n \n Statistical analyses\n \n
\n\n We describe discrete data as counts (percentages) and continuous data as mean (standard deviation) or median (interquartile range), as appropriate. In the first stage of the study, which examined the association between a history of brain health events (exposure) and epigenetic age (outcome), a history of brain health events was defined as having experienced a stroke, dementia, or late-life depression episode ascertained in waves 1 (1992) to 13 (2016). In the second stage of the study, which examined the association between epigenetic age (exposure) and the onset of new brain health events (outcomes), these events were defined as a stroke, dementia, or late-life depressive episode ascertained in waves 14 (2018) or 15 (2020). Participants who did not participate in both of these waves, due to loss to follow-up or death, were excluded from this analysis. Additionally, participants who had experienced brain health events between waves 1 and 13 were also excluded from this phase of the analysis.\n
\n\n In the first stage of our study, we explored the association between a history of brain health events and epigenetic age using multivariable linear regression models. These models were either unadjusted (Model 1), adjusted for potential demographic confounders such as age, sex, and race/ethnicity (Model 2), or adjusted for these demographic factors and cardiovascular risk factors (hypertension, diabetes, smoking, and body mass index), and comorbidities (history of heart events including heart attack, coronary artery disease, angina, and congestive heart failure, Model 3). In the second stage, we investigated the association between epigenetic age and the risk of new brain health events using multivariable logistic regression models. These models were either unadjusted (Model 1) or adjusted for the same sets of confounders as in the first stage (Model 2 and 3).\n
\n\n \n Mendelian Randomization\n \n
\n\n In both stages, our primary MR analyses used the inverse variance weighted (IVW) method. In secondary analyses, we tested for horizontal pleiotropy (the possibility that the effect of the instrument on the outcome of interest is exerted through a pathway other than the exposure) using the Mendelian Randomization Pleiotropy Residual Sum and Outlier (MR-PRESSO\n \n 89\n \n ) global test with 10,000 simulations and the MR-Egger intercept term\n \n 90\n \n . To account for this possible phenomenon, we implemented the weighted median method, a robust alternative to the IVW method that allows for up to 50% of the genetic variants used to be invalid instrumental variables without biasing the causal effect estimate\n \n 91\n \n . Additionally, the weighted median approach is less sensitive to outliers than the IVW method, which can be useful in the presence of genetic variants with extreme effect estimates\n \n 26\n \n .\n
\n\n \n Secondary and sensitivity analyses\n \n
\n\n In our secondary analyses, we repeated the epidemiological analyses for both stages, considering each brain health outcome individually (stroke, dementia, and depression), as well as a composite outcome that included only stroke and dementia. In addition to our main measure, the mean epigenetic age, we also report the association results for each epigenetic clock. In our sensitivity analyses, we: (1) tested the association between epigenetic age and the risk of new brain health events, excluding only participants missing data for waves 14 or 15, as opposed to excluding participants missing both waves; (2) repeated both stages using an age cutoff of 60 to ascertain late-life depression.\n
\n\n \n Software\n \n
\n\n Statistical analyses were performed using R 4.2.1\n \n 92\n \n and the following packages: dplyr, ggplot2, ggforestplot, tableone, TwoSampleMR, MR-PRESSO, gwasvcf, ieugswar.The current manuscript is written in line with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines (Supplementary Table X).\n
\n\n \n Table 1. Cohort characteristics\n \n
\n| \n \n Variable\n \n \n | \n \n \n Overall\n \n \n \n (n=4018)\n \n \n | \n \n \n Prevalent\n \n brain health events\n \n \n \n (n=806)\n \n \n | \n \n \n Incident\n \n \n \n Brain health events\n \n \n \n (n=261)\n \n \n | \n
| \n \n Demographics\n \n \n | \n \n \n | \n \n \n | \n \n \n | \n
| \n \n Age (mean (SD))\n \n \n | \n \n 69.9 (9.6)\n \n | \n \n 75.2 (10.1)\n \n | \n \n 73.0 (9.3)\n \n | \n
| \n \n Male gender\n \n \n | \n \n 1669 (41.5)\n \n | \n \n 334 (41.4)\n \n | \n \n 115 (44.1)\n \n | \n
| \n \n Race\n \n \n | \n \n \n | \n \n \n | \n \n \n | \n
| \n \n White\n \n \n | \n \n 3013 (75.0)\n \n | \n \n 572 (71.0)\n \n | \n \n 204 (78.2)\n \n | \n
| \n \n Black\n \n \n | \n \n 674 (16.8)\n \n | \n \n 170 (21.1)\n \n | \n \n 40 (15.3)\n \n | \n
| \n \n Hispanic\n \n \n | \n \n 207 (5.2)\n \n | \n \n 42 (5.2)\n \n | \n \n 12 (4.6)\n \n | \n
| \n \n Other\n \n \n | \n \n 122 (3.0)\n \n | \n \n 22 (2.7)\n \n | \n \n 5 (1.9)\n \n | \n
| \n \n Cardiovascular Risk factors\n \n \n | \n \n \n | \n \n \n | \n \n \n | \n
| \n \n Prevalent hypertension\n \n \n | \n \n 2559 (63.7)\n \n | \n \n 604 (74.9)\n \n | \n \n 186 (71.3)\n \n | \n
| \n \n Prevalent diabetes\n \n \n | \n \n 1151 (28.6)\n \n | \n \n 306 (38.0)\n \n | \n \n 85 (32.6)\n \n | \n
| \n \n BMI (mean (SD))\n \n \n | \n \n 28.92 (6.30)\n \n | \n \n 28.4 (6.4)\n \n | \n \n 28.9 (6.1)\n \n | \n
| \n \n Smoking\n \n \n | \n \n \n | \n \n \n | \n \n \n | \n
| \n \n Past\n \n \n | \n \n 1776 (44.2)\n \n | \n \n 399 (49.5)\n \n | \n \n 118 (45.2)\n \n | \n
| \n \n Never\n \n \n | \n \n 1764 (43.9)\n \n | \n \n 307 (38.1)\n \n | \n \n 113 (43.3)\n \n | \n
| \n \n Current\n \n \n | \n \n 455 (11.3)\n \n | \n \n 95 (11.8)\n \n | \n \n 29 (11.1)\n \n | \n
| \n \n Prevalent heart condition*\n \n \n | \n \n 1098 (27.3)\n \n | \n \n 356 (44.2)\n \n | \n \n 80 (30.7)\n \n | \n
\n *Heart conditions include: heart attack, coronary artery disease, angina, congestive heart failure\n
\n\n Note: The terms prevalent, respectively incident, refer to conditions having occurred before, respectively after, the epigenetic age estimation performed during the 2016 wave.\n
\n\n \n Table 2. Multivariable regression results: changes in mean epigenetic age following a brain health event and odds ratios of brain health events per one standard deviation increase in mean epigenetic age\n \n
\n| \n \n \n \n | \n \n \n 1\n \n st\n \n stage\n \n \n | \n \n \n 2\n \n nd\n \n stage\n \n \n | \n |||
| \n \n Outcome\n \n \n | \n \n \n Change in mean epigenetic age\n \n \n as a function of\n \n \n prevalent\n \n \n brain health\n \n \n events\n \n \n | \n \n \n Change in mean odds\n \n \n of incident brain health events\n \n \n as a function\n \n \n per\n \n \n 1\n \n \n standard deviation increase in mean epigenetic age\n \n \n | \n |||
| \n \n Statistical model\n \n \n | \n \n \n Linear regression\n \n \n | \n \n \n Logistic regression\n \n \n | \n |||
| \n \n Covariates\n \n \n | \n \n \n % change\n \n \n | \n \n \n \n \n \n Beta (SE)\n \n \n \n | \n \n \n P\n \n \n | \n \n \n Odds Ratios (95% CI)\n \n \n | \n \n \n P\n \n \n | \n
| \n \n Unadjusted\n \n \n \n Model 1\n \n \n | \n \n 42%\n \n | \n \n 0.42 (0.02)\n \n | \n \n <0.001\n \n | \n \n 2.62 (2.10-3.25)\n \n | \n \n <0.001\n \n | \n
| \n \n Multivariable\n \n \n \n Model 2\n \n \n | \n \n 8%\n \n | \n \n 0.08 (0.01)\n \n | \n \n <0.001\n \n | \n \n 1.70 (1.16-2.50)\n \n | \n \n 0.007\n \n | \n
| \n \n Multivariable\n \n \n \n Model 3\n \n \n | \n \n 4%\n \n | \n \n 0.04 (0.01)\n \n | \n \n 0.002\n \n | \n \n 1.48 (0.99-2.21)\n \n | \n \n 0.057\n \n | \n
\n Model 2: Adjusted for age, sex and race/ethnicity\n
\n\n Model 3: Adjusted for age, sex, race/ethnicity, hypertension, diabetes, smoking, BMI, history of heart attack, coronary artery disease, angina, or congestive heart failure\n
\n\n \n Table 3. Mendelian Randomization analyses.\n \n
\n| \n \n Analytical approach for Mendelian Randomization analyses\n \n \n | \n \n \n 1\n \n st\n \n stage\n \n \n \n \n \n \n Genetically modeled exposure =\n \n \n \n Risk of brain\n \n \n health events\n \n \n \n \n \n \n Outcome = Epigenetic age\n \n \n \n \n \n | \n \n \n 2\n \n nd\n \n stage\n \n \n \n \n \n \n Genetically modeled\n \n \n exposure =\n \n \n \n Epigenetic\n \n \n age\n \n \n \n \n \n \n Outcome = Risk of brain health events\n \n \n \n \n \n | \n ||||
| \n \n \n \n | \n \n \n Number of instruments\n \n \n | \n \n \n Beta (SE)\n \n \n | \n \n \n P\n \n \n | \n \n \n Number of instruments\n \n \n | \n \n \n OR (95% CI)\n \n \n | \n \n \n P\n \n \n | \n
| \n \n Primary IVW MR\n \n \n | \n \n 985\n \n | \n \n 0.11 (0.03)\n \n | \n \n <0.001\n \n | \n \n 777\n \n | \n \n 1.15 (1.06 \u2013 1.25)\n \n | \n \n <0.001\n \n | \n
| \n \n Weighted median MR\n \n \n | \n \n 985\n \n | \n \n 0.08 (0.04)\n \n | \n \n 0.052\n \n | \n \n 777\n \n | \n \n 1.15 (1.00 \u2013 1.31)\n \n | \n \n 0.048\n \n | \n
| \n \n MR-Egger\n \n \n | \n \n 985\n \n | \n \n 0.10 (0.04)\n \n | \n \n 0.01\n \n | \n \n 777\n \n | \n \n 1.15 (1.00 \u2013 1.31)\n \n | \n \n 0.047\n \n | \n
\n Abbreviations: IVW = Inverse probability weighted; MR = Mendelian Randomization; SE = Standard error; CI = confidence interval; OR = odds ratio.\n
\n![\n Scheme1.png\n](\"https://assets-eu.researchsquare.com/files/rs-3810968/v1/30d250ec6af1c639a3f082ce.png\"){kind=link}
![\n Onlinefloatimage6.png\n](\"https://assets-eu.researchsquare.com/files/rs-1092067/v1/77e8d621f38fb5f737f48338.png\"){kind=link}
![\n Scheme1.png\n](\"https://assets-eu.researchsquare.com/files/rs-4743287/v1/c8871d0867483188da3d3dbc.png\"){kind=link}
![\n Scheme2.png\n](\"https://assets-eu.researchsquare.com/files/rs-4743287/v1/fbe297cd7169b6e1e11824cb.png\"){kind=link}
![\n Scheme3.png\n](\"https://assets-eu.researchsquare.com/files/rs-4743287/v1/eb09224f65f8e5f090e9abc5.png\"){kind=link}
![\n Scheme4.png\n](\"https://assets-eu.researchsquare.com/files/rs-4743287/v1/7d59c7b4a6db5d9fffd558cf.png\"){kind=link}
![\n Scheme5.png\n](\"https://assets-eu.researchsquare.com/files/rs-4743287/v1/6f300cd3c721f360174e0132.png\"){kind=link}
![\n AdditionalFigure.png\n](\"https://assets-eu.researchsquare.com/files/rs-4743287/v1/dce786ff1343753dc7251beb.png\"){kind=link}