Scalable Deep Color Quantization: a Cluster Imitation Approach.

Color quantization reduces the number of colors used in an image while preserving its content, which is essential in pixel art and knitting art creation. Traditional methods primarily focus on visual fidelity and treat it as a clustering problem in the RGB space. While effective in large (5-6 bits) color spaces, these approaches cannot guarantee semantics in small (1-2 bits) color spaces. On the other hand, deep color quantization methods use network viewers such as AlexNet and ResNet for supervision, effectively preserving semantics in small color spaces. However, in large color spaces, they lag behind traditional methods in terms of visual fidelity. In this work, we propose ColorCNN+, a novel approach that combines the strengths of both. It uses network viewer signals for supervision in small color spaces and learns to cluster the colors in large color spaces. Noteworthily, it is non-trivial for neural networks to do clustering, where existing deep clustering methods often need K-means to cluster the features. In this work, through a newly introduced cluster imitation loss, ColorCNN+ learns to directly output the cluster assignment without any additional steps. Furthermore, ColorCNN+ supports multiple color space sizes and network viewers, offering scalability and easy deployment. Experimental results demonstrate competitive performance of ColorCNN+ across various settings. Code is available at link.

Endocytosis blocks the vesicular secretion of exosome marker proteins.

Exosomes are secreted vesicles of ~30 to 150 nm diameter that play important roles in human health and disease. To better understand how cells release these vesicles, we examined the biogenesis of the most highly enriched human exosome marker proteins, the exosomal tetraspanins CD81, CD9, and CD63. We show here that endocytosis inhibits their vesicular secretion and, in the case of CD9 and CD81, triggers their destruction. Furthermore, we show that syntenin, a previously described exosome biogenesis factor, drives the vesicular secretion of CD63 by blocking CD63 endocytosis and that other endocytosis inhibitors also induce the plasma membrane accumulation and vesicular secretion of CD63. Finally, we show that CD63 is an expression-dependent inhibitor of endocytosis that triggers the vesicular secretion of lysosomal proteins and the clathrin adaptor AP-2 mu2. These results suggest that the vesicular secretion of exosome marker proteins in exosome-sized vesicles occurs primarily by an endocytosis-independent pathway.

Antigen-display exosomes provide adjuvant-free protection against SARS-CoV-2 disease at nanogram levels of spike protein.

As the only bionormal nanovesicle, exosomes have high potential as a nanovesicle for delivering vaccines and therapeutics. We show here that the loading of type-1 membrane proteins into the exosome membrane is induced by exosome membrane anchor domains, EMADs, that maximize protein delivery to the plasma membrane, minimize protein sorting to other compartments, and direct proteins into exosome membranes. Using SARS-CoV-2 spike as an example and EMAD13 as our most effective exosome membrane anchor, we show that cells expressing a spike-EMAD13 fusion protein produced exosomes that carry dense arrays of spike trimers on 50% of all exosomes. Moreover, we find that immunization with spike-EMAD13 exosomes induced strong neutralizing antibody responses and protected hamsters against SARS-CoV-2 disease at doses of just 0.5-5 ng of spike protein, without adjuvant, demonstrating that antigen-display exosomes are particularly immunogenic, with important implications for both structural and expression-dependent vaccines.

Fixation strength in arthroscopic labral repair of the hip: A head-to-head comparison of the biomechanical performance of a biocompatible vs. all-suture anchor in the setting of acetabuloplasty.

Much is known about the biomechanical performance of various types of suture anchors commonly used for labral fixation in the shoulder; however, similar studies in the hip are less common. We sought to compare all-suture and polyether ether ketone small-diameter anchors in the setting of labral repair during hip arthroscopy, with and without acetabuloplasty. We hypothesized that the biomechanical properties of the all-suture group when compared to polyether ether ketone anchors would be similar amongst native acetabula and significantly less following acetabuloplasty and that pullout forces would be reduced in the anterior and inferior regions of the acetabulum compared to the superior region. Bone density was measured in nine matched pairs of fresh-frozen cadaveric acetabula in the superior, anterosuperior, and anterior regions. Acetabuloplasty was performed in all three regions, while the contralateral acetabulum was left in situ as a control. Suture anchors were placed such that one each of two different types was placed within each region. Specimens were tested in cyclic fatigue and loaded to failure. The all-suture group had significantly higher cyclic displacement compared to the polyether ether ketone, but there was no significant difference in ultimate load, regardless of acetabuloplasty. Amongst all non-resected specimens, the lowest bone density was observed consistently in the inferior region. Our results indicate that, with or without acetabuloplasty, a small-diameter polyether ether ketone anchor appears to be more stable than an all-suture anchor, which needs to be set first.

Application of machine learning models to identify serological predictors of COVID-19 severity and outcomes.

Critically ill people with COVID-19 have greater antibody titers than those with mild to moderate illness, but their association with recovery or death from COVID-19 has not been characterized. In 178 COVID-19 patients, 73 non-hospitalized and 105 hospitalized patients, mucosal swabs and plasma samples were collected at hospital enrollment and up to 3 months post-enrollment (MPE) to measure virus RNA, cytokines/chemokines, binding antibodies, ACE2 binding inhibition, and Fc effector antibody responses against SARS-CoV-2. The association of demographic variables and >20 serological antibody measures with intubation or death due to COVID-19 was determined using machine learning algorithms. Predictive models revealed that IgG binding and ACE2 binding inhibition responses at 1 MPE were positively and C1q complement activity at enrollment was negatively associated with an increased probability of intubation or death from COVID-19 within 3 MPE. Serological antibody measures were more predictive than demographic variables of intubation or death among COVID-19 patients.

Bidirectionally self-normalizing neural networks.

The problem of vanishing and exploding gradients has been a long-standing obstacle that hinders the effective training of neural networks. Despite various tricks and techniques that have been employed to alleviate the problem in practice, there still lacks satisfactory theories or provable solutions. In this paper, we address the problem from the perspective of high-dimensional probability theory. We provide a rigorous result that shows, under mild conditions, how the vanishing/exploding gradients problem disappears with high probability if the neural networks have sufficient width. Our main idea is to constrain both forward and backward signal propagation in a nonlinear neural network through a new class of activation functions, namely Gaussian-Poincaré normalized functions, and orthogonal weight matrices. Experiments on both synthetic and real-world data validate our theory and confirm its effectiveness on very deep neural networks when applied in practice.

Combined PD-L1/TGFß blockade allows expansion and differentiation of stem cell-like CD8 T cells in immune excluded tumors.

TGFß signaling is associated with non-response to immune checkpoint blockade in patients with advanced cancers, particularly in the immune-excluded phenotype. While previous work demonstrates that converting tumors from excluded to inflamed phenotypes requires attenuation of PD-L1 and TGFß signaling, the underlying cellular mechanisms remain unclear. Here, we show that TGFß and PD-L1 restrain intratumoral stem cell-like CD8 T cell (TSCL) expansion and replacement of progenitor-exhausted and dysfunctional CD8 T cells with non-exhausted T effector cells in the EMT6 tumor model in female mice. Upon combined TGFß/PD-L1 blockade IFNγhi CD8 T effector cells show enhanced motility and accumulate in the tumor. Ensuing IFNγ signaling transforms myeloid, stromal, and tumor niches to yield an immune-supportive ecosystem. Blocking IFNγ abolishes the anti-PD-L1/anti-TGFß therapy efficacy. Our data suggest that TGFß works with PD-L1 to prevent TSCL expansion and replacement of exhausted CD8 T cells, thereby maintaining the T cell compartment in a dysfunctional state.

Exomap1 mouse: a transgenic model for in vivo studies of exosome biology.

Exosomes are small extracellular vesicles (sEVs) of ~30-150 nm in diameter that have the same topology as the cell, are enriched in selected exosome cargo proteins, and play important roles in health and disease. To address large unanswered questions regarding exosome biology in vivo, we created the exomap1 transgenic mouse model. In response to Cre recombinase, exomap1 mice express HsCD81mNG, a fusion protein between human CD81, the most highly enriched exosome protein yet described, and the bright green fluorescent protein mNeonGreen. As expected, cell type-specific expression of Cre induced the cell type-specific expression of HsCD81mNG in diverse cell types, correctly localized HsCD81mNG to the plasma membrane, and selectively loaded HsCD81mNG into secreted vesicles that have the size (~80 nm), topology (outside out), and content (presence of mouse exosome markers) of exosomes. Furthermore, mouse cells expressing HsCD81mNG released HsCD81mNG-marked exosomes into blood and other biofluids. Using high-resolution, single-exosome analysis by quantitative single molecule localization microscopy, we show here that that hepatocytes contribute ~15% of the blood exosome population whereas neurons contribute <1% of blood exosomes. These estimates of cell type-specific contributions to blood EV population are consistent with the porosity of liver sinusoidal endothelial cells to particles of ~50-300 nm in diameter, as well as with the impermeability of blood-brain and blood-neuron barriers to particles >5 nm in size. Taken together, these results establish the exomap1 mouse as a useful tool for in vivo studies of exosome biology, and for mapping cell type-specific contributions to biofluid exosome populations. In addition, our data confirm that CD81 is a highly-specific marker for exosomes and is not enriched in the larger microvesicle class of EVs.

Syntenin and CD63 Promote Exosome Biogenesis from the Plasma Membrane by Blocking Cargo Endocytosis.

Exosomes are small extracellular vesicles important in health and disease. Syntenin is thought to drive the biogenesis of CD63 exosomes by recruiting Alix and the ESCRT machinery to endosomes, initiating an endosome-mediated pathway of exosome biogenesis. Contrary to this model, we show here that syntenin drives the biogenesis of CD63 exosomes by blocking CD63 endocytosis, thereby allowing CD63 to accumulate at the plasma membrane, the primary site of exosome biogenesis. Consistent with these results, we find that inhibitors of endocytosis induce the exosomal secretion of CD63, that endocytosis inhibits the vesicular secretion of exosome cargo proteins, and that high-level expression of CD63 itself also inhibits endocytosis. These and other results indicate that exosomes bud primarily from the plasma membrane, that endocytosis inhibits their loading into exosomes, that syntenin and CD63 are expression-dependent regulators of exosome biogenesis, and that syntenin drives the biogenesis of CD63 exosomes even in Alix knockout cells.

The D614G mutation redirects SARS-CoV-2 spike to lysosomes and suppresses deleterious traits of the furin cleavage site insertion mutation.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) egress occurs by lysosomal exocytosis. We show that the Spike D614G mutation enhances Spike trafficking to lysosomes, drives Spike-mediated reprogramming of lysosomes, and reduces cell surface Spike expression by ~3-fold. D614G is not a human-specific adaptation. Rather, it is an adaptation to the earlier furin cleavage site insertion (FCSI) mutation that occurred at the genesis of SARS-CoV-2. While advantageous to the virus, furin cleavage of spike has deleterious effects on spike structure and function, inhibiting its trafficking to lysosomes and impairing its infectivity by the transmembrane serine protease 2(TMPRSS2)-independent, endolysosomal pathway. D614G restores spike trafficking to lysosomes and enhances the earliest events in SARS-CoV-2 infectivity, while spike mutations that restore SARS-CoV-2's TMPRSS2-independent infectivity restore spike's trafficking to lysosomes. Together, these and other results show that D614G is an intragenic suppressor of deleterious traits linked to the FCSI and lend additional support to the endolysosomal model of SARS-CoV-2 egress and entry.

Phase 2 of extracellular RNA communication consortium charts next-generation approaches for extracellular RNA research.

The extracellular RNA communication consortium (ERCC) is an NIH-funded program aiming to promote the development of new technologies, resources, and knowledge about exRNAs and their carriers. After Phase 1 (2013-2018), Phase 2 of the program (ERCC2, 2019-2023) aims to fill critical gaps in knowledge and technology to enable rigorous and reproducible methods for separation and characterization of both bulk populations of exRNA carriers and single EVs. ERCC2 investigators are also developing new bioinformatic pipelines to promote data integration through the exRNA atlas database. ERCC2 has established several Working Groups (Resource Sharing, Reagent Development, Data Analysis and Coordination, Technology Development, nomenclature, and Scientific Outreach) to promote collaboration between ERCC2 members and the broader scientific community. We expect that ERCC2's current and future achievements will significantly improve our understanding of exRNA biology and the development of accurate and efficient exRNA-based diagnostic, prognostic, and theranostic biomarker assays.

A shared, stochastic pathway mediates exosome protein budding along plasma and endosome membranes.

Exosomes are small extracellular vesicles of ∼30 to 150 nm that are secreted by all cells, abundant in all biofluids, and play important roles in health and disease. However, details about the mechanism of exosome biogenesis are unclear. Here, we carried out a cargo-based analysis of exosome cargo protein biogenesis in which we identified the most highly enriched exosomal cargo proteins and then followed their biogenesis, trafficking, and exosomal secretion to test different hypotheses for how cells make exosomes. We show that exosome cargo proteins bud from cells (i) in exosome-sized vesicles regardless of whether they are localized to plasma or endosome membranes, (ii) ∼5-fold more efficiently when localized to the plasma membrane, (iii) ∼5-fold less efficiently when targeted to the endosome membrane, (iv) by a stochastic process that leads to ∼100-fold differences in their abundance from one exosome to another, and (v) independently of small GTPase Rab27a, the ESCRT complex-associated protein Alix, or the cargo protein CD63. Taken together, our results demonstrate that cells use a shared, stochastic mechanism to bud exosome cargoes along the spectrum of plasma and endosome membranes and far more efficiently from the plasma membrane than the endosome. Our observations also indicate that the pronounced variation in content between different exosome-sized vesicles is an inevitable consequence of a stochastic mechanism of small vesicle biogenesis, that the origin membrane of exosome-sized extracellular vesicles simply cannot be determined, and that most of what we currently know about exosomes has likely come from studies of plasma membrane-derived vesicles.

Fine-Grained Classification via Categorical Memory Networks.

Motivated by the desire to exploit patterns shared across classes, we present a simple yet effective class-specific memory module for fine-grained feature learning. The memory module stores the prototypical feature representation for each category as a moving average. We hypothesize that the combination of similarities with respect to each category is itself a useful discriminative cue. To detect these similarities, we use attention as a querying mechanism. The attention scores with respect to each class prototype are used as weights to combine prototypes via weighted sum, producing a uniquely tailored response feature representation for a given input. The original and response features are combined to produce an augmented feature for classification. We integrate our class-specific memory module into a standard convolutional neural network, yielding a Categorical Memory Network. Our memory module significantly improves accuracy over baseline CNNs, achieving competitive accuracy with state-of-the-art methods on four benchmarks, including CUB-200-2011, Stanford Cars, FGVC Aircraft, and NABirds.

Degron tagging of BleoR and other antibiotic-resistance genes selects for higher expression of linked transgenes and improved exosome engineering.

Five antibiotic resistance (AR) genes have been used to select for transgenic eukaryotic cell lines, with the BleoR, PuroR, HygR, NeoR, and BsdR cassettes conferring resistance to zeocin, puromycin, hygromycin, geneticin/G418, and blasticidin, respectively. We recently demonstrated that each AR gene establishes a distinct threshold of transgene expression below which no cell can survive, with BleoR selecting for the highest level of transgene expression, nearly ∼10-fold higher than in cells selected using the NeoR or BsdR markers. Here, we tested the hypothesis that there may be an inverse proportionality between AR protein function and the expression of linked, transgene-encoded, recombinant proteins. Specifically, we fused each AR protein to proteasome-targeting degron tags, used these to select for antibiotic-resistant cell lines, and then measured the expression of the linked, recombinant protein, mCherry, as a proxy marker of transgene expression. In each case, degron-tagged AR proteins selected for higher mCherry expression than their cognate WT AR proteins. ER50BleoR selected for the highest level of mCherry expression, greater than twofold higher than BleoR or any other AR gene. Interestingly, use of ER50BleoR as the selectable marker translated to an even higher, 3.5-fold increase in the exosomal loading of the exosomal cargo protein, CD63/Y235A. Although a putative CD63-binding peptide, CP05, has been used to decorate exosome membranes in a technology known as "exosome painting," we show here that CP05 binds equally well to CD63-/- cells, WT 293F cells, and CD63-overexpressing cells, indicating that CP05 may bind membranes nonspecifically. These results are of high significance for cell engineering and especially for exosome engineering.

Continued Virus-Specific Antibody-Secreting Cell Production, Avidity Maturation and B Cell Evolution in Patients Hospitalized with COVID-19.

Understanding the development and sustainability of the virus-specific protective immune response to infection with severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) remains incomplete with respect to the appearance and disappearance of virus-specific antibody-secreting cells (ASCs) in circulation. Therefore, we performed cross-sectional and longitudinal analyses of peripheral blood mononuclear cells and plasma collected from 55 hospitalized patients up to 4 months after onset of COVID-19 symptoms. Spike (S)- and nucleocapsid (N)-specific IgM and IgG ASCs appeared within 2 weeks accompanied by flow cytometry increases in double negative plasmablasts consistent with a rapid extrafollicular B cell response. Total and virus-specific IgM and IgG ASCs peaked at 3-4 weeks and were still being produced at 3-4 months accompanied by increasing antibody avidity consistent with a slower germinal center B cell response. N-specific ASCs were produced for longer than S-specific ASCs and avidity maturation was greater for antibody to N than S. Patients with more severe disease produced more S-specific IgM and IgG ASCs than those with mild disease and had higher levels of N- and S-specific antibody. Women had more B cells in circulation than men and produced more S-specific IgA and IgG and N-specific IgG ASCs. Flow cytometry analysis of B cell phenotypes showed an increase in circulating B cells at 4-6 weeks with decreased percentages of switched and unswitched memory B cells. These data indicate ongoing antigen-specific stimulation, maturation, and production of ASCs for several months after onset of symptoms in patients hospitalized with COVID-19.

Deep Declarative Networks.

We explore a class of end-to-end learnable models wherein data processing nodes (or network layers) are defined in terms of desired behavior rather than an explicit forward function. Specifically, the forward function is implicitly defined as the solution to a mathematical optimization problem. Consistent with nomenclature in the programming languages community, we name these models deep declarative networks. Importantly, it can be shown that the class of deep declarative networks subsumes current deep learning models. Moreover, invoking the implicit function theorem, we show how gradients can be back-propagated through many declaratively defined data processing nodes thereby enabling end-to-end learning. We discuss how these declarative processing nodes can be implemented in the popular PyTorch deep learning software library allowing declarative and imperative nodes to co-exist within the same network. We also provide numerous insights and illustrative examples of declarative nodes and demonstrate their application for image and point cloud classification tasks.

Exosome-mediated mRNA delivery in vivo is safe and can be used to induce SARS-CoV-2 immunity.

Functional delivery of mRNA has high clinical potential. Previous studies established that mRNAs can be delivered to cells in vitro and in vivo via RNA-loaded lipid nanoparticles (LNPs). Here we describe an alternative approach using exosomes, the only biologically normal nanovesicle. In contrast to LNPs, which elicited pronounced cellular toxicity, exosomes had no adverse effects in vitro or in vivo at any dose tested. Moreover, mRNA-loaded exosomes were characterized by efficient mRNA encapsulation (∼90%), high mRNA content, consistent size, and a polydispersity index under 0.2. Using an mRNA encoding the red light-emitting luciferase Antares2, we observed that mRNA-loaded exosomes were superior to mRNA-loaded LNPs at delivering functional mRNA into human cells in vitro. Injection of Antares2 mRNA-loaded exosomes also led to strong light emission following injection into the vitreous fluid of the eye or into the tissue of skeletal muscle in mice. Furthermore, we show that repeated injection of Antares2 mRNA-loaded exosomes drove sustained luciferase expression across six injections spanning at least 10 weeks, without evidence of signal attenuation or adverse injection site responses. Consistent with these findings, we observed that exosomes loaded with mRNAs encoding immunogenic forms of the SARS-CoV-2 Spike and Nucleocapsid proteins induced long-lasting cellular and humoral responses to both. Taken together, these results demonstrate that exosomes can be used to deliver functional mRNA to and into cells in vivo.

Extracellular vesicles derived from cardiosphere-derived cells as a potential antishock therapeutic.

BACKGROUND: Extracellular vesicles (EVs) isolated from cardiosphere-derived cells (CDC-EVs) are coming to light as a unique cell-free therapeutic. Because of their novelty, however, there still exist prominent gaps in knowledge regarding their therapeutic potential. Herein the therapeutic potential of CDC-EVs in a rat model of acute traumatic coagulopathy induced by multiple injuries and hemorrhagic shock is outlined. METHODS: Extracellular vesicle surface expression of procoagulant molecules (tissue factor and phosphatidylserine) was evaluated by flow cytometry. Extracellular vesicle thrombogenicity was tested using calibrated thrombogram, and clotting parameters were assessed using a flow-based adhesion model simulating blood flow over a collagen-expressing surface. The therapeutic efficacy of EVs was then determined in a rat model of acute traumatic coagulopathy induced by multiple injuries and hemorrhagic shock. RESULTS: Extracellular vesicles isolated from cardiosphere-derived cells are not functionally procoagulant and do not interfere with platelet function. In a rat model of multiple injuries and hemorrhagic shock, early administration of EVs significantly reduced the elevation of lactate and creatinine and did not significantly enhance coagulopathy in rats with acute traumatic coagulopathy. CONCLUSION: The results of this study are of great relevance to the development of EV products for use in combat casualty care, as our studies show that CDC-EVs have the potential to be an antishock therapeutic if administered early. These results demonstrate that research using CDC-EVs in trauma care needs to be considered and expanded beyond their reported cardioprotective benefits.

Choice of selectable marker affects recombinant protein expression in cells and exosomes.

Transgenic mammalian cells are used for numerous research, pharmaceutical, industrial, and clinical purposes, and dominant selectable markers are often used to enable the selection of transgenic cell lines. Using HEK293 cells, we show here that the choice of selectable marker gene has a significant impact on both the level of recombinant protein expression and the cell-to-cell variability in recombinant protein expression. Specifically, we observed that cell lines generated with the NeoR or BsdR selectable markers and selected in the antibiotics G418 or blasticidin, respectively, displayed the lowest level of recombinant protein expression as well as the greatest cell-to-cell variability in transgene expression. In contrast, cell lines generated with the BleoR marker and selected in zeocin yielded cell lines that expressed the highest levels of linked recombinant protein, approximately 10-fold higher than those selected using the NeoR or BsdR markers, as well as the lowest cell-to-cell variability in recombinant protein expression. Intermediate yet still-high levels of expression were observed in cells generated with the PuroR- or HygR-based vectors and that were selected in puromycin or hygromycin, respectively. Similar results were observed in the African green monkey cell line COS7. These data indicate that each combination of selectable marker and antibiotic establishes a threshold below which no cell can survive and that these thresholds vary significantly between different selectable markers. Moreover, we show that choice of selectable marker also affects recombinant protein expression in cell-derived exosomes, consistent with the hypothesis that exosome protein budding is a stochastic rather than determinative process.