ABSTRACT
Bioimage analysis of fluorescent labels is widely used in the life sciences. Recent advances in deep learning (DL) allow automating time-consuming manual image analysis processes based on annotated training data. However, manual annotation of fluorescent features with a low signal-to-noise ratio is somewhat subjective. Training DL models on subjective annotations may be instable or yield biased models. In turn, these models may be unable to reliably detect biological effects. An analysis pipeline integrating data annotation, ground truth estimation, and model training can mitigate this risk. To evaluate this integrated process, we compared different DL-based analysis approaches. With data from two model organisms (mice, zebrafish) and five laboratories, we show that ground truth estimation from multiple human annotators helps to establish objectivity in fluorescent feature annotations. Furthermore, ensembles of multiple models trained on the estimated ground truth establish reliability and validity. Our research provides guidelines for reproducible DL-based bioimage analyses.
Research in biology generates many image datasets, mostly from microscopy. These images have to be analyzed, and much of this analysis relies on a human expert looking at the images and manually annotating features. Image datasets are often large, and human annotation can be subjective, so automating image analysis is highly desirable. This is where machine learning algorithms, such as deep learning, have proven to be useful. In order for deep learning algorithms to work first they have to be 'trained'. Deep learning algorithms are trained by being given a training dataset that has been annotated by human experts. The algorithms extract the relevant features to look out for from this training dataset and can then look for these features in other image data. However, it is also worth noting that because these models try to mimic the annotation behavior presented to them during training as well as possible, they can sometimes also mimic an expert's subjectivity when annotating data. Segebarth, Griebel et al. asked whether this was the case, whether it had an impact on the outcome of the image data analysis, and whether it was possible to avoid this problem when using deep learning for imaging dataset analysis. For this research, Segebarth, Griebel et al. used microscopy images of mouse brain sections, where a protein called cFOS had been labeled with a fluorescent tag. This protein typically controls the rate at which DNA information is copied into RNA, leading to the production of proteins. Its activity can be influenced experimentally by testing the behaviors of mice. Thus, this experimental manipulation can be used to evaluate the results of deep learning-based image analyses. First, the fluorescent images were interpreted manually by a group of human experts. Then, their results were used to train a large variety of deep learning models. Models were trained either on the results of an individual expert or on the results pooled from all experts to come up with a consensus model, a deep learning model that learned from the personal annotation preferences of all experts. This made it possible to test whether training a model on multiple experts reduces the risk of subjectivity. As the training of deep learning models is random, Segebarth, Griebel et al. also tested whether combining the predictions from multiple models in a so-called model ensemble improves the consistency of the analyses. For evaluation, the annotations of the deep learning models were compared to those of the human experts, to ensure that the results were not influenced by the subjective behavior of one person. The results of all bioimage annotations were finally compared to the experimental results from analyzing the mice's behaviors in order to check whether the models were able to find the behavioral effect on cFOS. Segebarth, Griebel et al. concluded that combining the expert knowledge of multiple experts reduces the subjectivity of bioimage annotation by deep learning algorithms. Combining such consensus information in a group of deep learning models improves the quality of bioimage analysis, so that the results are reliable, transparent and less subjective.
Subject(s)
Image Processing, Computer-Assisted/methods , Animals , Deep Learning , Fear , Fluorescent Dyes , Male , Mice , Reproducibility of Results , Signal-To-Noise Ratio , ZebrafishABSTRACT
Impairments in neuronal circuits underly multiple neurodevelopmental and neurodegenerative disorders. 3D cell culture models enhance the complexity of in vitro systems and provide a microenvironment closer to the native situation than with 2D cultures. Such novel model systems will allow the assessment of neuronal network formation and their dysfunction under disease conditions. Here, mouse cortical neurons are cultured from embryonic day E17 within in a fiber-reinforced matrix. A soft Matrigel with a shear modulus of 31 ± 5.6 Pa is reinforced with scaffolds created by melt electrowriting, improving its mechanical properties and facilitating the handling. Cortical neurons display enhance cell viability and the neuronal network maturation in 3D, estimated by staining of dendrites and synapses over 21 days in vitro, is faster in 3D compared to 2D cultures. Using functional readouts with electrophysiological recordings, different firing patterns of action potentials are observed, which are absent in the presence of the sodium channel blocker, tetrodotoxin. Voltage-gated sodium currents display a current-voltage relationship with a maximum peak current at -25 mV. With its high customizability in terms of scaffold reinforcement and soft matrix formulation, this approach represents a new tool to study neuronal networks in 3D under normal and, potentially, disease conditions.
Subject(s)
Neurogenesis , Neurons , Action Potentials , Animals , Cell Culture Techniques , Cells, Cultured , Mice , Printing, Three-DimensionalSubject(s)
Analgesics/therapeutic use , Anti-Inflammatory Agents/therapeutic use , Antibodies, Monoclonal/therapeutic use , Autoantibodies/therapeutic use , Hyperalgesia/drug therapy , Pain/drug therapy , Phosphatidylcholines/antagonists & inhibitors , Aldehydes/pharmacology , Analgesics/pharmacology , Animals , Anti-Inflammatory Agents/pharmacology , Antibodies, Monoclonal/pharmacology , Autoantibodies/pharmacology , Capsaicin/pharmacology , Ganglia, Spinal , HEK293 Cells , Humans , Irritants/pharmacology , Irritants/urine , Isothiocyanates/pharmacology , Male , Mice , Mice, Inbred C57BL , Neurons/drug effects , Oxidation-Reduction , Phosphatidylcholines/immunology , Phosphatidylcholines/metabolism , Rats, Wistar , TRPA1 Cation Channel/agonists , TRPA1 Cation Channel/physiology , TRPV Cation Channels/agonists , TRPV Cation Channels/physiologyABSTRACT
Local and spontaneous calcium signals play important roles in neurons and neuronal networks. Spontaneous or cell-autonomous calcium signals may be difficult to assess because they appear in an unpredictable spatiotemporal pattern and in very small neuronal loci of axons or dendrites. We developed an open source bioinformatics tool for an unbiased assessment of calcium signals in x,y-t imaging series. The tool bases its algorithm on a continuous wavelet transform-guided peak detection to identify calcium signal candidates. The highly sensitive calcium event definition is based on identification of peaks in 1D data through analysis of a 2D wavelet transform surface. For spatial analysis, the tool uses a grid to separate the x,y-image field in independently analyzed grid windows. A document containing a graphical summary of the data is automatically created and displays the loci of activity for a wide range of signal intensities. Furthermore, the number of activity events is summed up to create an estimated total activity value, which can be used to compare different experimental situations, such as calcium activity before or after an experimental treatment. All traces and data of active loci become documented. The tool can also compute the signal variance in a sliding window to visualize activity-dependent signal fluctuations. We applied the calcium signal detector to monitor activity states of cultured mouse neurons. Our data show that both the total activity value and the variance area created by a sliding window can distinguish experimental manipulations of neuronal activity states. Notably, the tool is powerful enough to compute local calcium events and 'signal-close-to-noise' activity in small loci of distal neurites of neurons, which remain during pharmacological blockade of neuronal activity with inhibitors such as tetrodotoxin, to block action potential firing, or inhibitors of ionotropic glutamate receptors. The tool can also offer information about local homeostatic calcium activity events in neurites.
Subject(s)
Calcium Signaling/physiology , Computational Biology/methods , Molecular Imaging/methods , Algorithms , Animals , Axons/metabolism , Calcium/metabolism , Cells, Cultured , Mice , Neurons/physiology , Software , Spatio-Temporal AnalysisABSTRACT
Oxidized phospholipids (OxPL) like oxidized 1-palmitoyl-2-arachidonoyl-sn-glycero-3-phosphocholine (OxPAPC) were recently identified as novel proalgesic targets in acute and chronic inflammatory pain. These endogenous chemical irritants are generated in inflamed tissue and mediate their pain-inducing function by activating the transient receptor potential channels TRPA1 and TRPV1 expressed in sensory neurons. Notably, prototypical therapeutics interfering with OxPL were shown to inhibit TRP channel activation and pain behavior. Here, we asked how OxPL excite primary sensory neurons of dorsal root ganglia (DRG neurons from mice of either sex). Acute stimulation of sensory neurons with the prototypical OxPL 1-palmitoyl-2-glutaryl-sn-glycero-3-phosphocholine (PGPC) evoked repetitive calcium spikes in small-diameter neurons. As NaV1.9, a voltage-gated sodium channel involved in nociceptor excitability, was previously shown to be essential for the generation of calcium spikes in motoneurons, we asked if this channel is also important for OxPL mediated calcium spike and action potential generation in nociceptors. In wild-type and NaV1.9-deficient neurons, the action potential firing rate and the calcium spike frequency to an acute PGPC stimulus was similar. When preincubated with inflammatory mediators, both, the action potential firing rate and the calcium spike frequency were markedly increased in response to an acute PGPC stimulus. However, this potentiating effect was completely lost in NaV1.9-deficient small-diameter neurons. After treatment with inflammatory mediators, the resting membrane potential of NaV1.9 KO neurons was slightly more negative than that of wild-type control neurons. This suggests that NaV1.9 channels are active under this condition and therefore increases the ease with which action potentials are elicited after OxPL stimulation. In summary, our data suggest that NaV1.9 has a switch function to potentiate the receptor potentials induced by OxPL under inflammatory conditions. Since human NaV1.9 has been shown to mediate painful and painless channelopathies, this study provides new insights into the mechanism by which NaV1.9 amplifies stimuli of endogenous irritants under inflammatory conditions.
ABSTRACT
Phospholipids occurring in cell membranes and lipoproteins are converted into oxidized phospholipids (OxPL) by oxidative stress promoting atherosclerotic plaque formation. Here, OxPL were characterized as novel targets in acute and chronic inflammatory pain. Oxidized 1-palmitoyl-2-arachidonoyl-sn-glycero-3-phosphocholine (OxPAPC) and its derivatives were identified in inflamed tissue by mass spectrometry and binding assays. They elicited calcium influx, hyperalgesia and induced pro-nociceptive peptide release. Genetic, pharmacological and mass spectrometric evidence in vivo as well as in vitro confirmed the role of transient receptor potential channels (TRPA1 and TRPV1) as OxPAPC targets. Treatment with the monoclonal antibody E06 or with apolipoprotein A-I mimetic peptide D-4F, capturing OxPAPC in atherosclerosis, prevented inflammatory hyperalgesia, and in vitro TRPA1 activation. Administration of D-4F or E06 to rats profoundly ameliorated mechanical hyperalgesia and inflammation in collagen-induced arthritis. These data reveal a clinically relevant role for OxPAPC in inflammation offering therapy for acute and chronic inflammatory pain treatment by scavenging OxPAPC.
Subject(s)
Antibodies, Monoclonal/pharmacology , Apolipoprotein A-I/pharmacology , Arthritis, Experimental/drug therapy , Hyperalgesia/drug therapy , Pain/drug therapy , Phosphatidylcholines/antagonists & inhibitors , TRPA1 Cation Channel/genetics , TRPV Cation Channels/genetics , Animals , Arthritis, Experimental/chemically induced , Arthritis, Experimental/metabolism , Arthritis, Experimental/pathology , Calcium/metabolism , Calcium Signaling/drug effects , Collagen Type II/administration & dosage , Female , Gene Expression , HEK293 Cells , Hindlimb , Humans , Hyperalgesia/chemically induced , Hyperalgesia/metabolism , Hyperalgesia/pathology , Male , Nociception/drug effects , Nociception/physiology , Pain/chemically induced , Pain/metabolism , Pain/pathology , Patch-Clamp Techniques , Phosphatidylcholines/metabolism , Phosphatidylcholines/pharmacology , Rats , Rats, Inbred Lew , Rats, Wistar , TRPA1 Cation Channel/antagonists & inhibitors , TRPA1 Cation Channel/metabolism , TRPV Cation Channels/antagonists & inhibitors , TRPV Cation Channels/metabolismABSTRACT
Auto-antibodies against the paranodal proteins neurofascin-155 and contactin-1 have recently been described in patients with chronic inflammatory demyelinating polyradiculoneuropathy and are associated with a distinct clinical phenotype and response to treatment. Contactin-associated protein 1 (Caspr, encoded by CNTNAP1) is a paranodal protein that is attached to neurofascin-155 and contactin-1 (CNTN1) but has not yet been identified as a sole antigen in patients with inflammatory neuropathies. In the present study, we screened a cohort of 35 patients with chronic inflammatory demyelinating polyradiculoneuropathy (age range 20-80, 10 female, 25 male) and 22 patients with Guillain-Barré syndrome (age range 17-86, eight female, 14 male) for autoantibodies against paranodal antigens. We identified two patients, one with chronic inflammatory demyelinating polyradiculoneuropathy and one with Guillain-Barré syndrome, with autoantibodies against Caspr by binding assays using Caspr transfected human embryonic kidney cells and murine teased fibres. IgG3 was the predominant autoantibody subclass in the patient with Guillain-Barré syndrome, IgG4 was predominant in the patient with chronic inflammatory demyelinating polyradiculoneuropathy. Accordingly, complement deposition after binding to HEK293 cells was detectable in the patient with IgG3 autoantibodies only, not in the patient with IgG4. Severe disruption of the paranodal and nodal architecture was detectable in teased fibres of the sural nerve biopsy and in dermal myelinated fibres, supporting the notion of the paranodes being the site of pathology. Deposition of IgG at the paranodes was detected in teased fibre preparations of the sural nerve, further supporting the pathogenicity of anti-Caspr autoantibodies. Pain was one of the predominant findings in both patients, possibly reflected by binding of patients' IgG to TRPV1 immunoreactive dorsal root ganglia neurons. Our results demonstrate that the paranodal protein Caspr constitutes a new antigen that leads to autoantibody generation as part of the novel entity of neuropathies associated with autoantibodies against paranodal proteins.
ABSTRACT
Phycodnaviruses are dsDNA viruses, which infect algae. Their large genomes encode many gene products, like small K(+) channels, with homologs in prokaryotes and eukaryotes. Screening for K(+) channels revealed their abundance in viruses from fresh-water habitats. Recent sequencing of viruses from marine algae or from salt water in Antarctica revealed sequences with the predicted characteristics of K(+) channels but with some unexpected features. Two genes encode either 78 or 79 amino acid proteins, which are the smallest known K(+) channels. Also of interest is an unusual sequence in the canonical α-helixes in K(+) channels. Structural prediction algorithms indicate that the new channels have the conserved α-helix folds but the algorithms failed to identify the expected transmembrane domains flanking the K(+) channel pores. In spite of these unexpected properties electophysiological studies confirmed that the new proteins are functional K(+) channels.