Morphological Characterization of Antibiotic Combinations.

Combination therapies are common in many therapeutic contexts, including infectious diseases and cancer. A common approach for evaluating combinations in vitro is to assess effects on cell growth as synergistic, antagonistic, or neutral using "checkerboard" experiments to systematically sample combinations of agents in multiple doses. To further understand the effects of antibiotic combinations, we employed high-content imaging to study the morphological changes caused by combination treatments in checkerboard experiments. Using an automated, unsupervised image analysis approach to group morphologies, and an "expert-in-the-loop" to annotate them, we attributed the heterogeneous morphological changes ofEscherichia coli cells to varying doses of both single-agent and combination treatments. We identified patterns of morphological change, including morphological potentiation, competition, and the emergence of unexpected morphologies. We found these frequently did not correlate with synergistic or antagonistic effects on viability, suggesting morphological approaches may provide a distinctive signature of the biological interaction between compounds over a range of conditions. Among the unexpected morphologies we observed, there were transitional changes associated with intermediate doses of compounds and uncharacterized phenotypes associated with well-studied antibiotics. Our approach exemplifies how unsupervised image analysis and expert knowledge can be combined to reckon with complex phenotypic changes arising from combination screening, dose titrations, or polypharmacology. In this way, quantification of morphological diversity across treatment conditions could aid in analysis and prioritization of complementary pairings of antibiotic agents by more closely capturing the true signature of the integrated cellular response.

Automated detection and staging of malaria parasites from cytological smears using convolutional neural networks.

Microscopic examination of blood smears remains the gold standard for laboratory inspection and diagnosis of malaria. Smear inspection is, however, time-consuming and dependent on trained microscopists with results varying in accuracy. We sought to develop an automated image analysis method to improve accuracy and standardization of smear inspection that retains capacity for expert confirmation and image archiving. Here, we present a machine learning method that achieves red blood cell (RBC) detection, differentiation between infected/uninfected cells, and parasite life stage categorization from unprocessed, heterogeneous smear images. Based on a pretrained Faster Region-Based Convolutional Neural Networks (R-CNN) model for RBC detection, our model performs accurately, with an average precision of 0.99 at an intersection-over-union threshold of 0.5. Application of a residual neural network-50 model to infected cells also performs accurately, with an area under the receiver operating characteristic curve of 0.98. Finally, combining our method with a regression model successfully recapitulates intraerythrocytic developmental cycle with accurate lifecycle stage categorization. Combined with a mobile-friendly web-based interface, called PlasmoCount, our method permits rapid navigation through and review of results for quality assurance. By standardizing assessment of Giemsa smears, our method markedly improves inspection reproducibility and presents a realistic route to both routine lab and future field-based automated malaria diagnosis.

Batch equalization with a generative adversarial network.

MOTIVATION: Advances in automation and imaging have made it possible to capture a large image dataset that spans multiple experimental batches of data. However, accurate biological comparison across the batches is challenged by batch-to-batch variation (i.e. batch effect) due to uncontrollable experimental noise (e.g. varying stain intensity or cell density). Previous approaches to minimize the batch effect have commonly focused on normalizing the low-dimensional image measurements such as an embedding generated by a neural network. However, normalization of the embedding could suffer from over-correction and alter true biological features (e.g. cell size) due to our limited ability to interpret the effect of the normalization on the embedding space. Although techniques like flat-field correction can be applied to normalize the image values directly, they are limited transformations that handle only simple artifacts due to batch effect. RESULTS: We present a neural network-based batch equalization method that can transfer images from one batch to another while preserving the biological phenotype. The equalization method is trained as a generative adversarial network (GAN), using the StarGAN architecture that has shown considerable ability in style transfer. After incorporating new objectives that disentangle batch effect from biological features, we show that the equalized images have less batch information and preserve the biological information. We also demonstrate that the same model training parameters can generalize to two dramatically different types of cells, indicating this approach could be broadly applicable. AVAILABILITY AND IMPLEMENTATION: https://github.com/tensorflow/gan/tree/master/tensorflow_gan/examples/stargan. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A machine learning approach to define antimalarial drug action from heterogeneous cell-based screens.

Drug resistance threatens the effective prevention and treatment of an ever-increasing range of human infections. This highlights an urgent need for new and improved drugs with novel mechanisms of action to avoid cross-resistance. Current cell-based drug screens are, however, restricted to binary live/dead readouts with no provision for mechanism of action prediction. Machine learning methods are increasingly being used to improve information extraction from imaging data. These methods, however, work poorly with heterogeneous cellular phenotypes and generally require time-consuming human-led training. We have developed a semi-supervised machine learning approach, combining human- and machine-labeled training data from mixed human malaria parasite cultures. Designed for high-throughput and high-resolution screening, our semi-supervised approach is robust to natural parasite morphological heterogeneity and correctly orders parasite developmental stages. Our approach also reproducibly detects and clusters drug-induced morphological outliers by mechanism of action, demonstrating the potential power of machine learning for accelerating cell-based drug discovery.

Morphological profiling of tubercle bacilli identifies drug pathways of action.

Morphological profiling is a method to classify target pathways of antibacterials based on how bacteria respond to treatment through changes to cellular shape and spatial organization. Here we utilized the cell-to-cell variation in morphological features of Mycobacterium tuberculosis bacilli to develop a rapid profiling platform called Morphological Evaluation and Understanding of Stress (MorphEUS). MorphEUS classified 94% of tested drugs correctly into broad categories according to modes of action previously identified in the literature. In the other 6%, MorphEUS pointed to key off-target activities. We observed cell wall damage induced by bedaquiline and moxifloxacin through secondary effects downstream from their main target pathways. We implemented MorphEUS to correctly classify three compounds in a blinded study and identified an off-target effect for one compound that was not readily apparent in previous studies. We anticipate that the ability of MorphEUS to rapidly identify pathways of drug action and the proximal cause of cellular damage in tubercle bacilli will make it applicable to other pathogens and cell types where morphological responses are subtle and heterogeneous.

Applying Deep Neural Network Analysis to High-Content Image-Based Assays.

The etiological underpinnings of many CNS disorders are not well understood. This is likely due to the fact that individual diseases aggregate numerous pathological subtypes, each associated with a complex landscape of genetic risk factors. To overcome these challenges, researchers are integrating novel data types from numerous patients, including imaging studies capturing broadly applicable features from patient-derived materials. These datasets, when combined with machine learning, potentially hold the power to elucidate the subtle patterns that stratify patients by shared pathology. In this study, we interrogated whether high-content imaging of primary skin fibroblasts, using the Cell Painting method, could reveal disease-relevant information among patients. First, we showed that technical features such as batch/plate type, plate, and location within a plate lead to detectable nuisance signals, as revealed by a pre-trained deep neural network and analysis with deep image embeddings. Using a plate design and image acquisition strategy that accounts for these variables, we performed a pilot study with 12 healthy controls and 12 subjects affected by the severe genetic neurological disorder spinal muscular atrophy (SMA), and evaluated whether a convolutional neural network (CNN) generated using a subset of the cells could distinguish disease states on cells from the remaining unseen control-SMA pair. Our results indicate that these two populations could effectively be differentiated from one another and that model selectivity is insensitive to batch/plate type. One caveat is that the samples were also largely separated by source. These findings lay a foundation for how to conduct future studies exploring diseases with more complex genetic contributions and unknown subtypes.

In Silico Labeling: Predicting Fluorescent Labels in Unlabeled Images.

Microscopy is a central method in life sciences. Many popular methods, such as antibody labeling, are used to add physical fluorescent labels to specific cellular constituents. However, these approaches have significant drawbacks, including inconsistency; limitations in the number of simultaneous labels because of spectral overlap; and necessary perturbations of the experiment, such as fixing the cells, to generate the measurement. Here, we show that a computational machine-learning approach, which we call "in silico labeling" (ISL), reliably predicts some fluorescent labels from transmitted-light images of unlabeled fixed or live biological samples. ISL predicts a range of labels, such as those for nuclei, cell type (e.g., neural), and cell state (e.g., cell death). Because prediction happens in silico, the method is consistent, is not limited by spectral overlap, and does not disturb the experiment. ISL generates biological measurements that would otherwise be problematic or impossible to acquire.

Autophagy induction enhances TDP43 turnover and survival in neuronal ALS models.

Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) have distinct clinical features but a common pathology--cytoplasmic inclusions rich in transactive response element DNA-binding protein of 43 kDa (TDP43). Rare TDP43 mutations cause ALS or FTD, but abnormal TDP43 levels and localization may cause disease even if TDP43 lacks a mutation. Here we show that individual neurons vary in their ability to clear TDP43 and are exquisitely sensitive to TDP43 levels. To measure TDP43 clearance, we developed and validated a single-cell optical method that overcomes the confounding effects of aggregation and toxicity and discovered that pathogenic mutations shorten TDP43 half-life. New compounds that stimulate autophagy improved TDP43 clearance and localization and enhanced survival in primary murine neurons and in human stem cell-derived neurons and astrocytes harboring mutant TDP43. These findings indicate that the levels and localization of TDP43 critically determine neurotoxicity and show that autophagy induction mitigates neurodegeneration by acting directly on TDP43 clearance.

The RNA-binding protein TDP-43 selectively disrupts microRNA-1/206 incorporation into the RNA-induced silencing complex.

MicroRNA (miRNA) maturation is regulated by interaction of particular miRNA precursors with specific RNA-binding proteins. Following their biogenesis, mature miRNAs are incorporated into the RNA-induced silencing complex (RISC) where they interact with mRNAs to negatively regulate protein production. However, little is known about how mature miRNAs are regulated at the level of their activity. To address this, we screened for proteins differentially bound to the mature form of the miR-1 or miR-133 miRNA families. These muscle-enriched, co-transcribed miRNA pairs cooperate to suppress smooth muscle gene expression in the heart. However, they also have opposing roles, with the miR-1 family, composed of miR-1 and miR-206, promoting myogenic differentiation, whereas miR-133 maintains the progenitor state. Here, we describe a physical interaction between TDP-43, an RNA-binding protein that forms aggregates in the neuromuscular disease, amyotrophic lateral sclerosis, and the miR-1, but not miR-133, family. Deficiency of the TDP-43 Drosophila ortholog enhanced dmiR-1 activity in vivo. In mammalian cells, TDP-43 limited the activity of both miR-1 and miR-206, but not the miR-133 family, by disrupting their RISC association. Consistent with TDP-43 dampening miR-1/206 activity, protein levels of the miR-1/206 targets, IGF-1 and HDAC4, were elevated in TDP-43 transgenic mouse muscle. This occurred without corresponding Igf-1 or Hdac4 mRNA increases and despite higher miR-1 and miR-206 expression. Our findings reveal that TDP-43 negatively regulates the activity of the miR-1 family of miRNAs by limiting their bioavailability for RISC loading and suggest a processing-independent mechanism for differential regulation of miRNA activity.

Proteostasis of polyglutamine varies among neurons and predicts neurodegeneration.

In polyglutamine (polyQ) diseases, only certain neurons die, despite widespread expression of the offending protein. PolyQ expansion may induce neurodegeneration by impairing proteostasis, but protein aggregation and toxicity tend to confound conventional measurements of protein stability. Here, we used optical pulse labeling to measure effects of polyQ expansions on the mean lifetime of a fragment of huntingtin, the protein that causes Huntington's disease, in living neurons. We show that polyQ expansion reduced the mean lifetime of mutant huntingtin within a given neuron and that the mean lifetime varied among neurons, indicating differences in their capacity to clear the polypeptide. We found that neuronal longevity is predicted by the mean lifetime of huntingtin, as cortical neurons cleared mutant huntingtin faster and lived longer than striatal neurons. Thus, cell type-specific differences in turnover capacity may contribute to cellular susceptibility to toxic proteins, and efforts to bolster proteostasis in Huntington's disease, such as protein clearance, could be neuroprotective.

Longitudinal imaging and analysis of neurons expressing polyglutamine-expanded proteins.

Misfolded proteins have been implicated in most of the major neurodegenerative diseases, and identifying drugs and pathways that protect neurons from the toxicity of misfolded proteins is of paramount importance. We invented a form of automated imaging and analysis called robotic microscopy that is well suited to the study of neurodegeneration. It enables the monitoring of large cohorts of individual neurons over their lifetimes as they undergo neurodegeneration. With automated analysis, multiple endpoints in neurons can be measured, including survival. Statistical approaches, typically reserved for engineering and clinical medicine, can be applied to these data in an unbiased fashion to discover whether factors contribute positively or negatively to neuronal fate and to quantify the importance of their contribution. Ultimately, multivariate dynamic models can be constructed from these data, which can provide a systems-level understanding of the neurodegenerative disease process and guide the rationale for the development of therapies.

High-throughput screening in primary neurons.

Despite years of incremental progress in our understanding of diseases such as Alzheimer's disease (AD), Parkinson's disease (PD), Huntington's disease (HD), and amyotrophic lateral sclerosis (ALS), there are still no disease-modifying therapeutics. The discrepancy between the number of lead compounds and approved drugs may partially be a result of the methods used to generate the leads and highlights the need for new technology to obtain more detailed and physiologically relevant information on cellular processes in normal and diseased states. Our high-throughput screening (HTS) system in a primary neuron model can help address this unmet need. HTS allows scientists to assay thousands of conditions in a short period of time which can reveal completely new aspects of biology and identify potential therapeutics in the span of a few months when conventional methods could take years or fail all together. HTS in primary neurons combines the advantages of HTS with the biological relevance of intact, fully differentiated neurons which can capture the critical cellular events or homeostatic states that make neurons uniquely susceptible to disease-associated proteins. We detail methodologies of our primary neuron HTS assay workflow from sample preparation to data reporting. We also discuss the adaptation of our HTS system into high-content screening (HCS), a type of HTS that uses multichannel fluorescence images to capture biological events in situ, and is uniquely suited to study dynamical processes in living cells.

No difference in kinetics of tau or histone phosphorylation by CDK5/p25 versus CDK5/p35 in vitro.

CDK5/p35 is a cyclin-dependent kinase essential for normal neuron function. Proteolysis of the p35 subunit in vivo results in CDK5/p25 that causes neurotoxicity associated with a number of neurodegenerative diseases. Whereas the mechanism by which conversion of p35 to p25 leads to toxicity is unknown, there is common belief that CDK5/p25 is catalytically hyperactive compared to CDK5/p35. Here, we have compared the steady-state kinetic parameters of CDK5/p35 and CDK5/p25 towards both histone H1, the best known substrate for both enzymes, and the microtubule-associated protein, tau, a physiological substrate whose in vivo phosphorylation is relevant to Alzheimer's disease. We show that the kinetics of both enzymes are the same towards either substrate in vitro. Furthermore, both enzymes display virtually identical kinetics towards individual phosphorylation sites in tau monitored by NMR. We conclude that conversion of p35 to p25 does not alter the catalytic efficiency of the CDK5 catalytic subunit by using histone H1 or tau as substrates, and that neurotoxicity associated with CDK5/p25 is unlikely attributable to CDK5 hyperactivation, as measured in vitro.