Visualizing and Comparing Machine Learning Predictions to Improve Human-AI Teaming on the Example of Cell Lineage.

We visualize the predictions of multiple machine learning models to help biologists as they interactively make decisions about cell lineage-the development of a (plant) embryo from a single ovum cell. Based on a confocal microscopy dataset, traditionally biologists manually constructed the cell lineage, starting from this observation and reasoning backward in time to establish their inheritance. To speed up this tedious process, we make use of machine learning (ML) models trained on a database of manually established cell lineages to assist the biologist in cell assignment. Most biologists, however, are not familiar with ML, nor is it clear to them which model best predicts the embryo's development. We thus have developed a visualization system that is designed to support biologists in exploring and comparing ML models, checking the model predictions, detecting possible ML model mistakes, and deciding on the most likely embryo development. To evaluate our proposed system, we deployed our interface with six biologists in an observational study. Our results show that the visual representations of machine learning are easily understandable, and our tool, LineageD+, could potentially increase biologists' working efficiency and enhance the understanding of embryos.

TreeJ: an ImageJ plugin for interactive cell lineage reconstruction from static images.

BACKGROUND: With the emergence of deep-learning methods, tools are needed to capture and standardize image annotations made by experimentalists. In developmental biology, cell lineages are generally reconstructed from time-lapse data. However, some tissues need to be fixed to be accessible or to improve the staining. In this case, classical software do not offer the possibility of generating any lineage. Because of their rigid cell walls, plants present the advantage of keeping traces of the cell division history over successive generations in the cell patterns. To record this information despite having only a static image, dedicated tools are required. RESULTS: We developed an interface to assist users in the building and editing of a lineage tree from a 3D labeled image. Each cell within the tree can be tagged. From the created tree, cells of a sub-tree or cells sharing the same tag can be extracted. The tree can be exported in a format compatible with dedicated software for advanced graph visualization and manipulation. CONCLUSIONS: The TreeJ plugin for ImageJ/Fiji allows the user to generate and manipulate a lineage tree structure. The tree is compatible with other software to analyze the tree organization at the graphical level and at the cell pattern level. The code source is available at https://github.com/L-EL/TreeJ .

Cooperative cell-cell actin network remodeling to perform Gap junction endocytosis.

BACKGROUND: The endocytosis of Gap junction plaques (GJP) requires cytoskeletal forces to internalize such large membranous structures. Actin, which partners the connexin proteins constituting Gap junctions and is located close to Annular Gap Junctions (AGJ), could be actively involved in this physiological process. RESULTS: Electron Microscopy and Light Microscopy images, associated with time-lapse analysis and 3D reconstruction, used at high resolution and enhanced using ImageJ based software analysis, revealed that: i) actin cables, originating from Donor cells, insert on the edge of GJP and contribute to their invagination, giving rise to AGJ, whereas actin cables on the Acceptor cell side of the plaque are not modified; ii) actin cables from the Donor cell are continuous with the actin network present over the entire GJP surface. These actin cables fuse at a single point distant from the plaque, which then detaches itself from the membrane, condensing to form an actin mass during the final internalization process; iii) the Acceptor cell participates in the last step of the endocytic invagination process by forming an annular actin structure known as an actin ring. CONCLUSIONS: Together, these data suggest that the endocytosis of GJP is an example of a unique cooperative mechanism between the Donor (the traction of its actin cables) and the Acceptor cells (forming the actin ring).

RéSUMé: CONTEXTE: L'endocytose des plaques de jonctions communicantes ou jonctions gap (GJP) nécessite les forces du cytosquelette pour internaliser ces grandes structures membranaires. L'actine, partenaire des connexines, proteins constitutives des jonctions gap (Gj), localisée proche des jonctions gap annulaires (GJA), pourrait être impliquée dans ce processus physiologique. RéSULTATS: L' imagerie par microscopie optique et électronique, associées avec des analyses vidéo et des reconstructions en relief/3D, examinées à haute résolution et améliorées après traitement par des logiciels développés sous ImageJ, montrent que: i) des câbles d'actine, originaires des cellules donneuses, s'insèrent sur le bord des plaques jonctionnelles et facilitent leur invagination pour former les GJA tandis que les câbles d'actine des cellules receveuses ne sont pas modifies; ii) les câbles d'actine des cellules donneuses sont en continuité avec le réseau d'actine qui couvre la totalité de la surface de la plaque. De plus, ces câbles fusionnent en un point unique, à distance de la plaque, qui se détache de la région membranaire pour former une masse d'actine à la fin du processus d'endocytose; iii) la cellule receveuse participe à l'étape ultime du processus d'endocytose de la plaque en formant un anneau d'actine. CONCLUSIONS: L'ensemble de nos résultats montrent que l'endocytose des plaques jonctionnelles est un exemple de coopération unique entre la cellule donneuse (grâce à la traction des câbles d'actine) et la cellule receveuse (anneau d'actine).

Large-scale analysis and computer modeling reveal hidden regularities behind variability of cell division patterns in Arabidopsis thaliana embryogenesis.

Noise plays a major role in cellular processes and in the development of tissues and organs. Several studies have examined the origin, the integration or the accommodation of noise in gene expression, cell growth and elaboration of organ shape. By contrast, much less is known about variability in cell division plane positioning, its origin and links with cell geometry, and its impact on tissue organization. Taking advantage of the first-stereotyped-then-variable division patterns in the embryo of the model plant Arabidopsis thaliana, we combined 3D imaging and quantitative cell shape and cell lineage analysis together with mathematical and computer modeling to perform a large-scale, systematic analysis of variability in division plane orientation. Our results reveal that, paradoxically, variability in cell division patterns of Arabidopsis embryos is accompanied by a progressive reduction of heterogeneity in cell shape topology. The paradox is solved by showing that variability operates within a reduced repertoire of possible division plane orientations that is related to cell geometry. We show that in several domains of the embryo, a recently proposed geometrical division rule recapitulates observed variable patterns, suggesting that variable patterns emerge from deterministic principles operating in a variable geometrical context. Our work highlights the importance of emerging patterns in the plant embryo under iterated division principles, but also reveal domains where deviations between rule predictions and experimental observations point to additional regulatory mechanisms.

Challenges of intracellular visualization using virtual and augmented reality.

Microscopy image observation is commonly performed on 2D screens, which limits human capacities to grasp volumetric, complex, and discrete biological dynamics. With the massive production of multidimensional images (3D + time, multi-channels) and derived images (e.g., restored images, segmentation maps, and object tracks), scientists need appropriate visualization and navigation methods to better apprehend the amount of information in their content. New modes of visualization have emerged, including virtual reality (VR)/augmented reality (AR) approaches which should allow more accurate analysis and exploration of large time series of volumetric images, such as those produced by the latest 3D + time fluorescence microscopy. They include integrated algorithms that allow researchers to interactively explore complex spatiotemporal objects at the scale of single cells or multicellular systems, almost in a real time manner. In practice, however, immersion of the user within 3D + time microscopy data represents both a paradigm shift in human-image interaction and an acculturation challenge, for the concerned community. To promote a broader adoption of these approaches by biologists, further dialogue is needed between the bioimaging community and the VR&AR developers.

Inferring characteristics of bacterial swimming in biofilm matrix from time-lapse confocal laser scanning microscopy.

Biofilms are spatially organized communities of microorganisms embedded in a self-produced organic matrix, conferring to the population emerging properties such as an increased tolerance to the action of antimicrobials. It was shown that some bacilli were able to swim in the exogenous matrix of pathogenic biofilms and to counterbalance these properties. Swimming bacteria can deliver antimicrobial agents in situ, or potentiate the activity of antimicrobial by creating a transient vascularization network in the matrix. Hence, characterizing swimmer trajectories in the biofilm matrix is of particular interest to understand and optimize this new biocontrol strategy in particular, but also more generally to decipher ecological drivers of population spatial structure in natural biofilms ecosystems. In this study, a new methodology is developed to analyze time-lapse confocal laser scanning images to describe and compare the swimming trajectories of bacilli swimmers populations and their adaptations to the biofilm structure. The method is based on the inference of a kinetic model of swimmer populations including mechanistic interactions with the host biofilm. After validation on synthetic data, the methodology is implemented on images of three different species of motile bacillus species swimming in a Staphylococcus aureus biofilm. The fitted model allows to stratify the swimmer populations by their swimming behavior and provides insights into the mechanisms deployed by the micro-swimmers to adapt their swimming traits to the biofilm matrix.

Anyone who has ever cleaned a bathroom probably faced biofilms, the dark, slimy deposits that lurk around taps and pipes. These structures are created by bacteria which abandon their solitary lifestyle to work together as a community, secreting various substances that allow the cells to organise themselves in 3D and to better resist external aggression. Unwanted biofilms can impair industrial operations or endanger health, for example when they form inside medical equipment or water supplies. Removing these structures usually involves massive application of substances which can cause long-term damage to the environment. Recently, researchers have observed that a range of small rod-shaped bacteria ­ or 'bacilli' ­ can penetrate a harmful biofilm and dig transient tunnels in its 3D structure. These 'swimmers' can enhance the penetration of anti-microbial agents, or could even be modified to deliver these molecules right inside the biofilm. However, little is known about how the various types of bacilli, which have very different shapes and propelling systems, can navigate the complex environment that is a biofilm. This knowledge would be essential for scientists to select which swimmers could be the best to harness for industrial and medical applications. To investigate this question, Ravel et al. established a way to track how three species of bacilli swim inside a biofilm compared to in a simple fluid. A mathematical model was created which integrated several swimming behaviors such as speed adaptation and direction changes in response to the structure and density of the biofilm. This modelling was then fitted on microscopy images of the different species navigating the two types of environments. Different motion patterns for the three bacilli emerged, each showing different degrees of adapting to moving inside a biofilm. One species, in particular, was able to run straight in and out of this environment because it could adapt its speed to the biofilm density as well as randomly change direction. The new method developed by Ravel et al. can be redeployed to systematically study swimmer candidates in different types of biofilms. This would allow scientists to examine how various swimming characteristics impact how bacteria-killing chemicals can penetrate the altered biofilms. In addition, as the mathematical model can predict trajectories, it could be used in computational studies to examine which species of bacilli would be best suited in industrial settings.

Cell geometry determines symmetric and asymmetric division plane selection in Arabidopsis early embryos.

Plant tissue architecture and organ morphogenesis rely on the proper orientation of cell divisions. Previous attempts to predict division planes from cell geometry in plants mostly focused on 2D symmetric divisions. Using the stereotyped division patterns of Arabidopsis thaliana early embryogenesis, we investigated geometrical principles underlying plane selection in symmetric and in asymmetric divisions within complex 3D cell shapes. Introducing a 3D computational model of cell division, we show that area minimization constrained on passing through the cell centroid predicts observed divisions. Our results suggest that the positioning of division planes ensues from cell geometry and gives rise to spatially organized cell types with stereotyped shapes, thus underlining the role of self-organization in the developing architecture of the embryo. Our data further suggested the rule could be interpreted as surface minimization constrained by the nucleus position, which was validated using live imaging of cell divisions in the stomatal cell lineage.

Specialization of oleosins in oil body dynamics during seed development in Arabidopsis seeds.

Oil bodies (OBs) are seed-specific lipid storage organelles that allow the accumulation of neutral lipids that sustain plantlet development after the onset of germination. OBs are covered with specific proteins embedded in a single layer of phospholipids. Using fluorescent dyes and confocal microscopy, we monitored the dynamics of OBs in living Arabidopsis (Arabidopsis thaliana) embryos at different stages of development. Analyses were carried out with different genotypes: the wild type and three mutants affected in the accumulation of various oleosins (OLE1, OLE2, and OLE4), three major OB proteins. Image acquisition was followed by a detailed statistical analysis of OB size and distribution during seed development in the four dimensions (x, y, z, and t). Our results indicate that OB size increases sharply during seed maturation, in part by OB fusion, and then decreases until the end of the maturation process. In single, double, and triple mutant backgrounds, the size and spatial distribution of OBs are modified, affecting in turn the total lipid content, which suggests that the oleosins studied have specific functions in the dynamics of lipid accumulation.

Bacillus subtilis serine/threonine protein kinase YabT is involved in spore development via phosphorylation of a bacterial recombinase.

We characterized YabT, a serine/threonine kinase of the Hanks family, from Bacillus subtilis. YabT is a putative transmembrane kinase that lacks the canonical extracellular signal receptor domain. We demonstrate that YabT possesses a DNA-binding motif essential for its activation. In vivo YabT is expressed during sporulation and localizes to the asymmetric septum. Cells devoid of YabT sporulate more slowly and exhibit reduced resistance to DNA damage during sporulation. We established that YabT phosphorylates DNA-recombinase RecA at the residue serine 2. A non-phosphorylatable mutant of RecA exhibits the same phenotype as the ΔyabT mutant, and a phosphomimetic mutant of RecA complements ΔyabT, suggesting that YabT acts via RecA phosphorylation in vivo. During spore development, phosphorylation facilitates the formation of transient and mobile RecA foci that exhibit a scanning-like movement associated to the nucleoid in the mother cell. In some cells these foci persist at the end of spore development. We show that persistent RecA foci, which presumably coincide with irreparable lesions, are mutually exclusive with the completion of spore morphogenesis. Our results highlight similarities between the bacterial serine/threonine kinase YabT and eukaryal kinases C-Abl and Mec1, which are also activated by DNA, and phosphorylate proteins involved in DNA damage repair.

Role of bacterial peptidase F inferred by statistical analysis and further experimental validation.

Despite the quantity of high-throughput data available nowadays, the precise role of many proteins has not been elucidated. Available methods for classifying proteins and reconstructing metabolic networks are efficient for finding global categories, but do not answer the biologist's specific and targeted questions. Following Yamanishi et al. [Yamanishi, Y, Vert, JP, Nakaya, A, and Kaneisha, M (2003). "Extraction of correlated clusters from multiple genomic data by generalized kernel canonical correlation analysis." Bioinformatics 19, Suppl. 1, i323-i330] we used a kernel canonical correlation analysis (KCCA) to predict the role of the bacterial peptidase PepF. We integrated five existing data types: protein metabolic networks, microarray data, phylogenetic profiles, distances between proteins and incomplete two-dimensional-gel data (for which we propose a completion strategy), available for Lactococcus lactis to determine relationships between proteins. The predicted relationships were then used to guide our laboratory work which proved most of the predictions correct. PepF had previously been characterized as a zinc dependent endopeptidase [Nardi, M, Renault, P, and Monnet, V (1997). "Duplication of the pepF gene and shuffling of DNA fragments on the lactose plasmid of Lactococcus lactis." J. Bacteriol. 179, 4164-4171; Monnet, V, Nardi, M, Chopin, MC, and Gripon, JC (1994). "Biochemical and genetic characterization of PepF on oligoendopeptidase from Lactococcus lactis." J. Bio. Chem. 269, 32070-32076]. Analyzing a PepF mutant, we confirmed its participation in protein secretion through a strong relationship between the signal peptidase I and PepF predicted by the KCCA. The global nature of our approach made it possible to discover pleiotropic roles of the protein which had remained unknown using classical approaches.

Visualization and quantification of vesicle trafficking on a three-dimensional cytoskeleton network in living cells.

Recent progress in biology and microscopy has made it possible to acquire multidimensional data on rapid cellular activities. Unfortunately, the data analysis needed to describe the observed biological process still remains a major bottleneck. We here present a novel method of studying membrane trafficking by monitoring vesicular structures moving along a three-dimensional cytoskeleton network. It allows the dynamics of such structures to be qualitatively and quantitatively investigated. Our tracking method uses kymogram analysis to extract the consistent part of the temporal information and to allow the meaningful representation of vesicle dynamics. A fully automatic extension of this method, together with a statistical tool for dynamic comparisons, allows the precise analysis and comparison of overall speed distributions and directions. It can handle typical complex situations, such as a dense set of vesicles moving at various velocities, fusing and dissociating with each other or with other cell compartments. The overall approach has been characterized and validated on synthetic data. We chose the Rab6A protein, a GTPase involved in the regulation of intracellular membrane trafficking, as a molecular model. The application of our method to GFP-Rab6A stable cells acquired using fast four-dimensional deconvolution video-microscopy gives considerable cellular dynamic information unreachable using other techniques.

A model selection approach for robust spatio-temporal analysis of dynamics in 4D fluorescence videomicroscopy.

We describe a novel automatic approach for vesicle trafficking analysis in 3D+T videomicroscopy. Tracking individually objects in time in 3D+T videomicroscopy is known to be a very tedious job and leads generally to unreliable results. So instead, our method proceeds by first identifying trafficking regions in the 3D volume and next analysing at them the vesicle trafficking. The latter is viewed as significant change in the fluorescence of a region in the image. We embed the problem in a model selection framework and we resolve it using dynamic programming. We applied the proposed approach to analyse the vesicle dynamics related to the trafficking of the RAB6A protein between the Golgi apparatus and ER cell compartments.

Proteomic signature of Lactococcus lactis NCDO763 cultivated in milk.

We have compared the proteomic profiles of L. lactis subsp. cremoris NCDO763 growing in the synthetic medium M17Lac, skim milk microfiltrate (SMM), and skim milk. SMM was used as a simple model medium to reproduce the initial phase of growth of L. lactis in milk. To widen the analysis of the cytoplasmic proteome, we used two different gel systems (pH ranges of 4 to 7 and 4.5 to 5.5), and the proteins associated with the cell envelopes were also studied by two-dimensional electrophoresis. In the course of the study, we analyzed about 800 spots and identified 330 proteins by mass spectrometry. We observed that the levels of more than 50 and 30 proteins were significantly increased upon growth in SMM and milk, respectively. The large redeployment of protein synthesis was essentially associated with an activation of pathways involved in the metabolism of nitrogenous compounds: peptidolytic and peptide transport systems, amino acid biosynthesis and interconversion, and de novo biosynthesis of purines. We also showed that enzymes involved in reactions feeding the purine biosynthetic pathway in one-carbon units and amino acids have an increased level in SMM and milk. The analysis of the proteomic data suggested that the glutamine synthetase (GS) would play a pivotal role in the adaptation to SMM and milk. The analysis of glnA expression during growth in milk and the construction of a glnA-defective mutant confirmed that GS is an essential enzyme for the development of L. lactis in dairy media. This analysis thus provides a proteomic signature of L. lactis, a model lactic acid bacterium, growing in its technological environment.

Quantitative trait loci controlling root growth and architecture in Arabidopsis thaliana confirmed by heterogeneous inbred family.

Arabidopsis thaliana provides a scientifically attractive and simple model for studying root growth and architecture and, subsequently, for discovering new genes involved in the control of these characters in plants. We have used the natural variation available in Arabidopsis accessions and mapped quantitative trait loci (QTLs) for primary root length (PRL), lateral root number (LRN) and density (LRD) and for total length of the lateral root system (LRL) in the Bay-0 x Shahdara population. Total phenotypic variation was very large, and despite the importance of the environmental component we were able to map 13 QTLs and one epistatic interaction between QTLs. Our results highlight the biological relevance and genetic control of lateral root density in this material. We were also able to show that variation in the extent of the lateral root system depends mainly on the growth of the existing lateral roots rather than in a change in LRN. Factors controlling lateral root growth seemed to have no major effect on primary root growth. Moreover, Shahdara QTL alleles always increased the length of the lateral roots, which may be taken as an adaptation to its very dry natural environment in Tadjikistan. A QTL for PRL was confirmed using a type of near-isogenic line called a heterogeneous inbred family (HIF), and this QTL is a candidate for further fine-mapping and cloning.

PARIS: a proteomic analysis and resources indexation system.

UNLABELLED: We developed a system for managing data from two-dimensional electrophoresis-based proteomic experiments. Named PARIS, the system stores gel image and information about experiments and analysis procedures, allows the user to search and navigate in genomic and proteomic data, supports visual verification and validation of the analysis results, and provides tools for cross multi-experiment and multi-experimenter data validation and exploration. AVAILABILITY: The software is freely available from http://www.inra.fr/bia/J/imaste/Projets/PARIS/index.html

Three-dimensional binding of epidermal growth factor peptides in colonic tissues produced from rotating bioreactor.

Epidermal growth factor peptide binding was analyzed on primary cultures of colonic cells and along crypts by fluorescent laser-scanning confocal microscopy, using a three-dimensional image analysis software (Quant3D, Linux/Unix). Structural, proliferative units from primary cultures grown in rotating bioreactor for 41 d were arranged according to a tubular symmetry or on a parallelepiped sheet. Mean width, height, and depth of 23 tissue-like masses (+/- standard error) were 125 microm (+/-16), 152 microm (+/-23), and 29 microm (+/-3), respectively. Mean density of nuclei in tissue-like masses, expressed as the number of nuclei per cubic millimeter (+/- standard error of the mean), was 1.8 x 10(5) (+/-0.7 x 10(5)) nuclei per cubic millimeter, which corresponded to a density that was five to six times lower than that estimated for the colonic crypt isolated by chelation. Spots of high epidermal growth factor (EGF) peptide binding that corresponded to microlesions in crypt monolayers or to active colonization of microcarriers by epithelial and stromal cells in tissue-like masses were observed. The relative intensities of EGF peptide binding that were obtained below cell position 8 on crypts were very homogeneous and were representative of the profile obtained with crypts isolated from adult rats adapted to a normal diet and used to develop primary cultures of colonocytes in our laboratory. A microscopic multidimensional analytic system to record the expression profiles of biomarkers along intestinal tissues should enhance the use of primary cultures of colonocytes for in vitro testing of new food products.