Content-aware frame interpolation (CAFI): deep learning-based temporal super-resolution for fast bioimaging.

The development of high-resolution microscopes has made it possible to investigate cellular processes in 3D and over time. However, observing fast cellular dynamics remains challenging because of photobleaching and phototoxicity. Here we report the implementation of two content-aware frame interpolation (CAFI) deep learning networks, Zooming SlowMo and Depth-Aware Video Frame Interpolation, that are highly suited for accurately predicting images in between image pairs, therefore improving the temporal resolution of image series post-acquisition. We show that CAFI is capable of understanding the motion context of biological structures and can perform better than standard interpolation methods. We benchmark CAFI's performance on 12 different datasets, obtained from four different microscopy modalities, and demonstrate its capabilities for single-particle tracking and nuclear segmentation. CAFI potentially allows for reduced light exposure and phototoxicity on the sample for improved long-term live-cell imaging. The models and the training and testing data are available via the ZeroCostDL4Mic platform.

Altered Fhod3 expression involved in progressive high-frequency hearing loss via dysregulation of actin polymerization stoichiometry in the cuticular plate.

Age-related hearing loss (ARHL) is a common sensory impairment with complex underlying mechanisms. In our previous study, we performed a meta-analysis of genome-wide association studies (GWAS) in mice and identified a novel locus on chromosome 18 associated with ARHL specifically linked to a 32 kHz tone burst stimulus. Consequently, we investigated the role of Formin Homology 2 Domain Containing 3 (Fhod3), a newly discovered candidate gene for ARHL based on the GWAS results. We observed Fhod3 expression in auditory hair cells (HCs) primarily localized at the cuticular plate (CP). To understand the functional implications of Fhod3 in the cochlea, we generated Fhod3 overexpression mice (Pax2-Cre+/-; Fhod3Tg/+) (TG) and HC-specific conditional knockout mice (Atoh1-Cre+/-; Fhod3fl/fl) (KO). Audiological assessments in TG mice demonstrated progressive high-frequency hearing loss, characterized by predominant loss of outer hair cells, and a decreased phalloidin intensities of CP. Ultrastructural analysis revealed loss of the shortest row of stereocilia in the basal turn of the cochlea, and alterations in the cuticular plate surrounding stereocilia rootlets. Importantly, the hearing and HC phenotype in TG mice phenocopied that of the KO mice. These findings suggest that balanced expression of Fhod3 is critical for proper CP and stereocilia structure and function. Further investigation of Fhod3 related hearing impairment mechanisms may lend new insight towards the myriad mechanisms underlying ARHL, which in turn could facilitate the development of therapeutic strategies for ARHL.

Local shape descriptors for neuron segmentation.

We present an auxiliary learning task for the problem of neuron segmentation in electron microscopy volumes. The auxiliary task consists of the prediction of local shape descriptors (LSDs), which we combine with conventional voxel-wise direct neighbor affinities for neuron boundary detection. The shape descriptors capture local statistics about the neuron to be segmented, such as diameter, elongation, and direction. On a study comparing several existing methods across various specimen, imaging techniques, and resolutions, auxiliary learning of LSDs consistently increases segmentation accuracy of affinity-based methods over a range of metrics. Furthermore, the addition of LSDs promotes affinity-based segmentation methods to be on par with the current state of the art for neuron segmentation (flood-filling networks), while being two orders of magnitudes more efficient-a critical requirement for the processing of future petabyte-sized datasets.

Targeting LIF-mediated paracrine interaction for pancreatic cancer therapy and monitoring.

Pancreatic ductal adenocarcinoma (PDAC) has a dismal prognosis largely owing to inefficient diagnosis and tenacious drug resistance. Activation of pancreatic stellate cells (PSCs) and consequent development of dense stroma are prominent features accounting for this aggressive biology1,2. The reciprocal interplay between PSCs and pancreatic cancer cells (PCCs) not only enhances tumour progression and metastasis but also sustains their own activation, facilitating a vicious cycle to exacerbate tumorigenesis and drug resistance3-7. Furthermore, PSC activation occurs very early during PDAC tumorigenesis8-10, and activated PSCs comprise a substantial fraction of the tumour mass, providing a rich source of readily detectable factors. Therefore, we hypothesized that the communication between PSCs and PCCs could be an exploitable target to develop effective strategies for PDAC therapy and diagnosis. Here, starting with a systematic proteomic investigation of secreted disease mediators and underlying molecular mechanisms, we reveal that leukaemia inhibitory factor (LIF) is a key paracrine factor from activated PSCs acting on cancer cells. Both pharmacologic LIF blockade and genetic Lifr deletion markedly slow tumour progression and augment the efficacy of chemotherapy to prolong survival of PDAC mouse models, mainly by modulating cancer cell differentiation and epithelial-mesenchymal transition status. Moreover, in both mouse models and human PDAC, aberrant production of LIF in the pancreas is restricted to pathological conditions and correlates with PDAC pathogenesis, and changes in the levels of circulating LIF correlate well with tumour response to therapy. Collectively, these findings reveal a function of LIF in PDAC tumorigenesis, and suggest its translational potential as an attractive therapeutic target and circulating marker. Our studies underscore how a better understanding of cell-cell communication within the tumour microenvironment can suggest novel strategies for cancer therapy.

Deep learning-based point-scanning super-resolution imaging.

Point-scanning imaging systems are among the most widely used tools for high-resolution cellular and tissue imaging, benefiting from arbitrarily defined pixel sizes. The resolution, speed, sample preservation and signal-to-noise ratio (SNR) of point-scanning systems are difficult to optimize simultaneously. We show these limitations can be mitigated via the use of deep learning-based supersampling of undersampled images acquired on a point-scanning system, which we term point-scanning super-resolution (PSSR) imaging. We designed a 'crappifier' that computationally degrades high SNR, high-pixel resolution ground truth images to simulate low SNR, low-resolution counterparts for training PSSR models that can restore real-world undersampled images. For high spatiotemporal resolution fluorescence time-lapse data, we developed a 'multi-frame' PSSR approach that uses information in adjacent frames to improve model predictions. PSSR facilitates point-scanning image acquisition with otherwise unattainable resolution, speed and sensitivity. All the training data, models and code for PSSR are publicly available at 3DEM.org.

Obstructive Sleep Apnea-induced Endothelial Dysfunction Is Mediated by miR-210.

Rationale: Obstructive sleep apnea (OSA)-induced endothelial cell (EC) dysfunction contributes to OSA-related cardiovascular sequelae. The mechanistic basis of endothelial impairment by OSA is unclear. Objectives: The goals of this study were to identify the mechanism of OSA-induced EC dysfunction and explore the potential therapies for OSA-accelerated cardiovascular disease. Methods: The experimental methods include data mining, bioinformatics, EC functional analyses, OSA mouse models, and assessment of OSA human subjects. Measurements and Main Results: Using mined microRNA sequencing data, we found that microRNA 210 (miR-210) conferred the greatest induction by intermittent hypoxia in ECs. Consistently, the serum concentration of miR-210 was higher in individuals with OSA from two independent cohorts. Importantly, miR-210 concentration was positively correlated with the apnea-hypopnea index. RNA sequencing data collected from ECs transfected with miR-210 or treated with OSA serum showed a set of genes commonly altered by miR-210 and OSA serum, which are largely involved in mitochondrion-related pathways. ECs transfected with miR-210 or treated with OSA serum showed reduced [Formula: see text]o2 rate, mitochondrial membrane potential, and DNA abundance. Mechanistically, intermittent hypoxia-induced SREBP2 (sterol regulatory element-binding protein 2) bound to the promoter region of miR-210, which in turn inhibited the iron-sulfur cluster assembly enzyme and led to mitochondrial dysfunction. Moreover, the SREBP2 inhibitor betulin alleviated intermittent hypoxia-increased systolic blood pressure in the OSA mouse model. Conclusions: These results identify an axis involving SREBP2, miR-210, and mitochondrial dysfunction, representing a new mechanistic link between OSA and EC dysfunction that may have important implications for treating and preventing OSA-related cardiovascular sequelae.

Dynamic regulation of CTCF stability and sub-nuclear localization in response to stress.

The nuclear protein CCCTC-binding factor (CTCF) has diverse roles in chromatin architecture and gene regulation. Functionally, CTCF associates with thousands of genomic sites and interacts with proteins, such as cohesin, or non-coding RNAs to facilitate specific transcriptional programming. In this study, we examined CTCF during the cellular stress response in human primary cells using immune-blotting, quantitative real time-PCR, chromatin immunoprecipitation-sequence (ChIP-seq) analysis, mass spectrometry, RNA immunoprecipitation-sequence analysis (RIP-seq), and Airyscan confocal microscopy. Unexpectedly, we found that CTCF is exquisitely sensitive to diverse forms of stress in normal patient-derived human mammary epithelial cells (HMECs). In HMECs, a subset of CTCF protein forms complexes that localize to Serine/arginine-rich splicing factor (SC-35)-containing nuclear speckles. Upon stress, this species of CTCF protein is rapidly downregulated by changes in protein stability, resulting in loss of CTCF from SC-35 nuclear speckles and changes in CTCF-RNA interactions. Our ChIP-seq analysis indicated that CTCF binding to genomic DNA is largely unchanged. Restoration of the stress-sensitive pool of CTCF protein abundance and re-localization to nuclear speckles can be achieved by inhibition of proteasome-mediated degradation. Surprisingly, we observed the same characteristics of the stress response during neuronal differentiation of human pluripotent stem cells (hPSCs). CTCF forms stress-sensitive complexes that localize to SC-35 nuclear speckles during a specific stage of neuronal commitment/development but not in differentiated neurons. We speculate that these particular CTCF complexes serve a role in RNA processing that may be intimately linked with specific genes in the vicinity of nuclear speckles, potentially to maintain cells in a certain differentiation state, that is dynamically regulated by environmental signals. The stress-regulated activity of CTCF is uncoupled in persistently stressed, epigenetically re-programmed "variant" HMECs and certain cancer cell lines. These results reveal new insights into CTCF function in cell differentiation and the stress-response with implications for oxidative damage-induced cancer initiation and neuro-degenerative diseases.

Single-Cell Transcriptomics Reveals a Conserved Metaplasia Program in Pancreatic Injury.

BACKGROUND & AIMS: Acinar to ductal metaplasia (ADM) occurs in the pancreas in response to tissue injury and is a potential precursor for adenocarcinoma. The goal of these studies was to define the populations arising from ADM, the associated transcriptional changes, and markers of disease progression. METHODS: Acinar cells were lineage-traced with enhanced yellow fluorescent protein (EYFP) to follow their fate post-injury. Transcripts of more than 13,000 EYFP+ cells were determined using single-cell RNA sequencing (scRNA-seq). Developmental trajectories were generated. Data were compared with gastric metaplasia, KrasG12D-induced neoplasia, and human pancreatitis. Results were confirmed by immunostaining and electron microscopy. KrasG12D was expressed in injury-induced ADM using several inducible Cre drivers. Surgical specimens of chronic pancreatitis from 15 patients were evaluated by immunostaining. RESULTS: scRNA-seq of ADM revealed emergence of a mucin/ductal population resembling gastric pyloric metaplasia. Lineage trajectories suggest that some pyloric metaplasia cells can generate tuft and enteroendocrine cells (EECs). Comparison with KrasG12D-induced ADM identifies populations associated with disease progression. Activation of KrasG12D expression in HNF1B+ or POU2F3+ ADM populations leads to neoplastic transformation and formation of MUC5AC+ gastric-pit-like cells. Human pancreatitis samples also harbor pyloric metaplasia with a similar transcriptional phenotype. CONCLUSIONS: Under conditions of chronic injury, acinar cells undergo a pyloric-type metaplasia to mucinous progenitor-like populations, which seed disparate tuft cell and EEC lineages. ADM-derived EEC subtypes are diverse. KrasG12D expression is sufficient to drive neoplasia when targeted to injury-induced ADM populations and offers an alternative origin for tumorigenesis. This program is conserved in human pancreatitis, providing insight into early events in pancreas diseases.

Actin chromobody imaging reveals sub-organellar actin dynamics.

The actin cytoskeleton plays multiple critical roles in cells, from cell migration to organelle dynamics. The small and transient actin structures regulating organelle dynamics are challenging to detect with fluorescence microscopy, making it difficult to determine whether actin filaments are directly associated with specific membranes. To address these limitations, we developed fluorescent-protein-tagged actin nanobodies, termed 'actin chromobodies' (ACs), targeted to organelle membranes to enable high-resolution imaging of sub-organellar actin dynamics.

Association between Thrombin Generation and Clinical Characteristics in COVID-19 Patients.

INTRODUCTION: Coronavirus disease 2019 (COVID-19) disease is associated with coagulopathy and an increased risk of thrombosis. An association between thrombin generation (TG) capacity, disease severity, and outcomes has not been well described. METHODS: We assessed the correlation of TG with sequential organ failure assessment (SOFA) and sepsis-induced coagulopathy (SIC) scores and clinical outcomes by analysis of plasma samples obtained from hospitalized COVID-19 patients. RESULTS: 32 patients (68.8% male), whose median age was 69 years, were assessed, of whom only 3 patients did not receive anticoagulant therapy. D-dimers were uniformly increased. During hospitalization, 2 patients suffered thrombosis, 3 experienced bleeding, and 12 died. TG parameters from anticoagulated COVID-19 patients did not significantly differ from the values obtained from non-anticoagulated healthy controls. Patients who received higher than prophylactic doses of anticoagulant therapy had increased lag time (p = 0.003), lower endogenous thrombin potential (ETP) (p = 0.037), and a reduced peak height (p = 0.006). ETP correlated with the SIC score (p = 0.038). None of the TG parameters correlated with the SOFA score or were associated with mortality. CONCLUSION: TG was not associated with disease severity among patients hospitalized with COVID-19. However, a correlation between ETP and the SIC score was noted and deserves attention.

Safety of the BNT162b2 mRNA COVID-19 vaccine in patients with familial Mediterranean fever.

OBJECTIVES: Evidence suggests a possible association between the COVID-19 vaccine and autoimmune disease flares or new onset of various autoinflammatory manifestations, such as pericarditis and myocarditis. The objective of this study was to assess the safety of an mRNA-based BNT162b2 anti-COVID-19 vaccine in individuals with FMF, a prototypic autoinflammatory disease. METHODS: Patients participating in this study fulfilled the criteria for diagnosis of FMF, were older than 18 years and received at least one dose of the vaccine. Data on baseline characteristics, features of FMF, post-vaccination side effects, and disease flares were acquired using electronic medical files and telephone interviews. RESULTS: A total of 273 FMF patients were recruited for the study. >95% were vaccinated with two doses of the vaccine. The rates of local reactions following the first and second vaccine doses were 65.5% and 60%, respectively, and 26% and 50.4%, respectively, for systemic adverse events. These rates are lower than those reported for the general population from real-world and clinical trial settings. Postvaccination FMF activity remained stable in most patients. None of the patients reported an attack of pericarditis or myocarditis, considered the most serious vaccine-associated adverse events. Patients with a more active FMF disease and patients harboring the M694V mutation had a significantly higher rate of post-vaccination systemic side effects and attacks. CONCLUSION: The BNT162b2 mRNA COVID-19 vaccine is safe in patients with FMF. Our results support the administration of this vaccine to FMF patients according to guidelines applicable to the general population.

Tuft Cells Inhibit Pancreatic Tumorigenesis in Mice by Producing Prostaglandin D2.

BACKGROUND & AIMS: Development of pancreatic ductal adenocarcinoma (PDA) involves acinar to ductal metaplasia and genesis of tuft cells. It has been a challenge to study these rare cells because of the lack of animal models. We investigated the role of tuft cells in pancreatic tumorigenesis. METHODS: We performed studies with LSL-KrasG12D/+;Ptf1aCre/+ mice (KC; develop pancreatic tumors), KC mice crossed with mice with pancreatic disruption of Pou2f3 (KPouC mice; do not develop tuft cells), or mice with pancreatic disruption of the hematopoietic prostaglandin D synthase gene (Hpgds, KHC mice) and wild-type mice. Mice were allowed to age or were given caerulein to induce pancreatitis; pancreata were collected and analyzed by histology, immunohistochemistry, RNA sequencing, ultrastructural microscopy, and metabolic profiling. We performed laser-capture dissection and RNA-sequencing analysis of pancreatic tissues from 26 patients with pancreatic intraepithelial neoplasia (PanIN), 19 patients with intraductal papillary mucinous neoplasms (IPMNs), and 197 patients with PDA. RESULTS: Pancreata from KC mice had increased formation of tuft cells and higher levels of prostaglandin D2 than wild-type mice. Pancreas-specific deletion of POU2F3 in KC mice (KPouC mice) resulted in a loss of tuft cells and accelerated tumorigenesis. KPouC mice had increased fibrosis and activation of immune cells after administration of caerulein. Pancreata from KPouC and KHC mice had significantly lower levels of prostaglandin D2, compared with KC mice, and significantly increased numbers of PanINs and PDAs. KPouC and KHC mice had increased pancreatic injury after administration of caerulein, significantly less normal tissue, more extracellular matrix deposition, and higher PanIN grade than KC mice. Human PanIN and intraductal papillary mucinous neoplasm had gene expression signatures associated with tuft cells and increased expression of Hpgds messenger RNA compared with PDA. CONCLUSIONS: In mice with KRAS-induced pancreatic tumorigenesis, loss of tuft cells accelerates tumorigenesis and increases the severity of caerulein-induced pancreatic injury, via decreased production of prostaglandin D2. These data are consistent with the hypothesis that tuft cells are a metaplasia-induced tumor attenuating cell type.

Tectorins crosslink type II collagen fibrils and connect the tectorial membrane to the spiral limbus.

All inner ear organs possess extracellular matrix appendices over the sensory epithelia that are crucial for their proper function. The tectorial membrane (TM) is a gelatinous acellular membrane located above the hearing sensory epithelium and is composed mostly of type II collagen, and α and ß tectorins. TM molecules self-assemble in the endolymph fluid environment, interacting medially with the spiral limbus and distally with the outer hair cell stereocilia. Here, we used immunogold labeling in freeze-substituted mouse cochleae to assess the fine localization of both tectorins in distinct TM regions. We observed that the TM adheres to the spiral limbus through a dense thin matrix enriched in α- and ß-tectorin, both likely bound to the membranes of interdental cells. Freeze-etching images revealed that type II collagen fibrils were crosslinked by short thin filaments (4±1.5nm, width), resembling another collagen type protein, or chains of globular elements (15±3.2nm, diameter). Gold-particles for both tectorins also localized adjacent to the type II collagen fibrils, suggesting that these globules might be composed essentially of α- and ß-tectorins. Finally, the presence of gold-particles at the TM lower side suggests that the outer hair cell stereocilia membrane has a molecular partner to tectorins, probably stereocilin, allowing the physical connection between the TM and the organ of Corti.

Progressive hearing loss and gradual deterioration of sensory hair bundles in the ears of mice lacking the actin-binding protein Eps8L2.

Mechanotransduction in the mammalian auditory system depends on mechanosensitive channels in the hair bundles that project from the apical surface of the sensory hair cells. Individual stereocilia within each bundle contain a core of tightly packed actin filaments, whose length is dynamically regulated during development and in the adult. We show that the actin-binding protein epidermal growth factor receptor pathway substrate 8 (Eps8)L2, a member of the Eps8-like protein family, is a newly identified hair bundle protein that is localized at the tips of stereocilia of both cochlear and vestibular hair cells. It has a spatiotemporal expression pattern that complements that of Eps8. In the cochlea, whereas Eps8 is essential for the initial elongation of stereocilia, Eps8L2 is required for their maintenance in adult hair cells. In the absence of both proteins, the ordered staircase structure of the hair bundle in the cochlea decays. In contrast to the early profound hearing loss associated with an absence of Eps8, Eps8L2 null-mutant mice exhibit a late-onset, progressive hearing loss that is directly linked to a gradual deterioration in hair bundle morphology. We conclude that Eps8L2 is required for the long-term maintenance of the staircase structure and mechanosensory function of auditory hair bundles. It complements the developmental role of Eps8 and is a candidate gene for progressive age-related hearing loss.

Myosin 3A kinase activity is regulated by phosphorylation of the kinase domain activation loop.

Class III myosins are unique members of the myosin superfamily in that they contain both a motor and kinase domain. We have found that motor activity is decreased by autophosphorylation, although little is known about the regulation of the kinase domain. We demonstrate by mass spectrometry that Thr-178 and Thr-184 in the kinase domain activation loop and two threonines in the loop 2 region of the motor domain are autophosphorylated (Thr-908 and Thr-919). The kinase activity of MYO3A 2IQ with the phosphomimic (T184E) or phosphoblock (T184A) mutations demonstrates that kinase activity is reduced 30-fold as a result of the T184A mutation, although the Thr-178 site only had a minor impact on kinase activity. Interestingly, the actin-activated ATPase activity of MYO3A 2IQ is slightly reduced as a result of the T178A and T184A mutations suggesting coupling between motor and kinase domains. Full-length GFP-tagged T184A and T184E MYO3A constructs transfected into COS7 cells do not disrupt the ability of MYO3A to localize to filopodia structures. In addition, we demonstrate that T184E MYO3A reduces filopodia elongation in the presence of espin-1, whereas T184A enhances filopodia elongation in a similar fashion to kinase-dead MYO3A. Our results suggest that as MYO3A accumulates at the tips of actin protrusions, autophosphorylation of Thr-184 enhances kinase activity resulting in phosphorylation of the MYO3A motor and reducing motor activity. The differential regulation of the kinase and motor activities allows for MYO3A to precisely self-regulate its concentration in the actin bundle-based structures of cells.

PSSR2: a user-friendly Python package for democratizing deep learning-based point-scanning super-resolution microscopy.

PSSR2 improves and expands on the previously established PSSR (Point-Scanning Super-Resolution) workflow for simultaneous super-resolution and denoising of undersampled microscopy data. PSSR2 is designed to put state-of-the-art technology into the hands of the general microscopy and biology research community, enabling user-friendly implementation of PSSR workflows with little to no programming experience required, especially through its integrated CLI and Napari plugin.

A deep learning-based strategy for producing dense 3D segmentations from sparsely annotated 2D images.

Producing dense 3D reconstructions from biological imaging data is a challenging instance segmentation task that requires significant ground-truth training data for effective and accurate deep learning-based models. Generating training data requires intense human effort to annotate each instance of an object across serial section images. Our focus is on the especially complicated brain neuropil, comprising an extensive interdigitation of dendritic, axonal, and glial processes visualized through serial section electron microscopy. We developed a novel deep learning-based method to generate dense 3D segmentations rapidly from sparse 2D annotations of a few objects on single sections. Models trained on the rapidly generated segmentations achieved similar accuracy as those trained on expert dense ground-truth annotations. Human time to generate annotations was reduced by three orders of magnitude and could be produced by non-expert annotators. This capability will democratize generation of training data for large image volumes needed to achieve brain circuits and measures of circuit strengths.