Deep Artificial Neural Networks Reveal a Distributed Cortical Network Encoding Propositional Sentence-Level Meaning

Understanding how and where in the brain sentence-level meaning is constructed from words presents a major scientific challenge. Recent advances have begun to explain brain activation elicited by sentences using vector models of word meaning derived from patterns of word co-occurrence in text corpora. These studies have helped map out semantic representation across a distributed brain network spanning temporal, parietal, and frontal cortex. However, it remains unclear whether activation patterns within regions reflect unified representations of sentence-level meaning, as opposed to superpositions of context-independent component words. This is because models have typically represented sentences as “bags-of-words” that neglect sentence-level structure. To address this issue, we interrogated fMRI activation elicited as 240 sentences were read by 14 participants (9 female, 5 male), using sentences encoded by a recurrent deep artificial neural-network trained on a sentence inference task (InferSent). Recurrent connections and nonlinear filters enable InferSent to transform sequences of word vectors into unified “propositional” sentence representations suitable for evaluating intersentence entailment relations. Using voxelwise encoding modeling, we demonstrate that InferSent predicts elements of fMRI activation that cannot be predicted by bag-of-words models and sentence models using grammatical rules to assemble word vectors. This effect occurs throughout a distributed network, which suggests that propositional sentence-level meaning is represented within and across multiple cortical regions rather than at any single site. In follow-up analyses, we place results in the context of other deep network approaches (ELMo and BERT) and estimate the degree of unpredicted neural signal using an “experiential” semantic model and cross-participant encoding.SIGNIFICANCE STATEMENT A modern-day scientific challenge is to understand how the human brain transforms word sequences into representations of sentence meaning. A recent approach, emerging from advances in functional neuroimaging, big data, and machine learning, is to computationally model meaning, and use models to predict brain activity. Such models have helped map a cortical semantic information-processing network. However, how unified sentence-level information, as opposed to word-level units, is represented throughout this network remains unclear. This is because models have typically represented sentences as unordered “bags-of-words.” Using a deep artificial neural network that recurrently and nonlinearly combines word representations into unified propositional sentence representations, we provide evidence that sentence-level information is encoded throughout a cortical network, rather than in a single region.     

Gene transfer to primary normal and malignant human hemopoietic progenitors using recombinant retroviruses

To study the feasibility of using retroviruses for gene transfer into human hemopoietic cells, various cell types were exposed to virus carrying the gene for neomycin resistance (neor). In preliminary studies using K562 cells as targets, we found that high viral titer and co-cultivation with viral producer cells rather than incubation in medium exposed to viral producer cells were important variables for achieving high frequencies of G418 resistant (G418r) colonies. The maximum frequency of G418r K562 colonies after co-cultivation with cells producing a neor virus titer of 4 X 10(6) cfu/mL was 60%. When primary human progenitors from normal marrow, fetal liver, or chronic myelogenous leukemia blood were exposed to high titer viral stocks, both with and without helper virus, under conditions optimized for K562 cells, maximum frequencies of G418r colonies were 3% to 16% for granulocyte macrophage progenitors and 2% to 6% for primitive erythroid progenitors. The presence of the neor gene in both G418r K562 and primary hemopoietic colonies was verified by Southern blot. Expression of the neor gene was shown by RNA spot blot. These data demonstrate efficient transfer and expression of the neor gene in both K562 cells and primary human hemopoietic cells from normal and leukemic individuals.     

In Vitro Activity of Daptomycin in Combination with β-Lactams,Gentamicin, Rifampin,and Tigecycline against Daptomycin-Nonsusceptible Enterococci

Enterococci that are nonsusceptible (NS; MIC > 4 μg/ml) to daptomycin are an emerging clinical concern. The synergistic combination of daptomycin plus beta-lactams has been shown to be effective against vancomycin-resistant Enterococcus (VRE) species in vitro. This study systematically evaluated by in vitro time-kill studies the effect of daptomycin in combination with ampicillin, cefazolin, ceftriaxone, ceftaroline, ertapenem, gentamicin, tigecycline, and rifampin, for a collection of 9 daptomycin-NS enterococci that exhibited a broad range of MICs and different resistance-conferring mutations. We found that ampicillin plus daptomycin yielded the most consistent synergy but did so only for isolates with mutations to the liaFSR system. Daptomycin binding was found to be enhanced by ampicillin in a representative isolate with such mutations but not for an isolate with mutation to the yycFGHIJ system. In contrast, ampicillin enhanced the killing of the LL-37 human antimicrobial peptide against daptomycin-NS E. faecium with either the liaFSR or yycFGHIJ mutation. Antagonism was noted only for rifampin and tigecycline and only for 2 or 3 isolates. These data add support to the growing body of evidence indicating that therapy combining daptomycin and ampicillin may be helpful in eradicating refractory VRE infections.     

Ultrastructural changes to rabbit fibrin and platelets due to aspartame

The coagulation process, including thrombin, fibrin, as well as platelets, plays an important role in hemostasis, contributing to the general well-being of humans. Fibrin formation and platelet activation are delicate processes that are under the control of many small physiological events. Any one of these many processes may be influenced or changed by external factors, including pharmaceutical or nutritional products, e.g., the sweetener aspartame (L-aspartyl-L-phenylalanine methyl ester). It is known that phenylalanine is present at position P(9) and aspartate at position P(10) of the alpha-chain of human fibrinogen, and plays an important role in the conversion of fibrinogen to fibrin by the catalyst alpha-thrombin. The authors investigate the effect of aspartame on platelet and fibrin ultrastructure, by using the rabbit animal model and the scanning electron microscope. Animals were exposed to 34 mg/kg of aspartame 26x during a 2-month period. Aspartame-exposed fibrin networks appeared denser, with a thick matted fine fiber network covering thick major fibers. Also, the platelet aggregates appeared more granular than the globular control platelet aggregates. The authors conclude by suggesting that aspartame usage may interfere with the coagulation process and might cause delayed fibrin breakup after clot formation. They suggest this, as the fibrin networks from aspartame-exposed rabbits are more complex and dense, due to the netlike appearance of the minor, thin fibers. Aspartame usage should possibly be limited by people on anti-clotting medicine or those with prone to clot formation.     

Global Analysis Reveals the Complexity of the Human Glomerular Extracellular Matrix

The glomerulus contains unique cellular and extracellular matrix (ECM) components, which are required for intact barrier function. Studies of the cellular components have helped to build understanding of glomerular disease; however, the full composition and regulation of glomerular ECM remains poorly understood. We used mass spectrometry-based proteomics of enriched ECM extracts for a global analysis of human glomerular ECM in vivo and identified a tissue-specific proteome of 144 structural and regulatory ECM proteins. This catalog includes all previously identified glomerular components plus many new and abundant components. Relative protein quantification showed a dominance of collagen IV, collagen I, and laminin isoforms in the glomerular ECM together with abundant collagen VI and TINAGL1. Protein network analysis enabled the creation of a glomerular ECM interactome, which revealed a core of highly connected structural components. More than one half of the glomerular ECM proteome was validated using colocalization studies and data from the Human Protein Atlas. This study yields the greatest number of ECM proteins relative to previous investigations of whole glomerular extracts, highlighting the importance of sample enrichment. It also shows that the composition of glomerular ECM is far more complex than previously appreciated and suggests that many more ECM components may contribute to glomerular development and disease processes. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium with the dataset identifier PXD000456.The glomerulus is a sophisticated organelle comprising unique cellular and extracellular matrix (ECM) components. Fenestrated capillary endothelial cells and overlying podocytes are separated by a specialized glomerular basement membrane (GBM), and these three components together form the filtration barrier. Mesangial cells and their associated ECM, the mesangial matrix, exist between adjacent capillary loops and maintain the three-dimensional organization of the capillary bundle. In turn, the parietal epithelial cells and ECM of Bowman’s capsule enclose this network of capillaries. Cells adhere to ECM proteins by adhesion receptors, and these interactions are required to maintain intact barrier function of the glomerulus.^1,2In addition to operating as a signaling platform, ECM provides a structural scaffold for adjacent cells and has a tissue-specific molecular composition.^3,4 Candidate-based investigations of glomerular ECM have focused on the GBM and shown that it resembles the typical basal lamina found in multicellular organisms, containing a core of glycoproteins (collagen IV, laminins, and nidogens) and heparan sulfate proteoglycans (agrin, perlecan, and collagen XVIII).⁵ Mesangial and parietal cell ECMs have been less well investigated; nonetheless, they are also thought to contain similar core components in addition to other glycoproteins, including fibronectin.^6,7 Thus, the glomerulus consists of a combination of condensed ECM within the GBM and Bowman’s capsule and loose ECM supporting the mesangial cells.The ECM compartments in the glomerulus are thought to be distinct and exhibit different functional roles. The GBM is integral to the capillary wall and therefore, functionally linked to glomerular filtration.⁵ Mutations of tissue-restricted isoforms of collagen IV (COL4A3, COL4A4, and COL4A5) and laminin (LAMB2), which are found in the GBM, cause significant barrier dysfunction and ultimately, renal failure.^8,9 Less is understood about the functions of mesangial and parietal cell ECMs, although expansion of the mesangial compartment is a histologic pattern seen across the spectrum of glomerular disease.¹⁰Compositional investigation of the distinct glomerular ECM compartments is limited by the technical difficulties of separation. Early investigations of GBM constituents used the relative insolubility of ECM proteins to facilitate separation from cellular proteins in the glomerulus but did not separate the GBM from mesangial and parietal cells ECMs.^11,12 Recently, studies incorporating laser microdissection of glomerular sections have been coupled with proteomic analyses.^13,14 These studies report both cellular and ECM components and typically require pooled material from glomerular sections to improve protein identification. The ability of laser microdissection to separate glomerular ECM compartments has not yet been tested; however, this approach will be limited by the amount of protein that is possible to retrieve. To achieve good coverage of ECM proteins within a tissue, proteomic studies need to combine a reduction in sample complexity with maximal protein quantity. Currently, the inability to separate glomerular ECM compartments in sufficient quantity is a limitation that prohibits proteomic studies of these structures; however, for other tissues, proteomic analysis of ECM has been achieved by enrichment of ECM combined with sample fractionation.¹⁵Although the composition of the ECM in other tissues has been addressed using proteomic approaches,¹⁵ studies of glomerular ECM to date have used candidate-based technologies. These studies have identified key molecular changes during development and disease and highlighted the compositional and organizational dynamics of glomerular ECM. Nonetheless, the extracellular environment within the glomerulus is the setting for a complex series of interactions between both structural ECM proteins and ECM-associated proteins, such as growth factors^16–18 and proteases,¹⁹ which together provide a specialized niche to support glomerular cell function. Therefore, to interrogate this complexity effectively, a systems-level understanding of glomerular ECM is required. To address the need for a global analysis of the extracellular environment within the glomerulus, we used mass spectrometry (MS)-based proteomics of glomerular ECM fractions to define the human glomerular ECM proteome.